Contents of /trunk/kernel-alx/patches-4.9/0313-4.9.214-all-fixes.patch
Parent Directory
|
Revision Log
Revision 3588 -
(show annotations)
(download)
Thu Aug 13 10:21:30 2020 UTC (4 years, 1 month ago) by niro
File size: 355649 byte(s)
Thu Aug 13 10:21:30 2020 UTC (4 years, 1 month ago) by niro
File size: 355649 byte(s)
linux-214
1 | diff --git a/Makefile b/Makefile |
2 | index de79c801abcd..9a6aa41a9ec1 100644 |
3 | --- a/Makefile |
4 | +++ b/Makefile |
5 | @@ -1,6 +1,6 @@ |
6 | VERSION = 4 |
7 | PATCHLEVEL = 9 |
8 | -SUBLEVEL = 213 |
9 | +SUBLEVEL = 214 |
10 | EXTRAVERSION = |
11 | NAME = Roaring Lionus |
12 | |
13 | diff --git a/arch/arc/boot/dts/axs10x_mb.dtsi b/arch/arc/boot/dts/axs10x_mb.dtsi |
14 | index d6c1bbc98ac3..15698b3e490f 100644 |
15 | --- a/arch/arc/boot/dts/axs10x_mb.dtsi |
16 | +++ b/arch/arc/boot/dts/axs10x_mb.dtsi |
17 | @@ -63,6 +63,7 @@ |
18 | interrupt-names = "macirq"; |
19 | phy-mode = "rgmii"; |
20 | snps,pbl = < 32 >; |
21 | + snps,multicast-filter-bins = <256>; |
22 | clocks = <&apbclk>; |
23 | clock-names = "stmmaceth"; |
24 | max-speed = <100>; |
25 | diff --git a/arch/arm/boot/dts/sama5d3.dtsi b/arch/arm/boot/dts/sama5d3.dtsi |
26 | index 4c84d333fc7e..33c0d2668934 100644 |
27 | --- a/arch/arm/boot/dts/sama5d3.dtsi |
28 | +++ b/arch/arm/boot/dts/sama5d3.dtsi |
29 | @@ -1109,49 +1109,49 @@ |
30 | usart0_clk: usart0_clk { |
31 | #clock-cells = <0>; |
32 | reg = <12>; |
33 | - atmel,clk-output-range = <0 66000000>; |
34 | + atmel,clk-output-range = <0 83000000>; |
35 | }; |
36 | |
37 | usart1_clk: usart1_clk { |
38 | #clock-cells = <0>; |
39 | reg = <13>; |
40 | - atmel,clk-output-range = <0 66000000>; |
41 | + atmel,clk-output-range = <0 83000000>; |
42 | }; |
43 | |
44 | usart2_clk: usart2_clk { |
45 | #clock-cells = <0>; |
46 | reg = <14>; |
47 | - atmel,clk-output-range = <0 66000000>; |
48 | + atmel,clk-output-range = <0 83000000>; |
49 | }; |
50 | |
51 | usart3_clk: usart3_clk { |
52 | #clock-cells = <0>; |
53 | reg = <15>; |
54 | - atmel,clk-output-range = <0 66000000>; |
55 | + atmel,clk-output-range = <0 83000000>; |
56 | }; |
57 | |
58 | uart0_clk: uart0_clk { |
59 | #clock-cells = <0>; |
60 | reg = <16>; |
61 | - atmel,clk-output-range = <0 66000000>; |
62 | + atmel,clk-output-range = <0 83000000>; |
63 | }; |
64 | |
65 | twi0_clk: twi0_clk { |
66 | reg = <18>; |
67 | #clock-cells = <0>; |
68 | - atmel,clk-output-range = <0 16625000>; |
69 | + atmel,clk-output-range = <0 41500000>; |
70 | }; |
71 | |
72 | twi1_clk: twi1_clk { |
73 | #clock-cells = <0>; |
74 | reg = <19>; |
75 | - atmel,clk-output-range = <0 16625000>; |
76 | + atmel,clk-output-range = <0 41500000>; |
77 | }; |
78 | |
79 | twi2_clk: twi2_clk { |
80 | #clock-cells = <0>; |
81 | reg = <20>; |
82 | - atmel,clk-output-range = <0 16625000>; |
83 | + atmel,clk-output-range = <0 41500000>; |
84 | }; |
85 | |
86 | mci0_clk: mci0_clk { |
87 | @@ -1167,19 +1167,19 @@ |
88 | spi0_clk: spi0_clk { |
89 | #clock-cells = <0>; |
90 | reg = <24>; |
91 | - atmel,clk-output-range = <0 133000000>; |
92 | + atmel,clk-output-range = <0 166000000>; |
93 | }; |
94 | |
95 | spi1_clk: spi1_clk { |
96 | #clock-cells = <0>; |
97 | reg = <25>; |
98 | - atmel,clk-output-range = <0 133000000>; |
99 | + atmel,clk-output-range = <0 166000000>; |
100 | }; |
101 | |
102 | tcb0_clk: tcb0_clk { |
103 | #clock-cells = <0>; |
104 | reg = <26>; |
105 | - atmel,clk-output-range = <0 133000000>; |
106 | + atmel,clk-output-range = <0 166000000>; |
107 | }; |
108 | |
109 | pwm_clk: pwm_clk { |
110 | @@ -1190,7 +1190,7 @@ |
111 | adc_clk: adc_clk { |
112 | #clock-cells = <0>; |
113 | reg = <29>; |
114 | - atmel,clk-output-range = <0 66000000>; |
115 | + atmel,clk-output-range = <0 83000000>; |
116 | }; |
117 | |
118 | dma0_clk: dma0_clk { |
119 | @@ -1221,13 +1221,13 @@ |
120 | ssc0_clk: ssc0_clk { |
121 | #clock-cells = <0>; |
122 | reg = <38>; |
123 | - atmel,clk-output-range = <0 66000000>; |
124 | + atmel,clk-output-range = <0 83000000>; |
125 | }; |
126 | |
127 | ssc1_clk: ssc1_clk { |
128 | #clock-cells = <0>; |
129 | reg = <39>; |
130 | - atmel,clk-output-range = <0 66000000>; |
131 | + atmel,clk-output-range = <0 83000000>; |
132 | }; |
133 | |
134 | sha_clk: sha_clk { |
135 | diff --git a/arch/arm/boot/dts/sama5d3_can.dtsi b/arch/arm/boot/dts/sama5d3_can.dtsi |
136 | index c5a3772741bf..0fac79f75c06 100644 |
137 | --- a/arch/arm/boot/dts/sama5d3_can.dtsi |
138 | +++ b/arch/arm/boot/dts/sama5d3_can.dtsi |
139 | @@ -37,13 +37,13 @@ |
140 | can0_clk: can0_clk { |
141 | #clock-cells = <0>; |
142 | reg = <40>; |
143 | - atmel,clk-output-range = <0 66000000>; |
144 | + atmel,clk-output-range = <0 83000000>; |
145 | }; |
146 | |
147 | can1_clk: can1_clk { |
148 | #clock-cells = <0>; |
149 | reg = <41>; |
150 | - atmel,clk-output-range = <0 66000000>; |
151 | + atmel,clk-output-range = <0 83000000>; |
152 | }; |
153 | }; |
154 | }; |
155 | diff --git a/arch/arm/boot/dts/sama5d3_tcb1.dtsi b/arch/arm/boot/dts/sama5d3_tcb1.dtsi |
156 | index 801f9745e82f..b80dbc45a3c2 100644 |
157 | --- a/arch/arm/boot/dts/sama5d3_tcb1.dtsi |
158 | +++ b/arch/arm/boot/dts/sama5d3_tcb1.dtsi |
159 | @@ -23,6 +23,7 @@ |
160 | tcb1_clk: tcb1_clk { |
161 | #clock-cells = <0>; |
162 | reg = <27>; |
163 | + atmel,clk-output-range = <0 166000000>; |
164 | }; |
165 | }; |
166 | }; |
167 | diff --git a/arch/arm/boot/dts/sama5d3_uart.dtsi b/arch/arm/boot/dts/sama5d3_uart.dtsi |
168 | index 2511d748867b..71818c7bfb67 100644 |
169 | --- a/arch/arm/boot/dts/sama5d3_uart.dtsi |
170 | +++ b/arch/arm/boot/dts/sama5d3_uart.dtsi |
171 | @@ -42,13 +42,13 @@ |
172 | uart0_clk: uart0_clk { |
173 | #clock-cells = <0>; |
174 | reg = <16>; |
175 | - atmel,clk-output-range = <0 66000000>; |
176 | + atmel,clk-output-range = <0 83000000>; |
177 | }; |
178 | |
179 | uart1_clk: uart1_clk { |
180 | #clock-cells = <0>; |
181 | reg = <17>; |
182 | - atmel,clk-output-range = <0 66000000>; |
183 | + atmel,clk-output-range = <0 83000000>; |
184 | }; |
185 | }; |
186 | }; |
187 | diff --git a/arch/arm/mach-tegra/sleep-tegra30.S b/arch/arm/mach-tegra/sleep-tegra30.S |
188 | index 16e5ff03383c..91b3f06e5425 100644 |
189 | --- a/arch/arm/mach-tegra/sleep-tegra30.S |
190 | +++ b/arch/arm/mach-tegra/sleep-tegra30.S |
191 | @@ -382,6 +382,14 @@ _pll_m_c_x_done: |
192 | pll_locked r1, r0, CLK_RESET_PLLC_BASE |
193 | pll_locked r1, r0, CLK_RESET_PLLX_BASE |
194 | |
195 | + tegra_get_soc_id TEGRA_APB_MISC_BASE, r1 |
196 | + cmp r1, #TEGRA30 |
197 | + beq 1f |
198 | + ldr r1, [r0, #CLK_RESET_PLLP_BASE] |
199 | + bic r1, r1, #(1<<31) @ disable PllP bypass |
200 | + str r1, [r0, #CLK_RESET_PLLP_BASE] |
201 | +1: |
202 | + |
203 | mov32 r7, TEGRA_TMRUS_BASE |
204 | ldr r1, [r7] |
205 | add r1, r1, #LOCK_DELAY |
206 | @@ -641,7 +649,10 @@ tegra30_switch_cpu_to_clk32k: |
207 | str r0, [r4, #PMC_PLLP_WB0_OVERRIDE] |
208 | |
209 | /* disable PLLP, PLLA, PLLC and PLLX */ |
210 | + tegra_get_soc_id TEGRA_APB_MISC_BASE, r1 |
211 | + cmp r1, #TEGRA30 |
212 | ldr r0, [r5, #CLK_RESET_PLLP_BASE] |
213 | + orrne r0, r0, #(1 << 31) @ enable PllP bypass on fast cluster |
214 | bic r0, r0, #(1 << 30) |
215 | str r0, [r5, #CLK_RESET_PLLP_BASE] |
216 | ldr r0, [r5, #CLK_RESET_PLLA_BASE] |
217 | diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig |
218 | index fa8f2aa88189..f529d3d9d88d 100644 |
219 | --- a/arch/powerpc/Kconfig |
220 | +++ b/arch/powerpc/Kconfig |
221 | @@ -85,6 +85,7 @@ config PPC |
222 | select BINFMT_ELF |
223 | select ARCH_HAS_ELF_RANDOMIZE |
224 | select OF |
225 | + select OF_DMA_DEFAULT_COHERENT if !NOT_COHERENT_CACHE |
226 | select OF_EARLY_FLATTREE |
227 | select OF_RESERVED_MEM |
228 | select HAVE_FTRACE_MCOUNT_RECORD |
229 | diff --git a/arch/powerpc/boot/4xx.c b/arch/powerpc/boot/4xx.c |
230 | index 9d3bd4c45a24..1c4354f922fd 100644 |
231 | --- a/arch/powerpc/boot/4xx.c |
232 | +++ b/arch/powerpc/boot/4xx.c |
233 | @@ -232,7 +232,7 @@ void ibm4xx_denali_fixup_memsize(void) |
234 | dpath = 8; /* 64 bits */ |
235 | |
236 | /* get address pins (rows) */ |
237 | - val = SDRAM0_READ(DDR0_42); |
238 | + val = SDRAM0_READ(DDR0_42); |
239 | |
240 | row = DDR_GET_VAL(val, DDR_APIN, DDR_APIN_SHIFT); |
241 | if (row > max_row) |
242 | diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c |
243 | index e840f943cd2c..5cf1392dff96 100644 |
244 | --- a/arch/powerpc/kvm/book3s_hv.c |
245 | +++ b/arch/powerpc/kvm/book3s_hv.c |
246 | @@ -1766,7 +1766,7 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm, |
247 | mutex_unlock(&kvm->lock); |
248 | |
249 | if (!vcore) |
250 | - goto free_vcpu; |
251 | + goto uninit_vcpu; |
252 | |
253 | spin_lock(&vcore->lock); |
254 | ++vcore->num_threads; |
255 | @@ -1782,6 +1782,8 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm, |
256 | |
257 | return vcpu; |
258 | |
259 | +uninit_vcpu: |
260 | + kvm_vcpu_uninit(vcpu); |
261 | free_vcpu: |
262 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
263 | out: |
264 | diff --git a/arch/powerpc/kvm/book3s_pr.c b/arch/powerpc/kvm/book3s_pr.c |
265 | index e0d88d0890aa..8172021bcee6 100644 |
266 | --- a/arch/powerpc/kvm/book3s_pr.c |
267 | +++ b/arch/powerpc/kvm/book3s_pr.c |
268 | @@ -1482,10 +1482,12 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm, |
269 | |
270 | err = kvmppc_mmu_init(vcpu); |
271 | if (err < 0) |
272 | - goto uninit_vcpu; |
273 | + goto free_shared_page; |
274 | |
275 | return vcpu; |
276 | |
277 | +free_shared_page: |
278 | + free_page((unsigned long)vcpu->arch.shared); |
279 | uninit_vcpu: |
280 | kvm_vcpu_uninit(vcpu); |
281 | free_shadow_vcpu: |
282 | diff --git a/arch/powerpc/platforms/pseries/hotplug-memory.c b/arch/powerpc/platforms/pseries/hotplug-memory.c |
283 | index 6c12b02f4a61..eee45b9220e0 100644 |
284 | --- a/arch/powerpc/platforms/pseries/hotplug-memory.c |
285 | +++ b/arch/powerpc/platforms/pseries/hotplug-memory.c |
286 | @@ -398,8 +398,10 @@ static bool lmb_is_removable(struct of_drconf_cell *lmb) |
287 | |
288 | for (i = 0; i < scns_per_block; i++) { |
289 | pfn = PFN_DOWN(phys_addr); |
290 | - if (!pfn_present(pfn)) |
291 | + if (!pfn_present(pfn)) { |
292 | + phys_addr += MIN_MEMORY_BLOCK_SIZE; |
293 | continue; |
294 | + } |
295 | |
296 | rc &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
297 | phys_addr += MIN_MEMORY_BLOCK_SIZE; |
298 | diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c |
299 | index 0024e451bb36..c0f094c96cd6 100644 |
300 | --- a/arch/powerpc/platforms/pseries/iommu.c |
301 | +++ b/arch/powerpc/platforms/pseries/iommu.c |
302 | @@ -167,10 +167,10 @@ static unsigned long tce_get_pseries(struct iommu_table *tbl, long index) |
303 | return be64_to_cpu(*tcep); |
304 | } |
305 | |
306 | -static void tce_free_pSeriesLP(struct iommu_table*, long, long); |
307 | +static void tce_free_pSeriesLP(unsigned long liobn, long, long); |
308 | static void tce_freemulti_pSeriesLP(struct iommu_table*, long, long); |
309 | |
310 | -static int tce_build_pSeriesLP(struct iommu_table *tbl, long tcenum, |
311 | +static int tce_build_pSeriesLP(unsigned long liobn, long tcenum, long tceshift, |
312 | long npages, unsigned long uaddr, |
313 | enum dma_data_direction direction, |
314 | unsigned long attrs) |
315 | @@ -181,25 +181,25 @@ static int tce_build_pSeriesLP(struct iommu_table *tbl, long tcenum, |
316 | int ret = 0; |
317 | long tcenum_start = tcenum, npages_start = npages; |
318 | |
319 | - rpn = __pa(uaddr) >> TCE_SHIFT; |
320 | + rpn = __pa(uaddr) >> tceshift; |
321 | proto_tce = TCE_PCI_READ; |
322 | if (direction != DMA_TO_DEVICE) |
323 | proto_tce |= TCE_PCI_WRITE; |
324 | |
325 | while (npages--) { |
326 | - tce = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT; |
327 | - rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, tce); |
328 | + tce = proto_tce | (rpn & TCE_RPN_MASK) << tceshift; |
329 | + rc = plpar_tce_put((u64)liobn, (u64)tcenum << tceshift, tce); |
330 | |
331 | if (unlikely(rc == H_NOT_ENOUGH_RESOURCES)) { |
332 | ret = (int)rc; |
333 | - tce_free_pSeriesLP(tbl, tcenum_start, |
334 | + tce_free_pSeriesLP(liobn, tcenum_start, |
335 | (npages_start - (npages + 1))); |
336 | break; |
337 | } |
338 | |
339 | if (rc && printk_ratelimit()) { |
340 | printk("tce_build_pSeriesLP: plpar_tce_put failed. rc=%lld\n", rc); |
341 | - printk("\tindex = 0x%llx\n", (u64)tbl->it_index); |
342 | + printk("\tindex = 0x%llx\n", (u64)liobn); |
343 | printk("\ttcenum = 0x%llx\n", (u64)tcenum); |
344 | printk("\ttce val = 0x%llx\n", tce ); |
345 | dump_stack(); |
346 | @@ -228,7 +228,8 @@ static int tce_buildmulti_pSeriesLP(struct iommu_table *tbl, long tcenum, |
347 | unsigned long flags; |
348 | |
349 | if ((npages == 1) || !firmware_has_feature(FW_FEATURE_MULTITCE)) { |
350 | - return tce_build_pSeriesLP(tbl, tcenum, npages, uaddr, |
351 | + return tce_build_pSeriesLP(tbl->it_index, tcenum, |
352 | + tbl->it_page_shift, npages, uaddr, |
353 | direction, attrs); |
354 | } |
355 | |
356 | @@ -244,8 +245,9 @@ static int tce_buildmulti_pSeriesLP(struct iommu_table *tbl, long tcenum, |
357 | /* If allocation fails, fall back to the loop implementation */ |
358 | if (!tcep) { |
359 | local_irq_restore(flags); |
360 | - return tce_build_pSeriesLP(tbl, tcenum, npages, uaddr, |
361 | - direction, attrs); |
362 | + return tce_build_pSeriesLP(tbl->it_index, tcenum, |
363 | + tbl->it_page_shift, |
364 | + npages, uaddr, direction, attrs); |
365 | } |
366 | __this_cpu_write(tce_page, tcep); |
367 | } |
368 | @@ -296,16 +298,16 @@ static int tce_buildmulti_pSeriesLP(struct iommu_table *tbl, long tcenum, |
369 | return ret; |
370 | } |
371 | |
372 | -static void tce_free_pSeriesLP(struct iommu_table *tbl, long tcenum, long npages) |
373 | +static void tce_free_pSeriesLP(unsigned long liobn, long tcenum, long npages) |
374 | { |
375 | u64 rc; |
376 | |
377 | while (npages--) { |
378 | - rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, 0); |
379 | + rc = plpar_tce_put((u64)liobn, (u64)tcenum << 12, 0); |
380 | |
381 | if (rc && printk_ratelimit()) { |
382 | printk("tce_free_pSeriesLP: plpar_tce_put failed. rc=%lld\n", rc); |
383 | - printk("\tindex = 0x%llx\n", (u64)tbl->it_index); |
384 | + printk("\tindex = 0x%llx\n", (u64)liobn); |
385 | printk("\ttcenum = 0x%llx\n", (u64)tcenum); |
386 | dump_stack(); |
387 | } |
388 | @@ -320,7 +322,7 @@ static void tce_freemulti_pSeriesLP(struct iommu_table *tbl, long tcenum, long n |
389 | u64 rc; |
390 | |
391 | if (!firmware_has_feature(FW_FEATURE_MULTITCE)) |
392 | - return tce_free_pSeriesLP(tbl, tcenum, npages); |
393 | + return tce_free_pSeriesLP(tbl->it_index, tcenum, npages); |
394 | |
395 | rc = plpar_tce_stuff((u64)tbl->it_index, (u64)tcenum << 12, 0, npages); |
396 | |
397 | @@ -435,6 +437,19 @@ static int tce_setrange_multi_pSeriesLP(unsigned long start_pfn, |
398 | u64 rc = 0; |
399 | long l, limit; |
400 | |
401 | + if (!firmware_has_feature(FW_FEATURE_MULTITCE)) { |
402 | + unsigned long tceshift = be32_to_cpu(maprange->tce_shift); |
403 | + unsigned long dmastart = (start_pfn << PAGE_SHIFT) + |
404 | + be64_to_cpu(maprange->dma_base); |
405 | + unsigned long tcenum = dmastart >> tceshift; |
406 | + unsigned long npages = num_pfn << PAGE_SHIFT >> tceshift; |
407 | + void *uaddr = __va(start_pfn << PAGE_SHIFT); |
408 | + |
409 | + return tce_build_pSeriesLP(be32_to_cpu(maprange->liobn), |
410 | + tcenum, tceshift, npages, (unsigned long) uaddr, |
411 | + DMA_BIDIRECTIONAL, 0); |
412 | + } |
413 | + |
414 | local_irq_disable(); /* to protect tcep and the page behind it */ |
415 | tcep = __this_cpu_read(tce_page); |
416 | |
417 | diff --git a/arch/sparc/include/uapi/asm/ipcbuf.h b/arch/sparc/include/uapi/asm/ipcbuf.h |
418 | index 66013b4fe10d..58da9c4addb2 100644 |
419 | --- a/arch/sparc/include/uapi/asm/ipcbuf.h |
420 | +++ b/arch/sparc/include/uapi/asm/ipcbuf.h |
421 | @@ -14,19 +14,19 @@ |
422 | |
423 | struct ipc64_perm |
424 | { |
425 | - __kernel_key_t key; |
426 | - __kernel_uid_t uid; |
427 | - __kernel_gid_t gid; |
428 | - __kernel_uid_t cuid; |
429 | - __kernel_gid_t cgid; |
430 | + __kernel_key_t key; |
431 | + __kernel_uid32_t uid; |
432 | + __kernel_gid32_t gid; |
433 | + __kernel_uid32_t cuid; |
434 | + __kernel_gid32_t cgid; |
435 | #ifndef __arch64__ |
436 | - unsigned short __pad0; |
437 | + unsigned short __pad0; |
438 | #endif |
439 | - __kernel_mode_t mode; |
440 | - unsigned short __pad1; |
441 | - unsigned short seq; |
442 | - unsigned long long __unused1; |
443 | - unsigned long long __unused2; |
444 | + __kernel_mode_t mode; |
445 | + unsigned short __pad1; |
446 | + unsigned short seq; |
447 | + unsigned long long __unused1; |
448 | + unsigned long long __unused2; |
449 | }; |
450 | |
451 | #endif /* __SPARC_IPCBUF_H */ |
452 | diff --git a/arch/x86/kernel/cpu/tsx.c b/arch/x86/kernel/cpu/tsx.c |
453 | index 3e20d322bc98..032509adf9de 100644 |
454 | --- a/arch/x86/kernel/cpu/tsx.c |
455 | +++ b/arch/x86/kernel/cpu/tsx.c |
456 | @@ -115,11 +115,12 @@ void __init tsx_init(void) |
457 | tsx_disable(); |
458 | |
459 | /* |
460 | - * tsx_disable() will change the state of the |
461 | - * RTM CPUID bit. Clear it here since it is now |
462 | - * expected to be not set. |
463 | + * tsx_disable() will change the state of the RTM and HLE CPUID |
464 | + * bits. Clear them here since they are now expected to be not |
465 | + * set. |
466 | */ |
467 | setup_clear_cpu_cap(X86_FEATURE_RTM); |
468 | + setup_clear_cpu_cap(X86_FEATURE_HLE); |
469 | } else if (tsx_ctrl_state == TSX_CTRL_ENABLE) { |
470 | |
471 | /* |
472 | @@ -131,10 +132,10 @@ void __init tsx_init(void) |
473 | tsx_enable(); |
474 | |
475 | /* |
476 | - * tsx_enable() will change the state of the |
477 | - * RTM CPUID bit. Force it here since it is now |
478 | - * expected to be set. |
479 | + * tsx_enable() will change the state of the RTM and HLE CPUID |
480 | + * bits. Force them here since they are now expected to be set. |
481 | */ |
482 | setup_force_cpu_cap(X86_FEATURE_RTM); |
483 | + setup_force_cpu_cap(X86_FEATURE_HLE); |
484 | } |
485 | } |
486 | diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c |
487 | index 660c35f854f8..c456a9dbade8 100644 |
488 | --- a/arch/x86/kvm/emulate.c |
489 | +++ b/arch/x86/kvm/emulate.c |
490 | @@ -21,6 +21,7 @@ |
491 | */ |
492 | |
493 | #include <linux/kvm_host.h> |
494 | +#include <linux/nospec.h> |
495 | #include "kvm_cache_regs.h" |
496 | #include <asm/kvm_emulate.h> |
497 | #include <linux/stringify.h> |
498 | @@ -5053,16 +5054,28 @@ int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len) |
499 | ctxt->ad_bytes = def_ad_bytes ^ 6; |
500 | break; |
501 | case 0x26: /* ES override */ |
502 | + has_seg_override = true; |
503 | + ctxt->seg_override = VCPU_SREG_ES; |
504 | + break; |
505 | case 0x2e: /* CS override */ |
506 | + has_seg_override = true; |
507 | + ctxt->seg_override = VCPU_SREG_CS; |
508 | + break; |
509 | case 0x36: /* SS override */ |
510 | + has_seg_override = true; |
511 | + ctxt->seg_override = VCPU_SREG_SS; |
512 | + break; |
513 | case 0x3e: /* DS override */ |
514 | has_seg_override = true; |
515 | - ctxt->seg_override = (ctxt->b >> 3) & 3; |
516 | + ctxt->seg_override = VCPU_SREG_DS; |
517 | break; |
518 | case 0x64: /* FS override */ |
519 | + has_seg_override = true; |
520 | + ctxt->seg_override = VCPU_SREG_FS; |
521 | + break; |
522 | case 0x65: /* GS override */ |
523 | has_seg_override = true; |
524 | - ctxt->seg_override = ctxt->b & 7; |
525 | + ctxt->seg_override = VCPU_SREG_GS; |
526 | break; |
527 | case 0x40 ... 0x4f: /* REX */ |
528 | if (mode != X86EMUL_MODE_PROT64) |
529 | @@ -5146,10 +5159,15 @@ done_prefixes: |
530 | } |
531 | break; |
532 | case Escape: |
533 | - if (ctxt->modrm > 0xbf) |
534 | - opcode = opcode.u.esc->high[ctxt->modrm - 0xc0]; |
535 | - else |
536 | + if (ctxt->modrm > 0xbf) { |
537 | + size_t size = ARRAY_SIZE(opcode.u.esc->high); |
538 | + u32 index = array_index_nospec( |
539 | + ctxt->modrm - 0xc0, size); |
540 | + |
541 | + opcode = opcode.u.esc->high[index]; |
542 | + } else { |
543 | opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7]; |
544 | + } |
545 | break; |
546 | case InstrDual: |
547 | if ((ctxt->modrm >> 6) == 3) |
548 | diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c |
549 | index 42b1c83741c8..5e837c96e93f 100644 |
550 | --- a/arch/x86/kvm/hyperv.c |
551 | +++ b/arch/x86/kvm/hyperv.c |
552 | @@ -28,6 +28,7 @@ |
553 | |
554 | #include <linux/kvm_host.h> |
555 | #include <linux/highmem.h> |
556 | +#include <linux/nospec.h> |
557 | #include <asm/apicdef.h> |
558 | #include <trace/events/kvm.h> |
559 | |
560 | @@ -719,11 +720,12 @@ static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu, |
561 | u32 index, u64 *pdata) |
562 | { |
563 | struct kvm_hv *hv = &vcpu->kvm->arch.hyperv; |
564 | + size_t size = ARRAY_SIZE(hv->hv_crash_param); |
565 | |
566 | - if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param))) |
567 | + if (WARN_ON_ONCE(index >= size)) |
568 | return -EINVAL; |
569 | |
570 | - *pdata = hv->hv_crash_param[index]; |
571 | + *pdata = hv->hv_crash_param[array_index_nospec(index, size)]; |
572 | return 0; |
573 | } |
574 | |
575 | @@ -762,11 +764,12 @@ static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu, |
576 | u32 index, u64 data) |
577 | { |
578 | struct kvm_hv *hv = &vcpu->kvm->arch.hyperv; |
579 | + size_t size = ARRAY_SIZE(hv->hv_crash_param); |
580 | |
581 | - if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param))) |
582 | + if (WARN_ON_ONCE(index >= size)) |
583 | return -EINVAL; |
584 | |
585 | - hv->hv_crash_param[index] = data; |
586 | + hv->hv_crash_param[array_index_nospec(index, size)] = data; |
587 | return 0; |
588 | } |
589 | |
590 | diff --git a/arch/x86/kvm/ioapic.c b/arch/x86/kvm/ioapic.c |
591 | index 5f810bb80802..aa34b16e62c2 100644 |
592 | --- a/arch/x86/kvm/ioapic.c |
593 | +++ b/arch/x86/kvm/ioapic.c |
594 | @@ -36,6 +36,7 @@ |
595 | #include <linux/io.h> |
596 | #include <linux/slab.h> |
597 | #include <linux/export.h> |
598 | +#include <linux/nospec.h> |
599 | #include <asm/processor.h> |
600 | #include <asm/page.h> |
601 | #include <asm/current.h> |
602 | @@ -73,13 +74,14 @@ static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic, |
603 | default: |
604 | { |
605 | u32 redir_index = (ioapic->ioregsel - 0x10) >> 1; |
606 | - u64 redir_content; |
607 | + u64 redir_content = ~0ULL; |
608 | |
609 | - if (redir_index < IOAPIC_NUM_PINS) |
610 | - redir_content = |
611 | - ioapic->redirtbl[redir_index].bits; |
612 | - else |
613 | - redir_content = ~0ULL; |
614 | + if (redir_index < IOAPIC_NUM_PINS) { |
615 | + u32 index = array_index_nospec( |
616 | + redir_index, IOAPIC_NUM_PINS); |
617 | + |
618 | + redir_content = ioapic->redirtbl[index].bits; |
619 | + } |
620 | |
621 | result = (ioapic->ioregsel & 0x1) ? |
622 | (redir_content >> 32) & 0xffffffff : |
623 | @@ -299,6 +301,7 @@ static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val) |
624 | ioapic_debug("change redir index %x val %x\n", index, val); |
625 | if (index >= IOAPIC_NUM_PINS) |
626 | return; |
627 | + index = array_index_nospec(index, IOAPIC_NUM_PINS); |
628 | e = &ioapic->redirtbl[index]; |
629 | mask_before = e->fields.mask; |
630 | /* Preserve read-only fields */ |
631 | diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c |
632 | index cf32533225bb..caa17f8d4221 100644 |
633 | --- a/arch/x86/kvm/lapic.c |
634 | +++ b/arch/x86/kvm/lapic.c |
635 | @@ -28,6 +28,7 @@ |
636 | #include <linux/export.h> |
637 | #include <linux/math64.h> |
638 | #include <linux/slab.h> |
639 | +#include <linux/nospec.h> |
640 | #include <asm/processor.h> |
641 | #include <asm/msr.h> |
642 | #include <asm/page.h> |
643 | @@ -1587,15 +1588,20 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val) |
644 | case APIC_LVTTHMR: |
645 | case APIC_LVTPC: |
646 | case APIC_LVT1: |
647 | - case APIC_LVTERR: |
648 | + case APIC_LVTERR: { |
649 | /* TODO: Check vector */ |
650 | + size_t size; |
651 | + u32 index; |
652 | + |
653 | if (!kvm_apic_sw_enabled(apic)) |
654 | val |= APIC_LVT_MASKED; |
655 | - |
656 | - val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4]; |
657 | + size = ARRAY_SIZE(apic_lvt_mask); |
658 | + index = array_index_nospec( |
659 | + (reg - APIC_LVTT) >> 4, size); |
660 | + val &= apic_lvt_mask[index]; |
661 | kvm_lapic_set_reg(apic, reg, val); |
662 | - |
663 | break; |
664 | + } |
665 | |
666 | case APIC_LVTT: |
667 | if (!kvm_apic_sw_enabled(apic)) |
668 | diff --git a/arch/x86/kvm/mtrr.c b/arch/x86/kvm/mtrr.c |
669 | index 0149ac59c273..3e3016411020 100644 |
670 | --- a/arch/x86/kvm/mtrr.c |
671 | +++ b/arch/x86/kvm/mtrr.c |
672 | @@ -17,6 +17,7 @@ |
673 | */ |
674 | |
675 | #include <linux/kvm_host.h> |
676 | +#include <linux/nospec.h> |
677 | #include <asm/mtrr.h> |
678 | |
679 | #include "cpuid.h" |
680 | @@ -202,11 +203,15 @@ static bool fixed_msr_to_seg_unit(u32 msr, int *seg, int *unit) |
681 | break; |
682 | case MSR_MTRRfix16K_80000 ... MSR_MTRRfix16K_A0000: |
683 | *seg = 1; |
684 | - *unit = msr - MSR_MTRRfix16K_80000; |
685 | + *unit = array_index_nospec( |
686 | + msr - MSR_MTRRfix16K_80000, |
687 | + MSR_MTRRfix16K_A0000 - MSR_MTRRfix16K_80000 + 1); |
688 | break; |
689 | case MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000: |
690 | *seg = 2; |
691 | - *unit = msr - MSR_MTRRfix4K_C0000; |
692 | + *unit = array_index_nospec( |
693 | + msr - MSR_MTRRfix4K_C0000, |
694 | + MSR_MTRRfix4K_F8000 - MSR_MTRRfix4K_C0000 + 1); |
695 | break; |
696 | default: |
697 | return false; |
698 | diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h |
699 | index f96e1f962587..fbf3d25af765 100644 |
700 | --- a/arch/x86/kvm/pmu.h |
701 | +++ b/arch/x86/kvm/pmu.h |
702 | @@ -1,6 +1,8 @@ |
703 | #ifndef __KVM_X86_PMU_H |
704 | #define __KVM_X86_PMU_H |
705 | |
706 | +#include <linux/nospec.h> |
707 | + |
708 | #define vcpu_to_pmu(vcpu) (&(vcpu)->arch.pmu) |
709 | #define pmu_to_vcpu(pmu) (container_of((pmu), struct kvm_vcpu, arch.pmu)) |
710 | #define pmc_to_pmu(pmc) (&(pmc)->vcpu->arch.pmu) |
711 | @@ -80,8 +82,12 @@ static inline bool pmc_is_enabled(struct kvm_pmc *pmc) |
712 | static inline struct kvm_pmc *get_gp_pmc(struct kvm_pmu *pmu, u32 msr, |
713 | u32 base) |
714 | { |
715 | - if (msr >= base && msr < base + pmu->nr_arch_gp_counters) |
716 | - return &pmu->gp_counters[msr - base]; |
717 | + if (msr >= base && msr < base + pmu->nr_arch_gp_counters) { |
718 | + u32 index = array_index_nospec(msr - base, |
719 | + pmu->nr_arch_gp_counters); |
720 | + |
721 | + return &pmu->gp_counters[index]; |
722 | + } |
723 | |
724 | return NULL; |
725 | } |
726 | @@ -91,8 +97,12 @@ static inline struct kvm_pmc *get_fixed_pmc(struct kvm_pmu *pmu, u32 msr) |
727 | { |
728 | int base = MSR_CORE_PERF_FIXED_CTR0; |
729 | |
730 | - if (msr >= base && msr < base + pmu->nr_arch_fixed_counters) |
731 | - return &pmu->fixed_counters[msr - base]; |
732 | + if (msr >= base && msr < base + pmu->nr_arch_fixed_counters) { |
733 | + u32 index = array_index_nospec(msr - base, |
734 | + pmu->nr_arch_fixed_counters); |
735 | + |
736 | + return &pmu->fixed_counters[index]; |
737 | + } |
738 | |
739 | return NULL; |
740 | } |
741 | diff --git a/arch/x86/kvm/pmu_intel.c b/arch/x86/kvm/pmu_intel.c |
742 | index 2729131fe9bf..84ae4dd261ca 100644 |
743 | --- a/arch/x86/kvm/pmu_intel.c |
744 | +++ b/arch/x86/kvm/pmu_intel.c |
745 | @@ -87,10 +87,14 @@ static unsigned intel_find_arch_event(struct kvm_pmu *pmu, |
746 | |
747 | static unsigned intel_find_fixed_event(int idx) |
748 | { |
749 | - if (idx >= ARRAY_SIZE(fixed_pmc_events)) |
750 | + u32 event; |
751 | + size_t size = ARRAY_SIZE(fixed_pmc_events); |
752 | + |
753 | + if (idx >= size) |
754 | return PERF_COUNT_HW_MAX; |
755 | |
756 | - return intel_arch_events[fixed_pmc_events[idx]].event_type; |
757 | + event = fixed_pmc_events[array_index_nospec(idx, size)]; |
758 | + return intel_arch_events[event].event_type; |
759 | } |
760 | |
761 | /* check if a PMC is enabled by comparing it with globl_ctrl bits. */ |
762 | @@ -131,15 +135,19 @@ static struct kvm_pmc *intel_msr_idx_to_pmc(struct kvm_vcpu *vcpu, |
763 | struct kvm_pmu *pmu = vcpu_to_pmu(vcpu); |
764 | bool fixed = idx & (1u << 30); |
765 | struct kvm_pmc *counters; |
766 | + unsigned int num_counters; |
767 | |
768 | idx &= ~(3u << 30); |
769 | - if (!fixed && idx >= pmu->nr_arch_gp_counters) |
770 | - return NULL; |
771 | - if (fixed && idx >= pmu->nr_arch_fixed_counters) |
772 | + if (fixed) { |
773 | + counters = pmu->fixed_counters; |
774 | + num_counters = pmu->nr_arch_fixed_counters; |
775 | + } else { |
776 | + counters = pmu->gp_counters; |
777 | + num_counters = pmu->nr_arch_gp_counters; |
778 | + } |
779 | + if (idx >= num_counters) |
780 | return NULL; |
781 | - counters = fixed ? pmu->fixed_counters : pmu->gp_counters; |
782 | - |
783 | - return &counters[idx]; |
784 | + return &counters[array_index_nospec(idx, num_counters)]; |
785 | } |
786 | |
787 | static bool intel_is_valid_msr(struct kvm_vcpu *vcpu, u32 msr) |
788 | diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c |
789 | index f76caa03f4f8..67cdb08a736f 100644 |
790 | --- a/arch/x86/kvm/vmx.c |
791 | +++ b/arch/x86/kvm/vmx.c |
792 | @@ -7653,8 +7653,10 @@ static int handle_vmread(struct kvm_vcpu *vcpu) |
793 | /* _system ok, as nested_vmx_check_permission verified cpl=0 */ |
794 | if (kvm_write_guest_virt_system(vcpu, gva, &field_value, |
795 | (is_long_mode(vcpu) ? 8 : 4), |
796 | - &e)) |
797 | + &e)) { |
798 | kvm_inject_page_fault(vcpu, &e); |
799 | + return 1; |
800 | + } |
801 | } |
802 | |
803 | nested_vmx_succeed(vcpu); |
804 | diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c |
805 | new file mode 100644 |
806 | index 000000000000..3791ce8d269e |
807 | --- /dev/null |
808 | +++ b/arch/x86/kvm/vmx/vmx.c |
809 | @@ -0,0 +1,8033 @@ |
810 | +// SPDX-License-Identifier: GPL-2.0-only |
811 | +/* |
812 | + * Kernel-based Virtual Machine driver for Linux |
813 | + * |
814 | + * This module enables machines with Intel VT-x extensions to run virtual |
815 | + * machines without emulation or binary translation. |
816 | + * |
817 | + * Copyright (C) 2006 Qumranet, Inc. |
818 | + * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
819 | + * |
820 | + * Authors: |
821 | + * Avi Kivity <avi@qumranet.com> |
822 | + * Yaniv Kamay <yaniv@qumranet.com> |
823 | + */ |
824 | + |
825 | +#include <linux/frame.h> |
826 | +#include <linux/highmem.h> |
827 | +#include <linux/hrtimer.h> |
828 | +#include <linux/kernel.h> |
829 | +#include <linux/kvm_host.h> |
830 | +#include <linux/module.h> |
831 | +#include <linux/moduleparam.h> |
832 | +#include <linux/mod_devicetable.h> |
833 | +#include <linux/mm.h> |
834 | +#include <linux/sched.h> |
835 | +#include <linux/sched/smt.h> |
836 | +#include <linux/slab.h> |
837 | +#include <linux/tboot.h> |
838 | +#include <linux/trace_events.h> |
839 | + |
840 | +#include <asm/apic.h> |
841 | +#include <asm/asm.h> |
842 | +#include <asm/cpu.h> |
843 | +#include <asm/debugreg.h> |
844 | +#include <asm/desc.h> |
845 | +#include <asm/fpu/internal.h> |
846 | +#include <asm/io.h> |
847 | +#include <asm/irq_remapping.h> |
848 | +#include <asm/kexec.h> |
849 | +#include <asm/perf_event.h> |
850 | +#include <asm/mce.h> |
851 | +#include <asm/mmu_context.h> |
852 | +#include <asm/mshyperv.h> |
853 | +#include <asm/spec-ctrl.h> |
854 | +#include <asm/virtext.h> |
855 | +#include <asm/vmx.h> |
856 | + |
857 | +#include "capabilities.h" |
858 | +#include "cpuid.h" |
859 | +#include "evmcs.h" |
860 | +#include "irq.h" |
861 | +#include "kvm_cache_regs.h" |
862 | +#include "lapic.h" |
863 | +#include "mmu.h" |
864 | +#include "nested.h" |
865 | +#include "ops.h" |
866 | +#include "pmu.h" |
867 | +#include "trace.h" |
868 | +#include "vmcs.h" |
869 | +#include "vmcs12.h" |
870 | +#include "vmx.h" |
871 | +#include "x86.h" |
872 | + |
873 | +MODULE_AUTHOR("Qumranet"); |
874 | +MODULE_LICENSE("GPL"); |
875 | + |
876 | +static const struct x86_cpu_id vmx_cpu_id[] = { |
877 | + X86_FEATURE_MATCH(X86_FEATURE_VMX), |
878 | + {} |
879 | +}; |
880 | +MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id); |
881 | + |
882 | +bool __read_mostly enable_vpid = 1; |
883 | +module_param_named(vpid, enable_vpid, bool, 0444); |
884 | + |
885 | +static bool __read_mostly enable_vnmi = 1; |
886 | +module_param_named(vnmi, enable_vnmi, bool, S_IRUGO); |
887 | + |
888 | +bool __read_mostly flexpriority_enabled = 1; |
889 | +module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO); |
890 | + |
891 | +bool __read_mostly enable_ept = 1; |
892 | +module_param_named(ept, enable_ept, bool, S_IRUGO); |
893 | + |
894 | +bool __read_mostly enable_unrestricted_guest = 1; |
895 | +module_param_named(unrestricted_guest, |
896 | + enable_unrestricted_guest, bool, S_IRUGO); |
897 | + |
898 | +bool __read_mostly enable_ept_ad_bits = 1; |
899 | +module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO); |
900 | + |
901 | +static bool __read_mostly emulate_invalid_guest_state = true; |
902 | +module_param(emulate_invalid_guest_state, bool, S_IRUGO); |
903 | + |
904 | +static bool __read_mostly fasteoi = 1; |
905 | +module_param(fasteoi, bool, S_IRUGO); |
906 | + |
907 | +static bool __read_mostly enable_apicv = 1; |
908 | +module_param(enable_apicv, bool, S_IRUGO); |
909 | + |
910 | +/* |
911 | + * If nested=1, nested virtualization is supported, i.e., guests may use |
912 | + * VMX and be a hypervisor for its own guests. If nested=0, guests may not |
913 | + * use VMX instructions. |
914 | + */ |
915 | +static bool __read_mostly nested = 1; |
916 | +module_param(nested, bool, S_IRUGO); |
917 | + |
918 | +bool __read_mostly enable_pml = 1; |
919 | +module_param_named(pml, enable_pml, bool, S_IRUGO); |
920 | + |
921 | +static bool __read_mostly dump_invalid_vmcs = 0; |
922 | +module_param(dump_invalid_vmcs, bool, 0644); |
923 | + |
924 | +#define MSR_BITMAP_MODE_X2APIC 1 |
925 | +#define MSR_BITMAP_MODE_X2APIC_APICV 2 |
926 | + |
927 | +#define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL |
928 | + |
929 | +/* Guest_tsc -> host_tsc conversion requires 64-bit division. */ |
930 | +static int __read_mostly cpu_preemption_timer_multi; |
931 | +static bool __read_mostly enable_preemption_timer = 1; |
932 | +#ifdef CONFIG_X86_64 |
933 | +module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO); |
934 | +#endif |
935 | + |
936 | +#define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD) |
937 | +#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE |
938 | +#define KVM_VM_CR0_ALWAYS_ON \ |
939 | + (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | \ |
940 | + X86_CR0_WP | X86_CR0_PG | X86_CR0_PE) |
941 | +#define KVM_CR4_GUEST_OWNED_BITS \ |
942 | + (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \ |
943 | + | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD) |
944 | + |
945 | +#define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE |
946 | +#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE) |
947 | +#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE) |
948 | + |
949 | +#define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM)) |
950 | + |
951 | +#define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \ |
952 | + RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \ |
953 | + RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \ |
954 | + RTIT_STATUS_BYTECNT)) |
955 | + |
956 | +#define MSR_IA32_RTIT_OUTPUT_BASE_MASK \ |
957 | + (~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f) |
958 | + |
959 | +/* |
960 | + * These 2 parameters are used to config the controls for Pause-Loop Exiting: |
961 | + * ple_gap: upper bound on the amount of time between two successive |
962 | + * executions of PAUSE in a loop. Also indicate if ple enabled. |
963 | + * According to test, this time is usually smaller than 128 cycles. |
964 | + * ple_window: upper bound on the amount of time a guest is allowed to execute |
965 | + * in a PAUSE loop. Tests indicate that most spinlocks are held for |
966 | + * less than 2^12 cycles |
967 | + * Time is measured based on a counter that runs at the same rate as the TSC, |
968 | + * refer SDM volume 3b section 21.6.13 & 22.1.3. |
969 | + */ |
970 | +static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP; |
971 | +module_param(ple_gap, uint, 0444); |
972 | + |
973 | +static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW; |
974 | +module_param(ple_window, uint, 0444); |
975 | + |
976 | +/* Default doubles per-vcpu window every exit. */ |
977 | +static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW; |
978 | +module_param(ple_window_grow, uint, 0444); |
979 | + |
980 | +/* Default resets per-vcpu window every exit to ple_window. */ |
981 | +static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK; |
982 | +module_param(ple_window_shrink, uint, 0444); |
983 | + |
984 | +/* Default is to compute the maximum so we can never overflow. */ |
985 | +static unsigned int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX; |
986 | +module_param(ple_window_max, uint, 0444); |
987 | + |
988 | +/* Default is SYSTEM mode, 1 for host-guest mode */ |
989 | +int __read_mostly pt_mode = PT_MODE_SYSTEM; |
990 | +module_param(pt_mode, int, S_IRUGO); |
991 | + |
992 | +static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush); |
993 | +static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond); |
994 | +static DEFINE_MUTEX(vmx_l1d_flush_mutex); |
995 | + |
996 | +/* Storage for pre module init parameter parsing */ |
997 | +static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO; |
998 | + |
999 | +static const struct { |
1000 | + const char *option; |
1001 | + bool for_parse; |
1002 | +} vmentry_l1d_param[] = { |
1003 | + [VMENTER_L1D_FLUSH_AUTO] = {"auto", true}, |
1004 | + [VMENTER_L1D_FLUSH_NEVER] = {"never", true}, |
1005 | + [VMENTER_L1D_FLUSH_COND] = {"cond", true}, |
1006 | + [VMENTER_L1D_FLUSH_ALWAYS] = {"always", true}, |
1007 | + [VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false}, |
1008 | + [VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false}, |
1009 | +}; |
1010 | + |
1011 | +#define L1D_CACHE_ORDER 4 |
1012 | +static void *vmx_l1d_flush_pages; |
1013 | + |
1014 | +static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf) |
1015 | +{ |
1016 | + struct page *page; |
1017 | + unsigned int i; |
1018 | + |
1019 | + if (!boot_cpu_has_bug(X86_BUG_L1TF)) { |
1020 | + l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED; |
1021 | + return 0; |
1022 | + } |
1023 | + |
1024 | + if (!enable_ept) { |
1025 | + l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED; |
1026 | + return 0; |
1027 | + } |
1028 | + |
1029 | + if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) { |
1030 | + u64 msr; |
1031 | + |
1032 | + rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr); |
1033 | + if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) { |
1034 | + l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED; |
1035 | + return 0; |
1036 | + } |
1037 | + } |
1038 | + |
1039 | + /* If set to auto use the default l1tf mitigation method */ |
1040 | + if (l1tf == VMENTER_L1D_FLUSH_AUTO) { |
1041 | + switch (l1tf_mitigation) { |
1042 | + case L1TF_MITIGATION_OFF: |
1043 | + l1tf = VMENTER_L1D_FLUSH_NEVER; |
1044 | + break; |
1045 | + case L1TF_MITIGATION_FLUSH_NOWARN: |
1046 | + case L1TF_MITIGATION_FLUSH: |
1047 | + case L1TF_MITIGATION_FLUSH_NOSMT: |
1048 | + l1tf = VMENTER_L1D_FLUSH_COND; |
1049 | + break; |
1050 | + case L1TF_MITIGATION_FULL: |
1051 | + case L1TF_MITIGATION_FULL_FORCE: |
1052 | + l1tf = VMENTER_L1D_FLUSH_ALWAYS; |
1053 | + break; |
1054 | + } |
1055 | + } else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) { |
1056 | + l1tf = VMENTER_L1D_FLUSH_ALWAYS; |
1057 | + } |
1058 | + |
1059 | + if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages && |
1060 | + !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) { |
1061 | + /* |
1062 | + * This allocation for vmx_l1d_flush_pages is not tied to a VM |
1063 | + * lifetime and so should not be charged to a memcg. |
1064 | + */ |
1065 | + page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER); |
1066 | + if (!page) |
1067 | + return -ENOMEM; |
1068 | + vmx_l1d_flush_pages = page_address(page); |
1069 | + |
1070 | + /* |
1071 | + * Initialize each page with a different pattern in |
1072 | + * order to protect against KSM in the nested |
1073 | + * virtualization case. |
1074 | + */ |
1075 | + for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) { |
1076 | + memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1, |
1077 | + PAGE_SIZE); |
1078 | + } |
1079 | + } |
1080 | + |
1081 | + l1tf_vmx_mitigation = l1tf; |
1082 | + |
1083 | + if (l1tf != VMENTER_L1D_FLUSH_NEVER) |
1084 | + static_branch_enable(&vmx_l1d_should_flush); |
1085 | + else |
1086 | + static_branch_disable(&vmx_l1d_should_flush); |
1087 | + |
1088 | + if (l1tf == VMENTER_L1D_FLUSH_COND) |
1089 | + static_branch_enable(&vmx_l1d_flush_cond); |
1090 | + else |
1091 | + static_branch_disable(&vmx_l1d_flush_cond); |
1092 | + return 0; |
1093 | +} |
1094 | + |
1095 | +static int vmentry_l1d_flush_parse(const char *s) |
1096 | +{ |
1097 | + unsigned int i; |
1098 | + |
1099 | + if (s) { |
1100 | + for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) { |
1101 | + if (vmentry_l1d_param[i].for_parse && |
1102 | + sysfs_streq(s, vmentry_l1d_param[i].option)) |
1103 | + return i; |
1104 | + } |
1105 | + } |
1106 | + return -EINVAL; |
1107 | +} |
1108 | + |
1109 | +static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp) |
1110 | +{ |
1111 | + int l1tf, ret; |
1112 | + |
1113 | + l1tf = vmentry_l1d_flush_parse(s); |
1114 | + if (l1tf < 0) |
1115 | + return l1tf; |
1116 | + |
1117 | + if (!boot_cpu_has(X86_BUG_L1TF)) |
1118 | + return 0; |
1119 | + |
1120 | + /* |
1121 | + * Has vmx_init() run already? If not then this is the pre init |
1122 | + * parameter parsing. In that case just store the value and let |
1123 | + * vmx_init() do the proper setup after enable_ept has been |
1124 | + * established. |
1125 | + */ |
1126 | + if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) { |
1127 | + vmentry_l1d_flush_param = l1tf; |
1128 | + return 0; |
1129 | + } |
1130 | + |
1131 | + mutex_lock(&vmx_l1d_flush_mutex); |
1132 | + ret = vmx_setup_l1d_flush(l1tf); |
1133 | + mutex_unlock(&vmx_l1d_flush_mutex); |
1134 | + return ret; |
1135 | +} |
1136 | + |
1137 | +static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp) |
1138 | +{ |
1139 | + if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param))) |
1140 | + return sprintf(s, "???\n"); |
1141 | + |
1142 | + return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option); |
1143 | +} |
1144 | + |
1145 | +static const struct kernel_param_ops vmentry_l1d_flush_ops = { |
1146 | + .set = vmentry_l1d_flush_set, |
1147 | + .get = vmentry_l1d_flush_get, |
1148 | +}; |
1149 | +module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644); |
1150 | + |
1151 | +static bool guest_state_valid(struct kvm_vcpu *vcpu); |
1152 | +static u32 vmx_segment_access_rights(struct kvm_segment *var); |
1153 | +static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, |
1154 | + u32 msr, int type); |
1155 | + |
1156 | +void vmx_vmexit(void); |
1157 | + |
1158 | +#define vmx_insn_failed(fmt...) \ |
1159 | +do { \ |
1160 | + WARN_ONCE(1, fmt); \ |
1161 | + pr_warn_ratelimited(fmt); \ |
1162 | +} while (0) |
1163 | + |
1164 | +asmlinkage void vmread_error(unsigned long field, bool fault) |
1165 | +{ |
1166 | + if (fault) |
1167 | + kvm_spurious_fault(); |
1168 | + else |
1169 | + vmx_insn_failed("kvm: vmread failed: field=%lx\n", field); |
1170 | +} |
1171 | + |
1172 | +noinline void vmwrite_error(unsigned long field, unsigned long value) |
1173 | +{ |
1174 | + vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n", |
1175 | + field, value, vmcs_read32(VM_INSTRUCTION_ERROR)); |
1176 | +} |
1177 | + |
1178 | +noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr) |
1179 | +{ |
1180 | + vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr); |
1181 | +} |
1182 | + |
1183 | +noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr) |
1184 | +{ |
1185 | + vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr); |
1186 | +} |
1187 | + |
1188 | +noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva) |
1189 | +{ |
1190 | + vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n", |
1191 | + ext, vpid, gva); |
1192 | +} |
1193 | + |
1194 | +noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa) |
1195 | +{ |
1196 | + vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n", |
1197 | + ext, eptp, gpa); |
1198 | +} |
1199 | + |
1200 | +static DEFINE_PER_CPU(struct vmcs *, vmxarea); |
1201 | +DEFINE_PER_CPU(struct vmcs *, current_vmcs); |
1202 | +/* |
1203 | + * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed |
1204 | + * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it. |
1205 | + */ |
1206 | +static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu); |
1207 | + |
1208 | +/* |
1209 | + * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we |
1210 | + * can find which vCPU should be waken up. |
1211 | + */ |
1212 | +static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu); |
1213 | +static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock); |
1214 | + |
1215 | +static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS); |
1216 | +static DEFINE_SPINLOCK(vmx_vpid_lock); |
1217 | + |
1218 | +struct vmcs_config vmcs_config; |
1219 | +struct vmx_capability vmx_capability; |
1220 | + |
1221 | +#define VMX_SEGMENT_FIELD(seg) \ |
1222 | + [VCPU_SREG_##seg] = { \ |
1223 | + .selector = GUEST_##seg##_SELECTOR, \ |
1224 | + .base = GUEST_##seg##_BASE, \ |
1225 | + .limit = GUEST_##seg##_LIMIT, \ |
1226 | + .ar_bytes = GUEST_##seg##_AR_BYTES, \ |
1227 | + } |
1228 | + |
1229 | +static const struct kvm_vmx_segment_field { |
1230 | + unsigned selector; |
1231 | + unsigned base; |
1232 | + unsigned limit; |
1233 | + unsigned ar_bytes; |
1234 | +} kvm_vmx_segment_fields[] = { |
1235 | + VMX_SEGMENT_FIELD(CS), |
1236 | + VMX_SEGMENT_FIELD(DS), |
1237 | + VMX_SEGMENT_FIELD(ES), |
1238 | + VMX_SEGMENT_FIELD(FS), |
1239 | + VMX_SEGMENT_FIELD(GS), |
1240 | + VMX_SEGMENT_FIELD(SS), |
1241 | + VMX_SEGMENT_FIELD(TR), |
1242 | + VMX_SEGMENT_FIELD(LDTR), |
1243 | +}; |
1244 | + |
1245 | +u64 host_efer; |
1246 | +static unsigned long host_idt_base; |
1247 | + |
1248 | +/* |
1249 | + * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm |
1250 | + * will emulate SYSCALL in legacy mode if the vendor string in guest |
1251 | + * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To |
1252 | + * support this emulation, IA32_STAR must always be included in |
1253 | + * vmx_msr_index[], even in i386 builds. |
1254 | + */ |
1255 | +const u32 vmx_msr_index[] = { |
1256 | +#ifdef CONFIG_X86_64 |
1257 | + MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, |
1258 | +#endif |
1259 | + MSR_EFER, MSR_TSC_AUX, MSR_STAR, |
1260 | + MSR_IA32_TSX_CTRL, |
1261 | +}; |
1262 | + |
1263 | +#if IS_ENABLED(CONFIG_HYPERV) |
1264 | +static bool __read_mostly enlightened_vmcs = true; |
1265 | +module_param(enlightened_vmcs, bool, 0444); |
1266 | + |
1267 | +/* check_ept_pointer() should be under protection of ept_pointer_lock. */ |
1268 | +static void check_ept_pointer_match(struct kvm *kvm) |
1269 | +{ |
1270 | + struct kvm_vcpu *vcpu; |
1271 | + u64 tmp_eptp = INVALID_PAGE; |
1272 | + int i; |
1273 | + |
1274 | + kvm_for_each_vcpu(i, vcpu, kvm) { |
1275 | + if (!VALID_PAGE(tmp_eptp)) { |
1276 | + tmp_eptp = to_vmx(vcpu)->ept_pointer; |
1277 | + } else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) { |
1278 | + to_kvm_vmx(kvm)->ept_pointers_match |
1279 | + = EPT_POINTERS_MISMATCH; |
1280 | + return; |
1281 | + } |
1282 | + } |
1283 | + |
1284 | + to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH; |
1285 | +} |
1286 | + |
1287 | +static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush, |
1288 | + void *data) |
1289 | +{ |
1290 | + struct kvm_tlb_range *range = data; |
1291 | + |
1292 | + return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn, |
1293 | + range->pages); |
1294 | +} |
1295 | + |
1296 | +static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm, |
1297 | + struct kvm_vcpu *vcpu, struct kvm_tlb_range *range) |
1298 | +{ |
1299 | + u64 ept_pointer = to_vmx(vcpu)->ept_pointer; |
1300 | + |
1301 | + /* |
1302 | + * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address |
1303 | + * of the base of EPT PML4 table, strip off EPT configuration |
1304 | + * information. |
1305 | + */ |
1306 | + if (range) |
1307 | + return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK, |
1308 | + kvm_fill_hv_flush_list_func, (void *)range); |
1309 | + else |
1310 | + return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK); |
1311 | +} |
1312 | + |
1313 | +static int hv_remote_flush_tlb_with_range(struct kvm *kvm, |
1314 | + struct kvm_tlb_range *range) |
1315 | +{ |
1316 | + struct kvm_vcpu *vcpu; |
1317 | + int ret = 0, i; |
1318 | + |
1319 | + spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock); |
1320 | + |
1321 | + if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK) |
1322 | + check_ept_pointer_match(kvm); |
1323 | + |
1324 | + if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) { |
1325 | + kvm_for_each_vcpu(i, vcpu, kvm) { |
1326 | + /* If ept_pointer is invalid pointer, bypass flush request. */ |
1327 | + if (VALID_PAGE(to_vmx(vcpu)->ept_pointer)) |
1328 | + ret |= __hv_remote_flush_tlb_with_range( |
1329 | + kvm, vcpu, range); |
1330 | + } |
1331 | + } else { |
1332 | + ret = __hv_remote_flush_tlb_with_range(kvm, |
1333 | + kvm_get_vcpu(kvm, 0), range); |
1334 | + } |
1335 | + |
1336 | + spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock); |
1337 | + return ret; |
1338 | +} |
1339 | +static int hv_remote_flush_tlb(struct kvm *kvm) |
1340 | +{ |
1341 | + return hv_remote_flush_tlb_with_range(kvm, NULL); |
1342 | +} |
1343 | + |
1344 | +static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu) |
1345 | +{ |
1346 | + struct hv_enlightened_vmcs *evmcs; |
1347 | + struct hv_partition_assist_pg **p_hv_pa_pg = |
1348 | + &vcpu->kvm->arch.hyperv.hv_pa_pg; |
1349 | + /* |
1350 | + * Synthetic VM-Exit is not enabled in current code and so All |
1351 | + * evmcs in singe VM shares same assist page. |
1352 | + */ |
1353 | + if (!*p_hv_pa_pg) |
1354 | + *p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL); |
1355 | + |
1356 | + if (!*p_hv_pa_pg) |
1357 | + return -ENOMEM; |
1358 | + |
1359 | + evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs; |
1360 | + |
1361 | + evmcs->partition_assist_page = |
1362 | + __pa(*p_hv_pa_pg); |
1363 | + evmcs->hv_vm_id = (unsigned long)vcpu->kvm; |
1364 | + evmcs->hv_enlightenments_control.nested_flush_hypercall = 1; |
1365 | + |
1366 | + return 0; |
1367 | +} |
1368 | + |
1369 | +#endif /* IS_ENABLED(CONFIG_HYPERV) */ |
1370 | + |
1371 | +/* |
1372 | + * Comment's format: document - errata name - stepping - processor name. |
1373 | + * Refer from |
1374 | + * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp |
1375 | + */ |
1376 | +static u32 vmx_preemption_cpu_tfms[] = { |
1377 | +/* 323344.pdf - BA86 - D0 - Xeon 7500 Series */ |
1378 | +0x000206E6, |
1379 | +/* 323056.pdf - AAX65 - C2 - Xeon L3406 */ |
1380 | +/* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */ |
1381 | +/* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */ |
1382 | +0x00020652, |
1383 | +/* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */ |
1384 | +0x00020655, |
1385 | +/* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */ |
1386 | +/* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */ |
1387 | +/* |
1388 | + * 320767.pdf - AAP86 - B1 - |
1389 | + * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile |
1390 | + */ |
1391 | +0x000106E5, |
1392 | +/* 321333.pdf - AAM126 - C0 - Xeon 3500 */ |
1393 | +0x000106A0, |
1394 | +/* 321333.pdf - AAM126 - C1 - Xeon 3500 */ |
1395 | +0x000106A1, |
1396 | +/* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */ |
1397 | +0x000106A4, |
1398 | + /* 321333.pdf - AAM126 - D0 - Xeon 3500 */ |
1399 | + /* 321324.pdf - AAK139 - D0 - Xeon 5500 */ |
1400 | + /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */ |
1401 | +0x000106A5, |
1402 | + /* Xeon E3-1220 V2 */ |
1403 | +0x000306A8, |
1404 | +}; |
1405 | + |
1406 | +static inline bool cpu_has_broken_vmx_preemption_timer(void) |
1407 | +{ |
1408 | + u32 eax = cpuid_eax(0x00000001), i; |
1409 | + |
1410 | + /* Clear the reserved bits */ |
1411 | + eax &= ~(0x3U << 14 | 0xfU << 28); |
1412 | + for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++) |
1413 | + if (eax == vmx_preemption_cpu_tfms[i]) |
1414 | + return true; |
1415 | + |
1416 | + return false; |
1417 | +} |
1418 | + |
1419 | +static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu) |
1420 | +{ |
1421 | + return flexpriority_enabled && lapic_in_kernel(vcpu); |
1422 | +} |
1423 | + |
1424 | +static inline bool report_flexpriority(void) |
1425 | +{ |
1426 | + return flexpriority_enabled; |
1427 | +} |
1428 | + |
1429 | +static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr) |
1430 | +{ |
1431 | + int i; |
1432 | + |
1433 | + for (i = 0; i < vmx->nmsrs; ++i) |
1434 | + if (vmx_msr_index[vmx->guest_msrs[i].index] == msr) |
1435 | + return i; |
1436 | + return -1; |
1437 | +} |
1438 | + |
1439 | +struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr) |
1440 | +{ |
1441 | + int i; |
1442 | + |
1443 | + i = __find_msr_index(vmx, msr); |
1444 | + if (i >= 0) |
1445 | + return &vmx->guest_msrs[i]; |
1446 | + return NULL; |
1447 | +} |
1448 | + |
1449 | +static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data) |
1450 | +{ |
1451 | + int ret = 0; |
1452 | + |
1453 | + u64 old_msr_data = msr->data; |
1454 | + msr->data = data; |
1455 | + if (msr - vmx->guest_msrs < vmx->save_nmsrs) { |
1456 | + preempt_disable(); |
1457 | + ret = kvm_set_shared_msr(msr->index, msr->data, |
1458 | + msr->mask); |
1459 | + preempt_enable(); |
1460 | + if (ret) |
1461 | + msr->data = old_msr_data; |
1462 | + } |
1463 | + return ret; |
1464 | +} |
1465 | + |
1466 | +void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs) |
1467 | +{ |
1468 | + vmcs_clear(loaded_vmcs->vmcs); |
1469 | + if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched) |
1470 | + vmcs_clear(loaded_vmcs->shadow_vmcs); |
1471 | + loaded_vmcs->cpu = -1; |
1472 | + loaded_vmcs->launched = 0; |
1473 | +} |
1474 | + |
1475 | +#ifdef CONFIG_KEXEC_CORE |
1476 | +/* |
1477 | + * This bitmap is used to indicate whether the vmclear |
1478 | + * operation is enabled on all cpus. All disabled by |
1479 | + * default. |
1480 | + */ |
1481 | +static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE; |
1482 | + |
1483 | +static inline void crash_enable_local_vmclear(int cpu) |
1484 | +{ |
1485 | + cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap); |
1486 | +} |
1487 | + |
1488 | +static inline void crash_disable_local_vmclear(int cpu) |
1489 | +{ |
1490 | + cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap); |
1491 | +} |
1492 | + |
1493 | +static inline int crash_local_vmclear_enabled(int cpu) |
1494 | +{ |
1495 | + return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap); |
1496 | +} |
1497 | + |
1498 | +static void crash_vmclear_local_loaded_vmcss(void) |
1499 | +{ |
1500 | + int cpu = raw_smp_processor_id(); |
1501 | + struct loaded_vmcs *v; |
1502 | + |
1503 | + if (!crash_local_vmclear_enabled(cpu)) |
1504 | + return; |
1505 | + |
1506 | + list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu), |
1507 | + loaded_vmcss_on_cpu_link) |
1508 | + vmcs_clear(v->vmcs); |
1509 | +} |
1510 | +#else |
1511 | +static inline void crash_enable_local_vmclear(int cpu) { } |
1512 | +static inline void crash_disable_local_vmclear(int cpu) { } |
1513 | +#endif /* CONFIG_KEXEC_CORE */ |
1514 | + |
1515 | +static void __loaded_vmcs_clear(void *arg) |
1516 | +{ |
1517 | + struct loaded_vmcs *loaded_vmcs = arg; |
1518 | + int cpu = raw_smp_processor_id(); |
1519 | + |
1520 | + if (loaded_vmcs->cpu != cpu) |
1521 | + return; /* vcpu migration can race with cpu offline */ |
1522 | + if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs) |
1523 | + per_cpu(current_vmcs, cpu) = NULL; |
1524 | + crash_disable_local_vmclear(cpu); |
1525 | + list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link); |
1526 | + |
1527 | + /* |
1528 | + * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link |
1529 | + * is before setting loaded_vmcs->vcpu to -1 which is done in |
1530 | + * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist |
1531 | + * then adds the vmcs into percpu list before it is deleted. |
1532 | + */ |
1533 | + smp_wmb(); |
1534 | + |
1535 | + loaded_vmcs_init(loaded_vmcs); |
1536 | + crash_enable_local_vmclear(cpu); |
1537 | +} |
1538 | + |
1539 | +void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs) |
1540 | +{ |
1541 | + int cpu = loaded_vmcs->cpu; |
1542 | + |
1543 | + if (cpu != -1) |
1544 | + smp_call_function_single(cpu, |
1545 | + __loaded_vmcs_clear, loaded_vmcs, 1); |
1546 | +} |
1547 | + |
1548 | +static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg, |
1549 | + unsigned field) |
1550 | +{ |
1551 | + bool ret; |
1552 | + u32 mask = 1 << (seg * SEG_FIELD_NR + field); |
1553 | + |
1554 | + if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) { |
1555 | + kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS); |
1556 | + vmx->segment_cache.bitmask = 0; |
1557 | + } |
1558 | + ret = vmx->segment_cache.bitmask & mask; |
1559 | + vmx->segment_cache.bitmask |= mask; |
1560 | + return ret; |
1561 | +} |
1562 | + |
1563 | +static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg) |
1564 | +{ |
1565 | + u16 *p = &vmx->segment_cache.seg[seg].selector; |
1566 | + |
1567 | + if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL)) |
1568 | + *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector); |
1569 | + return *p; |
1570 | +} |
1571 | + |
1572 | +static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg) |
1573 | +{ |
1574 | + ulong *p = &vmx->segment_cache.seg[seg].base; |
1575 | + |
1576 | + if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE)) |
1577 | + *p = vmcs_readl(kvm_vmx_segment_fields[seg].base); |
1578 | + return *p; |
1579 | +} |
1580 | + |
1581 | +static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg) |
1582 | +{ |
1583 | + u32 *p = &vmx->segment_cache.seg[seg].limit; |
1584 | + |
1585 | + if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT)) |
1586 | + *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit); |
1587 | + return *p; |
1588 | +} |
1589 | + |
1590 | +static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg) |
1591 | +{ |
1592 | + u32 *p = &vmx->segment_cache.seg[seg].ar; |
1593 | + |
1594 | + if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR)) |
1595 | + *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes); |
1596 | + return *p; |
1597 | +} |
1598 | + |
1599 | +void update_exception_bitmap(struct kvm_vcpu *vcpu) |
1600 | +{ |
1601 | + u32 eb; |
1602 | + |
1603 | + eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) | |
1604 | + (1u << DB_VECTOR) | (1u << AC_VECTOR); |
1605 | + /* |
1606 | + * Guest access to VMware backdoor ports could legitimately |
1607 | + * trigger #GP because of TSS I/O permission bitmap. |
1608 | + * We intercept those #GP and allow access to them anyway |
1609 | + * as VMware does. |
1610 | + */ |
1611 | + if (enable_vmware_backdoor) |
1612 | + eb |= (1u << GP_VECTOR); |
1613 | + if ((vcpu->guest_debug & |
1614 | + (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) == |
1615 | + (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) |
1616 | + eb |= 1u << BP_VECTOR; |
1617 | + if (to_vmx(vcpu)->rmode.vm86_active) |
1618 | + eb = ~0; |
1619 | + if (enable_ept) |
1620 | + eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */ |
1621 | + |
1622 | + /* When we are running a nested L2 guest and L1 specified for it a |
1623 | + * certain exception bitmap, we must trap the same exceptions and pass |
1624 | + * them to L1. When running L2, we will only handle the exceptions |
1625 | + * specified above if L1 did not want them. |
1626 | + */ |
1627 | + if (is_guest_mode(vcpu)) |
1628 | + eb |= get_vmcs12(vcpu)->exception_bitmap; |
1629 | + |
1630 | + vmcs_write32(EXCEPTION_BITMAP, eb); |
1631 | +} |
1632 | + |
1633 | +/* |
1634 | + * Check if MSR is intercepted for currently loaded MSR bitmap. |
1635 | + */ |
1636 | +static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr) |
1637 | +{ |
1638 | + unsigned long *msr_bitmap; |
1639 | + int f = sizeof(unsigned long); |
1640 | + |
1641 | + if (!cpu_has_vmx_msr_bitmap()) |
1642 | + return true; |
1643 | + |
1644 | + msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap; |
1645 | + |
1646 | + if (msr <= 0x1fff) { |
1647 | + return !!test_bit(msr, msr_bitmap + 0x800 / f); |
1648 | + } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { |
1649 | + msr &= 0x1fff; |
1650 | + return !!test_bit(msr, msr_bitmap + 0xc00 / f); |
1651 | + } |
1652 | + |
1653 | + return true; |
1654 | +} |
1655 | + |
1656 | +static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx, |
1657 | + unsigned long entry, unsigned long exit) |
1658 | +{ |
1659 | + vm_entry_controls_clearbit(vmx, entry); |
1660 | + vm_exit_controls_clearbit(vmx, exit); |
1661 | +} |
1662 | + |
1663 | +int vmx_find_msr_index(struct vmx_msrs *m, u32 msr) |
1664 | +{ |
1665 | + unsigned int i; |
1666 | + |
1667 | + for (i = 0; i < m->nr; ++i) { |
1668 | + if (m->val[i].index == msr) |
1669 | + return i; |
1670 | + } |
1671 | + return -ENOENT; |
1672 | +} |
1673 | + |
1674 | +static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr) |
1675 | +{ |
1676 | + int i; |
1677 | + struct msr_autoload *m = &vmx->msr_autoload; |
1678 | + |
1679 | + switch (msr) { |
1680 | + case MSR_EFER: |
1681 | + if (cpu_has_load_ia32_efer()) { |
1682 | + clear_atomic_switch_msr_special(vmx, |
1683 | + VM_ENTRY_LOAD_IA32_EFER, |
1684 | + VM_EXIT_LOAD_IA32_EFER); |
1685 | + return; |
1686 | + } |
1687 | + break; |
1688 | + case MSR_CORE_PERF_GLOBAL_CTRL: |
1689 | + if (cpu_has_load_perf_global_ctrl()) { |
1690 | + clear_atomic_switch_msr_special(vmx, |
1691 | + VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL, |
1692 | + VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL); |
1693 | + return; |
1694 | + } |
1695 | + break; |
1696 | + } |
1697 | + i = vmx_find_msr_index(&m->guest, msr); |
1698 | + if (i < 0) |
1699 | + goto skip_guest; |
1700 | + --m->guest.nr; |
1701 | + m->guest.val[i] = m->guest.val[m->guest.nr]; |
1702 | + vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr); |
1703 | + |
1704 | +skip_guest: |
1705 | + i = vmx_find_msr_index(&m->host, msr); |
1706 | + if (i < 0) |
1707 | + return; |
1708 | + |
1709 | + --m->host.nr; |
1710 | + m->host.val[i] = m->host.val[m->host.nr]; |
1711 | + vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr); |
1712 | +} |
1713 | + |
1714 | +static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx, |
1715 | + unsigned long entry, unsigned long exit, |
1716 | + unsigned long guest_val_vmcs, unsigned long host_val_vmcs, |
1717 | + u64 guest_val, u64 host_val) |
1718 | +{ |
1719 | + vmcs_write64(guest_val_vmcs, guest_val); |
1720 | + if (host_val_vmcs != HOST_IA32_EFER) |
1721 | + vmcs_write64(host_val_vmcs, host_val); |
1722 | + vm_entry_controls_setbit(vmx, entry); |
1723 | + vm_exit_controls_setbit(vmx, exit); |
1724 | +} |
1725 | + |
1726 | +static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr, |
1727 | + u64 guest_val, u64 host_val, bool entry_only) |
1728 | +{ |
1729 | + int i, j = 0; |
1730 | + struct msr_autoload *m = &vmx->msr_autoload; |
1731 | + |
1732 | + switch (msr) { |
1733 | + case MSR_EFER: |
1734 | + if (cpu_has_load_ia32_efer()) { |
1735 | + add_atomic_switch_msr_special(vmx, |
1736 | + VM_ENTRY_LOAD_IA32_EFER, |
1737 | + VM_EXIT_LOAD_IA32_EFER, |
1738 | + GUEST_IA32_EFER, |
1739 | + HOST_IA32_EFER, |
1740 | + guest_val, host_val); |
1741 | + return; |
1742 | + } |
1743 | + break; |
1744 | + case MSR_CORE_PERF_GLOBAL_CTRL: |
1745 | + if (cpu_has_load_perf_global_ctrl()) { |
1746 | + add_atomic_switch_msr_special(vmx, |
1747 | + VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL, |
1748 | + VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL, |
1749 | + GUEST_IA32_PERF_GLOBAL_CTRL, |
1750 | + HOST_IA32_PERF_GLOBAL_CTRL, |
1751 | + guest_val, host_val); |
1752 | + return; |
1753 | + } |
1754 | + break; |
1755 | + case MSR_IA32_PEBS_ENABLE: |
1756 | + /* PEBS needs a quiescent period after being disabled (to write |
1757 | + * a record). Disabling PEBS through VMX MSR swapping doesn't |
1758 | + * provide that period, so a CPU could write host's record into |
1759 | + * guest's memory. |
1760 | + */ |
1761 | + wrmsrl(MSR_IA32_PEBS_ENABLE, 0); |
1762 | + } |
1763 | + |
1764 | + i = vmx_find_msr_index(&m->guest, msr); |
1765 | + if (!entry_only) |
1766 | + j = vmx_find_msr_index(&m->host, msr); |
1767 | + |
1768 | + if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) || |
1769 | + (j < 0 && m->host.nr == NR_LOADSTORE_MSRS)) { |
1770 | + printk_once(KERN_WARNING "Not enough msr switch entries. " |
1771 | + "Can't add msr %x\n", msr); |
1772 | + return; |
1773 | + } |
1774 | + if (i < 0) { |
1775 | + i = m->guest.nr++; |
1776 | + vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr); |
1777 | + } |
1778 | + m->guest.val[i].index = msr; |
1779 | + m->guest.val[i].value = guest_val; |
1780 | + |
1781 | + if (entry_only) |
1782 | + return; |
1783 | + |
1784 | + if (j < 0) { |
1785 | + j = m->host.nr++; |
1786 | + vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr); |
1787 | + } |
1788 | + m->host.val[j].index = msr; |
1789 | + m->host.val[j].value = host_val; |
1790 | +} |
1791 | + |
1792 | +static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset) |
1793 | +{ |
1794 | + u64 guest_efer = vmx->vcpu.arch.efer; |
1795 | + u64 ignore_bits = 0; |
1796 | + |
1797 | + /* Shadow paging assumes NX to be available. */ |
1798 | + if (!enable_ept) |
1799 | + guest_efer |= EFER_NX; |
1800 | + |
1801 | + /* |
1802 | + * LMA and LME handled by hardware; SCE meaningless outside long mode. |
1803 | + */ |
1804 | + ignore_bits |= EFER_SCE; |
1805 | +#ifdef CONFIG_X86_64 |
1806 | + ignore_bits |= EFER_LMA | EFER_LME; |
1807 | + /* SCE is meaningful only in long mode on Intel */ |
1808 | + if (guest_efer & EFER_LMA) |
1809 | + ignore_bits &= ~(u64)EFER_SCE; |
1810 | +#endif |
1811 | + |
1812 | + /* |
1813 | + * On EPT, we can't emulate NX, so we must switch EFER atomically. |
1814 | + * On CPUs that support "load IA32_EFER", always switch EFER |
1815 | + * atomically, since it's faster than switching it manually. |
1816 | + */ |
1817 | + if (cpu_has_load_ia32_efer() || |
1818 | + (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) { |
1819 | + if (!(guest_efer & EFER_LMA)) |
1820 | + guest_efer &= ~EFER_LME; |
1821 | + if (guest_efer != host_efer) |
1822 | + add_atomic_switch_msr(vmx, MSR_EFER, |
1823 | + guest_efer, host_efer, false); |
1824 | + else |
1825 | + clear_atomic_switch_msr(vmx, MSR_EFER); |
1826 | + return false; |
1827 | + } else { |
1828 | + clear_atomic_switch_msr(vmx, MSR_EFER); |
1829 | + |
1830 | + guest_efer &= ~ignore_bits; |
1831 | + guest_efer |= host_efer & ignore_bits; |
1832 | + |
1833 | + vmx->guest_msrs[efer_offset].data = guest_efer; |
1834 | + vmx->guest_msrs[efer_offset].mask = ~ignore_bits; |
1835 | + |
1836 | + return true; |
1837 | + } |
1838 | +} |
1839 | + |
1840 | +#ifdef CONFIG_X86_32 |
1841 | +/* |
1842 | + * On 32-bit kernels, VM exits still load the FS and GS bases from the |
1843 | + * VMCS rather than the segment table. KVM uses this helper to figure |
1844 | + * out the current bases to poke them into the VMCS before entry. |
1845 | + */ |
1846 | +static unsigned long segment_base(u16 selector) |
1847 | +{ |
1848 | + struct desc_struct *table; |
1849 | + unsigned long v; |
1850 | + |
1851 | + if (!(selector & ~SEGMENT_RPL_MASK)) |
1852 | + return 0; |
1853 | + |
1854 | + table = get_current_gdt_ro(); |
1855 | + |
1856 | + if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) { |
1857 | + u16 ldt_selector = kvm_read_ldt(); |
1858 | + |
1859 | + if (!(ldt_selector & ~SEGMENT_RPL_MASK)) |
1860 | + return 0; |
1861 | + |
1862 | + table = (struct desc_struct *)segment_base(ldt_selector); |
1863 | + } |
1864 | + v = get_desc_base(&table[selector >> 3]); |
1865 | + return v; |
1866 | +} |
1867 | +#endif |
1868 | + |
1869 | +static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range) |
1870 | +{ |
1871 | + u32 i; |
1872 | + |
1873 | + wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status); |
1874 | + wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base); |
1875 | + wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask); |
1876 | + wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match); |
1877 | + for (i = 0; i < addr_range; i++) { |
1878 | + wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]); |
1879 | + wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]); |
1880 | + } |
1881 | +} |
1882 | + |
1883 | +static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range) |
1884 | +{ |
1885 | + u32 i; |
1886 | + |
1887 | + rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status); |
1888 | + rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base); |
1889 | + rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask); |
1890 | + rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match); |
1891 | + for (i = 0; i < addr_range; i++) { |
1892 | + rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]); |
1893 | + rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]); |
1894 | + } |
1895 | +} |
1896 | + |
1897 | +static void pt_guest_enter(struct vcpu_vmx *vmx) |
1898 | +{ |
1899 | + if (pt_mode == PT_MODE_SYSTEM) |
1900 | + return; |
1901 | + |
1902 | + /* |
1903 | + * GUEST_IA32_RTIT_CTL is already set in the VMCS. |
1904 | + * Save host state before VM entry. |
1905 | + */ |
1906 | + rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl); |
1907 | + if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) { |
1908 | + wrmsrl(MSR_IA32_RTIT_CTL, 0); |
1909 | + pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range); |
1910 | + pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range); |
1911 | + } |
1912 | +} |
1913 | + |
1914 | +static void pt_guest_exit(struct vcpu_vmx *vmx) |
1915 | +{ |
1916 | + if (pt_mode == PT_MODE_SYSTEM) |
1917 | + return; |
1918 | + |
1919 | + if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) { |
1920 | + pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range); |
1921 | + pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range); |
1922 | + } |
1923 | + |
1924 | + /* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */ |
1925 | + wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl); |
1926 | +} |
1927 | + |
1928 | +void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel, |
1929 | + unsigned long fs_base, unsigned long gs_base) |
1930 | +{ |
1931 | + if (unlikely(fs_sel != host->fs_sel)) { |
1932 | + if (!(fs_sel & 7)) |
1933 | + vmcs_write16(HOST_FS_SELECTOR, fs_sel); |
1934 | + else |
1935 | + vmcs_write16(HOST_FS_SELECTOR, 0); |
1936 | + host->fs_sel = fs_sel; |
1937 | + } |
1938 | + if (unlikely(gs_sel != host->gs_sel)) { |
1939 | + if (!(gs_sel & 7)) |
1940 | + vmcs_write16(HOST_GS_SELECTOR, gs_sel); |
1941 | + else |
1942 | + vmcs_write16(HOST_GS_SELECTOR, 0); |
1943 | + host->gs_sel = gs_sel; |
1944 | + } |
1945 | + if (unlikely(fs_base != host->fs_base)) { |
1946 | + vmcs_writel(HOST_FS_BASE, fs_base); |
1947 | + host->fs_base = fs_base; |
1948 | + } |
1949 | + if (unlikely(gs_base != host->gs_base)) { |
1950 | + vmcs_writel(HOST_GS_BASE, gs_base); |
1951 | + host->gs_base = gs_base; |
1952 | + } |
1953 | +} |
1954 | + |
1955 | +void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu) |
1956 | +{ |
1957 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
1958 | + struct vmcs_host_state *host_state; |
1959 | +#ifdef CONFIG_X86_64 |
1960 | + int cpu = raw_smp_processor_id(); |
1961 | +#endif |
1962 | + unsigned long fs_base, gs_base; |
1963 | + u16 fs_sel, gs_sel; |
1964 | + int i; |
1965 | + |
1966 | + vmx->req_immediate_exit = false; |
1967 | + |
1968 | + /* |
1969 | + * Note that guest MSRs to be saved/restored can also be changed |
1970 | + * when guest state is loaded. This happens when guest transitions |
1971 | + * to/from long-mode by setting MSR_EFER.LMA. |
1972 | + */ |
1973 | + if (!vmx->guest_msrs_ready) { |
1974 | + vmx->guest_msrs_ready = true; |
1975 | + for (i = 0; i < vmx->save_nmsrs; ++i) |
1976 | + kvm_set_shared_msr(vmx->guest_msrs[i].index, |
1977 | + vmx->guest_msrs[i].data, |
1978 | + vmx->guest_msrs[i].mask); |
1979 | + |
1980 | + } |
1981 | + if (vmx->guest_state_loaded) |
1982 | + return; |
1983 | + |
1984 | + host_state = &vmx->loaded_vmcs->host_state; |
1985 | + |
1986 | + /* |
1987 | + * Set host fs and gs selectors. Unfortunately, 22.2.3 does not |
1988 | + * allow segment selectors with cpl > 0 or ti == 1. |
1989 | + */ |
1990 | + host_state->ldt_sel = kvm_read_ldt(); |
1991 | + |
1992 | +#ifdef CONFIG_X86_64 |
1993 | + savesegment(ds, host_state->ds_sel); |
1994 | + savesegment(es, host_state->es_sel); |
1995 | + |
1996 | + gs_base = cpu_kernelmode_gs_base(cpu); |
1997 | + if (likely(is_64bit_mm(current->mm))) { |
1998 | + save_fsgs_for_kvm(); |
1999 | + fs_sel = current->thread.fsindex; |
2000 | + gs_sel = current->thread.gsindex; |
2001 | + fs_base = current->thread.fsbase; |
2002 | + vmx->msr_host_kernel_gs_base = current->thread.gsbase; |
2003 | + } else { |
2004 | + savesegment(fs, fs_sel); |
2005 | + savesegment(gs, gs_sel); |
2006 | + fs_base = read_msr(MSR_FS_BASE); |
2007 | + vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE); |
2008 | + } |
2009 | + |
2010 | + wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base); |
2011 | +#else |
2012 | + savesegment(fs, fs_sel); |
2013 | + savesegment(gs, gs_sel); |
2014 | + fs_base = segment_base(fs_sel); |
2015 | + gs_base = segment_base(gs_sel); |
2016 | +#endif |
2017 | + |
2018 | + vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base); |
2019 | + vmx->guest_state_loaded = true; |
2020 | +} |
2021 | + |
2022 | +static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx) |
2023 | +{ |
2024 | + struct vmcs_host_state *host_state; |
2025 | + |
2026 | + if (!vmx->guest_state_loaded) |
2027 | + return; |
2028 | + |
2029 | + host_state = &vmx->loaded_vmcs->host_state; |
2030 | + |
2031 | + ++vmx->vcpu.stat.host_state_reload; |
2032 | + |
2033 | +#ifdef CONFIG_X86_64 |
2034 | + rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base); |
2035 | +#endif |
2036 | + if (host_state->ldt_sel || (host_state->gs_sel & 7)) { |
2037 | + kvm_load_ldt(host_state->ldt_sel); |
2038 | +#ifdef CONFIG_X86_64 |
2039 | + load_gs_index(host_state->gs_sel); |
2040 | +#else |
2041 | + loadsegment(gs, host_state->gs_sel); |
2042 | +#endif |
2043 | + } |
2044 | + if (host_state->fs_sel & 7) |
2045 | + loadsegment(fs, host_state->fs_sel); |
2046 | +#ifdef CONFIG_X86_64 |
2047 | + if (unlikely(host_state->ds_sel | host_state->es_sel)) { |
2048 | + loadsegment(ds, host_state->ds_sel); |
2049 | + loadsegment(es, host_state->es_sel); |
2050 | + } |
2051 | +#endif |
2052 | + invalidate_tss_limit(); |
2053 | +#ifdef CONFIG_X86_64 |
2054 | + wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base); |
2055 | +#endif |
2056 | + load_fixmap_gdt(raw_smp_processor_id()); |
2057 | + vmx->guest_state_loaded = false; |
2058 | + vmx->guest_msrs_ready = false; |
2059 | +} |
2060 | + |
2061 | +#ifdef CONFIG_X86_64 |
2062 | +static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx) |
2063 | +{ |
2064 | + preempt_disable(); |
2065 | + if (vmx->guest_state_loaded) |
2066 | + rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base); |
2067 | + preempt_enable(); |
2068 | + return vmx->msr_guest_kernel_gs_base; |
2069 | +} |
2070 | + |
2071 | +static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data) |
2072 | +{ |
2073 | + preempt_disable(); |
2074 | + if (vmx->guest_state_loaded) |
2075 | + wrmsrl(MSR_KERNEL_GS_BASE, data); |
2076 | + preempt_enable(); |
2077 | + vmx->msr_guest_kernel_gs_base = data; |
2078 | +} |
2079 | +#endif |
2080 | + |
2081 | +static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) |
2082 | +{ |
2083 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
2084 | + struct pi_desc old, new; |
2085 | + unsigned int dest; |
2086 | + |
2087 | + /* |
2088 | + * In case of hot-plug or hot-unplug, we may have to undo |
2089 | + * vmx_vcpu_pi_put even if there is no assigned device. And we |
2090 | + * always keep PI.NDST up to date for simplicity: it makes the |
2091 | + * code easier, and CPU migration is not a fast path. |
2092 | + */ |
2093 | + if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu) |
2094 | + return; |
2095 | + |
2096 | + /* |
2097 | + * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change |
2098 | + * PI.NDST: pi_post_block is the one expected to change PID.NDST and the |
2099 | + * wakeup handler expects the vCPU to be on the blocked_vcpu_list that |
2100 | + * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up |
2101 | + * correctly. |
2102 | + */ |
2103 | + if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) { |
2104 | + pi_clear_sn(pi_desc); |
2105 | + goto after_clear_sn; |
2106 | + } |
2107 | + |
2108 | + /* The full case. */ |
2109 | + do { |
2110 | + old.control = new.control = pi_desc->control; |
2111 | + |
2112 | + dest = cpu_physical_id(cpu); |
2113 | + |
2114 | + if (x2apic_enabled()) |
2115 | + new.ndst = dest; |
2116 | + else |
2117 | + new.ndst = (dest << 8) & 0xFF00; |
2118 | + |
2119 | + new.sn = 0; |
2120 | + } while (cmpxchg64(&pi_desc->control, old.control, |
2121 | + new.control) != old.control); |
2122 | + |
2123 | +after_clear_sn: |
2124 | + |
2125 | + /* |
2126 | + * Clear SN before reading the bitmap. The VT-d firmware |
2127 | + * writes the bitmap and reads SN atomically (5.2.3 in the |
2128 | + * spec), so it doesn't really have a memory barrier that |
2129 | + * pairs with this, but we cannot do that and we need one. |
2130 | + */ |
2131 | + smp_mb__after_atomic(); |
2132 | + |
2133 | + if (!pi_is_pir_empty(pi_desc)) |
2134 | + pi_set_on(pi_desc); |
2135 | +} |
2136 | + |
2137 | +void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu) |
2138 | +{ |
2139 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2140 | + bool already_loaded = vmx->loaded_vmcs->cpu == cpu; |
2141 | + |
2142 | + if (!already_loaded) { |
2143 | + loaded_vmcs_clear(vmx->loaded_vmcs); |
2144 | + local_irq_disable(); |
2145 | + crash_disable_local_vmclear(cpu); |
2146 | + |
2147 | + /* |
2148 | + * Read loaded_vmcs->cpu should be before fetching |
2149 | + * loaded_vmcs->loaded_vmcss_on_cpu_link. |
2150 | + * See the comments in __loaded_vmcs_clear(). |
2151 | + */ |
2152 | + smp_rmb(); |
2153 | + |
2154 | + list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link, |
2155 | + &per_cpu(loaded_vmcss_on_cpu, cpu)); |
2156 | + crash_enable_local_vmclear(cpu); |
2157 | + local_irq_enable(); |
2158 | + } |
2159 | + |
2160 | + if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) { |
2161 | + per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs; |
2162 | + vmcs_load(vmx->loaded_vmcs->vmcs); |
2163 | + indirect_branch_prediction_barrier(); |
2164 | + } |
2165 | + |
2166 | + if (!already_loaded) { |
2167 | + void *gdt = get_current_gdt_ro(); |
2168 | + unsigned long sysenter_esp; |
2169 | + |
2170 | + kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
2171 | + |
2172 | + /* |
2173 | + * Linux uses per-cpu TSS and GDT, so set these when switching |
2174 | + * processors. See 22.2.4. |
2175 | + */ |
2176 | + vmcs_writel(HOST_TR_BASE, |
2177 | + (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss); |
2178 | + vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */ |
2179 | + |
2180 | + rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp); |
2181 | + vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */ |
2182 | + |
2183 | + vmx->loaded_vmcs->cpu = cpu; |
2184 | + } |
2185 | + |
2186 | + /* Setup TSC multiplier */ |
2187 | + if (kvm_has_tsc_control && |
2188 | + vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) |
2189 | + decache_tsc_multiplier(vmx); |
2190 | +} |
2191 | + |
2192 | +/* |
2193 | + * Switches to specified vcpu, until a matching vcpu_put(), but assumes |
2194 | + * vcpu mutex is already taken. |
2195 | + */ |
2196 | +void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
2197 | +{ |
2198 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2199 | + |
2200 | + vmx_vcpu_load_vmcs(vcpu, cpu); |
2201 | + |
2202 | + vmx_vcpu_pi_load(vcpu, cpu); |
2203 | + |
2204 | + vmx->host_pkru = read_pkru(); |
2205 | + vmx->host_debugctlmsr = get_debugctlmsr(); |
2206 | +} |
2207 | + |
2208 | +static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu) |
2209 | +{ |
2210 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
2211 | + |
2212 | + if (!kvm_arch_has_assigned_device(vcpu->kvm) || |
2213 | + !irq_remapping_cap(IRQ_POSTING_CAP) || |
2214 | + !kvm_vcpu_apicv_active(vcpu)) |
2215 | + return; |
2216 | + |
2217 | + /* Set SN when the vCPU is preempted */ |
2218 | + if (vcpu->preempted) |
2219 | + pi_set_sn(pi_desc); |
2220 | +} |
2221 | + |
2222 | +static void vmx_vcpu_put(struct kvm_vcpu *vcpu) |
2223 | +{ |
2224 | + vmx_vcpu_pi_put(vcpu); |
2225 | + |
2226 | + vmx_prepare_switch_to_host(to_vmx(vcpu)); |
2227 | +} |
2228 | + |
2229 | +static bool emulation_required(struct kvm_vcpu *vcpu) |
2230 | +{ |
2231 | + return emulate_invalid_guest_state && !guest_state_valid(vcpu); |
2232 | +} |
2233 | + |
2234 | +static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu); |
2235 | + |
2236 | +unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu) |
2237 | +{ |
2238 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2239 | + unsigned long rflags, save_rflags; |
2240 | + |
2241 | + if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) { |
2242 | + kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS); |
2243 | + rflags = vmcs_readl(GUEST_RFLAGS); |
2244 | + if (vmx->rmode.vm86_active) { |
2245 | + rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS; |
2246 | + save_rflags = vmx->rmode.save_rflags; |
2247 | + rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS; |
2248 | + } |
2249 | + vmx->rflags = rflags; |
2250 | + } |
2251 | + return vmx->rflags; |
2252 | +} |
2253 | + |
2254 | +void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
2255 | +{ |
2256 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2257 | + unsigned long old_rflags; |
2258 | + |
2259 | + if (enable_unrestricted_guest) { |
2260 | + kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS); |
2261 | + vmx->rflags = rflags; |
2262 | + vmcs_writel(GUEST_RFLAGS, rflags); |
2263 | + return; |
2264 | + } |
2265 | + |
2266 | + old_rflags = vmx_get_rflags(vcpu); |
2267 | + vmx->rflags = rflags; |
2268 | + if (vmx->rmode.vm86_active) { |
2269 | + vmx->rmode.save_rflags = rflags; |
2270 | + rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
2271 | + } |
2272 | + vmcs_writel(GUEST_RFLAGS, rflags); |
2273 | + |
2274 | + if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM) |
2275 | + vmx->emulation_required = emulation_required(vcpu); |
2276 | +} |
2277 | + |
2278 | +u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu) |
2279 | +{ |
2280 | + u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); |
2281 | + int ret = 0; |
2282 | + |
2283 | + if (interruptibility & GUEST_INTR_STATE_STI) |
2284 | + ret |= KVM_X86_SHADOW_INT_STI; |
2285 | + if (interruptibility & GUEST_INTR_STATE_MOV_SS) |
2286 | + ret |= KVM_X86_SHADOW_INT_MOV_SS; |
2287 | + |
2288 | + return ret; |
2289 | +} |
2290 | + |
2291 | +void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) |
2292 | +{ |
2293 | + u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); |
2294 | + u32 interruptibility = interruptibility_old; |
2295 | + |
2296 | + interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS); |
2297 | + |
2298 | + if (mask & KVM_X86_SHADOW_INT_MOV_SS) |
2299 | + interruptibility |= GUEST_INTR_STATE_MOV_SS; |
2300 | + else if (mask & KVM_X86_SHADOW_INT_STI) |
2301 | + interruptibility |= GUEST_INTR_STATE_STI; |
2302 | + |
2303 | + if ((interruptibility != interruptibility_old)) |
2304 | + vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility); |
2305 | +} |
2306 | + |
2307 | +static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data) |
2308 | +{ |
2309 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2310 | + unsigned long value; |
2311 | + |
2312 | + /* |
2313 | + * Any MSR write that attempts to change bits marked reserved will |
2314 | + * case a #GP fault. |
2315 | + */ |
2316 | + if (data & vmx->pt_desc.ctl_bitmask) |
2317 | + return 1; |
2318 | + |
2319 | + /* |
2320 | + * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will |
2321 | + * result in a #GP unless the same write also clears TraceEn. |
2322 | + */ |
2323 | + if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) && |
2324 | + ((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN)) |
2325 | + return 1; |
2326 | + |
2327 | + /* |
2328 | + * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit |
2329 | + * and FabricEn would cause #GP, if |
2330 | + * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0 |
2331 | + */ |
2332 | + if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) && |
2333 | + !(data & RTIT_CTL_FABRIC_EN) && |
2334 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2335 | + PT_CAP_single_range_output)) |
2336 | + return 1; |
2337 | + |
2338 | + /* |
2339 | + * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that |
2340 | + * utilize encodings marked reserved will casue a #GP fault. |
2341 | + */ |
2342 | + value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods); |
2343 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) && |
2344 | + !test_bit((data & RTIT_CTL_MTC_RANGE) >> |
2345 | + RTIT_CTL_MTC_RANGE_OFFSET, &value)) |
2346 | + return 1; |
2347 | + value = intel_pt_validate_cap(vmx->pt_desc.caps, |
2348 | + PT_CAP_cycle_thresholds); |
2349 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) && |
2350 | + !test_bit((data & RTIT_CTL_CYC_THRESH) >> |
2351 | + RTIT_CTL_CYC_THRESH_OFFSET, &value)) |
2352 | + return 1; |
2353 | + value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods); |
2354 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) && |
2355 | + !test_bit((data & RTIT_CTL_PSB_FREQ) >> |
2356 | + RTIT_CTL_PSB_FREQ_OFFSET, &value)) |
2357 | + return 1; |
2358 | + |
2359 | + /* |
2360 | + * If ADDRx_CFG is reserved or the encodings is >2 will |
2361 | + * cause a #GP fault. |
2362 | + */ |
2363 | + value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET; |
2364 | + if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2)) |
2365 | + return 1; |
2366 | + value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET; |
2367 | + if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2)) |
2368 | + return 1; |
2369 | + value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET; |
2370 | + if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2)) |
2371 | + return 1; |
2372 | + value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET; |
2373 | + if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2)) |
2374 | + return 1; |
2375 | + |
2376 | + return 0; |
2377 | +} |
2378 | + |
2379 | +static int skip_emulated_instruction(struct kvm_vcpu *vcpu) |
2380 | +{ |
2381 | + unsigned long rip; |
2382 | + |
2383 | + /* |
2384 | + * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on |
2385 | + * undefined behavior: Intel's SDM doesn't mandate the VMCS field be |
2386 | + * set when EPT misconfig occurs. In practice, real hardware updates |
2387 | + * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors |
2388 | + * (namely Hyper-V) don't set it due to it being undefined behavior, |
2389 | + * i.e. we end up advancing IP with some random value. |
2390 | + */ |
2391 | + if (!static_cpu_has(X86_FEATURE_HYPERVISOR) || |
2392 | + to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) { |
2393 | + rip = kvm_rip_read(vcpu); |
2394 | + rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
2395 | + kvm_rip_write(vcpu, rip); |
2396 | + } else { |
2397 | + if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP)) |
2398 | + return 0; |
2399 | + } |
2400 | + |
2401 | + /* skipping an emulated instruction also counts */ |
2402 | + vmx_set_interrupt_shadow(vcpu, 0); |
2403 | + |
2404 | + return 1; |
2405 | +} |
2406 | + |
2407 | +static void vmx_clear_hlt(struct kvm_vcpu *vcpu) |
2408 | +{ |
2409 | + /* |
2410 | + * Ensure that we clear the HLT state in the VMCS. We don't need to |
2411 | + * explicitly skip the instruction because if the HLT state is set, |
2412 | + * then the instruction is already executing and RIP has already been |
2413 | + * advanced. |
2414 | + */ |
2415 | + if (kvm_hlt_in_guest(vcpu->kvm) && |
2416 | + vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT) |
2417 | + vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE); |
2418 | +} |
2419 | + |
2420 | +static void vmx_queue_exception(struct kvm_vcpu *vcpu) |
2421 | +{ |
2422 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2423 | + unsigned nr = vcpu->arch.exception.nr; |
2424 | + bool has_error_code = vcpu->arch.exception.has_error_code; |
2425 | + u32 error_code = vcpu->arch.exception.error_code; |
2426 | + u32 intr_info = nr | INTR_INFO_VALID_MASK; |
2427 | + |
2428 | + kvm_deliver_exception_payload(vcpu); |
2429 | + |
2430 | + if (has_error_code) { |
2431 | + vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code); |
2432 | + intr_info |= INTR_INFO_DELIVER_CODE_MASK; |
2433 | + } |
2434 | + |
2435 | + if (vmx->rmode.vm86_active) { |
2436 | + int inc_eip = 0; |
2437 | + if (kvm_exception_is_soft(nr)) |
2438 | + inc_eip = vcpu->arch.event_exit_inst_len; |
2439 | + kvm_inject_realmode_interrupt(vcpu, nr, inc_eip); |
2440 | + return; |
2441 | + } |
2442 | + |
2443 | + WARN_ON_ONCE(vmx->emulation_required); |
2444 | + |
2445 | + if (kvm_exception_is_soft(nr)) { |
2446 | + vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, |
2447 | + vmx->vcpu.arch.event_exit_inst_len); |
2448 | + intr_info |= INTR_TYPE_SOFT_EXCEPTION; |
2449 | + } else |
2450 | + intr_info |= INTR_TYPE_HARD_EXCEPTION; |
2451 | + |
2452 | + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info); |
2453 | + |
2454 | + vmx_clear_hlt(vcpu); |
2455 | +} |
2456 | + |
2457 | +static bool vmx_rdtscp_supported(void) |
2458 | +{ |
2459 | + return cpu_has_vmx_rdtscp(); |
2460 | +} |
2461 | + |
2462 | +static bool vmx_invpcid_supported(void) |
2463 | +{ |
2464 | + return cpu_has_vmx_invpcid(); |
2465 | +} |
2466 | + |
2467 | +/* |
2468 | + * Swap MSR entry in host/guest MSR entry array. |
2469 | + */ |
2470 | +static void move_msr_up(struct vcpu_vmx *vmx, int from, int to) |
2471 | +{ |
2472 | + struct shared_msr_entry tmp; |
2473 | + |
2474 | + tmp = vmx->guest_msrs[to]; |
2475 | + vmx->guest_msrs[to] = vmx->guest_msrs[from]; |
2476 | + vmx->guest_msrs[from] = tmp; |
2477 | +} |
2478 | + |
2479 | +/* |
2480 | + * Set up the vmcs to automatically save and restore system |
2481 | + * msrs. Don't touch the 64-bit msrs if the guest is in legacy |
2482 | + * mode, as fiddling with msrs is very expensive. |
2483 | + */ |
2484 | +static void setup_msrs(struct vcpu_vmx *vmx) |
2485 | +{ |
2486 | + int save_nmsrs, index; |
2487 | + |
2488 | + save_nmsrs = 0; |
2489 | +#ifdef CONFIG_X86_64 |
2490 | + /* |
2491 | + * The SYSCALL MSRs are only needed on long mode guests, and only |
2492 | + * when EFER.SCE is set. |
2493 | + */ |
2494 | + if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) { |
2495 | + index = __find_msr_index(vmx, MSR_STAR); |
2496 | + if (index >= 0) |
2497 | + move_msr_up(vmx, index, save_nmsrs++); |
2498 | + index = __find_msr_index(vmx, MSR_LSTAR); |
2499 | + if (index >= 0) |
2500 | + move_msr_up(vmx, index, save_nmsrs++); |
2501 | + index = __find_msr_index(vmx, MSR_SYSCALL_MASK); |
2502 | + if (index >= 0) |
2503 | + move_msr_up(vmx, index, save_nmsrs++); |
2504 | + } |
2505 | +#endif |
2506 | + index = __find_msr_index(vmx, MSR_EFER); |
2507 | + if (index >= 0 && update_transition_efer(vmx, index)) |
2508 | + move_msr_up(vmx, index, save_nmsrs++); |
2509 | + index = __find_msr_index(vmx, MSR_TSC_AUX); |
2510 | + if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP)) |
2511 | + move_msr_up(vmx, index, save_nmsrs++); |
2512 | + index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL); |
2513 | + if (index >= 0) |
2514 | + move_msr_up(vmx, index, save_nmsrs++); |
2515 | + |
2516 | + vmx->save_nmsrs = save_nmsrs; |
2517 | + vmx->guest_msrs_ready = false; |
2518 | + |
2519 | + if (cpu_has_vmx_msr_bitmap()) |
2520 | + vmx_update_msr_bitmap(&vmx->vcpu); |
2521 | +} |
2522 | + |
2523 | +static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu) |
2524 | +{ |
2525 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
2526 | + |
2527 | + if (is_guest_mode(vcpu) && |
2528 | + (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)) |
2529 | + return vcpu->arch.tsc_offset - vmcs12->tsc_offset; |
2530 | + |
2531 | + return vcpu->arch.tsc_offset; |
2532 | +} |
2533 | + |
2534 | +static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
2535 | +{ |
2536 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
2537 | + u64 g_tsc_offset = 0; |
2538 | + |
2539 | + /* |
2540 | + * We're here if L1 chose not to trap WRMSR to TSC. According |
2541 | + * to the spec, this should set L1's TSC; The offset that L1 |
2542 | + * set for L2 remains unchanged, and still needs to be added |
2543 | + * to the newly set TSC to get L2's TSC. |
2544 | + */ |
2545 | + if (is_guest_mode(vcpu) && |
2546 | + (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)) |
2547 | + g_tsc_offset = vmcs12->tsc_offset; |
2548 | + |
2549 | + trace_kvm_write_tsc_offset(vcpu->vcpu_id, |
2550 | + vcpu->arch.tsc_offset - g_tsc_offset, |
2551 | + offset); |
2552 | + vmcs_write64(TSC_OFFSET, offset + g_tsc_offset); |
2553 | + return offset + g_tsc_offset; |
2554 | +} |
2555 | + |
2556 | +/* |
2557 | + * nested_vmx_allowed() checks whether a guest should be allowed to use VMX |
2558 | + * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for |
2559 | + * all guests if the "nested" module option is off, and can also be disabled |
2560 | + * for a single guest by disabling its VMX cpuid bit. |
2561 | + */ |
2562 | +bool nested_vmx_allowed(struct kvm_vcpu *vcpu) |
2563 | +{ |
2564 | + return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX); |
2565 | +} |
2566 | + |
2567 | +static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu, |
2568 | + uint64_t val) |
2569 | +{ |
2570 | + uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits; |
2571 | + |
2572 | + return !(val & ~valid_bits); |
2573 | +} |
2574 | + |
2575 | +static int vmx_get_msr_feature(struct kvm_msr_entry *msr) |
2576 | +{ |
2577 | + switch (msr->index) { |
2578 | + case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: |
2579 | + if (!nested) |
2580 | + return 1; |
2581 | + return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data); |
2582 | + default: |
2583 | + return 1; |
2584 | + } |
2585 | +} |
2586 | + |
2587 | +/* |
2588 | + * Reads an msr value (of 'msr_index') into 'pdata'. |
2589 | + * Returns 0 on success, non-0 otherwise. |
2590 | + * Assumes vcpu_load() was already called. |
2591 | + */ |
2592 | +static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
2593 | +{ |
2594 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2595 | + struct shared_msr_entry *msr; |
2596 | + u32 index; |
2597 | + |
2598 | + switch (msr_info->index) { |
2599 | +#ifdef CONFIG_X86_64 |
2600 | + case MSR_FS_BASE: |
2601 | + msr_info->data = vmcs_readl(GUEST_FS_BASE); |
2602 | + break; |
2603 | + case MSR_GS_BASE: |
2604 | + msr_info->data = vmcs_readl(GUEST_GS_BASE); |
2605 | + break; |
2606 | + case MSR_KERNEL_GS_BASE: |
2607 | + msr_info->data = vmx_read_guest_kernel_gs_base(vmx); |
2608 | + break; |
2609 | +#endif |
2610 | + case MSR_EFER: |
2611 | + return kvm_get_msr_common(vcpu, msr_info); |
2612 | + case MSR_IA32_TSX_CTRL: |
2613 | + if (!msr_info->host_initiated && |
2614 | + !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR)) |
2615 | + return 1; |
2616 | + goto find_shared_msr; |
2617 | + case MSR_IA32_UMWAIT_CONTROL: |
2618 | + if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx)) |
2619 | + return 1; |
2620 | + |
2621 | + msr_info->data = vmx->msr_ia32_umwait_control; |
2622 | + break; |
2623 | + case MSR_IA32_SPEC_CTRL: |
2624 | + if (!msr_info->host_initiated && |
2625 | + !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL)) |
2626 | + return 1; |
2627 | + |
2628 | + msr_info->data = to_vmx(vcpu)->spec_ctrl; |
2629 | + break; |
2630 | + case MSR_IA32_SYSENTER_CS: |
2631 | + msr_info->data = vmcs_read32(GUEST_SYSENTER_CS); |
2632 | + break; |
2633 | + case MSR_IA32_SYSENTER_EIP: |
2634 | + msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP); |
2635 | + break; |
2636 | + case MSR_IA32_SYSENTER_ESP: |
2637 | + msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP); |
2638 | + break; |
2639 | + case MSR_IA32_BNDCFGS: |
2640 | + if (!kvm_mpx_supported() || |
2641 | + (!msr_info->host_initiated && |
2642 | + !guest_cpuid_has(vcpu, X86_FEATURE_MPX))) |
2643 | + return 1; |
2644 | + msr_info->data = vmcs_read64(GUEST_BNDCFGS); |
2645 | + break; |
2646 | + case MSR_IA32_MCG_EXT_CTL: |
2647 | + if (!msr_info->host_initiated && |
2648 | + !(vmx->msr_ia32_feature_control & |
2649 | + FEATURE_CONTROL_LMCE)) |
2650 | + return 1; |
2651 | + msr_info->data = vcpu->arch.mcg_ext_ctl; |
2652 | + break; |
2653 | + case MSR_IA32_FEATURE_CONTROL: |
2654 | + msr_info->data = vmx->msr_ia32_feature_control; |
2655 | + break; |
2656 | + case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: |
2657 | + if (!nested_vmx_allowed(vcpu)) |
2658 | + return 1; |
2659 | + return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index, |
2660 | + &msr_info->data); |
2661 | + case MSR_IA32_RTIT_CTL: |
2662 | + if (pt_mode != PT_MODE_HOST_GUEST) |
2663 | + return 1; |
2664 | + msr_info->data = vmx->pt_desc.guest.ctl; |
2665 | + break; |
2666 | + case MSR_IA32_RTIT_STATUS: |
2667 | + if (pt_mode != PT_MODE_HOST_GUEST) |
2668 | + return 1; |
2669 | + msr_info->data = vmx->pt_desc.guest.status; |
2670 | + break; |
2671 | + case MSR_IA32_RTIT_CR3_MATCH: |
2672 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2673 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2674 | + PT_CAP_cr3_filtering)) |
2675 | + return 1; |
2676 | + msr_info->data = vmx->pt_desc.guest.cr3_match; |
2677 | + break; |
2678 | + case MSR_IA32_RTIT_OUTPUT_BASE: |
2679 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2680 | + (!intel_pt_validate_cap(vmx->pt_desc.caps, |
2681 | + PT_CAP_topa_output) && |
2682 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2683 | + PT_CAP_single_range_output))) |
2684 | + return 1; |
2685 | + msr_info->data = vmx->pt_desc.guest.output_base; |
2686 | + break; |
2687 | + case MSR_IA32_RTIT_OUTPUT_MASK: |
2688 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2689 | + (!intel_pt_validate_cap(vmx->pt_desc.caps, |
2690 | + PT_CAP_topa_output) && |
2691 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2692 | + PT_CAP_single_range_output))) |
2693 | + return 1; |
2694 | + msr_info->data = vmx->pt_desc.guest.output_mask; |
2695 | + break; |
2696 | + case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: |
2697 | + index = msr_info->index - MSR_IA32_RTIT_ADDR0_A; |
2698 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2699 | + (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps, |
2700 | + PT_CAP_num_address_ranges))) |
2701 | + return 1; |
2702 | + if (is_noncanonical_address(data, vcpu)) |
2703 | + return 1; |
2704 | + if (index % 2) |
2705 | + msr_info->data = vmx->pt_desc.guest.addr_b[index / 2]; |
2706 | + else |
2707 | + msr_info->data = vmx->pt_desc.guest.addr_a[index / 2]; |
2708 | + break; |
2709 | + case MSR_TSC_AUX: |
2710 | + if (!msr_info->host_initiated && |
2711 | + !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP)) |
2712 | + return 1; |
2713 | + goto find_shared_msr; |
2714 | + default: |
2715 | + find_shared_msr: |
2716 | + msr = find_msr_entry(vmx, msr_info->index); |
2717 | + if (msr) { |
2718 | + msr_info->data = msr->data; |
2719 | + break; |
2720 | + } |
2721 | + return kvm_get_msr_common(vcpu, msr_info); |
2722 | + } |
2723 | + |
2724 | + return 0; |
2725 | +} |
2726 | + |
2727 | +/* |
2728 | + * Writes msr value into the appropriate "register". |
2729 | + * Returns 0 on success, non-0 otherwise. |
2730 | + * Assumes vcpu_load() was already called. |
2731 | + */ |
2732 | +static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
2733 | +{ |
2734 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
2735 | + struct shared_msr_entry *msr; |
2736 | + int ret = 0; |
2737 | + u32 msr_index = msr_info->index; |
2738 | + u64 data = msr_info->data; |
2739 | + u32 index; |
2740 | + |
2741 | + switch (msr_index) { |
2742 | + case MSR_EFER: |
2743 | + ret = kvm_set_msr_common(vcpu, msr_info); |
2744 | + break; |
2745 | +#ifdef CONFIG_X86_64 |
2746 | + case MSR_FS_BASE: |
2747 | + vmx_segment_cache_clear(vmx); |
2748 | + vmcs_writel(GUEST_FS_BASE, data); |
2749 | + break; |
2750 | + case MSR_GS_BASE: |
2751 | + vmx_segment_cache_clear(vmx); |
2752 | + vmcs_writel(GUEST_GS_BASE, data); |
2753 | + break; |
2754 | + case MSR_KERNEL_GS_BASE: |
2755 | + vmx_write_guest_kernel_gs_base(vmx, data); |
2756 | + break; |
2757 | +#endif |
2758 | + case MSR_IA32_SYSENTER_CS: |
2759 | + if (is_guest_mode(vcpu)) |
2760 | + get_vmcs12(vcpu)->guest_sysenter_cs = data; |
2761 | + vmcs_write32(GUEST_SYSENTER_CS, data); |
2762 | + break; |
2763 | + case MSR_IA32_SYSENTER_EIP: |
2764 | + if (is_guest_mode(vcpu)) |
2765 | + get_vmcs12(vcpu)->guest_sysenter_eip = data; |
2766 | + vmcs_writel(GUEST_SYSENTER_EIP, data); |
2767 | + break; |
2768 | + case MSR_IA32_SYSENTER_ESP: |
2769 | + if (is_guest_mode(vcpu)) |
2770 | + get_vmcs12(vcpu)->guest_sysenter_esp = data; |
2771 | + vmcs_writel(GUEST_SYSENTER_ESP, data); |
2772 | + break; |
2773 | + case MSR_IA32_DEBUGCTLMSR: |
2774 | + if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls & |
2775 | + VM_EXIT_SAVE_DEBUG_CONTROLS) |
2776 | + get_vmcs12(vcpu)->guest_ia32_debugctl = data; |
2777 | + |
2778 | + ret = kvm_set_msr_common(vcpu, msr_info); |
2779 | + break; |
2780 | + |
2781 | + case MSR_IA32_BNDCFGS: |
2782 | + if (!kvm_mpx_supported() || |
2783 | + (!msr_info->host_initiated && |
2784 | + !guest_cpuid_has(vcpu, X86_FEATURE_MPX))) |
2785 | + return 1; |
2786 | + if (is_noncanonical_address(data & PAGE_MASK, vcpu) || |
2787 | + (data & MSR_IA32_BNDCFGS_RSVD)) |
2788 | + return 1; |
2789 | + vmcs_write64(GUEST_BNDCFGS, data); |
2790 | + break; |
2791 | + case MSR_IA32_UMWAIT_CONTROL: |
2792 | + if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx)) |
2793 | + return 1; |
2794 | + |
2795 | + /* The reserved bit 1 and non-32 bit [63:32] should be zero */ |
2796 | + if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32))) |
2797 | + return 1; |
2798 | + |
2799 | + vmx->msr_ia32_umwait_control = data; |
2800 | + break; |
2801 | + case MSR_IA32_SPEC_CTRL: |
2802 | + if (!msr_info->host_initiated && |
2803 | + !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL)) |
2804 | + return 1; |
2805 | + |
2806 | + /* The STIBP bit doesn't fault even if it's not advertised */ |
2807 | + if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD)) |
2808 | + return 1; |
2809 | + |
2810 | + vmx->spec_ctrl = data; |
2811 | + |
2812 | + if (!data) |
2813 | + break; |
2814 | + |
2815 | + /* |
2816 | + * For non-nested: |
2817 | + * When it's written (to non-zero) for the first time, pass |
2818 | + * it through. |
2819 | + * |
2820 | + * For nested: |
2821 | + * The handling of the MSR bitmap for L2 guests is done in |
2822 | + * nested_vmx_prepare_msr_bitmap. We should not touch the |
2823 | + * vmcs02.msr_bitmap here since it gets completely overwritten |
2824 | + * in the merging. We update the vmcs01 here for L1 as well |
2825 | + * since it will end up touching the MSR anyway now. |
2826 | + */ |
2827 | + vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, |
2828 | + MSR_IA32_SPEC_CTRL, |
2829 | + MSR_TYPE_RW); |
2830 | + break; |
2831 | + case MSR_IA32_TSX_CTRL: |
2832 | + if (!msr_info->host_initiated && |
2833 | + !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR)) |
2834 | + return 1; |
2835 | + if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR)) |
2836 | + return 1; |
2837 | + goto find_shared_msr; |
2838 | + case MSR_IA32_PRED_CMD: |
2839 | + if (!msr_info->host_initiated && |
2840 | + !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL)) |
2841 | + return 1; |
2842 | + |
2843 | + if (data & ~PRED_CMD_IBPB) |
2844 | + return 1; |
2845 | + |
2846 | + if (!data) |
2847 | + break; |
2848 | + |
2849 | + wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB); |
2850 | + |
2851 | + /* |
2852 | + * For non-nested: |
2853 | + * When it's written (to non-zero) for the first time, pass |
2854 | + * it through. |
2855 | + * |
2856 | + * For nested: |
2857 | + * The handling of the MSR bitmap for L2 guests is done in |
2858 | + * nested_vmx_prepare_msr_bitmap. We should not touch the |
2859 | + * vmcs02.msr_bitmap here since it gets completely overwritten |
2860 | + * in the merging. |
2861 | + */ |
2862 | + vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD, |
2863 | + MSR_TYPE_W); |
2864 | + break; |
2865 | + case MSR_IA32_CR_PAT: |
2866 | + if (!kvm_pat_valid(data)) |
2867 | + return 1; |
2868 | + |
2869 | + if (is_guest_mode(vcpu) && |
2870 | + get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT) |
2871 | + get_vmcs12(vcpu)->guest_ia32_pat = data; |
2872 | + |
2873 | + if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) { |
2874 | + vmcs_write64(GUEST_IA32_PAT, data); |
2875 | + vcpu->arch.pat = data; |
2876 | + break; |
2877 | + } |
2878 | + ret = kvm_set_msr_common(vcpu, msr_info); |
2879 | + break; |
2880 | + case MSR_IA32_TSC_ADJUST: |
2881 | + ret = kvm_set_msr_common(vcpu, msr_info); |
2882 | + break; |
2883 | + case MSR_IA32_MCG_EXT_CTL: |
2884 | + if ((!msr_info->host_initiated && |
2885 | + !(to_vmx(vcpu)->msr_ia32_feature_control & |
2886 | + FEATURE_CONTROL_LMCE)) || |
2887 | + (data & ~MCG_EXT_CTL_LMCE_EN)) |
2888 | + return 1; |
2889 | + vcpu->arch.mcg_ext_ctl = data; |
2890 | + break; |
2891 | + case MSR_IA32_FEATURE_CONTROL: |
2892 | + if (!vmx_feature_control_msr_valid(vcpu, data) || |
2893 | + (to_vmx(vcpu)->msr_ia32_feature_control & |
2894 | + FEATURE_CONTROL_LOCKED && !msr_info->host_initiated)) |
2895 | + return 1; |
2896 | + vmx->msr_ia32_feature_control = data; |
2897 | + if (msr_info->host_initiated && data == 0) |
2898 | + vmx_leave_nested(vcpu); |
2899 | + break; |
2900 | + case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: |
2901 | + if (!msr_info->host_initiated) |
2902 | + return 1; /* they are read-only */ |
2903 | + if (!nested_vmx_allowed(vcpu)) |
2904 | + return 1; |
2905 | + return vmx_set_vmx_msr(vcpu, msr_index, data); |
2906 | + case MSR_IA32_RTIT_CTL: |
2907 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2908 | + vmx_rtit_ctl_check(vcpu, data) || |
2909 | + vmx->nested.vmxon) |
2910 | + return 1; |
2911 | + vmcs_write64(GUEST_IA32_RTIT_CTL, data); |
2912 | + vmx->pt_desc.guest.ctl = data; |
2913 | + pt_update_intercept_for_msr(vmx); |
2914 | + break; |
2915 | + case MSR_IA32_RTIT_STATUS: |
2916 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2917 | + (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) || |
2918 | + (data & MSR_IA32_RTIT_STATUS_MASK)) |
2919 | + return 1; |
2920 | + vmx->pt_desc.guest.status = data; |
2921 | + break; |
2922 | + case MSR_IA32_RTIT_CR3_MATCH: |
2923 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2924 | + (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) || |
2925 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2926 | + PT_CAP_cr3_filtering)) |
2927 | + return 1; |
2928 | + vmx->pt_desc.guest.cr3_match = data; |
2929 | + break; |
2930 | + case MSR_IA32_RTIT_OUTPUT_BASE: |
2931 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2932 | + (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) || |
2933 | + (!intel_pt_validate_cap(vmx->pt_desc.caps, |
2934 | + PT_CAP_topa_output) && |
2935 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2936 | + PT_CAP_single_range_output)) || |
2937 | + (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)) |
2938 | + return 1; |
2939 | + vmx->pt_desc.guest.output_base = data; |
2940 | + break; |
2941 | + case MSR_IA32_RTIT_OUTPUT_MASK: |
2942 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2943 | + (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) || |
2944 | + (!intel_pt_validate_cap(vmx->pt_desc.caps, |
2945 | + PT_CAP_topa_output) && |
2946 | + !intel_pt_validate_cap(vmx->pt_desc.caps, |
2947 | + PT_CAP_single_range_output))) |
2948 | + return 1; |
2949 | + vmx->pt_desc.guest.output_mask = data; |
2950 | + break; |
2951 | + case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: |
2952 | + index = msr_info->index - MSR_IA32_RTIT_ADDR0_A; |
2953 | + if ((pt_mode != PT_MODE_HOST_GUEST) || |
2954 | + (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) || |
2955 | + (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps, |
2956 | + PT_CAP_num_address_ranges))) |
2957 | + return 1; |
2958 | + if (is_noncanonical_address(data, vcpu)) |
2959 | + return 1; |
2960 | + if (index % 2) |
2961 | + vmx->pt_desc.guest.addr_b[index / 2] = data; |
2962 | + else |
2963 | + vmx->pt_desc.guest.addr_a[index / 2] = data; |
2964 | + break; |
2965 | + case MSR_TSC_AUX: |
2966 | + if (!msr_info->host_initiated && |
2967 | + !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP)) |
2968 | + return 1; |
2969 | + /* Check reserved bit, higher 32 bits should be zero */ |
2970 | + if ((data >> 32) != 0) |
2971 | + return 1; |
2972 | + goto find_shared_msr; |
2973 | + |
2974 | + default: |
2975 | + find_shared_msr: |
2976 | + msr = find_msr_entry(vmx, msr_index); |
2977 | + if (msr) |
2978 | + ret = vmx_set_guest_msr(vmx, msr, data); |
2979 | + else |
2980 | + ret = kvm_set_msr_common(vcpu, msr_info); |
2981 | + } |
2982 | + |
2983 | + return ret; |
2984 | +} |
2985 | + |
2986 | +static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
2987 | +{ |
2988 | + kvm_register_mark_available(vcpu, reg); |
2989 | + |
2990 | + switch (reg) { |
2991 | + case VCPU_REGS_RSP: |
2992 | + vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP); |
2993 | + break; |
2994 | + case VCPU_REGS_RIP: |
2995 | + vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP); |
2996 | + break; |
2997 | + case VCPU_EXREG_PDPTR: |
2998 | + if (enable_ept) |
2999 | + ept_save_pdptrs(vcpu); |
3000 | + break; |
3001 | + case VCPU_EXREG_CR3: |
3002 | + if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu))) |
3003 | + vcpu->arch.cr3 = vmcs_readl(GUEST_CR3); |
3004 | + break; |
3005 | + default: |
3006 | + WARN_ON_ONCE(1); |
3007 | + break; |
3008 | + } |
3009 | +} |
3010 | + |
3011 | +static __init int cpu_has_kvm_support(void) |
3012 | +{ |
3013 | + return cpu_has_vmx(); |
3014 | +} |
3015 | + |
3016 | +static __init int vmx_disabled_by_bios(void) |
3017 | +{ |
3018 | + u64 msr; |
3019 | + |
3020 | + rdmsrl(MSR_IA32_FEATURE_CONTROL, msr); |
3021 | + if (msr & FEATURE_CONTROL_LOCKED) { |
3022 | + /* launched w/ TXT and VMX disabled */ |
3023 | + if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX) |
3024 | + && tboot_enabled()) |
3025 | + return 1; |
3026 | + /* launched w/o TXT and VMX only enabled w/ TXT */ |
3027 | + if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX) |
3028 | + && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX) |
3029 | + && !tboot_enabled()) { |
3030 | + printk(KERN_WARNING "kvm: disable TXT in the BIOS or " |
3031 | + "activate TXT before enabling KVM\n"); |
3032 | + return 1; |
3033 | + } |
3034 | + /* launched w/o TXT and VMX disabled */ |
3035 | + if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX) |
3036 | + && !tboot_enabled()) |
3037 | + return 1; |
3038 | + } |
3039 | + |
3040 | + return 0; |
3041 | +} |
3042 | + |
3043 | +static void kvm_cpu_vmxon(u64 addr) |
3044 | +{ |
3045 | + cr4_set_bits(X86_CR4_VMXE); |
3046 | + intel_pt_handle_vmx(1); |
3047 | + |
3048 | + asm volatile ("vmxon %0" : : "m"(addr)); |
3049 | +} |
3050 | + |
3051 | +static int hardware_enable(void) |
3052 | +{ |
3053 | + int cpu = raw_smp_processor_id(); |
3054 | + u64 phys_addr = __pa(per_cpu(vmxarea, cpu)); |
3055 | + u64 old, test_bits; |
3056 | + |
3057 | + if (cr4_read_shadow() & X86_CR4_VMXE) |
3058 | + return -EBUSY; |
3059 | + |
3060 | + /* |
3061 | + * This can happen if we hot-added a CPU but failed to allocate |
3062 | + * VP assist page for it. |
3063 | + */ |
3064 | + if (static_branch_unlikely(&enable_evmcs) && |
3065 | + !hv_get_vp_assist_page(cpu)) |
3066 | + return -EFAULT; |
3067 | + |
3068 | + INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu)); |
3069 | + INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu)); |
3070 | + spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); |
3071 | + |
3072 | + /* |
3073 | + * Now we can enable the vmclear operation in kdump |
3074 | + * since the loaded_vmcss_on_cpu list on this cpu |
3075 | + * has been initialized. |
3076 | + * |
3077 | + * Though the cpu is not in VMX operation now, there |
3078 | + * is no problem to enable the vmclear operation |
3079 | + * for the loaded_vmcss_on_cpu list is empty! |
3080 | + */ |
3081 | + crash_enable_local_vmclear(cpu); |
3082 | + |
3083 | + rdmsrl(MSR_IA32_FEATURE_CONTROL, old); |
3084 | + |
3085 | + test_bits = FEATURE_CONTROL_LOCKED; |
3086 | + test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX; |
3087 | + if (tboot_enabled()) |
3088 | + test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX; |
3089 | + |
3090 | + if ((old & test_bits) != test_bits) { |
3091 | + /* enable and lock */ |
3092 | + wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits); |
3093 | + } |
3094 | + kvm_cpu_vmxon(phys_addr); |
3095 | + if (enable_ept) |
3096 | + ept_sync_global(); |
3097 | + |
3098 | + return 0; |
3099 | +} |
3100 | + |
3101 | +static void vmclear_local_loaded_vmcss(void) |
3102 | +{ |
3103 | + int cpu = raw_smp_processor_id(); |
3104 | + struct loaded_vmcs *v, *n; |
3105 | + |
3106 | + list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu), |
3107 | + loaded_vmcss_on_cpu_link) |
3108 | + __loaded_vmcs_clear(v); |
3109 | +} |
3110 | + |
3111 | + |
3112 | +/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot() |
3113 | + * tricks. |
3114 | + */ |
3115 | +static void kvm_cpu_vmxoff(void) |
3116 | +{ |
3117 | + asm volatile (__ex("vmxoff")); |
3118 | + |
3119 | + intel_pt_handle_vmx(0); |
3120 | + cr4_clear_bits(X86_CR4_VMXE); |
3121 | +} |
3122 | + |
3123 | +static void hardware_disable(void) |
3124 | +{ |
3125 | + vmclear_local_loaded_vmcss(); |
3126 | + kvm_cpu_vmxoff(); |
3127 | +} |
3128 | + |
3129 | +static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, |
3130 | + u32 msr, u32 *result) |
3131 | +{ |
3132 | + u32 vmx_msr_low, vmx_msr_high; |
3133 | + u32 ctl = ctl_min | ctl_opt; |
3134 | + |
3135 | + rdmsr(msr, vmx_msr_low, vmx_msr_high); |
3136 | + |
3137 | + ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */ |
3138 | + ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */ |
3139 | + |
3140 | + /* Ensure minimum (required) set of control bits are supported. */ |
3141 | + if (ctl_min & ~ctl) |
3142 | + return -EIO; |
3143 | + |
3144 | + *result = ctl; |
3145 | + return 0; |
3146 | +} |
3147 | + |
3148 | +static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf, |
3149 | + struct vmx_capability *vmx_cap) |
3150 | +{ |
3151 | + u32 vmx_msr_low, vmx_msr_high; |
3152 | + u32 min, opt, min2, opt2; |
3153 | + u32 _pin_based_exec_control = 0; |
3154 | + u32 _cpu_based_exec_control = 0; |
3155 | + u32 _cpu_based_2nd_exec_control = 0; |
3156 | + u32 _vmexit_control = 0; |
3157 | + u32 _vmentry_control = 0; |
3158 | + |
3159 | + memset(vmcs_conf, 0, sizeof(*vmcs_conf)); |
3160 | + min = CPU_BASED_HLT_EXITING | |
3161 | +#ifdef CONFIG_X86_64 |
3162 | + CPU_BASED_CR8_LOAD_EXITING | |
3163 | + CPU_BASED_CR8_STORE_EXITING | |
3164 | +#endif |
3165 | + CPU_BASED_CR3_LOAD_EXITING | |
3166 | + CPU_BASED_CR3_STORE_EXITING | |
3167 | + CPU_BASED_UNCOND_IO_EXITING | |
3168 | + CPU_BASED_MOV_DR_EXITING | |
3169 | + CPU_BASED_USE_TSC_OFFSETTING | |
3170 | + CPU_BASED_MWAIT_EXITING | |
3171 | + CPU_BASED_MONITOR_EXITING | |
3172 | + CPU_BASED_INVLPG_EXITING | |
3173 | + CPU_BASED_RDPMC_EXITING; |
3174 | + |
3175 | + opt = CPU_BASED_TPR_SHADOW | |
3176 | + CPU_BASED_USE_MSR_BITMAPS | |
3177 | + CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; |
3178 | + if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS, |
3179 | + &_cpu_based_exec_control) < 0) |
3180 | + return -EIO; |
3181 | +#ifdef CONFIG_X86_64 |
3182 | + if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW)) |
3183 | + _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING & |
3184 | + ~CPU_BASED_CR8_STORE_EXITING; |
3185 | +#endif |
3186 | + if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) { |
3187 | + min2 = 0; |
3188 | + opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | |
3189 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
3190 | + SECONDARY_EXEC_WBINVD_EXITING | |
3191 | + SECONDARY_EXEC_ENABLE_VPID | |
3192 | + SECONDARY_EXEC_ENABLE_EPT | |
3193 | + SECONDARY_EXEC_UNRESTRICTED_GUEST | |
3194 | + SECONDARY_EXEC_PAUSE_LOOP_EXITING | |
3195 | + SECONDARY_EXEC_DESC | |
3196 | + SECONDARY_EXEC_RDTSCP | |
3197 | + SECONDARY_EXEC_ENABLE_INVPCID | |
3198 | + SECONDARY_EXEC_APIC_REGISTER_VIRT | |
3199 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY | |
3200 | + SECONDARY_EXEC_SHADOW_VMCS | |
3201 | + SECONDARY_EXEC_XSAVES | |
3202 | + SECONDARY_EXEC_RDSEED_EXITING | |
3203 | + SECONDARY_EXEC_RDRAND_EXITING | |
3204 | + SECONDARY_EXEC_ENABLE_PML | |
3205 | + SECONDARY_EXEC_TSC_SCALING | |
3206 | + SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE | |
3207 | + SECONDARY_EXEC_PT_USE_GPA | |
3208 | + SECONDARY_EXEC_PT_CONCEAL_VMX | |
3209 | + SECONDARY_EXEC_ENABLE_VMFUNC | |
3210 | + SECONDARY_EXEC_ENCLS_EXITING; |
3211 | + if (adjust_vmx_controls(min2, opt2, |
3212 | + MSR_IA32_VMX_PROCBASED_CTLS2, |
3213 | + &_cpu_based_2nd_exec_control) < 0) |
3214 | + return -EIO; |
3215 | + } |
3216 | +#ifndef CONFIG_X86_64 |
3217 | + if (!(_cpu_based_2nd_exec_control & |
3218 | + SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) |
3219 | + _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW; |
3220 | +#endif |
3221 | + |
3222 | + if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW)) |
3223 | + _cpu_based_2nd_exec_control &= ~( |
3224 | + SECONDARY_EXEC_APIC_REGISTER_VIRT | |
3225 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
3226 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); |
3227 | + |
3228 | + rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP, |
3229 | + &vmx_cap->ept, &vmx_cap->vpid); |
3230 | + |
3231 | + if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) { |
3232 | + /* CR3 accesses and invlpg don't need to cause VM Exits when EPT |
3233 | + enabled */ |
3234 | + _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING | |
3235 | + CPU_BASED_CR3_STORE_EXITING | |
3236 | + CPU_BASED_INVLPG_EXITING); |
3237 | + } else if (vmx_cap->ept) { |
3238 | + vmx_cap->ept = 0; |
3239 | + pr_warn_once("EPT CAP should not exist if not support " |
3240 | + "1-setting enable EPT VM-execution control\n"); |
3241 | + } |
3242 | + if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) && |
3243 | + vmx_cap->vpid) { |
3244 | + vmx_cap->vpid = 0; |
3245 | + pr_warn_once("VPID CAP should not exist if not support " |
3246 | + "1-setting enable VPID VM-execution control\n"); |
3247 | + } |
3248 | + |
3249 | + min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT; |
3250 | +#ifdef CONFIG_X86_64 |
3251 | + min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; |
3252 | +#endif |
3253 | + opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | |
3254 | + VM_EXIT_LOAD_IA32_PAT | |
3255 | + VM_EXIT_LOAD_IA32_EFER | |
3256 | + VM_EXIT_CLEAR_BNDCFGS | |
3257 | + VM_EXIT_PT_CONCEAL_PIP | |
3258 | + VM_EXIT_CLEAR_IA32_RTIT_CTL; |
3259 | + if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS, |
3260 | + &_vmexit_control) < 0) |
3261 | + return -EIO; |
3262 | + |
3263 | + min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING; |
3264 | + opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR | |
3265 | + PIN_BASED_VMX_PREEMPTION_TIMER; |
3266 | + if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS, |
3267 | + &_pin_based_exec_control) < 0) |
3268 | + return -EIO; |
3269 | + |
3270 | + if (cpu_has_broken_vmx_preemption_timer()) |
3271 | + _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER; |
3272 | + if (!(_cpu_based_2nd_exec_control & |
3273 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)) |
3274 | + _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR; |
3275 | + |
3276 | + min = VM_ENTRY_LOAD_DEBUG_CONTROLS; |
3277 | + opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | |
3278 | + VM_ENTRY_LOAD_IA32_PAT | |
3279 | + VM_ENTRY_LOAD_IA32_EFER | |
3280 | + VM_ENTRY_LOAD_BNDCFGS | |
3281 | + VM_ENTRY_PT_CONCEAL_PIP | |
3282 | + VM_ENTRY_LOAD_IA32_RTIT_CTL; |
3283 | + if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS, |
3284 | + &_vmentry_control) < 0) |
3285 | + return -EIO; |
3286 | + |
3287 | + /* |
3288 | + * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they |
3289 | + * can't be used due to an errata where VM Exit may incorrectly clear |
3290 | + * IA32_PERF_GLOBAL_CTRL[34:32]. Workaround the errata by using the |
3291 | + * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL. |
3292 | + */ |
3293 | + if (boot_cpu_data.x86 == 0x6) { |
3294 | + switch (boot_cpu_data.x86_model) { |
3295 | + case 26: /* AAK155 */ |
3296 | + case 30: /* AAP115 */ |
3297 | + case 37: /* AAT100 */ |
3298 | + case 44: /* BC86,AAY89,BD102 */ |
3299 | + case 46: /* BA97 */ |
3300 | + _vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL; |
3301 | + _vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL; |
3302 | + pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL " |
3303 | + "does not work properly. Using workaround\n"); |
3304 | + break; |
3305 | + default: |
3306 | + break; |
3307 | + } |
3308 | + } |
3309 | + |
3310 | + |
3311 | + rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high); |
3312 | + |
3313 | + /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */ |
3314 | + if ((vmx_msr_high & 0x1fff) > PAGE_SIZE) |
3315 | + return -EIO; |
3316 | + |
3317 | +#ifdef CONFIG_X86_64 |
3318 | + /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */ |
3319 | + if (vmx_msr_high & (1u<<16)) |
3320 | + return -EIO; |
3321 | +#endif |
3322 | + |
3323 | + /* Require Write-Back (WB) memory type for VMCS accesses. */ |
3324 | + if (((vmx_msr_high >> 18) & 15) != 6) |
3325 | + return -EIO; |
3326 | + |
3327 | + vmcs_conf->size = vmx_msr_high & 0x1fff; |
3328 | + vmcs_conf->order = get_order(vmcs_conf->size); |
3329 | + vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff; |
3330 | + |
3331 | + vmcs_conf->revision_id = vmx_msr_low; |
3332 | + |
3333 | + vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control; |
3334 | + vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control; |
3335 | + vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control; |
3336 | + vmcs_conf->vmexit_ctrl = _vmexit_control; |
3337 | + vmcs_conf->vmentry_ctrl = _vmentry_control; |
3338 | + |
3339 | + if (static_branch_unlikely(&enable_evmcs)) |
3340 | + evmcs_sanitize_exec_ctrls(vmcs_conf); |
3341 | + |
3342 | + return 0; |
3343 | +} |
3344 | + |
3345 | +struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags) |
3346 | +{ |
3347 | + int node = cpu_to_node(cpu); |
3348 | + struct page *pages; |
3349 | + struct vmcs *vmcs; |
3350 | + |
3351 | + pages = __alloc_pages_node(node, flags, vmcs_config.order); |
3352 | + if (!pages) |
3353 | + return NULL; |
3354 | + vmcs = page_address(pages); |
3355 | + memset(vmcs, 0, vmcs_config.size); |
3356 | + |
3357 | + /* KVM supports Enlightened VMCS v1 only */ |
3358 | + if (static_branch_unlikely(&enable_evmcs)) |
3359 | + vmcs->hdr.revision_id = KVM_EVMCS_VERSION; |
3360 | + else |
3361 | + vmcs->hdr.revision_id = vmcs_config.revision_id; |
3362 | + |
3363 | + if (shadow) |
3364 | + vmcs->hdr.shadow_vmcs = 1; |
3365 | + return vmcs; |
3366 | +} |
3367 | + |
3368 | +void free_vmcs(struct vmcs *vmcs) |
3369 | +{ |
3370 | + free_pages((unsigned long)vmcs, vmcs_config.order); |
3371 | +} |
3372 | + |
3373 | +/* |
3374 | + * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded |
3375 | + */ |
3376 | +void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs) |
3377 | +{ |
3378 | + if (!loaded_vmcs->vmcs) |
3379 | + return; |
3380 | + loaded_vmcs_clear(loaded_vmcs); |
3381 | + free_vmcs(loaded_vmcs->vmcs); |
3382 | + loaded_vmcs->vmcs = NULL; |
3383 | + if (loaded_vmcs->msr_bitmap) |
3384 | + free_page((unsigned long)loaded_vmcs->msr_bitmap); |
3385 | + WARN_ON(loaded_vmcs->shadow_vmcs != NULL); |
3386 | +} |
3387 | + |
3388 | +int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs) |
3389 | +{ |
3390 | + loaded_vmcs->vmcs = alloc_vmcs(false); |
3391 | + if (!loaded_vmcs->vmcs) |
3392 | + return -ENOMEM; |
3393 | + |
3394 | + loaded_vmcs->shadow_vmcs = NULL; |
3395 | + loaded_vmcs->hv_timer_soft_disabled = false; |
3396 | + loaded_vmcs_init(loaded_vmcs); |
3397 | + |
3398 | + if (cpu_has_vmx_msr_bitmap()) { |
3399 | + loaded_vmcs->msr_bitmap = (unsigned long *) |
3400 | + __get_free_page(GFP_KERNEL_ACCOUNT); |
3401 | + if (!loaded_vmcs->msr_bitmap) |
3402 | + goto out_vmcs; |
3403 | + memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE); |
3404 | + |
3405 | + if (IS_ENABLED(CONFIG_HYPERV) && |
3406 | + static_branch_unlikely(&enable_evmcs) && |
3407 | + (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) { |
3408 | + struct hv_enlightened_vmcs *evmcs = |
3409 | + (struct hv_enlightened_vmcs *)loaded_vmcs->vmcs; |
3410 | + |
3411 | + evmcs->hv_enlightenments_control.msr_bitmap = 1; |
3412 | + } |
3413 | + } |
3414 | + |
3415 | + memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state)); |
3416 | + memset(&loaded_vmcs->controls_shadow, 0, |
3417 | + sizeof(struct vmcs_controls_shadow)); |
3418 | + |
3419 | + return 0; |
3420 | + |
3421 | +out_vmcs: |
3422 | + free_loaded_vmcs(loaded_vmcs); |
3423 | + return -ENOMEM; |
3424 | +} |
3425 | + |
3426 | +static void free_kvm_area(void) |
3427 | +{ |
3428 | + int cpu; |
3429 | + |
3430 | + for_each_possible_cpu(cpu) { |
3431 | + free_vmcs(per_cpu(vmxarea, cpu)); |
3432 | + per_cpu(vmxarea, cpu) = NULL; |
3433 | + } |
3434 | +} |
3435 | + |
3436 | +static __init int alloc_kvm_area(void) |
3437 | +{ |
3438 | + int cpu; |
3439 | + |
3440 | + for_each_possible_cpu(cpu) { |
3441 | + struct vmcs *vmcs; |
3442 | + |
3443 | + vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL); |
3444 | + if (!vmcs) { |
3445 | + free_kvm_area(); |
3446 | + return -ENOMEM; |
3447 | + } |
3448 | + |
3449 | + /* |
3450 | + * When eVMCS is enabled, alloc_vmcs_cpu() sets |
3451 | + * vmcs->revision_id to KVM_EVMCS_VERSION instead of |
3452 | + * revision_id reported by MSR_IA32_VMX_BASIC. |
3453 | + * |
3454 | + * However, even though not explicitly documented by |
3455 | + * TLFS, VMXArea passed as VMXON argument should |
3456 | + * still be marked with revision_id reported by |
3457 | + * physical CPU. |
3458 | + */ |
3459 | + if (static_branch_unlikely(&enable_evmcs)) |
3460 | + vmcs->hdr.revision_id = vmcs_config.revision_id; |
3461 | + |
3462 | + per_cpu(vmxarea, cpu) = vmcs; |
3463 | + } |
3464 | + return 0; |
3465 | +} |
3466 | + |
3467 | +static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg, |
3468 | + struct kvm_segment *save) |
3469 | +{ |
3470 | + if (!emulate_invalid_guest_state) { |
3471 | + /* |
3472 | + * CS and SS RPL should be equal during guest entry according |
3473 | + * to VMX spec, but in reality it is not always so. Since vcpu |
3474 | + * is in the middle of the transition from real mode to |
3475 | + * protected mode it is safe to assume that RPL 0 is a good |
3476 | + * default value. |
3477 | + */ |
3478 | + if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS) |
3479 | + save->selector &= ~SEGMENT_RPL_MASK; |
3480 | + save->dpl = save->selector & SEGMENT_RPL_MASK; |
3481 | + save->s = 1; |
3482 | + } |
3483 | + vmx_set_segment(vcpu, save, seg); |
3484 | +} |
3485 | + |
3486 | +static void enter_pmode(struct kvm_vcpu *vcpu) |
3487 | +{ |
3488 | + unsigned long flags; |
3489 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3490 | + |
3491 | + /* |
3492 | + * Update real mode segment cache. It may be not up-to-date if sement |
3493 | + * register was written while vcpu was in a guest mode. |
3494 | + */ |
3495 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES); |
3496 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS); |
3497 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS); |
3498 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS); |
3499 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS); |
3500 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS); |
3501 | + |
3502 | + vmx->rmode.vm86_active = 0; |
3503 | + |
3504 | + vmx_segment_cache_clear(vmx); |
3505 | + |
3506 | + vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR); |
3507 | + |
3508 | + flags = vmcs_readl(GUEST_RFLAGS); |
3509 | + flags &= RMODE_GUEST_OWNED_EFLAGS_BITS; |
3510 | + flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS; |
3511 | + vmcs_writel(GUEST_RFLAGS, flags); |
3512 | + |
3513 | + vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) | |
3514 | + (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME)); |
3515 | + |
3516 | + update_exception_bitmap(vcpu); |
3517 | + |
3518 | + fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]); |
3519 | + fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]); |
3520 | + fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]); |
3521 | + fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]); |
3522 | + fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]); |
3523 | + fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]); |
3524 | +} |
3525 | + |
3526 | +static void fix_rmode_seg(int seg, struct kvm_segment *save) |
3527 | +{ |
3528 | + const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
3529 | + struct kvm_segment var = *save; |
3530 | + |
3531 | + var.dpl = 0x3; |
3532 | + if (seg == VCPU_SREG_CS) |
3533 | + var.type = 0x3; |
3534 | + |
3535 | + if (!emulate_invalid_guest_state) { |
3536 | + var.selector = var.base >> 4; |
3537 | + var.base = var.base & 0xffff0; |
3538 | + var.limit = 0xffff; |
3539 | + var.g = 0; |
3540 | + var.db = 0; |
3541 | + var.present = 1; |
3542 | + var.s = 1; |
3543 | + var.l = 0; |
3544 | + var.unusable = 0; |
3545 | + var.type = 0x3; |
3546 | + var.avl = 0; |
3547 | + if (save->base & 0xf) |
3548 | + printk_once(KERN_WARNING "kvm: segment base is not " |
3549 | + "paragraph aligned when entering " |
3550 | + "protected mode (seg=%d)", seg); |
3551 | + } |
3552 | + |
3553 | + vmcs_write16(sf->selector, var.selector); |
3554 | + vmcs_writel(sf->base, var.base); |
3555 | + vmcs_write32(sf->limit, var.limit); |
3556 | + vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var)); |
3557 | +} |
3558 | + |
3559 | +static void enter_rmode(struct kvm_vcpu *vcpu) |
3560 | +{ |
3561 | + unsigned long flags; |
3562 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3563 | + struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm); |
3564 | + |
3565 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR); |
3566 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES); |
3567 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS); |
3568 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS); |
3569 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS); |
3570 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS); |
3571 | + vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS); |
3572 | + |
3573 | + vmx->rmode.vm86_active = 1; |
3574 | + |
3575 | + /* |
3576 | + * Very old userspace does not call KVM_SET_TSS_ADDR before entering |
3577 | + * vcpu. Warn the user that an update is overdue. |
3578 | + */ |
3579 | + if (!kvm_vmx->tss_addr) |
3580 | + printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be " |
3581 | + "called before entering vcpu\n"); |
3582 | + |
3583 | + vmx_segment_cache_clear(vmx); |
3584 | + |
3585 | + vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr); |
3586 | + vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1); |
3587 | + vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); |
3588 | + |
3589 | + flags = vmcs_readl(GUEST_RFLAGS); |
3590 | + vmx->rmode.save_rflags = flags; |
3591 | + |
3592 | + flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
3593 | + |
3594 | + vmcs_writel(GUEST_RFLAGS, flags); |
3595 | + vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME); |
3596 | + update_exception_bitmap(vcpu); |
3597 | + |
3598 | + fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]); |
3599 | + fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]); |
3600 | + fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]); |
3601 | + fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]); |
3602 | + fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]); |
3603 | + fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]); |
3604 | + |
3605 | + kvm_mmu_reset_context(vcpu); |
3606 | +} |
3607 | + |
3608 | +void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer) |
3609 | +{ |
3610 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3611 | + struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER); |
3612 | + |
3613 | + if (!msr) |
3614 | + return; |
3615 | + |
3616 | + vcpu->arch.efer = efer; |
3617 | + if (efer & EFER_LMA) { |
3618 | + vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
3619 | + msr->data = efer; |
3620 | + } else { |
3621 | + vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
3622 | + |
3623 | + msr->data = efer & ~EFER_LME; |
3624 | + } |
3625 | + setup_msrs(vmx); |
3626 | +} |
3627 | + |
3628 | +#ifdef CONFIG_X86_64 |
3629 | + |
3630 | +static void enter_lmode(struct kvm_vcpu *vcpu) |
3631 | +{ |
3632 | + u32 guest_tr_ar; |
3633 | + |
3634 | + vmx_segment_cache_clear(to_vmx(vcpu)); |
3635 | + |
3636 | + guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES); |
3637 | + if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) { |
3638 | + pr_debug_ratelimited("%s: tss fixup for long mode. \n", |
3639 | + __func__); |
3640 | + vmcs_write32(GUEST_TR_AR_BYTES, |
3641 | + (guest_tr_ar & ~VMX_AR_TYPE_MASK) |
3642 | + | VMX_AR_TYPE_BUSY_64_TSS); |
3643 | + } |
3644 | + vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA); |
3645 | +} |
3646 | + |
3647 | +static void exit_lmode(struct kvm_vcpu *vcpu) |
3648 | +{ |
3649 | + vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
3650 | + vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA); |
3651 | +} |
3652 | + |
3653 | +#endif |
3654 | + |
3655 | +static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr) |
3656 | +{ |
3657 | + int vpid = to_vmx(vcpu)->vpid; |
3658 | + |
3659 | + if (!vpid_sync_vcpu_addr(vpid, addr)) |
3660 | + vpid_sync_context(vpid); |
3661 | + |
3662 | + /* |
3663 | + * If VPIDs are not supported or enabled, then the above is a no-op. |
3664 | + * But we don't really need a TLB flush in that case anyway, because |
3665 | + * each VM entry/exit includes an implicit flush when VPID is 0. |
3666 | + */ |
3667 | +} |
3668 | + |
3669 | +static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) |
3670 | +{ |
3671 | + ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits; |
3672 | + |
3673 | + vcpu->arch.cr0 &= ~cr0_guest_owned_bits; |
3674 | + vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits; |
3675 | +} |
3676 | + |
3677 | +static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
3678 | +{ |
3679 | + ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits; |
3680 | + |
3681 | + vcpu->arch.cr4 &= ~cr4_guest_owned_bits; |
3682 | + vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits; |
3683 | +} |
3684 | + |
3685 | +static void ept_load_pdptrs(struct kvm_vcpu *vcpu) |
3686 | +{ |
3687 | + struct kvm_mmu *mmu = vcpu->arch.walk_mmu; |
3688 | + |
3689 | + if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR)) |
3690 | + return; |
3691 | + |
3692 | + if (is_pae_paging(vcpu)) { |
3693 | + vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]); |
3694 | + vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]); |
3695 | + vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]); |
3696 | + vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]); |
3697 | + } |
3698 | +} |
3699 | + |
3700 | +void ept_save_pdptrs(struct kvm_vcpu *vcpu) |
3701 | +{ |
3702 | + struct kvm_mmu *mmu = vcpu->arch.walk_mmu; |
3703 | + |
3704 | + if (is_pae_paging(vcpu)) { |
3705 | + mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0); |
3706 | + mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1); |
3707 | + mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2); |
3708 | + mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3); |
3709 | + } |
3710 | + |
3711 | + kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR); |
3712 | +} |
3713 | + |
3714 | +static void ept_update_paging_mode_cr0(unsigned long *hw_cr0, |
3715 | + unsigned long cr0, |
3716 | + struct kvm_vcpu *vcpu) |
3717 | +{ |
3718 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3719 | + |
3720 | + if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3)) |
3721 | + vmx_cache_reg(vcpu, VCPU_EXREG_CR3); |
3722 | + if (!(cr0 & X86_CR0_PG)) { |
3723 | + /* From paging/starting to nonpaging */ |
3724 | + exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING | |
3725 | + CPU_BASED_CR3_STORE_EXITING); |
3726 | + vcpu->arch.cr0 = cr0; |
3727 | + vmx_set_cr4(vcpu, kvm_read_cr4(vcpu)); |
3728 | + } else if (!is_paging(vcpu)) { |
3729 | + /* From nonpaging to paging */ |
3730 | + exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING | |
3731 | + CPU_BASED_CR3_STORE_EXITING); |
3732 | + vcpu->arch.cr0 = cr0; |
3733 | + vmx_set_cr4(vcpu, kvm_read_cr4(vcpu)); |
3734 | + } |
3735 | + |
3736 | + if (!(cr0 & X86_CR0_WP)) |
3737 | + *hw_cr0 &= ~X86_CR0_WP; |
3738 | +} |
3739 | + |
3740 | +void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
3741 | +{ |
3742 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3743 | + unsigned long hw_cr0; |
3744 | + |
3745 | + hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF); |
3746 | + if (enable_unrestricted_guest) |
3747 | + hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST; |
3748 | + else { |
3749 | + hw_cr0 |= KVM_VM_CR0_ALWAYS_ON; |
3750 | + |
3751 | + if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE)) |
3752 | + enter_pmode(vcpu); |
3753 | + |
3754 | + if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE)) |
3755 | + enter_rmode(vcpu); |
3756 | + } |
3757 | + |
3758 | +#ifdef CONFIG_X86_64 |
3759 | + if (vcpu->arch.efer & EFER_LME) { |
3760 | + if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) |
3761 | + enter_lmode(vcpu); |
3762 | + if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) |
3763 | + exit_lmode(vcpu); |
3764 | + } |
3765 | +#endif |
3766 | + |
3767 | + if (enable_ept && !enable_unrestricted_guest) |
3768 | + ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu); |
3769 | + |
3770 | + vmcs_writel(CR0_READ_SHADOW, cr0); |
3771 | + vmcs_writel(GUEST_CR0, hw_cr0); |
3772 | + vcpu->arch.cr0 = cr0; |
3773 | + |
3774 | + /* depends on vcpu->arch.cr0 to be set to a new value */ |
3775 | + vmx->emulation_required = emulation_required(vcpu); |
3776 | +} |
3777 | + |
3778 | +static int get_ept_level(struct kvm_vcpu *vcpu) |
3779 | +{ |
3780 | + if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48)) |
3781 | + return 5; |
3782 | + return 4; |
3783 | +} |
3784 | + |
3785 | +u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa) |
3786 | +{ |
3787 | + u64 eptp = VMX_EPTP_MT_WB; |
3788 | + |
3789 | + eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4; |
3790 | + |
3791 | + if (enable_ept_ad_bits && |
3792 | + (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu))) |
3793 | + eptp |= VMX_EPTP_AD_ENABLE_BIT; |
3794 | + eptp |= (root_hpa & PAGE_MASK); |
3795 | + |
3796 | + return eptp; |
3797 | +} |
3798 | + |
3799 | +void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
3800 | +{ |
3801 | + struct kvm *kvm = vcpu->kvm; |
3802 | + bool update_guest_cr3 = true; |
3803 | + unsigned long guest_cr3; |
3804 | + u64 eptp; |
3805 | + |
3806 | + guest_cr3 = cr3; |
3807 | + if (enable_ept) { |
3808 | + eptp = construct_eptp(vcpu, cr3); |
3809 | + vmcs_write64(EPT_POINTER, eptp); |
3810 | + |
3811 | + if (kvm_x86_ops->tlb_remote_flush) { |
3812 | + spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock); |
3813 | + to_vmx(vcpu)->ept_pointer = eptp; |
3814 | + to_kvm_vmx(kvm)->ept_pointers_match |
3815 | + = EPT_POINTERS_CHECK; |
3816 | + spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock); |
3817 | + } |
3818 | + |
3819 | + /* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */ |
3820 | + if (is_guest_mode(vcpu)) |
3821 | + update_guest_cr3 = false; |
3822 | + else if (!enable_unrestricted_guest && !is_paging(vcpu)) |
3823 | + guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr; |
3824 | + else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail)) |
3825 | + guest_cr3 = vcpu->arch.cr3; |
3826 | + else /* vmcs01.GUEST_CR3 is already up-to-date. */ |
3827 | + update_guest_cr3 = false; |
3828 | + ept_load_pdptrs(vcpu); |
3829 | + } |
3830 | + |
3831 | + if (update_guest_cr3) |
3832 | + vmcs_writel(GUEST_CR3, guest_cr3); |
3833 | +} |
3834 | + |
3835 | +int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
3836 | +{ |
3837 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3838 | + /* |
3839 | + * Pass through host's Machine Check Enable value to hw_cr4, which |
3840 | + * is in force while we are in guest mode. Do not let guests control |
3841 | + * this bit, even if host CR4.MCE == 0. |
3842 | + */ |
3843 | + unsigned long hw_cr4; |
3844 | + |
3845 | + hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE); |
3846 | + if (enable_unrestricted_guest) |
3847 | + hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST; |
3848 | + else if (vmx->rmode.vm86_active) |
3849 | + hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON; |
3850 | + else |
3851 | + hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON; |
3852 | + |
3853 | + if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) { |
3854 | + if (cr4 & X86_CR4_UMIP) { |
3855 | + secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC); |
3856 | + hw_cr4 &= ~X86_CR4_UMIP; |
3857 | + } else if (!is_guest_mode(vcpu) || |
3858 | + !nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) { |
3859 | + secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC); |
3860 | + } |
3861 | + } |
3862 | + |
3863 | + if (cr4 & X86_CR4_VMXE) { |
3864 | + /* |
3865 | + * To use VMXON (and later other VMX instructions), a guest |
3866 | + * must first be able to turn on cr4.VMXE (see handle_vmon()). |
3867 | + * So basically the check on whether to allow nested VMX |
3868 | + * is here. We operate under the default treatment of SMM, |
3869 | + * so VMX cannot be enabled under SMM. |
3870 | + */ |
3871 | + if (!nested_vmx_allowed(vcpu) || is_smm(vcpu)) |
3872 | + return 1; |
3873 | + } |
3874 | + |
3875 | + if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4)) |
3876 | + return 1; |
3877 | + |
3878 | + vcpu->arch.cr4 = cr4; |
3879 | + |
3880 | + if (!enable_unrestricted_guest) { |
3881 | + if (enable_ept) { |
3882 | + if (!is_paging(vcpu)) { |
3883 | + hw_cr4 &= ~X86_CR4_PAE; |
3884 | + hw_cr4 |= X86_CR4_PSE; |
3885 | + } else if (!(cr4 & X86_CR4_PAE)) { |
3886 | + hw_cr4 &= ~X86_CR4_PAE; |
3887 | + } |
3888 | + } |
3889 | + |
3890 | + /* |
3891 | + * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in |
3892 | + * hardware. To emulate this behavior, SMEP/SMAP/PKU needs |
3893 | + * to be manually disabled when guest switches to non-paging |
3894 | + * mode. |
3895 | + * |
3896 | + * If !enable_unrestricted_guest, the CPU is always running |
3897 | + * with CR0.PG=1 and CR4 needs to be modified. |
3898 | + * If enable_unrestricted_guest, the CPU automatically |
3899 | + * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0. |
3900 | + */ |
3901 | + if (!is_paging(vcpu)) |
3902 | + hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE); |
3903 | + } |
3904 | + |
3905 | + vmcs_writel(CR4_READ_SHADOW, cr4); |
3906 | + vmcs_writel(GUEST_CR4, hw_cr4); |
3907 | + return 0; |
3908 | +} |
3909 | + |
3910 | +void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg) |
3911 | +{ |
3912 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3913 | + u32 ar; |
3914 | + |
3915 | + if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) { |
3916 | + *var = vmx->rmode.segs[seg]; |
3917 | + if (seg == VCPU_SREG_TR |
3918 | + || var->selector == vmx_read_guest_seg_selector(vmx, seg)) |
3919 | + return; |
3920 | + var->base = vmx_read_guest_seg_base(vmx, seg); |
3921 | + var->selector = vmx_read_guest_seg_selector(vmx, seg); |
3922 | + return; |
3923 | + } |
3924 | + var->base = vmx_read_guest_seg_base(vmx, seg); |
3925 | + var->limit = vmx_read_guest_seg_limit(vmx, seg); |
3926 | + var->selector = vmx_read_guest_seg_selector(vmx, seg); |
3927 | + ar = vmx_read_guest_seg_ar(vmx, seg); |
3928 | + var->unusable = (ar >> 16) & 1; |
3929 | + var->type = ar & 15; |
3930 | + var->s = (ar >> 4) & 1; |
3931 | + var->dpl = (ar >> 5) & 3; |
3932 | + /* |
3933 | + * Some userspaces do not preserve unusable property. Since usable |
3934 | + * segment has to be present according to VMX spec we can use present |
3935 | + * property to amend userspace bug by making unusable segment always |
3936 | + * nonpresent. vmx_segment_access_rights() already marks nonpresent |
3937 | + * segment as unusable. |
3938 | + */ |
3939 | + var->present = !var->unusable; |
3940 | + var->avl = (ar >> 12) & 1; |
3941 | + var->l = (ar >> 13) & 1; |
3942 | + var->db = (ar >> 14) & 1; |
3943 | + var->g = (ar >> 15) & 1; |
3944 | +} |
3945 | + |
3946 | +static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg) |
3947 | +{ |
3948 | + struct kvm_segment s; |
3949 | + |
3950 | + if (to_vmx(vcpu)->rmode.vm86_active) { |
3951 | + vmx_get_segment(vcpu, &s, seg); |
3952 | + return s.base; |
3953 | + } |
3954 | + return vmx_read_guest_seg_base(to_vmx(vcpu), seg); |
3955 | +} |
3956 | + |
3957 | +int vmx_get_cpl(struct kvm_vcpu *vcpu) |
3958 | +{ |
3959 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3960 | + |
3961 | + if (unlikely(vmx->rmode.vm86_active)) |
3962 | + return 0; |
3963 | + else { |
3964 | + int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS); |
3965 | + return VMX_AR_DPL(ar); |
3966 | + } |
3967 | +} |
3968 | + |
3969 | +static u32 vmx_segment_access_rights(struct kvm_segment *var) |
3970 | +{ |
3971 | + u32 ar; |
3972 | + |
3973 | + if (var->unusable || !var->present) |
3974 | + ar = 1 << 16; |
3975 | + else { |
3976 | + ar = var->type & 15; |
3977 | + ar |= (var->s & 1) << 4; |
3978 | + ar |= (var->dpl & 3) << 5; |
3979 | + ar |= (var->present & 1) << 7; |
3980 | + ar |= (var->avl & 1) << 12; |
3981 | + ar |= (var->l & 1) << 13; |
3982 | + ar |= (var->db & 1) << 14; |
3983 | + ar |= (var->g & 1) << 15; |
3984 | + } |
3985 | + |
3986 | + return ar; |
3987 | +} |
3988 | + |
3989 | +void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg) |
3990 | +{ |
3991 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
3992 | + const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
3993 | + |
3994 | + vmx_segment_cache_clear(vmx); |
3995 | + |
3996 | + if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) { |
3997 | + vmx->rmode.segs[seg] = *var; |
3998 | + if (seg == VCPU_SREG_TR) |
3999 | + vmcs_write16(sf->selector, var->selector); |
4000 | + else if (var->s) |
4001 | + fix_rmode_seg(seg, &vmx->rmode.segs[seg]); |
4002 | + goto out; |
4003 | + } |
4004 | + |
4005 | + vmcs_writel(sf->base, var->base); |
4006 | + vmcs_write32(sf->limit, var->limit); |
4007 | + vmcs_write16(sf->selector, var->selector); |
4008 | + |
4009 | + /* |
4010 | + * Fix the "Accessed" bit in AR field of segment registers for older |
4011 | + * qemu binaries. |
4012 | + * IA32 arch specifies that at the time of processor reset the |
4013 | + * "Accessed" bit in the AR field of segment registers is 1. And qemu |
4014 | + * is setting it to 0 in the userland code. This causes invalid guest |
4015 | + * state vmexit when "unrestricted guest" mode is turned on. |
4016 | + * Fix for this setup issue in cpu_reset is being pushed in the qemu |
4017 | + * tree. Newer qemu binaries with that qemu fix would not need this |
4018 | + * kvm hack. |
4019 | + */ |
4020 | + if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR)) |
4021 | + var->type |= 0x1; /* Accessed */ |
4022 | + |
4023 | + vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var)); |
4024 | + |
4025 | +out: |
4026 | + vmx->emulation_required = emulation_required(vcpu); |
4027 | +} |
4028 | + |
4029 | +static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
4030 | +{ |
4031 | + u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS); |
4032 | + |
4033 | + *db = (ar >> 14) & 1; |
4034 | + *l = (ar >> 13) & 1; |
4035 | +} |
4036 | + |
4037 | +static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
4038 | +{ |
4039 | + dt->size = vmcs_read32(GUEST_IDTR_LIMIT); |
4040 | + dt->address = vmcs_readl(GUEST_IDTR_BASE); |
4041 | +} |
4042 | + |
4043 | +static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
4044 | +{ |
4045 | + vmcs_write32(GUEST_IDTR_LIMIT, dt->size); |
4046 | + vmcs_writel(GUEST_IDTR_BASE, dt->address); |
4047 | +} |
4048 | + |
4049 | +static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
4050 | +{ |
4051 | + dt->size = vmcs_read32(GUEST_GDTR_LIMIT); |
4052 | + dt->address = vmcs_readl(GUEST_GDTR_BASE); |
4053 | +} |
4054 | + |
4055 | +static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
4056 | +{ |
4057 | + vmcs_write32(GUEST_GDTR_LIMIT, dt->size); |
4058 | + vmcs_writel(GUEST_GDTR_BASE, dt->address); |
4059 | +} |
4060 | + |
4061 | +static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg) |
4062 | +{ |
4063 | + struct kvm_segment var; |
4064 | + u32 ar; |
4065 | + |
4066 | + vmx_get_segment(vcpu, &var, seg); |
4067 | + var.dpl = 0x3; |
4068 | + if (seg == VCPU_SREG_CS) |
4069 | + var.type = 0x3; |
4070 | + ar = vmx_segment_access_rights(&var); |
4071 | + |
4072 | + if (var.base != (var.selector << 4)) |
4073 | + return false; |
4074 | + if (var.limit != 0xffff) |
4075 | + return false; |
4076 | + if (ar != 0xf3) |
4077 | + return false; |
4078 | + |
4079 | + return true; |
4080 | +} |
4081 | + |
4082 | +static bool code_segment_valid(struct kvm_vcpu *vcpu) |
4083 | +{ |
4084 | + struct kvm_segment cs; |
4085 | + unsigned int cs_rpl; |
4086 | + |
4087 | + vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); |
4088 | + cs_rpl = cs.selector & SEGMENT_RPL_MASK; |
4089 | + |
4090 | + if (cs.unusable) |
4091 | + return false; |
4092 | + if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK)) |
4093 | + return false; |
4094 | + if (!cs.s) |
4095 | + return false; |
4096 | + if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) { |
4097 | + if (cs.dpl > cs_rpl) |
4098 | + return false; |
4099 | + } else { |
4100 | + if (cs.dpl != cs_rpl) |
4101 | + return false; |
4102 | + } |
4103 | + if (!cs.present) |
4104 | + return false; |
4105 | + |
4106 | + /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */ |
4107 | + return true; |
4108 | +} |
4109 | + |
4110 | +static bool stack_segment_valid(struct kvm_vcpu *vcpu) |
4111 | +{ |
4112 | + struct kvm_segment ss; |
4113 | + unsigned int ss_rpl; |
4114 | + |
4115 | + vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); |
4116 | + ss_rpl = ss.selector & SEGMENT_RPL_MASK; |
4117 | + |
4118 | + if (ss.unusable) |
4119 | + return true; |
4120 | + if (ss.type != 3 && ss.type != 7) |
4121 | + return false; |
4122 | + if (!ss.s) |
4123 | + return false; |
4124 | + if (ss.dpl != ss_rpl) /* DPL != RPL */ |
4125 | + return false; |
4126 | + if (!ss.present) |
4127 | + return false; |
4128 | + |
4129 | + return true; |
4130 | +} |
4131 | + |
4132 | +static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg) |
4133 | +{ |
4134 | + struct kvm_segment var; |
4135 | + unsigned int rpl; |
4136 | + |
4137 | + vmx_get_segment(vcpu, &var, seg); |
4138 | + rpl = var.selector & SEGMENT_RPL_MASK; |
4139 | + |
4140 | + if (var.unusable) |
4141 | + return true; |
4142 | + if (!var.s) |
4143 | + return false; |
4144 | + if (!var.present) |
4145 | + return false; |
4146 | + if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) { |
4147 | + if (var.dpl < rpl) /* DPL < RPL */ |
4148 | + return false; |
4149 | + } |
4150 | + |
4151 | + /* TODO: Add other members to kvm_segment_field to allow checking for other access |
4152 | + * rights flags |
4153 | + */ |
4154 | + return true; |
4155 | +} |
4156 | + |
4157 | +static bool tr_valid(struct kvm_vcpu *vcpu) |
4158 | +{ |
4159 | + struct kvm_segment tr; |
4160 | + |
4161 | + vmx_get_segment(vcpu, &tr, VCPU_SREG_TR); |
4162 | + |
4163 | + if (tr.unusable) |
4164 | + return false; |
4165 | + if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */ |
4166 | + return false; |
4167 | + if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */ |
4168 | + return false; |
4169 | + if (!tr.present) |
4170 | + return false; |
4171 | + |
4172 | + return true; |
4173 | +} |
4174 | + |
4175 | +static bool ldtr_valid(struct kvm_vcpu *vcpu) |
4176 | +{ |
4177 | + struct kvm_segment ldtr; |
4178 | + |
4179 | + vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR); |
4180 | + |
4181 | + if (ldtr.unusable) |
4182 | + return true; |
4183 | + if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */ |
4184 | + return false; |
4185 | + if (ldtr.type != 2) |
4186 | + return false; |
4187 | + if (!ldtr.present) |
4188 | + return false; |
4189 | + |
4190 | + return true; |
4191 | +} |
4192 | + |
4193 | +static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu) |
4194 | +{ |
4195 | + struct kvm_segment cs, ss; |
4196 | + |
4197 | + vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); |
4198 | + vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); |
4199 | + |
4200 | + return ((cs.selector & SEGMENT_RPL_MASK) == |
4201 | + (ss.selector & SEGMENT_RPL_MASK)); |
4202 | +} |
4203 | + |
4204 | +/* |
4205 | + * Check if guest state is valid. Returns true if valid, false if |
4206 | + * not. |
4207 | + * We assume that registers are always usable |
4208 | + */ |
4209 | +static bool guest_state_valid(struct kvm_vcpu *vcpu) |
4210 | +{ |
4211 | + if (enable_unrestricted_guest) |
4212 | + return true; |
4213 | + |
4214 | + /* real mode guest state checks */ |
4215 | + if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) { |
4216 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_CS)) |
4217 | + return false; |
4218 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_SS)) |
4219 | + return false; |
4220 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_DS)) |
4221 | + return false; |
4222 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_ES)) |
4223 | + return false; |
4224 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_FS)) |
4225 | + return false; |
4226 | + if (!rmode_segment_valid(vcpu, VCPU_SREG_GS)) |
4227 | + return false; |
4228 | + } else { |
4229 | + /* protected mode guest state checks */ |
4230 | + if (!cs_ss_rpl_check(vcpu)) |
4231 | + return false; |
4232 | + if (!code_segment_valid(vcpu)) |
4233 | + return false; |
4234 | + if (!stack_segment_valid(vcpu)) |
4235 | + return false; |
4236 | + if (!data_segment_valid(vcpu, VCPU_SREG_DS)) |
4237 | + return false; |
4238 | + if (!data_segment_valid(vcpu, VCPU_SREG_ES)) |
4239 | + return false; |
4240 | + if (!data_segment_valid(vcpu, VCPU_SREG_FS)) |
4241 | + return false; |
4242 | + if (!data_segment_valid(vcpu, VCPU_SREG_GS)) |
4243 | + return false; |
4244 | + if (!tr_valid(vcpu)) |
4245 | + return false; |
4246 | + if (!ldtr_valid(vcpu)) |
4247 | + return false; |
4248 | + } |
4249 | + /* TODO: |
4250 | + * - Add checks on RIP |
4251 | + * - Add checks on RFLAGS |
4252 | + */ |
4253 | + |
4254 | + return true; |
4255 | +} |
4256 | + |
4257 | +static int init_rmode_tss(struct kvm *kvm) |
4258 | +{ |
4259 | + gfn_t fn; |
4260 | + u16 data = 0; |
4261 | + int idx, r; |
4262 | + |
4263 | + idx = srcu_read_lock(&kvm->srcu); |
4264 | + fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT; |
4265 | + r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
4266 | + if (r < 0) |
4267 | + goto out; |
4268 | + data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE; |
4269 | + r = kvm_write_guest_page(kvm, fn++, &data, |
4270 | + TSS_IOPB_BASE_OFFSET, sizeof(u16)); |
4271 | + if (r < 0) |
4272 | + goto out; |
4273 | + r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE); |
4274 | + if (r < 0) |
4275 | + goto out; |
4276 | + r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
4277 | + if (r < 0) |
4278 | + goto out; |
4279 | + data = ~0; |
4280 | + r = kvm_write_guest_page(kvm, fn, &data, |
4281 | + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1, |
4282 | + sizeof(u8)); |
4283 | +out: |
4284 | + srcu_read_unlock(&kvm->srcu, idx); |
4285 | + return r; |
4286 | +} |
4287 | + |
4288 | +static int init_rmode_identity_map(struct kvm *kvm) |
4289 | +{ |
4290 | + struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm); |
4291 | + int i, idx, r = 0; |
4292 | + kvm_pfn_t identity_map_pfn; |
4293 | + u32 tmp; |
4294 | + |
4295 | + /* Protect kvm_vmx->ept_identity_pagetable_done. */ |
4296 | + mutex_lock(&kvm->slots_lock); |
4297 | + |
4298 | + if (likely(kvm_vmx->ept_identity_pagetable_done)) |
4299 | + goto out2; |
4300 | + |
4301 | + if (!kvm_vmx->ept_identity_map_addr) |
4302 | + kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR; |
4303 | + identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT; |
4304 | + |
4305 | + r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, |
4306 | + kvm_vmx->ept_identity_map_addr, PAGE_SIZE); |
4307 | + if (r < 0) |
4308 | + goto out2; |
4309 | + |
4310 | + idx = srcu_read_lock(&kvm->srcu); |
4311 | + r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE); |
4312 | + if (r < 0) |
4313 | + goto out; |
4314 | + /* Set up identity-mapping pagetable for EPT in real mode */ |
4315 | + for (i = 0; i < PT32_ENT_PER_PAGE; i++) { |
4316 | + tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | |
4317 | + _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE); |
4318 | + r = kvm_write_guest_page(kvm, identity_map_pfn, |
4319 | + &tmp, i * sizeof(tmp), sizeof(tmp)); |
4320 | + if (r < 0) |
4321 | + goto out; |
4322 | + } |
4323 | + kvm_vmx->ept_identity_pagetable_done = true; |
4324 | + |
4325 | +out: |
4326 | + srcu_read_unlock(&kvm->srcu, idx); |
4327 | + |
4328 | +out2: |
4329 | + mutex_unlock(&kvm->slots_lock); |
4330 | + return r; |
4331 | +} |
4332 | + |
4333 | +static void seg_setup(int seg) |
4334 | +{ |
4335 | + const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
4336 | + unsigned int ar; |
4337 | + |
4338 | + vmcs_write16(sf->selector, 0); |
4339 | + vmcs_writel(sf->base, 0); |
4340 | + vmcs_write32(sf->limit, 0xffff); |
4341 | + ar = 0x93; |
4342 | + if (seg == VCPU_SREG_CS) |
4343 | + ar |= 0x08; /* code segment */ |
4344 | + |
4345 | + vmcs_write32(sf->ar_bytes, ar); |
4346 | +} |
4347 | + |
4348 | +static int alloc_apic_access_page(struct kvm *kvm) |
4349 | +{ |
4350 | + struct page *page; |
4351 | + int r = 0; |
4352 | + |
4353 | + mutex_lock(&kvm->slots_lock); |
4354 | + if (kvm->arch.apic_access_page_done) |
4355 | + goto out; |
4356 | + r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, |
4357 | + APIC_DEFAULT_PHYS_BASE, PAGE_SIZE); |
4358 | + if (r) |
4359 | + goto out; |
4360 | + |
4361 | + page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
4362 | + if (is_error_page(page)) { |
4363 | + r = -EFAULT; |
4364 | + goto out; |
4365 | + } |
4366 | + |
4367 | + /* |
4368 | + * Do not pin the page in memory, so that memory hot-unplug |
4369 | + * is able to migrate it. |
4370 | + */ |
4371 | + put_page(page); |
4372 | + kvm->arch.apic_access_page_done = true; |
4373 | +out: |
4374 | + mutex_unlock(&kvm->slots_lock); |
4375 | + return r; |
4376 | +} |
4377 | + |
4378 | +int allocate_vpid(void) |
4379 | +{ |
4380 | + int vpid; |
4381 | + |
4382 | + if (!enable_vpid) |
4383 | + return 0; |
4384 | + spin_lock(&vmx_vpid_lock); |
4385 | + vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS); |
4386 | + if (vpid < VMX_NR_VPIDS) |
4387 | + __set_bit(vpid, vmx_vpid_bitmap); |
4388 | + else |
4389 | + vpid = 0; |
4390 | + spin_unlock(&vmx_vpid_lock); |
4391 | + return vpid; |
4392 | +} |
4393 | + |
4394 | +void free_vpid(int vpid) |
4395 | +{ |
4396 | + if (!enable_vpid || vpid == 0) |
4397 | + return; |
4398 | + spin_lock(&vmx_vpid_lock); |
4399 | + __clear_bit(vpid, vmx_vpid_bitmap); |
4400 | + spin_unlock(&vmx_vpid_lock); |
4401 | +} |
4402 | + |
4403 | +static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, |
4404 | + u32 msr, int type) |
4405 | +{ |
4406 | + int f = sizeof(unsigned long); |
4407 | + |
4408 | + if (!cpu_has_vmx_msr_bitmap()) |
4409 | + return; |
4410 | + |
4411 | + if (static_branch_unlikely(&enable_evmcs)) |
4412 | + evmcs_touch_msr_bitmap(); |
4413 | + |
4414 | + /* |
4415 | + * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals |
4416 | + * have the write-low and read-high bitmap offsets the wrong way round. |
4417 | + * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. |
4418 | + */ |
4419 | + if (msr <= 0x1fff) { |
4420 | + if (type & MSR_TYPE_R) |
4421 | + /* read-low */ |
4422 | + __clear_bit(msr, msr_bitmap + 0x000 / f); |
4423 | + |
4424 | + if (type & MSR_TYPE_W) |
4425 | + /* write-low */ |
4426 | + __clear_bit(msr, msr_bitmap + 0x800 / f); |
4427 | + |
4428 | + } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { |
4429 | + msr &= 0x1fff; |
4430 | + if (type & MSR_TYPE_R) |
4431 | + /* read-high */ |
4432 | + __clear_bit(msr, msr_bitmap + 0x400 / f); |
4433 | + |
4434 | + if (type & MSR_TYPE_W) |
4435 | + /* write-high */ |
4436 | + __clear_bit(msr, msr_bitmap + 0xc00 / f); |
4437 | + |
4438 | + } |
4439 | +} |
4440 | + |
4441 | +static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap, |
4442 | + u32 msr, int type) |
4443 | +{ |
4444 | + int f = sizeof(unsigned long); |
4445 | + |
4446 | + if (!cpu_has_vmx_msr_bitmap()) |
4447 | + return; |
4448 | + |
4449 | + if (static_branch_unlikely(&enable_evmcs)) |
4450 | + evmcs_touch_msr_bitmap(); |
4451 | + |
4452 | + /* |
4453 | + * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals |
4454 | + * have the write-low and read-high bitmap offsets the wrong way round. |
4455 | + * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. |
4456 | + */ |
4457 | + if (msr <= 0x1fff) { |
4458 | + if (type & MSR_TYPE_R) |
4459 | + /* read-low */ |
4460 | + __set_bit(msr, msr_bitmap + 0x000 / f); |
4461 | + |
4462 | + if (type & MSR_TYPE_W) |
4463 | + /* write-low */ |
4464 | + __set_bit(msr, msr_bitmap + 0x800 / f); |
4465 | + |
4466 | + } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { |
4467 | + msr &= 0x1fff; |
4468 | + if (type & MSR_TYPE_R) |
4469 | + /* read-high */ |
4470 | + __set_bit(msr, msr_bitmap + 0x400 / f); |
4471 | + |
4472 | + if (type & MSR_TYPE_W) |
4473 | + /* write-high */ |
4474 | + __set_bit(msr, msr_bitmap + 0xc00 / f); |
4475 | + |
4476 | + } |
4477 | +} |
4478 | + |
4479 | +static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap, |
4480 | + u32 msr, int type, bool value) |
4481 | +{ |
4482 | + if (value) |
4483 | + vmx_enable_intercept_for_msr(msr_bitmap, msr, type); |
4484 | + else |
4485 | + vmx_disable_intercept_for_msr(msr_bitmap, msr, type); |
4486 | +} |
4487 | + |
4488 | +static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu) |
4489 | +{ |
4490 | + u8 mode = 0; |
4491 | + |
4492 | + if (cpu_has_secondary_exec_ctrls() && |
4493 | + (secondary_exec_controls_get(to_vmx(vcpu)) & |
4494 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) { |
4495 | + mode |= MSR_BITMAP_MODE_X2APIC; |
4496 | + if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) |
4497 | + mode |= MSR_BITMAP_MODE_X2APIC_APICV; |
4498 | + } |
4499 | + |
4500 | + return mode; |
4501 | +} |
4502 | + |
4503 | +static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap, |
4504 | + u8 mode) |
4505 | +{ |
4506 | + int msr; |
4507 | + |
4508 | + for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) { |
4509 | + unsigned word = msr / BITS_PER_LONG; |
4510 | + msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0; |
4511 | + msr_bitmap[word + (0x800 / sizeof(long))] = ~0; |
4512 | + } |
4513 | + |
4514 | + if (mode & MSR_BITMAP_MODE_X2APIC) { |
4515 | + /* |
4516 | + * TPR reads and writes can be virtualized even if virtual interrupt |
4517 | + * delivery is not in use. |
4518 | + */ |
4519 | + vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW); |
4520 | + if (mode & MSR_BITMAP_MODE_X2APIC_APICV) { |
4521 | + vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R); |
4522 | + vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W); |
4523 | + vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W); |
4524 | + } |
4525 | + } |
4526 | +} |
4527 | + |
4528 | +void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu) |
4529 | +{ |
4530 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
4531 | + unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap; |
4532 | + u8 mode = vmx_msr_bitmap_mode(vcpu); |
4533 | + u8 changed = mode ^ vmx->msr_bitmap_mode; |
4534 | + |
4535 | + if (!changed) |
4536 | + return; |
4537 | + |
4538 | + if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV)) |
4539 | + vmx_update_msr_bitmap_x2apic(msr_bitmap, mode); |
4540 | + |
4541 | + vmx->msr_bitmap_mode = mode; |
4542 | +} |
4543 | + |
4544 | +void pt_update_intercept_for_msr(struct vcpu_vmx *vmx) |
4545 | +{ |
4546 | + unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap; |
4547 | + bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN); |
4548 | + u32 i; |
4549 | + |
4550 | + vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS, |
4551 | + MSR_TYPE_RW, flag); |
4552 | + vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE, |
4553 | + MSR_TYPE_RW, flag); |
4554 | + vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK, |
4555 | + MSR_TYPE_RW, flag); |
4556 | + vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH, |
4557 | + MSR_TYPE_RW, flag); |
4558 | + for (i = 0; i < vmx->pt_desc.addr_range; i++) { |
4559 | + vmx_set_intercept_for_msr(msr_bitmap, |
4560 | + MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag); |
4561 | + vmx_set_intercept_for_msr(msr_bitmap, |
4562 | + MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag); |
4563 | + } |
4564 | +} |
4565 | + |
4566 | +static bool vmx_get_enable_apicv(struct kvm *kvm) |
4567 | +{ |
4568 | + return enable_apicv; |
4569 | +} |
4570 | + |
4571 | +static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu) |
4572 | +{ |
4573 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
4574 | + void *vapic_page; |
4575 | + u32 vppr; |
4576 | + int rvi; |
4577 | + |
4578 | + if (WARN_ON_ONCE(!is_guest_mode(vcpu)) || |
4579 | + !nested_cpu_has_vid(get_vmcs12(vcpu)) || |
4580 | + WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn)) |
4581 | + return false; |
4582 | + |
4583 | + rvi = vmx_get_rvi(); |
4584 | + |
4585 | + vapic_page = vmx->nested.virtual_apic_map.hva; |
4586 | + vppr = *((u32 *)(vapic_page + APIC_PROCPRI)); |
4587 | + |
4588 | + return ((rvi & 0xf0) > (vppr & 0xf0)); |
4589 | +} |
4590 | + |
4591 | +static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu, |
4592 | + bool nested) |
4593 | +{ |
4594 | +#ifdef CONFIG_SMP |
4595 | + int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR; |
4596 | + |
4597 | + if (vcpu->mode == IN_GUEST_MODE) { |
4598 | + /* |
4599 | + * The vector of interrupt to be delivered to vcpu had |
4600 | + * been set in PIR before this function. |
4601 | + * |
4602 | + * Following cases will be reached in this block, and |
4603 | + * we always send a notification event in all cases as |
4604 | + * explained below. |
4605 | + * |
4606 | + * Case 1: vcpu keeps in non-root mode. Sending a |
4607 | + * notification event posts the interrupt to vcpu. |
4608 | + * |
4609 | + * Case 2: vcpu exits to root mode and is still |
4610 | + * runnable. PIR will be synced to vIRR before the |
4611 | + * next vcpu entry. Sending a notification event in |
4612 | + * this case has no effect, as vcpu is not in root |
4613 | + * mode. |
4614 | + * |
4615 | + * Case 3: vcpu exits to root mode and is blocked. |
4616 | + * vcpu_block() has already synced PIR to vIRR and |
4617 | + * never blocks vcpu if vIRR is not cleared. Therefore, |
4618 | + * a blocked vcpu here does not wait for any requested |
4619 | + * interrupts in PIR, and sending a notification event |
4620 | + * which has no effect is safe here. |
4621 | + */ |
4622 | + |
4623 | + apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec); |
4624 | + return true; |
4625 | + } |
4626 | +#endif |
4627 | + return false; |
4628 | +} |
4629 | + |
4630 | +static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu, |
4631 | + int vector) |
4632 | +{ |
4633 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
4634 | + |
4635 | + if (is_guest_mode(vcpu) && |
4636 | + vector == vmx->nested.posted_intr_nv) { |
4637 | + /* |
4638 | + * If a posted intr is not recognized by hardware, |
4639 | + * we will accomplish it in the next vmentry. |
4640 | + */ |
4641 | + vmx->nested.pi_pending = true; |
4642 | + kvm_make_request(KVM_REQ_EVENT, vcpu); |
4643 | + /* the PIR and ON have been set by L1. */ |
4644 | + if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true)) |
4645 | + kvm_vcpu_kick(vcpu); |
4646 | + return 0; |
4647 | + } |
4648 | + return -1; |
4649 | +} |
4650 | +/* |
4651 | + * Send interrupt to vcpu via posted interrupt way. |
4652 | + * 1. If target vcpu is running(non-root mode), send posted interrupt |
4653 | + * notification to vcpu and hardware will sync PIR to vIRR atomically. |
4654 | + * 2. If target vcpu isn't running(root mode), kick it to pick up the |
4655 | + * interrupt from PIR in next vmentry. |
4656 | + */ |
4657 | +static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector) |
4658 | +{ |
4659 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
4660 | + int r; |
4661 | + |
4662 | + r = vmx_deliver_nested_posted_interrupt(vcpu, vector); |
4663 | + if (!r) |
4664 | + return; |
4665 | + |
4666 | + if (pi_test_and_set_pir(vector, &vmx->pi_desc)) |
4667 | + return; |
4668 | + |
4669 | + /* If a previous notification has sent the IPI, nothing to do. */ |
4670 | + if (pi_test_and_set_on(&vmx->pi_desc)) |
4671 | + return; |
4672 | + |
4673 | + if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false)) |
4674 | + kvm_vcpu_kick(vcpu); |
4675 | +} |
4676 | + |
4677 | +/* |
4678 | + * Set up the vmcs's constant host-state fields, i.e., host-state fields that |
4679 | + * will not change in the lifetime of the guest. |
4680 | + * Note that host-state that does change is set elsewhere. E.g., host-state |
4681 | + * that is set differently for each CPU is set in vmx_vcpu_load(), not here. |
4682 | + */ |
4683 | +void vmx_set_constant_host_state(struct vcpu_vmx *vmx) |
4684 | +{ |
4685 | + u32 low32, high32; |
4686 | + unsigned long tmpl; |
4687 | + unsigned long cr0, cr3, cr4; |
4688 | + |
4689 | + cr0 = read_cr0(); |
4690 | + WARN_ON(cr0 & X86_CR0_TS); |
4691 | + vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */ |
4692 | + |
4693 | + /* |
4694 | + * Save the most likely value for this task's CR3 in the VMCS. |
4695 | + * We can't use __get_current_cr3_fast() because we're not atomic. |
4696 | + */ |
4697 | + cr3 = __read_cr3(); |
4698 | + vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */ |
4699 | + vmx->loaded_vmcs->host_state.cr3 = cr3; |
4700 | + |
4701 | + /* Save the most likely value for this task's CR4 in the VMCS. */ |
4702 | + cr4 = cr4_read_shadow(); |
4703 | + vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */ |
4704 | + vmx->loaded_vmcs->host_state.cr4 = cr4; |
4705 | + |
4706 | + vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */ |
4707 | +#ifdef CONFIG_X86_64 |
4708 | + /* |
4709 | + * Load null selectors, so we can avoid reloading them in |
4710 | + * vmx_prepare_switch_to_host(), in case userspace uses |
4711 | + * the null selectors too (the expected case). |
4712 | + */ |
4713 | + vmcs_write16(HOST_DS_SELECTOR, 0); |
4714 | + vmcs_write16(HOST_ES_SELECTOR, 0); |
4715 | +#else |
4716 | + vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
4717 | + vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
4718 | +#endif |
4719 | + vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
4720 | + vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */ |
4721 | + |
4722 | + vmcs_writel(HOST_IDTR_BASE, host_idt_base); /* 22.2.4 */ |
4723 | + |
4724 | + vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */ |
4725 | + |
4726 | + rdmsr(MSR_IA32_SYSENTER_CS, low32, high32); |
4727 | + vmcs_write32(HOST_IA32_SYSENTER_CS, low32); |
4728 | + rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl); |
4729 | + vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */ |
4730 | + |
4731 | + if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) { |
4732 | + rdmsr(MSR_IA32_CR_PAT, low32, high32); |
4733 | + vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32)); |
4734 | + } |
4735 | + |
4736 | + if (cpu_has_load_ia32_efer()) |
4737 | + vmcs_write64(HOST_IA32_EFER, host_efer); |
4738 | +} |
4739 | + |
4740 | +void set_cr4_guest_host_mask(struct vcpu_vmx *vmx) |
4741 | +{ |
4742 | + vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS; |
4743 | + if (enable_ept) |
4744 | + vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE; |
4745 | + if (is_guest_mode(&vmx->vcpu)) |
4746 | + vmx->vcpu.arch.cr4_guest_owned_bits &= |
4747 | + ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask; |
4748 | + vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits); |
4749 | +} |
4750 | + |
4751 | +u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx) |
4752 | +{ |
4753 | + u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl; |
4754 | + |
4755 | + if (!kvm_vcpu_apicv_active(&vmx->vcpu)) |
4756 | + pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR; |
4757 | + |
4758 | + if (!enable_vnmi) |
4759 | + pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS; |
4760 | + |
4761 | + if (!enable_preemption_timer) |
4762 | + pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER; |
4763 | + |
4764 | + return pin_based_exec_ctrl; |
4765 | +} |
4766 | + |
4767 | +static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu) |
4768 | +{ |
4769 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
4770 | + |
4771 | + pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx)); |
4772 | + if (cpu_has_secondary_exec_ctrls()) { |
4773 | + if (kvm_vcpu_apicv_active(vcpu)) |
4774 | + secondary_exec_controls_setbit(vmx, |
4775 | + SECONDARY_EXEC_APIC_REGISTER_VIRT | |
4776 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); |
4777 | + else |
4778 | + secondary_exec_controls_clearbit(vmx, |
4779 | + SECONDARY_EXEC_APIC_REGISTER_VIRT | |
4780 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); |
4781 | + } |
4782 | + |
4783 | + if (cpu_has_vmx_msr_bitmap()) |
4784 | + vmx_update_msr_bitmap(vcpu); |
4785 | +} |
4786 | + |
4787 | +u32 vmx_exec_control(struct vcpu_vmx *vmx) |
4788 | +{ |
4789 | + u32 exec_control = vmcs_config.cpu_based_exec_ctrl; |
4790 | + |
4791 | + if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) |
4792 | + exec_control &= ~CPU_BASED_MOV_DR_EXITING; |
4793 | + |
4794 | + if (!cpu_need_tpr_shadow(&vmx->vcpu)) { |
4795 | + exec_control &= ~CPU_BASED_TPR_SHADOW; |
4796 | +#ifdef CONFIG_X86_64 |
4797 | + exec_control |= CPU_BASED_CR8_STORE_EXITING | |
4798 | + CPU_BASED_CR8_LOAD_EXITING; |
4799 | +#endif |
4800 | + } |
4801 | + if (!enable_ept) |
4802 | + exec_control |= CPU_BASED_CR3_STORE_EXITING | |
4803 | + CPU_BASED_CR3_LOAD_EXITING | |
4804 | + CPU_BASED_INVLPG_EXITING; |
4805 | + if (kvm_mwait_in_guest(vmx->vcpu.kvm)) |
4806 | + exec_control &= ~(CPU_BASED_MWAIT_EXITING | |
4807 | + CPU_BASED_MONITOR_EXITING); |
4808 | + if (kvm_hlt_in_guest(vmx->vcpu.kvm)) |
4809 | + exec_control &= ~CPU_BASED_HLT_EXITING; |
4810 | + return exec_control; |
4811 | +} |
4812 | + |
4813 | + |
4814 | +static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx) |
4815 | +{ |
4816 | + struct kvm_vcpu *vcpu = &vmx->vcpu; |
4817 | + |
4818 | + u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl; |
4819 | + |
4820 | + if (pt_mode == PT_MODE_SYSTEM) |
4821 | + exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX); |
4822 | + if (!cpu_need_virtualize_apic_accesses(vcpu)) |
4823 | + exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; |
4824 | + if (vmx->vpid == 0) |
4825 | + exec_control &= ~SECONDARY_EXEC_ENABLE_VPID; |
4826 | + if (!enable_ept) { |
4827 | + exec_control &= ~SECONDARY_EXEC_ENABLE_EPT; |
4828 | + enable_unrestricted_guest = 0; |
4829 | + } |
4830 | + if (!enable_unrestricted_guest) |
4831 | + exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST; |
4832 | + if (kvm_pause_in_guest(vmx->vcpu.kvm)) |
4833 | + exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING; |
4834 | + if (!kvm_vcpu_apicv_active(vcpu)) |
4835 | + exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT | |
4836 | + SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); |
4837 | + exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; |
4838 | + |
4839 | + /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP, |
4840 | + * in vmx_set_cr4. */ |
4841 | + exec_control &= ~SECONDARY_EXEC_DESC; |
4842 | + |
4843 | + /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD |
4844 | + (handle_vmptrld). |
4845 | + We can NOT enable shadow_vmcs here because we don't have yet |
4846 | + a current VMCS12 |
4847 | + */ |
4848 | + exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS; |
4849 | + |
4850 | + if (!enable_pml) |
4851 | + exec_control &= ~SECONDARY_EXEC_ENABLE_PML; |
4852 | + |
4853 | + if (vmx_xsaves_supported()) { |
4854 | + /* Exposing XSAVES only when XSAVE is exposed */ |
4855 | + bool xsaves_enabled = |
4856 | + guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && |
4857 | + guest_cpuid_has(vcpu, X86_FEATURE_XSAVES); |
4858 | + |
4859 | + vcpu->arch.xsaves_enabled = xsaves_enabled; |
4860 | + |
4861 | + if (!xsaves_enabled) |
4862 | + exec_control &= ~SECONDARY_EXEC_XSAVES; |
4863 | + |
4864 | + if (nested) { |
4865 | + if (xsaves_enabled) |
4866 | + vmx->nested.msrs.secondary_ctls_high |= |
4867 | + SECONDARY_EXEC_XSAVES; |
4868 | + else |
4869 | + vmx->nested.msrs.secondary_ctls_high &= |
4870 | + ~SECONDARY_EXEC_XSAVES; |
4871 | + } |
4872 | + } |
4873 | + |
4874 | + if (vmx_rdtscp_supported()) { |
4875 | + bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP); |
4876 | + if (!rdtscp_enabled) |
4877 | + exec_control &= ~SECONDARY_EXEC_RDTSCP; |
4878 | + |
4879 | + if (nested) { |
4880 | + if (rdtscp_enabled) |
4881 | + vmx->nested.msrs.secondary_ctls_high |= |
4882 | + SECONDARY_EXEC_RDTSCP; |
4883 | + else |
4884 | + vmx->nested.msrs.secondary_ctls_high &= |
4885 | + ~SECONDARY_EXEC_RDTSCP; |
4886 | + } |
4887 | + } |
4888 | + |
4889 | + if (vmx_invpcid_supported()) { |
4890 | + /* Exposing INVPCID only when PCID is exposed */ |
4891 | + bool invpcid_enabled = |
4892 | + guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) && |
4893 | + guest_cpuid_has(vcpu, X86_FEATURE_PCID); |
4894 | + |
4895 | + if (!invpcid_enabled) { |
4896 | + exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID; |
4897 | + guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID); |
4898 | + } |
4899 | + |
4900 | + if (nested) { |
4901 | + if (invpcid_enabled) |
4902 | + vmx->nested.msrs.secondary_ctls_high |= |
4903 | + SECONDARY_EXEC_ENABLE_INVPCID; |
4904 | + else |
4905 | + vmx->nested.msrs.secondary_ctls_high &= |
4906 | + ~SECONDARY_EXEC_ENABLE_INVPCID; |
4907 | + } |
4908 | + } |
4909 | + |
4910 | + if (vmx_rdrand_supported()) { |
4911 | + bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND); |
4912 | + if (rdrand_enabled) |
4913 | + exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING; |
4914 | + |
4915 | + if (nested) { |
4916 | + if (rdrand_enabled) |
4917 | + vmx->nested.msrs.secondary_ctls_high |= |
4918 | + SECONDARY_EXEC_RDRAND_EXITING; |
4919 | + else |
4920 | + vmx->nested.msrs.secondary_ctls_high &= |
4921 | + ~SECONDARY_EXEC_RDRAND_EXITING; |
4922 | + } |
4923 | + } |
4924 | + |
4925 | + if (vmx_rdseed_supported()) { |
4926 | + bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED); |
4927 | + if (rdseed_enabled) |
4928 | + exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING; |
4929 | + |
4930 | + if (nested) { |
4931 | + if (rdseed_enabled) |
4932 | + vmx->nested.msrs.secondary_ctls_high |= |
4933 | + SECONDARY_EXEC_RDSEED_EXITING; |
4934 | + else |
4935 | + vmx->nested.msrs.secondary_ctls_high &= |
4936 | + ~SECONDARY_EXEC_RDSEED_EXITING; |
4937 | + } |
4938 | + } |
4939 | + |
4940 | + if (vmx_waitpkg_supported()) { |
4941 | + bool waitpkg_enabled = |
4942 | + guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG); |
4943 | + |
4944 | + if (!waitpkg_enabled) |
4945 | + exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE; |
4946 | + |
4947 | + if (nested) { |
4948 | + if (waitpkg_enabled) |
4949 | + vmx->nested.msrs.secondary_ctls_high |= |
4950 | + SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE; |
4951 | + else |
4952 | + vmx->nested.msrs.secondary_ctls_high &= |
4953 | + ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE; |
4954 | + } |
4955 | + } |
4956 | + |
4957 | + vmx->secondary_exec_control = exec_control; |
4958 | +} |
4959 | + |
4960 | +static void ept_set_mmio_spte_mask(void) |
4961 | +{ |
4962 | + /* |
4963 | + * EPT Misconfigurations can be generated if the value of bits 2:0 |
4964 | + * of an EPT paging-structure entry is 110b (write/execute). |
4965 | + */ |
4966 | + kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK, |
4967 | + VMX_EPT_MISCONFIG_WX_VALUE, 0); |
4968 | +} |
4969 | + |
4970 | +#define VMX_XSS_EXIT_BITMAP 0 |
4971 | + |
4972 | +/* |
4973 | + * Noting that the initialization of Guest-state Area of VMCS is in |
4974 | + * vmx_vcpu_reset(). |
4975 | + */ |
4976 | +static void init_vmcs(struct vcpu_vmx *vmx) |
4977 | +{ |
4978 | + if (nested) |
4979 | + nested_vmx_set_vmcs_shadowing_bitmap(); |
4980 | + |
4981 | + if (cpu_has_vmx_msr_bitmap()) |
4982 | + vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap)); |
4983 | + |
4984 | + vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */ |
4985 | + |
4986 | + /* Control */ |
4987 | + pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx)); |
4988 | + |
4989 | + exec_controls_set(vmx, vmx_exec_control(vmx)); |
4990 | + |
4991 | + if (cpu_has_secondary_exec_ctrls()) { |
4992 | + vmx_compute_secondary_exec_control(vmx); |
4993 | + secondary_exec_controls_set(vmx, vmx->secondary_exec_control); |
4994 | + } |
4995 | + |
4996 | + if (kvm_vcpu_apicv_active(&vmx->vcpu)) { |
4997 | + vmcs_write64(EOI_EXIT_BITMAP0, 0); |
4998 | + vmcs_write64(EOI_EXIT_BITMAP1, 0); |
4999 | + vmcs_write64(EOI_EXIT_BITMAP2, 0); |
5000 | + vmcs_write64(EOI_EXIT_BITMAP3, 0); |
5001 | + |
5002 | + vmcs_write16(GUEST_INTR_STATUS, 0); |
5003 | + |
5004 | + vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR); |
5005 | + vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc))); |
5006 | + } |
5007 | + |
5008 | + if (!kvm_pause_in_guest(vmx->vcpu.kvm)) { |
5009 | + vmcs_write32(PLE_GAP, ple_gap); |
5010 | + vmx->ple_window = ple_window; |
5011 | + vmx->ple_window_dirty = true; |
5012 | + } |
5013 | + |
5014 | + vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0); |
5015 | + vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0); |
5016 | + vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */ |
5017 | + |
5018 | + vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */ |
5019 | + vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */ |
5020 | + vmx_set_constant_host_state(vmx); |
5021 | + vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */ |
5022 | + vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */ |
5023 | + |
5024 | + if (cpu_has_vmx_vmfunc()) |
5025 | + vmcs_write64(VM_FUNCTION_CONTROL, 0); |
5026 | + |
5027 | + vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0); |
5028 | + vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0); |
5029 | + vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val)); |
5030 | + vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0); |
5031 | + vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val)); |
5032 | + |
5033 | + if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) |
5034 | + vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat); |
5035 | + |
5036 | + vm_exit_controls_set(vmx, vmx_vmexit_ctrl()); |
5037 | + |
5038 | + /* 22.2.1, 20.8.1 */ |
5039 | + vm_entry_controls_set(vmx, vmx_vmentry_ctrl()); |
5040 | + |
5041 | + vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS; |
5042 | + vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS); |
5043 | + |
5044 | + set_cr4_guest_host_mask(vmx); |
5045 | + |
5046 | + if (vmx->vpid != 0) |
5047 | + vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid); |
5048 | + |
5049 | + if (vmx_xsaves_supported()) |
5050 | + vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP); |
5051 | + |
5052 | + if (enable_pml) { |
5053 | + vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg)); |
5054 | + vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1); |
5055 | + } |
5056 | + |
5057 | + if (cpu_has_vmx_encls_vmexit()) |
5058 | + vmcs_write64(ENCLS_EXITING_BITMAP, -1ull); |
5059 | + |
5060 | + if (pt_mode == PT_MODE_HOST_GUEST) { |
5061 | + memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc)); |
5062 | + /* Bit[6~0] are forced to 1, writes are ignored. */ |
5063 | + vmx->pt_desc.guest.output_mask = 0x7F; |
5064 | + vmcs_write64(GUEST_IA32_RTIT_CTL, 0); |
5065 | + } |
5066 | +} |
5067 | + |
5068 | +static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
5069 | +{ |
5070 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5071 | + struct msr_data apic_base_msr; |
5072 | + u64 cr0; |
5073 | + |
5074 | + vmx->rmode.vm86_active = 0; |
5075 | + vmx->spec_ctrl = 0; |
5076 | + |
5077 | + vmx->msr_ia32_umwait_control = 0; |
5078 | + |
5079 | + vcpu->arch.microcode_version = 0x100000000ULL; |
5080 | + vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val(); |
5081 | + vmx->hv_deadline_tsc = -1; |
5082 | + kvm_set_cr8(vcpu, 0); |
5083 | + |
5084 | + if (!init_event) { |
5085 | + apic_base_msr.data = APIC_DEFAULT_PHYS_BASE | |
5086 | + MSR_IA32_APICBASE_ENABLE; |
5087 | + if (kvm_vcpu_is_reset_bsp(vcpu)) |
5088 | + apic_base_msr.data |= MSR_IA32_APICBASE_BSP; |
5089 | + apic_base_msr.host_initiated = true; |
5090 | + kvm_set_apic_base(vcpu, &apic_base_msr); |
5091 | + } |
5092 | + |
5093 | + vmx_segment_cache_clear(vmx); |
5094 | + |
5095 | + seg_setup(VCPU_SREG_CS); |
5096 | + vmcs_write16(GUEST_CS_SELECTOR, 0xf000); |
5097 | + vmcs_writel(GUEST_CS_BASE, 0xffff0000ul); |
5098 | + |
5099 | + seg_setup(VCPU_SREG_DS); |
5100 | + seg_setup(VCPU_SREG_ES); |
5101 | + seg_setup(VCPU_SREG_FS); |
5102 | + seg_setup(VCPU_SREG_GS); |
5103 | + seg_setup(VCPU_SREG_SS); |
5104 | + |
5105 | + vmcs_write16(GUEST_TR_SELECTOR, 0); |
5106 | + vmcs_writel(GUEST_TR_BASE, 0); |
5107 | + vmcs_write32(GUEST_TR_LIMIT, 0xffff); |
5108 | + vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); |
5109 | + |
5110 | + vmcs_write16(GUEST_LDTR_SELECTOR, 0); |
5111 | + vmcs_writel(GUEST_LDTR_BASE, 0); |
5112 | + vmcs_write32(GUEST_LDTR_LIMIT, 0xffff); |
5113 | + vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082); |
5114 | + |
5115 | + if (!init_event) { |
5116 | + vmcs_write32(GUEST_SYSENTER_CS, 0); |
5117 | + vmcs_writel(GUEST_SYSENTER_ESP, 0); |
5118 | + vmcs_writel(GUEST_SYSENTER_EIP, 0); |
5119 | + vmcs_write64(GUEST_IA32_DEBUGCTL, 0); |
5120 | + } |
5121 | + |
5122 | + kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); |
5123 | + kvm_rip_write(vcpu, 0xfff0); |
5124 | + |
5125 | + vmcs_writel(GUEST_GDTR_BASE, 0); |
5126 | + vmcs_write32(GUEST_GDTR_LIMIT, 0xffff); |
5127 | + |
5128 | + vmcs_writel(GUEST_IDTR_BASE, 0); |
5129 | + vmcs_write32(GUEST_IDTR_LIMIT, 0xffff); |
5130 | + |
5131 | + vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE); |
5132 | + vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0); |
5133 | + vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0); |
5134 | + if (kvm_mpx_supported()) |
5135 | + vmcs_write64(GUEST_BNDCFGS, 0); |
5136 | + |
5137 | + setup_msrs(vmx); |
5138 | + |
5139 | + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */ |
5140 | + |
5141 | + if (cpu_has_vmx_tpr_shadow() && !init_event) { |
5142 | + vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0); |
5143 | + if (cpu_need_tpr_shadow(vcpu)) |
5144 | + vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, |
5145 | + __pa(vcpu->arch.apic->regs)); |
5146 | + vmcs_write32(TPR_THRESHOLD, 0); |
5147 | + } |
5148 | + |
5149 | + kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu); |
5150 | + |
5151 | + cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET; |
5152 | + vmx->vcpu.arch.cr0 = cr0; |
5153 | + vmx_set_cr0(vcpu, cr0); /* enter rmode */ |
5154 | + vmx_set_cr4(vcpu, 0); |
5155 | + vmx_set_efer(vcpu, 0); |
5156 | + |
5157 | + update_exception_bitmap(vcpu); |
5158 | + |
5159 | + vpid_sync_context(vmx->vpid); |
5160 | + if (init_event) |
5161 | + vmx_clear_hlt(vcpu); |
5162 | +} |
5163 | + |
5164 | +static void enable_irq_window(struct kvm_vcpu *vcpu) |
5165 | +{ |
5166 | + exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING); |
5167 | +} |
5168 | + |
5169 | +static void enable_nmi_window(struct kvm_vcpu *vcpu) |
5170 | +{ |
5171 | + if (!enable_vnmi || |
5172 | + vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) { |
5173 | + enable_irq_window(vcpu); |
5174 | + return; |
5175 | + } |
5176 | + |
5177 | + exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING); |
5178 | +} |
5179 | + |
5180 | +static void vmx_inject_irq(struct kvm_vcpu *vcpu) |
5181 | +{ |
5182 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5183 | + uint32_t intr; |
5184 | + int irq = vcpu->arch.interrupt.nr; |
5185 | + |
5186 | + trace_kvm_inj_virq(irq); |
5187 | + |
5188 | + ++vcpu->stat.irq_injections; |
5189 | + if (vmx->rmode.vm86_active) { |
5190 | + int inc_eip = 0; |
5191 | + if (vcpu->arch.interrupt.soft) |
5192 | + inc_eip = vcpu->arch.event_exit_inst_len; |
5193 | + kvm_inject_realmode_interrupt(vcpu, irq, inc_eip); |
5194 | + return; |
5195 | + } |
5196 | + intr = irq | INTR_INFO_VALID_MASK; |
5197 | + if (vcpu->arch.interrupt.soft) { |
5198 | + intr |= INTR_TYPE_SOFT_INTR; |
5199 | + vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, |
5200 | + vmx->vcpu.arch.event_exit_inst_len); |
5201 | + } else |
5202 | + intr |= INTR_TYPE_EXT_INTR; |
5203 | + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr); |
5204 | + |
5205 | + vmx_clear_hlt(vcpu); |
5206 | +} |
5207 | + |
5208 | +static void vmx_inject_nmi(struct kvm_vcpu *vcpu) |
5209 | +{ |
5210 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5211 | + |
5212 | + if (!enable_vnmi) { |
5213 | + /* |
5214 | + * Tracking the NMI-blocked state in software is built upon |
5215 | + * finding the next open IRQ window. This, in turn, depends on |
5216 | + * well-behaving guests: They have to keep IRQs disabled at |
5217 | + * least as long as the NMI handler runs. Otherwise we may |
5218 | + * cause NMI nesting, maybe breaking the guest. But as this is |
5219 | + * highly unlikely, we can live with the residual risk. |
5220 | + */ |
5221 | + vmx->loaded_vmcs->soft_vnmi_blocked = 1; |
5222 | + vmx->loaded_vmcs->vnmi_blocked_time = 0; |
5223 | + } |
5224 | + |
5225 | + ++vcpu->stat.nmi_injections; |
5226 | + vmx->loaded_vmcs->nmi_known_unmasked = false; |
5227 | + |
5228 | + if (vmx->rmode.vm86_active) { |
5229 | + kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0); |
5230 | + return; |
5231 | + } |
5232 | + |
5233 | + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, |
5234 | + INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR); |
5235 | + |
5236 | + vmx_clear_hlt(vcpu); |
5237 | +} |
5238 | + |
5239 | +bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu) |
5240 | +{ |
5241 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5242 | + bool masked; |
5243 | + |
5244 | + if (!enable_vnmi) |
5245 | + return vmx->loaded_vmcs->soft_vnmi_blocked; |
5246 | + if (vmx->loaded_vmcs->nmi_known_unmasked) |
5247 | + return false; |
5248 | + masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI; |
5249 | + vmx->loaded_vmcs->nmi_known_unmasked = !masked; |
5250 | + return masked; |
5251 | +} |
5252 | + |
5253 | +void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) |
5254 | +{ |
5255 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5256 | + |
5257 | + if (!enable_vnmi) { |
5258 | + if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) { |
5259 | + vmx->loaded_vmcs->soft_vnmi_blocked = masked; |
5260 | + vmx->loaded_vmcs->vnmi_blocked_time = 0; |
5261 | + } |
5262 | + } else { |
5263 | + vmx->loaded_vmcs->nmi_known_unmasked = !masked; |
5264 | + if (masked) |
5265 | + vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, |
5266 | + GUEST_INTR_STATE_NMI); |
5267 | + else |
5268 | + vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO, |
5269 | + GUEST_INTR_STATE_NMI); |
5270 | + } |
5271 | +} |
5272 | + |
5273 | +static int vmx_nmi_allowed(struct kvm_vcpu *vcpu) |
5274 | +{ |
5275 | + if (to_vmx(vcpu)->nested.nested_run_pending) |
5276 | + return 0; |
5277 | + |
5278 | + if (!enable_vnmi && |
5279 | + to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked) |
5280 | + return 0; |
5281 | + |
5282 | + return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & |
5283 | + (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
5284 | + | GUEST_INTR_STATE_NMI)); |
5285 | +} |
5286 | + |
5287 | +static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu) |
5288 | +{ |
5289 | + return (!to_vmx(vcpu)->nested.nested_run_pending && |
5290 | + vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) && |
5291 | + !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & |
5292 | + (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)); |
5293 | +} |
5294 | + |
5295 | +static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr) |
5296 | +{ |
5297 | + int ret; |
5298 | + |
5299 | + if (enable_unrestricted_guest) |
5300 | + return 0; |
5301 | + |
5302 | + ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr, |
5303 | + PAGE_SIZE * 3); |
5304 | + if (ret) |
5305 | + return ret; |
5306 | + to_kvm_vmx(kvm)->tss_addr = addr; |
5307 | + return init_rmode_tss(kvm); |
5308 | +} |
5309 | + |
5310 | +static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr) |
5311 | +{ |
5312 | + to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr; |
5313 | + return 0; |
5314 | +} |
5315 | + |
5316 | +static bool rmode_exception(struct kvm_vcpu *vcpu, int vec) |
5317 | +{ |
5318 | + switch (vec) { |
5319 | + case BP_VECTOR: |
5320 | + /* |
5321 | + * Update instruction length as we may reinject the exception |
5322 | + * from user space while in guest debugging mode. |
5323 | + */ |
5324 | + to_vmx(vcpu)->vcpu.arch.event_exit_inst_len = |
5325 | + vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
5326 | + if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
5327 | + return false; |
5328 | + /* fall through */ |
5329 | + case DB_VECTOR: |
5330 | + if (vcpu->guest_debug & |
5331 | + (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) |
5332 | + return false; |
5333 | + /* fall through */ |
5334 | + case DE_VECTOR: |
5335 | + case OF_VECTOR: |
5336 | + case BR_VECTOR: |
5337 | + case UD_VECTOR: |
5338 | + case DF_VECTOR: |
5339 | + case SS_VECTOR: |
5340 | + case GP_VECTOR: |
5341 | + case MF_VECTOR: |
5342 | + return true; |
5343 | + break; |
5344 | + } |
5345 | + return false; |
5346 | +} |
5347 | + |
5348 | +static int handle_rmode_exception(struct kvm_vcpu *vcpu, |
5349 | + int vec, u32 err_code) |
5350 | +{ |
5351 | + /* |
5352 | + * Instruction with address size override prefix opcode 0x67 |
5353 | + * Cause the #SS fault with 0 error code in VM86 mode. |
5354 | + */ |
5355 | + if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) { |
5356 | + if (kvm_emulate_instruction(vcpu, 0)) { |
5357 | + if (vcpu->arch.halt_request) { |
5358 | + vcpu->arch.halt_request = 0; |
5359 | + return kvm_vcpu_halt(vcpu); |
5360 | + } |
5361 | + return 1; |
5362 | + } |
5363 | + return 0; |
5364 | + } |
5365 | + |
5366 | + /* |
5367 | + * Forward all other exceptions that are valid in real mode. |
5368 | + * FIXME: Breaks guest debugging in real mode, needs to be fixed with |
5369 | + * the required debugging infrastructure rework. |
5370 | + */ |
5371 | + kvm_queue_exception(vcpu, vec); |
5372 | + return 1; |
5373 | +} |
5374 | + |
5375 | +/* |
5376 | + * Trigger machine check on the host. We assume all the MSRs are already set up |
5377 | + * by the CPU and that we still run on the same CPU as the MCE occurred on. |
5378 | + * We pass a fake environment to the machine check handler because we want |
5379 | + * the guest to be always treated like user space, no matter what context |
5380 | + * it used internally. |
5381 | + */ |
5382 | +static void kvm_machine_check(void) |
5383 | +{ |
5384 | +#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64) |
5385 | + struct pt_regs regs = { |
5386 | + .cs = 3, /* Fake ring 3 no matter what the guest ran on */ |
5387 | + .flags = X86_EFLAGS_IF, |
5388 | + }; |
5389 | + |
5390 | + do_machine_check(®s, 0); |
5391 | +#endif |
5392 | +} |
5393 | + |
5394 | +static int handle_machine_check(struct kvm_vcpu *vcpu) |
5395 | +{ |
5396 | + /* handled by vmx_vcpu_run() */ |
5397 | + return 1; |
5398 | +} |
5399 | + |
5400 | +static int handle_exception_nmi(struct kvm_vcpu *vcpu) |
5401 | +{ |
5402 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5403 | + struct kvm_run *kvm_run = vcpu->run; |
5404 | + u32 intr_info, ex_no, error_code; |
5405 | + unsigned long cr2, rip, dr6; |
5406 | + u32 vect_info; |
5407 | + |
5408 | + vect_info = vmx->idt_vectoring_info; |
5409 | + intr_info = vmx->exit_intr_info; |
5410 | + |
5411 | + if (is_machine_check(intr_info) || is_nmi(intr_info)) |
5412 | + return 1; /* handled by handle_exception_nmi_irqoff() */ |
5413 | + |
5414 | + if (is_invalid_opcode(intr_info)) |
5415 | + return handle_ud(vcpu); |
5416 | + |
5417 | + error_code = 0; |
5418 | + if (intr_info & INTR_INFO_DELIVER_CODE_MASK) |
5419 | + error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE); |
5420 | + |
5421 | + if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) { |
5422 | + WARN_ON_ONCE(!enable_vmware_backdoor); |
5423 | + |
5424 | + /* |
5425 | + * VMware backdoor emulation on #GP interception only handles |
5426 | + * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero |
5427 | + * error code on #GP. |
5428 | + */ |
5429 | + if (error_code) { |
5430 | + kvm_queue_exception_e(vcpu, GP_VECTOR, error_code); |
5431 | + return 1; |
5432 | + } |
5433 | + return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP); |
5434 | + } |
5435 | + |
5436 | + /* |
5437 | + * The #PF with PFEC.RSVD = 1 indicates the guest is accessing |
5438 | + * MMIO, it is better to report an internal error. |
5439 | + * See the comments in vmx_handle_exit. |
5440 | + */ |
5441 | + if ((vect_info & VECTORING_INFO_VALID_MASK) && |
5442 | + !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) { |
5443 | + vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
5444 | + vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX; |
5445 | + vcpu->run->internal.ndata = 3; |
5446 | + vcpu->run->internal.data[0] = vect_info; |
5447 | + vcpu->run->internal.data[1] = intr_info; |
5448 | + vcpu->run->internal.data[2] = error_code; |
5449 | + return 0; |
5450 | + } |
5451 | + |
5452 | + if (is_page_fault(intr_info)) { |
5453 | + cr2 = vmcs_readl(EXIT_QUALIFICATION); |
5454 | + /* EPT won't cause page fault directly */ |
5455 | + WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept); |
5456 | + return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0); |
5457 | + } |
5458 | + |
5459 | + ex_no = intr_info & INTR_INFO_VECTOR_MASK; |
5460 | + |
5461 | + if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no)) |
5462 | + return handle_rmode_exception(vcpu, ex_no, error_code); |
5463 | + |
5464 | + switch (ex_no) { |
5465 | + case AC_VECTOR: |
5466 | + kvm_queue_exception_e(vcpu, AC_VECTOR, error_code); |
5467 | + return 1; |
5468 | + case DB_VECTOR: |
5469 | + dr6 = vmcs_readl(EXIT_QUALIFICATION); |
5470 | + if (!(vcpu->guest_debug & |
5471 | + (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) { |
5472 | + vcpu->arch.dr6 &= ~DR_TRAP_BITS; |
5473 | + vcpu->arch.dr6 |= dr6 | DR6_RTM; |
5474 | + if (is_icebp(intr_info)) |
5475 | + WARN_ON(!skip_emulated_instruction(vcpu)); |
5476 | + |
5477 | + kvm_queue_exception(vcpu, DB_VECTOR); |
5478 | + return 1; |
5479 | + } |
5480 | + kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1; |
5481 | + kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7); |
5482 | + /* fall through */ |
5483 | + case BP_VECTOR: |
5484 | + /* |
5485 | + * Update instruction length as we may reinject #BP from |
5486 | + * user space while in guest debugging mode. Reading it for |
5487 | + * #DB as well causes no harm, it is not used in that case. |
5488 | + */ |
5489 | + vmx->vcpu.arch.event_exit_inst_len = |
5490 | + vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
5491 | + kvm_run->exit_reason = KVM_EXIT_DEBUG; |
5492 | + rip = kvm_rip_read(vcpu); |
5493 | + kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip; |
5494 | + kvm_run->debug.arch.exception = ex_no; |
5495 | + break; |
5496 | + default: |
5497 | + kvm_run->exit_reason = KVM_EXIT_EXCEPTION; |
5498 | + kvm_run->ex.exception = ex_no; |
5499 | + kvm_run->ex.error_code = error_code; |
5500 | + break; |
5501 | + } |
5502 | + return 0; |
5503 | +} |
5504 | + |
5505 | +static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu) |
5506 | +{ |
5507 | + ++vcpu->stat.irq_exits; |
5508 | + return 1; |
5509 | +} |
5510 | + |
5511 | +static int handle_triple_fault(struct kvm_vcpu *vcpu) |
5512 | +{ |
5513 | + vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
5514 | + vcpu->mmio_needed = 0; |
5515 | + return 0; |
5516 | +} |
5517 | + |
5518 | +static int handle_io(struct kvm_vcpu *vcpu) |
5519 | +{ |
5520 | + unsigned long exit_qualification; |
5521 | + int size, in, string; |
5522 | + unsigned port; |
5523 | + |
5524 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5525 | + string = (exit_qualification & 16) != 0; |
5526 | + |
5527 | + ++vcpu->stat.io_exits; |
5528 | + |
5529 | + if (string) |
5530 | + return kvm_emulate_instruction(vcpu, 0); |
5531 | + |
5532 | + port = exit_qualification >> 16; |
5533 | + size = (exit_qualification & 7) + 1; |
5534 | + in = (exit_qualification & 8) != 0; |
5535 | + |
5536 | + return kvm_fast_pio(vcpu, size, port, in); |
5537 | +} |
5538 | + |
5539 | +static void |
5540 | +vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) |
5541 | +{ |
5542 | + /* |
5543 | + * Patch in the VMCALL instruction: |
5544 | + */ |
5545 | + hypercall[0] = 0x0f; |
5546 | + hypercall[1] = 0x01; |
5547 | + hypercall[2] = 0xc1; |
5548 | +} |
5549 | + |
5550 | +/* called to set cr0 as appropriate for a mov-to-cr0 exit. */ |
5551 | +static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val) |
5552 | +{ |
5553 | + if (is_guest_mode(vcpu)) { |
5554 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
5555 | + unsigned long orig_val = val; |
5556 | + |
5557 | + /* |
5558 | + * We get here when L2 changed cr0 in a way that did not change |
5559 | + * any of L1's shadowed bits (see nested_vmx_exit_handled_cr), |
5560 | + * but did change L0 shadowed bits. So we first calculate the |
5561 | + * effective cr0 value that L1 would like to write into the |
5562 | + * hardware. It consists of the L2-owned bits from the new |
5563 | + * value combined with the L1-owned bits from L1's guest_cr0. |
5564 | + */ |
5565 | + val = (val & ~vmcs12->cr0_guest_host_mask) | |
5566 | + (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask); |
5567 | + |
5568 | + if (!nested_guest_cr0_valid(vcpu, val)) |
5569 | + return 1; |
5570 | + |
5571 | + if (kvm_set_cr0(vcpu, val)) |
5572 | + return 1; |
5573 | + vmcs_writel(CR0_READ_SHADOW, orig_val); |
5574 | + return 0; |
5575 | + } else { |
5576 | + if (to_vmx(vcpu)->nested.vmxon && |
5577 | + !nested_host_cr0_valid(vcpu, val)) |
5578 | + return 1; |
5579 | + |
5580 | + return kvm_set_cr0(vcpu, val); |
5581 | + } |
5582 | +} |
5583 | + |
5584 | +static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val) |
5585 | +{ |
5586 | + if (is_guest_mode(vcpu)) { |
5587 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
5588 | + unsigned long orig_val = val; |
5589 | + |
5590 | + /* analogously to handle_set_cr0 */ |
5591 | + val = (val & ~vmcs12->cr4_guest_host_mask) | |
5592 | + (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask); |
5593 | + if (kvm_set_cr4(vcpu, val)) |
5594 | + return 1; |
5595 | + vmcs_writel(CR4_READ_SHADOW, orig_val); |
5596 | + return 0; |
5597 | + } else |
5598 | + return kvm_set_cr4(vcpu, val); |
5599 | +} |
5600 | + |
5601 | +static int handle_desc(struct kvm_vcpu *vcpu) |
5602 | +{ |
5603 | + WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP)); |
5604 | + return kvm_emulate_instruction(vcpu, 0); |
5605 | +} |
5606 | + |
5607 | +static int handle_cr(struct kvm_vcpu *vcpu) |
5608 | +{ |
5609 | + unsigned long exit_qualification, val; |
5610 | + int cr; |
5611 | + int reg; |
5612 | + int err; |
5613 | + int ret; |
5614 | + |
5615 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5616 | + cr = exit_qualification & 15; |
5617 | + reg = (exit_qualification >> 8) & 15; |
5618 | + switch ((exit_qualification >> 4) & 3) { |
5619 | + case 0: /* mov to cr */ |
5620 | + val = kvm_register_readl(vcpu, reg); |
5621 | + trace_kvm_cr_write(cr, val); |
5622 | + switch (cr) { |
5623 | + case 0: |
5624 | + err = handle_set_cr0(vcpu, val); |
5625 | + return kvm_complete_insn_gp(vcpu, err); |
5626 | + case 3: |
5627 | + WARN_ON_ONCE(enable_unrestricted_guest); |
5628 | + err = kvm_set_cr3(vcpu, val); |
5629 | + return kvm_complete_insn_gp(vcpu, err); |
5630 | + case 4: |
5631 | + err = handle_set_cr4(vcpu, val); |
5632 | + return kvm_complete_insn_gp(vcpu, err); |
5633 | + case 8: { |
5634 | + u8 cr8_prev = kvm_get_cr8(vcpu); |
5635 | + u8 cr8 = (u8)val; |
5636 | + err = kvm_set_cr8(vcpu, cr8); |
5637 | + ret = kvm_complete_insn_gp(vcpu, err); |
5638 | + if (lapic_in_kernel(vcpu)) |
5639 | + return ret; |
5640 | + if (cr8_prev <= cr8) |
5641 | + return ret; |
5642 | + /* |
5643 | + * TODO: we might be squashing a |
5644 | + * KVM_GUESTDBG_SINGLESTEP-triggered |
5645 | + * KVM_EXIT_DEBUG here. |
5646 | + */ |
5647 | + vcpu->run->exit_reason = KVM_EXIT_SET_TPR; |
5648 | + return 0; |
5649 | + } |
5650 | + } |
5651 | + break; |
5652 | + case 2: /* clts */ |
5653 | + WARN_ONCE(1, "Guest should always own CR0.TS"); |
5654 | + vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS)); |
5655 | + trace_kvm_cr_write(0, kvm_read_cr0(vcpu)); |
5656 | + return kvm_skip_emulated_instruction(vcpu); |
5657 | + case 1: /*mov from cr*/ |
5658 | + switch (cr) { |
5659 | + case 3: |
5660 | + WARN_ON_ONCE(enable_unrestricted_guest); |
5661 | + val = kvm_read_cr3(vcpu); |
5662 | + kvm_register_write(vcpu, reg, val); |
5663 | + trace_kvm_cr_read(cr, val); |
5664 | + return kvm_skip_emulated_instruction(vcpu); |
5665 | + case 8: |
5666 | + val = kvm_get_cr8(vcpu); |
5667 | + kvm_register_write(vcpu, reg, val); |
5668 | + trace_kvm_cr_read(cr, val); |
5669 | + return kvm_skip_emulated_instruction(vcpu); |
5670 | + } |
5671 | + break; |
5672 | + case 3: /* lmsw */ |
5673 | + val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f; |
5674 | + trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val); |
5675 | + kvm_lmsw(vcpu, val); |
5676 | + |
5677 | + return kvm_skip_emulated_instruction(vcpu); |
5678 | + default: |
5679 | + break; |
5680 | + } |
5681 | + vcpu->run->exit_reason = 0; |
5682 | + vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n", |
5683 | + (int)(exit_qualification >> 4) & 3, cr); |
5684 | + return 0; |
5685 | +} |
5686 | + |
5687 | +static int handle_dr(struct kvm_vcpu *vcpu) |
5688 | +{ |
5689 | + unsigned long exit_qualification; |
5690 | + int dr, dr7, reg; |
5691 | + |
5692 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5693 | + dr = exit_qualification & DEBUG_REG_ACCESS_NUM; |
5694 | + |
5695 | + /* First, if DR does not exist, trigger UD */ |
5696 | + if (!kvm_require_dr(vcpu, dr)) |
5697 | + return 1; |
5698 | + |
5699 | + /* Do not handle if the CPL > 0, will trigger GP on re-entry */ |
5700 | + if (!kvm_require_cpl(vcpu, 0)) |
5701 | + return 1; |
5702 | + dr7 = vmcs_readl(GUEST_DR7); |
5703 | + if (dr7 & DR7_GD) { |
5704 | + /* |
5705 | + * As the vm-exit takes precedence over the debug trap, we |
5706 | + * need to emulate the latter, either for the host or the |
5707 | + * guest debugging itself. |
5708 | + */ |
5709 | + if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { |
5710 | + vcpu->run->debug.arch.dr6 = vcpu->arch.dr6; |
5711 | + vcpu->run->debug.arch.dr7 = dr7; |
5712 | + vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu); |
5713 | + vcpu->run->debug.arch.exception = DB_VECTOR; |
5714 | + vcpu->run->exit_reason = KVM_EXIT_DEBUG; |
5715 | + return 0; |
5716 | + } else { |
5717 | + vcpu->arch.dr6 &= ~DR_TRAP_BITS; |
5718 | + vcpu->arch.dr6 |= DR6_BD | DR6_RTM; |
5719 | + kvm_queue_exception(vcpu, DB_VECTOR); |
5720 | + return 1; |
5721 | + } |
5722 | + } |
5723 | + |
5724 | + if (vcpu->guest_debug == 0) { |
5725 | + exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING); |
5726 | + |
5727 | + /* |
5728 | + * No more DR vmexits; force a reload of the debug registers |
5729 | + * and reenter on this instruction. The next vmexit will |
5730 | + * retrieve the full state of the debug registers. |
5731 | + */ |
5732 | + vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; |
5733 | + return 1; |
5734 | + } |
5735 | + |
5736 | + reg = DEBUG_REG_ACCESS_REG(exit_qualification); |
5737 | + if (exit_qualification & TYPE_MOV_FROM_DR) { |
5738 | + unsigned long val; |
5739 | + |
5740 | + if (kvm_get_dr(vcpu, dr, &val)) |
5741 | + return 1; |
5742 | + kvm_register_write(vcpu, reg, val); |
5743 | + } else |
5744 | + if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg))) |
5745 | + return 1; |
5746 | + |
5747 | + return kvm_skip_emulated_instruction(vcpu); |
5748 | +} |
5749 | + |
5750 | +static u64 vmx_get_dr6(struct kvm_vcpu *vcpu) |
5751 | +{ |
5752 | + return vcpu->arch.dr6; |
5753 | +} |
5754 | + |
5755 | +static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val) |
5756 | +{ |
5757 | +} |
5758 | + |
5759 | +static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) |
5760 | +{ |
5761 | + get_debugreg(vcpu->arch.db[0], 0); |
5762 | + get_debugreg(vcpu->arch.db[1], 1); |
5763 | + get_debugreg(vcpu->arch.db[2], 2); |
5764 | + get_debugreg(vcpu->arch.db[3], 3); |
5765 | + get_debugreg(vcpu->arch.dr6, 6); |
5766 | + vcpu->arch.dr7 = vmcs_readl(GUEST_DR7); |
5767 | + |
5768 | + vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; |
5769 | + exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING); |
5770 | +} |
5771 | + |
5772 | +static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val) |
5773 | +{ |
5774 | + vmcs_writel(GUEST_DR7, val); |
5775 | +} |
5776 | + |
5777 | +static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu) |
5778 | +{ |
5779 | + kvm_apic_update_ppr(vcpu); |
5780 | + return 1; |
5781 | +} |
5782 | + |
5783 | +static int handle_interrupt_window(struct kvm_vcpu *vcpu) |
5784 | +{ |
5785 | + exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING); |
5786 | + |
5787 | + kvm_make_request(KVM_REQ_EVENT, vcpu); |
5788 | + |
5789 | + ++vcpu->stat.irq_window_exits; |
5790 | + return 1; |
5791 | +} |
5792 | + |
5793 | +static int handle_vmcall(struct kvm_vcpu *vcpu) |
5794 | +{ |
5795 | + return kvm_emulate_hypercall(vcpu); |
5796 | +} |
5797 | + |
5798 | +static int handle_invd(struct kvm_vcpu *vcpu) |
5799 | +{ |
5800 | + return kvm_emulate_instruction(vcpu, 0); |
5801 | +} |
5802 | + |
5803 | +static int handle_invlpg(struct kvm_vcpu *vcpu) |
5804 | +{ |
5805 | + unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5806 | + |
5807 | + kvm_mmu_invlpg(vcpu, exit_qualification); |
5808 | + return kvm_skip_emulated_instruction(vcpu); |
5809 | +} |
5810 | + |
5811 | +static int handle_rdpmc(struct kvm_vcpu *vcpu) |
5812 | +{ |
5813 | + int err; |
5814 | + |
5815 | + err = kvm_rdpmc(vcpu); |
5816 | + return kvm_complete_insn_gp(vcpu, err); |
5817 | +} |
5818 | + |
5819 | +static int handle_wbinvd(struct kvm_vcpu *vcpu) |
5820 | +{ |
5821 | + return kvm_emulate_wbinvd(vcpu); |
5822 | +} |
5823 | + |
5824 | +static int handle_xsetbv(struct kvm_vcpu *vcpu) |
5825 | +{ |
5826 | + u64 new_bv = kvm_read_edx_eax(vcpu); |
5827 | + u32 index = kvm_rcx_read(vcpu); |
5828 | + |
5829 | + if (kvm_set_xcr(vcpu, index, new_bv) == 0) |
5830 | + return kvm_skip_emulated_instruction(vcpu); |
5831 | + return 1; |
5832 | +} |
5833 | + |
5834 | +static int handle_apic_access(struct kvm_vcpu *vcpu) |
5835 | +{ |
5836 | + if (likely(fasteoi)) { |
5837 | + unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5838 | + int access_type, offset; |
5839 | + |
5840 | + access_type = exit_qualification & APIC_ACCESS_TYPE; |
5841 | + offset = exit_qualification & APIC_ACCESS_OFFSET; |
5842 | + /* |
5843 | + * Sane guest uses MOV to write EOI, with written value |
5844 | + * not cared. So make a short-circuit here by avoiding |
5845 | + * heavy instruction emulation. |
5846 | + */ |
5847 | + if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) && |
5848 | + (offset == APIC_EOI)) { |
5849 | + kvm_lapic_set_eoi(vcpu); |
5850 | + return kvm_skip_emulated_instruction(vcpu); |
5851 | + } |
5852 | + } |
5853 | + return kvm_emulate_instruction(vcpu, 0); |
5854 | +} |
5855 | + |
5856 | +static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu) |
5857 | +{ |
5858 | + unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5859 | + int vector = exit_qualification & 0xff; |
5860 | + |
5861 | + /* EOI-induced VM exit is trap-like and thus no need to adjust IP */ |
5862 | + kvm_apic_set_eoi_accelerated(vcpu, vector); |
5863 | + return 1; |
5864 | +} |
5865 | + |
5866 | +static int handle_apic_write(struct kvm_vcpu *vcpu) |
5867 | +{ |
5868 | + unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5869 | + u32 offset = exit_qualification & 0xfff; |
5870 | + |
5871 | + /* APIC-write VM exit is trap-like and thus no need to adjust IP */ |
5872 | + kvm_apic_write_nodecode(vcpu, offset); |
5873 | + return 1; |
5874 | +} |
5875 | + |
5876 | +static int handle_task_switch(struct kvm_vcpu *vcpu) |
5877 | +{ |
5878 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
5879 | + unsigned long exit_qualification; |
5880 | + bool has_error_code = false; |
5881 | + u32 error_code = 0; |
5882 | + u16 tss_selector; |
5883 | + int reason, type, idt_v, idt_index; |
5884 | + |
5885 | + idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK); |
5886 | + idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK); |
5887 | + type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK); |
5888 | + |
5889 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5890 | + |
5891 | + reason = (u32)exit_qualification >> 30; |
5892 | + if (reason == TASK_SWITCH_GATE && idt_v) { |
5893 | + switch (type) { |
5894 | + case INTR_TYPE_NMI_INTR: |
5895 | + vcpu->arch.nmi_injected = false; |
5896 | + vmx_set_nmi_mask(vcpu, true); |
5897 | + break; |
5898 | + case INTR_TYPE_EXT_INTR: |
5899 | + case INTR_TYPE_SOFT_INTR: |
5900 | + kvm_clear_interrupt_queue(vcpu); |
5901 | + break; |
5902 | + case INTR_TYPE_HARD_EXCEPTION: |
5903 | + if (vmx->idt_vectoring_info & |
5904 | + VECTORING_INFO_DELIVER_CODE_MASK) { |
5905 | + has_error_code = true; |
5906 | + error_code = |
5907 | + vmcs_read32(IDT_VECTORING_ERROR_CODE); |
5908 | + } |
5909 | + /* fall through */ |
5910 | + case INTR_TYPE_SOFT_EXCEPTION: |
5911 | + kvm_clear_exception_queue(vcpu); |
5912 | + break; |
5913 | + default: |
5914 | + break; |
5915 | + } |
5916 | + } |
5917 | + tss_selector = exit_qualification; |
5918 | + |
5919 | + if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION && |
5920 | + type != INTR_TYPE_EXT_INTR && |
5921 | + type != INTR_TYPE_NMI_INTR)) |
5922 | + WARN_ON(!skip_emulated_instruction(vcpu)); |
5923 | + |
5924 | + /* |
5925 | + * TODO: What about debug traps on tss switch? |
5926 | + * Are we supposed to inject them and update dr6? |
5927 | + */ |
5928 | + return kvm_task_switch(vcpu, tss_selector, |
5929 | + type == INTR_TYPE_SOFT_INTR ? idt_index : -1, |
5930 | + reason, has_error_code, error_code); |
5931 | +} |
5932 | + |
5933 | +static int handle_ept_violation(struct kvm_vcpu *vcpu) |
5934 | +{ |
5935 | + unsigned long exit_qualification; |
5936 | + gpa_t gpa; |
5937 | + u64 error_code; |
5938 | + |
5939 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
5940 | + |
5941 | + /* |
5942 | + * EPT violation happened while executing iret from NMI, |
5943 | + * "blocked by NMI" bit has to be set before next VM entry. |
5944 | + * There are errata that may cause this bit to not be set: |
5945 | + * AAK134, BY25. |
5946 | + */ |
5947 | + if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) && |
5948 | + enable_vnmi && |
5949 | + (exit_qualification & INTR_INFO_UNBLOCK_NMI)) |
5950 | + vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI); |
5951 | + |
5952 | + gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); |
5953 | + trace_kvm_page_fault(gpa, exit_qualification); |
5954 | + |
5955 | + /* Is it a read fault? */ |
5956 | + error_code = (exit_qualification & EPT_VIOLATION_ACC_READ) |
5957 | + ? PFERR_USER_MASK : 0; |
5958 | + /* Is it a write fault? */ |
5959 | + error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE) |
5960 | + ? PFERR_WRITE_MASK : 0; |
5961 | + /* Is it a fetch fault? */ |
5962 | + error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR) |
5963 | + ? PFERR_FETCH_MASK : 0; |
5964 | + /* ept page table entry is present? */ |
5965 | + error_code |= (exit_qualification & |
5966 | + (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE | |
5967 | + EPT_VIOLATION_EXECUTABLE)) |
5968 | + ? PFERR_PRESENT_MASK : 0; |
5969 | + |
5970 | + error_code |= (exit_qualification & 0x100) != 0 ? |
5971 | + PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK; |
5972 | + |
5973 | + vcpu->arch.exit_qualification = exit_qualification; |
5974 | + return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0); |
5975 | +} |
5976 | + |
5977 | +static int handle_ept_misconfig(struct kvm_vcpu *vcpu) |
5978 | +{ |
5979 | + gpa_t gpa; |
5980 | + |
5981 | + /* |
5982 | + * A nested guest cannot optimize MMIO vmexits, because we have an |
5983 | + * nGPA here instead of the required GPA. |
5984 | + */ |
5985 | + gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); |
5986 | + if (!is_guest_mode(vcpu) && |
5987 | + !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) { |
5988 | + trace_kvm_fast_mmio(gpa); |
5989 | + return kvm_skip_emulated_instruction(vcpu); |
5990 | + } |
5991 | + |
5992 | + return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0); |
5993 | +} |
5994 | + |
5995 | +static int handle_nmi_window(struct kvm_vcpu *vcpu) |
5996 | +{ |
5997 | + WARN_ON_ONCE(!enable_vnmi); |
5998 | + exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING); |
5999 | + ++vcpu->stat.nmi_window_exits; |
6000 | + kvm_make_request(KVM_REQ_EVENT, vcpu); |
6001 | + |
6002 | + return 1; |
6003 | +} |
6004 | + |
6005 | +static int handle_invalid_guest_state(struct kvm_vcpu *vcpu) |
6006 | +{ |
6007 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6008 | + bool intr_window_requested; |
6009 | + unsigned count = 130; |
6010 | + |
6011 | + /* |
6012 | + * We should never reach the point where we are emulating L2 |
6013 | + * due to invalid guest state as that means we incorrectly |
6014 | + * allowed a nested VMEntry with an invalid vmcs12. |
6015 | + */ |
6016 | + WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending); |
6017 | + |
6018 | + intr_window_requested = exec_controls_get(vmx) & |
6019 | + CPU_BASED_INTR_WINDOW_EXITING; |
6020 | + |
6021 | + while (vmx->emulation_required && count-- != 0) { |
6022 | + if (intr_window_requested && vmx_interrupt_allowed(vcpu)) |
6023 | + return handle_interrupt_window(&vmx->vcpu); |
6024 | + |
6025 | + if (kvm_test_request(KVM_REQ_EVENT, vcpu)) |
6026 | + return 1; |
6027 | + |
6028 | + if (!kvm_emulate_instruction(vcpu, 0)) |
6029 | + return 0; |
6030 | + |
6031 | + if (vmx->emulation_required && !vmx->rmode.vm86_active && |
6032 | + vcpu->arch.exception.pending) { |
6033 | + vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6034 | + vcpu->run->internal.suberror = |
6035 | + KVM_INTERNAL_ERROR_EMULATION; |
6036 | + vcpu->run->internal.ndata = 0; |
6037 | + return 0; |
6038 | + } |
6039 | + |
6040 | + if (vcpu->arch.halt_request) { |
6041 | + vcpu->arch.halt_request = 0; |
6042 | + return kvm_vcpu_halt(vcpu); |
6043 | + } |
6044 | + |
6045 | + /* |
6046 | + * Note, return 1 and not 0, vcpu_run() is responsible for |
6047 | + * morphing the pending signal into the proper return code. |
6048 | + */ |
6049 | + if (signal_pending(current)) |
6050 | + return 1; |
6051 | + |
6052 | + if (need_resched()) |
6053 | + schedule(); |
6054 | + } |
6055 | + |
6056 | + return 1; |
6057 | +} |
6058 | + |
6059 | +static void grow_ple_window(struct kvm_vcpu *vcpu) |
6060 | +{ |
6061 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6062 | + unsigned int old = vmx->ple_window; |
6063 | + |
6064 | + vmx->ple_window = __grow_ple_window(old, ple_window, |
6065 | + ple_window_grow, |
6066 | + ple_window_max); |
6067 | + |
6068 | + if (vmx->ple_window != old) { |
6069 | + vmx->ple_window_dirty = true; |
6070 | + trace_kvm_ple_window_update(vcpu->vcpu_id, |
6071 | + vmx->ple_window, old); |
6072 | + } |
6073 | +} |
6074 | + |
6075 | +static void shrink_ple_window(struct kvm_vcpu *vcpu) |
6076 | +{ |
6077 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6078 | + unsigned int old = vmx->ple_window; |
6079 | + |
6080 | + vmx->ple_window = __shrink_ple_window(old, ple_window, |
6081 | + ple_window_shrink, |
6082 | + ple_window); |
6083 | + |
6084 | + if (vmx->ple_window != old) { |
6085 | + vmx->ple_window_dirty = true; |
6086 | + trace_kvm_ple_window_update(vcpu->vcpu_id, |
6087 | + vmx->ple_window, old); |
6088 | + } |
6089 | +} |
6090 | + |
6091 | +/* |
6092 | + * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR. |
6093 | + */ |
6094 | +static void wakeup_handler(void) |
6095 | +{ |
6096 | + struct kvm_vcpu *vcpu; |
6097 | + int cpu = smp_processor_id(); |
6098 | + |
6099 | + spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); |
6100 | + list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu), |
6101 | + blocked_vcpu_list) { |
6102 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
6103 | + |
6104 | + if (pi_test_on(pi_desc) == 1) |
6105 | + kvm_vcpu_kick(vcpu); |
6106 | + } |
6107 | + spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); |
6108 | +} |
6109 | + |
6110 | +static void vmx_enable_tdp(void) |
6111 | +{ |
6112 | + kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK, |
6113 | + enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull, |
6114 | + enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull, |
6115 | + 0ull, VMX_EPT_EXECUTABLE_MASK, |
6116 | + cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK, |
6117 | + VMX_EPT_RWX_MASK, 0ull); |
6118 | + |
6119 | + ept_set_mmio_spte_mask(); |
6120 | + kvm_enable_tdp(); |
6121 | +} |
6122 | + |
6123 | +/* |
6124 | + * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE |
6125 | + * exiting, so only get here on cpu with PAUSE-Loop-Exiting. |
6126 | + */ |
6127 | +static int handle_pause(struct kvm_vcpu *vcpu) |
6128 | +{ |
6129 | + if (!kvm_pause_in_guest(vcpu->kvm)) |
6130 | + grow_ple_window(vcpu); |
6131 | + |
6132 | + /* |
6133 | + * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting" |
6134 | + * VM-execution control is ignored if CPL > 0. OTOH, KVM |
6135 | + * never set PAUSE_EXITING and just set PLE if supported, |
6136 | + * so the vcpu must be CPL=0 if it gets a PAUSE exit. |
6137 | + */ |
6138 | + kvm_vcpu_on_spin(vcpu, true); |
6139 | + return kvm_skip_emulated_instruction(vcpu); |
6140 | +} |
6141 | + |
6142 | +static int handle_nop(struct kvm_vcpu *vcpu) |
6143 | +{ |
6144 | + return kvm_skip_emulated_instruction(vcpu); |
6145 | +} |
6146 | + |
6147 | +static int handle_mwait(struct kvm_vcpu *vcpu) |
6148 | +{ |
6149 | + printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n"); |
6150 | + return handle_nop(vcpu); |
6151 | +} |
6152 | + |
6153 | +static int handle_invalid_op(struct kvm_vcpu *vcpu) |
6154 | +{ |
6155 | + kvm_queue_exception(vcpu, UD_VECTOR); |
6156 | + return 1; |
6157 | +} |
6158 | + |
6159 | +static int handle_monitor_trap(struct kvm_vcpu *vcpu) |
6160 | +{ |
6161 | + return 1; |
6162 | +} |
6163 | + |
6164 | +static int handle_monitor(struct kvm_vcpu *vcpu) |
6165 | +{ |
6166 | + printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n"); |
6167 | + return handle_nop(vcpu); |
6168 | +} |
6169 | + |
6170 | +static int handle_invpcid(struct kvm_vcpu *vcpu) |
6171 | +{ |
6172 | + u32 vmx_instruction_info; |
6173 | + unsigned long type; |
6174 | + bool pcid_enabled; |
6175 | + gva_t gva; |
6176 | + struct x86_exception e; |
6177 | + unsigned i; |
6178 | + unsigned long roots_to_free = 0; |
6179 | + struct { |
6180 | + u64 pcid; |
6181 | + u64 gla; |
6182 | + } operand; |
6183 | + |
6184 | + if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) { |
6185 | + kvm_queue_exception(vcpu, UD_VECTOR); |
6186 | + return 1; |
6187 | + } |
6188 | + |
6189 | + vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); |
6190 | + type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf); |
6191 | + |
6192 | + if (type > 3) { |
6193 | + kvm_inject_gp(vcpu, 0); |
6194 | + return 1; |
6195 | + } |
6196 | + |
6197 | + /* According to the Intel instruction reference, the memory operand |
6198 | + * is read even if it isn't needed (e.g., for type==all) |
6199 | + */ |
6200 | + if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION), |
6201 | + vmx_instruction_info, false, |
6202 | + sizeof(operand), &gva)) |
6203 | + return 1; |
6204 | + |
6205 | + if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) { |
6206 | + kvm_inject_page_fault(vcpu, &e); |
6207 | + return 1; |
6208 | + } |
6209 | + |
6210 | + if (operand.pcid >> 12 != 0) { |
6211 | + kvm_inject_gp(vcpu, 0); |
6212 | + return 1; |
6213 | + } |
6214 | + |
6215 | + pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); |
6216 | + |
6217 | + switch (type) { |
6218 | + case INVPCID_TYPE_INDIV_ADDR: |
6219 | + if ((!pcid_enabled && (operand.pcid != 0)) || |
6220 | + is_noncanonical_address(operand.gla, vcpu)) { |
6221 | + kvm_inject_gp(vcpu, 0); |
6222 | + return 1; |
6223 | + } |
6224 | + kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid); |
6225 | + return kvm_skip_emulated_instruction(vcpu); |
6226 | + |
6227 | + case INVPCID_TYPE_SINGLE_CTXT: |
6228 | + if (!pcid_enabled && (operand.pcid != 0)) { |
6229 | + kvm_inject_gp(vcpu, 0); |
6230 | + return 1; |
6231 | + } |
6232 | + |
6233 | + if (kvm_get_active_pcid(vcpu) == operand.pcid) { |
6234 | + kvm_mmu_sync_roots(vcpu); |
6235 | + kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
6236 | + } |
6237 | + |
6238 | + for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) |
6239 | + if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3) |
6240 | + == operand.pcid) |
6241 | + roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); |
6242 | + |
6243 | + kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free); |
6244 | + /* |
6245 | + * If neither the current cr3 nor any of the prev_roots use the |
6246 | + * given PCID, then nothing needs to be done here because a |
6247 | + * resync will happen anyway before switching to any other CR3. |
6248 | + */ |
6249 | + |
6250 | + return kvm_skip_emulated_instruction(vcpu); |
6251 | + |
6252 | + case INVPCID_TYPE_ALL_NON_GLOBAL: |
6253 | + /* |
6254 | + * Currently, KVM doesn't mark global entries in the shadow |
6255 | + * page tables, so a non-global flush just degenerates to a |
6256 | + * global flush. If needed, we could optimize this later by |
6257 | + * keeping track of global entries in shadow page tables. |
6258 | + */ |
6259 | + |
6260 | + /* fall-through */ |
6261 | + case INVPCID_TYPE_ALL_INCL_GLOBAL: |
6262 | + kvm_mmu_unload(vcpu); |
6263 | + return kvm_skip_emulated_instruction(vcpu); |
6264 | + |
6265 | + default: |
6266 | + BUG(); /* We have already checked above that type <= 3 */ |
6267 | + } |
6268 | +} |
6269 | + |
6270 | +static int handle_pml_full(struct kvm_vcpu *vcpu) |
6271 | +{ |
6272 | + unsigned long exit_qualification; |
6273 | + |
6274 | + trace_kvm_pml_full(vcpu->vcpu_id); |
6275 | + |
6276 | + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
6277 | + |
6278 | + /* |
6279 | + * PML buffer FULL happened while executing iret from NMI, |
6280 | + * "blocked by NMI" bit has to be set before next VM entry. |
6281 | + */ |
6282 | + if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) && |
6283 | + enable_vnmi && |
6284 | + (exit_qualification & INTR_INFO_UNBLOCK_NMI)) |
6285 | + vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, |
6286 | + GUEST_INTR_STATE_NMI); |
6287 | + |
6288 | + /* |
6289 | + * PML buffer already flushed at beginning of VMEXIT. Nothing to do |
6290 | + * here.., and there's no userspace involvement needed for PML. |
6291 | + */ |
6292 | + return 1; |
6293 | +} |
6294 | + |
6295 | +static int handle_preemption_timer(struct kvm_vcpu *vcpu) |
6296 | +{ |
6297 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6298 | + |
6299 | + if (!vmx->req_immediate_exit && |
6300 | + !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) |
6301 | + kvm_lapic_expired_hv_timer(vcpu); |
6302 | + |
6303 | + return 1; |
6304 | +} |
6305 | + |
6306 | +/* |
6307 | + * When nested=0, all VMX instruction VM Exits filter here. The handlers |
6308 | + * are overwritten by nested_vmx_setup() when nested=1. |
6309 | + */ |
6310 | +static int handle_vmx_instruction(struct kvm_vcpu *vcpu) |
6311 | +{ |
6312 | + kvm_queue_exception(vcpu, UD_VECTOR); |
6313 | + return 1; |
6314 | +} |
6315 | + |
6316 | +static int handle_encls(struct kvm_vcpu *vcpu) |
6317 | +{ |
6318 | + /* |
6319 | + * SGX virtualization is not yet supported. There is no software |
6320 | + * enable bit for SGX, so we have to trap ENCLS and inject a #UD |
6321 | + * to prevent the guest from executing ENCLS. |
6322 | + */ |
6323 | + kvm_queue_exception(vcpu, UD_VECTOR); |
6324 | + return 1; |
6325 | +} |
6326 | + |
6327 | +/* |
6328 | + * The exit handlers return 1 if the exit was handled fully and guest execution |
6329 | + * may resume. Otherwise they set the kvm_run parameter to indicate what needs |
6330 | + * to be done to userspace and return 0. |
6331 | + */ |
6332 | +static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = { |
6333 | + [EXIT_REASON_EXCEPTION_NMI] = handle_exception_nmi, |
6334 | + [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt, |
6335 | + [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault, |
6336 | + [EXIT_REASON_NMI_WINDOW] = handle_nmi_window, |
6337 | + [EXIT_REASON_IO_INSTRUCTION] = handle_io, |
6338 | + [EXIT_REASON_CR_ACCESS] = handle_cr, |
6339 | + [EXIT_REASON_DR_ACCESS] = handle_dr, |
6340 | + [EXIT_REASON_CPUID] = kvm_emulate_cpuid, |
6341 | + [EXIT_REASON_MSR_READ] = kvm_emulate_rdmsr, |
6342 | + [EXIT_REASON_MSR_WRITE] = kvm_emulate_wrmsr, |
6343 | + [EXIT_REASON_INTERRUPT_WINDOW] = handle_interrupt_window, |
6344 | + [EXIT_REASON_HLT] = kvm_emulate_halt, |
6345 | + [EXIT_REASON_INVD] = handle_invd, |
6346 | + [EXIT_REASON_INVLPG] = handle_invlpg, |
6347 | + [EXIT_REASON_RDPMC] = handle_rdpmc, |
6348 | + [EXIT_REASON_VMCALL] = handle_vmcall, |
6349 | + [EXIT_REASON_VMCLEAR] = handle_vmx_instruction, |
6350 | + [EXIT_REASON_VMLAUNCH] = handle_vmx_instruction, |
6351 | + [EXIT_REASON_VMPTRLD] = handle_vmx_instruction, |
6352 | + [EXIT_REASON_VMPTRST] = handle_vmx_instruction, |
6353 | + [EXIT_REASON_VMREAD] = handle_vmx_instruction, |
6354 | + [EXIT_REASON_VMRESUME] = handle_vmx_instruction, |
6355 | + [EXIT_REASON_VMWRITE] = handle_vmx_instruction, |
6356 | + [EXIT_REASON_VMOFF] = handle_vmx_instruction, |
6357 | + [EXIT_REASON_VMON] = handle_vmx_instruction, |
6358 | + [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold, |
6359 | + [EXIT_REASON_APIC_ACCESS] = handle_apic_access, |
6360 | + [EXIT_REASON_APIC_WRITE] = handle_apic_write, |
6361 | + [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced, |
6362 | + [EXIT_REASON_WBINVD] = handle_wbinvd, |
6363 | + [EXIT_REASON_XSETBV] = handle_xsetbv, |
6364 | + [EXIT_REASON_TASK_SWITCH] = handle_task_switch, |
6365 | + [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check, |
6366 | + [EXIT_REASON_GDTR_IDTR] = handle_desc, |
6367 | + [EXIT_REASON_LDTR_TR] = handle_desc, |
6368 | + [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation, |
6369 | + [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig, |
6370 | + [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause, |
6371 | + [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait, |
6372 | + [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap, |
6373 | + [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor, |
6374 | + [EXIT_REASON_INVEPT] = handle_vmx_instruction, |
6375 | + [EXIT_REASON_INVVPID] = handle_vmx_instruction, |
6376 | + [EXIT_REASON_RDRAND] = handle_invalid_op, |
6377 | + [EXIT_REASON_RDSEED] = handle_invalid_op, |
6378 | + [EXIT_REASON_PML_FULL] = handle_pml_full, |
6379 | + [EXIT_REASON_INVPCID] = handle_invpcid, |
6380 | + [EXIT_REASON_VMFUNC] = handle_vmx_instruction, |
6381 | + [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer, |
6382 | + [EXIT_REASON_ENCLS] = handle_encls, |
6383 | +}; |
6384 | + |
6385 | +static const int kvm_vmx_max_exit_handlers = |
6386 | + ARRAY_SIZE(kvm_vmx_exit_handlers); |
6387 | + |
6388 | +static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2) |
6389 | +{ |
6390 | + *info1 = vmcs_readl(EXIT_QUALIFICATION); |
6391 | + *info2 = vmcs_read32(VM_EXIT_INTR_INFO); |
6392 | +} |
6393 | + |
6394 | +static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx) |
6395 | +{ |
6396 | + if (vmx->pml_pg) { |
6397 | + __free_page(vmx->pml_pg); |
6398 | + vmx->pml_pg = NULL; |
6399 | + } |
6400 | +} |
6401 | + |
6402 | +static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu) |
6403 | +{ |
6404 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6405 | + u64 *pml_buf; |
6406 | + u16 pml_idx; |
6407 | + |
6408 | + pml_idx = vmcs_read16(GUEST_PML_INDEX); |
6409 | + |
6410 | + /* Do nothing if PML buffer is empty */ |
6411 | + if (pml_idx == (PML_ENTITY_NUM - 1)) |
6412 | + return; |
6413 | + |
6414 | + /* PML index always points to next available PML buffer entity */ |
6415 | + if (pml_idx >= PML_ENTITY_NUM) |
6416 | + pml_idx = 0; |
6417 | + else |
6418 | + pml_idx++; |
6419 | + |
6420 | + pml_buf = page_address(vmx->pml_pg); |
6421 | + for (; pml_idx < PML_ENTITY_NUM; pml_idx++) { |
6422 | + u64 gpa; |
6423 | + |
6424 | + gpa = pml_buf[pml_idx]; |
6425 | + WARN_ON(gpa & (PAGE_SIZE - 1)); |
6426 | + kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT); |
6427 | + } |
6428 | + |
6429 | + /* reset PML index */ |
6430 | + vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1); |
6431 | +} |
6432 | + |
6433 | +/* |
6434 | + * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap. |
6435 | + * Called before reporting dirty_bitmap to userspace. |
6436 | + */ |
6437 | +static void kvm_flush_pml_buffers(struct kvm *kvm) |
6438 | +{ |
6439 | + int i; |
6440 | + struct kvm_vcpu *vcpu; |
6441 | + /* |
6442 | + * We only need to kick vcpu out of guest mode here, as PML buffer |
6443 | + * is flushed at beginning of all VMEXITs, and it's obvious that only |
6444 | + * vcpus running in guest are possible to have unflushed GPAs in PML |
6445 | + * buffer. |
6446 | + */ |
6447 | + kvm_for_each_vcpu(i, vcpu, kvm) |
6448 | + kvm_vcpu_kick(vcpu); |
6449 | +} |
6450 | + |
6451 | +static void vmx_dump_sel(char *name, uint32_t sel) |
6452 | +{ |
6453 | + pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n", |
6454 | + name, vmcs_read16(sel), |
6455 | + vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR), |
6456 | + vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR), |
6457 | + vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR)); |
6458 | +} |
6459 | + |
6460 | +static void vmx_dump_dtsel(char *name, uint32_t limit) |
6461 | +{ |
6462 | + pr_err("%s limit=0x%08x, base=0x%016lx\n", |
6463 | + name, vmcs_read32(limit), |
6464 | + vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT)); |
6465 | +} |
6466 | + |
6467 | +void dump_vmcs(void) |
6468 | +{ |
6469 | + u32 vmentry_ctl, vmexit_ctl; |
6470 | + u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control; |
6471 | + unsigned long cr4; |
6472 | + u64 efer; |
6473 | + int i, n; |
6474 | + |
6475 | + if (!dump_invalid_vmcs) { |
6476 | + pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n"); |
6477 | + return; |
6478 | + } |
6479 | + |
6480 | + vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS); |
6481 | + vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS); |
6482 | + cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); |
6483 | + pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL); |
6484 | + cr4 = vmcs_readl(GUEST_CR4); |
6485 | + efer = vmcs_read64(GUEST_IA32_EFER); |
6486 | + secondary_exec_control = 0; |
6487 | + if (cpu_has_secondary_exec_ctrls()) |
6488 | + secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL); |
6489 | + |
6490 | + pr_err("*** Guest State ***\n"); |
6491 | + pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n", |
6492 | + vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW), |
6493 | + vmcs_readl(CR0_GUEST_HOST_MASK)); |
6494 | + pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n", |
6495 | + cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK)); |
6496 | + pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3)); |
6497 | + if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) && |
6498 | + (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA)) |
6499 | + { |
6500 | + pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n", |
6501 | + vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1)); |
6502 | + pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n", |
6503 | + vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3)); |
6504 | + } |
6505 | + pr_err("RSP = 0x%016lx RIP = 0x%016lx\n", |
6506 | + vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP)); |
6507 | + pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n", |
6508 | + vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7)); |
6509 | + pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n", |
6510 | + vmcs_readl(GUEST_SYSENTER_ESP), |
6511 | + vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP)); |
6512 | + vmx_dump_sel("CS: ", GUEST_CS_SELECTOR); |
6513 | + vmx_dump_sel("DS: ", GUEST_DS_SELECTOR); |
6514 | + vmx_dump_sel("SS: ", GUEST_SS_SELECTOR); |
6515 | + vmx_dump_sel("ES: ", GUEST_ES_SELECTOR); |
6516 | + vmx_dump_sel("FS: ", GUEST_FS_SELECTOR); |
6517 | + vmx_dump_sel("GS: ", GUEST_GS_SELECTOR); |
6518 | + vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT); |
6519 | + vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR); |
6520 | + vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT); |
6521 | + vmx_dump_sel("TR: ", GUEST_TR_SELECTOR); |
6522 | + if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) || |
6523 | + (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER))) |
6524 | + pr_err("EFER = 0x%016llx PAT = 0x%016llx\n", |
6525 | + efer, vmcs_read64(GUEST_IA32_PAT)); |
6526 | + pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n", |
6527 | + vmcs_read64(GUEST_IA32_DEBUGCTL), |
6528 | + vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS)); |
6529 | + if (cpu_has_load_perf_global_ctrl() && |
6530 | + vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) |
6531 | + pr_err("PerfGlobCtl = 0x%016llx\n", |
6532 | + vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL)); |
6533 | + if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS) |
6534 | + pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS)); |
6535 | + pr_err("Interruptibility = %08x ActivityState = %08x\n", |
6536 | + vmcs_read32(GUEST_INTERRUPTIBILITY_INFO), |
6537 | + vmcs_read32(GUEST_ACTIVITY_STATE)); |
6538 | + if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) |
6539 | + pr_err("InterruptStatus = %04x\n", |
6540 | + vmcs_read16(GUEST_INTR_STATUS)); |
6541 | + |
6542 | + pr_err("*** Host State ***\n"); |
6543 | + pr_err("RIP = 0x%016lx RSP = 0x%016lx\n", |
6544 | + vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP)); |
6545 | + pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n", |
6546 | + vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR), |
6547 | + vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR), |
6548 | + vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR), |
6549 | + vmcs_read16(HOST_TR_SELECTOR)); |
6550 | + pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n", |
6551 | + vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE), |
6552 | + vmcs_readl(HOST_TR_BASE)); |
6553 | + pr_err("GDTBase=%016lx IDTBase=%016lx\n", |
6554 | + vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE)); |
6555 | + pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n", |
6556 | + vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3), |
6557 | + vmcs_readl(HOST_CR4)); |
6558 | + pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n", |
6559 | + vmcs_readl(HOST_IA32_SYSENTER_ESP), |
6560 | + vmcs_read32(HOST_IA32_SYSENTER_CS), |
6561 | + vmcs_readl(HOST_IA32_SYSENTER_EIP)); |
6562 | + if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER)) |
6563 | + pr_err("EFER = 0x%016llx PAT = 0x%016llx\n", |
6564 | + vmcs_read64(HOST_IA32_EFER), |
6565 | + vmcs_read64(HOST_IA32_PAT)); |
6566 | + if (cpu_has_load_perf_global_ctrl() && |
6567 | + vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) |
6568 | + pr_err("PerfGlobCtl = 0x%016llx\n", |
6569 | + vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL)); |
6570 | + |
6571 | + pr_err("*** Control State ***\n"); |
6572 | + pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n", |
6573 | + pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control); |
6574 | + pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl); |
6575 | + pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n", |
6576 | + vmcs_read32(EXCEPTION_BITMAP), |
6577 | + vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK), |
6578 | + vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH)); |
6579 | + pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n", |
6580 | + vmcs_read32(VM_ENTRY_INTR_INFO_FIELD), |
6581 | + vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE), |
6582 | + vmcs_read32(VM_ENTRY_INSTRUCTION_LEN)); |
6583 | + pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n", |
6584 | + vmcs_read32(VM_EXIT_INTR_INFO), |
6585 | + vmcs_read32(VM_EXIT_INTR_ERROR_CODE), |
6586 | + vmcs_read32(VM_EXIT_INSTRUCTION_LEN)); |
6587 | + pr_err(" reason=%08x qualification=%016lx\n", |
6588 | + vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION)); |
6589 | + pr_err("IDTVectoring: info=%08x errcode=%08x\n", |
6590 | + vmcs_read32(IDT_VECTORING_INFO_FIELD), |
6591 | + vmcs_read32(IDT_VECTORING_ERROR_CODE)); |
6592 | + pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET)); |
6593 | + if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING) |
6594 | + pr_err("TSC Multiplier = 0x%016llx\n", |
6595 | + vmcs_read64(TSC_MULTIPLIER)); |
6596 | + if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) { |
6597 | + if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) { |
6598 | + u16 status = vmcs_read16(GUEST_INTR_STATUS); |
6599 | + pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff); |
6600 | + } |
6601 | + pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD)); |
6602 | + if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) |
6603 | + pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR)); |
6604 | + pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR)); |
6605 | + } |
6606 | + if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR) |
6607 | + pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV)); |
6608 | + if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT)) |
6609 | + pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER)); |
6610 | + n = vmcs_read32(CR3_TARGET_COUNT); |
6611 | + for (i = 0; i + 1 < n; i += 4) |
6612 | + pr_err("CR3 target%u=%016lx target%u=%016lx\n", |
6613 | + i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2), |
6614 | + i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2)); |
6615 | + if (i < n) |
6616 | + pr_err("CR3 target%u=%016lx\n", |
6617 | + i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2)); |
6618 | + if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING) |
6619 | + pr_err("PLE Gap=%08x Window=%08x\n", |
6620 | + vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW)); |
6621 | + if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID) |
6622 | + pr_err("Virtual processor ID = 0x%04x\n", |
6623 | + vmcs_read16(VIRTUAL_PROCESSOR_ID)); |
6624 | +} |
6625 | + |
6626 | +/* |
6627 | + * The guest has exited. See if we can fix it or if we need userspace |
6628 | + * assistance. |
6629 | + */ |
6630 | +static int vmx_handle_exit(struct kvm_vcpu *vcpu, |
6631 | + enum exit_fastpath_completion exit_fastpath) |
6632 | +{ |
6633 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6634 | + u32 exit_reason = vmx->exit_reason; |
6635 | + u32 vectoring_info = vmx->idt_vectoring_info; |
6636 | + |
6637 | + trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX); |
6638 | + |
6639 | + /* |
6640 | + * Flush logged GPAs PML buffer, this will make dirty_bitmap more |
6641 | + * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before |
6642 | + * querying dirty_bitmap, we only need to kick all vcpus out of guest |
6643 | + * mode as if vcpus is in root mode, the PML buffer must has been |
6644 | + * flushed already. |
6645 | + */ |
6646 | + if (enable_pml) |
6647 | + vmx_flush_pml_buffer(vcpu); |
6648 | + |
6649 | + /* If guest state is invalid, start emulating */ |
6650 | + if (vmx->emulation_required) |
6651 | + return handle_invalid_guest_state(vcpu); |
6652 | + |
6653 | + if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason)) |
6654 | + return nested_vmx_reflect_vmexit(vcpu, exit_reason); |
6655 | + |
6656 | + if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) { |
6657 | + dump_vmcs(); |
6658 | + vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
6659 | + vcpu->run->fail_entry.hardware_entry_failure_reason |
6660 | + = exit_reason; |
6661 | + return 0; |
6662 | + } |
6663 | + |
6664 | + if (unlikely(vmx->fail)) { |
6665 | + dump_vmcs(); |
6666 | + vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
6667 | + vcpu->run->fail_entry.hardware_entry_failure_reason |
6668 | + = vmcs_read32(VM_INSTRUCTION_ERROR); |
6669 | + return 0; |
6670 | + } |
6671 | + |
6672 | + /* |
6673 | + * Note: |
6674 | + * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by |
6675 | + * delivery event since it indicates guest is accessing MMIO. |
6676 | + * The vm-exit can be triggered again after return to guest that |
6677 | + * will cause infinite loop. |
6678 | + */ |
6679 | + if ((vectoring_info & VECTORING_INFO_VALID_MASK) && |
6680 | + (exit_reason != EXIT_REASON_EXCEPTION_NMI && |
6681 | + exit_reason != EXIT_REASON_EPT_VIOLATION && |
6682 | + exit_reason != EXIT_REASON_PML_FULL && |
6683 | + exit_reason != EXIT_REASON_TASK_SWITCH)) { |
6684 | + vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6685 | + vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV; |
6686 | + vcpu->run->internal.ndata = 3; |
6687 | + vcpu->run->internal.data[0] = vectoring_info; |
6688 | + vcpu->run->internal.data[1] = exit_reason; |
6689 | + vcpu->run->internal.data[2] = vcpu->arch.exit_qualification; |
6690 | + if (exit_reason == EXIT_REASON_EPT_MISCONFIG) { |
6691 | + vcpu->run->internal.ndata++; |
6692 | + vcpu->run->internal.data[3] = |
6693 | + vmcs_read64(GUEST_PHYSICAL_ADDRESS); |
6694 | + } |
6695 | + return 0; |
6696 | + } |
6697 | + |
6698 | + if (unlikely(!enable_vnmi && |
6699 | + vmx->loaded_vmcs->soft_vnmi_blocked)) { |
6700 | + if (vmx_interrupt_allowed(vcpu)) { |
6701 | + vmx->loaded_vmcs->soft_vnmi_blocked = 0; |
6702 | + } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL && |
6703 | + vcpu->arch.nmi_pending) { |
6704 | + /* |
6705 | + * This CPU don't support us in finding the end of an |
6706 | + * NMI-blocked window if the guest runs with IRQs |
6707 | + * disabled. So we pull the trigger after 1 s of |
6708 | + * futile waiting, but inform the user about this. |
6709 | + */ |
6710 | + printk(KERN_WARNING "%s: Breaking out of NMI-blocked " |
6711 | + "state on VCPU %d after 1 s timeout\n", |
6712 | + __func__, vcpu->vcpu_id); |
6713 | + vmx->loaded_vmcs->soft_vnmi_blocked = 0; |
6714 | + } |
6715 | + } |
6716 | + |
6717 | + if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) { |
6718 | + kvm_skip_emulated_instruction(vcpu); |
6719 | + return 1; |
6720 | + } else if (exit_reason < kvm_vmx_max_exit_handlers |
6721 | + && kvm_vmx_exit_handlers[exit_reason]) { |
6722 | +#ifdef CONFIG_RETPOLINE |
6723 | + if (exit_reason == EXIT_REASON_MSR_WRITE) |
6724 | + return kvm_emulate_wrmsr(vcpu); |
6725 | + else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER) |
6726 | + return handle_preemption_timer(vcpu); |
6727 | + else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW) |
6728 | + return handle_interrupt_window(vcpu); |
6729 | + else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT) |
6730 | + return handle_external_interrupt(vcpu); |
6731 | + else if (exit_reason == EXIT_REASON_HLT) |
6732 | + return kvm_emulate_halt(vcpu); |
6733 | + else if (exit_reason == EXIT_REASON_EPT_MISCONFIG) |
6734 | + return handle_ept_misconfig(vcpu); |
6735 | +#endif |
6736 | + return kvm_vmx_exit_handlers[exit_reason](vcpu); |
6737 | + } else { |
6738 | + vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", |
6739 | + exit_reason); |
6740 | + dump_vmcs(); |
6741 | + vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6742 | + vcpu->run->internal.suberror = |
6743 | + KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON; |
6744 | + vcpu->run->internal.ndata = 1; |
6745 | + vcpu->run->internal.data[0] = exit_reason; |
6746 | + return 0; |
6747 | + } |
6748 | +} |
6749 | + |
6750 | +/* |
6751 | + * Software based L1D cache flush which is used when microcode providing |
6752 | + * the cache control MSR is not loaded. |
6753 | + * |
6754 | + * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to |
6755 | + * flush it is required to read in 64 KiB because the replacement algorithm |
6756 | + * is not exactly LRU. This could be sized at runtime via topology |
6757 | + * information but as all relevant affected CPUs have 32KiB L1D cache size |
6758 | + * there is no point in doing so. |
6759 | + */ |
6760 | +static void vmx_l1d_flush(struct kvm_vcpu *vcpu) |
6761 | +{ |
6762 | + int size = PAGE_SIZE << L1D_CACHE_ORDER; |
6763 | + |
6764 | + /* |
6765 | + * This code is only executed when the the flush mode is 'cond' or |
6766 | + * 'always' |
6767 | + */ |
6768 | + if (static_branch_likely(&vmx_l1d_flush_cond)) { |
6769 | + bool flush_l1d; |
6770 | + |
6771 | + /* |
6772 | + * Clear the per-vcpu flush bit, it gets set again |
6773 | + * either from vcpu_run() or from one of the unsafe |
6774 | + * VMEXIT handlers. |
6775 | + */ |
6776 | + flush_l1d = vcpu->arch.l1tf_flush_l1d; |
6777 | + vcpu->arch.l1tf_flush_l1d = false; |
6778 | + |
6779 | + /* |
6780 | + * Clear the per-cpu flush bit, it gets set again from |
6781 | + * the interrupt handlers. |
6782 | + */ |
6783 | + flush_l1d |= kvm_get_cpu_l1tf_flush_l1d(); |
6784 | + kvm_clear_cpu_l1tf_flush_l1d(); |
6785 | + |
6786 | + if (!flush_l1d) |
6787 | + return; |
6788 | + } |
6789 | + |
6790 | + vcpu->stat.l1d_flush++; |
6791 | + |
6792 | + if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) { |
6793 | + wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH); |
6794 | + return; |
6795 | + } |
6796 | + |
6797 | + asm volatile( |
6798 | + /* First ensure the pages are in the TLB */ |
6799 | + "xorl %%eax, %%eax\n" |
6800 | + ".Lpopulate_tlb:\n\t" |
6801 | + "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t" |
6802 | + "addl $4096, %%eax\n\t" |
6803 | + "cmpl %%eax, %[size]\n\t" |
6804 | + "jne .Lpopulate_tlb\n\t" |
6805 | + "xorl %%eax, %%eax\n\t" |
6806 | + "cpuid\n\t" |
6807 | + /* Now fill the cache */ |
6808 | + "xorl %%eax, %%eax\n" |
6809 | + ".Lfill_cache:\n" |
6810 | + "movzbl (%[flush_pages], %%" _ASM_AX "), %%ecx\n\t" |
6811 | + "addl $64, %%eax\n\t" |
6812 | + "cmpl %%eax, %[size]\n\t" |
6813 | + "jne .Lfill_cache\n\t" |
6814 | + "lfence\n" |
6815 | + :: [flush_pages] "r" (vmx_l1d_flush_pages), |
6816 | + [size] "r" (size) |
6817 | + : "eax", "ebx", "ecx", "edx"); |
6818 | +} |
6819 | + |
6820 | +static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
6821 | +{ |
6822 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
6823 | + int tpr_threshold; |
6824 | + |
6825 | + if (is_guest_mode(vcpu) && |
6826 | + nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) |
6827 | + return; |
6828 | + |
6829 | + tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr; |
6830 | + if (is_guest_mode(vcpu)) |
6831 | + to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold; |
6832 | + else |
6833 | + vmcs_write32(TPR_THRESHOLD, tpr_threshold); |
6834 | +} |
6835 | + |
6836 | +void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu) |
6837 | +{ |
6838 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6839 | + u32 sec_exec_control; |
6840 | + |
6841 | + if (!lapic_in_kernel(vcpu)) |
6842 | + return; |
6843 | + |
6844 | + if (!flexpriority_enabled && |
6845 | + !cpu_has_vmx_virtualize_x2apic_mode()) |
6846 | + return; |
6847 | + |
6848 | + /* Postpone execution until vmcs01 is the current VMCS. */ |
6849 | + if (is_guest_mode(vcpu)) { |
6850 | + vmx->nested.change_vmcs01_virtual_apic_mode = true; |
6851 | + return; |
6852 | + } |
6853 | + |
6854 | + sec_exec_control = secondary_exec_controls_get(vmx); |
6855 | + sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | |
6856 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE); |
6857 | + |
6858 | + switch (kvm_get_apic_mode(vcpu)) { |
6859 | + case LAPIC_MODE_INVALID: |
6860 | + WARN_ONCE(true, "Invalid local APIC state"); |
6861 | + case LAPIC_MODE_DISABLED: |
6862 | + break; |
6863 | + case LAPIC_MODE_XAPIC: |
6864 | + if (flexpriority_enabled) { |
6865 | + sec_exec_control |= |
6866 | + SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; |
6867 | + vmx_flush_tlb(vcpu, true); |
6868 | + } |
6869 | + break; |
6870 | + case LAPIC_MODE_X2APIC: |
6871 | + if (cpu_has_vmx_virtualize_x2apic_mode()) |
6872 | + sec_exec_control |= |
6873 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; |
6874 | + break; |
6875 | + } |
6876 | + secondary_exec_controls_set(vmx, sec_exec_control); |
6877 | + |
6878 | + vmx_update_msr_bitmap(vcpu); |
6879 | +} |
6880 | + |
6881 | +static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa) |
6882 | +{ |
6883 | + if (!is_guest_mode(vcpu)) { |
6884 | + vmcs_write64(APIC_ACCESS_ADDR, hpa); |
6885 | + vmx_flush_tlb(vcpu, true); |
6886 | + } |
6887 | +} |
6888 | + |
6889 | +static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr) |
6890 | +{ |
6891 | + u16 status; |
6892 | + u8 old; |
6893 | + |
6894 | + if (max_isr == -1) |
6895 | + max_isr = 0; |
6896 | + |
6897 | + status = vmcs_read16(GUEST_INTR_STATUS); |
6898 | + old = status >> 8; |
6899 | + if (max_isr != old) { |
6900 | + status &= 0xff; |
6901 | + status |= max_isr << 8; |
6902 | + vmcs_write16(GUEST_INTR_STATUS, status); |
6903 | + } |
6904 | +} |
6905 | + |
6906 | +static void vmx_set_rvi(int vector) |
6907 | +{ |
6908 | + u16 status; |
6909 | + u8 old; |
6910 | + |
6911 | + if (vector == -1) |
6912 | + vector = 0; |
6913 | + |
6914 | + status = vmcs_read16(GUEST_INTR_STATUS); |
6915 | + old = (u8)status & 0xff; |
6916 | + if ((u8)vector != old) { |
6917 | + status &= ~0xff; |
6918 | + status |= (u8)vector; |
6919 | + vmcs_write16(GUEST_INTR_STATUS, status); |
6920 | + } |
6921 | +} |
6922 | + |
6923 | +static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr) |
6924 | +{ |
6925 | + /* |
6926 | + * When running L2, updating RVI is only relevant when |
6927 | + * vmcs12 virtual-interrupt-delivery enabled. |
6928 | + * However, it can be enabled only when L1 also |
6929 | + * intercepts external-interrupts and in that case |
6930 | + * we should not update vmcs02 RVI but instead intercept |
6931 | + * interrupt. Therefore, do nothing when running L2. |
6932 | + */ |
6933 | + if (!is_guest_mode(vcpu)) |
6934 | + vmx_set_rvi(max_irr); |
6935 | +} |
6936 | + |
6937 | +static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu) |
6938 | +{ |
6939 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6940 | + int max_irr; |
6941 | + bool max_irr_updated; |
6942 | + |
6943 | + WARN_ON(!vcpu->arch.apicv_active); |
6944 | + if (pi_test_on(&vmx->pi_desc)) { |
6945 | + pi_clear_on(&vmx->pi_desc); |
6946 | + /* |
6947 | + * IOMMU can write to PID.ON, so the barrier matters even on UP. |
6948 | + * But on x86 this is just a compiler barrier anyway. |
6949 | + */ |
6950 | + smp_mb__after_atomic(); |
6951 | + max_irr_updated = |
6952 | + kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr); |
6953 | + |
6954 | + /* |
6955 | + * If we are running L2 and L1 has a new pending interrupt |
6956 | + * which can be injected, we should re-evaluate |
6957 | + * what should be done with this new L1 interrupt. |
6958 | + * If L1 intercepts external-interrupts, we should |
6959 | + * exit from L2 to L1. Otherwise, interrupt should be |
6960 | + * delivered directly to L2. |
6961 | + */ |
6962 | + if (is_guest_mode(vcpu) && max_irr_updated) { |
6963 | + if (nested_exit_on_intr(vcpu)) |
6964 | + kvm_vcpu_exiting_guest_mode(vcpu); |
6965 | + else |
6966 | + kvm_make_request(KVM_REQ_EVENT, vcpu); |
6967 | + } |
6968 | + } else { |
6969 | + max_irr = kvm_lapic_find_highest_irr(vcpu); |
6970 | + } |
6971 | + vmx_hwapic_irr_update(vcpu, max_irr); |
6972 | + return max_irr; |
6973 | +} |
6974 | + |
6975 | +static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu) |
6976 | +{ |
6977 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
6978 | + |
6979 | + return pi_test_on(pi_desc) || |
6980 | + (pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc)); |
6981 | +} |
6982 | + |
6983 | +static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap) |
6984 | +{ |
6985 | + if (!kvm_vcpu_apicv_active(vcpu)) |
6986 | + return; |
6987 | + |
6988 | + vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]); |
6989 | + vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]); |
6990 | + vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]); |
6991 | + vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]); |
6992 | +} |
6993 | + |
6994 | +static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu) |
6995 | +{ |
6996 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
6997 | + |
6998 | + pi_clear_on(&vmx->pi_desc); |
6999 | + memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir)); |
7000 | +} |
7001 | + |
7002 | +static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx) |
7003 | +{ |
7004 | + vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
7005 | + |
7006 | + /* if exit due to PF check for async PF */ |
7007 | + if (is_page_fault(vmx->exit_intr_info)) |
7008 | + vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason(); |
7009 | + |
7010 | + /* Handle machine checks before interrupts are enabled */ |
7011 | + if (is_machine_check(vmx->exit_intr_info)) |
7012 | + kvm_machine_check(); |
7013 | + |
7014 | + /* We need to handle NMIs before interrupts are enabled */ |
7015 | + if (is_nmi(vmx->exit_intr_info)) { |
7016 | + kvm_before_interrupt(&vmx->vcpu); |
7017 | + asm("int $2"); |
7018 | + kvm_after_interrupt(&vmx->vcpu); |
7019 | + } |
7020 | +} |
7021 | + |
7022 | +static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu) |
7023 | +{ |
7024 | + unsigned int vector; |
7025 | + unsigned long entry; |
7026 | +#ifdef CONFIG_X86_64 |
7027 | + unsigned long tmp; |
7028 | +#endif |
7029 | + gate_desc *desc; |
7030 | + u32 intr_info; |
7031 | + |
7032 | + intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
7033 | + if (WARN_ONCE(!is_external_intr(intr_info), |
7034 | + "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info)) |
7035 | + return; |
7036 | + |
7037 | + vector = intr_info & INTR_INFO_VECTOR_MASK; |
7038 | + desc = (gate_desc *)host_idt_base + vector; |
7039 | + entry = gate_offset(desc); |
7040 | + |
7041 | + kvm_before_interrupt(vcpu); |
7042 | + |
7043 | + asm volatile( |
7044 | +#ifdef CONFIG_X86_64 |
7045 | + "mov %%" _ASM_SP ", %[sp]\n\t" |
7046 | + "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t" |
7047 | + "push $%c[ss]\n\t" |
7048 | + "push %[sp]\n\t" |
7049 | +#endif |
7050 | + "pushf\n\t" |
7051 | + __ASM_SIZE(push) " $%c[cs]\n\t" |
7052 | + CALL_NOSPEC |
7053 | + : |
7054 | +#ifdef CONFIG_X86_64 |
7055 | + [sp]"=&r"(tmp), |
7056 | +#endif |
7057 | + ASM_CALL_CONSTRAINT |
7058 | + : |
7059 | + THUNK_TARGET(entry), |
7060 | + [ss]"i"(__KERNEL_DS), |
7061 | + [cs]"i"(__KERNEL_CS) |
7062 | + ); |
7063 | + |
7064 | + kvm_after_interrupt(vcpu); |
7065 | +} |
7066 | +STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff); |
7067 | + |
7068 | +static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu, |
7069 | + enum exit_fastpath_completion *exit_fastpath) |
7070 | +{ |
7071 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7072 | + |
7073 | + if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT) |
7074 | + handle_external_interrupt_irqoff(vcpu); |
7075 | + else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI) |
7076 | + handle_exception_nmi_irqoff(vmx); |
7077 | + else if (!is_guest_mode(vcpu) && |
7078 | + vmx->exit_reason == EXIT_REASON_MSR_WRITE) |
7079 | + *exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu); |
7080 | +} |
7081 | + |
7082 | +static bool vmx_has_emulated_msr(int index) |
7083 | +{ |
7084 | + switch (index) { |
7085 | + case MSR_IA32_SMBASE: |
7086 | + /* |
7087 | + * We cannot do SMM unless we can run the guest in big |
7088 | + * real mode. |
7089 | + */ |
7090 | + return enable_unrestricted_guest || emulate_invalid_guest_state; |
7091 | + case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: |
7092 | + return nested; |
7093 | + case MSR_AMD64_VIRT_SPEC_CTRL: |
7094 | + /* This is AMD only. */ |
7095 | + return false; |
7096 | + default: |
7097 | + return true; |
7098 | + } |
7099 | +} |
7100 | + |
7101 | +static bool vmx_pt_supported(void) |
7102 | +{ |
7103 | + return pt_mode == PT_MODE_HOST_GUEST; |
7104 | +} |
7105 | + |
7106 | +static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx) |
7107 | +{ |
7108 | + u32 exit_intr_info; |
7109 | + bool unblock_nmi; |
7110 | + u8 vector; |
7111 | + bool idtv_info_valid; |
7112 | + |
7113 | + idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK; |
7114 | + |
7115 | + if (enable_vnmi) { |
7116 | + if (vmx->loaded_vmcs->nmi_known_unmasked) |
7117 | + return; |
7118 | + /* |
7119 | + * Can't use vmx->exit_intr_info since we're not sure what |
7120 | + * the exit reason is. |
7121 | + */ |
7122 | + exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
7123 | + unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0; |
7124 | + vector = exit_intr_info & INTR_INFO_VECTOR_MASK; |
7125 | + /* |
7126 | + * SDM 3: 27.7.1.2 (September 2008) |
7127 | + * Re-set bit "block by NMI" before VM entry if vmexit caused by |
7128 | + * a guest IRET fault. |
7129 | + * SDM 3: 23.2.2 (September 2008) |
7130 | + * Bit 12 is undefined in any of the following cases: |
7131 | + * If the VM exit sets the valid bit in the IDT-vectoring |
7132 | + * information field. |
7133 | + * If the VM exit is due to a double fault. |
7134 | + */ |
7135 | + if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi && |
7136 | + vector != DF_VECTOR && !idtv_info_valid) |
7137 | + vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, |
7138 | + GUEST_INTR_STATE_NMI); |
7139 | + else |
7140 | + vmx->loaded_vmcs->nmi_known_unmasked = |
7141 | + !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) |
7142 | + & GUEST_INTR_STATE_NMI); |
7143 | + } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked)) |
7144 | + vmx->loaded_vmcs->vnmi_blocked_time += |
7145 | + ktime_to_ns(ktime_sub(ktime_get(), |
7146 | + vmx->loaded_vmcs->entry_time)); |
7147 | +} |
7148 | + |
7149 | +static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu, |
7150 | + u32 idt_vectoring_info, |
7151 | + int instr_len_field, |
7152 | + int error_code_field) |
7153 | +{ |
7154 | + u8 vector; |
7155 | + int type; |
7156 | + bool idtv_info_valid; |
7157 | + |
7158 | + idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK; |
7159 | + |
7160 | + vcpu->arch.nmi_injected = false; |
7161 | + kvm_clear_exception_queue(vcpu); |
7162 | + kvm_clear_interrupt_queue(vcpu); |
7163 | + |
7164 | + if (!idtv_info_valid) |
7165 | + return; |
7166 | + |
7167 | + kvm_make_request(KVM_REQ_EVENT, vcpu); |
7168 | + |
7169 | + vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK; |
7170 | + type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK; |
7171 | + |
7172 | + switch (type) { |
7173 | + case INTR_TYPE_NMI_INTR: |
7174 | + vcpu->arch.nmi_injected = true; |
7175 | + /* |
7176 | + * SDM 3: 27.7.1.2 (September 2008) |
7177 | + * Clear bit "block by NMI" before VM entry if a NMI |
7178 | + * delivery faulted. |
7179 | + */ |
7180 | + vmx_set_nmi_mask(vcpu, false); |
7181 | + break; |
7182 | + case INTR_TYPE_SOFT_EXCEPTION: |
7183 | + vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field); |
7184 | + /* fall through */ |
7185 | + case INTR_TYPE_HARD_EXCEPTION: |
7186 | + if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) { |
7187 | + u32 err = vmcs_read32(error_code_field); |
7188 | + kvm_requeue_exception_e(vcpu, vector, err); |
7189 | + } else |
7190 | + kvm_requeue_exception(vcpu, vector); |
7191 | + break; |
7192 | + case INTR_TYPE_SOFT_INTR: |
7193 | + vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field); |
7194 | + /* fall through */ |
7195 | + case INTR_TYPE_EXT_INTR: |
7196 | + kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR); |
7197 | + break; |
7198 | + default: |
7199 | + break; |
7200 | + } |
7201 | +} |
7202 | + |
7203 | +static void vmx_complete_interrupts(struct vcpu_vmx *vmx) |
7204 | +{ |
7205 | + __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info, |
7206 | + VM_EXIT_INSTRUCTION_LEN, |
7207 | + IDT_VECTORING_ERROR_CODE); |
7208 | +} |
7209 | + |
7210 | +static void vmx_cancel_injection(struct kvm_vcpu *vcpu) |
7211 | +{ |
7212 | + __vmx_complete_interrupts(vcpu, |
7213 | + vmcs_read32(VM_ENTRY_INTR_INFO_FIELD), |
7214 | + VM_ENTRY_INSTRUCTION_LEN, |
7215 | + VM_ENTRY_EXCEPTION_ERROR_CODE); |
7216 | + |
7217 | + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); |
7218 | +} |
7219 | + |
7220 | +static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx) |
7221 | +{ |
7222 | + int i, nr_msrs; |
7223 | + struct perf_guest_switch_msr *msrs; |
7224 | + |
7225 | + msrs = perf_guest_get_msrs(&nr_msrs); |
7226 | + |
7227 | + if (!msrs) |
7228 | + return; |
7229 | + |
7230 | + for (i = 0; i < nr_msrs; i++) |
7231 | + if (msrs[i].host == msrs[i].guest) |
7232 | + clear_atomic_switch_msr(vmx, msrs[i].msr); |
7233 | + else |
7234 | + add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest, |
7235 | + msrs[i].host, false); |
7236 | +} |
7237 | + |
7238 | +static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx) |
7239 | +{ |
7240 | + u32 host_umwait_control; |
7241 | + |
7242 | + if (!vmx_has_waitpkg(vmx)) |
7243 | + return; |
7244 | + |
7245 | + host_umwait_control = get_umwait_control_msr(); |
7246 | + |
7247 | + if (vmx->msr_ia32_umwait_control != host_umwait_control) |
7248 | + add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL, |
7249 | + vmx->msr_ia32_umwait_control, |
7250 | + host_umwait_control, false); |
7251 | + else |
7252 | + clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL); |
7253 | +} |
7254 | + |
7255 | +static void vmx_update_hv_timer(struct kvm_vcpu *vcpu) |
7256 | +{ |
7257 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7258 | + u64 tscl; |
7259 | + u32 delta_tsc; |
7260 | + |
7261 | + if (vmx->req_immediate_exit) { |
7262 | + vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0); |
7263 | + vmx->loaded_vmcs->hv_timer_soft_disabled = false; |
7264 | + } else if (vmx->hv_deadline_tsc != -1) { |
7265 | + tscl = rdtsc(); |
7266 | + if (vmx->hv_deadline_tsc > tscl) |
7267 | + /* set_hv_timer ensures the delta fits in 32-bits */ |
7268 | + delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >> |
7269 | + cpu_preemption_timer_multi); |
7270 | + else |
7271 | + delta_tsc = 0; |
7272 | + |
7273 | + vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc); |
7274 | + vmx->loaded_vmcs->hv_timer_soft_disabled = false; |
7275 | + } else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) { |
7276 | + vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1); |
7277 | + vmx->loaded_vmcs->hv_timer_soft_disabled = true; |
7278 | + } |
7279 | +} |
7280 | + |
7281 | +void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp) |
7282 | +{ |
7283 | + if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) { |
7284 | + vmx->loaded_vmcs->host_state.rsp = host_rsp; |
7285 | + vmcs_writel(HOST_RSP, host_rsp); |
7286 | + } |
7287 | +} |
7288 | + |
7289 | +bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched); |
7290 | + |
7291 | +static void vmx_vcpu_run(struct kvm_vcpu *vcpu) |
7292 | +{ |
7293 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7294 | + unsigned long cr3, cr4; |
7295 | + |
7296 | + /* Record the guest's net vcpu time for enforced NMI injections. */ |
7297 | + if (unlikely(!enable_vnmi && |
7298 | + vmx->loaded_vmcs->soft_vnmi_blocked)) |
7299 | + vmx->loaded_vmcs->entry_time = ktime_get(); |
7300 | + |
7301 | + /* Don't enter VMX if guest state is invalid, let the exit handler |
7302 | + start emulation until we arrive back to a valid state */ |
7303 | + if (vmx->emulation_required) |
7304 | + return; |
7305 | + |
7306 | + if (vmx->ple_window_dirty) { |
7307 | + vmx->ple_window_dirty = false; |
7308 | + vmcs_write32(PLE_WINDOW, vmx->ple_window); |
7309 | + } |
7310 | + |
7311 | + if (vmx->nested.need_vmcs12_to_shadow_sync) |
7312 | + nested_sync_vmcs12_to_shadow(vcpu); |
7313 | + |
7314 | + if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP)) |
7315 | + vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]); |
7316 | + if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP)) |
7317 | + vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]); |
7318 | + |
7319 | + cr3 = __get_current_cr3_fast(); |
7320 | + if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) { |
7321 | + vmcs_writel(HOST_CR3, cr3); |
7322 | + vmx->loaded_vmcs->host_state.cr3 = cr3; |
7323 | + } |
7324 | + |
7325 | + cr4 = cr4_read_shadow(); |
7326 | + if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) { |
7327 | + vmcs_writel(HOST_CR4, cr4); |
7328 | + vmx->loaded_vmcs->host_state.cr4 = cr4; |
7329 | + } |
7330 | + |
7331 | + /* When single-stepping over STI and MOV SS, we must clear the |
7332 | + * corresponding interruptibility bits in the guest state. Otherwise |
7333 | + * vmentry fails as it then expects bit 14 (BS) in pending debug |
7334 | + * exceptions being set, but that's not correct for the guest debugging |
7335 | + * case. */ |
7336 | + if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
7337 | + vmx_set_interrupt_shadow(vcpu, 0); |
7338 | + |
7339 | + kvm_load_guest_xsave_state(vcpu); |
7340 | + |
7341 | + if (static_cpu_has(X86_FEATURE_PKU) && |
7342 | + kvm_read_cr4_bits(vcpu, X86_CR4_PKE) && |
7343 | + vcpu->arch.pkru != vmx->host_pkru) |
7344 | + __write_pkru(vcpu->arch.pkru); |
7345 | + |
7346 | + pt_guest_enter(vmx); |
7347 | + |
7348 | + atomic_switch_perf_msrs(vmx); |
7349 | + atomic_switch_umwait_control_msr(vmx); |
7350 | + |
7351 | + if (enable_preemption_timer) |
7352 | + vmx_update_hv_timer(vcpu); |
7353 | + |
7354 | + if (lapic_in_kernel(vcpu) && |
7355 | + vcpu->arch.apic->lapic_timer.timer_advance_ns) |
7356 | + kvm_wait_lapic_expire(vcpu); |
7357 | + |
7358 | + /* |
7359 | + * If this vCPU has touched SPEC_CTRL, restore the guest's value if |
7360 | + * it's non-zero. Since vmentry is serialising on affected CPUs, there |
7361 | + * is no need to worry about the conditional branch over the wrmsr |
7362 | + * being speculatively taken. |
7363 | + */ |
7364 | + x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0); |
7365 | + |
7366 | + /* L1D Flush includes CPU buffer clear to mitigate MDS */ |
7367 | + if (static_branch_unlikely(&vmx_l1d_should_flush)) |
7368 | + vmx_l1d_flush(vcpu); |
7369 | + else if (static_branch_unlikely(&mds_user_clear)) |
7370 | + mds_clear_cpu_buffers(); |
7371 | + |
7372 | + if (vcpu->arch.cr2 != read_cr2()) |
7373 | + write_cr2(vcpu->arch.cr2); |
7374 | + |
7375 | + vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs, |
7376 | + vmx->loaded_vmcs->launched); |
7377 | + |
7378 | + vcpu->arch.cr2 = read_cr2(); |
7379 | + |
7380 | + /* |
7381 | + * We do not use IBRS in the kernel. If this vCPU has used the |
7382 | + * SPEC_CTRL MSR it may have left it on; save the value and |
7383 | + * turn it off. This is much more efficient than blindly adding |
7384 | + * it to the atomic save/restore list. Especially as the former |
7385 | + * (Saving guest MSRs on vmexit) doesn't even exist in KVM. |
7386 | + * |
7387 | + * For non-nested case: |
7388 | + * If the L01 MSR bitmap does not intercept the MSR, then we need to |
7389 | + * save it. |
7390 | + * |
7391 | + * For nested case: |
7392 | + * If the L02 MSR bitmap does not intercept the MSR, then we need to |
7393 | + * save it. |
7394 | + */ |
7395 | + if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))) |
7396 | + vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL); |
7397 | + |
7398 | + x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0); |
7399 | + |
7400 | + /* All fields are clean at this point */ |
7401 | + if (static_branch_unlikely(&enable_evmcs)) |
7402 | + current_evmcs->hv_clean_fields |= |
7403 | + HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL; |
7404 | + |
7405 | + if (static_branch_unlikely(&enable_evmcs)) |
7406 | + current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index; |
7407 | + |
7408 | + /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */ |
7409 | + if (vmx->host_debugctlmsr) |
7410 | + update_debugctlmsr(vmx->host_debugctlmsr); |
7411 | + |
7412 | +#ifndef CONFIG_X86_64 |
7413 | + /* |
7414 | + * The sysexit path does not restore ds/es, so we must set them to |
7415 | + * a reasonable value ourselves. |
7416 | + * |
7417 | + * We can't defer this to vmx_prepare_switch_to_host() since that |
7418 | + * function may be executed in interrupt context, which saves and |
7419 | + * restore segments around it, nullifying its effect. |
7420 | + */ |
7421 | + loadsegment(ds, __USER_DS); |
7422 | + loadsegment(es, __USER_DS); |
7423 | +#endif |
7424 | + |
7425 | + vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP) |
7426 | + | (1 << VCPU_EXREG_RFLAGS) |
7427 | + | (1 << VCPU_EXREG_PDPTR) |
7428 | + | (1 << VCPU_EXREG_SEGMENTS) |
7429 | + | (1 << VCPU_EXREG_CR3)); |
7430 | + vcpu->arch.regs_dirty = 0; |
7431 | + |
7432 | + pt_guest_exit(vmx); |
7433 | + |
7434 | + /* |
7435 | + * eager fpu is enabled if PKEY is supported and CR4 is switched |
7436 | + * back on host, so it is safe to read guest PKRU from current |
7437 | + * XSAVE. |
7438 | + */ |
7439 | + if (static_cpu_has(X86_FEATURE_PKU) && |
7440 | + kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) { |
7441 | + vcpu->arch.pkru = rdpkru(); |
7442 | + if (vcpu->arch.pkru != vmx->host_pkru) |
7443 | + __write_pkru(vmx->host_pkru); |
7444 | + } |
7445 | + |
7446 | + kvm_load_host_xsave_state(vcpu); |
7447 | + |
7448 | + vmx->nested.nested_run_pending = 0; |
7449 | + vmx->idt_vectoring_info = 0; |
7450 | + |
7451 | + vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON); |
7452 | + if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY) |
7453 | + kvm_machine_check(); |
7454 | + |
7455 | + if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) |
7456 | + return; |
7457 | + |
7458 | + vmx->loaded_vmcs->launched = 1; |
7459 | + vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD); |
7460 | + |
7461 | + vmx_recover_nmi_blocking(vmx); |
7462 | + vmx_complete_interrupts(vmx); |
7463 | +} |
7464 | + |
7465 | +static struct kvm *vmx_vm_alloc(void) |
7466 | +{ |
7467 | + struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx), |
7468 | + GFP_KERNEL_ACCOUNT | __GFP_ZERO, |
7469 | + PAGE_KERNEL); |
7470 | + return &kvm_vmx->kvm; |
7471 | +} |
7472 | + |
7473 | +static void vmx_vm_free(struct kvm *kvm) |
7474 | +{ |
7475 | + kfree(kvm->arch.hyperv.hv_pa_pg); |
7476 | + vfree(to_kvm_vmx(kvm)); |
7477 | +} |
7478 | + |
7479 | +static void vmx_free_vcpu(struct kvm_vcpu *vcpu) |
7480 | +{ |
7481 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7482 | + |
7483 | + if (enable_pml) |
7484 | + vmx_destroy_pml_buffer(vmx); |
7485 | + free_vpid(vmx->vpid); |
7486 | + nested_vmx_free_vcpu(vcpu); |
7487 | + free_loaded_vmcs(vmx->loaded_vmcs); |
7488 | + kvm_vcpu_uninit(vcpu); |
7489 | + kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu); |
7490 | + kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu); |
7491 | + kmem_cache_free(kvm_vcpu_cache, vmx); |
7492 | +} |
7493 | + |
7494 | +static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id) |
7495 | +{ |
7496 | + int err; |
7497 | + struct vcpu_vmx *vmx; |
7498 | + unsigned long *msr_bitmap; |
7499 | + int i, cpu; |
7500 | + |
7501 | + BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0, |
7502 | + "struct kvm_vcpu must be at offset 0 for arch usercopy region"); |
7503 | + |
7504 | + vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
7505 | + if (!vmx) |
7506 | + return ERR_PTR(-ENOMEM); |
7507 | + |
7508 | + vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, |
7509 | + GFP_KERNEL_ACCOUNT); |
7510 | + if (!vmx->vcpu.arch.user_fpu) { |
7511 | + printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n"); |
7512 | + err = -ENOMEM; |
7513 | + goto free_partial_vcpu; |
7514 | + } |
7515 | + |
7516 | + vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, |
7517 | + GFP_KERNEL_ACCOUNT); |
7518 | + if (!vmx->vcpu.arch.guest_fpu) { |
7519 | + printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n"); |
7520 | + err = -ENOMEM; |
7521 | + goto free_user_fpu; |
7522 | + } |
7523 | + |
7524 | + vmx->vpid = allocate_vpid(); |
7525 | + |
7526 | + err = kvm_vcpu_init(&vmx->vcpu, kvm, id); |
7527 | + if (err) |
7528 | + goto free_vcpu; |
7529 | + |
7530 | + err = -ENOMEM; |
7531 | + |
7532 | + /* |
7533 | + * If PML is turned on, failure on enabling PML just results in failure |
7534 | + * of creating the vcpu, therefore we can simplify PML logic (by |
7535 | + * avoiding dealing with cases, such as enabling PML partially on vcpus |
7536 | + * for the guest), etc. |
7537 | + */ |
7538 | + if (enable_pml) { |
7539 | + vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
7540 | + if (!vmx->pml_pg) |
7541 | + goto uninit_vcpu; |
7542 | + } |
7543 | + |
7544 | + BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS); |
7545 | + |
7546 | + for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) { |
7547 | + u32 index = vmx_msr_index[i]; |
7548 | + u32 data_low, data_high; |
7549 | + int j = vmx->nmsrs; |
7550 | + |
7551 | + if (rdmsr_safe(index, &data_low, &data_high) < 0) |
7552 | + continue; |
7553 | + if (wrmsr_safe(index, data_low, data_high) < 0) |
7554 | + continue; |
7555 | + |
7556 | + vmx->guest_msrs[j].index = i; |
7557 | + vmx->guest_msrs[j].data = 0; |
7558 | + switch (index) { |
7559 | + case MSR_IA32_TSX_CTRL: |
7560 | + /* |
7561 | + * No need to pass TSX_CTRL_CPUID_CLEAR through, so |
7562 | + * let's avoid changing CPUID bits under the host |
7563 | + * kernel's feet. |
7564 | + */ |
7565 | + vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR; |
7566 | + break; |
7567 | + default: |
7568 | + vmx->guest_msrs[j].mask = -1ull; |
7569 | + break; |
7570 | + } |
7571 | + ++vmx->nmsrs; |
7572 | + } |
7573 | + |
7574 | + err = alloc_loaded_vmcs(&vmx->vmcs01); |
7575 | + if (err < 0) |
7576 | + goto free_pml; |
7577 | + |
7578 | + msr_bitmap = vmx->vmcs01.msr_bitmap; |
7579 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R); |
7580 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW); |
7581 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW); |
7582 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW); |
7583 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW); |
7584 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW); |
7585 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW); |
7586 | + if (kvm_cstate_in_guest(kvm)) { |
7587 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R); |
7588 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R); |
7589 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R); |
7590 | + vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R); |
7591 | + } |
7592 | + vmx->msr_bitmap_mode = 0; |
7593 | + |
7594 | + vmx->loaded_vmcs = &vmx->vmcs01; |
7595 | + cpu = get_cpu(); |
7596 | + vmx_vcpu_load(&vmx->vcpu, cpu); |
7597 | + vmx->vcpu.cpu = cpu; |
7598 | + init_vmcs(vmx); |
7599 | + vmx_vcpu_put(&vmx->vcpu); |
7600 | + put_cpu(); |
7601 | + if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) { |
7602 | + err = alloc_apic_access_page(kvm); |
7603 | + if (err) |
7604 | + goto free_vmcs; |
7605 | + } |
7606 | + |
7607 | + if (enable_ept && !enable_unrestricted_guest) { |
7608 | + err = init_rmode_identity_map(kvm); |
7609 | + if (err) |
7610 | + goto free_vmcs; |
7611 | + } |
7612 | + |
7613 | + if (nested) |
7614 | + nested_vmx_setup_ctls_msrs(&vmx->nested.msrs, |
7615 | + vmx_capability.ept, |
7616 | + kvm_vcpu_apicv_active(&vmx->vcpu)); |
7617 | + else |
7618 | + memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs)); |
7619 | + |
7620 | + vmx->nested.posted_intr_nv = -1; |
7621 | + vmx->nested.current_vmptr = -1ull; |
7622 | + |
7623 | + vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED; |
7624 | + |
7625 | + /* |
7626 | + * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR |
7627 | + * or POSTED_INTR_WAKEUP_VECTOR. |
7628 | + */ |
7629 | + vmx->pi_desc.nv = POSTED_INTR_VECTOR; |
7630 | + vmx->pi_desc.sn = 1; |
7631 | + |
7632 | + vmx->ept_pointer = INVALID_PAGE; |
7633 | + |
7634 | + return &vmx->vcpu; |
7635 | + |
7636 | +free_vmcs: |
7637 | + free_loaded_vmcs(vmx->loaded_vmcs); |
7638 | +free_pml: |
7639 | + vmx_destroy_pml_buffer(vmx); |
7640 | +uninit_vcpu: |
7641 | + kvm_vcpu_uninit(&vmx->vcpu); |
7642 | +free_vcpu: |
7643 | + free_vpid(vmx->vpid); |
7644 | + kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu); |
7645 | +free_user_fpu: |
7646 | + kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu); |
7647 | +free_partial_vcpu: |
7648 | + kmem_cache_free(kvm_vcpu_cache, vmx); |
7649 | + return ERR_PTR(err); |
7650 | +} |
7651 | + |
7652 | +#define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n" |
7653 | +#define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n" |
7654 | + |
7655 | +static int vmx_vm_init(struct kvm *kvm) |
7656 | +{ |
7657 | + spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock); |
7658 | + |
7659 | + if (!ple_gap) |
7660 | + kvm->arch.pause_in_guest = true; |
7661 | + |
7662 | + if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) { |
7663 | + switch (l1tf_mitigation) { |
7664 | + case L1TF_MITIGATION_OFF: |
7665 | + case L1TF_MITIGATION_FLUSH_NOWARN: |
7666 | + /* 'I explicitly don't care' is set */ |
7667 | + break; |
7668 | + case L1TF_MITIGATION_FLUSH: |
7669 | + case L1TF_MITIGATION_FLUSH_NOSMT: |
7670 | + case L1TF_MITIGATION_FULL: |
7671 | + /* |
7672 | + * Warn upon starting the first VM in a potentially |
7673 | + * insecure environment. |
7674 | + */ |
7675 | + if (sched_smt_active()) |
7676 | + pr_warn_once(L1TF_MSG_SMT); |
7677 | + if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER) |
7678 | + pr_warn_once(L1TF_MSG_L1D); |
7679 | + break; |
7680 | + case L1TF_MITIGATION_FULL_FORCE: |
7681 | + /* Flush is enforced */ |
7682 | + break; |
7683 | + } |
7684 | + } |
7685 | + return 0; |
7686 | +} |
7687 | + |
7688 | +static int __init vmx_check_processor_compat(void) |
7689 | +{ |
7690 | + struct vmcs_config vmcs_conf; |
7691 | + struct vmx_capability vmx_cap; |
7692 | + |
7693 | + if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0) |
7694 | + return -EIO; |
7695 | + if (nested) |
7696 | + nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept, |
7697 | + enable_apicv); |
7698 | + if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) { |
7699 | + printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n", |
7700 | + smp_processor_id()); |
7701 | + return -EIO; |
7702 | + } |
7703 | + return 0; |
7704 | +} |
7705 | + |
7706 | +static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
7707 | +{ |
7708 | + u8 cache; |
7709 | + u64 ipat = 0; |
7710 | + |
7711 | + /* For VT-d and EPT combination |
7712 | + * 1. MMIO: always map as UC |
7713 | + * 2. EPT with VT-d: |
7714 | + * a. VT-d without snooping control feature: can't guarantee the |
7715 | + * result, try to trust guest. |
7716 | + * b. VT-d with snooping control feature: snooping control feature of |
7717 | + * VT-d engine can guarantee the cache correctness. Just set it |
7718 | + * to WB to keep consistent with host. So the same as item 3. |
7719 | + * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep |
7720 | + * consistent with host MTRR |
7721 | + */ |
7722 | + if (is_mmio) { |
7723 | + cache = MTRR_TYPE_UNCACHABLE; |
7724 | + goto exit; |
7725 | + } |
7726 | + |
7727 | + if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) { |
7728 | + ipat = VMX_EPT_IPAT_BIT; |
7729 | + cache = MTRR_TYPE_WRBACK; |
7730 | + goto exit; |
7731 | + } |
7732 | + |
7733 | + if (kvm_read_cr0(vcpu) & X86_CR0_CD) { |
7734 | + ipat = VMX_EPT_IPAT_BIT; |
7735 | + if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) |
7736 | + cache = MTRR_TYPE_WRBACK; |
7737 | + else |
7738 | + cache = MTRR_TYPE_UNCACHABLE; |
7739 | + goto exit; |
7740 | + } |
7741 | + |
7742 | + cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn); |
7743 | + |
7744 | +exit: |
7745 | + return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat; |
7746 | +} |
7747 | + |
7748 | +static int vmx_get_lpage_level(void) |
7749 | +{ |
7750 | + if (enable_ept && !cpu_has_vmx_ept_1g_page()) |
7751 | + return PT_DIRECTORY_LEVEL; |
7752 | + else |
7753 | + /* For shadow and EPT supported 1GB page */ |
7754 | + return PT_PDPE_LEVEL; |
7755 | +} |
7756 | + |
7757 | +static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx) |
7758 | +{ |
7759 | + /* |
7760 | + * These bits in the secondary execution controls field |
7761 | + * are dynamic, the others are mostly based on the hypervisor |
7762 | + * architecture and the guest's CPUID. Do not touch the |
7763 | + * dynamic bits. |
7764 | + */ |
7765 | + u32 mask = |
7766 | + SECONDARY_EXEC_SHADOW_VMCS | |
7767 | + SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
7768 | + SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | |
7769 | + SECONDARY_EXEC_DESC; |
7770 | + |
7771 | + u32 new_ctl = vmx->secondary_exec_control; |
7772 | + u32 cur_ctl = secondary_exec_controls_get(vmx); |
7773 | + |
7774 | + secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask)); |
7775 | +} |
7776 | + |
7777 | +/* |
7778 | + * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits |
7779 | + * (indicating "allowed-1") if they are supported in the guest's CPUID. |
7780 | + */ |
7781 | +static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu) |
7782 | +{ |
7783 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7784 | + struct kvm_cpuid_entry2 *entry; |
7785 | + |
7786 | + vmx->nested.msrs.cr0_fixed1 = 0xffffffff; |
7787 | + vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE; |
7788 | + |
7789 | +#define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \ |
7790 | + if (entry && (entry->_reg & (_cpuid_mask))) \ |
7791 | + vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask); \ |
7792 | +} while (0) |
7793 | + |
7794 | + entry = kvm_find_cpuid_entry(vcpu, 0x1, 0); |
7795 | + cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME)); |
7796 | + cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME)); |
7797 | + cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC)); |
7798 | + cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE)); |
7799 | + cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE)); |
7800 | + cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE)); |
7801 | + cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE)); |
7802 | + cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE)); |
7803 | + cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR)); |
7804 | + cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM)); |
7805 | + cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX)); |
7806 | + cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX)); |
7807 | + cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID)); |
7808 | + cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE)); |
7809 | + |
7810 | + entry = kvm_find_cpuid_entry(vcpu, 0x7, 0); |
7811 | + cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE)); |
7812 | + cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP)); |
7813 | + cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP)); |
7814 | + cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU)); |
7815 | + cr4_fixed1_update(X86_CR4_UMIP, ecx, bit(X86_FEATURE_UMIP)); |
7816 | + cr4_fixed1_update(X86_CR4_LA57, ecx, bit(X86_FEATURE_LA57)); |
7817 | + |
7818 | +#undef cr4_fixed1_update |
7819 | +} |
7820 | + |
7821 | +static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu) |
7822 | +{ |
7823 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7824 | + |
7825 | + if (kvm_mpx_supported()) { |
7826 | + bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX); |
7827 | + |
7828 | + if (mpx_enabled) { |
7829 | + vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS; |
7830 | + vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS; |
7831 | + } else { |
7832 | + vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS; |
7833 | + vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS; |
7834 | + } |
7835 | + } |
7836 | +} |
7837 | + |
7838 | +static void update_intel_pt_cfg(struct kvm_vcpu *vcpu) |
7839 | +{ |
7840 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7841 | + struct kvm_cpuid_entry2 *best = NULL; |
7842 | + int i; |
7843 | + |
7844 | + for (i = 0; i < PT_CPUID_LEAVES; i++) { |
7845 | + best = kvm_find_cpuid_entry(vcpu, 0x14, i); |
7846 | + if (!best) |
7847 | + return; |
7848 | + vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax; |
7849 | + vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx; |
7850 | + vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx; |
7851 | + vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx; |
7852 | + } |
7853 | + |
7854 | + /* Get the number of configurable Address Ranges for filtering */ |
7855 | + vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps, |
7856 | + PT_CAP_num_address_ranges); |
7857 | + |
7858 | + /* Initialize and clear the no dependency bits */ |
7859 | + vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS | |
7860 | + RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC); |
7861 | + |
7862 | + /* |
7863 | + * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise |
7864 | + * will inject an #GP |
7865 | + */ |
7866 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering)) |
7867 | + vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN; |
7868 | + |
7869 | + /* |
7870 | + * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and |
7871 | + * PSBFreq can be set |
7872 | + */ |
7873 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc)) |
7874 | + vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC | |
7875 | + RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ); |
7876 | + |
7877 | + /* |
7878 | + * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and |
7879 | + * MTCFreq can be set |
7880 | + */ |
7881 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc)) |
7882 | + vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN | |
7883 | + RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE); |
7884 | + |
7885 | + /* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */ |
7886 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite)) |
7887 | + vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW | |
7888 | + RTIT_CTL_PTW_EN); |
7889 | + |
7890 | + /* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */ |
7891 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace)) |
7892 | + vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN; |
7893 | + |
7894 | + /* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */ |
7895 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output)) |
7896 | + vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA; |
7897 | + |
7898 | + /* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */ |
7899 | + if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys)) |
7900 | + vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN; |
7901 | + |
7902 | + /* unmask address range configure area */ |
7903 | + for (i = 0; i < vmx->pt_desc.addr_range; i++) |
7904 | + vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4)); |
7905 | +} |
7906 | + |
7907 | +static void vmx_cpuid_update(struct kvm_vcpu *vcpu) |
7908 | +{ |
7909 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
7910 | + |
7911 | + /* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */ |
7912 | + vcpu->arch.xsaves_enabled = false; |
7913 | + |
7914 | + if (cpu_has_secondary_exec_ctrls()) { |
7915 | + vmx_compute_secondary_exec_control(vmx); |
7916 | + vmcs_set_secondary_exec_control(vmx); |
7917 | + } |
7918 | + |
7919 | + if (nested_vmx_allowed(vcpu)) |
7920 | + to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |= |
7921 | + FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX | |
7922 | + FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX; |
7923 | + else |
7924 | + to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &= |
7925 | + ~(FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX | |
7926 | + FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX); |
7927 | + |
7928 | + if (nested_vmx_allowed(vcpu)) { |
7929 | + nested_vmx_cr_fixed1_bits_update(vcpu); |
7930 | + nested_vmx_entry_exit_ctls_update(vcpu); |
7931 | + } |
7932 | + |
7933 | + if (boot_cpu_has(X86_FEATURE_INTEL_PT) && |
7934 | + guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT)) |
7935 | + update_intel_pt_cfg(vcpu); |
7936 | + |
7937 | + if (boot_cpu_has(X86_FEATURE_RTM)) { |
7938 | + struct shared_msr_entry *msr; |
7939 | + msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL); |
7940 | + if (msr) { |
7941 | + bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM); |
7942 | + vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE); |
7943 | + } |
7944 | + } |
7945 | +} |
7946 | + |
7947 | +static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) |
7948 | +{ |
7949 | + if (func == 1 && nested) |
7950 | + entry->ecx |= bit(X86_FEATURE_VMX); |
7951 | +} |
7952 | + |
7953 | +static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu) |
7954 | +{ |
7955 | + to_vmx(vcpu)->req_immediate_exit = true; |
7956 | +} |
7957 | + |
7958 | +static int vmx_check_intercept(struct kvm_vcpu *vcpu, |
7959 | + struct x86_instruction_info *info, |
7960 | + enum x86_intercept_stage stage) |
7961 | +{ |
7962 | + struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
7963 | + struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
7964 | + |
7965 | + /* |
7966 | + * RDPID causes #UD if disabled through secondary execution controls. |
7967 | + * Because it is marked as EmulateOnUD, we need to intercept it here. |
7968 | + */ |
7969 | + if (info->intercept == x86_intercept_rdtscp && |
7970 | + !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) { |
7971 | + ctxt->exception.vector = UD_VECTOR; |
7972 | + ctxt->exception.error_code_valid = false; |
7973 | + return X86EMUL_PROPAGATE_FAULT; |
7974 | + } |
7975 | + |
7976 | + /* TODO: check more intercepts... */ |
7977 | + return X86EMUL_CONTINUE; |
7978 | +} |
7979 | + |
7980 | +#ifdef CONFIG_X86_64 |
7981 | +/* (a << shift) / divisor, return 1 if overflow otherwise 0 */ |
7982 | +static inline int u64_shl_div_u64(u64 a, unsigned int shift, |
7983 | + u64 divisor, u64 *result) |
7984 | +{ |
7985 | + u64 low = a << shift, high = a >> (64 - shift); |
7986 | + |
7987 | + /* To avoid the overflow on divq */ |
7988 | + if (high >= divisor) |
7989 | + return 1; |
7990 | + |
7991 | + /* Low hold the result, high hold rem which is discarded */ |
7992 | + asm("divq %2\n\t" : "=a" (low), "=d" (high) : |
7993 | + "rm" (divisor), "0" (low), "1" (high)); |
7994 | + *result = low; |
7995 | + |
7996 | + return 0; |
7997 | +} |
7998 | + |
7999 | +static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc, |
8000 | + bool *expired) |
8001 | +{ |
8002 | + struct vcpu_vmx *vmx; |
8003 | + u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles; |
8004 | + struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer; |
8005 | + |
8006 | + if (kvm_mwait_in_guest(vcpu->kvm) || |
8007 | + kvm_can_post_timer_interrupt(vcpu)) |
8008 | + return -EOPNOTSUPP; |
8009 | + |
8010 | + vmx = to_vmx(vcpu); |
8011 | + tscl = rdtsc(); |
8012 | + guest_tscl = kvm_read_l1_tsc(vcpu, tscl); |
8013 | + delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl; |
8014 | + lapic_timer_advance_cycles = nsec_to_cycles(vcpu, |
8015 | + ktimer->timer_advance_ns); |
8016 | + |
8017 | + if (delta_tsc > lapic_timer_advance_cycles) |
8018 | + delta_tsc -= lapic_timer_advance_cycles; |
8019 | + else |
8020 | + delta_tsc = 0; |
8021 | + |
8022 | + /* Convert to host delta tsc if tsc scaling is enabled */ |
8023 | + if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio && |
8024 | + delta_tsc && u64_shl_div_u64(delta_tsc, |
8025 | + kvm_tsc_scaling_ratio_frac_bits, |
8026 | + vcpu->arch.tsc_scaling_ratio, &delta_tsc)) |
8027 | + return -ERANGE; |
8028 | + |
8029 | + /* |
8030 | + * If the delta tsc can't fit in the 32 bit after the multi shift, |
8031 | + * we can't use the preemption timer. |
8032 | + * It's possible that it fits on later vmentries, but checking |
8033 | + * on every vmentry is costly so we just use an hrtimer. |
8034 | + */ |
8035 | + if (delta_tsc >> (cpu_preemption_timer_multi + 32)) |
8036 | + return -ERANGE; |
8037 | + |
8038 | + vmx->hv_deadline_tsc = tscl + delta_tsc; |
8039 | + *expired = !delta_tsc; |
8040 | + return 0; |
8041 | +} |
8042 | + |
8043 | +static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu) |
8044 | +{ |
8045 | + to_vmx(vcpu)->hv_deadline_tsc = -1; |
8046 | +} |
8047 | +#endif |
8048 | + |
8049 | +static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu) |
8050 | +{ |
8051 | + if (!kvm_pause_in_guest(vcpu->kvm)) |
8052 | + shrink_ple_window(vcpu); |
8053 | +} |
8054 | + |
8055 | +static void vmx_slot_enable_log_dirty(struct kvm *kvm, |
8056 | + struct kvm_memory_slot *slot) |
8057 | +{ |
8058 | + kvm_mmu_slot_leaf_clear_dirty(kvm, slot); |
8059 | + kvm_mmu_slot_largepage_remove_write_access(kvm, slot); |
8060 | +} |
8061 | + |
8062 | +static void vmx_slot_disable_log_dirty(struct kvm *kvm, |
8063 | + struct kvm_memory_slot *slot) |
8064 | +{ |
8065 | + kvm_mmu_slot_set_dirty(kvm, slot); |
8066 | +} |
8067 | + |
8068 | +static void vmx_flush_log_dirty(struct kvm *kvm) |
8069 | +{ |
8070 | + kvm_flush_pml_buffers(kvm); |
8071 | +} |
8072 | + |
8073 | +static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu) |
8074 | +{ |
8075 | + struct vmcs12 *vmcs12; |
8076 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
8077 | + gpa_t gpa, dst; |
8078 | + |
8079 | + if (is_guest_mode(vcpu)) { |
8080 | + WARN_ON_ONCE(vmx->nested.pml_full); |
8081 | + |
8082 | + /* |
8083 | + * Check if PML is enabled for the nested guest. |
8084 | + * Whether eptp bit 6 is set is already checked |
8085 | + * as part of A/D emulation. |
8086 | + */ |
8087 | + vmcs12 = get_vmcs12(vcpu); |
8088 | + if (!nested_cpu_has_pml(vmcs12)) |
8089 | + return 0; |
8090 | + |
8091 | + if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) { |
8092 | + vmx->nested.pml_full = true; |
8093 | + return 1; |
8094 | + } |
8095 | + |
8096 | + gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull; |
8097 | + dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index; |
8098 | + |
8099 | + if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa, |
8100 | + offset_in_page(dst), sizeof(gpa))) |
8101 | + return 0; |
8102 | + |
8103 | + vmcs12->guest_pml_index--; |
8104 | + } |
8105 | + |
8106 | + return 0; |
8107 | +} |
8108 | + |
8109 | +static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm, |
8110 | + struct kvm_memory_slot *memslot, |
8111 | + gfn_t offset, unsigned long mask) |
8112 | +{ |
8113 | + kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask); |
8114 | +} |
8115 | + |
8116 | +static void __pi_post_block(struct kvm_vcpu *vcpu) |
8117 | +{ |
8118 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
8119 | + struct pi_desc old, new; |
8120 | + unsigned int dest; |
8121 | + |
8122 | + do { |
8123 | + old.control = new.control = pi_desc->control; |
8124 | + WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR, |
8125 | + "Wakeup handler not enabled while the VCPU is blocked\n"); |
8126 | + |
8127 | + dest = cpu_physical_id(vcpu->cpu); |
8128 | + |
8129 | + if (x2apic_enabled()) |
8130 | + new.ndst = dest; |
8131 | + else |
8132 | + new.ndst = (dest << 8) & 0xFF00; |
8133 | + |
8134 | + /* set 'NV' to 'notification vector' */ |
8135 | + new.nv = POSTED_INTR_VECTOR; |
8136 | + } while (cmpxchg64(&pi_desc->control, old.control, |
8137 | + new.control) != old.control); |
8138 | + |
8139 | + if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) { |
8140 | + spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu)); |
8141 | + list_del(&vcpu->blocked_vcpu_list); |
8142 | + spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu)); |
8143 | + vcpu->pre_pcpu = -1; |
8144 | + } |
8145 | +} |
8146 | + |
8147 | +/* |
8148 | + * This routine does the following things for vCPU which is going |
8149 | + * to be blocked if VT-d PI is enabled. |
8150 | + * - Store the vCPU to the wakeup list, so when interrupts happen |
8151 | + * we can find the right vCPU to wake up. |
8152 | + * - Change the Posted-interrupt descriptor as below: |
8153 | + * 'NDST' <-- vcpu->pre_pcpu |
8154 | + * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR |
8155 | + * - If 'ON' is set during this process, which means at least one |
8156 | + * interrupt is posted for this vCPU, we cannot block it, in |
8157 | + * this case, return 1, otherwise, return 0. |
8158 | + * |
8159 | + */ |
8160 | +static int pi_pre_block(struct kvm_vcpu *vcpu) |
8161 | +{ |
8162 | + unsigned int dest; |
8163 | + struct pi_desc old, new; |
8164 | + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); |
8165 | + |
8166 | + if (!kvm_arch_has_assigned_device(vcpu->kvm) || |
8167 | + !irq_remapping_cap(IRQ_POSTING_CAP) || |
8168 | + !kvm_vcpu_apicv_active(vcpu)) |
8169 | + return 0; |
8170 | + |
8171 | + WARN_ON(irqs_disabled()); |
8172 | + local_irq_disable(); |
8173 | + if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) { |
8174 | + vcpu->pre_pcpu = vcpu->cpu; |
8175 | + spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu)); |
8176 | + list_add_tail(&vcpu->blocked_vcpu_list, |
8177 | + &per_cpu(blocked_vcpu_on_cpu, |
8178 | + vcpu->pre_pcpu)); |
8179 | + spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu)); |
8180 | + } |
8181 | + |
8182 | + do { |
8183 | + old.control = new.control = pi_desc->control; |
8184 | + |
8185 | + WARN((pi_desc->sn == 1), |
8186 | + "Warning: SN field of posted-interrupts " |
8187 | + "is set before blocking\n"); |
8188 | + |
8189 | + /* |
8190 | + * Since vCPU can be preempted during this process, |
8191 | + * vcpu->cpu could be different with pre_pcpu, we |
8192 | + * need to set pre_pcpu as the destination of wakeup |
8193 | + * notification event, then we can find the right vCPU |
8194 | + * to wakeup in wakeup handler if interrupts happen |
8195 | + * when the vCPU is in blocked state. |
8196 | + */ |
8197 | + dest = cpu_physical_id(vcpu->pre_pcpu); |
8198 | + |
8199 | + if (x2apic_enabled()) |
8200 | + new.ndst = dest; |
8201 | + else |
8202 | + new.ndst = (dest << 8) & 0xFF00; |
8203 | + |
8204 | + /* set 'NV' to 'wakeup vector' */ |
8205 | + new.nv = POSTED_INTR_WAKEUP_VECTOR; |
8206 | + } while (cmpxchg64(&pi_desc->control, old.control, |
8207 | + new.control) != old.control); |
8208 | + |
8209 | + /* We should not block the vCPU if an interrupt is posted for it. */ |
8210 | + if (pi_test_on(pi_desc) == 1) |
8211 | + __pi_post_block(vcpu); |
8212 | + |
8213 | + local_irq_enable(); |
8214 | + return (vcpu->pre_pcpu == -1); |
8215 | +} |
8216 | + |
8217 | +static int vmx_pre_block(struct kvm_vcpu *vcpu) |
8218 | +{ |
8219 | + if (pi_pre_block(vcpu)) |
8220 | + return 1; |
8221 | + |
8222 | + if (kvm_lapic_hv_timer_in_use(vcpu)) |
8223 | + kvm_lapic_switch_to_sw_timer(vcpu); |
8224 | + |
8225 | + return 0; |
8226 | +} |
8227 | + |
8228 | +static void pi_post_block(struct kvm_vcpu *vcpu) |
8229 | +{ |
8230 | + if (vcpu->pre_pcpu == -1) |
8231 | + return; |
8232 | + |
8233 | + WARN_ON(irqs_disabled()); |
8234 | + local_irq_disable(); |
8235 | + __pi_post_block(vcpu); |
8236 | + local_irq_enable(); |
8237 | +} |
8238 | + |
8239 | +static void vmx_post_block(struct kvm_vcpu *vcpu) |
8240 | +{ |
8241 | + if (kvm_x86_ops->set_hv_timer) |
8242 | + kvm_lapic_switch_to_hv_timer(vcpu); |
8243 | + |
8244 | + pi_post_block(vcpu); |
8245 | +} |
8246 | + |
8247 | +/* |
8248 | + * vmx_update_pi_irte - set IRTE for Posted-Interrupts |
8249 | + * |
8250 | + * @kvm: kvm |
8251 | + * @host_irq: host irq of the interrupt |
8252 | + * @guest_irq: gsi of the interrupt |
8253 | + * @set: set or unset PI |
8254 | + * returns 0 on success, < 0 on failure |
8255 | + */ |
8256 | +static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq, |
8257 | + uint32_t guest_irq, bool set) |
8258 | +{ |
8259 | + struct kvm_kernel_irq_routing_entry *e; |
8260 | + struct kvm_irq_routing_table *irq_rt; |
8261 | + struct kvm_lapic_irq irq; |
8262 | + struct kvm_vcpu *vcpu; |
8263 | + struct vcpu_data vcpu_info; |
8264 | + int idx, ret = 0; |
8265 | + |
8266 | + if (!kvm_arch_has_assigned_device(kvm) || |
8267 | + !irq_remapping_cap(IRQ_POSTING_CAP) || |
8268 | + !kvm_vcpu_apicv_active(kvm->vcpus[0])) |
8269 | + return 0; |
8270 | + |
8271 | + idx = srcu_read_lock(&kvm->irq_srcu); |
8272 | + irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); |
8273 | + if (guest_irq >= irq_rt->nr_rt_entries || |
8274 | + hlist_empty(&irq_rt->map[guest_irq])) { |
8275 | + pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n", |
8276 | + guest_irq, irq_rt->nr_rt_entries); |
8277 | + goto out; |
8278 | + } |
8279 | + |
8280 | + hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) { |
8281 | + if (e->type != KVM_IRQ_ROUTING_MSI) |
8282 | + continue; |
8283 | + /* |
8284 | + * VT-d PI cannot support posting multicast/broadcast |
8285 | + * interrupts to a vCPU, we still use interrupt remapping |
8286 | + * for these kind of interrupts. |
8287 | + * |
8288 | + * For lowest-priority interrupts, we only support |
8289 | + * those with single CPU as the destination, e.g. user |
8290 | + * configures the interrupts via /proc/irq or uses |
8291 | + * irqbalance to make the interrupts single-CPU. |
8292 | + * |
8293 | + * We will support full lowest-priority interrupt later. |
8294 | + * |
8295 | + * In addition, we can only inject generic interrupts using |
8296 | + * the PI mechanism, refuse to route others through it. |
8297 | + */ |
8298 | + |
8299 | + kvm_set_msi_irq(kvm, e, &irq); |
8300 | + if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) || |
8301 | + !kvm_irq_is_postable(&irq)) { |
8302 | + /* |
8303 | + * Make sure the IRTE is in remapped mode if |
8304 | + * we don't handle it in posted mode. |
8305 | + */ |
8306 | + ret = irq_set_vcpu_affinity(host_irq, NULL); |
8307 | + if (ret < 0) { |
8308 | + printk(KERN_INFO |
8309 | + "failed to back to remapped mode, irq: %u\n", |
8310 | + host_irq); |
8311 | + goto out; |
8312 | + } |
8313 | + |
8314 | + continue; |
8315 | + } |
8316 | + |
8317 | + vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu)); |
8318 | + vcpu_info.vector = irq.vector; |
8319 | + |
8320 | + trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi, |
8321 | + vcpu_info.vector, vcpu_info.pi_desc_addr, set); |
8322 | + |
8323 | + if (set) |
8324 | + ret = irq_set_vcpu_affinity(host_irq, &vcpu_info); |
8325 | + else |
8326 | + ret = irq_set_vcpu_affinity(host_irq, NULL); |
8327 | + |
8328 | + if (ret < 0) { |
8329 | + printk(KERN_INFO "%s: failed to update PI IRTE\n", |
8330 | + __func__); |
8331 | + goto out; |
8332 | + } |
8333 | + } |
8334 | + |
8335 | + ret = 0; |
8336 | +out: |
8337 | + srcu_read_unlock(&kvm->irq_srcu, idx); |
8338 | + return ret; |
8339 | +} |
8340 | + |
8341 | +static void vmx_setup_mce(struct kvm_vcpu *vcpu) |
8342 | +{ |
8343 | + if (vcpu->arch.mcg_cap & MCG_LMCE_P) |
8344 | + to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |= |
8345 | + FEATURE_CONTROL_LMCE; |
8346 | + else |
8347 | + to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &= |
8348 | + ~FEATURE_CONTROL_LMCE; |
8349 | +} |
8350 | + |
8351 | +static int vmx_smi_allowed(struct kvm_vcpu *vcpu) |
8352 | +{ |
8353 | + /* we need a nested vmexit to enter SMM, postpone if run is pending */ |
8354 | + if (to_vmx(vcpu)->nested.nested_run_pending) |
8355 | + return 0; |
8356 | + return 1; |
8357 | +} |
8358 | + |
8359 | +static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate) |
8360 | +{ |
8361 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
8362 | + |
8363 | + vmx->nested.smm.guest_mode = is_guest_mode(vcpu); |
8364 | + if (vmx->nested.smm.guest_mode) |
8365 | + nested_vmx_vmexit(vcpu, -1, 0, 0); |
8366 | + |
8367 | + vmx->nested.smm.vmxon = vmx->nested.vmxon; |
8368 | + vmx->nested.vmxon = false; |
8369 | + vmx_clear_hlt(vcpu); |
8370 | + return 0; |
8371 | +} |
8372 | + |
8373 | +static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate) |
8374 | +{ |
8375 | + struct vcpu_vmx *vmx = to_vmx(vcpu); |
8376 | + int ret; |
8377 | + |
8378 | + if (vmx->nested.smm.vmxon) { |
8379 | + vmx->nested.vmxon = true; |
8380 | + vmx->nested.smm.vmxon = false; |
8381 | + } |
8382 | + |
8383 | + if (vmx->nested.smm.guest_mode) { |
8384 | + ret = nested_vmx_enter_non_root_mode(vcpu, false); |
8385 | + if (ret) |
8386 | + return ret; |
8387 | + |
8388 | + vmx->nested.smm.guest_mode = false; |
8389 | + } |
8390 | + return 0; |
8391 | +} |
8392 | + |
8393 | +static int enable_smi_window(struct kvm_vcpu *vcpu) |
8394 | +{ |
8395 | + return 0; |
8396 | +} |
8397 | + |
8398 | +static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu) |
8399 | +{ |
8400 | + return false; |
8401 | +} |
8402 | + |
8403 | +static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu) |
8404 | +{ |
8405 | + return to_vmx(vcpu)->nested.vmxon; |
8406 | +} |
8407 | + |
8408 | +static __init int hardware_setup(void) |
8409 | +{ |
8410 | + unsigned long host_bndcfgs; |
8411 | + struct desc_ptr dt; |
8412 | + int r, i; |
8413 | + |
8414 | + rdmsrl_safe(MSR_EFER, &host_efer); |
8415 | + |
8416 | + store_idt(&dt); |
8417 | + host_idt_base = dt.address; |
8418 | + |
8419 | + for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) |
8420 | + kvm_define_shared_msr(i, vmx_msr_index[i]); |
8421 | + |
8422 | + if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0) |
8423 | + return -EIO; |
8424 | + |
8425 | + if (boot_cpu_has(X86_FEATURE_NX)) |
8426 | + kvm_enable_efer_bits(EFER_NX); |
8427 | + |
8428 | + if (boot_cpu_has(X86_FEATURE_MPX)) { |
8429 | + rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs); |
8430 | + WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost"); |
8431 | + } |
8432 | + |
8433 | + if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() || |
8434 | + !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global())) |
8435 | + enable_vpid = 0; |
8436 | + |
8437 | + if (!cpu_has_vmx_ept() || |
8438 | + !cpu_has_vmx_ept_4levels() || |
8439 | + !cpu_has_vmx_ept_mt_wb() || |
8440 | + !cpu_has_vmx_invept_global()) |
8441 | + enable_ept = 0; |
8442 | + |
8443 | + if (!cpu_has_vmx_ept_ad_bits() || !enable_ept) |
8444 | + enable_ept_ad_bits = 0; |
8445 | + |
8446 | + if (!cpu_has_vmx_unrestricted_guest() || !enable_ept) |
8447 | + enable_unrestricted_guest = 0; |
8448 | + |
8449 | + if (!cpu_has_vmx_flexpriority()) |
8450 | + flexpriority_enabled = 0; |
8451 | + |
8452 | + if (!cpu_has_virtual_nmis()) |
8453 | + enable_vnmi = 0; |
8454 | + |
8455 | + /* |
8456 | + * set_apic_access_page_addr() is used to reload apic access |
8457 | + * page upon invalidation. No need to do anything if not |
8458 | + * using the APIC_ACCESS_ADDR VMCS field. |
8459 | + */ |
8460 | + if (!flexpriority_enabled) |
8461 | + kvm_x86_ops->set_apic_access_page_addr = NULL; |
8462 | + |
8463 | + if (!cpu_has_vmx_tpr_shadow()) |
8464 | + kvm_x86_ops->update_cr8_intercept = NULL; |
8465 | + |
8466 | + if (enable_ept && !cpu_has_vmx_ept_2m_page()) |
8467 | + kvm_disable_largepages(); |
8468 | + |
8469 | +#if IS_ENABLED(CONFIG_HYPERV) |
8470 | + if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH |
8471 | + && enable_ept) { |
8472 | + kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb; |
8473 | + kvm_x86_ops->tlb_remote_flush_with_range = |
8474 | + hv_remote_flush_tlb_with_range; |
8475 | + } |
8476 | +#endif |
8477 | + |
8478 | + if (!cpu_has_vmx_ple()) { |
8479 | + ple_gap = 0; |
8480 | + ple_window = 0; |
8481 | + ple_window_grow = 0; |
8482 | + ple_window_max = 0; |
8483 | + ple_window_shrink = 0; |
8484 | + } |
8485 | + |
8486 | + if (!cpu_has_vmx_apicv()) { |
8487 | + enable_apicv = 0; |
8488 | + kvm_x86_ops->sync_pir_to_irr = NULL; |
8489 | + } |
8490 | + |
8491 | + if (cpu_has_vmx_tsc_scaling()) { |
8492 | + kvm_has_tsc_control = true; |
8493 | + kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX; |
8494 | + kvm_tsc_scaling_ratio_frac_bits = 48; |
8495 | + } |
8496 | + |
8497 | + set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */ |
8498 | + |
8499 | + if (enable_ept) |
8500 | + vmx_enable_tdp(); |
8501 | + else |
8502 | + kvm_disable_tdp(); |
8503 | + |
8504 | + /* |
8505 | + * Only enable PML when hardware supports PML feature, and both EPT |
8506 | + * and EPT A/D bit features are enabled -- PML depends on them to work. |
8507 | + */ |
8508 | + if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml()) |
8509 | + enable_pml = 0; |
8510 | + |
8511 | + if (!enable_pml) { |
8512 | + kvm_x86_ops->slot_enable_log_dirty = NULL; |
8513 | + kvm_x86_ops->slot_disable_log_dirty = NULL; |
8514 | + kvm_x86_ops->flush_log_dirty = NULL; |
8515 | + kvm_x86_ops->enable_log_dirty_pt_masked = NULL; |
8516 | + } |
8517 | + |
8518 | + if (!cpu_has_vmx_preemption_timer()) |
8519 | + enable_preemption_timer = false; |
8520 | + |
8521 | + if (enable_preemption_timer) { |
8522 | + u64 use_timer_freq = 5000ULL * 1000 * 1000; |
8523 | + u64 vmx_msr; |
8524 | + |
8525 | + rdmsrl(MSR_IA32_VMX_MISC, vmx_msr); |
8526 | + cpu_preemption_timer_multi = |
8527 | + vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK; |
8528 | + |
8529 | + if (tsc_khz) |
8530 | + use_timer_freq = (u64)tsc_khz * 1000; |
8531 | + use_timer_freq >>= cpu_preemption_timer_multi; |
8532 | + |
8533 | + /* |
8534 | + * KVM "disables" the preemption timer by setting it to its max |
8535 | + * value. Don't use the timer if it might cause spurious exits |
8536 | + * at a rate faster than 0.1 Hz (of uninterrupted guest time). |
8537 | + */ |
8538 | + if (use_timer_freq > 0xffffffffu / 10) |
8539 | + enable_preemption_timer = false; |
8540 | + } |
8541 | + |
8542 | + if (!enable_preemption_timer) { |
8543 | + kvm_x86_ops->set_hv_timer = NULL; |
8544 | + kvm_x86_ops->cancel_hv_timer = NULL; |
8545 | + kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit; |
8546 | + } |
8547 | + |
8548 | + kvm_set_posted_intr_wakeup_handler(wakeup_handler); |
8549 | + |
8550 | + kvm_mce_cap_supported |= MCG_LMCE_P; |
8551 | + |
8552 | + if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST) |
8553 | + return -EINVAL; |
8554 | + if (!enable_ept || !cpu_has_vmx_intel_pt()) |
8555 | + pt_mode = PT_MODE_SYSTEM; |
8556 | + |
8557 | + if (nested) { |
8558 | + nested_vmx_setup_ctls_msrs(&vmcs_config.nested, |
8559 | + vmx_capability.ept, enable_apicv); |
8560 | + |
8561 | + r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers); |
8562 | + if (r) |
8563 | + return r; |
8564 | + } |
8565 | + |
8566 | + r = alloc_kvm_area(); |
8567 | + if (r) |
8568 | + nested_vmx_hardware_unsetup(); |
8569 | + return r; |
8570 | +} |
8571 | + |
8572 | +static __exit void hardware_unsetup(void) |
8573 | +{ |
8574 | + if (nested) |
8575 | + nested_vmx_hardware_unsetup(); |
8576 | + |
8577 | + free_kvm_area(); |
8578 | +} |
8579 | + |
8580 | +static struct kvm_x86_ops vmx_x86_ops __ro_after_init = { |
8581 | + .cpu_has_kvm_support = cpu_has_kvm_support, |
8582 | + .disabled_by_bios = vmx_disabled_by_bios, |
8583 | + .hardware_setup = hardware_setup, |
8584 | + .hardware_unsetup = hardware_unsetup, |
8585 | + .check_processor_compatibility = vmx_check_processor_compat, |
8586 | + .hardware_enable = hardware_enable, |
8587 | + .hardware_disable = hardware_disable, |
8588 | + .cpu_has_accelerated_tpr = report_flexpriority, |
8589 | + .has_emulated_msr = vmx_has_emulated_msr, |
8590 | + |
8591 | + .vm_init = vmx_vm_init, |
8592 | + .vm_alloc = vmx_vm_alloc, |
8593 | + .vm_free = vmx_vm_free, |
8594 | + |
8595 | + .vcpu_create = vmx_create_vcpu, |
8596 | + .vcpu_free = vmx_free_vcpu, |
8597 | + .vcpu_reset = vmx_vcpu_reset, |
8598 | + |
8599 | + .prepare_guest_switch = vmx_prepare_switch_to_guest, |
8600 | + .vcpu_load = vmx_vcpu_load, |
8601 | + .vcpu_put = vmx_vcpu_put, |
8602 | + |
8603 | + .update_bp_intercept = update_exception_bitmap, |
8604 | + .get_msr_feature = vmx_get_msr_feature, |
8605 | + .get_msr = vmx_get_msr, |
8606 | + .set_msr = vmx_set_msr, |
8607 | + .get_segment_base = vmx_get_segment_base, |
8608 | + .get_segment = vmx_get_segment, |
8609 | + .set_segment = vmx_set_segment, |
8610 | + .get_cpl = vmx_get_cpl, |
8611 | + .get_cs_db_l_bits = vmx_get_cs_db_l_bits, |
8612 | + .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits, |
8613 | + .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits, |
8614 | + .set_cr0 = vmx_set_cr0, |
8615 | + .set_cr3 = vmx_set_cr3, |
8616 | + .set_cr4 = vmx_set_cr4, |
8617 | + .set_efer = vmx_set_efer, |
8618 | + .get_idt = vmx_get_idt, |
8619 | + .set_idt = vmx_set_idt, |
8620 | + .get_gdt = vmx_get_gdt, |
8621 | + .set_gdt = vmx_set_gdt, |
8622 | + .get_dr6 = vmx_get_dr6, |
8623 | + .set_dr6 = vmx_set_dr6, |
8624 | + .set_dr7 = vmx_set_dr7, |
8625 | + .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs, |
8626 | + .cache_reg = vmx_cache_reg, |
8627 | + .get_rflags = vmx_get_rflags, |
8628 | + .set_rflags = vmx_set_rflags, |
8629 | + |
8630 | + .tlb_flush = vmx_flush_tlb, |
8631 | + .tlb_flush_gva = vmx_flush_tlb_gva, |
8632 | + |
8633 | + .run = vmx_vcpu_run, |
8634 | + .handle_exit = vmx_handle_exit, |
8635 | + .skip_emulated_instruction = skip_emulated_instruction, |
8636 | + .set_interrupt_shadow = vmx_set_interrupt_shadow, |
8637 | + .get_interrupt_shadow = vmx_get_interrupt_shadow, |
8638 | + .patch_hypercall = vmx_patch_hypercall, |
8639 | + .set_irq = vmx_inject_irq, |
8640 | + .set_nmi = vmx_inject_nmi, |
8641 | + .queue_exception = vmx_queue_exception, |
8642 | + .cancel_injection = vmx_cancel_injection, |
8643 | + .interrupt_allowed = vmx_interrupt_allowed, |
8644 | + .nmi_allowed = vmx_nmi_allowed, |
8645 | + .get_nmi_mask = vmx_get_nmi_mask, |
8646 | + .set_nmi_mask = vmx_set_nmi_mask, |
8647 | + .enable_nmi_window = enable_nmi_window, |
8648 | + .enable_irq_window = enable_irq_window, |
8649 | + .update_cr8_intercept = update_cr8_intercept, |
8650 | + .set_virtual_apic_mode = vmx_set_virtual_apic_mode, |
8651 | + .set_apic_access_page_addr = vmx_set_apic_access_page_addr, |
8652 | + .get_enable_apicv = vmx_get_enable_apicv, |
8653 | + .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl, |
8654 | + .load_eoi_exitmap = vmx_load_eoi_exitmap, |
8655 | + .apicv_post_state_restore = vmx_apicv_post_state_restore, |
8656 | + .hwapic_irr_update = vmx_hwapic_irr_update, |
8657 | + .hwapic_isr_update = vmx_hwapic_isr_update, |
8658 | + .guest_apic_has_interrupt = vmx_guest_apic_has_interrupt, |
8659 | + .sync_pir_to_irr = vmx_sync_pir_to_irr, |
8660 | + .deliver_posted_interrupt = vmx_deliver_posted_interrupt, |
8661 | + .dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt, |
8662 | + |
8663 | + .set_tss_addr = vmx_set_tss_addr, |
8664 | + .set_identity_map_addr = vmx_set_identity_map_addr, |
8665 | + .get_tdp_level = get_ept_level, |
8666 | + .get_mt_mask = vmx_get_mt_mask, |
8667 | + |
8668 | + .get_exit_info = vmx_get_exit_info, |
8669 | + |
8670 | + .get_lpage_level = vmx_get_lpage_level, |
8671 | + |
8672 | + .cpuid_update = vmx_cpuid_update, |
8673 | + |
8674 | + .rdtscp_supported = vmx_rdtscp_supported, |
8675 | + .invpcid_supported = vmx_invpcid_supported, |
8676 | + |
8677 | + .set_supported_cpuid = vmx_set_supported_cpuid, |
8678 | + |
8679 | + .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit, |
8680 | + |
8681 | + .read_l1_tsc_offset = vmx_read_l1_tsc_offset, |
8682 | + .write_l1_tsc_offset = vmx_write_l1_tsc_offset, |
8683 | + |
8684 | + .set_tdp_cr3 = vmx_set_cr3, |
8685 | + |
8686 | + .check_intercept = vmx_check_intercept, |
8687 | + .handle_exit_irqoff = vmx_handle_exit_irqoff, |
8688 | + .mpx_supported = vmx_mpx_supported, |
8689 | + .xsaves_supported = vmx_xsaves_supported, |
8690 | + .umip_emulated = vmx_umip_emulated, |
8691 | + .pt_supported = vmx_pt_supported, |
8692 | + |
8693 | + .request_immediate_exit = vmx_request_immediate_exit, |
8694 | + |
8695 | + .sched_in = vmx_sched_in, |
8696 | + |
8697 | + .slot_enable_log_dirty = vmx_slot_enable_log_dirty, |
8698 | + .slot_disable_log_dirty = vmx_slot_disable_log_dirty, |
8699 | + .flush_log_dirty = vmx_flush_log_dirty, |
8700 | + .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked, |
8701 | + .write_log_dirty = vmx_write_pml_buffer, |
8702 | + |
8703 | + .pre_block = vmx_pre_block, |
8704 | + .post_block = vmx_post_block, |
8705 | + |
8706 | + .pmu_ops = &intel_pmu_ops, |
8707 | + |
8708 | + .update_pi_irte = vmx_update_pi_irte, |
8709 | + |
8710 | +#ifdef CONFIG_X86_64 |
8711 | + .set_hv_timer = vmx_set_hv_timer, |
8712 | + .cancel_hv_timer = vmx_cancel_hv_timer, |
8713 | +#endif |
8714 | + |
8715 | + .setup_mce = vmx_setup_mce, |
8716 | + |
8717 | + .smi_allowed = vmx_smi_allowed, |
8718 | + .pre_enter_smm = vmx_pre_enter_smm, |
8719 | + .pre_leave_smm = vmx_pre_leave_smm, |
8720 | + .enable_smi_window = enable_smi_window, |
8721 | + |
8722 | + .check_nested_events = NULL, |
8723 | + .get_nested_state = NULL, |
8724 | + .set_nested_state = NULL, |
8725 | + .get_vmcs12_pages = NULL, |
8726 | + .nested_enable_evmcs = NULL, |
8727 | + .nested_get_evmcs_version = NULL, |
8728 | + .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault, |
8729 | + .apic_init_signal_blocked = vmx_apic_init_signal_blocked, |
8730 | +}; |
8731 | + |
8732 | +static void vmx_cleanup_l1d_flush(void) |
8733 | +{ |
8734 | + if (vmx_l1d_flush_pages) { |
8735 | + free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER); |
8736 | + vmx_l1d_flush_pages = NULL; |
8737 | + } |
8738 | + /* Restore state so sysfs ignores VMX */ |
8739 | + l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO; |
8740 | +} |
8741 | + |
8742 | +static void vmx_exit(void) |
8743 | +{ |
8744 | +#ifdef CONFIG_KEXEC_CORE |
8745 | + RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL); |
8746 | + synchronize_rcu(); |
8747 | +#endif |
8748 | + |
8749 | + kvm_exit(); |
8750 | + |
8751 | +#if IS_ENABLED(CONFIG_HYPERV) |
8752 | + if (static_branch_unlikely(&enable_evmcs)) { |
8753 | + int cpu; |
8754 | + struct hv_vp_assist_page *vp_ap; |
8755 | + /* |
8756 | + * Reset everything to support using non-enlightened VMCS |
8757 | + * access later (e.g. when we reload the module with |
8758 | + * enlightened_vmcs=0) |
8759 | + */ |
8760 | + for_each_online_cpu(cpu) { |
8761 | + vp_ap = hv_get_vp_assist_page(cpu); |
8762 | + |
8763 | + if (!vp_ap) |
8764 | + continue; |
8765 | + |
8766 | + vp_ap->nested_control.features.directhypercall = 0; |
8767 | + vp_ap->current_nested_vmcs = 0; |
8768 | + vp_ap->enlighten_vmentry = 0; |
8769 | + } |
8770 | + |
8771 | + static_branch_disable(&enable_evmcs); |
8772 | + } |
8773 | +#endif |
8774 | + vmx_cleanup_l1d_flush(); |
8775 | +} |
8776 | +module_exit(vmx_exit); |
8777 | + |
8778 | +static int __init vmx_init(void) |
8779 | +{ |
8780 | + int r; |
8781 | + |
8782 | +#if IS_ENABLED(CONFIG_HYPERV) |
8783 | + /* |
8784 | + * Enlightened VMCS usage should be recommended and the host needs |
8785 | + * to support eVMCS v1 or above. We can also disable eVMCS support |
8786 | + * with module parameter. |
8787 | + */ |
8788 | + if (enlightened_vmcs && |
8789 | + ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED && |
8790 | + (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >= |
8791 | + KVM_EVMCS_VERSION) { |
8792 | + int cpu; |
8793 | + |
8794 | + /* Check that we have assist pages on all online CPUs */ |
8795 | + for_each_online_cpu(cpu) { |
8796 | + if (!hv_get_vp_assist_page(cpu)) { |
8797 | + enlightened_vmcs = false; |
8798 | + break; |
8799 | + } |
8800 | + } |
8801 | + |
8802 | + if (enlightened_vmcs) { |
8803 | + pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n"); |
8804 | + static_branch_enable(&enable_evmcs); |
8805 | + } |
8806 | + |
8807 | + if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH) |
8808 | + vmx_x86_ops.enable_direct_tlbflush |
8809 | + = hv_enable_direct_tlbflush; |
8810 | + |
8811 | + } else { |
8812 | + enlightened_vmcs = false; |
8813 | + } |
8814 | +#endif |
8815 | + |
8816 | + r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), |
8817 | + __alignof__(struct vcpu_vmx), THIS_MODULE); |
8818 | + if (r) |
8819 | + return r; |
8820 | + |
8821 | + /* |
8822 | + * Must be called after kvm_init() so enable_ept is properly set |
8823 | + * up. Hand the parameter mitigation value in which was stored in |
8824 | + * the pre module init parser. If no parameter was given, it will |
8825 | + * contain 'auto' which will be turned into the default 'cond' |
8826 | + * mitigation mode. |
8827 | + */ |
8828 | + r = vmx_setup_l1d_flush(vmentry_l1d_flush_param); |
8829 | + if (r) { |
8830 | + vmx_exit(); |
8831 | + return r; |
8832 | + } |
8833 | + |
8834 | +#ifdef CONFIG_KEXEC_CORE |
8835 | + rcu_assign_pointer(crash_vmclear_loaded_vmcss, |
8836 | + crash_vmclear_local_loaded_vmcss); |
8837 | +#endif |
8838 | + vmx_check_vmcs12_offsets(); |
8839 | + |
8840 | + return 0; |
8841 | +} |
8842 | +module_init(vmx_init); |
8843 | diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c |
8844 | index c9c533370e88..43aabd72019b 100644 |
8845 | --- a/arch/x86/kvm/x86.c |
8846 | +++ b/arch/x86/kvm/x86.c |
8847 | @@ -54,6 +54,7 @@ |
8848 | #include <linux/pvclock_gtod.h> |
8849 | #include <linux/kvm_irqfd.h> |
8850 | #include <linux/irqbypass.h> |
8851 | +#include <linux/nospec.h> |
8852 | #include <trace/events/kvm.h> |
8853 | |
8854 | #include <asm/debugreg.h> |
8855 | @@ -889,9 +890,11 @@ static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
8856 | |
8857 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
8858 | { |
8859 | + size_t size = ARRAY_SIZE(vcpu->arch.db); |
8860 | + |
8861 | switch (dr) { |
8862 | case 0 ... 3: |
8863 | - vcpu->arch.db[dr] = val; |
8864 | + vcpu->arch.db[array_index_nospec(dr, size)] = val; |
8865 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) |
8866 | vcpu->arch.eff_db[dr] = val; |
8867 | break; |
8868 | @@ -928,9 +931,11 @@ EXPORT_SYMBOL_GPL(kvm_set_dr); |
8869 | |
8870 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
8871 | { |
8872 | + size_t size = ARRAY_SIZE(vcpu->arch.db); |
8873 | + |
8874 | switch (dr) { |
8875 | case 0 ... 3: |
8876 | - *val = vcpu->arch.db[dr]; |
8877 | + *val = vcpu->arch.db[array_index_nospec(dr, size)]; |
8878 | break; |
8879 | case 4: |
8880 | /* fall through */ |
8881 | @@ -2125,7 +2130,10 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data) |
8882 | default: |
8883 | if (msr >= MSR_IA32_MC0_CTL && |
8884 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
8885 | - u32 offset = msr - MSR_IA32_MC0_CTL; |
8886 | + u32 offset = array_index_nospec( |
8887 | + msr - MSR_IA32_MC0_CTL, |
8888 | + MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); |
8889 | + |
8890 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
8891 | * some Linux kernels though clear bit 10 in bank 4 to |
8892 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore |
8893 | @@ -2493,7 +2501,10 @@ static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
8894 | default: |
8895 | if (msr >= MSR_IA32_MC0_CTL && |
8896 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
8897 | - u32 offset = msr - MSR_IA32_MC0_CTL; |
8898 | + u32 offset = array_index_nospec( |
8899 | + msr - MSR_IA32_MC0_CTL, |
8900 | + MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); |
8901 | + |
8902 | data = vcpu->arch.mce_banks[offset]; |
8903 | break; |
8904 | } |
8905 | @@ -6121,14 +6132,12 @@ static void kvm_set_mmio_spte_mask(void) |
8906 | /* Set the present bit. */ |
8907 | mask |= 1ull; |
8908 | |
8909 | -#ifdef CONFIG_X86_64 |
8910 | /* |
8911 | * If reserved bit is not supported, clear the present bit to disable |
8912 | * mmio page fault. |
8913 | */ |
8914 | if (maxphyaddr == 52) |
8915 | mask &= ~1ull; |
8916 | -#endif |
8917 | |
8918 | kvm_mmu_set_mmio_spte_mask(mask); |
8919 | } |
8920 | @@ -7798,7 +7807,7 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
8921 | kvm_mmu_unload(vcpu); |
8922 | vcpu_put(vcpu); |
8923 | |
8924 | - kvm_x86_ops->vcpu_free(vcpu); |
8925 | + kvm_arch_vcpu_free(vcpu); |
8926 | } |
8927 | |
8928 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
8929 | diff --git a/crypto/algapi.c b/crypto/algapi.c |
8930 | index 5c098ffa7d3d..9e5b24329b41 100644 |
8931 | --- a/crypto/algapi.c |
8932 | +++ b/crypto/algapi.c |
8933 | @@ -652,11 +652,9 @@ EXPORT_SYMBOL_GPL(crypto_grab_spawn); |
8934 | |
8935 | void crypto_drop_spawn(struct crypto_spawn *spawn) |
8936 | { |
8937 | - if (!spawn->alg) |
8938 | - return; |
8939 | - |
8940 | down_write(&crypto_alg_sem); |
8941 | - list_del(&spawn->list); |
8942 | + if (spawn->alg) |
8943 | + list_del(&spawn->list); |
8944 | up_write(&crypto_alg_sem); |
8945 | } |
8946 | EXPORT_SYMBOL_GPL(crypto_drop_spawn); |
8947 | @@ -664,22 +662,16 @@ EXPORT_SYMBOL_GPL(crypto_drop_spawn); |
8948 | static struct crypto_alg *crypto_spawn_alg(struct crypto_spawn *spawn) |
8949 | { |
8950 | struct crypto_alg *alg; |
8951 | - struct crypto_alg *alg2; |
8952 | |
8953 | down_read(&crypto_alg_sem); |
8954 | alg = spawn->alg; |
8955 | - alg2 = alg; |
8956 | - if (alg2) |
8957 | - alg2 = crypto_mod_get(alg2); |
8958 | - up_read(&crypto_alg_sem); |
8959 | - |
8960 | - if (!alg2) { |
8961 | - if (alg) |
8962 | - crypto_shoot_alg(alg); |
8963 | - return ERR_PTR(-EAGAIN); |
8964 | + if (alg && !crypto_mod_get(alg)) { |
8965 | + alg->cra_flags |= CRYPTO_ALG_DYING; |
8966 | + alg = NULL; |
8967 | } |
8968 | + up_read(&crypto_alg_sem); |
8969 | |
8970 | - return alg; |
8971 | + return alg ?: ERR_PTR(-EAGAIN); |
8972 | } |
8973 | |
8974 | struct crypto_tfm *crypto_spawn_tfm(struct crypto_spawn *spawn, u32 type, |
8975 | diff --git a/crypto/api.c b/crypto/api.c |
8976 | index abf53e67e3d8..b273b3a726a9 100644 |
8977 | --- a/crypto/api.c |
8978 | +++ b/crypto/api.c |
8979 | @@ -355,13 +355,12 @@ static unsigned int crypto_ctxsize(struct crypto_alg *alg, u32 type, u32 mask) |
8980 | return len; |
8981 | } |
8982 | |
8983 | -void crypto_shoot_alg(struct crypto_alg *alg) |
8984 | +static void crypto_shoot_alg(struct crypto_alg *alg) |
8985 | { |
8986 | down_write(&crypto_alg_sem); |
8987 | alg->cra_flags |= CRYPTO_ALG_DYING; |
8988 | up_write(&crypto_alg_sem); |
8989 | } |
8990 | -EXPORT_SYMBOL_GPL(crypto_shoot_alg); |
8991 | |
8992 | struct crypto_tfm *__crypto_alloc_tfm(struct crypto_alg *alg, u32 type, |
8993 | u32 mask) |
8994 | diff --git a/crypto/internal.h b/crypto/internal.h |
8995 | index 7eefcdb00227..6184c4226a8f 100644 |
8996 | --- a/crypto/internal.h |
8997 | +++ b/crypto/internal.h |
8998 | @@ -87,7 +87,6 @@ void crypto_alg_tested(const char *name, int err); |
8999 | void crypto_remove_spawns(struct crypto_alg *alg, struct list_head *list, |
9000 | struct crypto_alg *nalg); |
9001 | void crypto_remove_final(struct list_head *list); |
9002 | -void crypto_shoot_alg(struct crypto_alg *alg); |
9003 | struct crypto_tfm *__crypto_alloc_tfm(struct crypto_alg *alg, u32 type, |
9004 | u32 mask); |
9005 | void *crypto_create_tfm(struct crypto_alg *alg, |
9006 | diff --git a/crypto/pcrypt.c b/crypto/pcrypt.c |
9007 | index 1348541da463..85082574c515 100644 |
9008 | --- a/crypto/pcrypt.c |
9009 | +++ b/crypto/pcrypt.c |
9010 | @@ -130,7 +130,6 @@ static void pcrypt_aead_done(struct crypto_async_request *areq, int err) |
9011 | struct padata_priv *padata = pcrypt_request_padata(preq); |
9012 | |
9013 | padata->info = err; |
9014 | - req->base.flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; |
9015 | |
9016 | padata_do_serial(padata); |
9017 | } |
9018 | diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c |
9019 | index 4ce4e7fb1124..d9c1f229c644 100644 |
9020 | --- a/drivers/clk/tegra/clk-tegra-periph.c |
9021 | +++ b/drivers/clk/tegra/clk-tegra-periph.c |
9022 | @@ -797,7 +797,11 @@ static struct tegra_periph_init_data gate_clks[] = { |
9023 | GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0), |
9024 | GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0), |
9025 | GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0), |
9026 | - GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0), |
9027 | + /* |
9028 | + * Critical for RAM re-repair operation, which must occur on resume |
9029 | + * from LP1 system suspend and as part of CCPLEX cluster switching. |
9030 | + */ |
9031 | + GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, CLK_IS_CRITICAL), |
9032 | GATE("fuse_burn", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse_burn, 0), |
9033 | GATE("kfuse", "clk_m", 40, TEGRA_PERIPH_ON_APB, tegra_clk_kfuse, 0), |
9034 | GATE("apbif", "clk_m", 107, TEGRA_PERIPH_ON_APB, tegra_clk_apbif, 0), |
9035 | diff --git a/drivers/crypto/atmel-aes.c b/drivers/crypto/atmel-aes.c |
9036 | index e3d40a8dfffb..915253b9c912 100644 |
9037 | --- a/drivers/crypto/atmel-aes.c |
9038 | +++ b/drivers/crypto/atmel-aes.c |
9039 | @@ -87,7 +87,6 @@ |
9040 | struct atmel_aes_caps { |
9041 | bool has_dualbuff; |
9042 | bool has_cfb64; |
9043 | - bool has_ctr32; |
9044 | bool has_gcm; |
9045 | u32 max_burst_size; |
9046 | }; |
9047 | @@ -923,8 +922,9 @@ static int atmel_aes_ctr_transfer(struct atmel_aes_dev *dd) |
9048 | struct atmel_aes_ctr_ctx *ctx = atmel_aes_ctr_ctx_cast(dd->ctx); |
9049 | struct ablkcipher_request *req = ablkcipher_request_cast(dd->areq); |
9050 | struct scatterlist *src, *dst; |
9051 | - u32 ctr, blocks; |
9052 | size_t datalen; |
9053 | + u32 ctr; |
9054 | + u16 blocks, start, end; |
9055 | bool use_dma, fragmented = false; |
9056 | |
9057 | /* Check for transfer completion. */ |
9058 | @@ -936,27 +936,17 @@ static int atmel_aes_ctr_transfer(struct atmel_aes_dev *dd) |
9059 | datalen = req->nbytes - ctx->offset; |
9060 | blocks = DIV_ROUND_UP(datalen, AES_BLOCK_SIZE); |
9061 | ctr = be32_to_cpu(ctx->iv[3]); |
9062 | - if (dd->caps.has_ctr32) { |
9063 | - /* Check 32bit counter overflow. */ |
9064 | - u32 start = ctr; |
9065 | - u32 end = start + blocks - 1; |
9066 | - |
9067 | - if (end < start) { |
9068 | - ctr |= 0xffffffff; |
9069 | - datalen = AES_BLOCK_SIZE * -start; |
9070 | - fragmented = true; |
9071 | - } |
9072 | - } else { |
9073 | - /* Check 16bit counter overflow. */ |
9074 | - u16 start = ctr & 0xffff; |
9075 | - u16 end = start + (u16)blocks - 1; |
9076 | - |
9077 | - if (blocks >> 16 || end < start) { |
9078 | - ctr |= 0xffff; |
9079 | - datalen = AES_BLOCK_SIZE * (0x10000-start); |
9080 | - fragmented = true; |
9081 | - } |
9082 | + |
9083 | + /* Check 16bit counter overflow. */ |
9084 | + start = ctr & 0xffff; |
9085 | + end = start + blocks - 1; |
9086 | + |
9087 | + if (blocks >> 16 || end < start) { |
9088 | + ctr |= 0xffff; |
9089 | + datalen = AES_BLOCK_SIZE * (0x10000 - start); |
9090 | + fragmented = true; |
9091 | } |
9092 | + |
9093 | use_dma = (datalen >= ATMEL_AES_DMA_THRESHOLD); |
9094 | |
9095 | /* Jump to offset. */ |
9096 | @@ -1926,7 +1916,6 @@ static void atmel_aes_get_cap(struct atmel_aes_dev *dd) |
9097 | { |
9098 | dd->caps.has_dualbuff = 0; |
9099 | dd->caps.has_cfb64 = 0; |
9100 | - dd->caps.has_ctr32 = 0; |
9101 | dd->caps.has_gcm = 0; |
9102 | dd->caps.max_burst_size = 1; |
9103 | |
9104 | @@ -1935,14 +1924,12 @@ static void atmel_aes_get_cap(struct atmel_aes_dev *dd) |
9105 | case 0x500: |
9106 | dd->caps.has_dualbuff = 1; |
9107 | dd->caps.has_cfb64 = 1; |
9108 | - dd->caps.has_ctr32 = 1; |
9109 | dd->caps.has_gcm = 1; |
9110 | dd->caps.max_burst_size = 4; |
9111 | break; |
9112 | case 0x200: |
9113 | dd->caps.has_dualbuff = 1; |
9114 | dd->caps.has_cfb64 = 1; |
9115 | - dd->caps.has_ctr32 = 1; |
9116 | dd->caps.has_gcm = 1; |
9117 | dd->caps.max_burst_size = 4; |
9118 | break; |
9119 | diff --git a/drivers/crypto/picoxcell_crypto.c b/drivers/crypto/picoxcell_crypto.c |
9120 | index 47576098831f..b3ea6d60c458 100644 |
9121 | --- a/drivers/crypto/picoxcell_crypto.c |
9122 | +++ b/drivers/crypto/picoxcell_crypto.c |
9123 | @@ -1632,6 +1632,11 @@ static bool spacc_is_compatible(struct platform_device *pdev, |
9124 | return false; |
9125 | } |
9126 | |
9127 | +static void spacc_tasklet_kill(void *data) |
9128 | +{ |
9129 | + tasklet_kill(data); |
9130 | +} |
9131 | + |
9132 | static int spacc_probe(struct platform_device *pdev) |
9133 | { |
9134 | int i, err, ret = -EINVAL; |
9135 | @@ -1674,6 +1679,14 @@ static int spacc_probe(struct platform_device *pdev) |
9136 | return -ENXIO; |
9137 | } |
9138 | |
9139 | + tasklet_init(&engine->complete, spacc_spacc_complete, |
9140 | + (unsigned long)engine); |
9141 | + |
9142 | + ret = devm_add_action(&pdev->dev, spacc_tasklet_kill, |
9143 | + &engine->complete); |
9144 | + if (ret) |
9145 | + return ret; |
9146 | + |
9147 | if (devm_request_irq(&pdev->dev, irq->start, spacc_spacc_irq, 0, |
9148 | engine->name, engine)) { |
9149 | dev_err(engine->dev, "failed to request IRQ\n"); |
9150 | @@ -1736,8 +1749,6 @@ static int spacc_probe(struct platform_device *pdev) |
9151 | INIT_LIST_HEAD(&engine->completed); |
9152 | INIT_LIST_HEAD(&engine->in_progress); |
9153 | engine->in_flight = 0; |
9154 | - tasklet_init(&engine->complete, spacc_spacc_complete, |
9155 | - (unsigned long)engine); |
9156 | |
9157 | platform_set_drvdata(pdev, engine); |
9158 | |
9159 | diff --git a/drivers/gpu/drm/atmel-hlcdc/atmel_hlcdc_crtc.c b/drivers/gpu/drm/atmel-hlcdc/atmel_hlcdc_crtc.c |
9160 | index 9b17a66cf0e1..aa54a6a2ad1d 100644 |
9161 | --- a/drivers/gpu/drm/atmel-hlcdc/atmel_hlcdc_crtc.c |
9162 | +++ b/drivers/gpu/drm/atmel-hlcdc/atmel_hlcdc_crtc.c |
9163 | @@ -81,7 +81,11 @@ static void atmel_hlcdc_crtc_mode_set_nofb(struct drm_crtc *c) |
9164 | struct videomode vm; |
9165 | unsigned long prate; |
9166 | unsigned int cfg; |
9167 | - int div; |
9168 | + int div, ret; |
9169 | + |
9170 | + ret = clk_prepare_enable(crtc->dc->hlcdc->sys_clk); |
9171 | + if (ret) |
9172 | + return; |
9173 | |
9174 | vm.vfront_porch = adj->crtc_vsync_start - adj->crtc_vdisplay; |
9175 | vm.vback_porch = adj->crtc_vtotal - adj->crtc_vsync_end; |
9176 | @@ -140,6 +144,8 @@ static void atmel_hlcdc_crtc_mode_set_nofb(struct drm_crtc *c) |
9177 | ATMEL_HLCDC_VSPSU | ATMEL_HLCDC_VSPHO | |
9178 | ATMEL_HLCDC_GUARDTIME_MASK | ATMEL_HLCDC_MODE_MASK, |
9179 | cfg); |
9180 | + |
9181 | + clk_disable_unprepare(crtc->dc->hlcdc->sys_clk); |
9182 | } |
9183 | |
9184 | static bool atmel_hlcdc_crtc_mode_fixup(struct drm_crtc *c, |
9185 | diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c |
9186 | index 1baa25e82bdd..f7d23c1081dc 100644 |
9187 | --- a/drivers/infiniband/core/addr.c |
9188 | +++ b/drivers/infiniband/core/addr.c |
9189 | @@ -141,7 +141,7 @@ int ib_nl_handle_ip_res_resp(struct sk_buff *skb, |
9190 | if (ib_nl_is_good_ip_resp(nlh)) |
9191 | ib_nl_process_good_ip_rsep(nlh); |
9192 | |
9193 | - return skb->len; |
9194 | + return 0; |
9195 | } |
9196 | |
9197 | static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr, |
9198 | diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c |
9199 | index 5879a06ada93..1c459725d64e 100644 |
9200 | --- a/drivers/infiniband/core/sa_query.c |
9201 | +++ b/drivers/infiniband/core/sa_query.c |
9202 | @@ -848,7 +848,7 @@ int ib_nl_handle_set_timeout(struct sk_buff *skb, |
9203 | } |
9204 | |
9205 | settimeout_out: |
9206 | - return skb->len; |
9207 | + return 0; |
9208 | } |
9209 | |
9210 | static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh) |
9211 | @@ -920,7 +920,7 @@ int ib_nl_handle_resolve_resp(struct sk_buff *skb, |
9212 | } |
9213 | |
9214 | resp_out: |
9215 | - return skb->len; |
9216 | + return 0; |
9217 | } |
9218 | |
9219 | static void free_sm_ah(struct kref *kref) |
9220 | diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c |
9221 | index 79e6309460dc..262c18b2f525 100644 |
9222 | --- a/drivers/infiniband/hw/mlx5/gsi.c |
9223 | +++ b/drivers/infiniband/hw/mlx5/gsi.c |
9224 | @@ -507,8 +507,7 @@ int mlx5_ib_gsi_post_send(struct ib_qp *qp, struct ib_send_wr *wr, |
9225 | ret = ib_post_send(tx_qp, &cur_wr.wr, bad_wr); |
9226 | if (ret) { |
9227 | /* Undo the effect of adding the outstanding wr */ |
9228 | - gsi->outstanding_pi = (gsi->outstanding_pi - 1) % |
9229 | - gsi->cap.max_send_wr; |
9230 | + gsi->outstanding_pi--; |
9231 | goto err; |
9232 | } |
9233 | spin_unlock_irqrestore(&gsi->lock, flags); |
9234 | diff --git a/drivers/md/dm.c b/drivers/md/dm.c |
9235 | index 36e6221fabab..dd154027adc9 100644 |
9236 | --- a/drivers/md/dm.c |
9237 | +++ b/drivers/md/dm.c |
9238 | @@ -1457,7 +1457,6 @@ void dm_init_md_queue(struct mapped_device *md) |
9239 | * - must do so here (in alloc_dev callchain) before queue is used |
9240 | */ |
9241 | md->queue->queuedata = md; |
9242 | - md->queue->backing_dev_info.congested_data = md; |
9243 | } |
9244 | |
9245 | void dm_init_normal_md_queue(struct mapped_device *md) |
9246 | @@ -1468,6 +1467,7 @@ void dm_init_normal_md_queue(struct mapped_device *md) |
9247 | /* |
9248 | * Initialize aspects of queue that aren't relevant for blk-mq |
9249 | */ |
9250 | + md->queue->backing_dev_info.congested_data = md; |
9251 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
9252 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
9253 | } |
9254 | @@ -1555,6 +1555,12 @@ static struct mapped_device *alloc_dev(int minor) |
9255 | goto bad; |
9256 | |
9257 | dm_init_md_queue(md); |
9258 | + /* |
9259 | + * default to bio-based required ->make_request_fn until DM |
9260 | + * table is loaded and md->type established. If request-based |
9261 | + * table is loaded: blk-mq will override accordingly. |
9262 | + */ |
9263 | + blk_queue_make_request(md->queue, dm_make_request); |
9264 | |
9265 | md->disk = alloc_disk_node(1, numa_node_id); |
9266 | if (!md->disk) |
9267 | @@ -1853,7 +1859,6 @@ int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
9268 | case DM_TYPE_BIO_BASED: |
9269 | case DM_TYPE_DAX_BIO_BASED: |
9270 | dm_init_normal_md_queue(md); |
9271 | - blk_queue_make_request(md->queue, dm_make_request); |
9272 | /* |
9273 | * DM handles splitting bios as needed. Free the bio_split bioset |
9274 | * since it won't be used (saves 1 process per bio-based DM device). |
9275 | diff --git a/drivers/md/persistent-data/dm-space-map-common.c b/drivers/md/persistent-data/dm-space-map-common.c |
9276 | index 306d2e4502c4..22729fd92a1b 100644 |
9277 | --- a/drivers/md/persistent-data/dm-space-map-common.c |
9278 | +++ b/drivers/md/persistent-data/dm-space-map-common.c |
9279 | @@ -382,6 +382,33 @@ int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin, |
9280 | return -ENOSPC; |
9281 | } |
9282 | |
9283 | +int sm_ll_find_common_free_block(struct ll_disk *old_ll, struct ll_disk *new_ll, |
9284 | + dm_block_t begin, dm_block_t end, dm_block_t *b) |
9285 | +{ |
9286 | + int r; |
9287 | + uint32_t count; |
9288 | + |
9289 | + do { |
9290 | + r = sm_ll_find_free_block(new_ll, begin, new_ll->nr_blocks, b); |
9291 | + if (r) |
9292 | + break; |
9293 | + |
9294 | + /* double check this block wasn't used in the old transaction */ |
9295 | + if (*b >= old_ll->nr_blocks) |
9296 | + count = 0; |
9297 | + else { |
9298 | + r = sm_ll_lookup(old_ll, *b, &count); |
9299 | + if (r) |
9300 | + break; |
9301 | + |
9302 | + if (count) |
9303 | + begin = *b + 1; |
9304 | + } |
9305 | + } while (count); |
9306 | + |
9307 | + return r; |
9308 | +} |
9309 | + |
9310 | static int sm_ll_mutate(struct ll_disk *ll, dm_block_t b, |
9311 | int (*mutator)(void *context, uint32_t old, uint32_t *new), |
9312 | void *context, enum allocation_event *ev) |
9313 | diff --git a/drivers/md/persistent-data/dm-space-map-common.h b/drivers/md/persistent-data/dm-space-map-common.h |
9314 | index b3078d5eda0c..8de63ce39bdd 100644 |
9315 | --- a/drivers/md/persistent-data/dm-space-map-common.h |
9316 | +++ b/drivers/md/persistent-data/dm-space-map-common.h |
9317 | @@ -109,6 +109,8 @@ int sm_ll_lookup_bitmap(struct ll_disk *ll, dm_block_t b, uint32_t *result); |
9318 | int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result); |
9319 | int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin, |
9320 | dm_block_t end, dm_block_t *result); |
9321 | +int sm_ll_find_common_free_block(struct ll_disk *old_ll, struct ll_disk *new_ll, |
9322 | + dm_block_t begin, dm_block_t end, dm_block_t *result); |
9323 | int sm_ll_insert(struct ll_disk *ll, dm_block_t b, uint32_t ref_count, enum allocation_event *ev); |
9324 | int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev); |
9325 | int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev); |
9326 | diff --git a/drivers/md/persistent-data/dm-space-map-disk.c b/drivers/md/persistent-data/dm-space-map-disk.c |
9327 | index 32adf6b4a9c7..bf4c5e2ccb6f 100644 |
9328 | --- a/drivers/md/persistent-data/dm-space-map-disk.c |
9329 | +++ b/drivers/md/persistent-data/dm-space-map-disk.c |
9330 | @@ -167,8 +167,10 @@ static int sm_disk_new_block(struct dm_space_map *sm, dm_block_t *b) |
9331 | enum allocation_event ev; |
9332 | struct sm_disk *smd = container_of(sm, struct sm_disk, sm); |
9333 | |
9334 | - /* FIXME: we should loop round a couple of times */ |
9335 | - r = sm_ll_find_free_block(&smd->old_ll, smd->begin, smd->old_ll.nr_blocks, b); |
9336 | + /* |
9337 | + * Any block we allocate has to be free in both the old and current ll. |
9338 | + */ |
9339 | + r = sm_ll_find_common_free_block(&smd->old_ll, &smd->ll, smd->begin, smd->ll.nr_blocks, b); |
9340 | if (r) |
9341 | return r; |
9342 | |
9343 | diff --git a/drivers/md/persistent-data/dm-space-map-metadata.c b/drivers/md/persistent-data/dm-space-map-metadata.c |
9344 | index 1d29771af380..967d8f2a731f 100644 |
9345 | --- a/drivers/md/persistent-data/dm-space-map-metadata.c |
9346 | +++ b/drivers/md/persistent-data/dm-space-map-metadata.c |
9347 | @@ -447,7 +447,10 @@ static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b) |
9348 | enum allocation_event ev; |
9349 | struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm); |
9350 | |
9351 | - r = sm_ll_find_free_block(&smm->old_ll, smm->begin, smm->old_ll.nr_blocks, b); |
9352 | + /* |
9353 | + * Any block we allocate has to be free in both the old and current ll. |
9354 | + */ |
9355 | + r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, smm->begin, smm->ll.nr_blocks, b); |
9356 | if (r) |
9357 | return r; |
9358 | |
9359 | diff --git a/drivers/media/rc/iguanair.c b/drivers/media/rc/iguanair.c |
9360 | index 25470395c43f..246795c31553 100644 |
9361 | --- a/drivers/media/rc/iguanair.c |
9362 | +++ b/drivers/media/rc/iguanair.c |
9363 | @@ -430,7 +430,7 @@ static int iguanair_probe(struct usb_interface *intf, |
9364 | int ret, pipein, pipeout; |
9365 | struct usb_host_interface *idesc; |
9366 | |
9367 | - idesc = intf->altsetting; |
9368 | + idesc = intf->cur_altsetting; |
9369 | if (idesc->desc.bNumEndpoints < 2) |
9370 | return -ENODEV; |
9371 | |
9372 | diff --git a/drivers/media/usb/uvc/uvc_driver.c b/drivers/media/usb/uvc/uvc_driver.c |
9373 | index 7c375b6dd318..9803135f2e59 100644 |
9374 | --- a/drivers/media/usb/uvc/uvc_driver.c |
9375 | +++ b/drivers/media/usb/uvc/uvc_driver.c |
9376 | @@ -1411,6 +1411,11 @@ static int uvc_scan_chain_forward(struct uvc_video_chain *chain, |
9377 | break; |
9378 | if (forward == prev) |
9379 | continue; |
9380 | + if (forward->chain.next || forward->chain.prev) { |
9381 | + uvc_trace(UVC_TRACE_DESCR, "Found reference to " |
9382 | + "entity %d already in chain.\n", forward->id); |
9383 | + return -EINVAL; |
9384 | + } |
9385 | |
9386 | switch (UVC_ENTITY_TYPE(forward)) { |
9387 | case UVC_VC_EXTENSION_UNIT: |
9388 | @@ -1492,6 +1497,13 @@ static int uvc_scan_chain_backward(struct uvc_video_chain *chain, |
9389 | return -1; |
9390 | } |
9391 | |
9392 | + if (term->chain.next || term->chain.prev) { |
9393 | + uvc_trace(UVC_TRACE_DESCR, "Found reference to " |
9394 | + "entity %d already in chain.\n", |
9395 | + term->id); |
9396 | + return -EINVAL; |
9397 | + } |
9398 | + |
9399 | if (uvc_trace_param & UVC_TRACE_PROBE) |
9400 | printk(" %d", term->id); |
9401 | |
9402 | diff --git a/drivers/media/v4l2-core/videobuf-dma-sg.c b/drivers/media/v4l2-core/videobuf-dma-sg.c |
9403 | index b6189a4958c5..025822bc6941 100644 |
9404 | --- a/drivers/media/v4l2-core/videobuf-dma-sg.c |
9405 | +++ b/drivers/media/v4l2-core/videobuf-dma-sg.c |
9406 | @@ -352,8 +352,11 @@ int videobuf_dma_free(struct videobuf_dmabuf *dma) |
9407 | BUG_ON(dma->sglen); |
9408 | |
9409 | if (dma->pages) { |
9410 | - for (i = 0; i < dma->nr_pages; i++) |
9411 | + for (i = 0; i < dma->nr_pages; i++) { |
9412 | + if (dma->direction == DMA_FROM_DEVICE) |
9413 | + set_page_dirty_lock(dma->pages[i]); |
9414 | put_page(dma->pages[i]); |
9415 | + } |
9416 | kfree(dma->pages); |
9417 | dma->pages = NULL; |
9418 | } |
9419 | diff --git a/drivers/mfd/da9062-core.c b/drivers/mfd/da9062-core.c |
9420 | index 8f873866ea60..86aab322da33 100644 |
9421 | --- a/drivers/mfd/da9062-core.c |
9422 | +++ b/drivers/mfd/da9062-core.c |
9423 | @@ -142,7 +142,7 @@ static const struct mfd_cell da9062_devs[] = { |
9424 | .name = "da9062-watchdog", |
9425 | .num_resources = ARRAY_SIZE(da9062_wdt_resources), |
9426 | .resources = da9062_wdt_resources, |
9427 | - .of_compatible = "dlg,da9062-wdt", |
9428 | + .of_compatible = "dlg,da9062-watchdog", |
9429 | }, |
9430 | { |
9431 | .name = "da9062-thermal", |
9432 | diff --git a/drivers/mfd/dln2.c b/drivers/mfd/dln2.c |
9433 | index 704e189ca162..95d0f2df0ad4 100644 |
9434 | --- a/drivers/mfd/dln2.c |
9435 | +++ b/drivers/mfd/dln2.c |
9436 | @@ -729,6 +729,8 @@ static int dln2_probe(struct usb_interface *interface, |
9437 | const struct usb_device_id *usb_id) |
9438 | { |
9439 | struct usb_host_interface *hostif = interface->cur_altsetting; |
9440 | + struct usb_endpoint_descriptor *epin; |
9441 | + struct usb_endpoint_descriptor *epout; |
9442 | struct device *dev = &interface->dev; |
9443 | struct dln2_dev *dln2; |
9444 | int ret; |
9445 | @@ -738,12 +740,19 @@ static int dln2_probe(struct usb_interface *interface, |
9446 | hostif->desc.bNumEndpoints < 2) |
9447 | return -ENODEV; |
9448 | |
9449 | + epin = &hostif->endpoint[0].desc; |
9450 | + epout = &hostif->endpoint[1].desc; |
9451 | + if (!usb_endpoint_is_bulk_out(epout)) |
9452 | + return -ENODEV; |
9453 | + if (!usb_endpoint_is_bulk_in(epin)) |
9454 | + return -ENODEV; |
9455 | + |
9456 | dln2 = kzalloc(sizeof(*dln2), GFP_KERNEL); |
9457 | if (!dln2) |
9458 | return -ENOMEM; |
9459 | |
9460 | - dln2->ep_out = hostif->endpoint[0].desc.bEndpointAddress; |
9461 | - dln2->ep_in = hostif->endpoint[1].desc.bEndpointAddress; |
9462 | + dln2->ep_out = epout->bEndpointAddress; |
9463 | + dln2->ep_in = epin->bEndpointAddress; |
9464 | dln2->usb_dev = usb_get_dev(interface_to_usbdev(interface)); |
9465 | dln2->interface = interface; |
9466 | usb_set_intfdata(interface, dln2); |
9467 | diff --git a/drivers/mfd/rn5t618.c b/drivers/mfd/rn5t618.c |
9468 | index ee94080e1cbb..dd20c3e32352 100644 |
9469 | --- a/drivers/mfd/rn5t618.c |
9470 | +++ b/drivers/mfd/rn5t618.c |
9471 | @@ -32,6 +32,7 @@ static bool rn5t618_volatile_reg(struct device *dev, unsigned int reg) |
9472 | case RN5T618_WATCHDOGCNT: |
9473 | case RN5T618_DCIRQ: |
9474 | case RN5T618_ILIMDATAH ... RN5T618_AIN0DATAL: |
9475 | + case RN5T618_ADCCNT3: |
9476 | case RN5T618_IR_ADC1 ... RN5T618_IR_ADC3: |
9477 | case RN5T618_IR_GPR: |
9478 | case RN5T618_IR_GPF: |
9479 | diff --git a/drivers/mmc/host/mmc_spi.c b/drivers/mmc/host/mmc_spi.c |
9480 | index c2df68e958b3..279d5da6e54b 100644 |
9481 | --- a/drivers/mmc/host/mmc_spi.c |
9482 | +++ b/drivers/mmc/host/mmc_spi.c |
9483 | @@ -1157,17 +1157,22 @@ static void mmc_spi_initsequence(struct mmc_spi_host *host) |
9484 | * SPI protocol. Another is that when chipselect is released while |
9485 | * the card returns BUSY status, the clock must issue several cycles |
9486 | * with chipselect high before the card will stop driving its output. |
9487 | + * |
9488 | + * SPI_CS_HIGH means "asserted" here. In some cases like when using |
9489 | + * GPIOs for chip select, SPI_CS_HIGH is set but this will be logically |
9490 | + * inverted by gpiolib, so if we want to ascertain to drive it high |
9491 | + * we should toggle the default with an XOR as we do here. |
9492 | */ |
9493 | - host->spi->mode |= SPI_CS_HIGH; |
9494 | + host->spi->mode ^= SPI_CS_HIGH; |
9495 | if (spi_setup(host->spi) != 0) { |
9496 | /* Just warn; most cards work without it. */ |
9497 | dev_warn(&host->spi->dev, |
9498 | "can't change chip-select polarity\n"); |
9499 | - host->spi->mode &= ~SPI_CS_HIGH; |
9500 | + host->spi->mode ^= SPI_CS_HIGH; |
9501 | } else { |
9502 | mmc_spi_readbytes(host, 18); |
9503 | |
9504 | - host->spi->mode &= ~SPI_CS_HIGH; |
9505 | + host->spi->mode ^= SPI_CS_HIGH; |
9506 | if (spi_setup(host->spi) != 0) { |
9507 | /* Wot, we can't get the same setup we had before? */ |
9508 | dev_err(&host->spi->dev, |
9509 | diff --git a/drivers/mtd/ubi/fastmap.c b/drivers/mtd/ubi/fastmap.c |
9510 | index b44c8d348e78..e7b177c61642 100644 |
9511 | --- a/drivers/mtd/ubi/fastmap.c |
9512 | +++ b/drivers/mtd/ubi/fastmap.c |
9513 | @@ -73,7 +73,7 @@ static int self_check_seen(struct ubi_device *ubi, unsigned long *seen) |
9514 | return 0; |
9515 | |
9516 | for (pnum = 0; pnum < ubi->peb_count; pnum++) { |
9517 | - if (test_bit(pnum, seen) && ubi->lookuptbl[pnum]) { |
9518 | + if (!test_bit(pnum, seen) && ubi->lookuptbl[pnum]) { |
9519 | ubi_err(ubi, "self-check failed for PEB %d, fastmap didn't see it", pnum); |
9520 | ret = -EINVAL; |
9521 | } |
9522 | @@ -1127,7 +1127,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9523 | struct rb_node *tmp_rb; |
9524 | int ret, i, j, free_peb_count, used_peb_count, vol_count; |
9525 | int scrub_peb_count, erase_peb_count; |
9526 | - unsigned long *seen_pebs = NULL; |
9527 | + unsigned long *seen_pebs; |
9528 | |
9529 | fm_raw = ubi->fm_buf; |
9530 | memset(ubi->fm_buf, 0, ubi->fm_size); |
9531 | @@ -1141,7 +1141,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9532 | dvbuf = new_fm_vbuf(ubi, UBI_FM_DATA_VOLUME_ID); |
9533 | if (!dvbuf) { |
9534 | ret = -ENOMEM; |
9535 | - goto out_kfree; |
9536 | + goto out_free_avbuf; |
9537 | } |
9538 | |
9539 | avhdr = ubi_get_vid_hdr(avbuf); |
9540 | @@ -1150,7 +1150,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9541 | seen_pebs = init_seen(ubi); |
9542 | if (IS_ERR(seen_pebs)) { |
9543 | ret = PTR_ERR(seen_pebs); |
9544 | - goto out_kfree; |
9545 | + goto out_free_dvbuf; |
9546 | } |
9547 | |
9548 | spin_lock(&ubi->volumes_lock); |
9549 | @@ -1318,7 +1318,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9550 | ret = ubi_io_write_vid_hdr(ubi, new_fm->e[0]->pnum, avbuf); |
9551 | if (ret) { |
9552 | ubi_err(ubi, "unable to write vid_hdr to fastmap SB!"); |
9553 | - goto out_kfree; |
9554 | + goto out_free_seen; |
9555 | } |
9556 | |
9557 | for (i = 0; i < new_fm->used_blocks; i++) { |
9558 | @@ -1340,7 +1340,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9559 | if (ret) { |
9560 | ubi_err(ubi, "unable to write vid_hdr to PEB %i!", |
9561 | new_fm->e[i]->pnum); |
9562 | - goto out_kfree; |
9563 | + goto out_free_seen; |
9564 | } |
9565 | } |
9566 | |
9567 | @@ -1350,7 +1350,7 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9568 | if (ret) { |
9569 | ubi_err(ubi, "unable to write fastmap to PEB %i!", |
9570 | new_fm->e[i]->pnum); |
9571 | - goto out_kfree; |
9572 | + goto out_free_seen; |
9573 | } |
9574 | } |
9575 | |
9576 | @@ -1360,10 +1360,13 @@ static int ubi_write_fastmap(struct ubi_device *ubi, |
9577 | ret = self_check_seen(ubi, seen_pebs); |
9578 | dbg_bld("fastmap written!"); |
9579 | |
9580 | -out_kfree: |
9581 | - ubi_free_vid_buf(avbuf); |
9582 | - ubi_free_vid_buf(dvbuf); |
9583 | +out_free_seen: |
9584 | free_seen(seen_pebs); |
9585 | +out_free_dvbuf: |
9586 | + ubi_free_vid_buf(dvbuf); |
9587 | +out_free_avbuf: |
9588 | + ubi_free_vid_buf(avbuf); |
9589 | + |
9590 | out: |
9591 | return ret; |
9592 | } |
9593 | diff --git a/drivers/net/bonding/bond_alb.c b/drivers/net/bonding/bond_alb.c |
9594 | index 91d8a48e53c3..9834d28d52e8 100644 |
9595 | --- a/drivers/net/bonding/bond_alb.c |
9596 | +++ b/drivers/net/bonding/bond_alb.c |
9597 | @@ -1371,26 +1371,31 @@ int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
9598 | bool do_tx_balance = true; |
9599 | u32 hash_index = 0; |
9600 | const u8 *hash_start = NULL; |
9601 | - struct ipv6hdr *ip6hdr; |
9602 | |
9603 | skb_reset_mac_header(skb); |
9604 | eth_data = eth_hdr(skb); |
9605 | |
9606 | switch (ntohs(skb->protocol)) { |
9607 | case ETH_P_IP: { |
9608 | - const struct iphdr *iph = ip_hdr(skb); |
9609 | + const struct iphdr *iph; |
9610 | |
9611 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) || |
9612 | - (iph->daddr == ip_bcast) || |
9613 | - (iph->protocol == IPPROTO_IGMP)) { |
9614 | + (!pskb_network_may_pull(skb, sizeof(*iph)))) { |
9615 | + do_tx_balance = false; |
9616 | + break; |
9617 | + } |
9618 | + iph = ip_hdr(skb); |
9619 | + if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) { |
9620 | do_tx_balance = false; |
9621 | break; |
9622 | } |
9623 | hash_start = (char *)&(iph->daddr); |
9624 | hash_size = sizeof(iph->daddr); |
9625 | - } |
9626 | break; |
9627 | - case ETH_P_IPV6: |
9628 | + } |
9629 | + case ETH_P_IPV6: { |
9630 | + const struct ipv6hdr *ip6hdr; |
9631 | + |
9632 | /* IPv6 doesn't really use broadcast mac address, but leave |
9633 | * that here just in case. |
9634 | */ |
9635 | @@ -1407,7 +1412,11 @@ int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
9636 | break; |
9637 | } |
9638 | |
9639 | - /* Additianally, DAD probes should not be tx-balanced as that |
9640 | + if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) { |
9641 | + do_tx_balance = false; |
9642 | + break; |
9643 | + } |
9644 | + /* Additionally, DAD probes should not be tx-balanced as that |
9645 | * will lead to false positives for duplicate addresses and |
9646 | * prevent address configuration from working. |
9647 | */ |
9648 | @@ -1417,17 +1426,26 @@ int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
9649 | break; |
9650 | } |
9651 | |
9652 | - hash_start = (char *)&(ipv6_hdr(skb)->daddr); |
9653 | - hash_size = sizeof(ipv6_hdr(skb)->daddr); |
9654 | + hash_start = (char *)&ip6hdr->daddr; |
9655 | + hash_size = sizeof(ip6hdr->daddr); |
9656 | break; |
9657 | - case ETH_P_IPX: |
9658 | - if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) { |
9659 | + } |
9660 | + case ETH_P_IPX: { |
9661 | + const struct ipxhdr *ipxhdr; |
9662 | + |
9663 | + if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) { |
9664 | + do_tx_balance = false; |
9665 | + break; |
9666 | + } |
9667 | + ipxhdr = (struct ipxhdr *)skb_network_header(skb); |
9668 | + |
9669 | + if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) { |
9670 | /* something is wrong with this packet */ |
9671 | do_tx_balance = false; |
9672 | break; |
9673 | } |
9674 | |
9675 | - if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) { |
9676 | + if (ipxhdr->ipx_type != IPX_TYPE_NCP) { |
9677 | /* The only protocol worth balancing in |
9678 | * this family since it has an "ARP" like |
9679 | * mechanism |
9680 | @@ -1436,9 +1454,11 @@ int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
9681 | break; |
9682 | } |
9683 | |
9684 | + eth_data = eth_hdr(skb); |
9685 | hash_start = (char *)eth_data->h_dest; |
9686 | hash_size = ETH_ALEN; |
9687 | break; |
9688 | + } |
9689 | case ETH_P_ARP: |
9690 | do_tx_balance = false; |
9691 | if (bond_info->rlb_enabled) |
9692 | diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c |
9693 | index e3b41af65d18..6519dd33c7ca 100644 |
9694 | --- a/drivers/net/ethernet/broadcom/bcmsysport.c |
9695 | +++ b/drivers/net/ethernet/broadcom/bcmsysport.c |
9696 | @@ -1983,6 +1983,9 @@ static int bcm_sysport_resume(struct device *d) |
9697 | |
9698 | umac_reset(priv); |
9699 | |
9700 | + /* Disable the UniMAC RX/TX */ |
9701 | + umac_enable_set(priv, CMD_RX_EN | CMD_TX_EN, 0); |
9702 | + |
9703 | /* We may have been suspended and never received a WOL event that |
9704 | * would turn off MPD detection, take care of that now |
9705 | */ |
9706 | diff --git a/drivers/net/ethernet/dec/tulip/dmfe.c b/drivers/net/ethernet/dec/tulip/dmfe.c |
9707 | index 8ed0fd8b1dda..74cb9b3c2f41 100644 |
9708 | --- a/drivers/net/ethernet/dec/tulip/dmfe.c |
9709 | +++ b/drivers/net/ethernet/dec/tulip/dmfe.c |
9710 | @@ -2225,15 +2225,16 @@ static int __init dmfe_init_module(void) |
9711 | if (cr6set) |
9712 | dmfe_cr6_user_set = cr6set; |
9713 | |
9714 | - switch(mode) { |
9715 | - case DMFE_10MHF: |
9716 | + switch (mode) { |
9717 | + case DMFE_10MHF: |
9718 | case DMFE_100MHF: |
9719 | case DMFE_10MFD: |
9720 | case DMFE_100MFD: |
9721 | case DMFE_1M_HPNA: |
9722 | dmfe_media_mode = mode; |
9723 | break; |
9724 | - default:dmfe_media_mode = DMFE_AUTO; |
9725 | + default: |
9726 | + dmfe_media_mode = DMFE_AUTO; |
9727 | break; |
9728 | } |
9729 | |
9730 | diff --git a/drivers/net/ethernet/dec/tulip/uli526x.c b/drivers/net/ethernet/dec/tulip/uli526x.c |
9731 | index e750b5ddc0fb..5f79e2731b76 100644 |
9732 | --- a/drivers/net/ethernet/dec/tulip/uli526x.c |
9733 | +++ b/drivers/net/ethernet/dec/tulip/uli526x.c |
9734 | @@ -1813,8 +1813,8 @@ static int __init uli526x_init_module(void) |
9735 | if (cr6set) |
9736 | uli526x_cr6_user_set = cr6set; |
9737 | |
9738 | - switch (mode) { |
9739 | - case ULI526X_10MHF: |
9740 | + switch (mode) { |
9741 | + case ULI526X_10MHF: |
9742 | case ULI526X_100MHF: |
9743 | case ULI526X_10MFD: |
9744 | case ULI526X_100MFD: |
9745 | diff --git a/drivers/net/ethernet/smsc/smc911x.c b/drivers/net/ethernet/smsc/smc911x.c |
9746 | index 323b3ac16bc0..d0cf971aa4eb 100644 |
9747 | --- a/drivers/net/ethernet/smsc/smc911x.c |
9748 | +++ b/drivers/net/ethernet/smsc/smc911x.c |
9749 | @@ -948,7 +948,7 @@ static void smc911x_phy_configure(struct work_struct *work) |
9750 | if (lp->ctl_rspeed != 100) |
9751 | my_ad_caps &= ~(ADVERTISE_100BASE4|ADVERTISE_100FULL|ADVERTISE_100HALF); |
9752 | |
9753 | - if (!lp->ctl_rfduplx) |
9754 | + if (!lp->ctl_rfduplx) |
9755 | my_ad_caps &= ~(ADVERTISE_100FULL|ADVERTISE_10FULL); |
9756 | |
9757 | /* Update our Auto-Neg Advertisement Register */ |
9758 | diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c |
9759 | index 5077c69eb652..a9e8a7356c41 100644 |
9760 | --- a/drivers/net/gtp.c |
9761 | +++ b/drivers/net/gtp.c |
9762 | @@ -784,11 +784,13 @@ static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize) |
9763 | { |
9764 | int i; |
9765 | |
9766 | - gtp->addr_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL); |
9767 | + gtp->addr_hash = kmalloc(sizeof(struct hlist_head) * hsize, |
9768 | + GFP_KERNEL | __GFP_NOWARN); |
9769 | if (gtp->addr_hash == NULL) |
9770 | return -ENOMEM; |
9771 | |
9772 | - gtp->tid_hash = kmalloc(sizeof(struct hlist_head) * hsize, GFP_KERNEL); |
9773 | + gtp->tid_hash = kmalloc(sizeof(struct hlist_head) * hsize, |
9774 | + GFP_KERNEL | __GFP_NOWARN); |
9775 | if (gtp->tid_hash == NULL) |
9776 | goto err1; |
9777 | |
9778 | diff --git a/drivers/net/ppp/ppp_async.c b/drivers/net/ppp/ppp_async.c |
9779 | index 9c889e0303dd..cef40de1bd05 100644 |
9780 | --- a/drivers/net/ppp/ppp_async.c |
9781 | +++ b/drivers/net/ppp/ppp_async.c |
9782 | @@ -878,15 +878,15 @@ ppp_async_input(struct asyncppp *ap, const unsigned char *buf, |
9783 | skb = dev_alloc_skb(ap->mru + PPP_HDRLEN + 2); |
9784 | if (!skb) |
9785 | goto nomem; |
9786 | - ap->rpkt = skb; |
9787 | - } |
9788 | - if (skb->len == 0) { |
9789 | - /* Try to get the payload 4-byte aligned. |
9790 | - * This should match the |
9791 | - * PPP_ALLSTATIONS/PPP_UI/compressed tests in |
9792 | - * process_input_packet, but we do not have |
9793 | - * enough chars here to test buf[1] and buf[2]. |
9794 | - */ |
9795 | + ap->rpkt = skb; |
9796 | + } |
9797 | + if (skb->len == 0) { |
9798 | + /* Try to get the payload 4-byte aligned. |
9799 | + * This should match the |
9800 | + * PPP_ALLSTATIONS/PPP_UI/compressed tests in |
9801 | + * process_input_packet, but we do not have |
9802 | + * enough chars here to test buf[1] and buf[2]. |
9803 | + */ |
9804 | if (buf[0] != PPP_ALLSTATIONS) |
9805 | skb_reserve(skb, 2 + (buf[0] & 1)); |
9806 | } |
9807 | diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c |
9808 | index 05df9d8f76e9..31727f34381f 100644 |
9809 | --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c |
9810 | +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c |
9811 | @@ -438,6 +438,7 @@ fail: |
9812 | usb_free_urb(req->urb); |
9813 | list_del(q->next); |
9814 | } |
9815 | + kfree(reqs); |
9816 | return NULL; |
9817 | |
9818 | } |
9819 | diff --git a/drivers/net/wireless/marvell/libertas/cfg.c b/drivers/net/wireless/marvell/libertas/cfg.c |
9820 | index 3eab802c7d3f..ece6d72cf90c 100644 |
9821 | --- a/drivers/net/wireless/marvell/libertas/cfg.c |
9822 | +++ b/drivers/net/wireless/marvell/libertas/cfg.c |
9823 | @@ -1859,6 +1859,8 @@ static int lbs_ibss_join_existing(struct lbs_private *priv, |
9824 | rates_max = rates_eid[1]; |
9825 | if (rates_max > MAX_RATES) { |
9826 | lbs_deb_join("invalid rates"); |
9827 | + rcu_read_unlock(); |
9828 | + ret = -EINVAL; |
9829 | goto out; |
9830 | } |
9831 | rates = cmd.bss.rates; |
9832 | diff --git a/drivers/net/wireless/marvell/mwifiex/scan.c b/drivers/net/wireless/marvell/mwifiex/scan.c |
9833 | index 828c6f5eb83c..5fde2e2f1fea 100644 |
9834 | --- a/drivers/net/wireless/marvell/mwifiex/scan.c |
9835 | +++ b/drivers/net/wireless/marvell/mwifiex/scan.c |
9836 | @@ -2878,6 +2878,13 @@ mwifiex_cmd_append_vsie_tlv(struct mwifiex_private *priv, |
9837 | vs_param_set->header.len = |
9838 | cpu_to_le16((((u16) priv->vs_ie[id].ie[1]) |
9839 | & 0x00FF) + 2); |
9840 | + if (le16_to_cpu(vs_param_set->header.len) > |
9841 | + MWIFIEX_MAX_VSIE_LEN) { |
9842 | + mwifiex_dbg(priv->adapter, ERROR, |
9843 | + "Invalid param length!\n"); |
9844 | + break; |
9845 | + } |
9846 | + |
9847 | memcpy(vs_param_set->ie, priv->vs_ie[id].ie, |
9848 | le16_to_cpu(vs_param_set->header.len)); |
9849 | *buffer += le16_to_cpu(vs_param_set->header.len) + |
9850 | diff --git a/drivers/net/wireless/marvell/mwifiex/sta_ioctl.c b/drivers/net/wireless/marvell/mwifiex/sta_ioctl.c |
9851 | index be3be7a63cf0..f2d10ba19920 100644 |
9852 | --- a/drivers/net/wireless/marvell/mwifiex/sta_ioctl.c |
9853 | +++ b/drivers/net/wireless/marvell/mwifiex/sta_ioctl.c |
9854 | @@ -274,6 +274,7 @@ static int mwifiex_process_country_ie(struct mwifiex_private *priv, |
9855 | |
9856 | if (country_ie_len > |
9857 | (IEEE80211_COUNTRY_STRING_LEN + MWIFIEX_MAX_TRIPLET_802_11D)) { |
9858 | + rcu_read_unlock(); |
9859 | mwifiex_dbg(priv->adapter, ERROR, |
9860 | "11D: country_ie_len overflow!, deauth AP\n"); |
9861 | return -EINVAL; |
9862 | diff --git a/drivers/net/wireless/marvell/mwifiex/wmm.c b/drivers/net/wireless/marvell/mwifiex/wmm.c |
9863 | index 9843560e784f..c93fcafbcc7a 100644 |
9864 | --- a/drivers/net/wireless/marvell/mwifiex/wmm.c |
9865 | +++ b/drivers/net/wireless/marvell/mwifiex/wmm.c |
9866 | @@ -980,6 +980,10 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv, |
9867 | "WMM Parameter Set Count: %d\n", |
9868 | wmm_param_ie->qos_info_bitmap & mask); |
9869 | |
9870 | + if (wmm_param_ie->vend_hdr.len + 2 > |
9871 | + sizeof(struct ieee_types_wmm_parameter)) |
9872 | + break; |
9873 | + |
9874 | memcpy((u8 *) &priv->curr_bss_params.bss_descriptor. |
9875 | wmm_ie, wmm_param_ie, |
9876 | wmm_param_ie->vend_hdr.len + 2); |
9877 | diff --git a/drivers/nfc/pn544/pn544.c b/drivers/nfc/pn544/pn544.c |
9878 | index 12e819ddf17a..3afc53ff7369 100644 |
9879 | --- a/drivers/nfc/pn544/pn544.c |
9880 | +++ b/drivers/nfc/pn544/pn544.c |
9881 | @@ -704,7 +704,7 @@ static int pn544_hci_check_presence(struct nfc_hci_dev *hdev, |
9882 | target->nfcid1_len != 10) |
9883 | return -EOPNOTSUPP; |
9884 | |
9885 | - return nfc_hci_send_cmd(hdev, NFC_HCI_RF_READER_A_GATE, |
9886 | + return nfc_hci_send_cmd(hdev, NFC_HCI_RF_READER_A_GATE, |
9887 | PN544_RF_READER_CMD_ACTIVATE_NEXT, |
9888 | target->nfcid1, target->nfcid1_len, NULL); |
9889 | } else if (target->supported_protocols & (NFC_PROTO_JEWEL_MASK | |
9890 | diff --git a/drivers/of/Kconfig b/drivers/of/Kconfig |
9891 | index ba7b034b2b91..6b8646db110c 100644 |
9892 | --- a/drivers/of/Kconfig |
9893 | +++ b/drivers/of/Kconfig |
9894 | @@ -112,4 +112,8 @@ config OF_OVERLAY |
9895 | config OF_NUMA |
9896 | bool |
9897 | |
9898 | +config OF_DMA_DEFAULT_COHERENT |
9899 | + # arches should select this if DMA is coherent by default for OF devices |
9900 | + bool |
9901 | + |
9902 | endif # OF |
9903 | diff --git a/drivers/of/address.c b/drivers/of/address.c |
9904 | index 72914cdfce2a..37619bb2c97a 100644 |
9905 | --- a/drivers/of/address.c |
9906 | +++ b/drivers/of/address.c |
9907 | @@ -896,12 +896,16 @@ EXPORT_SYMBOL_GPL(of_dma_get_range); |
9908 | * @np: device node |
9909 | * |
9910 | * It returns true if "dma-coherent" property was found |
9911 | - * for this device in DT. |
9912 | + * for this device in the DT, or if DMA is coherent by |
9913 | + * default for OF devices on the current platform. |
9914 | */ |
9915 | bool of_dma_is_coherent(struct device_node *np) |
9916 | { |
9917 | struct device_node *node = of_node_get(np); |
9918 | |
9919 | + if (IS_ENABLED(CONFIG_OF_DMA_DEFAULT_COHERENT)) |
9920 | + return true; |
9921 | + |
9922 | while (node) { |
9923 | if (of_property_read_bool(node, "dma-coherent")) { |
9924 | of_node_put(node); |
9925 | diff --git a/drivers/pci/host/pci-keystone-dw.c b/drivers/pci/host/pci-keystone-dw.c |
9926 | index 9397c4667106..f011a8780ff5 100644 |
9927 | --- a/drivers/pci/host/pci-keystone-dw.c |
9928 | +++ b/drivers/pci/host/pci-keystone-dw.c |
9929 | @@ -502,7 +502,7 @@ void ks_dw_pcie_initiate_link_train(struct keystone_pcie *ks_pcie) |
9930 | /* Disable Link training */ |
9931 | val = ks_dw_app_readl(ks_pcie, CMD_STATUS); |
9932 | val &= ~LTSSM_EN_VAL; |
9933 | - ks_dw_app_writel(ks_pcie, CMD_STATUS, LTSSM_EN_VAL | val); |
9934 | + ks_dw_app_writel(ks_pcie, CMD_STATUS, val); |
9935 | |
9936 | /* Initiate Link Training */ |
9937 | val = ks_dw_app_readl(ks_pcie, CMD_STATUS); |
9938 | diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c |
9939 | index f30ca75b5b6c..e631636a0aa5 100644 |
9940 | --- a/drivers/pci/setup-bus.c |
9941 | +++ b/drivers/pci/setup-bus.c |
9942 | @@ -1833,12 +1833,18 @@ again: |
9943 | /* restore size and flags */ |
9944 | list_for_each_entry(fail_res, &fail_head, list) { |
9945 | struct resource *res = fail_res->res; |
9946 | + int idx; |
9947 | |
9948 | res->start = fail_res->start; |
9949 | res->end = fail_res->end; |
9950 | res->flags = fail_res->flags; |
9951 | - if (fail_res->dev->subordinate) |
9952 | - res->flags = 0; |
9953 | + |
9954 | + if (pci_is_bridge(fail_res->dev)) { |
9955 | + idx = res - &fail_res->dev->resource[0]; |
9956 | + if (idx >= PCI_BRIDGE_RESOURCES && |
9957 | + idx <= PCI_BRIDGE_RESOURCE_END) |
9958 | + res->flags = 0; |
9959 | + } |
9960 | } |
9961 | free_list(&fail_head); |
9962 | |
9963 | @@ -1904,12 +1910,18 @@ again: |
9964 | /* restore size and flags */ |
9965 | list_for_each_entry(fail_res, &fail_head, list) { |
9966 | struct resource *res = fail_res->res; |
9967 | + int idx; |
9968 | |
9969 | res->start = fail_res->start; |
9970 | res->end = fail_res->end; |
9971 | res->flags = fail_res->flags; |
9972 | - if (fail_res->dev->subordinate) |
9973 | - res->flags = 0; |
9974 | + |
9975 | + if (pci_is_bridge(fail_res->dev)) { |
9976 | + idx = res - &fail_res->dev->resource[0]; |
9977 | + if (idx >= PCI_BRIDGE_RESOURCES && |
9978 | + idx <= PCI_BRIDGE_RESOURCE_END) |
9979 | + res->flags = 0; |
9980 | + } |
9981 | } |
9982 | free_list(&fail_head); |
9983 | |
9984 | diff --git a/drivers/pinctrl/sh-pfc/pfc-r8a7778.c b/drivers/pinctrl/sh-pfc/pfc-r8a7778.c |
9985 | index 18ef7042b3d1..771689a41dbf 100644 |
9986 | --- a/drivers/pinctrl/sh-pfc/pfc-r8a7778.c |
9987 | +++ b/drivers/pinctrl/sh-pfc/pfc-r8a7778.c |
9988 | @@ -2324,7 +2324,7 @@ static const struct pinmux_cfg_reg pinmux_config_regs[] = { |
9989 | FN_ATAG0_A, 0, FN_REMOCON_B, 0, |
9990 | /* IP0_11_8 [4] */ |
9991 | FN_SD1_DAT2_A, FN_MMC_D2, 0, FN_BS, |
9992 | - FN_ATADIR0_A, 0, FN_SDSELF_B, 0, |
9993 | + FN_ATADIR0_A, 0, FN_SDSELF_A, 0, |
9994 | FN_PWM4_B, 0, 0, 0, |
9995 | 0, 0, 0, 0, |
9996 | /* IP0_7_5 [3] */ |
9997 | @@ -2366,7 +2366,7 @@ static const struct pinmux_cfg_reg pinmux_config_regs[] = { |
9998 | FN_TS_SDAT0_A, 0, 0, 0, |
9999 | 0, 0, 0, 0, |
10000 | /* IP1_10_8 [3] */ |
10001 | - FN_SD1_CLK_B, FN_MMC_D6, 0, FN_A24, |
10002 | + FN_SD1_CD_A, FN_MMC_D6, 0, FN_A24, |
10003 | FN_DREQ1_A, 0, FN_HRX0_B, FN_TS_SPSYNC0_A, |
10004 | /* IP1_7_5 [3] */ |
10005 | FN_A23, FN_HTX0_B, FN_TX2_B, FN_DACK2_A, |
10006 | diff --git a/drivers/power/supply/ltc2941-battery-gauge.c b/drivers/power/supply/ltc2941-battery-gauge.c |
10007 | index 4adf2ba021ce..043de9d039d5 100644 |
10008 | --- a/drivers/power/supply/ltc2941-battery-gauge.c |
10009 | +++ b/drivers/power/supply/ltc2941-battery-gauge.c |
10010 | @@ -364,7 +364,7 @@ static int ltc294x_i2c_remove(struct i2c_client *client) |
10011 | { |
10012 | struct ltc294x_info *info = i2c_get_clientdata(client); |
10013 | |
10014 | - cancel_delayed_work(&info->work); |
10015 | + cancel_delayed_work_sync(&info->work); |
10016 | power_supply_unregister(info->supply); |
10017 | return 0; |
10018 | } |
10019 | diff --git a/drivers/rtc/rtc-cmos.c b/drivers/rtc/rtc-cmos.c |
10020 | index c554e529fc4e..b962dbe51750 100644 |
10021 | --- a/drivers/rtc/rtc-cmos.c |
10022 | +++ b/drivers/rtc/rtc-cmos.c |
10023 | @@ -730,7 +730,7 @@ cmos_do_probe(struct device *dev, struct resource *ports, int rtc_irq) |
10024 | rtc_cmos_int_handler = cmos_interrupt; |
10025 | |
10026 | retval = request_irq(rtc_irq, rtc_cmos_int_handler, |
10027 | - IRQF_SHARED, dev_name(&cmos_rtc.rtc->dev), |
10028 | + 0, dev_name(&cmos_rtc.rtc->dev), |
10029 | cmos_rtc.rtc); |
10030 | if (retval < 0) { |
10031 | dev_dbg(dev, "IRQ %d is already in use\n", rtc_irq); |
10032 | diff --git a/drivers/rtc/rtc-hym8563.c b/drivers/rtc/rtc-hym8563.c |
10033 | index e5ad527cb75e..a8c2d38b2411 100644 |
10034 | --- a/drivers/rtc/rtc-hym8563.c |
10035 | +++ b/drivers/rtc/rtc-hym8563.c |
10036 | @@ -105,7 +105,7 @@ static int hym8563_rtc_read_time(struct device *dev, struct rtc_time *tm) |
10037 | |
10038 | if (!hym8563->valid) { |
10039 | dev_warn(&client->dev, "no valid clock/calendar values available\n"); |
10040 | - return -EPERM; |
10041 | + return -EINVAL; |
10042 | } |
10043 | |
10044 | ret = i2c_smbus_read_i2c_block_data(client, HYM8563_SEC, 7, buf); |
10045 | diff --git a/drivers/scsi/csiostor/csio_scsi.c b/drivers/scsi/csiostor/csio_scsi.c |
10046 | index 894d97e4ace5..5db57671fa28 100644 |
10047 | --- a/drivers/scsi/csiostor/csio_scsi.c |
10048 | +++ b/drivers/scsi/csiostor/csio_scsi.c |
10049 | @@ -1383,7 +1383,7 @@ csio_device_reset(struct device *dev, |
10050 | return -EINVAL; |
10051 | |
10052 | /* Delete NPIV lnodes */ |
10053 | - csio_lnodes_exit(hw, 1); |
10054 | + csio_lnodes_exit(hw, 1); |
10055 | |
10056 | /* Block upper IOs */ |
10057 | csio_lnodes_block_request(hw); |
10058 | diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c |
10059 | index 10ae624dd266..9fa6a560b162 100644 |
10060 | --- a/drivers/scsi/megaraid/megaraid_sas_base.c |
10061 | +++ b/drivers/scsi/megaraid/megaraid_sas_base.c |
10062 | @@ -3978,7 +3978,8 @@ dcmd_timeout_ocr_possible(struct megasas_instance *instance) { |
10063 | if (!instance->ctrl_context) |
10064 | return KILL_ADAPTER; |
10065 | else if (instance->unload || |
10066 | - test_bit(MEGASAS_FUSION_IN_RESET, &instance->reset_flags)) |
10067 | + test_bit(MEGASAS_FUSION_OCR_NOT_POSSIBLE, |
10068 | + &instance->reset_flags)) |
10069 | return IGNORE_TIMEOUT; |
10070 | else |
10071 | return INITIATE_OCR; |
10072 | diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c |
10073 | index fe1a20973e47..874e5a7f7998 100644 |
10074 | --- a/drivers/scsi/megaraid/megaraid_sas_fusion.c |
10075 | +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c |
10076 | @@ -3438,6 +3438,7 @@ int megasas_reset_fusion(struct Scsi_Host *shost, int reason) |
10077 | if (instance->requestorId && !instance->skip_heartbeat_timer_del) |
10078 | del_timer_sync(&instance->sriov_heartbeat_timer); |
10079 | set_bit(MEGASAS_FUSION_IN_RESET, &instance->reset_flags); |
10080 | + set_bit(MEGASAS_FUSION_OCR_NOT_POSSIBLE, &instance->reset_flags); |
10081 | atomic_set(&instance->adprecovery, MEGASAS_ADPRESET_SM_POLLING); |
10082 | instance->instancet->disable_intr(instance); |
10083 | msleep(1000); |
10084 | @@ -3594,7 +3595,7 @@ fail_kill_adapter: |
10085 | atomic_set(&instance->adprecovery, MEGASAS_HBA_OPERATIONAL); |
10086 | } |
10087 | out: |
10088 | - clear_bit(MEGASAS_FUSION_IN_RESET, &instance->reset_flags); |
10089 | + clear_bit(MEGASAS_FUSION_OCR_NOT_POSSIBLE, &instance->reset_flags); |
10090 | mutex_unlock(&instance->reset_mutex); |
10091 | return retval; |
10092 | } |
10093 | diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.h b/drivers/scsi/megaraid/megaraid_sas_fusion.h |
10094 | index e3bee04c1eb1..034653d93365 100644 |
10095 | --- a/drivers/scsi/megaraid/megaraid_sas_fusion.h |
10096 | +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.h |
10097 | @@ -93,6 +93,7 @@ enum MR_RAID_FLAGS_IO_SUB_TYPE { |
10098 | |
10099 | #define MEGASAS_FP_CMD_LEN 16 |
10100 | #define MEGASAS_FUSION_IN_RESET 0 |
10101 | +#define MEGASAS_FUSION_OCR_NOT_POSSIBLE 1 |
10102 | #define THRESHOLD_REPLY_COUNT 50 |
10103 | #define JBOD_MAPS_COUNT 2 |
10104 | |
10105 | diff --git a/drivers/scsi/qla2xxx/qla_mbx.c b/drivers/scsi/qla2xxx/qla_mbx.c |
10106 | index bf29ad454118..9bbea4223917 100644 |
10107 | --- a/drivers/scsi/qla2xxx/qla_mbx.c |
10108 | +++ b/drivers/scsi/qla2xxx/qla_mbx.c |
10109 | @@ -5723,9 +5723,8 @@ qla2x00_dump_mctp_data(scsi_qla_host_t *vha, dma_addr_t req_dma, uint32_t addr, |
10110 | mcp->mb[7] = LSW(MSD(req_dma)); |
10111 | mcp->mb[8] = MSW(addr); |
10112 | /* Setting RAM ID to valid */ |
10113 | - mcp->mb[10] |= BIT_7; |
10114 | /* For MCTP RAM ID is 0x40 */ |
10115 | - mcp->mb[10] |= 0x40; |
10116 | + mcp->mb[10] = BIT_7 | 0x40; |
10117 | |
10118 | mcp->out_mb |= MBX_10|MBX_8|MBX_7|MBX_6|MBX_5|MBX_4|MBX_3|MBX_2|MBX_1| |
10119 | MBX_0; |
10120 | diff --git a/drivers/scsi/qla2xxx/qla_nx.c b/drivers/scsi/qla2xxx/qla_nx.c |
10121 | index 54380b434b30..104e13ae3428 100644 |
10122 | --- a/drivers/scsi/qla2xxx/qla_nx.c |
10123 | +++ b/drivers/scsi/qla2xxx/qla_nx.c |
10124 | @@ -1600,8 +1600,7 @@ qla82xx_get_bootld_offset(struct qla_hw_data *ha) |
10125 | return (u8 *)&ha->hablob->fw->data[offset]; |
10126 | } |
10127 | |
10128 | -static __le32 |
10129 | -qla82xx_get_fw_size(struct qla_hw_data *ha) |
10130 | +static u32 qla82xx_get_fw_size(struct qla_hw_data *ha) |
10131 | { |
10132 | struct qla82xx_uri_data_desc *uri_desc = NULL; |
10133 | |
10134 | @@ -1612,7 +1611,7 @@ qla82xx_get_fw_size(struct qla_hw_data *ha) |
10135 | return cpu_to_le32(uri_desc->size); |
10136 | } |
10137 | |
10138 | - return cpu_to_le32(*(u32 *)&ha->hablob->fw->data[FW_SIZE_OFFSET]); |
10139 | + return get_unaligned_le32(&ha->hablob->fw->data[FW_SIZE_OFFSET]); |
10140 | } |
10141 | |
10142 | static u8 * |
10143 | @@ -1803,7 +1802,7 @@ qla82xx_fw_load_from_blob(struct qla_hw_data *ha) |
10144 | } |
10145 | |
10146 | flashaddr = FLASH_ADDR_START; |
10147 | - size = (__force u32)qla82xx_get_fw_size(ha) / 8; |
10148 | + size = qla82xx_get_fw_size(ha) / 8; |
10149 | ptr64 = (u64 *)qla82xx_get_fw_offs(ha); |
10150 | |
10151 | for (i = 0; i < size; i++) { |
10152 | diff --git a/drivers/scsi/qla4xxx/ql4_os.c b/drivers/scsi/qla4xxx/ql4_os.c |
10153 | index f714d5f917d1..3fda5836aac6 100644 |
10154 | --- a/drivers/scsi/qla4xxx/ql4_os.c |
10155 | +++ b/drivers/scsi/qla4xxx/ql4_os.c |
10156 | @@ -4150,7 +4150,7 @@ static void qla4xxx_mem_free(struct scsi_qla_host *ha) |
10157 | dma_free_coherent(&ha->pdev->dev, ha->queues_len, ha->queues, |
10158 | ha->queues_dma); |
10159 | |
10160 | - if (ha->fw_dump) |
10161 | + if (ha->fw_dump) |
10162 | vfree(ha->fw_dump); |
10163 | |
10164 | ha->queues_len = 0; |
10165 | diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c |
10166 | index 094e879af121..394df57894e6 100644 |
10167 | --- a/drivers/scsi/ufs/ufshcd.c |
10168 | +++ b/drivers/scsi/ufs/ufshcd.c |
10169 | @@ -5347,7 +5347,8 @@ static int ufshcd_probe_hba(struct ufs_hba *hba) |
10170 | ufshcd_init_icc_levels(hba); |
10171 | |
10172 | /* Add required well known logical units to scsi mid layer */ |
10173 | - if (ufshcd_scsi_add_wlus(hba)) |
10174 | + ret = ufshcd_scsi_add_wlus(hba); |
10175 | + if (ret) |
10176 | goto out; |
10177 | |
10178 | scsi_scan_host(hba->host); |
10179 | diff --git a/drivers/usb/gadget/function/f_ecm.c b/drivers/usb/gadget/function/f_ecm.c |
10180 | index dc99ed94f03d..8e3e44382785 100644 |
10181 | --- a/drivers/usb/gadget/function/f_ecm.c |
10182 | +++ b/drivers/usb/gadget/function/f_ecm.c |
10183 | @@ -56,6 +56,7 @@ struct f_ecm { |
10184 | struct usb_ep *notify; |
10185 | struct usb_request *notify_req; |
10186 | u8 notify_state; |
10187 | + atomic_t notify_count; |
10188 | bool is_open; |
10189 | |
10190 | /* FIXME is_open needs some irq-ish locking |
10191 | @@ -384,7 +385,7 @@ static void ecm_do_notify(struct f_ecm *ecm) |
10192 | int status; |
10193 | |
10194 | /* notification already in flight? */ |
10195 | - if (!req) |
10196 | + if (atomic_read(&ecm->notify_count)) |
10197 | return; |
10198 | |
10199 | event = req->buf; |
10200 | @@ -424,10 +425,10 @@ static void ecm_do_notify(struct f_ecm *ecm) |
10201 | event->bmRequestType = 0xA1; |
10202 | event->wIndex = cpu_to_le16(ecm->ctrl_id); |
10203 | |
10204 | - ecm->notify_req = NULL; |
10205 | + atomic_inc(&ecm->notify_count); |
10206 | status = usb_ep_queue(ecm->notify, req, GFP_ATOMIC); |
10207 | if (status < 0) { |
10208 | - ecm->notify_req = req; |
10209 | + atomic_dec(&ecm->notify_count); |
10210 | DBG(cdev, "notify --> %d\n", status); |
10211 | } |
10212 | } |
10213 | @@ -452,17 +453,19 @@ static void ecm_notify_complete(struct usb_ep *ep, struct usb_request *req) |
10214 | switch (req->status) { |
10215 | case 0: |
10216 | /* no fault */ |
10217 | + atomic_dec(&ecm->notify_count); |
10218 | break; |
10219 | case -ECONNRESET: |
10220 | case -ESHUTDOWN: |
10221 | + atomic_set(&ecm->notify_count, 0); |
10222 | ecm->notify_state = ECM_NOTIFY_NONE; |
10223 | break; |
10224 | default: |
10225 | DBG(cdev, "event %02x --> %d\n", |
10226 | event->bNotificationType, req->status); |
10227 | + atomic_dec(&ecm->notify_count); |
10228 | break; |
10229 | } |
10230 | - ecm->notify_req = req; |
10231 | ecm_do_notify(ecm); |
10232 | } |
10233 | |
10234 | @@ -909,6 +912,11 @@ static void ecm_unbind(struct usb_configuration *c, struct usb_function *f) |
10235 | |
10236 | usb_free_all_descriptors(f); |
10237 | |
10238 | + if (atomic_read(&ecm->notify_count)) { |
10239 | + usb_ep_dequeue(ecm->notify, ecm->notify_req); |
10240 | + atomic_set(&ecm->notify_count, 0); |
10241 | + } |
10242 | + |
10243 | kfree(ecm->notify_req->buf); |
10244 | usb_ep_free_request(ecm->notify, ecm->notify_req); |
10245 | } |
10246 | diff --git a/drivers/usb/gadget/function/f_ncm.c b/drivers/usb/gadget/function/f_ncm.c |
10247 | index 639603722709..6399923239e7 100644 |
10248 | --- a/drivers/usb/gadget/function/f_ncm.c |
10249 | +++ b/drivers/usb/gadget/function/f_ncm.c |
10250 | @@ -57,6 +57,7 @@ struct f_ncm { |
10251 | struct usb_ep *notify; |
10252 | struct usb_request *notify_req; |
10253 | u8 notify_state; |
10254 | + atomic_t notify_count; |
10255 | bool is_open; |
10256 | |
10257 | const struct ndp_parser_opts *parser_opts; |
10258 | @@ -552,7 +553,7 @@ static void ncm_do_notify(struct f_ncm *ncm) |
10259 | int status; |
10260 | |
10261 | /* notification already in flight? */ |
10262 | - if (!req) |
10263 | + if (atomic_read(&ncm->notify_count)) |
10264 | return; |
10265 | |
10266 | event = req->buf; |
10267 | @@ -592,7 +593,8 @@ static void ncm_do_notify(struct f_ncm *ncm) |
10268 | event->bmRequestType = 0xA1; |
10269 | event->wIndex = cpu_to_le16(ncm->ctrl_id); |
10270 | |
10271 | - ncm->notify_req = NULL; |
10272 | + atomic_inc(&ncm->notify_count); |
10273 | + |
10274 | /* |
10275 | * In double buffering if there is a space in FIFO, |
10276 | * completion callback can be called right after the call, |
10277 | @@ -602,7 +604,7 @@ static void ncm_do_notify(struct f_ncm *ncm) |
10278 | status = usb_ep_queue(ncm->notify, req, GFP_ATOMIC); |
10279 | spin_lock(&ncm->lock); |
10280 | if (status < 0) { |
10281 | - ncm->notify_req = req; |
10282 | + atomic_dec(&ncm->notify_count); |
10283 | DBG(cdev, "notify --> %d\n", status); |
10284 | } |
10285 | } |
10286 | @@ -637,17 +639,19 @@ static void ncm_notify_complete(struct usb_ep *ep, struct usb_request *req) |
10287 | case 0: |
10288 | VDBG(cdev, "Notification %02x sent\n", |
10289 | event->bNotificationType); |
10290 | + atomic_dec(&ncm->notify_count); |
10291 | break; |
10292 | case -ECONNRESET: |
10293 | case -ESHUTDOWN: |
10294 | + atomic_set(&ncm->notify_count, 0); |
10295 | ncm->notify_state = NCM_NOTIFY_NONE; |
10296 | break; |
10297 | default: |
10298 | DBG(cdev, "event %02x --> %d\n", |
10299 | event->bNotificationType, req->status); |
10300 | + atomic_dec(&ncm->notify_count); |
10301 | break; |
10302 | } |
10303 | - ncm->notify_req = req; |
10304 | ncm_do_notify(ncm); |
10305 | spin_unlock(&ncm->lock); |
10306 | } |
10307 | @@ -1639,6 +1643,11 @@ static void ncm_unbind(struct usb_configuration *c, struct usb_function *f) |
10308 | ncm_string_defs[0].id = 0; |
10309 | usb_free_all_descriptors(f); |
10310 | |
10311 | + if (atomic_read(&ncm->notify_count)) { |
10312 | + usb_ep_dequeue(ncm->notify, ncm->notify_req); |
10313 | + atomic_set(&ncm->notify_count, 0); |
10314 | + } |
10315 | + |
10316 | kfree(ncm->notify_req->buf); |
10317 | usb_ep_free_request(ncm->notify, ncm->notify_req); |
10318 | } |
10319 | diff --git a/drivers/usb/gadget/legacy/cdc2.c b/drivers/usb/gadget/legacy/cdc2.c |
10320 | index 51c08682de84..5ee25beb52f0 100644 |
10321 | --- a/drivers/usb/gadget/legacy/cdc2.c |
10322 | +++ b/drivers/usb/gadget/legacy/cdc2.c |
10323 | @@ -229,7 +229,7 @@ static struct usb_composite_driver cdc_driver = { |
10324 | .name = "g_cdc", |
10325 | .dev = &device_desc, |
10326 | .strings = dev_strings, |
10327 | - .max_speed = USB_SPEED_HIGH, |
10328 | + .max_speed = USB_SPEED_SUPER, |
10329 | .bind = cdc_bind, |
10330 | .unbind = cdc_unbind, |
10331 | }; |
10332 | diff --git a/drivers/usb/gadget/legacy/g_ffs.c b/drivers/usb/gadget/legacy/g_ffs.c |
10333 | index 6da7316f8e87..54ee4e31645b 100644 |
10334 | --- a/drivers/usb/gadget/legacy/g_ffs.c |
10335 | +++ b/drivers/usb/gadget/legacy/g_ffs.c |
10336 | @@ -153,7 +153,7 @@ static struct usb_composite_driver gfs_driver = { |
10337 | .name = DRIVER_NAME, |
10338 | .dev = &gfs_dev_desc, |
10339 | .strings = gfs_dev_strings, |
10340 | - .max_speed = USB_SPEED_HIGH, |
10341 | + .max_speed = USB_SPEED_SUPER, |
10342 | .bind = gfs_bind, |
10343 | .unbind = gfs_unbind, |
10344 | }; |
10345 | diff --git a/drivers/usb/gadget/legacy/multi.c b/drivers/usb/gadget/legacy/multi.c |
10346 | index a70a406580ea..3b7fc5c7e9c3 100644 |
10347 | --- a/drivers/usb/gadget/legacy/multi.c |
10348 | +++ b/drivers/usb/gadget/legacy/multi.c |
10349 | @@ -486,7 +486,7 @@ static struct usb_composite_driver multi_driver = { |
10350 | .name = "g_multi", |
10351 | .dev = &device_desc, |
10352 | .strings = dev_strings, |
10353 | - .max_speed = USB_SPEED_HIGH, |
10354 | + .max_speed = USB_SPEED_SUPER, |
10355 | .bind = multi_bind, |
10356 | .unbind = multi_unbind, |
10357 | .needs_serial = 1, |
10358 | diff --git a/drivers/usb/gadget/legacy/ncm.c b/drivers/usb/gadget/legacy/ncm.c |
10359 | index 0aba68253e3d..2fb4a847dd52 100644 |
10360 | --- a/drivers/usb/gadget/legacy/ncm.c |
10361 | +++ b/drivers/usb/gadget/legacy/ncm.c |
10362 | @@ -203,7 +203,7 @@ static struct usb_composite_driver ncm_driver = { |
10363 | .name = "g_ncm", |
10364 | .dev = &device_desc, |
10365 | .strings = dev_strings, |
10366 | - .max_speed = USB_SPEED_HIGH, |
10367 | + .max_speed = USB_SPEED_SUPER, |
10368 | .bind = gncm_bind, |
10369 | .unbind = gncm_unbind, |
10370 | }; |
10371 | diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c |
10372 | index 305deb6e59c3..b5ebb43b1824 100644 |
10373 | --- a/fs/btrfs/ctree.c |
10374 | +++ b/fs/btrfs/ctree.c |
10375 | @@ -331,26 +331,6 @@ struct tree_mod_elem { |
10376 | struct tree_mod_root old_root; |
10377 | }; |
10378 | |
10379 | -static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
10380 | -{ |
10381 | - read_lock(&fs_info->tree_mod_log_lock); |
10382 | -} |
10383 | - |
10384 | -static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
10385 | -{ |
10386 | - read_unlock(&fs_info->tree_mod_log_lock); |
10387 | -} |
10388 | - |
10389 | -static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) |
10390 | -{ |
10391 | - write_lock(&fs_info->tree_mod_log_lock); |
10392 | -} |
10393 | - |
10394 | -static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) |
10395 | -{ |
10396 | - write_unlock(&fs_info->tree_mod_log_lock); |
10397 | -} |
10398 | - |
10399 | /* |
10400 | * Pull a new tree mod seq number for our operation. |
10401 | */ |
10402 | @@ -370,14 +350,12 @@ static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
10403 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, |
10404 | struct seq_list *elem) |
10405 | { |
10406 | - tree_mod_log_write_lock(fs_info); |
10407 | - spin_lock(&fs_info->tree_mod_seq_lock); |
10408 | + write_lock(&fs_info->tree_mod_log_lock); |
10409 | if (!elem->seq) { |
10410 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
10411 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
10412 | } |
10413 | - spin_unlock(&fs_info->tree_mod_seq_lock); |
10414 | - tree_mod_log_write_unlock(fs_info); |
10415 | + write_unlock(&fs_info->tree_mod_log_lock); |
10416 | |
10417 | return elem->seq; |
10418 | } |
10419 | @@ -396,7 +374,7 @@ void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, |
10420 | if (!seq_putting) |
10421 | return; |
10422 | |
10423 | - spin_lock(&fs_info->tree_mod_seq_lock); |
10424 | + write_lock(&fs_info->tree_mod_log_lock); |
10425 | list_del(&elem->list); |
10426 | elem->seq = 0; |
10427 | |
10428 | @@ -407,19 +385,17 @@ void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, |
10429 | * blocker with lower sequence number exists, we |
10430 | * cannot remove anything from the log |
10431 | */ |
10432 | - spin_unlock(&fs_info->tree_mod_seq_lock); |
10433 | + write_unlock(&fs_info->tree_mod_log_lock); |
10434 | return; |
10435 | } |
10436 | min_seq = cur_elem->seq; |
10437 | } |
10438 | } |
10439 | - spin_unlock(&fs_info->tree_mod_seq_lock); |
10440 | |
10441 | /* |
10442 | * anything that's lower than the lowest existing (read: blocked) |
10443 | * sequence number can be removed from the tree. |
10444 | */ |
10445 | - tree_mod_log_write_lock(fs_info); |
10446 | tm_root = &fs_info->tree_mod_log; |
10447 | for (node = rb_first(tm_root); node; node = next) { |
10448 | next = rb_next(node); |
10449 | @@ -429,7 +405,7 @@ void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, |
10450 | rb_erase(node, tm_root); |
10451 | kfree(tm); |
10452 | } |
10453 | - tree_mod_log_write_unlock(fs_info); |
10454 | + write_unlock(&fs_info->tree_mod_log_lock); |
10455 | } |
10456 | |
10457 | /* |
10458 | @@ -440,7 +416,7 @@ void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, |
10459 | * for root replace operations, or the logical address of the affected |
10460 | * block for all other operations. |
10461 | * |
10462 | - * Note: must be called with write lock (tree_mod_log_write_lock). |
10463 | + * Note: must be called with write lock for fs_info::tree_mod_log_lock. |
10464 | */ |
10465 | static noinline int |
10466 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) |
10467 | @@ -480,7 +456,7 @@ __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) |
10468 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it |
10469 | * returns zero with the tree_mod_log_lock acquired. The caller must hold |
10470 | * this until all tree mod log insertions are recorded in the rb tree and then |
10471 | - * call tree_mod_log_write_unlock() to release. |
10472 | + * write unlock fs_info::tree_mod_log_lock. |
10473 | */ |
10474 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
10475 | struct extent_buffer *eb) { |
10476 | @@ -490,9 +466,9 @@ static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
10477 | if (eb && btrfs_header_level(eb) == 0) |
10478 | return 1; |
10479 | |
10480 | - tree_mod_log_write_lock(fs_info); |
10481 | + write_lock(&fs_info->tree_mod_log_lock); |
10482 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
10483 | - tree_mod_log_write_unlock(fs_info); |
10484 | + write_unlock(&fs_info->tree_mod_log_lock); |
10485 | return 1; |
10486 | } |
10487 | |
10488 | @@ -556,7 +532,7 @@ tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
10489 | } |
10490 | |
10491 | ret = __tree_mod_log_insert(fs_info, tm); |
10492 | - tree_mod_log_write_unlock(fs_info); |
10493 | + write_unlock(&eb->fs_info->tree_mod_log_lock); |
10494 | if (ret) |
10495 | kfree(tm); |
10496 | |
10497 | @@ -620,7 +596,7 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, |
10498 | ret = __tree_mod_log_insert(fs_info, tm); |
10499 | if (ret) |
10500 | goto free_tms; |
10501 | - tree_mod_log_write_unlock(fs_info); |
10502 | + write_unlock(&eb->fs_info->tree_mod_log_lock); |
10503 | kfree(tm_list); |
10504 | |
10505 | return 0; |
10506 | @@ -631,7 +607,7 @@ free_tms: |
10507 | kfree(tm_list[i]); |
10508 | } |
10509 | if (locked) |
10510 | - tree_mod_log_write_unlock(fs_info); |
10511 | + write_unlock(&eb->fs_info->tree_mod_log_lock); |
10512 | kfree(tm_list); |
10513 | kfree(tm); |
10514 | |
10515 | @@ -712,7 +688,7 @@ tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, |
10516 | if (!ret) |
10517 | ret = __tree_mod_log_insert(fs_info, tm); |
10518 | |
10519 | - tree_mod_log_write_unlock(fs_info); |
10520 | + write_unlock(&fs_info->tree_mod_log_lock); |
10521 | if (ret) |
10522 | goto free_tms; |
10523 | kfree(tm_list); |
10524 | @@ -739,7 +715,7 @@ __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, |
10525 | struct tree_mod_elem *cur = NULL; |
10526 | struct tree_mod_elem *found = NULL; |
10527 | |
10528 | - tree_mod_log_read_lock(fs_info); |
10529 | + read_lock(&fs_info->tree_mod_log_lock); |
10530 | tm_root = &fs_info->tree_mod_log; |
10531 | node = tm_root->rb_node; |
10532 | while (node) { |
10533 | @@ -767,7 +743,7 @@ __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, |
10534 | break; |
10535 | } |
10536 | } |
10537 | - tree_mod_log_read_unlock(fs_info); |
10538 | + read_unlock(&fs_info->tree_mod_log_lock); |
10539 | |
10540 | return found; |
10541 | } |
10542 | @@ -848,7 +824,7 @@ tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
10543 | goto free_tms; |
10544 | } |
10545 | |
10546 | - tree_mod_log_write_unlock(fs_info); |
10547 | + write_unlock(&fs_info->tree_mod_log_lock); |
10548 | kfree(tm_list); |
10549 | |
10550 | return 0; |
10551 | @@ -860,7 +836,7 @@ free_tms: |
10552 | kfree(tm_list[i]); |
10553 | } |
10554 | if (locked) |
10555 | - tree_mod_log_write_unlock(fs_info); |
10556 | + write_unlock(&fs_info->tree_mod_log_lock); |
10557 | kfree(tm_list); |
10558 | |
10559 | return ret; |
10560 | @@ -920,7 +896,7 @@ tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
10561 | goto free_tms; |
10562 | |
10563 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); |
10564 | - tree_mod_log_write_unlock(fs_info); |
10565 | + write_unlock(&eb->fs_info->tree_mod_log_lock); |
10566 | if (ret) |
10567 | goto free_tms; |
10568 | kfree(tm_list); |
10569 | @@ -1271,7 +1247,7 @@ __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
10570 | unsigned long p_size = sizeof(struct btrfs_key_ptr); |
10571 | |
10572 | n = btrfs_header_nritems(eb); |
10573 | - tree_mod_log_read_lock(fs_info); |
10574 | + read_lock(&fs_info->tree_mod_log_lock); |
10575 | while (tm && tm->seq >= time_seq) { |
10576 | /* |
10577 | * all the operations are recorded with the operator used for |
10578 | @@ -1326,7 +1302,7 @@ __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
10579 | if (tm->logical != first_tm->logical) |
10580 | break; |
10581 | } |
10582 | - tree_mod_log_read_unlock(fs_info); |
10583 | + read_unlock(&fs_info->tree_mod_log_lock); |
10584 | btrfs_set_header_nritems(eb, n); |
10585 | } |
10586 | |
10587 | diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h |
10588 | index a423c36bcd72..2bc37d03d407 100644 |
10589 | --- a/fs/btrfs/ctree.h |
10590 | +++ b/fs/btrfs/ctree.h |
10591 | @@ -851,14 +851,12 @@ struct btrfs_fs_info { |
10592 | struct list_head delayed_iputs; |
10593 | struct mutex cleaner_delayed_iput_mutex; |
10594 | |
10595 | - /* this protects tree_mod_seq_list */ |
10596 | - spinlock_t tree_mod_seq_lock; |
10597 | atomic64_t tree_mod_seq; |
10598 | - struct list_head tree_mod_seq_list; |
10599 | |
10600 | - /* this protects tree_mod_log */ |
10601 | + /* this protects tree_mod_log and tree_mod_seq_list */ |
10602 | rwlock_t tree_mod_log_lock; |
10603 | struct rb_root tree_mod_log; |
10604 | + struct list_head tree_mod_seq_list; |
10605 | |
10606 | atomic_t nr_async_submits; |
10607 | atomic_t async_submit_draining; |
10608 | diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c |
10609 | index 74c17db75201..c1ca4ce11e69 100644 |
10610 | --- a/fs/btrfs/delayed-ref.c |
10611 | +++ b/fs/btrfs/delayed-ref.c |
10612 | @@ -279,7 +279,7 @@ void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, |
10613 | if (head->is_data) |
10614 | return; |
10615 | |
10616 | - spin_lock(&fs_info->tree_mod_seq_lock); |
10617 | + read_lock(&fs_info->tree_mod_log_lock); |
10618 | if (!list_empty(&fs_info->tree_mod_seq_list)) { |
10619 | struct seq_list *elem; |
10620 | |
10621 | @@ -287,7 +287,7 @@ void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, |
10622 | struct seq_list, list); |
10623 | seq = elem->seq; |
10624 | } |
10625 | - spin_unlock(&fs_info->tree_mod_seq_lock); |
10626 | + read_unlock(&fs_info->tree_mod_log_lock); |
10627 | |
10628 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, |
10629 | list); |
10630 | @@ -315,7 +315,7 @@ int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, |
10631 | struct seq_list *elem; |
10632 | int ret = 0; |
10633 | |
10634 | - spin_lock(&fs_info->tree_mod_seq_lock); |
10635 | + read_lock(&fs_info->tree_mod_log_lock); |
10636 | if (!list_empty(&fs_info->tree_mod_seq_list)) { |
10637 | elem = list_first_entry(&fs_info->tree_mod_seq_list, |
10638 | struct seq_list, list); |
10639 | @@ -329,7 +329,7 @@ int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, |
10640 | } |
10641 | } |
10642 | |
10643 | - spin_unlock(&fs_info->tree_mod_seq_lock); |
10644 | + read_unlock(&fs_info->tree_mod_log_lock); |
10645 | return ret; |
10646 | } |
10647 | |
10648 | diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c |
10649 | index b37519241eb1..e3524ecce3d7 100644 |
10650 | --- a/fs/btrfs/disk-io.c |
10651 | +++ b/fs/btrfs/disk-io.c |
10652 | @@ -2104,7 +2104,7 @@ static void free_root_extent_buffers(struct btrfs_root *root) |
10653 | } |
10654 | |
10655 | /* helper to cleanup tree roots */ |
10656 | -static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) |
10657 | +static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) |
10658 | { |
10659 | free_root_extent_buffers(info->tree_root); |
10660 | |
10661 | @@ -2113,7 +2113,7 @@ static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) |
10662 | free_root_extent_buffers(info->csum_root); |
10663 | free_root_extent_buffers(info->quota_root); |
10664 | free_root_extent_buffers(info->uuid_root); |
10665 | - if (chunk_root) |
10666 | + if (free_chunk_root) |
10667 | free_root_extent_buffers(info->chunk_root); |
10668 | free_root_extent_buffers(info->free_space_root); |
10669 | } |
10670 | @@ -2519,7 +2519,6 @@ int open_ctree(struct super_block *sb, |
10671 | spin_lock_init(&fs_info->delayed_iput_lock); |
10672 | spin_lock_init(&fs_info->defrag_inodes_lock); |
10673 | spin_lock_init(&fs_info->free_chunk_lock); |
10674 | - spin_lock_init(&fs_info->tree_mod_seq_lock); |
10675 | spin_lock_init(&fs_info->super_lock); |
10676 | spin_lock_init(&fs_info->qgroup_op_lock); |
10677 | spin_lock_init(&fs_info->buffer_lock); |
10678 | @@ -3136,7 +3135,7 @@ fail_block_groups: |
10679 | btrfs_free_block_groups(fs_info); |
10680 | |
10681 | fail_tree_roots: |
10682 | - free_root_pointers(fs_info, 1); |
10683 | + free_root_pointers(fs_info, true); |
10684 | invalidate_inode_pages2(fs_info->btree_inode->i_mapping); |
10685 | |
10686 | fail_sb_buffer: |
10687 | @@ -3165,7 +3164,7 @@ recovery_tree_root: |
10688 | if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT)) |
10689 | goto fail_tree_roots; |
10690 | |
10691 | - free_root_pointers(fs_info, 0); |
10692 | + free_root_pointers(fs_info, false); |
10693 | |
10694 | /* don't use the log in recovery mode, it won't be valid */ |
10695 | btrfs_set_super_log_root(disk_super, 0); |
10696 | @@ -3862,7 +3861,7 @@ void close_ctree(struct btrfs_root *root) |
10697 | btrfs_stop_all_workers(fs_info); |
10698 | |
10699 | clear_bit(BTRFS_FS_OPEN, &fs_info->flags); |
10700 | - free_root_pointers(fs_info, 1); |
10701 | + free_root_pointers(fs_info, true); |
10702 | |
10703 | iput(fs_info->btree_inode); |
10704 | |
10705 | diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c |
10706 | index 37a28e2369b9..1372d3e5d90b 100644 |
10707 | --- a/fs/btrfs/extent_io.c |
10708 | +++ b/fs/btrfs/extent_io.c |
10709 | @@ -4060,6 +4060,14 @@ retry: |
10710 | */ |
10711 | scanned = 1; |
10712 | index = 0; |
10713 | + |
10714 | + /* |
10715 | + * If we're looping we could run into a page that is locked by a |
10716 | + * writer and that writer could be waiting on writeback for a |
10717 | + * page in our current bio, and thus deadlock, so flush the |
10718 | + * write bio here. |
10719 | + */ |
10720 | + flush_write_bio(data); |
10721 | goto retry; |
10722 | } |
10723 | |
10724 | diff --git a/fs/btrfs/tests/btrfs-tests.c b/fs/btrfs/tests/btrfs-tests.c |
10725 | index bf62ad919a95..9edc2674b8a7 100644 |
10726 | --- a/fs/btrfs/tests/btrfs-tests.c |
10727 | +++ b/fs/btrfs/tests/btrfs-tests.c |
10728 | @@ -112,7 +112,6 @@ struct btrfs_fs_info *btrfs_alloc_dummy_fs_info(void) |
10729 | spin_lock_init(&fs_info->qgroup_op_lock); |
10730 | spin_lock_init(&fs_info->super_lock); |
10731 | spin_lock_init(&fs_info->fs_roots_radix_lock); |
10732 | - spin_lock_init(&fs_info->tree_mod_seq_lock); |
10733 | mutex_init(&fs_info->qgroup_ioctl_lock); |
10734 | mutex_init(&fs_info->qgroup_rescan_lock); |
10735 | rwlock_init(&fs_info->tree_mod_log_lock); |
10736 | diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c |
10737 | index fd6c74662e9a..31df020634cd 100644 |
10738 | --- a/fs/btrfs/transaction.c |
10739 | +++ b/fs/btrfs/transaction.c |
10740 | @@ -1917,6 +1917,14 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
10741 | struct btrfs_transaction *prev_trans = NULL; |
10742 | int ret; |
10743 | |
10744 | + /* |
10745 | + * Some places just start a transaction to commit it. We need to make |
10746 | + * sure that if this commit fails that the abort code actually marks the |
10747 | + * transaction as failed, so set trans->dirty to make the abort code do |
10748 | + * the right thing. |
10749 | + */ |
10750 | + trans->dirty = true; |
10751 | + |
10752 | /* Stop the commit early if ->aborted is set */ |
10753 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { |
10754 | ret = cur_trans->aborted; |
10755 | diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c |
10756 | index 7ee573cddde9..f79682937faf 100644 |
10757 | --- a/fs/btrfs/tree-log.c |
10758 | +++ b/fs/btrfs/tree-log.c |
10759 | @@ -4443,13 +4443,8 @@ static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, |
10760 | struct btrfs_file_extent_item); |
10761 | |
10762 | if (btrfs_file_extent_type(leaf, extent) == |
10763 | - BTRFS_FILE_EXTENT_INLINE) { |
10764 | - len = btrfs_file_extent_inline_len(leaf, |
10765 | - path->slots[0], |
10766 | - extent); |
10767 | - ASSERT(len == i_size); |
10768 | + BTRFS_FILE_EXTENT_INLINE) |
10769 | return 0; |
10770 | - } |
10771 | |
10772 | len = btrfs_file_extent_num_bytes(leaf, extent); |
10773 | /* Last extent goes beyond i_size, no need to log a hole. */ |
10774 | diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c |
10775 | index 5255deac86b2..e8dc28dbe563 100644 |
10776 | --- a/fs/cifs/smb2pdu.c |
10777 | +++ b/fs/cifs/smb2pdu.c |
10778 | @@ -247,9 +247,14 @@ smb2_reconnect(__le16 smb2_command, struct cifs_tcon *tcon) |
10779 | */ |
10780 | mutex_lock(&tcon->ses->session_mutex); |
10781 | rc = cifs_negotiate_protocol(0, tcon->ses); |
10782 | - if (!rc && tcon->ses->need_reconnect) |
10783 | + if (!rc && tcon->ses->need_reconnect) { |
10784 | rc = cifs_setup_session(0, tcon->ses, nls_codepage); |
10785 | - |
10786 | + if ((rc == -EACCES) && !tcon->retry) { |
10787 | + rc = -EHOSTDOWN; |
10788 | + mutex_unlock(&tcon->ses->session_mutex); |
10789 | + goto failed; |
10790 | + } |
10791 | + } |
10792 | if (rc || !tcon->need_reconnect) { |
10793 | mutex_unlock(&tcon->ses->session_mutex); |
10794 | goto out; |
10795 | @@ -291,6 +296,7 @@ out: |
10796 | case SMB2_SET_INFO: |
10797 | rc = -EAGAIN; |
10798 | } |
10799 | +failed: |
10800 | unload_nls(nls_codepage); |
10801 | return rc; |
10802 | } |
10803 | diff --git a/fs/ext2/super.c b/fs/ext2/super.c |
10804 | index 6fcb29b393d3..186912c9bf56 100644 |
10805 | --- a/fs/ext2/super.c |
10806 | +++ b/fs/ext2/super.c |
10807 | @@ -1047,9 +1047,9 @@ static int ext2_fill_super(struct super_block *sb, void *data, int silent) |
10808 | |
10809 | if (EXT2_BLOCKS_PER_GROUP(sb) == 0) |
10810 | goto cantfind_ext2; |
10811 | - sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) - |
10812 | - le32_to_cpu(es->s_first_data_block) - 1) |
10813 | - / EXT2_BLOCKS_PER_GROUP(sb)) + 1; |
10814 | + sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) - |
10815 | + le32_to_cpu(es->s_first_data_block) - 1) |
10816 | + / EXT2_BLOCKS_PER_GROUP(sb)) + 1; |
10817 | db_count = (sbi->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) / |
10818 | EXT2_DESC_PER_BLOCK(sb); |
10819 | sbi->s_group_desc = kmalloc (db_count * sizeof (struct buffer_head *), GFP_KERNEL); |
10820 | diff --git a/fs/ext4/page-io.c b/fs/ext4/page-io.c |
10821 | index 0094923e5ebf..94f60f9d57fd 100644 |
10822 | --- a/fs/ext4/page-io.c |
10823 | +++ b/fs/ext4/page-io.c |
10824 | @@ -469,16 +469,25 @@ int ext4_bio_write_page(struct ext4_io_submit *io, |
10825 | nr_to_submit) { |
10826 | gfp_t gfp_flags = GFP_NOFS; |
10827 | |
10828 | + /* |
10829 | + * Since bounce page allocation uses a mempool, we can only use |
10830 | + * a waiting mask (i.e. request guaranteed allocation) on the |
10831 | + * first page of the bio. Otherwise it can deadlock. |
10832 | + */ |
10833 | + if (io->io_bio) |
10834 | + gfp_flags = GFP_NOWAIT | __GFP_NOWARN; |
10835 | retry_encrypt: |
10836 | data_page = fscrypt_encrypt_page(inode, page, gfp_flags); |
10837 | if (IS_ERR(data_page)) { |
10838 | ret = PTR_ERR(data_page); |
10839 | - if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) { |
10840 | - if (io->io_bio) { |
10841 | + if (ret == -ENOMEM && |
10842 | + (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) { |
10843 | + gfp_flags = GFP_NOFS; |
10844 | + if (io->io_bio) |
10845 | ext4_io_submit(io); |
10846 | - congestion_wait(BLK_RW_ASYNC, HZ/50); |
10847 | - } |
10848 | - gfp_flags |= __GFP_NOFAIL; |
10849 | + else |
10850 | + gfp_flags |= __GFP_NOFAIL; |
10851 | + congestion_wait(BLK_RW_ASYNC, HZ/50); |
10852 | goto retry_encrypt; |
10853 | } |
10854 | data_page = NULL; |
10855 | diff --git a/fs/nfs/Kconfig b/fs/nfs/Kconfig |
10856 | index b1daeafbea92..c3428767332c 100644 |
10857 | --- a/fs/nfs/Kconfig |
10858 | +++ b/fs/nfs/Kconfig |
10859 | @@ -89,7 +89,7 @@ config NFS_V4 |
10860 | config NFS_SWAP |
10861 | bool "Provide swap over NFS support" |
10862 | default n |
10863 | - depends on NFS_FS |
10864 | + depends on NFS_FS && SWAP |
10865 | select SUNRPC_SWAP |
10866 | help |
10867 | This option enables swapon to work on files located on NFS mounts. |
10868 | diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c |
10869 | index 9d7537446260..0d4a56c77a1a 100644 |
10870 | --- a/fs/nfs/callback_proc.c |
10871 | +++ b/fs/nfs/callback_proc.c |
10872 | @@ -419,7 +419,7 @@ static bool referring_call_exists(struct nfs_client *clp, |
10873 | uint32_t nrclists, |
10874 | struct referring_call_list *rclists) |
10875 | { |
10876 | - bool status = 0; |
10877 | + bool status = false; |
10878 | int i, j; |
10879 | struct nfs4_session *session; |
10880 | struct nfs4_slot_table *tbl; |
10881 | diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c |
10882 | index 1e5321d1ed22..c2665d920cf8 100644 |
10883 | --- a/fs/nfs/dir.c |
10884 | +++ b/fs/nfs/dir.c |
10885 | @@ -57,7 +57,7 @@ static void nfs_readdir_clear_array(struct page*); |
10886 | const struct file_operations nfs_dir_operations = { |
10887 | .llseek = nfs_llseek_dir, |
10888 | .read = generic_read_dir, |
10889 | - .iterate_shared = nfs_readdir, |
10890 | + .iterate = nfs_readdir, |
10891 | .open = nfs_opendir, |
10892 | .release = nfs_closedir, |
10893 | .fsync = nfs_fsync_dir, |
10894 | @@ -145,7 +145,6 @@ struct nfs_cache_array_entry { |
10895 | }; |
10896 | |
10897 | struct nfs_cache_array { |
10898 | - atomic_t refcount; |
10899 | int size; |
10900 | int eof_index; |
10901 | u64 last_cookie; |
10902 | @@ -170,6 +169,17 @@ typedef struct { |
10903 | unsigned int eof:1; |
10904 | } nfs_readdir_descriptor_t; |
10905 | |
10906 | +static |
10907 | +void nfs_readdir_init_array(struct page *page) |
10908 | +{ |
10909 | + struct nfs_cache_array *array; |
10910 | + |
10911 | + array = kmap_atomic(page); |
10912 | + memset(array, 0, sizeof(struct nfs_cache_array)); |
10913 | + array->eof_index = -1; |
10914 | + kunmap_atomic(array); |
10915 | +} |
10916 | + |
10917 | /* |
10918 | * The caller is responsible for calling nfs_readdir_release_array(page) |
10919 | */ |
10920 | @@ -201,20 +211,12 @@ void nfs_readdir_clear_array(struct page *page) |
10921 | int i; |
10922 | |
10923 | array = kmap_atomic(page); |
10924 | - if (atomic_dec_and_test(&array->refcount)) |
10925 | - for (i = 0; i < array->size; i++) |
10926 | - kfree(array->array[i].string.name); |
10927 | + for (i = 0; i < array->size; i++) |
10928 | + kfree(array->array[i].string.name); |
10929 | + array->size = 0; |
10930 | kunmap_atomic(array); |
10931 | } |
10932 | |
10933 | -static bool grab_page(struct page *page) |
10934 | -{ |
10935 | - struct nfs_cache_array *array = kmap_atomic(page); |
10936 | - bool res = atomic_inc_not_zero(&array->refcount); |
10937 | - kunmap_atomic(array); |
10938 | - return res; |
10939 | -} |
10940 | - |
10941 | /* |
10942 | * the caller is responsible for freeing qstr.name |
10943 | * when called by nfs_readdir_add_to_array, the strings will be freed in |
10944 | @@ -287,7 +289,7 @@ int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descri |
10945 | desc->cache_entry_index = index; |
10946 | return 0; |
10947 | out_eof: |
10948 | - desc->eof = 1; |
10949 | + desc->eof = true; |
10950 | return -EBADCOOKIE; |
10951 | } |
10952 | |
10953 | @@ -341,7 +343,7 @@ int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_des |
10954 | if (array->eof_index >= 0) { |
10955 | status = -EBADCOOKIE; |
10956 | if (*desc->dir_cookie == array->last_cookie) |
10957 | - desc->eof = 1; |
10958 | + desc->eof = true; |
10959 | } |
10960 | out: |
10961 | return status; |
10962 | @@ -653,6 +655,8 @@ int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, |
10963 | int status = -ENOMEM; |
10964 | unsigned int array_size = ARRAY_SIZE(pages); |
10965 | |
10966 | + nfs_readdir_init_array(page); |
10967 | + |
10968 | entry.prev_cookie = 0; |
10969 | entry.cookie = desc->last_cookie; |
10970 | entry.eof = 0; |
10971 | @@ -673,9 +677,8 @@ int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, |
10972 | status = PTR_ERR(array); |
10973 | goto out_label_free; |
10974 | } |
10975 | - memset(array, 0, sizeof(struct nfs_cache_array)); |
10976 | - atomic_set(&array->refcount, 1); |
10977 | - array->eof_index = -1; |
10978 | + |
10979 | + array = kmap(page); |
10980 | |
10981 | status = nfs_readdir_alloc_pages(pages, array_size); |
10982 | if (status < 0) |
10983 | @@ -730,6 +733,7 @@ int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) |
10984 | unlock_page(page); |
10985 | return 0; |
10986 | error: |
10987 | + nfs_readdir_clear_array(page); |
10988 | unlock_page(page); |
10989 | return ret; |
10990 | } |
10991 | @@ -737,7 +741,6 @@ int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) |
10992 | static |
10993 | void cache_page_release(nfs_readdir_descriptor_t *desc) |
10994 | { |
10995 | - nfs_readdir_clear_array(desc->page); |
10996 | put_page(desc->page); |
10997 | desc->page = NULL; |
10998 | } |
10999 | @@ -745,33 +748,34 @@ void cache_page_release(nfs_readdir_descriptor_t *desc) |
11000 | static |
11001 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) |
11002 | { |
11003 | - struct page *page; |
11004 | - |
11005 | - for (;;) { |
11006 | - page = read_cache_page(desc->file->f_mapping, |
11007 | + return read_cache_page(desc->file->f_mapping, |
11008 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
11009 | - if (IS_ERR(page) || grab_page(page)) |
11010 | - break; |
11011 | - put_page(page); |
11012 | - } |
11013 | - return page; |
11014 | } |
11015 | |
11016 | /* |
11017 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
11018 | + * and locks the page to prevent removal from the page cache. |
11019 | */ |
11020 | static |
11021 | -int find_cache_page(nfs_readdir_descriptor_t *desc) |
11022 | +int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc) |
11023 | { |
11024 | int res; |
11025 | |
11026 | desc->page = get_cache_page(desc); |
11027 | if (IS_ERR(desc->page)) |
11028 | return PTR_ERR(desc->page); |
11029 | - |
11030 | - res = nfs_readdir_search_array(desc); |
11031 | + res = lock_page_killable(desc->page); |
11032 | if (res != 0) |
11033 | - cache_page_release(desc); |
11034 | + goto error; |
11035 | + res = -EAGAIN; |
11036 | + if (desc->page->mapping != NULL) { |
11037 | + res = nfs_readdir_search_array(desc); |
11038 | + if (res == 0) |
11039 | + return 0; |
11040 | + } |
11041 | + unlock_page(desc->page); |
11042 | +error: |
11043 | + cache_page_release(desc); |
11044 | return res; |
11045 | } |
11046 | |
11047 | @@ -786,7 +790,7 @@ int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) |
11048 | desc->last_cookie = 0; |
11049 | } |
11050 | do { |
11051 | - res = find_cache_page(desc); |
11052 | + res = find_and_lock_cache_page(desc); |
11053 | } while (res == -EAGAIN); |
11054 | return res; |
11055 | } |
11056 | @@ -815,7 +819,7 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc) |
11057 | ent = &array->array[i]; |
11058 | if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, |
11059 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { |
11060 | - desc->eof = 1; |
11061 | + desc->eof = true; |
11062 | break; |
11063 | } |
11064 | desc->ctx->pos++; |
11065 | @@ -827,11 +831,10 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc) |
11066 | ctx->duped = 1; |
11067 | } |
11068 | if (array->eof_index >= 0) |
11069 | - desc->eof = 1; |
11070 | + desc->eof = true; |
11071 | |
11072 | nfs_readdir_release_array(desc->page); |
11073 | out: |
11074 | - cache_page_release(desc); |
11075 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
11076 | (unsigned long long)*desc->dir_cookie, res); |
11077 | return res; |
11078 | @@ -877,13 +880,13 @@ int uncached_readdir(nfs_readdir_descriptor_t *desc) |
11079 | |
11080 | status = nfs_do_filldir(desc); |
11081 | |
11082 | + out_release: |
11083 | + nfs_readdir_clear_array(desc->page); |
11084 | + cache_page_release(desc); |
11085 | out: |
11086 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
11087 | __func__, status); |
11088 | return status; |
11089 | - out_release: |
11090 | - cache_page_release(desc); |
11091 | - goto out; |
11092 | } |
11093 | |
11094 | /* The file offset position represents the dirent entry number. A |
11095 | @@ -928,7 +931,7 @@ static int nfs_readdir(struct file *file, struct dir_context *ctx) |
11096 | if (res == -EBADCOOKIE) { |
11097 | res = 0; |
11098 | /* This means either end of directory */ |
11099 | - if (*desc->dir_cookie && desc->eof == 0) { |
11100 | + if (*desc->dir_cookie && !desc->eof) { |
11101 | /* Or that the server has 'lost' a cookie */ |
11102 | res = uncached_readdir(desc); |
11103 | if (res == 0) |
11104 | @@ -948,6 +951,8 @@ static int nfs_readdir(struct file *file, struct dir_context *ctx) |
11105 | break; |
11106 | |
11107 | res = nfs_do_filldir(desc); |
11108 | + unlock_page(desc->page); |
11109 | + cache_page_release(desc); |
11110 | if (res < 0) |
11111 | break; |
11112 | } while (!desc->eof); |
11113 | @@ -960,11 +965,13 @@ out: |
11114 | |
11115 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
11116 | { |
11117 | + struct inode *inode = file_inode(filp); |
11118 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
11119 | |
11120 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", |
11121 | filp, offset, whence); |
11122 | |
11123 | + inode_lock(inode); |
11124 | switch (whence) { |
11125 | case 1: |
11126 | offset += filp->f_pos; |
11127 | @@ -972,13 +979,16 @@ static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
11128 | if (offset >= 0) |
11129 | break; |
11130 | default: |
11131 | - return -EINVAL; |
11132 | + offset = -EINVAL; |
11133 | + goto out; |
11134 | } |
11135 | if (offset != filp->f_pos) { |
11136 | filp->f_pos = offset; |
11137 | dir_ctx->dir_cookie = 0; |
11138 | dir_ctx->duped = 0; |
11139 | } |
11140 | +out: |
11141 | + inode_unlock(inode); |
11142 | return offset; |
11143 | } |
11144 | |
11145 | diff --git a/fs/nfs/nfs4client.c b/fs/nfs/nfs4client.c |
11146 | index 1ec6dd4f3e2e..3ee60c533217 100644 |
11147 | --- a/fs/nfs/nfs4client.c |
11148 | +++ b/fs/nfs/nfs4client.c |
11149 | @@ -847,7 +847,7 @@ nfs4_find_client_sessionid(struct net *net, const struct sockaddr *addr, |
11150 | |
11151 | spin_lock(&nn->nfs_client_lock); |
11152 | list_for_each_entry(clp, &nn->nfs_client_list, cl_share_link) { |
11153 | - if (nfs4_cb_match_client(addr, clp, minorversion) == false) |
11154 | + if (!nfs4_cb_match_client(addr, clp, minorversion)) |
11155 | continue; |
11156 | |
11157 | if (!nfs4_has_session(clp)) |
11158 | diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c |
11159 | index ca4249ae644f..632d3c3f8dfb 100644 |
11160 | --- a/fs/nfs/nfs4proc.c |
11161 | +++ b/fs/nfs/nfs4proc.c |
11162 | @@ -2916,6 +2916,11 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, |
11163 | exception.retry = 1; |
11164 | continue; |
11165 | } |
11166 | + if (status == -NFS4ERR_EXPIRED) { |
11167 | + nfs4_schedule_lease_recovery(server->nfs_client); |
11168 | + exception.retry = 1; |
11169 | + continue; |
11170 | + } |
11171 | if (status == -EAGAIN) { |
11172 | /* We must have found a delegation */ |
11173 | exception.retry = 1; |
11174 | diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c |
11175 | index 0e008db16b16..c3abf92adfb7 100644 |
11176 | --- a/fs/nfs/pnfs.c |
11177 | +++ b/fs/nfs/pnfs.c |
11178 | @@ -1436,7 +1436,7 @@ pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
11179 | if ((range->iomode == IOMODE_RW && |
11180 | ls_range->iomode != IOMODE_RW) || |
11181 | (range->iomode != ls_range->iomode && |
11182 | - strict_iomode == true) || |
11183 | + strict_iomode) || |
11184 | !pnfs_lseg_range_intersecting(ls_range, range)) |
11185 | return 0; |
11186 | |
11187 | diff --git a/fs/nfsd/nfs4layouts.c b/fs/nfsd/nfs4layouts.c |
11188 | index 64813697f4c4..f6cc2fddb78b 100644 |
11189 | --- a/fs/nfsd/nfs4layouts.c |
11190 | +++ b/fs/nfsd/nfs4layouts.c |
11191 | @@ -680,7 +680,7 @@ nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task) |
11192 | |
11193 | /* Client gets 2 lease periods to return it */ |
11194 | cutoff = ktime_add_ns(task->tk_start, |
11195 | - nn->nfsd4_lease * NSEC_PER_SEC * 2); |
11196 | + (u64)nn->nfsd4_lease * NSEC_PER_SEC * 2); |
11197 | |
11198 | if (ktime_before(now, cutoff)) { |
11199 | rpc_delay(task, HZ/100); /* 10 mili-seconds */ |
11200 | diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c |
11201 | index db4bd70b62d0..4509c76716e3 100644 |
11202 | --- a/fs/nfsd/nfs4state.c |
11203 | +++ b/fs/nfsd/nfs4state.c |
11204 | @@ -6034,7 +6034,7 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, |
11205 | } |
11206 | |
11207 | if (fl_flags & FL_SLEEP) { |
11208 | - nbl->nbl_time = jiffies; |
11209 | + nbl->nbl_time = get_seconds(); |
11210 | spin_lock(&nn->blocked_locks_lock); |
11211 | list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked); |
11212 | list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru); |
11213 | diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h |
11214 | index 133d8bf62a5c..7872b1ead885 100644 |
11215 | --- a/fs/nfsd/state.h |
11216 | +++ b/fs/nfsd/state.h |
11217 | @@ -591,7 +591,7 @@ static inline bool nfsd4_stateid_generation_after(stateid_t *a, stateid_t *b) |
11218 | struct nfsd4_blocked_lock { |
11219 | struct list_head nbl_list; |
11220 | struct list_head nbl_lru; |
11221 | - unsigned long nbl_time; |
11222 | + time_t nbl_time; |
11223 | struct file_lock nbl_lock; |
11224 | struct knfsd_fh nbl_fh; |
11225 | struct nfsd4_callback nbl_cb; |
11226 | diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c |
11227 | index b4fbeefba246..5ef0d1d60743 100644 |
11228 | --- a/fs/ubifs/file.c |
11229 | +++ b/fs/ubifs/file.c |
11230 | @@ -721,6 +721,7 @@ static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu, |
11231 | int err, page_idx, page_cnt, ret = 0, n = 0; |
11232 | int allocate = bu->buf ? 0 : 1; |
11233 | loff_t isize; |
11234 | + gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS; |
11235 | |
11236 | err = ubifs_tnc_get_bu_keys(c, bu); |
11237 | if (err) |
11238 | @@ -782,8 +783,9 @@ static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu, |
11239 | |
11240 | if (page_offset > end_index) |
11241 | break; |
11242 | - page = find_or_create_page(mapping, page_offset, |
11243 | - GFP_NOFS | __GFP_COLD); |
11244 | + page = pagecache_get_page(mapping, page_offset, |
11245 | + FGP_LOCK|FGP_ACCESSED|FGP_CREAT|FGP_NOWAIT, |
11246 | + ra_gfp_mask); |
11247 | if (!page) |
11248 | break; |
11249 | if (!PageUptodate(page)) |
11250 | diff --git a/include/media/v4l2-rect.h b/include/media/v4l2-rect.h |
11251 | index d2125f0cc7cd..1584c760b993 100644 |
11252 | --- a/include/media/v4l2-rect.h |
11253 | +++ b/include/media/v4l2-rect.h |
11254 | @@ -75,10 +75,10 @@ static inline void v4l2_rect_map_inside(struct v4l2_rect *r, |
11255 | r->left = boundary->left; |
11256 | if (r->top < boundary->top) |
11257 | r->top = boundary->top; |
11258 | - if (r->left + r->width > boundary->width) |
11259 | - r->left = boundary->width - r->width; |
11260 | - if (r->top + r->height > boundary->height) |
11261 | - r->top = boundary->height - r->height; |
11262 | + if (r->left + r->width > boundary->left + boundary->width) |
11263 | + r->left = boundary->left + boundary->width - r->width; |
11264 | + if (r->top + r->height > boundary->top + boundary->height) |
11265 | + r->top = boundary->top + boundary->height - r->height; |
11266 | } |
11267 | |
11268 | /** |
11269 | diff --git a/kernel/events/core.c b/kernel/events/core.c |
11270 | index 64ace5e9af2a..97b90faceb97 100644 |
11271 | --- a/kernel/events/core.c |
11272 | +++ b/kernel/events/core.c |
11273 | @@ -5303,7 +5303,15 @@ accounting: |
11274 | */ |
11275 | user_lock_limit *= num_online_cpus(); |
11276 | |
11277 | - user_locked = atomic_long_read(&user->locked_vm) + user_extra; |
11278 | + user_locked = atomic_long_read(&user->locked_vm); |
11279 | + |
11280 | + /* |
11281 | + * sysctl_perf_event_mlock may have changed, so that |
11282 | + * user->locked_vm > user_lock_limit |
11283 | + */ |
11284 | + if (user_locked > user_lock_limit) |
11285 | + user_locked = user_lock_limit; |
11286 | + user_locked += user_extra; |
11287 | |
11288 | if (user_locked > user_lock_limit) |
11289 | extra = user_locked - user_lock_limit; |
11290 | diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c |
11291 | index 7e4fad75acaa..2924ff544c9e 100644 |
11292 | --- a/kernel/time/clocksource.c |
11293 | +++ b/kernel/time/clocksource.c |
11294 | @@ -272,8 +272,15 @@ static void clocksource_watchdog(unsigned long data) |
11295 | next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); |
11296 | if (next_cpu >= nr_cpu_ids) |
11297 | next_cpu = cpumask_first(cpu_online_mask); |
11298 | - watchdog_timer.expires += WATCHDOG_INTERVAL; |
11299 | - add_timer_on(&watchdog_timer, next_cpu); |
11300 | + |
11301 | + /* |
11302 | + * Arm timer if not already pending: could race with concurrent |
11303 | + * pair clocksource_stop_watchdog() clocksource_start_watchdog(). |
11304 | + */ |
11305 | + if (!timer_pending(&watchdog_timer)) { |
11306 | + watchdog_timer.expires += WATCHDOG_INTERVAL; |
11307 | + add_timer_on(&watchdog_timer, next_cpu); |
11308 | + } |
11309 | out: |
11310 | spin_unlock(&watchdog_lock); |
11311 | } |
11312 | diff --git a/lib/test_kasan.c b/lib/test_kasan.c |
11313 | index 4ba4cbe169a8..6e76a448867d 100644 |
11314 | --- a/lib/test_kasan.c |
11315 | +++ b/lib/test_kasan.c |
11316 | @@ -124,6 +124,7 @@ static noinline void __init kmalloc_oob_krealloc_more(void) |
11317 | if (!ptr1 || !ptr2) { |
11318 | pr_err("Allocation failed\n"); |
11319 | kfree(ptr1); |
11320 | + kfree(ptr2); |
11321 | return; |
11322 | } |
11323 | |
11324 | diff --git a/net/hsr/hsr_slave.c b/net/hsr/hsr_slave.c |
11325 | index f5b60388d02f..4ff6e02d8b73 100644 |
11326 | --- a/net/hsr/hsr_slave.c |
11327 | +++ b/net/hsr/hsr_slave.c |
11328 | @@ -31,6 +31,8 @@ static rx_handler_result_t hsr_handle_frame(struct sk_buff **pskb) |
11329 | |
11330 | rcu_read_lock(); /* hsr->node_db, hsr->ports */ |
11331 | port = hsr_port_get_rcu(skb->dev); |
11332 | + if (!port) |
11333 | + goto finish_pass; |
11334 | |
11335 | if (hsr_addr_is_self(port->hsr, eth_hdr(skb)->h_source)) { |
11336 | /* Directly kill frames sent by ourselves */ |
11337 | diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c |
11338 | index 6e25524c6a74..02918e0d635e 100644 |
11339 | --- a/net/ipv4/tcp.c |
11340 | +++ b/net/ipv4/tcp.c |
11341 | @@ -2298,9 +2298,11 @@ int tcp_disconnect(struct sock *sk, int flags) |
11342 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
11343 | tp->snd_cwnd_cnt = 0; |
11344 | tp->window_clamp = 0; |
11345 | + tp->delivered = 0; |
11346 | tcp_set_ca_state(sk, TCP_CA_Open); |
11347 | tp->is_sack_reneg = 0; |
11348 | tcp_clear_retrans(tp); |
11349 | + tp->total_retrans = 0; |
11350 | inet_csk_delack_init(sk); |
11351 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
11352 | * issue in __tcp_select_window() |
11353 | @@ -2312,8 +2314,12 @@ int tcp_disconnect(struct sock *sk, int flags) |
11354 | dst_release(sk->sk_rx_dst); |
11355 | sk->sk_rx_dst = NULL; |
11356 | tcp_saved_syn_free(tp); |
11357 | + tp->segs_in = 0; |
11358 | + tp->segs_out = 0; |
11359 | tp->bytes_acked = 0; |
11360 | tp->bytes_received = 0; |
11361 | + tp->data_segs_in = 0; |
11362 | + tp->data_segs_out = 0; |
11363 | |
11364 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
11365 | |
11366 | diff --git a/net/rxrpc/ar-internal.h b/net/rxrpc/ar-internal.h |
11367 | index f60e35576526..d6a64771463f 100644 |
11368 | --- a/net/rxrpc/ar-internal.h |
11369 | +++ b/net/rxrpc/ar-internal.h |
11370 | @@ -401,6 +401,7 @@ enum rxrpc_call_flag { |
11371 | RXRPC_CALL_SEND_PING, /* A ping will need to be sent */ |
11372 | RXRPC_CALL_PINGING, /* Ping in process */ |
11373 | RXRPC_CALL_RETRANS_TIMEOUT, /* Retransmission due to timeout occurred */ |
11374 | + RXRPC_CALL_DISCONNECTED, /* The call has been disconnected */ |
11375 | }; |
11376 | |
11377 | /* |
11378 | diff --git a/net/rxrpc/call_object.c b/net/rxrpc/call_object.c |
11379 | index 1ed18d8c9c9f..88bcd146142f 100644 |
11380 | --- a/net/rxrpc/call_object.c |
11381 | +++ b/net/rxrpc/call_object.c |
11382 | @@ -464,7 +464,7 @@ void rxrpc_release_call(struct rxrpc_sock *rx, struct rxrpc_call *call) |
11383 | |
11384 | _debug("RELEASE CALL %p (%d CONN %p)", call, call->debug_id, conn); |
11385 | |
11386 | - if (conn) |
11387 | + if (conn && !test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) |
11388 | rxrpc_disconnect_call(call); |
11389 | |
11390 | for (i = 0; i < RXRPC_RXTX_BUFF_SIZE; i++) { |
11391 | @@ -539,6 +539,7 @@ static void rxrpc_rcu_destroy_call(struct rcu_head *rcu) |
11392 | { |
11393 | struct rxrpc_call *call = container_of(rcu, struct rxrpc_call, rcu); |
11394 | |
11395 | + rxrpc_put_connection(call->conn); |
11396 | rxrpc_put_peer(call->peer); |
11397 | kfree(call->rxtx_buffer); |
11398 | kfree(call->rxtx_annotations); |
11399 | @@ -560,7 +561,6 @@ void rxrpc_cleanup_call(struct rxrpc_call *call) |
11400 | |
11401 | ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE); |
11402 | ASSERT(test_bit(RXRPC_CALL_RELEASED, &call->flags)); |
11403 | - ASSERTCMP(call->conn, ==, NULL); |
11404 | |
11405 | /* Clean up the Rx/Tx buffer */ |
11406 | for (i = 0; i < RXRPC_RXTX_BUFF_SIZE; i++) |
11407 | diff --git a/net/rxrpc/conn_client.c b/net/rxrpc/conn_client.c |
11408 | index 0fce919bf47d..dd41733f182a 100644 |
11409 | --- a/net/rxrpc/conn_client.c |
11410 | +++ b/net/rxrpc/conn_client.c |
11411 | @@ -736,9 +736,9 @@ void rxrpc_disconnect_client_call(struct rxrpc_call *call) |
11412 | struct rxrpc_channel *chan = &conn->channels[channel]; |
11413 | |
11414 | trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect); |
11415 | - call->conn = NULL; |
11416 | |
11417 | spin_lock(&conn->channel_lock); |
11418 | + set_bit(RXRPC_CALL_DISCONNECTED, &call->flags); |
11419 | |
11420 | /* Calls that have never actually been assigned a channel can simply be |
11421 | * discarded. If the conn didn't get used either, it will follow |
11422 | @@ -828,7 +828,6 @@ out: |
11423 | spin_unlock(&rxrpc_client_conn_cache_lock); |
11424 | out_2: |
11425 | spin_unlock(&conn->channel_lock); |
11426 | - rxrpc_put_connection(conn); |
11427 | _leave(""); |
11428 | return; |
11429 | |
11430 | diff --git a/net/rxrpc/conn_object.c b/net/rxrpc/conn_object.c |
11431 | index e1e83af47866..e7c89b978587 100644 |
11432 | --- a/net/rxrpc/conn_object.c |
11433 | +++ b/net/rxrpc/conn_object.c |
11434 | @@ -211,9 +211,8 @@ void rxrpc_disconnect_call(struct rxrpc_call *call) |
11435 | __rxrpc_disconnect_call(conn, call); |
11436 | spin_unlock(&conn->channel_lock); |
11437 | |
11438 | - call->conn = NULL; |
11439 | + set_bit(RXRPC_CALL_DISCONNECTED, &call->flags); |
11440 | conn->idle_timestamp = jiffies; |
11441 | - rxrpc_put_connection(conn); |
11442 | } |
11443 | |
11444 | /* |
11445 | diff --git a/net/rxrpc/input.c b/net/rxrpc/input.c |
11446 | index a4380e182e6c..f0ccc6a04c7a 100644 |
11447 | --- a/net/rxrpc/input.c |
11448 | +++ b/net/rxrpc/input.c |
11449 | @@ -582,8 +582,7 @@ ack: |
11450 | immediate_ack, true, |
11451 | rxrpc_propose_ack_input_data); |
11452 | |
11453 | - if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1) |
11454 | - rxrpc_notify_socket(call); |
11455 | + rxrpc_notify_socket(call); |
11456 | _leave(" [queued]"); |
11457 | } |
11458 | |
11459 | diff --git a/net/rxrpc/output.c b/net/rxrpc/output.c |
11460 | index 64389f493bb2..d568c96f5262 100644 |
11461 | --- a/net/rxrpc/output.c |
11462 | +++ b/net/rxrpc/output.c |
11463 | @@ -96,7 +96,7 @@ static size_t rxrpc_fill_out_ack(struct rxrpc_call *call, |
11464 | */ |
11465 | int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping) |
11466 | { |
11467 | - struct rxrpc_connection *conn = NULL; |
11468 | + struct rxrpc_connection *conn; |
11469 | struct rxrpc_ack_buffer *pkt; |
11470 | struct msghdr msg; |
11471 | struct kvec iov[2]; |
11472 | @@ -106,18 +106,14 @@ int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping) |
11473 | int ret; |
11474 | u8 reason; |
11475 | |
11476 | - spin_lock_bh(&call->lock); |
11477 | - if (call->conn) |
11478 | - conn = rxrpc_get_connection_maybe(call->conn); |
11479 | - spin_unlock_bh(&call->lock); |
11480 | - if (!conn) |
11481 | + if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) |
11482 | return -ECONNRESET; |
11483 | |
11484 | pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); |
11485 | - if (!pkt) { |
11486 | - rxrpc_put_connection(conn); |
11487 | + if (!pkt) |
11488 | return -ENOMEM; |
11489 | - } |
11490 | + |
11491 | + conn = call->conn; |
11492 | |
11493 | msg.msg_name = &call->peer->srx.transport; |
11494 | msg.msg_namelen = call->peer->srx.transport_len; |
11495 | @@ -204,7 +200,6 @@ int rxrpc_send_ack_packet(struct rxrpc_call *call, bool ping) |
11496 | } |
11497 | |
11498 | out: |
11499 | - rxrpc_put_connection(conn); |
11500 | kfree(pkt); |
11501 | return ret; |
11502 | } |
11503 | @@ -214,20 +209,18 @@ out: |
11504 | */ |
11505 | int rxrpc_send_abort_packet(struct rxrpc_call *call) |
11506 | { |
11507 | - struct rxrpc_connection *conn = NULL; |
11508 | + struct rxrpc_connection *conn; |
11509 | struct rxrpc_abort_buffer pkt; |
11510 | struct msghdr msg; |
11511 | struct kvec iov[1]; |
11512 | rxrpc_serial_t serial; |
11513 | int ret; |
11514 | |
11515 | - spin_lock_bh(&call->lock); |
11516 | - if (call->conn) |
11517 | - conn = rxrpc_get_connection_maybe(call->conn); |
11518 | - spin_unlock_bh(&call->lock); |
11519 | - if (!conn) |
11520 | + if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) |
11521 | return -ECONNRESET; |
11522 | |
11523 | + conn = call->conn; |
11524 | + |
11525 | msg.msg_name = &call->peer->srx.transport; |
11526 | msg.msg_namelen = call->peer->srx.transport_len; |
11527 | msg.msg_control = NULL; |
11528 | @@ -255,7 +248,6 @@ int rxrpc_send_abort_packet(struct rxrpc_call *call) |
11529 | ret = kernel_sendmsg(conn->params.local->socket, |
11530 | &msg, iov, 1, sizeof(pkt)); |
11531 | |
11532 | - rxrpc_put_connection(conn); |
11533 | return ret; |
11534 | } |
11535 | |
11536 | diff --git a/net/sched/cls_rsvp.h b/net/sched/cls_rsvp.h |
11537 | index 322438fb3ffc..8673f9817f91 100644 |
11538 | --- a/net/sched/cls_rsvp.h |
11539 | +++ b/net/sched/cls_rsvp.h |
11540 | @@ -455,10 +455,8 @@ static u32 gen_tunnel(struct rsvp_head *data) |
11541 | |
11542 | static const struct nla_policy rsvp_policy[TCA_RSVP_MAX + 1] = { |
11543 | [TCA_RSVP_CLASSID] = { .type = NLA_U32 }, |
11544 | - [TCA_RSVP_DST] = { .type = NLA_BINARY, |
11545 | - .len = RSVP_DST_LEN * sizeof(u32) }, |
11546 | - [TCA_RSVP_SRC] = { .type = NLA_BINARY, |
11547 | - .len = RSVP_DST_LEN * sizeof(u32) }, |
11548 | + [TCA_RSVP_DST] = { .len = RSVP_DST_LEN * sizeof(u32) }, |
11549 | + [TCA_RSVP_SRC] = { .len = RSVP_DST_LEN * sizeof(u32) }, |
11550 | [TCA_RSVP_PINFO] = { .len = sizeof(struct tc_rsvp_pinfo) }, |
11551 | }; |
11552 | |
11553 | diff --git a/net/sched/cls_tcindex.c b/net/sched/cls_tcindex.c |
11554 | index db80a6440f37..3e1695b66e31 100644 |
11555 | --- a/net/sched/cls_tcindex.c |
11556 | +++ b/net/sched/cls_tcindex.c |
11557 | @@ -301,12 +301,31 @@ tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base, |
11558 | cp->fall_through = p->fall_through; |
11559 | cp->tp = tp; |
11560 | |
11561 | + if (tb[TCA_TCINDEX_HASH]) |
11562 | + cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]); |
11563 | + |
11564 | + if (tb[TCA_TCINDEX_MASK]) |
11565 | + cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]); |
11566 | + |
11567 | + if (tb[TCA_TCINDEX_SHIFT]) |
11568 | + cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]); |
11569 | + |
11570 | + if (!cp->hash) { |
11571 | + /* Hash not specified, use perfect hash if the upper limit |
11572 | + * of the hashing index is below the threshold. |
11573 | + */ |
11574 | + if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD) |
11575 | + cp->hash = (cp->mask >> cp->shift) + 1; |
11576 | + else |
11577 | + cp->hash = DEFAULT_HASH_SIZE; |
11578 | + } |
11579 | + |
11580 | if (p->perfect) { |
11581 | int i; |
11582 | |
11583 | if (tcindex_alloc_perfect_hash(cp) < 0) |
11584 | goto errout; |
11585 | - for (i = 0; i < cp->hash; i++) |
11586 | + for (i = 0; i < min(cp->hash, p->hash); i++) |
11587 | cp->perfect[i].res = p->perfect[i].res; |
11588 | balloc = 1; |
11589 | } |
11590 | @@ -321,15 +340,6 @@ tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base, |
11591 | if (old_r) |
11592 | cr.res = r->res; |
11593 | |
11594 | - if (tb[TCA_TCINDEX_HASH]) |
11595 | - cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]); |
11596 | - |
11597 | - if (tb[TCA_TCINDEX_MASK]) |
11598 | - cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]); |
11599 | - |
11600 | - if (tb[TCA_TCINDEX_SHIFT]) |
11601 | - cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]); |
11602 | - |
11603 | err = -EBUSY; |
11604 | |
11605 | /* Hash already allocated, make sure that we still meet the |
11606 | @@ -347,16 +357,6 @@ tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base, |
11607 | if (tb[TCA_TCINDEX_FALL_THROUGH]) |
11608 | cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]); |
11609 | |
11610 | - if (!cp->hash) { |
11611 | - /* Hash not specified, use perfect hash if the upper limit |
11612 | - * of the hashing index is below the threshold. |
11613 | - */ |
11614 | - if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD) |
11615 | - cp->hash = (cp->mask >> cp->shift) + 1; |
11616 | - else |
11617 | - cp->hash = DEFAULT_HASH_SIZE; |
11618 | - } |
11619 | - |
11620 | if (!cp->perfect && !cp->h) |
11621 | cp->alloc_hash = cp->hash; |
11622 | |
11623 | diff --git a/net/sunrpc/auth_gss/svcauth_gss.c b/net/sunrpc/auth_gss/svcauth_gss.c |
11624 | index b4b68c6e3f8b..d7775ca2fbb9 100644 |
11625 | --- a/net/sunrpc/auth_gss/svcauth_gss.c |
11626 | +++ b/net/sunrpc/auth_gss/svcauth_gss.c |
11627 | @@ -1180,6 +1180,7 @@ static int gss_proxy_save_rsc(struct cache_detail *cd, |
11628 | dprintk("RPC: No creds found!\n"); |
11629 | goto out; |
11630 | } else { |
11631 | + struct timespec64 boot; |
11632 | |
11633 | /* steal creds */ |
11634 | rsci.cred = ud->creds; |
11635 | @@ -1200,6 +1201,9 @@ static int gss_proxy_save_rsc(struct cache_detail *cd, |
11636 | &expiry, GFP_KERNEL); |
11637 | if (status) |
11638 | goto out; |
11639 | + |
11640 | + getboottime64(&boot); |
11641 | + expiry -= boot.tv_sec; |
11642 | } |
11643 | |
11644 | rsci.h.expiry_time = expiry; |
11645 | diff --git a/sound/drivers/dummy.c b/sound/drivers/dummy.c |
11646 | index 172dacd925f5..c182341c1714 100644 |
11647 | --- a/sound/drivers/dummy.c |
11648 | +++ b/sound/drivers/dummy.c |
11649 | @@ -925,7 +925,7 @@ static void print_formats(struct snd_dummy *dummy, |
11650 | { |
11651 | int i; |
11652 | |
11653 | - for (i = 0; i < SNDRV_PCM_FORMAT_LAST; i++) { |
11654 | + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) { |
11655 | if (dummy->pcm_hw.formats & (1ULL << i)) |
11656 | snd_iprintf(buffer, " %s", snd_pcm_format_name(i)); |
11657 | } |
11658 | diff --git a/sound/soc/qcom/apq8016_sbc.c b/sound/soc/qcom/apq8016_sbc.c |
11659 | index 754742018515..3d91cef3704a 100644 |
11660 | --- a/sound/soc/qcom/apq8016_sbc.c |
11661 | +++ b/sound/soc/qcom/apq8016_sbc.c |
11662 | @@ -128,7 +128,8 @@ static struct apq8016_sbc_data *apq8016_sbc_parse_of(struct snd_soc_card *card) |
11663 | link->codec_of_node = of_parse_phandle(codec, "sound-dai", 0); |
11664 | if (!link->codec_of_node) { |
11665 | dev_err(card->dev, "error getting codec phandle\n"); |
11666 | - return ERR_PTR(-EINVAL); |
11667 | + ret = -EINVAL; |
11668 | + goto error; |
11669 | } |
11670 | |
11671 | ret = snd_soc_of_get_dai_name(cpu, &link->cpu_dai_name); |
11672 | diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c |
11673 | index 635b22fa1101..280bb5cab87f 100644 |
11674 | --- a/sound/soc/soc-pcm.c |
11675 | +++ b/sound/soc/soc-pcm.c |
11676 | @@ -2137,42 +2137,81 @@ int dpcm_be_dai_trigger(struct snd_soc_pcm_runtime *fe, int stream, |
11677 | } |
11678 | EXPORT_SYMBOL_GPL(dpcm_be_dai_trigger); |
11679 | |
11680 | +static int dpcm_dai_trigger_fe_be(struct snd_pcm_substream *substream, |
11681 | + int cmd, bool fe_first) |
11682 | +{ |
11683 | + struct snd_soc_pcm_runtime *fe = substream->private_data; |
11684 | + int ret; |
11685 | + |
11686 | + /* call trigger on the frontend before the backend. */ |
11687 | + if (fe_first) { |
11688 | + dev_dbg(fe->dev, "ASoC: pre trigger FE %s cmd %d\n", |
11689 | + fe->dai_link->name, cmd); |
11690 | + |
11691 | + ret = soc_pcm_trigger(substream, cmd); |
11692 | + if (ret < 0) |
11693 | + return ret; |
11694 | + |
11695 | + ret = dpcm_be_dai_trigger(fe, substream->stream, cmd); |
11696 | + return ret; |
11697 | + } |
11698 | + |
11699 | + /* call trigger on the frontend after the backend. */ |
11700 | + ret = dpcm_be_dai_trigger(fe, substream->stream, cmd); |
11701 | + if (ret < 0) |
11702 | + return ret; |
11703 | + |
11704 | + dev_dbg(fe->dev, "ASoC: post trigger FE %s cmd %d\n", |
11705 | + fe->dai_link->name, cmd); |
11706 | + |
11707 | + ret = soc_pcm_trigger(substream, cmd); |
11708 | + |
11709 | + return ret; |
11710 | +} |
11711 | + |
11712 | static int dpcm_fe_dai_do_trigger(struct snd_pcm_substream *substream, int cmd) |
11713 | { |
11714 | struct snd_soc_pcm_runtime *fe = substream->private_data; |
11715 | - int stream = substream->stream, ret; |
11716 | + int stream = substream->stream; |
11717 | + int ret = 0; |
11718 | enum snd_soc_dpcm_trigger trigger = fe->dai_link->trigger[stream]; |
11719 | |
11720 | fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; |
11721 | |
11722 | switch (trigger) { |
11723 | case SND_SOC_DPCM_TRIGGER_PRE: |
11724 | - /* call trigger on the frontend before the backend. */ |
11725 | - |
11726 | - dev_dbg(fe->dev, "ASoC: pre trigger FE %s cmd %d\n", |
11727 | - fe->dai_link->name, cmd); |
11728 | - |
11729 | - ret = soc_pcm_trigger(substream, cmd); |
11730 | - if (ret < 0) { |
11731 | - dev_err(fe->dev,"ASoC: trigger FE failed %d\n", ret); |
11732 | - goto out; |
11733 | + switch (cmd) { |
11734 | + case SNDRV_PCM_TRIGGER_START: |
11735 | + case SNDRV_PCM_TRIGGER_RESUME: |
11736 | + case SNDRV_PCM_TRIGGER_PAUSE_RELEASE: |
11737 | + ret = dpcm_dai_trigger_fe_be(substream, cmd, true); |
11738 | + break; |
11739 | + case SNDRV_PCM_TRIGGER_STOP: |
11740 | + case SNDRV_PCM_TRIGGER_SUSPEND: |
11741 | + case SNDRV_PCM_TRIGGER_PAUSE_PUSH: |
11742 | + ret = dpcm_dai_trigger_fe_be(substream, cmd, false); |
11743 | + break; |
11744 | + default: |
11745 | + ret = -EINVAL; |
11746 | + break; |
11747 | } |
11748 | - |
11749 | - ret = dpcm_be_dai_trigger(fe, substream->stream, cmd); |
11750 | break; |
11751 | case SND_SOC_DPCM_TRIGGER_POST: |
11752 | - /* call trigger on the frontend after the backend. */ |
11753 | - |
11754 | - ret = dpcm_be_dai_trigger(fe, substream->stream, cmd); |
11755 | - if (ret < 0) { |
11756 | - dev_err(fe->dev,"ASoC: trigger FE failed %d\n", ret); |
11757 | - goto out; |
11758 | + switch (cmd) { |
11759 | + case SNDRV_PCM_TRIGGER_START: |
11760 | + case SNDRV_PCM_TRIGGER_RESUME: |
11761 | + case SNDRV_PCM_TRIGGER_PAUSE_RELEASE: |
11762 | + ret = dpcm_dai_trigger_fe_be(substream, cmd, false); |
11763 | + break; |
11764 | + case SNDRV_PCM_TRIGGER_STOP: |
11765 | + case SNDRV_PCM_TRIGGER_SUSPEND: |
11766 | + case SNDRV_PCM_TRIGGER_PAUSE_PUSH: |
11767 | + ret = dpcm_dai_trigger_fe_be(substream, cmd, true); |
11768 | + break; |
11769 | + default: |
11770 | + ret = -EINVAL; |
11771 | + break; |
11772 | } |
11773 | - |
11774 | - dev_dbg(fe->dev, "ASoC: post trigger FE %s cmd %d\n", |
11775 | - fe->dai_link->name, cmd); |
11776 | - |
11777 | - ret = soc_pcm_trigger(substream, cmd); |
11778 | break; |
11779 | case SND_SOC_DPCM_TRIGGER_BESPOKE: |
11780 | /* bespoke trigger() - handles both FE and BEs */ |
11781 | @@ -2181,10 +2220,6 @@ static int dpcm_fe_dai_do_trigger(struct snd_pcm_substream *substream, int cmd) |
11782 | fe->dai_link->name, cmd); |
11783 | |
11784 | ret = soc_pcm_bespoke_trigger(substream, cmd); |
11785 | - if (ret < 0) { |
11786 | - dev_err(fe->dev,"ASoC: trigger FE failed %d\n", ret); |
11787 | - goto out; |
11788 | - } |
11789 | break; |
11790 | default: |
11791 | dev_err(fe->dev, "ASoC: invalid trigger cmd %d for %s\n", cmd, |
11792 | @@ -2193,6 +2228,12 @@ static int dpcm_fe_dai_do_trigger(struct snd_pcm_substream *substream, int cmd) |
11793 | goto out; |
11794 | } |
11795 | |
11796 | + if (ret < 0) { |
11797 | + dev_err(fe->dev, "ASoC: trigger FE cmd: %d failed: %d\n", |
11798 | + cmd, ret); |
11799 | + goto out; |
11800 | + } |
11801 | + |
11802 | switch (cmd) { |
11803 | case SNDRV_PCM_TRIGGER_START: |
11804 | case SNDRV_PCM_TRIGGER_RESUME: |
11805 | diff --git a/tools/power/acpi/Makefile.config b/tools/power/acpi/Makefile.config |
11806 | index a1883bbb0144..fb5559f9819a 100644 |
11807 | --- a/tools/power/acpi/Makefile.config |
11808 | +++ b/tools/power/acpi/Makefile.config |
11809 | @@ -18,7 +18,7 @@ include $(srctree)/../../scripts/Makefile.include |
11810 | |
11811 | OUTPUT=$(srctree)/ |
11812 | ifeq ("$(origin O)", "command line") |
11813 | - OUTPUT := $(O)/power/acpi/ |
11814 | + OUTPUT := $(O)/tools/power/acpi/ |
11815 | endif |
11816 | #$(info Determined 'OUTPUT' to be $(OUTPUT)) |
11817 |