Contents of /trunk/kernel26-xen/patches-2.6.25-r1/1023-2.6.25-xen-patch-2.6.22.patch
Parent Directory | Revision Log
Revision 613 -
(show annotations)
(download)
Sat May 24 01:13:37 2008 UTC (16 years, 4 months ago) by niro
File size: 211855 byte(s)
Sat May 24 01:13:37 2008 UTC (16 years, 4 months ago) by niro
File size: 211855 byte(s)
-fixed patch
1 | From: www.kernel.org |
2 | Subject: Update to 2.6.22 |
3 | Patch-mainline: 2.6.22 |
4 | |
5 | Automatically created from "patches.kernel.org/patch-2.6.22" by xen-port-patches.py |
6 | |
7 | Acked-by: jbeulich@novell.com |
8 | |
9 | --- |
10 | arch/x86/Kconfig | 5 |
11 | arch/x86/ia32/ia32entry-xen.S | 18 - |
12 | arch/x86/kernel/Makefile | 2 |
13 | arch/x86/kernel/acpi/sleep_64-xen.c | 26 - |
14 | arch/x86/kernel/apic_32-xen.c | 1 |
15 | arch/x86/kernel/apic_64-xen.c | 1 |
16 | arch/x86/kernel/cpu/common-xen.c | 224 ++++--------- |
17 | arch/x86/kernel/cpu/mtrr/main-xen.c | 2 |
18 | arch/x86/kernel/e820_32-xen.c | 46 +- |
19 | arch/x86/kernel/e820_64-xen.c | 28 - |
20 | arch/x86/kernel/early_printk-xen.c | 27 - |
21 | arch/x86/kernel/entry_32-xen.S | 30 - |
22 | arch/x86/kernel/entry_64-xen.S | 7 |
23 | arch/x86/kernel/genapic_64-xen.c | 106 +----- |
24 | arch/x86/kernel/genapic_xen_64.c | 3 |
25 | arch/x86/kernel/head64-xen.c | 32 + |
26 | arch/x86/kernel/head_32-xen.S | 101 ------ |
27 | arch/x86/kernel/head_64-xen.S | 37 -- |
28 | arch/x86/kernel/io_apic_32-xen.c | 43 -- |
29 | arch/x86/kernel/io_apic_64-xen.c | 39 -- |
30 | arch/x86/kernel/ioport_32-xen.c | 2 |
31 | arch/x86/kernel/ioport_64-xen.c | 2 |
32 | arch/x86/kernel/irq_32-xen.c | 3 |
33 | arch/x86/kernel/irq_64-xen.c | 34 +- |
34 | arch/x86/kernel/ldt_32-xen.c | 1 |
35 | arch/x86/kernel/ldt_64-xen.c | 1 |
36 | arch/x86/kernel/microcode-xen.c | 2 |
37 | arch/x86/kernel/mpparse_32-xen.c | 3 |
38 | arch/x86/kernel/mpparse_64-xen.c | 3 |
39 | arch/x86/kernel/pci-dma_32-xen.c | 29 + |
40 | arch/x86/kernel/pci-swiotlb_64-xen.c | 2 |
41 | arch/x86/kernel/process_32-xen.c | 27 + |
42 | arch/x86/kernel/process_64-xen.c | 16 |
43 | arch/x86/kernel/quirks-xen.c | 63 --- |
44 | arch/x86/kernel/setup64-xen.c | 17 - |
45 | arch/x86/kernel/setup_64-xen.c | 30 - |
46 | arch/x86/kernel/smp_32-xen.c | 191 ++++------- |
47 | arch/x86/kernel/smp_64-xen.c | 29 - |
48 | arch/x86/kernel/time_32-xen.c | 62 +-- |
49 | arch/x86/kernel/traps_32-xen.c | 46 +- |
50 | arch/x86/kernel/traps_64-xen.c | 55 +-- |
51 | arch/x86/kernel/vsyscall_64-xen.c | 73 +++- |
52 | arch/x86/mm/fault_32-xen.c | 42 +- |
53 | arch/x86/mm/fault_64-xen.c | 15 |
54 | arch/x86/mm/highmem_32-xen.c | 14 |
55 | arch/x86/mm/init_32-xen.c | 157 ++++++--- |
56 | arch/x86/mm/init_64-xen.c | 132 ++++--- |
57 | arch/x86/mm/ioremap_32-xen.c | 1 |
58 | arch/x86/mm/pageattr_64-xen.c | 27 + |
59 | arch/x86/mm/pgtable_32-xen.c | 210 +++++++----- |
60 | drivers/char/tpm/tpm_xen.c | 2 |
61 | drivers/xen/blkfront/blkfront.c | 2 |
62 | drivers/xen/char/mem.c | 1 |
63 | drivers/xen/core/hypervisor_sysfs.c | 2 |
64 | drivers/xen/core/smpboot.c | 49 +- |
65 | drivers/xen/core/xen_sysfs.c | 20 - |
66 | drivers/xen/netback/netback.c | 14 |
67 | drivers/xen/netfront/netfront.c | 2 |
68 | drivers/xen/pciback/xenbus.c | 2 |
69 | drivers/xen/pcifront/xenbus.c | 4 |
70 | drivers/xen/sfc_netback/accel_fwd.c | 7 |
71 | drivers/xen/sfc_netback/accel_solarflare.c | 2 |
72 | drivers/xen/sfc_netfront/accel_tso.c | 28 - |
73 | drivers/xen/sfc_netfront/accel_vi.c | 4 |
74 | drivers/xen/sfc_netfront/accel_xenbus.c | 4 |
75 | drivers/xen/xenoprof/xenoprofile.c | 2 |
76 | fs/aio.c | 7 |
77 | include/asm-x86/mach-xen/asm/desc_32.h | 119 ++++--- |
78 | include/asm-x86/mach-xen/asm/desc_64.h | 30 - |
79 | include/asm-x86/mach-xen/asm/dma-mapping_64.h | 2 |
80 | include/asm-x86/mach-xen/asm/fixmap_32.h | 9 |
81 | include/asm-x86/mach-xen/asm/fixmap_64.h | 1 |
82 | include/asm-x86/mach-xen/asm/highmem.h | 6 |
83 | include/asm-x86/mach-xen/asm/io_32.h | 13 |
84 | include/asm-x86/mach-xen/asm/irqflags_32.h | 78 ++-- |
85 | include/asm-x86/mach-xen/asm/irqflags_64.h | 19 - |
86 | include/asm-x86/mach-xen/asm/mmu.h | 8 |
87 | include/asm-x86/mach-xen/asm/mmu_64.h | 8 |
88 | include/asm-x86/mach-xen/asm/mmu_context_32.h | 29 + |
89 | include/asm-x86/mach-xen/asm/mmu_context_64.h | 3 |
90 | include/asm-x86/mach-xen/asm/page_64.h | 61 +-- |
91 | include/asm-x86/mach-xen/asm/pgalloc_32.h | 3 |
92 | include/asm-x86/mach-xen/asm/pgalloc_64.h | 15 |
93 | include/asm-x86/mach-xen/asm/pgtable-2level.h | 43 +- |
94 | include/asm-x86/mach-xen/asm/pgtable-3level-defs.h | 2 |
95 | include/asm-x86/mach-xen/asm/pgtable-3level.h | 61 ++- |
96 | include/asm-x86/mach-xen/asm/pgtable_32.h | 80 ++-- |
97 | include/asm-x86/mach-xen/asm/pgtable_64.h | 83 ++--- |
98 | include/asm-x86/mach-xen/asm/processor_32.h | 141 +++----- |
99 | include/asm-x86/mach-xen/asm/processor_64.h | 55 --- |
100 | include/asm-x86/mach-xen/asm/scatterlist_32.h | 2 |
101 | include/asm-x86/mach-xen/asm/segment_32.h | 10 |
102 | include/asm-x86/mach-xen/asm/smp_32.h | 117 +++++-- |
103 | include/asm-x86/mach-xen/asm/smp_64.h | 20 - |
104 | include/asm-x86/mach-xen/asm/system_32.h | 348 ++++----------------- |
105 | include/asm-x86/mach-xen/asm/system_64.h | 106 ------ |
106 | include/asm-x86/mach-xen/asm/tlbflush_32.h | 11 |
107 | include/asm-x86/mach-xen/asm/tlbflush_64.h | 2 |
108 | lib/swiotlb-xen.c | 1 |
109 | net/core/dev.c | 15 |
110 | scripts/Makefile.xen.awk | 2 |
111 | 101 files changed, 1642 insertions(+), 2080 deletions(-) |
112 | |
113 | --- a/arch/x86/Kconfig |
114 | +++ b/arch/x86/Kconfig |
115 | @@ -1222,7 +1222,7 @@ |
116 | |
117 | config RELOCATABLE |
118 | bool "Build a relocatable kernel (EXPERIMENTAL)" |
119 | - depends on EXPERIMENTAL && !X86_XEN |
120 | + depends on EXPERIMENTAL && !X86_XEN && !X86_64_XEN |
121 | help |
122 | This builds a kernel image that retains relocation information |
123 | so it can be loaded someplace besides the default 1MB. |
124 | @@ -1276,7 +1276,6 @@ |
125 | def_bool y |
126 | prompt "Compat VDSO support" |
127 | depends on X86_32 || IA32_EMULATION |
128 | - depends on !X86_XEN |
129 | help |
130 | Map the 32-bit VDSO to the predictable old-style address too. |
131 | ---help--- |
132 | @@ -1453,7 +1452,7 @@ |
133 | bool "PCI support" if !X86_VISWS |
134 | depends on !X86_VOYAGER |
135 | default y |
136 | - select ARCH_SUPPORTS_MSI if (X86_LOCAL_APIC && X86_IO_APIC) |
137 | + select ARCH_SUPPORTS_MSI if (X86_LOCAL_APIC && X86_IO_APIC && !X86_XEN && !X86_64_XEN) |
138 | help |
139 | Find out whether you have a PCI motherboard. PCI is the name of a |
140 | bus system, i.e. the way the CPU talks to the other stuff inside |
141 | --- a/arch/x86/ia32/ia32entry-xen.S |
142 | +++ b/arch/x86/ia32/ia32entry-xen.S |
143 | @@ -431,11 +431,7 @@ |
144 | .quad sys_symlink |
145 | .quad sys_lstat |
146 | .quad sys_readlink /* 85 */ |
147 | -#ifdef CONFIG_IA32_AOUT |
148 | .quad sys_uselib |
149 | -#else |
150 | - .quad quiet_ni_syscall |
151 | -#endif |
152 | .quad sys_swapon |
153 | .quad sys_reboot |
154 | .quad compat_sys_old_readdir |
155 | @@ -574,7 +570,7 @@ |
156 | .quad quiet_ni_syscall /* tux */ |
157 | .quad quiet_ni_syscall /* security */ |
158 | .quad sys_gettid |
159 | - .quad sys_readahead /* 225 */ |
160 | + .quad sys32_readahead /* 225 */ |
161 | .quad sys_setxattr |
162 | .quad sys_lsetxattr |
163 | .quad sys_fsetxattr |
164 | @@ -599,7 +595,7 @@ |
165 | .quad compat_sys_io_getevents |
166 | .quad compat_sys_io_submit |
167 | .quad sys_io_cancel |
168 | - .quad sys_fadvise64 /* 250 */ |
169 | + .quad sys32_fadvise64 /* 250 */ |
170 | .quad quiet_ni_syscall /* free_huge_pages */ |
171 | .quad sys_exit_group |
172 | .quad sys32_lookup_dcookie |
173 | @@ -663,10 +659,14 @@ |
174 | .quad compat_sys_set_robust_list |
175 | .quad compat_sys_get_robust_list |
176 | .quad sys_splice |
177 | - .quad sys_sync_file_range |
178 | - .quad sys_tee |
179 | + .quad sys32_sync_file_range |
180 | + .quad sys_tee /* 315 */ |
181 | .quad compat_sys_vmsplice |
182 | .quad compat_sys_move_pages |
183 | .quad sys_getcpu |
184 | .quad sys_epoll_pwait |
185 | -ia32_syscall_end: |
186 | + .quad compat_sys_utimensat /* 320 */ |
187 | + .quad compat_sys_signalfd |
188 | + .quad compat_sys_timerfd |
189 | + .quad sys_eventfd |
190 | +ia32_syscall_end: |
191 | --- a/arch/x86/kernel/Makefile |
192 | +++ b/arch/x86/kernel/Makefile |
193 | @@ -106,4 +106,4 @@ |
194 | |
195 | disabled-obj-$(CONFIG_XEN) := early-quirks.o hpet.o i8253.o i8259_$(BITS).o reboot.o \ |
196 | smpboot_$(BITS).o tsc_$(BITS).o tsc_sync.o |
197 | -%/head_$(BITS).o %/head_$(BITS).s: $(if $(CONFIG_XEN),EXTRA_AFLAGS,dummy) := |
198 | +%/head_64.o %/head_64.s: $(if $(CONFIG_XEN),EXTRA_AFLAGS,dummy) := |
199 | --- a/arch/x86/kernel/acpi/sleep_64-xen.c |
200 | +++ b/arch/x86/kernel/acpi/sleep_64-xen.c |
201 | @@ -60,19 +60,6 @@ |
202 | extern char wakeup_start, wakeup_end; |
203 | |
204 | extern unsigned long acpi_copy_wakeup_routine(unsigned long); |
205 | - |
206 | -static pgd_t low_ptr; |
207 | - |
208 | -static void init_low_mapping(void) |
209 | -{ |
210 | - pgd_t *slot0 = pgd_offset(current->mm, 0UL); |
211 | - low_ptr = *slot0; |
212 | - /* FIXME: We're playing with the current task's page tables here, which |
213 | - * is potentially dangerous on SMP systems. |
214 | - */ |
215 | - set_pgd(slot0, *pgd_offset(current->mm, PAGE_OFFSET)); |
216 | - local_flush_tlb(); |
217 | -} |
218 | #endif |
219 | |
220 | /** |
221 | @@ -84,8 +71,6 @@ |
222 | int acpi_save_state_mem(void) |
223 | { |
224 | #ifndef CONFIG_ACPI_PV_SLEEP |
225 | - init_low_mapping(); |
226 | - |
227 | memcpy((void *)acpi_wakeup_address, &wakeup_start, |
228 | &wakeup_end - &wakeup_start); |
229 | acpi_copy_wakeup_routine(acpi_wakeup_address); |
230 | @@ -98,10 +83,6 @@ |
231 | */ |
232 | void acpi_restore_state_mem(void) |
233 | { |
234 | -#ifndef CONFIG_ACPI_PV_SLEEP |
235 | - set_pgd(pgd_offset(current->mm, 0UL), low_ptr); |
236 | - local_flush_tlb(); |
237 | -#endif |
238 | } |
239 | |
240 | /** |
241 | @@ -115,10 +96,11 @@ |
242 | void __init acpi_reserve_bootmem(void) |
243 | { |
244 | #ifndef CONFIG_ACPI_PV_SLEEP |
245 | - acpi_wakeup_address = (unsigned long)alloc_bootmem_low(PAGE_SIZE); |
246 | - if ((&wakeup_end - &wakeup_start) > PAGE_SIZE) |
247 | + acpi_wakeup_address = (unsigned long)alloc_bootmem_low(PAGE_SIZE*2); |
248 | + if ((&wakeup_end - &wakeup_start) > (PAGE_SIZE*2)) |
249 | printk(KERN_CRIT |
250 | - "ACPI: Wakeup code way too big, will crash on attempt to suspend\n"); |
251 | + "ACPI: Wakeup code way too big, will crash on attempt" |
252 | + " to suspend\n"); |
253 | #endif |
254 | } |
255 | |
256 | --- a/arch/x86/kernel/apic_32-xen.c |
257 | +++ b/arch/x86/kernel/apic_32-xen.c |
258 | @@ -19,7 +19,6 @@ |
259 | #include <linux/mm.h> |
260 | #include <linux/delay.h> |
261 | #include <linux/bootmem.h> |
262 | -#include <linux/smp_lock.h> |
263 | #include <linux/interrupt.h> |
264 | #include <linux/mc146818rtc.h> |
265 | #include <linux/kernel_stat.h> |
266 | --- a/arch/x86/kernel/apic_64-xen.c |
267 | +++ b/arch/x86/kernel/apic_64-xen.c |
268 | @@ -19,7 +19,6 @@ |
269 | #include <linux/mm.h> |
270 | #include <linux/delay.h> |
271 | #include <linux/bootmem.h> |
272 | -#include <linux/smp_lock.h> |
273 | #include <linux/interrupt.h> |
274 | #include <linux/mc146818rtc.h> |
275 | #include <linux/kernel_stat.h> |
276 | --- a/arch/x86/kernel/cpu/common-xen.c |
277 | +++ b/arch/x86/kernel/cpu/common-xen.c |
278 | @@ -22,16 +22,40 @@ |
279 | #define phys_pkg_id(a,b) a |
280 | #endif |
281 | #endif |
282 | -#include <asm/pda.h> |
283 | #include <asm/hypervisor.h> |
284 | |
285 | #include "cpu.h" |
286 | |
287 | -DEFINE_PER_CPU(struct Xgt_desc_struct, cpu_gdt_descr); |
288 | -EXPORT_PER_CPU_SYMBOL(cpu_gdt_descr); |
289 | +DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = { |
290 | + [GDT_ENTRY_KERNEL_CS] = { 0x0000ffff, 0x00cf9a00 }, |
291 | + [GDT_ENTRY_KERNEL_DS] = { 0x0000ffff, 0x00cf9200 }, |
292 | + [GDT_ENTRY_DEFAULT_USER_CS] = { 0x0000ffff, 0x00cffa00 }, |
293 | + [GDT_ENTRY_DEFAULT_USER_DS] = { 0x0000ffff, 0x00cff200 }, |
294 | +#ifndef CONFIG_XEN |
295 | + /* |
296 | + * Segments used for calling PnP BIOS have byte granularity. |
297 | + * They code segments and data segments have fixed 64k limits, |
298 | + * the transfer segment sizes are set at run time. |
299 | + */ |
300 | + [GDT_ENTRY_PNPBIOS_CS32] = { 0x0000ffff, 0x00409a00 },/* 32-bit code */ |
301 | + [GDT_ENTRY_PNPBIOS_CS16] = { 0x0000ffff, 0x00009a00 },/* 16-bit code */ |
302 | + [GDT_ENTRY_PNPBIOS_DS] = { 0x0000ffff, 0x00009200 }, /* 16-bit data */ |
303 | + [GDT_ENTRY_PNPBIOS_TS1] = { 0x00000000, 0x00009200 },/* 16-bit data */ |
304 | + [GDT_ENTRY_PNPBIOS_TS2] = { 0x00000000, 0x00009200 },/* 16-bit data */ |
305 | + /* |
306 | + * The APM segments have byte granularity and their bases |
307 | + * are set at run time. All have 64k limits. |
308 | + */ |
309 | + [GDT_ENTRY_APMBIOS_BASE] = { 0x0000ffff, 0x00409a00 },/* 32-bit code */ |
310 | + /* 16-bit code */ |
311 | + [GDT_ENTRY_APMBIOS_BASE+1] = { 0x0000ffff, 0x00009a00 }, |
312 | + [GDT_ENTRY_APMBIOS_BASE+2] = { 0x0000ffff, 0x00409200 }, /* data */ |
313 | |
314 | -struct i386_pda *_cpu_pda[NR_CPUS] __read_mostly; |
315 | -EXPORT_SYMBOL(_cpu_pda); |
316 | + [GDT_ENTRY_ESPFIX_SS] = { 0x00000000, 0x00c09200 }, |
317 | +#endif |
318 | + [GDT_ENTRY_PERCPU] = { 0x00000000, 0x00000000 }, |
319 | +} }; |
320 | +EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); |
321 | |
322 | static int cachesize_override __cpuinitdata = -1; |
323 | static int disable_x86_fxsr __cpuinitdata; |
324 | @@ -373,7 +397,7 @@ |
325 | /* |
326 | * This does the hard work of actually picking apart the CPU stuff... |
327 | */ |
328 | -void __cpuinit identify_cpu(struct cpuinfo_x86 *c) |
329 | +static void __cpuinit identify_cpu(struct cpuinfo_x86 *c) |
330 | { |
331 | int i; |
332 | |
333 | @@ -484,15 +508,22 @@ |
334 | |
335 | /* Init Machine Check Exception if available. */ |
336 | mcheck_init(c); |
337 | +} |
338 | |
339 | - if (c == &boot_cpu_data) |
340 | - sysenter_setup(); |
341 | +void __init identify_boot_cpu(void) |
342 | +{ |
343 | + identify_cpu(&boot_cpu_data); |
344 | + sysenter_setup(); |
345 | enable_sep_cpu(); |
346 | + mtrr_bp_init(); |
347 | +} |
348 | |
349 | - if (c == &boot_cpu_data) |
350 | - mtrr_bp_init(); |
351 | - else |
352 | - mtrr_ap_init(); |
353 | +void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) |
354 | +{ |
355 | + BUG_ON(c == &boot_cpu_data); |
356 | + identify_cpu(c); |
357 | + enable_sep_cpu(); |
358 | + mtrr_ap_init(); |
359 | } |
360 | |
361 | #ifdef CONFIG_X86_HT |
362 | @@ -606,136 +637,47 @@ |
363 | #endif |
364 | } |
365 | |
366 | -/* Make sure %gs is initialized properly in idle threads */ |
367 | +/* Make sure %fs is initialized properly in idle threads */ |
368 | struct pt_regs * __devinit idle_regs(struct pt_regs *regs) |
369 | { |
370 | memset(regs, 0, sizeof(struct pt_regs)); |
371 | - regs->xfs = __KERNEL_PDA; |
372 | + regs->xfs = __KERNEL_PERCPU; |
373 | return regs; |
374 | } |
375 | |
376 | -static __cpuinit int alloc_gdt(int cpu) |
377 | +/* Current gdt points %fs at the "master" per-cpu area: after this, |
378 | + * it's on the real one. */ |
379 | +void switch_to_new_gdt(void) |
380 | { |
381 | - struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
382 | - struct desc_struct *gdt; |
383 | - struct i386_pda *pda; |
384 | - |
385 | - gdt = (struct desc_struct *)cpu_gdt_descr->address; |
386 | - pda = cpu_pda(cpu); |
387 | - |
388 | - /* |
389 | - * This is a horrible hack to allocate the GDT. The problem |
390 | - * is that cpu_init() is called really early for the boot CPU |
391 | - * (and hence needs bootmem) but much later for the secondary |
392 | - * CPUs, when bootmem will have gone away |
393 | - */ |
394 | - if (NODE_DATA(0)->bdata->node_bootmem_map) { |
395 | - BUG_ON(gdt != NULL || pda != NULL); |
396 | - |
397 | - gdt = alloc_bootmem_pages(PAGE_SIZE); |
398 | - pda = alloc_bootmem(sizeof(*pda)); |
399 | - /* alloc_bootmem(_pages) panics on failure, so no check */ |
400 | - |
401 | - memset(gdt, 0, PAGE_SIZE); |
402 | - memset(pda, 0, sizeof(*pda)); |
403 | - } else { |
404 | - /* GDT and PDA might already have been allocated if |
405 | - this is a CPU hotplug re-insertion. */ |
406 | - if (gdt == NULL) |
407 | - gdt = (struct desc_struct *)get_zeroed_page(GFP_KERNEL); |
408 | - |
409 | - if (pda == NULL) |
410 | - pda = kmalloc_node(sizeof(*pda), GFP_KERNEL, cpu_to_node(cpu)); |
411 | - |
412 | - if (unlikely(!gdt || !pda)) { |
413 | - free_pages((unsigned long)gdt, 0); |
414 | - kfree(pda); |
415 | - return 0; |
416 | - } |
417 | - } |
418 | - |
419 | - cpu_gdt_descr->address = (unsigned long)gdt; |
420 | - cpu_pda(cpu) = pda; |
421 | - |
422 | - return 1; |
423 | -} |
424 | - |
425 | -/* Initial PDA used by boot CPU */ |
426 | -struct i386_pda boot_pda = { |
427 | - ._pda = &boot_pda, |
428 | - .cpu_number = 0, |
429 | - .pcurrent = &init_task, |
430 | -}; |
431 | - |
432 | -static inline void set_kernel_fs(void) |
433 | -{ |
434 | - /* Set %fs for this CPU's PDA. Memory clobber is to create a |
435 | - barrier with respect to any PDA operations, so the compiler |
436 | - doesn't move any before here. */ |
437 | - asm volatile ("mov %0, %%fs" : : "r" (__KERNEL_PDA) : "memory"); |
438 | -} |
439 | - |
440 | -/* Initialize the CPU's GDT and PDA. The boot CPU does this for |
441 | - itself, but secondaries find this done for them. */ |
442 | -__cpuinit int init_gdt(int cpu, struct task_struct *idle) |
443 | -{ |
444 | - struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
445 | - struct desc_struct *gdt; |
446 | - struct i386_pda *pda; |
447 | - |
448 | - /* For non-boot CPUs, the GDT and PDA should already have been |
449 | - allocated. */ |
450 | - if (!alloc_gdt(cpu)) { |
451 | - printk(KERN_CRIT "CPU%d failed to allocate GDT or PDA\n", cpu); |
452 | - return 0; |
453 | - } |
454 | - |
455 | - gdt = (struct desc_struct *)cpu_gdt_descr->address; |
456 | - pda = cpu_pda(cpu); |
457 | - |
458 | - BUG_ON(gdt == NULL || pda == NULL); |
459 | - |
460 | - /* |
461 | - * Initialize the per-CPU GDT with the boot GDT, |
462 | - * and set up the GDT descriptor: |
463 | - */ |
464 | - memcpy(gdt, cpu_gdt_table, GDT_SIZE); |
465 | - cpu_gdt_descr->size = GDT_SIZE - 1; |
466 | - |
467 | - pack_descriptor((u32 *)&gdt[GDT_ENTRY_PDA].a, |
468 | - (u32 *)&gdt[GDT_ENTRY_PDA].b, |
469 | - (unsigned long)pda, sizeof(*pda) - 1, |
470 | - 0x80 | DESCTYPE_S | 0x2, 0); /* present read-write data segment */ |
471 | - |
472 | - memset(pda, 0, sizeof(*pda)); |
473 | - pda->_pda = pda; |
474 | - pda->cpu_number = cpu; |
475 | - pda->pcurrent = idle; |
476 | - |
477 | - return 1; |
478 | -} |
479 | - |
480 | -void __cpuinit cpu_set_gdt(int cpu) |
481 | -{ |
482 | - struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
483 | + struct Xgt_desc_struct gdt_descr; |
484 | unsigned long va, frames[16]; |
485 | int f; |
486 | |
487 | - for (va = cpu_gdt_descr->address, f = 0; |
488 | - va < cpu_gdt_descr->address + cpu_gdt_descr->size; |
489 | + gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id()); |
490 | + gdt_descr.size = GDT_SIZE - 1; |
491 | + |
492 | + for (va = gdt_descr.address, f = 0; |
493 | + va < gdt_descr.address + gdt_descr.size; |
494 | va += PAGE_SIZE, f++) { |
495 | frames[f] = virt_to_mfn(va); |
496 | make_lowmem_page_readonly( |
497 | (void *)va, XENFEAT_writable_descriptor_tables); |
498 | } |
499 | - BUG_ON(HYPERVISOR_set_gdt(frames, (cpu_gdt_descr->size + 1) / 8)); |
500 | - |
501 | - set_kernel_fs(); |
502 | + if (HYPERVISOR_set_gdt(frames, (gdt_descr.size + 1) / 8)) |
503 | + BUG(); |
504 | + asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory"); |
505 | } |
506 | |
507 | -/* Common CPU init for both boot and secondary CPUs */ |
508 | -static void __cpuinit _cpu_init(int cpu, struct task_struct *curr) |
509 | +/* |
510 | + * cpu_init() initializes state that is per-CPU. Some data is already |
511 | + * initialized (naturally) in the bootstrap process, such as the GDT |
512 | + * and IDT. We reload them nevertheless, this function acts as a |
513 | + * 'CPU state barrier', nothing should get across. |
514 | + */ |
515 | +void __cpuinit cpu_init(void) |
516 | { |
517 | + int cpu = smp_processor_id(); |
518 | + struct task_struct *curr = current; |
519 | #ifndef CONFIG_X86_NO_TSS |
520 | struct tss_struct * t = &per_cpu(init_tss, cpu); |
521 | #endif |
522 | @@ -757,6 +699,8 @@ |
523 | set_in_cr4(X86_CR4_TSD); |
524 | } |
525 | |
526 | + switch_to_new_gdt(); |
527 | + |
528 | /* |
529 | * Set up and load the per-CPU TSS and LDT |
530 | */ |
531 | @@ -794,38 +738,6 @@ |
532 | mxcsr_feature_mask_init(); |
533 | } |
534 | |
535 | -/* Entrypoint to initialize secondary CPU */ |
536 | -void __cpuinit secondary_cpu_init(void) |
537 | -{ |
538 | - int cpu = smp_processor_id(); |
539 | - struct task_struct *curr = current; |
540 | - |
541 | - _cpu_init(cpu, curr); |
542 | -} |
543 | - |
544 | -/* |
545 | - * cpu_init() initializes state that is per-CPU. Some data is already |
546 | - * initialized (naturally) in the bootstrap process, such as the GDT |
547 | - * and IDT. We reload them nevertheless, this function acts as a |
548 | - * 'CPU state barrier', nothing should get across. |
549 | - */ |
550 | -void __cpuinit cpu_init(void) |
551 | -{ |
552 | - int cpu = smp_processor_id(); |
553 | - struct task_struct *curr = current; |
554 | - |
555 | - /* Set up the real GDT and PDA, so we can transition from the |
556 | - boot versions. */ |
557 | - if (!init_gdt(cpu, curr)) { |
558 | - /* failed to allocate something; not much we can do... */ |
559 | - for (;;) |
560 | - local_irq_enable(); |
561 | - } |
562 | - |
563 | - cpu_set_gdt(cpu); |
564 | - _cpu_init(cpu, curr); |
565 | -} |
566 | - |
567 | #ifdef CONFIG_HOTPLUG_CPU |
568 | void __cpuinit cpu_uninit(void) |
569 | { |
570 | --- a/arch/x86/kernel/cpu/mtrr/main-xen.c |
571 | +++ b/arch/x86/kernel/cpu/mtrr/main-xen.c |
572 | @@ -167,7 +167,7 @@ |
573 | EXPORT_SYMBOL(mtrr_add); |
574 | EXPORT_SYMBOL(mtrr_del); |
575 | |
576 | -void __init mtrr_bp_init(void) |
577 | +__init void mtrr_bp_init(void) |
578 | { |
579 | } |
580 | |
581 | --- a/arch/x86/kernel/e820_32-xen.c |
582 | +++ b/arch/x86/kernel/e820_32-xen.c |
583 | @@ -162,26 +162,27 @@ |
584 | |
585 | static int __init romsignature(const unsigned char *rom) |
586 | { |
587 | + const unsigned short * const ptr = (const unsigned short *)rom; |
588 | unsigned short sig; |
589 | |
590 | - return probe_kernel_address((const unsigned short *)rom, sig) == 0 && |
591 | - sig == ROMSIGNATURE; |
592 | + return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE; |
593 | } |
594 | |
595 | -static int __init romchecksum(unsigned char *rom, unsigned long length) |
596 | +static int __init romchecksum(const unsigned char *rom, unsigned long length) |
597 | { |
598 | - unsigned char sum; |
599 | + unsigned char sum, c; |
600 | |
601 | - for (sum = 0; length; length--) |
602 | - sum += *rom++; |
603 | - return sum == 0; |
604 | + for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--) |
605 | + sum += c; |
606 | + return !length && !sum; |
607 | } |
608 | |
609 | static void __init probe_roms(void) |
610 | { |
611 | + const unsigned char *rom; |
612 | unsigned long start, length, upper; |
613 | - unsigned char *rom; |
614 | - int i; |
615 | + unsigned char c; |
616 | + int i; |
617 | |
618 | #ifdef CONFIG_XEN |
619 | /* Nothing to do if not running in dom0. */ |
620 | @@ -198,8 +199,11 @@ |
621 | |
622 | video_rom_resource.start = start; |
623 | |
624 | + if (probe_kernel_address(rom + 2, c) != 0) |
625 | + continue; |
626 | + |
627 | /* 0 < length <= 0x7f * 512, historically */ |
628 | - length = rom[2] * 512; |
629 | + length = c * 512; |
630 | |
631 | /* if checksum okay, trust length byte */ |
632 | if (length && romchecksum(rom, length)) |
633 | @@ -233,8 +237,11 @@ |
634 | if (!romsignature(rom)) |
635 | continue; |
636 | |
637 | + if (probe_kernel_address(rom + 2, c) != 0) |
638 | + continue; |
639 | + |
640 | /* 0 < length <= 0x7f * 512, historically */ |
641 | - length = rom[2] * 512; |
642 | + length = c * 512; |
643 | |
644 | /* but accept any length that fits if checksum okay */ |
645 | if (!length || start + length > upper || !romchecksum(rom, length)) |
646 | @@ -249,7 +256,7 @@ |
647 | } |
648 | |
649 | #ifdef CONFIG_XEN |
650 | -static struct e820map machine_e820 __initdata; |
651 | +static struct e820map machine_e820; |
652 | #define e820 machine_e820 |
653 | #endif |
654 | |
655 | @@ -409,10 +416,8 @@ |
656 | ____________________33__ |
657 | ______________________4_ |
658 | */ |
659 | - printk("sanitize start\n"); |
660 | /* if there's only one memory region, don't bother */ |
661 | if (*pnr_map < 2) { |
662 | - printk("sanitize bail 0\n"); |
663 | return -1; |
664 | } |
665 | |
666 | @@ -421,7 +426,6 @@ |
667 | /* bail out if we find any unreasonable addresses in bios map */ |
668 | for (i=0; i<old_nr; i++) |
669 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) { |
670 | - printk("sanitize bail 1\n"); |
671 | return -1; |
672 | } |
673 | |
674 | @@ -517,7 +521,6 @@ |
675 | memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry)); |
676 | *pnr_map = new_nr; |
677 | |
678 | - printk("sanitize end\n"); |
679 | return 0; |
680 | } |
681 | |
682 | @@ -552,7 +555,6 @@ |
683 | unsigned long long size = biosmap->size; |
684 | unsigned long long end = start + size; |
685 | unsigned long type = biosmap->type; |
686 | - printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start, size, end, type); |
687 | |
688 | /* Overflow in 64 bits? Ignore the memory map. */ |
689 | if (start > end) |
690 | @@ -564,17 +566,11 @@ |
691 | * Not right. Fix it up. |
692 | */ |
693 | if (type == E820_RAM) { |
694 | - printk("copy_e820_map() type is E820_RAM\n"); |
695 | if (start < 0x100000ULL && end > 0xA0000ULL) { |
696 | - printk("copy_e820_map() lies in range...\n"); |
697 | - if (start < 0xA0000ULL) { |
698 | - printk("copy_e820_map() start < 0xA0000ULL\n"); |
699 | + if (start < 0xA0000ULL) |
700 | add_memory_region(start, 0xA0000ULL-start, type); |
701 | - } |
702 | - if (end <= 0x100000ULL) { |
703 | - printk("copy_e820_map() end <= 0x100000ULL\n"); |
704 | + if (end <= 0x100000ULL) |
705 | continue; |
706 | - } |
707 | start = 0x100000ULL; |
708 | size = end - start; |
709 | } |
710 | --- a/arch/x86/kernel/e820_64-xen.c |
711 | +++ b/arch/x86/kernel/e820_64-xen.c |
712 | @@ -17,6 +17,8 @@ |
713 | #include <linux/kexec.h> |
714 | #include <linux/module.h> |
715 | #include <linux/mm.h> |
716 | +#include <linux/suspend.h> |
717 | +#include <linux/pfn.h> |
718 | |
719 | #include <asm/pgtable.h> |
720 | #include <asm/page.h> |
721 | @@ -28,7 +30,7 @@ |
722 | |
723 | struct e820map e820 __initdata; |
724 | #ifdef CONFIG_XEN |
725 | -struct e820map machine_e820 __initdata; |
726 | +struct e820map machine_e820; |
727 | #endif |
728 | |
729 | /* |
730 | @@ -293,22 +295,6 @@ |
731 | } |
732 | |
733 | #ifndef CONFIG_XEN |
734 | -/* Mark pages corresponding to given address range as nosave */ |
735 | -static void __init |
736 | -e820_mark_nosave_range(unsigned long start, unsigned long end) |
737 | -{ |
738 | - unsigned long pfn, max_pfn; |
739 | - |
740 | - if (start >= end) |
741 | - return; |
742 | - |
743 | - printk("Nosave address range: %016lx - %016lx\n", start, end); |
744 | - max_pfn = end >> PAGE_SHIFT; |
745 | - for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++) |
746 | - if (pfn_valid(pfn)) |
747 | - SetPageNosave(pfn_to_page(pfn)); |
748 | -} |
749 | - |
750 | /* |
751 | * Find the ranges of physical addresses that do not correspond to |
752 | * e820 RAM areas and mark the corresponding pages as nosave for software |
753 | @@ -327,13 +313,13 @@ |
754 | struct e820entry *ei = &e820.map[i]; |
755 | |
756 | if (paddr < ei->addr) |
757 | - e820_mark_nosave_range(paddr, |
758 | - round_up(ei->addr, PAGE_SIZE)); |
759 | + register_nosave_region(PFN_DOWN(paddr), |
760 | + PFN_UP(ei->addr)); |
761 | |
762 | paddr = round_down(ei->addr + ei->size, PAGE_SIZE); |
763 | if (ei->type != E820_RAM) |
764 | - e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE), |
765 | - paddr); |
766 | + register_nosave_region(PFN_UP(ei->addr), |
767 | + PFN_DOWN(paddr)); |
768 | |
769 | if (paddr >= (end_pfn << PAGE_SHIFT)) |
770 | break; |
771 | --- a/arch/x86/kernel/early_printk-xen.c |
772 | +++ b/arch/x86/kernel/early_printk-xen.c |
773 | @@ -11,11 +11,10 @@ |
774 | |
775 | #ifdef __i386__ |
776 | #include <asm/setup.h> |
777 | -#define VGABASE (__ISA_IO_base + 0xb8000) |
778 | #else |
779 | #include <asm/bootsetup.h> |
780 | -#define VGABASE ((void __iomem *)0xffffffff800b8000UL) |
781 | #endif |
782 | +#define VGABASE (__ISA_IO_base + 0xb8000) |
783 | |
784 | #ifndef CONFIG_XEN |
785 | static int max_ypos = 25, max_xpos = 80; |
786 | @@ -93,9 +92,9 @@ |
787 | static void early_serial_write(struct console *con, const char *s, unsigned n) |
788 | { |
789 | while (*s && n-- > 0) { |
790 | - early_serial_putc(*s); |
791 | if (*s == '\n') |
792 | early_serial_putc('\r'); |
793 | + early_serial_putc(*s); |
794 | s++; |
795 | } |
796 | } |
797 | @@ -205,7 +204,7 @@ |
798 | return ret; |
799 | } |
800 | |
801 | -void __init simnow_init(char *str) |
802 | +static void __init simnow_init(char *str) |
803 | { |
804 | char *fn = "klog"; |
805 | if (*str == '=') |
806 | @@ -277,22 +276,12 @@ |
807 | early_console = &simnow_console; |
808 | keep_early = 1; |
809 | } |
810 | + |
811 | + if (keep_early) |
812 | + early_console->flags &= ~CON_BOOT; |
813 | + else |
814 | + early_console->flags |= CON_BOOT; |
815 | register_console(early_console); |
816 | return 0; |
817 | } |
818 | - |
819 | early_param("earlyprintk", setup_early_printk); |
820 | - |
821 | -void __init disable_early_printk(void) |
822 | -{ |
823 | - if (!early_console_initialized || !early_console) |
824 | - return; |
825 | - if (!keep_early) { |
826 | - printk("disabling early console\n"); |
827 | - unregister_console(early_console); |
828 | - early_console_initialized = 0; |
829 | - } else { |
830 | - printk("keeping early console\n"); |
831 | - } |
832 | -} |
833 | - |
834 | --- a/arch/x86/kernel/entry_32-xen.S |
835 | +++ b/arch/x86/kernel/entry_32-xen.S |
836 | @@ -15,7 +15,7 @@ |
837 | * I changed all the .align's to 4 (16 byte alignment), as that's faster |
838 | * on a 486. |
839 | * |
840 | - * Stack layout in 'ret_from_system_call': |
841 | + * Stack layout in 'syscall_exit': |
842 | * ptrace needs to have all regs on the stack. |
843 | * if the order here is changed, it needs to be |
844 | * updated in fork.c:copy_process, signal.c:do_signal, |
845 | @@ -135,7 +135,7 @@ |
846 | movl $(__USER_DS), %edx; \ |
847 | movl %edx, %ds; \ |
848 | movl %edx, %es; \ |
849 | - movl $(__KERNEL_PDA), %edx; \ |
850 | + movl $(__KERNEL_PERCPU), %edx; \ |
851 | movl %edx, %fs |
852 | |
853 | #define RESTORE_INT_REGS \ |
854 | @@ -308,16 +308,12 @@ |
855 | pushl $(__USER_CS) |
856 | CFI_ADJUST_CFA_OFFSET 4 |
857 | /*CFI_REL_OFFSET cs, 0*/ |
858 | -#ifndef CONFIG_COMPAT_VDSO |
859 | /* |
860 | * Push current_thread_info()->sysenter_return to the stack. |
861 | * A tiny bit of offset fixup is necessary - 4*4 means the 4 words |
862 | * pushed above; +8 corresponds to copy_thread's esp0 setting. |
863 | */ |
864 | pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp) |
865 | -#else |
866 | - pushl $SYSENTER_RETURN |
867 | -#endif |
868 | CFI_ADJUST_CFA_OFFSET 4 |
869 | CFI_REL_OFFSET eip, 0 |
870 | |
871 | @@ -345,7 +341,7 @@ |
872 | jae syscall_badsys |
873 | call *sys_call_table(,%eax,4) |
874 | movl %eax,PT_EAX(%esp) |
875 | - DISABLE_INTERRUPTS(CLBR_ECX|CLBR_EDX) |
876 | + DISABLE_INTERRUPTS(CLBR_ANY) |
877 | TRACE_IRQS_OFF |
878 | movl TI_flags(%ebp), %ecx |
879 | testw $_TIF_ALLWORK_MASK, %cx |
880 | @@ -400,10 +396,6 @@ |
881 | CFI_ADJUST_CFA_OFFSET 4 |
882 | SAVE_ALL |
883 | GET_THREAD_INFO(%ebp) |
884 | - testl $TF_MASK,PT_EFLAGS(%esp) |
885 | - jz no_singlestep |
886 | - orl $_TIF_SINGLESTEP,TI_flags(%ebp) |
887 | -no_singlestep: |
888 | # system call tracing in operation / emulation |
889 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
890 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
891 | @@ -418,6 +410,10 @@ |
892 | # setting need_resched or sigpending |
893 | # between sampling and the iret |
894 | TRACE_IRQS_OFF |
895 | + testl $TF_MASK,PT_EFLAGS(%esp) # If tracing set singlestep flag on exit |
896 | + jz no_singlestep |
897 | + orl $_TIF_SINGLESTEP,TI_flags(%ebp) |
898 | +no_singlestep: |
899 | movl TI_flags(%ebp), %ecx |
900 | testw $_TIF_ALLWORK_MASK, %cx # current->work |
901 | jne syscall_exit_work |
902 | @@ -635,9 +631,7 @@ |
903 | #ifndef CONFIG_XEN |
904 | #define FIXUP_ESPFIX_STACK \ |
905 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
906 | - movl %fs:PDA_cpu, %ebx; \ |
907 | - PER_CPU(cpu_gdt_descr, %ebx); \ |
908 | - movl GDS_address(%ebx), %ebx; \ |
909 | + PER_CPU(gdt_page, %ebx); \ |
910 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ |
911 | addl %esp, %eax; \ |
912 | pushl $__KERNEL_DS; \ |
913 | @@ -710,7 +704,7 @@ |
914 | SAVE_ALL; \ |
915 | TRACE_IRQS_OFF \ |
916 | movl %esp,%eax; \ |
917 | - call smp_/**/name; \ |
918 | + call smp_##name; \ |
919 | jmp ret_from_intr; \ |
920 | CFI_ENDPROC; \ |
921 | ENDPROC(name) |
922 | @@ -718,10 +712,6 @@ |
923 | /* The include is where all of the SMP etc. interrupts come from */ |
924 | #include "entry_arch.h" |
925 | |
926 | -/* This alternate entry is needed because we hijack the apic LVTT */ |
927 | -#if defined(CONFIG_VMI) && defined(CONFIG_X86_LOCAL_APIC) |
928 | -BUILD_INTERRUPT(apic_vmi_timer_interrupt,LOCAL_TIMER_VECTOR) |
929 | -#endif |
930 | #else |
931 | #define UNWIND_ESPFIX_STACK |
932 | #endif |
933 | @@ -764,7 +754,7 @@ |
934 | pushl %fs |
935 | CFI_ADJUST_CFA_OFFSET 4 |
936 | /*CFI_REL_OFFSET fs, 0*/ |
937 | - movl $(__KERNEL_PDA), %ecx |
938 | + movl $(__KERNEL_PERCPU), %ecx |
939 | movl %ecx, %fs |
940 | UNWIND_ESPFIX_STACK |
941 | popl %ecx |
942 | --- a/arch/x86/kernel/entry_64-xen.S |
943 | +++ b/arch/x86/kernel/entry_64-xen.S |
944 | @@ -1254,3 +1254,10 @@ |
945 | ret |
946 | CFI_ENDPROC |
947 | ENDPROC(call_softirq) |
948 | + |
949 | +KPROBE_ENTRY(ignore_sysret) |
950 | + CFI_STARTPROC |
951 | + mov $-ENOSYS,%eax |
952 | + HYPERVISOR_IRET 0 |
953 | + CFI_ENDPROC |
954 | +ENDPROC(ignore_sysret) |
955 | --- a/arch/x86/kernel/genapic_64-xen.c |
956 | +++ b/arch/x86/kernel/genapic_64-xen.c |
957 | @@ -11,123 +11,57 @@ |
958 | #include <linux/threads.h> |
959 | #include <linux/cpumask.h> |
960 | #include <linux/string.h> |
961 | +#include <linux/module.h> |
962 | #include <linux/kernel.h> |
963 | #include <linux/ctype.h> |
964 | #include <linux/init.h> |
965 | -#include <linux/module.h> |
966 | |
967 | #include <asm/smp.h> |
968 | #include <asm/ipi.h> |
969 | +#include <asm/genapic.h> |
970 | |
971 | -#if defined(CONFIG_ACPI) |
972 | +#ifdef CONFIG_ACPI |
973 | #include <acpi/acpi_bus.h> |
974 | #endif |
975 | |
976 | /* which logical CPU number maps to which CPU (physical APIC ID) */ |
977 | -u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID }; |
978 | +u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly |
979 | + = { [0 ... NR_CPUS-1] = BAD_APICID }; |
980 | EXPORT_SYMBOL(x86_cpu_to_apicid); |
981 | -u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID }; |
982 | |
983 | -extern struct genapic apic_cluster; |
984 | -extern struct genapic apic_flat; |
985 | -extern struct genapic apic_physflat; |
986 | +u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID }; |
987 | |
988 | #ifndef CONFIG_XEN |
989 | -struct genapic *genapic = &apic_flat; |
990 | -struct genapic *genapic_force; |
991 | +struct genapic __read_mostly *genapic = &apic_flat; |
992 | #else |
993 | extern struct genapic apic_xen; |
994 | -struct genapic *genapic = &apic_xen; |
995 | +struct genapic __read_mostly *genapic = &apic_xen; |
996 | #endif |
997 | |
998 | |
999 | /* |
1000 | * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode. |
1001 | */ |
1002 | -void __init clustered_apic_check(void) |
1003 | +void __init setup_apic_routing(void) |
1004 | { |
1005 | #ifndef CONFIG_XEN |
1006 | - long i; |
1007 | - u8 clusters, max_cluster; |
1008 | - u8 id; |
1009 | - u8 cluster_cnt[NUM_APIC_CLUSTERS]; |
1010 | - int max_apic = 0; |
1011 | - |
1012 | - /* genapic selection can be forced because of certain quirks. |
1013 | - */ |
1014 | - if (genapic_force) { |
1015 | - genapic = genapic_force; |
1016 | - goto print; |
1017 | - } |
1018 | - |
1019 | -#if defined(CONFIG_ACPI) |
1020 | +#ifdef CONFIG_ACPI |
1021 | /* |
1022 | - * Some x86_64 machines use physical APIC mode regardless of how many |
1023 | - * procs/clusters are present (x86_64 ES7000 is an example). |
1024 | + * Quirk: some x86_64 machines can only use physical APIC mode |
1025 | + * regardless of how many processors are present (x86_64 ES7000 |
1026 | + * is an example). |
1027 | */ |
1028 | - if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID) |
1029 | - if (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL) { |
1030 | - genapic = &apic_cluster; |
1031 | - goto print; |
1032 | - } |
1033 | -#endif |
1034 | - |
1035 | - memset(cluster_cnt, 0, sizeof(cluster_cnt)); |
1036 | - for (i = 0; i < NR_CPUS; i++) { |
1037 | - id = bios_cpu_apicid[i]; |
1038 | - if (id == BAD_APICID) |
1039 | - continue; |
1040 | - if (id > max_apic) |
1041 | - max_apic = id; |
1042 | - cluster_cnt[APIC_CLUSTERID(id)]++; |
1043 | - } |
1044 | - |
1045 | - /* Don't use clustered mode on AMD platforms. */ |
1046 | - if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) { |
1047 | + if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID && |
1048 | + (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL)) |
1049 | genapic = &apic_physflat; |
1050 | -#ifndef CONFIG_HOTPLUG_CPU |
1051 | - /* In the CPU hotplug case we cannot use broadcast mode |
1052 | - because that opens a race when a CPU is removed. |
1053 | - Stay at physflat mode in this case. |
1054 | - It is bad to do this unconditionally though. Once |
1055 | - we have ACPI platform support for CPU hotplug |
1056 | - we should detect hotplug capablity from ACPI tables and |
1057 | - only do this when really needed. -AK */ |
1058 | - if (max_apic <= 8) |
1059 | - genapic = &apic_flat; |
1060 | + else |
1061 | #endif |
1062 | - goto print; |
1063 | - } |
1064 | |
1065 | - clusters = 0; |
1066 | - max_cluster = 0; |
1067 | - |
1068 | - for (i = 0; i < NUM_APIC_CLUSTERS; i++) { |
1069 | - if (cluster_cnt[i] > 0) { |
1070 | - ++clusters; |
1071 | - if (cluster_cnt[i] > max_cluster) |
1072 | - max_cluster = cluster_cnt[i]; |
1073 | - } |
1074 | - } |
1075 | - |
1076 | - /* |
1077 | - * If we have clusters <= 1 and CPUs <= 8 in cluster 0, then flat mode, |
1078 | - * else if max_cluster <= 4 and cluster_cnt[15] == 0, clustered logical |
1079 | - * else physical mode. |
1080 | - * (We don't use lowest priority delivery + HW APIC IRQ steering, so |
1081 | - * can ignore the clustered logical case and go straight to physical.) |
1082 | - */ |
1083 | - if (clusters <= 1 && max_cluster <= 8 && cluster_cnt[0] == max_cluster) { |
1084 | -#ifdef CONFIG_HOTPLUG_CPU |
1085 | - /* Don't use APIC shortcuts in CPU hotplug to avoid races */ |
1086 | - genapic = &apic_physflat; |
1087 | -#else |
1088 | + if (cpus_weight(cpu_possible_map) <= 8) |
1089 | genapic = &apic_flat; |
1090 | -#endif |
1091 | - } else |
1092 | - genapic = &apic_cluster; |
1093 | + else |
1094 | + genapic = &apic_physflat; |
1095 | |
1096 | -print: |
1097 | #else |
1098 | /* hardcode to xen apic functions */ |
1099 | genapic = &apic_xen; |
1100 | @@ -135,7 +69,7 @@ |
1101 | printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name); |
1102 | } |
1103 | |
1104 | -/* Same for both flat and clustered. */ |
1105 | +/* Same for both flat and physical. */ |
1106 | |
1107 | #ifdef CONFIG_XEN |
1108 | extern void xen_send_IPI_shortcut(unsigned int shortcut, int vector, unsigned int dest); |
1109 | --- a/arch/x86/kernel/genapic_xen_64.c |
1110 | +++ b/arch/x86/kernel/genapic_xen_64.c |
1111 | @@ -21,9 +21,8 @@ |
1112 | #include <asm/ipi.h> |
1113 | #else |
1114 | #include <asm/apic.h> |
1115 | -#include <asm/apicdef.h> |
1116 | -#include <asm/genapic.h> |
1117 | #endif |
1118 | +#include <asm/genapic.h> |
1119 | #include <xen/evtchn.h> |
1120 | |
1121 | DECLARE_PER_CPU(int, ipi_to_irq[NR_IPIS]); |
1122 | --- a/arch/x86/kernel/head64-xen.c |
1123 | +++ b/arch/x86/kernel/head64-xen.c |
1124 | @@ -22,13 +22,21 @@ |
1125 | #include <asm/setup.h> |
1126 | #include <asm/desc.h> |
1127 | #include <asm/pgtable.h> |
1128 | +#include <asm/tlbflush.h> |
1129 | #include <asm/sections.h> |
1130 | |
1131 | unsigned long start_pfn; |
1132 | |
1133 | +#ifndef CONFIG_XEN |
1134 | +static void __init zap_identity_mappings(void) |
1135 | +{ |
1136 | + pgd_t *pgd = pgd_offset_k(0UL); |
1137 | + pgd_clear(pgd); |
1138 | + __flush_tlb(); |
1139 | +} |
1140 | + |
1141 | /* Don't add a printk in there. printk relies on the PDA which is not initialized |
1142 | yet. */ |
1143 | -#if 0 |
1144 | static void __init clear_bss(void) |
1145 | { |
1146 | memset(__bss_start, 0, |
1147 | @@ -37,26 +45,25 @@ |
1148 | #endif |
1149 | |
1150 | #define NEW_CL_POINTER 0x228 /* Relative to real mode data */ |
1151 | -#define OLD_CL_MAGIC_ADDR 0x90020 |
1152 | +#define OLD_CL_MAGIC_ADDR 0x20 |
1153 | #define OLD_CL_MAGIC 0xA33F |
1154 | -#define OLD_CL_BASE_ADDR 0x90000 |
1155 | -#define OLD_CL_OFFSET 0x90022 |
1156 | +#define OLD_CL_OFFSET 0x22 |
1157 | |
1158 | static void __init copy_bootdata(char *real_mode_data) |
1159 | { |
1160 | #ifndef CONFIG_XEN |
1161 | - int new_data; |
1162 | + unsigned long new_data; |
1163 | char * command_line; |
1164 | |
1165 | memcpy(x86_boot_params, real_mode_data, BOOT_PARAM_SIZE); |
1166 | - new_data = *(int *) (x86_boot_params + NEW_CL_POINTER); |
1167 | + new_data = *(u32 *) (x86_boot_params + NEW_CL_POINTER); |
1168 | if (!new_data) { |
1169 | - if (OLD_CL_MAGIC != * (u16 *) OLD_CL_MAGIC_ADDR) { |
1170 | + if (OLD_CL_MAGIC != *(u16 *)(real_mode_data + OLD_CL_MAGIC_ADDR)) { |
1171 | return; |
1172 | } |
1173 | - new_data = OLD_CL_BASE_ADDR + * (u16 *) OLD_CL_OFFSET; |
1174 | + new_data = __pa(real_mode_data) + *(u16 *)(real_mode_data + OLD_CL_OFFSET); |
1175 | } |
1176 | - command_line = (char *) ((u64)(new_data)); |
1177 | + command_line = __va(new_data); |
1178 | memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE); |
1179 | #else |
1180 | int max_cmdline; |
1181 | @@ -98,10 +105,13 @@ |
1182 | while ((1UL << machine_to_phys_order) < machine_to_phys_nr_ents ) |
1183 | machine_to_phys_order++; |
1184 | |
1185 | -#if 0 |
1186 | +#ifndef CONFIG_XEN |
1187 | /* clear bss before set_intr_gate with early_idt_handler */ |
1188 | clear_bss(); |
1189 | |
1190 | + /* Make NULL pointers segfault */ |
1191 | + zap_identity_mappings(); |
1192 | + |
1193 | for (i = 0; i < IDT_ENTRIES; i++) |
1194 | set_intr_gate(i, early_idt_handler); |
1195 | asm volatile("lidt %0" :: "m" (idt_descr)); |
1196 | @@ -113,7 +123,7 @@ |
1197 | cpu_pda(i) = &boot_cpu_pda[i]; |
1198 | |
1199 | pda_init(0); |
1200 | - copy_bootdata(real_mode_data); |
1201 | + copy_bootdata(__va(real_mode_data)); |
1202 | #ifdef CONFIG_SMP |
1203 | cpu_set(0, cpu_online_map); |
1204 | #endif |
1205 | --- a/arch/x86/kernel/head_32-xen.S |
1206 | +++ b/arch/x86/kernel/head_32-xen.S |
1207 | @@ -37,7 +37,8 @@ |
1208 | /* Set up the stack pointer */ |
1209 | movl $(init_thread_union+THREAD_SIZE),%esp |
1210 | |
1211 | - call setup_pda |
1212 | + movl %ss,%eax |
1213 | + movl %eax,%fs # gets reset once there's real percpu |
1214 | |
1215 | /* get vendor info */ |
1216 | xorl %eax,%eax # call CPUID with 0 -> return vendor ID |
1217 | @@ -64,55 +65,11 @@ |
1218 | xorl %eax,%eax # Clear GS |
1219 | movl %eax,%gs |
1220 | |
1221 | - movl $(__KERNEL_PDA),%eax |
1222 | - mov %eax,%fs |
1223 | - |
1224 | cld # gcc2 wants the direction flag cleared at all times |
1225 | |
1226 | pushl $0 # fake return address for unwinder |
1227 | jmp start_kernel |
1228 | |
1229 | -/* |
1230 | - * Point the GDT at this CPU's PDA. This will be |
1231 | - * cpu_gdt_table and boot_pda. |
1232 | - */ |
1233 | -ENTRY(setup_pda) |
1234 | - /* get the PDA pointer */ |
1235 | - movl $boot_pda, %eax |
1236 | - |
1237 | - /* slot the PDA address into the GDT */ |
1238 | - mov $cpu_gdt_table, %ecx |
1239 | - mov %ax, (__KERNEL_PDA+0+2)(%ecx) /* base & 0x0000ffff */ |
1240 | - shr $16, %eax |
1241 | - mov %al, (__KERNEL_PDA+4+0)(%ecx) /* base & 0x00ff0000 */ |
1242 | - mov %ah, (__KERNEL_PDA+4+3)(%ecx) /* base & 0xff000000 */ |
1243 | - |
1244 | - # %esi still points to start_info, and no registers |
1245 | - # need to be preserved. |
1246 | - |
1247 | - movl XEN_START_mfn_list(%esi), %ebx |
1248 | - movl $(cpu_gdt_table - __PAGE_OFFSET), %eax |
1249 | - shrl $PAGE_SHIFT, %eax |
1250 | - movl (%ebx,%eax,4), %ecx |
1251 | - pushl %ecx # frame number for set_gdt below |
1252 | - |
1253 | - xorl %esi, %esi |
1254 | - xorl %edx, %edx |
1255 | - shldl $PAGE_SHIFT, %ecx, %edx |
1256 | - shll $PAGE_SHIFT, %ecx |
1257 | - orl $0x61, %ecx |
1258 | - movl $cpu_gdt_table, %ebx |
1259 | - movl $__HYPERVISOR_update_va_mapping, %eax |
1260 | - int $0x82 |
1261 | - |
1262 | - movl $(PAGE_SIZE_asm / 8), %ecx |
1263 | - movl %esp, %ebx |
1264 | - movl $__HYPERVISOR_set_gdt, %eax |
1265 | - int $0x82 |
1266 | - |
1267 | - popl %ecx |
1268 | - ret |
1269 | - |
1270 | #define HYPERCALL_PAGE_OFFSET 0x1000 |
1271 | .org HYPERCALL_PAGE_OFFSET |
1272 | ENTRY(hypercall_page) |
1273 | @@ -138,60 +95,6 @@ |
1274 | */ |
1275 | .data |
1276 | |
1277 | -/* |
1278 | - * The Global Descriptor Table contains 28 quadwords, per-CPU. |
1279 | - */ |
1280 | - .section .data.page_aligned, "aw" |
1281 | - .align PAGE_SIZE_asm |
1282 | -ENTRY(cpu_gdt_table) |
1283 | - .quad 0x0000000000000000 /* NULL descriptor */ |
1284 | - .quad 0x0000000000000000 /* 0x0b reserved */ |
1285 | - .quad 0x0000000000000000 /* 0x13 reserved */ |
1286 | - .quad 0x0000000000000000 /* 0x1b reserved */ |
1287 | - .quad 0x0000000000000000 /* 0x20 unused */ |
1288 | - .quad 0x0000000000000000 /* 0x28 unused */ |
1289 | - .quad 0x0000000000000000 /* 0x33 TLS entry 1 */ |
1290 | - .quad 0x0000000000000000 /* 0x3b TLS entry 2 */ |
1291 | - .quad 0x0000000000000000 /* 0x43 TLS entry 3 */ |
1292 | - .quad 0x0000000000000000 /* 0x4b reserved */ |
1293 | - .quad 0x0000000000000000 /* 0x53 reserved */ |
1294 | - .quad 0x0000000000000000 /* 0x5b reserved */ |
1295 | - |
1296 | - .quad 0x00cf9a000000ffff /* 0x60 kernel 4GB code at 0x00000000 */ |
1297 | - .quad 0x00cf92000000ffff /* 0x68 kernel 4GB data at 0x00000000 */ |
1298 | - .quad 0x00cffa000000ffff /* 0x73 user 4GB code at 0x00000000 */ |
1299 | - .quad 0x00cff2000000ffff /* 0x7b user 4GB data at 0x00000000 */ |
1300 | - |
1301 | - .quad 0x0000000000000000 /* 0x80 TSS descriptor */ |
1302 | - .quad 0x0000000000000000 /* 0x88 LDT descriptor */ |
1303 | - |
1304 | - /* |
1305 | - * Segments used for calling PnP BIOS have byte granularity. |
1306 | - * They code segments and data segments have fixed 64k limits, |
1307 | - * the transfer segment sizes are set at run time. |
1308 | - */ |
1309 | - .quad 0x0000000000000000 /* 0x90 32-bit code */ |
1310 | - .quad 0x0000000000000000 /* 0x98 16-bit code */ |
1311 | - .quad 0x0000000000000000 /* 0xa0 16-bit data */ |
1312 | - .quad 0x0000000000000000 /* 0xa8 16-bit data */ |
1313 | - .quad 0x0000000000000000 /* 0xb0 16-bit data */ |
1314 | - |
1315 | - /* |
1316 | - * The APM segments have byte granularity and their bases |
1317 | - * are set at run time. All have 64k limits. |
1318 | - */ |
1319 | - .quad 0x0000000000000000 /* 0xb8 APM CS code */ |
1320 | - .quad 0x0000000000000000 /* 0xc0 APM CS 16 code (16 bit) */ |
1321 | - .quad 0x0000000000000000 /* 0xc8 APM DS data */ |
1322 | - |
1323 | - .quad 0x0000000000000000 /* 0xd0 - ESPFIX SS */ |
1324 | - .quad 0x00cf92000000ffff /* 0xd8 - PDA */ |
1325 | - .quad 0x0000000000000000 /* 0xe0 - unused */ |
1326 | - .quad 0x0000000000000000 /* 0xe8 - unused */ |
1327 | - .quad 0x0000000000000000 /* 0xf0 - unused */ |
1328 | - .quad 0x0000000000000000 /* 0xf8 - GDT entry 31: double-fault TSS */ |
1329 | - .align PAGE_SIZE_asm |
1330 | - |
1331 | #if CONFIG_XEN_COMPAT <= 0x030002 |
1332 | /* |
1333 | * __xen_guest information |
1334 | --- a/arch/x86/kernel/head_64-xen.S |
1335 | +++ b/arch/x86/kernel/head_64-xen.S |
1336 | @@ -41,18 +42,15 @@ |
1337 | .word gdt_end-cpu_gdt_table-1 |
1338 | .long cpu_gdt_table-__START_KERNEL_map |
1339 | #endif |
1340 | -ENTRY(stext) |
1341 | -ENTRY(_stext) |
1342 | |
1343 | - $page = 0 |
1344 | +.balign PAGE_SIZE |
1345 | + |
1346 | #define NEXT_PAGE(name) \ |
1347 | - $page = $page + 1; \ |
1348 | - .org $page * 0x1000; \ |
1349 | - phys_##name = $page * 0x1000 + __PHYSICAL_START; \ |
1350 | + .balign PAGE_SIZE; \ |
1351 | + phys_##name = . - .bootstrap.text; \ |
1352 | ENTRY(name) |
1353 | |
1354 | NEXT_PAGE(init_level4_pgt) |
1355 | - /* This gets initialized in x86_64_start_kernel */ |
1356 | .fill 512,8,0 |
1357 | NEXT_PAGE(init_level4_user_pgt) |
1358 | /* |
1359 | @@ -136,13 +134,13 @@ |
1360 | |
1361 | ENTRY(cpu_gdt_table) |
1362 | .quad 0x0000000000000000 /* NULL descriptor */ |
1363 | + .quad 0x00cf9b000000ffff /* __KERNEL32_CS */ |
1364 | + .quad 0x00af9b000000ffff /* __KERNEL_CS */ |
1365 | + .quad 0x00cf93000000ffff /* __KERNEL_DS */ |
1366 | + .quad 0x00cffb000000ffff /* __USER32_CS */ |
1367 | + .quad 0x00cff3000000ffff /* __USER_DS, __USER32_DS */ |
1368 | + .quad 0x00affb000000ffff /* __USER_CS */ |
1369 | .quad 0x0 /* unused */ |
1370 | - .quad 0x00af9a000000ffff /* __KERNEL_CS */ |
1371 | - .quad 0x00cf92000000ffff /* __KERNEL_DS */ |
1372 | - .quad 0x00cffa000000ffff /* __USER32_CS */ |
1373 | - .quad 0x00cff2000000ffff /* __USER_DS, __USER32_DS */ |
1374 | - .quad 0x00affa000000ffff /* __USER_CS */ |
1375 | - .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ |
1376 | .quad 0,0 /* TSS */ |
1377 | .quad 0,0 /* LDT */ |
1378 | .quad 0,0,0 /* three TLS descriptors */ |
1379 | @@ -165,14 +163,11 @@ |
1380 | * __xen_guest information |
1381 | */ |
1382 | .macro utoh value |
1383 | - .if (\value) < 0 || (\value) >= 0x10 |
1384 | - utoh (((\value)>>4)&0x0fffffffffffffff) |
1385 | - .endif |
1386 | - .if ((\value) & 0xf) < 10 |
1387 | - .byte '0' + ((\value) & 0xf) |
1388 | - .else |
1389 | - .byte 'A' + ((\value) & 0xf) - 10 |
1390 | - .endif |
1391 | + i = 64 |
1392 | + .rept 16 |
1393 | + i = i - 4 |
1394 | + .byte '0' + ((((\value) >> i) & 0xf) > 9) * ('0' - 'A' + 10) + (((\value) >> i) & 0xf) |
1395 | + .endr |
1396 | .endm |
1397 | |
1398 | .section __xen_guest |
1399 | --- a/arch/x86/kernel/io_apic_32-xen.c |
1400 | +++ b/arch/x86/kernel/io_apic_32-xen.c |
1401 | @@ -25,7 +25,6 @@ |
1402 | #include <linux/init.h> |
1403 | #include <linux/delay.h> |
1404 | #include <linux/sched.h> |
1405 | -#include <linux/smp_lock.h> |
1406 | #include <linux/mc146818rtc.h> |
1407 | #include <linux/compiler.h> |
1408 | #include <linux/acpi.h> |
1409 | @@ -35,6 +34,7 @@ |
1410 | #include <linux/msi.h> |
1411 | #include <linux/htirq.h> |
1412 | #include <linux/freezer.h> |
1413 | +#include <linux/kthread.h> |
1414 | |
1415 | #include <asm/io.h> |
1416 | #include <asm/smp.h> |
1417 | @@ -705,8 +705,6 @@ |
1418 | unsigned long prev_balance_time = jiffies; |
1419 | long time_remaining = balanced_irq_interval; |
1420 | |
1421 | - daemonize("kirqd"); |
1422 | - |
1423 | /* push everything to CPU 0 to give us a starting point. */ |
1424 | for (i = 0 ; i < NR_IRQS ; i++) { |
1425 | irq_desc[i].pending_mask = cpumask_of_cpu(0); |
1426 | @@ -766,10 +764,9 @@ |
1427 | } |
1428 | |
1429 | printk(KERN_INFO "Starting balanced_irq\n"); |
1430 | - if (kernel_thread(balanced_irq, NULL, CLONE_KERNEL) >= 0) |
1431 | + if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd"))) |
1432 | return 0; |
1433 | - else |
1434 | - printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq"); |
1435 | + printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq"); |
1436 | failed: |
1437 | for_each_possible_cpu(i) { |
1438 | kfree(irq_cpu_data[i].irq_delta); |
1439 | @@ -1445,10 +1442,6 @@ |
1440 | enable_8259A_irq(0); |
1441 | } |
1442 | |
1443 | -static inline void UNEXPECTED_IO_APIC(void) |
1444 | -{ |
1445 | -} |
1446 | - |
1447 | void __init print_IO_APIC(void) |
1448 | { |
1449 | int apic, i; |
1450 | @@ -1488,34 +1481,12 @@ |
1451 | printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID); |
1452 | printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type); |
1453 | printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS); |
1454 | - if (reg_00.bits.ID >= get_physical_broadcast()) |
1455 | - UNEXPECTED_IO_APIC(); |
1456 | - if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2) |
1457 | - UNEXPECTED_IO_APIC(); |
1458 | |
1459 | printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw); |
1460 | printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries); |
1461 | - if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */ |
1462 | - (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */ |
1463 | - (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */ |
1464 | - (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */ |
1465 | - (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */ |
1466 | - (reg_01.bits.entries != 0x2E) && |
1467 | - (reg_01.bits.entries != 0x3F) |
1468 | - ) |
1469 | - UNEXPECTED_IO_APIC(); |
1470 | |
1471 | printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ); |
1472 | printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version); |
1473 | - if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */ |
1474 | - (reg_01.bits.version != 0x10) && /* oldest IO-APICs */ |
1475 | - (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */ |
1476 | - (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */ |
1477 | - (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */ |
1478 | - ) |
1479 | - UNEXPECTED_IO_APIC(); |
1480 | - if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2) |
1481 | - UNEXPECTED_IO_APIC(); |
1482 | |
1483 | /* |
1484 | * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02, |
1485 | @@ -1525,8 +1496,6 @@ |
1486 | if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) { |
1487 | printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw); |
1488 | printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration); |
1489 | - if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2) |
1490 | - UNEXPECTED_IO_APIC(); |
1491 | } |
1492 | |
1493 | /* |
1494 | @@ -1538,8 +1507,6 @@ |
1495 | reg_03.raw != reg_01.raw) { |
1496 | printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw); |
1497 | printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT); |
1498 | - if (reg_03.bits.__reserved_1) |
1499 | - UNEXPECTED_IO_APIC(); |
1500 | } |
1501 | |
1502 | printk(KERN_DEBUG ".... IRQ redirection table:\n"); |
1503 | @@ -2670,19 +2637,19 @@ |
1504 | if (irq < 0) |
1505 | return irq; |
1506 | |
1507 | - set_irq_msi(irq, desc); |
1508 | ret = msi_compose_msg(dev, irq, &msg); |
1509 | if (ret < 0) { |
1510 | destroy_irq(irq); |
1511 | return ret; |
1512 | } |
1513 | |
1514 | + set_irq_msi(irq, desc); |
1515 | write_msi_msg(irq, &msg); |
1516 | |
1517 | set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, |
1518 | "edge"); |
1519 | |
1520 | - return irq; |
1521 | + return 0; |
1522 | } |
1523 | |
1524 | void arch_teardown_msi_irq(unsigned int irq) |
1525 | --- a/arch/x86/kernel/io_apic_64-xen.c |
1526 | +++ b/arch/x86/kernel/io_apic_64-xen.c |
1527 | @@ -25,7 +25,6 @@ |
1528 | #include <linux/init.h> |
1529 | #include <linux/delay.h> |
1530 | #include <linux/sched.h> |
1531 | -#include <linux/smp_lock.h> |
1532 | #include <linux/pci.h> |
1533 | #include <linux/mc146818rtc.h> |
1534 | #include <linux/acpi.h> |
1535 | @@ -897,10 +896,6 @@ |
1536 | enable_8259A_irq(0); |
1537 | } |
1538 | |
1539 | -void __init UNEXPECTED_IO_APIC(void) |
1540 | -{ |
1541 | -} |
1542 | - |
1543 | void __apicdebuginit print_IO_APIC(void) |
1544 | { |
1545 | int apic, i; |
1546 | @@ -936,40 +931,16 @@ |
1547 | printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid); |
1548 | printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw); |
1549 | printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID); |
1550 | - if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2) |
1551 | - UNEXPECTED_IO_APIC(); |
1552 | |
1553 | printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01); |
1554 | printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries); |
1555 | - if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */ |
1556 | - (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */ |
1557 | - (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */ |
1558 | - (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */ |
1559 | - (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */ |
1560 | - (reg_01.bits.entries != 0x2E) && |
1561 | - (reg_01.bits.entries != 0x3F) && |
1562 | - (reg_01.bits.entries != 0x03) |
1563 | - ) |
1564 | - UNEXPECTED_IO_APIC(); |
1565 | |
1566 | printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ); |
1567 | printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version); |
1568 | - if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */ |
1569 | - (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */ |
1570 | - (reg_01.bits.version != 0x10) && /* oldest IO-APICs */ |
1571 | - (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */ |
1572 | - (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */ |
1573 | - (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */ |
1574 | - ) |
1575 | - UNEXPECTED_IO_APIC(); |
1576 | - if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2) |
1577 | - UNEXPECTED_IO_APIC(); |
1578 | |
1579 | if (reg_01.bits.version >= 0x10) { |
1580 | printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw); |
1581 | printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration); |
1582 | - if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2) |
1583 | - UNEXPECTED_IO_APIC(); |
1584 | } |
1585 | |
1586 | printk(KERN_DEBUG ".... IRQ redirection table:\n"); |
1587 | @@ -1401,8 +1372,7 @@ |
1588 | |
1589 | vector = ~get_irq_regs()->orig_rax; |
1590 | me = smp_processor_id(); |
1591 | - if ((vector == cfg->vector) && |
1592 | - cpu_isset(smp_processor_id(), cfg->domain)) { |
1593 | + if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) { |
1594 | cpumask_t cleanup_mask; |
1595 | |
1596 | cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map); |
1597 | @@ -1437,7 +1407,7 @@ |
1598 | |
1599 | /* |
1600 | * We must acknowledge the irq before we move it or the acknowledge will |
1601 | - * not propogate properly. |
1602 | + * not propagate properly. |
1603 | */ |
1604 | ack_APIC_irq(); |
1605 | |
1606 | @@ -1520,6 +1490,7 @@ |
1607 | static void end_lapic_irq (unsigned int i) { /* nothing */ } |
1608 | |
1609 | static struct hw_interrupt_type lapic_irq_type __read_mostly = { |
1610 | + .name = "local-APIC", |
1611 | .typename = "local-APIC-edge", |
1612 | .startup = NULL, /* startup_irq() not used for IRQ0 */ |
1613 | .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */ |
1614 | @@ -1989,18 +1960,18 @@ |
1615 | if (irq < 0) |
1616 | return irq; |
1617 | |
1618 | - set_irq_msi(irq, desc); |
1619 | ret = msi_compose_msg(dev, irq, &msg); |
1620 | if (ret < 0) { |
1621 | destroy_irq(irq); |
1622 | return ret; |
1623 | } |
1624 | |
1625 | + set_irq_msi(irq, desc); |
1626 | write_msi_msg(irq, &msg); |
1627 | |
1628 | set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge"); |
1629 | |
1630 | - return irq; |
1631 | + return 0; |
1632 | } |
1633 | |
1634 | void arch_teardown_msi_irq(unsigned int irq) |
1635 | --- a/arch/x86/kernel/ioport_32-xen.c |
1636 | +++ b/arch/x86/kernel/ioport_32-xen.c |
1637 | @@ -12,10 +12,10 @@ |
1638 | #include <linux/types.h> |
1639 | #include <linux/ioport.h> |
1640 | #include <linux/smp.h> |
1641 | -#include <linux/smp_lock.h> |
1642 | #include <linux/stddef.h> |
1643 | #include <linux/slab.h> |
1644 | #include <linux/thread_info.h> |
1645 | +#include <linux/syscalls.h> |
1646 | #include <xen/interface/physdev.h> |
1647 | |
1648 | /* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */ |
1649 | --- a/arch/x86/kernel/ioport_64-xen.c |
1650 | +++ b/arch/x86/kernel/ioport_64-xen.c |
1651 | @@ -13,10 +13,10 @@ |
1652 | #include <linux/ioport.h> |
1653 | #include <linux/mm.h> |
1654 | #include <linux/smp.h> |
1655 | -#include <linux/smp_lock.h> |
1656 | #include <linux/stddef.h> |
1657 | #include <linux/slab.h> |
1658 | #include <linux/thread_info.h> |
1659 | +#include <linux/syscalls.h> |
1660 | #include <xen/interface/physdev.h> |
1661 | |
1662 | /* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */ |
1663 | --- a/arch/x86/kernel/irq_32-xen.c |
1664 | +++ b/arch/x86/kernel/irq_32-xen.c |
1665 | @@ -24,6 +24,9 @@ |
1666 | DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp; |
1667 | EXPORT_PER_CPU_SYMBOL(irq_stat); |
1668 | |
1669 | +DEFINE_PER_CPU(struct pt_regs *, irq_regs); |
1670 | +EXPORT_PER_CPU_SYMBOL(irq_regs); |
1671 | + |
1672 | /* |
1673 | * 'what should we do if we get a hw irq event on an illegal vector'. |
1674 | * each architecture has to answer this themselves. |
1675 | --- a/arch/x86/kernel/irq_64-xen.c |
1676 | +++ b/arch/x86/kernel/irq_64-xen.c |
1677 | @@ -32,7 +32,7 @@ |
1678 | */ |
1679 | static inline void stack_overflow_check(struct pt_regs *regs) |
1680 | { |
1681 | - u64 curbase = (u64) current->thread_info; |
1682 | + u64 curbase = (u64)task_stack_page(current); |
1683 | static unsigned long warned = -60*HZ; |
1684 | |
1685 | if (regs->rsp >= curbase && regs->rsp <= curbase + THREAD_SIZE && |
1686 | @@ -145,17 +145,43 @@ |
1687 | |
1688 | for (irq = 0; irq < NR_IRQS; irq++) { |
1689 | cpumask_t mask; |
1690 | + int break_affinity = 0; |
1691 | + int set_affinity = 1; |
1692 | + |
1693 | if (irq == 2) |
1694 | continue; |
1695 | |
1696 | + /* interrupt's are disabled at this point */ |
1697 | + spin_lock(&irq_desc[irq].lock); |
1698 | + |
1699 | + if (!irq_has_action(irq) || |
1700 | + cpus_equal(irq_desc[irq].affinity, map)) { |
1701 | + spin_unlock(&irq_desc[irq].lock); |
1702 | + continue; |
1703 | + } |
1704 | + |
1705 | cpus_and(mask, irq_desc[irq].affinity, map); |
1706 | - if (any_online_cpu(mask) == NR_CPUS) { |
1707 | - printk("Breaking affinity for irq %i\n", irq); |
1708 | + if (cpus_empty(mask)) { |
1709 | + break_affinity = 1; |
1710 | mask = map; |
1711 | } |
1712 | + |
1713 | + if (irq_desc[irq].chip->mask) |
1714 | + irq_desc[irq].chip->mask(irq); |
1715 | + |
1716 | if (irq_desc[irq].chip->set_affinity) |
1717 | irq_desc[irq].chip->set_affinity(irq, mask); |
1718 | - else if (irq_desc[irq].action && !(warned++)) |
1719 | + else if (!(warned++)) |
1720 | + set_affinity = 0; |
1721 | + |
1722 | + if (irq_desc[irq].chip->unmask) |
1723 | + irq_desc[irq].chip->unmask(irq); |
1724 | + |
1725 | + spin_unlock(&irq_desc[irq].lock); |
1726 | + |
1727 | + if (break_affinity && set_affinity) |
1728 | + printk("Broke affinity for irq %i\n", irq); |
1729 | + else if (!set_affinity) |
1730 | printk("Cannot set affinity for irq %i\n", irq); |
1731 | } |
1732 | |
1733 | --- a/arch/x86/kernel/ldt_32-xen.c |
1734 | +++ b/arch/x86/kernel/ldt_32-xen.c |
1735 | @@ -10,7 +10,6 @@ |
1736 | #include <linux/string.h> |
1737 | #include <linux/mm.h> |
1738 | #include <linux/smp.h> |
1739 | -#include <linux/smp_lock.h> |
1740 | #include <linux/vmalloc.h> |
1741 | #include <linux/slab.h> |
1742 | |
1743 | --- a/arch/x86/kernel/ldt_64-xen.c |
1744 | +++ b/arch/x86/kernel/ldt_64-xen.c |
1745 | @@ -13,7 +13,6 @@ |
1746 | #include <linux/string.h> |
1747 | #include <linux/mm.h> |
1748 | #include <linux/smp.h> |
1749 | -#include <linux/smp_lock.h> |
1750 | #include <linux/vmalloc.h> |
1751 | #include <linux/slab.h> |
1752 | |
1753 | --- a/arch/x86/kernel/microcode-xen.c |
1754 | +++ b/arch/x86/kernel/microcode-xen.c |
1755 | @@ -135,7 +135,7 @@ |
1756 | return 0; |
1757 | } |
1758 | |
1759 | -static void __exit microcode_dev_exit (void) |
1760 | +static void microcode_dev_exit (void) |
1761 | { |
1762 | misc_deregister(µcode_dev); |
1763 | } |
1764 | --- a/arch/x86/kernel/mpparse_32-xen.c |
1765 | +++ b/arch/x86/kernel/mpparse_32-xen.c |
1766 | @@ -18,7 +18,6 @@ |
1767 | #include <linux/acpi.h> |
1768 | #include <linux/delay.h> |
1769 | #include <linux/bootmem.h> |
1770 | -#include <linux/smp_lock.h> |
1771 | #include <linux/kernel_stat.h> |
1772 | #include <linux/mc146818rtc.h> |
1773 | #include <linux/bitops.h> |
1774 | @@ -484,7 +483,7 @@ |
1775 | } |
1776 | ++mpc_record; |
1777 | } |
1778 | - clustered_apic_check(); |
1779 | + setup_apic_routing(); |
1780 | if (!num_processors) |
1781 | printk(KERN_ERR "SMP mptable: no processors registered!\n"); |
1782 | return num_processors; |
1783 | --- a/arch/x86/kernel/mpparse_64-xen.c |
1784 | +++ b/arch/x86/kernel/mpparse_64-xen.c |
1785 | @@ -17,7 +17,6 @@ |
1786 | #include <linux/init.h> |
1787 | #include <linux/delay.h> |
1788 | #include <linux/bootmem.h> |
1789 | -#include <linux/smp_lock.h> |
1790 | #include <linux/kernel_stat.h> |
1791 | #include <linux/mc146818rtc.h> |
1792 | #include <linux/acpi.h> |
1793 | @@ -307,7 +306,7 @@ |
1794 | } |
1795 | } |
1796 | } |
1797 | - clustered_apic_check(); |
1798 | + setup_apic_routing(); |
1799 | if (!num_processors) |
1800 | printk(KERN_ERR "MPTABLE: no processors registered!\n"); |
1801 | return num_processors; |
1802 | --- a/arch/x86/kernel/pci-dma_32-xen.c |
1803 | +++ b/arch/x86/kernel/pci-dma_32-xen.c |
1804 | @@ -13,6 +13,7 @@ |
1805 | #include <linux/pci.h> |
1806 | #include <linux/module.h> |
1807 | #include <linux/version.h> |
1808 | +#include <linux/pci.h> |
1809 | #include <asm/io.h> |
1810 | #include <xen/balloon.h> |
1811 | #include <xen/gnttab.h> |
1812 | @@ -284,7 +285,7 @@ |
1813 | { |
1814 | void __iomem *mem_base = NULL; |
1815 | int pages = size >> PAGE_SHIFT; |
1816 | - int bitmap_size = (pages + 31)/32; |
1817 | + int bitmap_size = BITS_TO_LONGS(pages) * sizeof(long); |
1818 | |
1819 | if ((flags & (DMA_MEMORY_MAP | DMA_MEMORY_IO)) == 0) |
1820 | goto out; |
1821 | @@ -357,6 +358,32 @@ |
1822 | EXPORT_SYMBOL(dma_mark_declared_memory_occupied); |
1823 | #endif /* ARCH_HAS_DMA_DECLARE_COHERENT_MEMORY */ |
1824 | |
1825 | +#if defined(CONFIG_PCI) && !defined(CONFIG_XEN) |
1826 | +/* Many VIA bridges seem to corrupt data for DAC. Disable it here */ |
1827 | + |
1828 | +int forbid_dac; |
1829 | +EXPORT_SYMBOL(forbid_dac); |
1830 | + |
1831 | +static __devinit void via_no_dac(struct pci_dev *dev) |
1832 | +{ |
1833 | + if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI && forbid_dac == 0) { |
1834 | + printk(KERN_INFO "PCI: VIA PCI bridge detected. Disabling DAC.\n"); |
1835 | + forbid_dac = 1; |
1836 | + } |
1837 | +} |
1838 | +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, PCI_ANY_ID, via_no_dac); |
1839 | + |
1840 | +static int check_iommu(char *s) |
1841 | +{ |
1842 | + if (!strcmp(s, "usedac")) { |
1843 | + forbid_dac = -1; |
1844 | + return 1; |
1845 | + } |
1846 | + return 0; |
1847 | +} |
1848 | +__setup("iommu=", check_iommu); |
1849 | +#endif |
1850 | + |
1851 | dma_addr_t |
1852 | dma_map_single(struct device *dev, void *ptr, size_t size, |
1853 | enum dma_data_direction direction) |
1854 | --- a/arch/x86/kernel/pci-swiotlb_64-xen.c |
1855 | +++ b/arch/x86/kernel/pci-swiotlb_64-xen.c |
1856 | @@ -16,7 +16,7 @@ |
1857 | |
1858 | void swiotlb_init(void); |
1859 | |
1860 | -struct dma_mapping_ops swiotlb_dma_ops = { |
1861 | +const struct dma_mapping_ops swiotlb_dma_ops = { |
1862 | #if 0 |
1863 | .mapping_error = swiotlb_dma_mapping_error, |
1864 | .alloc_coherent = swiotlb_alloc_coherent, |
1865 | --- a/arch/x86/kernel/process_32-xen.c |
1866 | +++ b/arch/x86/kernel/process_32-xen.c |
1867 | @@ -21,7 +21,6 @@ |
1868 | #include <linux/mm.h> |
1869 | #include <linux/elfcore.h> |
1870 | #include <linux/smp.h> |
1871 | -#include <linux/smp_lock.h> |
1872 | #include <linux/stddef.h> |
1873 | #include <linux/slab.h> |
1874 | #include <linux/vmalloc.h> |
1875 | @@ -39,6 +38,7 @@ |
1876 | #include <linux/random.h> |
1877 | #include <linux/personality.h> |
1878 | #include <linux/tick.h> |
1879 | +#include <linux/percpu.h> |
1880 | |
1881 | #include <asm/uaccess.h> |
1882 | #include <asm/pgtable.h> |
1883 | @@ -61,7 +61,6 @@ |
1884 | |
1885 | #include <asm/tlbflush.h> |
1886 | #include <asm/cpu.h> |
1887 | -#include <asm/pda.h> |
1888 | |
1889 | asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); |
1890 | |
1891 | @@ -70,6 +69,12 @@ |
1892 | unsigned long boot_option_idle_override = 0; |
1893 | EXPORT_SYMBOL(boot_option_idle_override); |
1894 | |
1895 | +DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; |
1896 | +EXPORT_PER_CPU_SYMBOL(current_task); |
1897 | + |
1898 | +DEFINE_PER_CPU(int, cpu_number); |
1899 | +EXPORT_PER_CPU_SYMBOL(cpu_number); |
1900 | + |
1901 | /* |
1902 | * Return saved PC of a blocked thread. |
1903 | */ |
1904 | @@ -168,6 +173,7 @@ |
1905 | if (__get_cpu_var(cpu_idle_state)) |
1906 | __get_cpu_var(cpu_idle_state) = 0; |
1907 | |
1908 | + check_pgt_cache(); |
1909 | rmb(); |
1910 | idle = xen_idle; /* no alternatives */ |
1911 | |
1912 | @@ -218,18 +224,19 @@ |
1913 | { |
1914 | } |
1915 | |
1916 | -static int __init idle_setup (char *str) |
1917 | +static int __init idle_setup(char *str) |
1918 | { |
1919 | - if (!strncmp(str, "poll", 4)) { |
1920 | + if (!strcmp(str, "poll")) { |
1921 | printk("using polling idle threads.\n"); |
1922 | pm_idle = poll_idle; |
1923 | } |
1924 | + else |
1925 | + return -1; |
1926 | |
1927 | boot_option_idle_override = 1; |
1928 | - return 1; |
1929 | + return 0; |
1930 | } |
1931 | - |
1932 | -__setup("idle=", idle_setup); |
1933 | +early_param("idle", idle_setup); |
1934 | |
1935 | void show_regs(struct pt_regs * regs) |
1936 | { |
1937 | @@ -282,7 +289,7 @@ |
1938 | |
1939 | regs.xds = __USER_DS; |
1940 | regs.xes = __USER_DS; |
1941 | - regs.xfs = __KERNEL_PDA; |
1942 | + regs.xfs = __KERNEL_PERCPU; |
1943 | regs.orig_eax = -1; |
1944 | regs.eip = (unsigned long) kernel_thread_helper; |
1945 | regs.xcs = __KERNEL_CS | get_kernel_rpl(); |
1946 | @@ -556,7 +563,7 @@ |
1947 | * multicall to indicate FPU task switch, rather than |
1948 | * synchronously trapping to Xen. |
1949 | */ |
1950 | - if (prev_p->thread_info->status & TS_USEDFPU) { |
1951 | + if (task_thread_info(prev_p)->status & TS_USEDFPU) { |
1952 | __save_init_fpu(prev_p); /* _not_ save_init_fpu() */ |
1953 | mcl->op = __HYPERVISOR_fpu_taskswitch; |
1954 | mcl->args[0] = 1; |
1955 | @@ -648,7 +655,7 @@ |
1956 | if (prev->gs | next->gs) |
1957 | loadsegment(gs, next->gs); |
1958 | |
1959 | - write_pda(pcurrent, next_p); |
1960 | + x86_write_percpu(current_task, next_p); |
1961 | |
1962 | return prev_p; |
1963 | } |
1964 | --- a/arch/x86/kernel/process_64-xen.c |
1965 | +++ b/arch/x86/kernel/process_64-xen.c |
1966 | @@ -39,6 +39,7 @@ |
1967 | #include <linux/random.h> |
1968 | #include <linux/notifier.h> |
1969 | #include <linux/kprobes.h> |
1970 | +#include <linux/kdebug.h> |
1971 | |
1972 | #include <asm/uaccess.h> |
1973 | #include <asm/pgtable.h> |
1974 | @@ -49,7 +50,6 @@ |
1975 | #include <asm/mmu_context.h> |
1976 | #include <asm/pda.h> |
1977 | #include <asm/prctl.h> |
1978 | -#include <asm/kdebug.h> |
1979 | #include <xen/interface/platform.h> |
1980 | #include <xen/interface/physdev.h> |
1981 | #include <xen/interface/vcpu.h> |
1982 | @@ -232,16 +232,18 @@ |
1983 | |
1984 | static int __init idle_setup (char *str) |
1985 | { |
1986 | - if (!strncmp(str, "poll", 4)) { |
1987 | + if (!strcmp(str, "poll")) { |
1988 | printk("using polling idle threads.\n"); |
1989 | pm_idle = poll_idle; |
1990 | - } |
1991 | + } else if (!strcmp(str, "mwait")) |
1992 | + force_mwait = 1; |
1993 | + else |
1994 | + return -1; |
1995 | |
1996 | boot_option_idle_override = 1; |
1997 | - return 1; |
1998 | + return 0; |
1999 | } |
2000 | - |
2001 | -__setup("idle=", idle_setup); |
2002 | +early_param("idle", idle_setup); |
2003 | |
2004 | /* Prints also some state that isn't saved in the pt_regs */ |
2005 | void __show_regs(struct pt_regs * regs) |
2006 | @@ -540,7 +542,7 @@ |
2007 | * The AMD workaround requires it to be after DS reload, or |
2008 | * after DS has been cleared, which we do in __prepare_arch_switch. |
2009 | */ |
2010 | - if (prev_p->thread_info->status & TS_USEDFPU) { |
2011 | + if (task_thread_info(prev_p)->status & TS_USEDFPU) { |
2012 | __save_init_fpu(prev_p); /* _not_ save_init_fpu() */ |
2013 | mcl->op = __HYPERVISOR_fpu_taskswitch; |
2014 | mcl->args[0] = 1; |
2015 | --- a/arch/x86/kernel/quirks-xen.c |
2016 | +++ b/arch/x86/kernel/quirks-xen.c |
2017 | @@ -3,12 +3,10 @@ |
2018 | */ |
2019 | #include <linux/pci.h> |
2020 | #include <linux/irq.h> |
2021 | -#include <asm/pci-direct.h> |
2022 | -#include <asm/genapic.h> |
2023 | -#include <asm/cpu.h> |
2024 | |
2025 | #if defined(CONFIG_X86_IO_APIC) && (defined(CONFIG_SMP) || defined(CONFIG_XEN)) && defined(CONFIG_PCI) |
2026 | -static void __devinit verify_quirk_intel_irqbalance(struct pci_dev *dev) |
2027 | + |
2028 | +static void __devinit quirk_intel_irqbalance(struct pci_dev *dev) |
2029 | { |
2030 | u8 config, rev; |
2031 | u32 word; |
2032 | @@ -16,7 +14,7 @@ |
2033 | /* BIOS may enable hardware IRQ balancing for |
2034 | * E7520/E7320/E7525(revision ID 0x9 and below) |
2035 | * based platforms. |
2036 | - * For those platforms, make sure that the genapic is set to 'flat' |
2037 | + * Disable SW irqbalance/affinity on those platforms. |
2038 | */ |
2039 | pci_read_config_byte(dev, PCI_CLASS_REVISION, &rev); |
2040 | if (rev > 0x9) |
2041 | @@ -30,59 +28,20 @@ |
2042 | raw_pci_ops->read(0, 0, 0x40, 0x4c, 2, &word); |
2043 | |
2044 | if (!(word & (1 << 13))) { |
2045 | -#ifndef CONFIG_XEN |
2046 | -#ifdef CONFIG_X86_64 |
2047 | - if (genapic != &apic_flat) |
2048 | - panic("APIC mode must be flat on this system\n"); |
2049 | -#elif defined(CONFIG_X86_GENERICARCH) |
2050 | - if (genapic != &apic_default) |
2051 | - panic("APIC mode must be default(flat) on this system. Use apic=default\n"); |
2052 | -#endif |
2053 | -#endif |
2054 | - } |
2055 | - |
2056 | - /* put back the original value for config space*/ |
2057 | - if (!(config & 0x2)) |
2058 | - pci_write_config_byte(dev, 0xf4, config); |
2059 | -} |
2060 | - |
2061 | -void __init quirk_intel_irqbalance(void) |
2062 | -{ |
2063 | - u8 config, rev; |
2064 | - u32 word; |
2065 | - |
2066 | - /* BIOS may enable hardware IRQ balancing for |
2067 | - * E7520/E7320/E7525(revision ID 0x9 and below) |
2068 | - * based platforms. |
2069 | - * Disable SW irqbalance/affinity on those platforms. |
2070 | - */ |
2071 | - rev = read_pci_config_byte(0, 0, 0, PCI_CLASS_REVISION); |
2072 | - if (rev > 0x9) |
2073 | - return; |
2074 | - |
2075 | - printk(KERN_INFO "Intel E7520/7320/7525 detected."); |
2076 | - |
2077 | - /* enable access to config space */ |
2078 | - config = read_pci_config_byte(0, 0, 0, 0xf4); |
2079 | - write_pci_config_byte(0, 0, 0, 0xf4, config|0x2); |
2080 | - |
2081 | - /* read xTPR register */ |
2082 | - word = read_pci_config_16(0, 0, 0x40, 0x4c); |
2083 | - |
2084 | - if (!(word & (1 << 13))) { |
2085 | struct xen_platform_op op; |
2086 | - printk(KERN_INFO "Disabling irq balancing and affinity\n"); |
2087 | + |
2088 | + printk(KERN_INFO "Intel E7520/7320/7525 detected. " |
2089 | + "Disabling irq balancing and affinity\n"); |
2090 | op.cmd = XENPF_platform_quirk; |
2091 | op.u.platform_quirk.quirk_id = QUIRK_NOIRQBALANCING; |
2092 | WARN_ON(HYPERVISOR_platform_op(&op)); |
2093 | } |
2094 | |
2095 | - /* put back the original value for config space */ |
2096 | + /* put back the original value for config space*/ |
2097 | if (!(config & 0x2)) |
2098 | - write_pci_config_byte(0, 0, 0, 0xf4, config); |
2099 | + pci_write_config_byte(dev, 0xf4, config); |
2100 | } |
2101 | -DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7320_MCH, verify_quirk_intel_irqbalance); |
2102 | -DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7525_MCH, verify_quirk_intel_irqbalance); |
2103 | -DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7520_MCH, verify_quirk_intel_irqbalance); |
2104 | - |
2105 | +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7320_MCH, quirk_intel_irqbalance); |
2106 | +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7525_MCH, quirk_intel_irqbalance); |
2107 | +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_E7520_MCH, quirk_intel_irqbalance); |
2108 | #endif |
2109 | --- a/arch/x86/kernel/setup64-xen.c |
2110 | +++ b/arch/x86/kernel/setup64-xen.c |
2111 | @@ -113,9 +113,9 @@ |
2112 | if (!NODE_DATA(cpu_to_node(i))) { |
2113 | printk("cpu with no node %d, num_online_nodes %d\n", |
2114 | i, num_online_nodes()); |
2115 | - ptr = alloc_bootmem(size); |
2116 | + ptr = alloc_bootmem_pages(size); |
2117 | } else { |
2118 | - ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size); |
2119 | + ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size); |
2120 | } |
2121 | if (!ptr) |
2122 | panic("Cannot allocate cpu data for CPU %d\n", i); |
2123 | @@ -208,6 +208,8 @@ |
2124 | __attribute__((section(".bss.page_aligned"))); |
2125 | #endif |
2126 | |
2127 | +extern asmlinkage void ignore_sysret(void); |
2128 | + |
2129 | /* May not be marked __init: used by software suspend */ |
2130 | void syscall_init(void) |
2131 | { |
2132 | @@ -219,12 +221,22 @@ |
2133 | */ |
2134 | wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32); |
2135 | wrmsrl(MSR_LSTAR, system_call); |
2136 | + wrmsrl(MSR_CSTAR, ignore_sysret); |
2137 | |
2138 | /* Flags to clear on syscall */ |
2139 | wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000); |
2140 | #endif |
2141 | #ifdef CONFIG_IA32_EMULATION |
2142 | syscall32_cpu_init (); |
2143 | +#else |
2144 | + { |
2145 | + static const struct callback_register cstar = { |
2146 | + .type = CALLBACKTYPE_syscall32, |
2147 | + .address = (unsigned long)ignore_sysret |
2148 | + }; |
2149 | + if (HYPERVISOR_callback_op(CALLBACKOP_register, &cstar)) |
2150 | + printk(KERN_WARN "Unable to register CSTAR callback\n"); |
2151 | + } |
2152 | #endif |
2153 | } |
2154 | |
2155 | @@ -262,7 +274,6 @@ |
2156 | /* CPU 0 is initialised in head64.c */ |
2157 | if (cpu != 0) { |
2158 | pda_init(cpu); |
2159 | - zap_low_mappings(cpu); |
2160 | } |
2161 | #ifndef CONFIG_X86_NO_TSS |
2162 | else |
2163 | --- a/arch/x86/kernel/setup_64-xen.c |
2164 | +++ b/arch/x86/kernel/setup_64-xen.c |
2165 | @@ -123,6 +123,8 @@ |
2166 | |
2167 | unsigned long saved_video_mode; |
2168 | |
2169 | +int force_mwait __cpuinitdata; |
2170 | + |
2171 | /* |
2172 | * Early DMI memory |
2173 | */ |
2174 | @@ -256,10 +258,10 @@ |
2175 | * there is a real-mode segmented pointer pointing to the |
2176 | * 4K EBDA area at 0x40E |
2177 | */ |
2178 | - ebda_addr = *(unsigned short *)EBDA_ADDR_POINTER; |
2179 | + ebda_addr = *(unsigned short *)__va(EBDA_ADDR_POINTER); |
2180 | ebda_addr <<= 4; |
2181 | |
2182 | - ebda_size = *(unsigned short *)(unsigned long)ebda_addr; |
2183 | + ebda_size = *(unsigned short *)__va(ebda_addr); |
2184 | |
2185 | /* Round EBDA up to pages */ |
2186 | if (ebda_size == 0) |
2187 | @@ -413,15 +415,8 @@ |
2188 | #endif |
2189 | |
2190 | #ifdef CONFIG_SMP |
2191 | - /* |
2192 | - * But first pinch a few for the stack/trampoline stuff |
2193 | - * FIXME: Don't need the extra page at 4K, but need to fix |
2194 | - * trampoline before removing it. (see the GDT stuff) |
2195 | - */ |
2196 | - reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE); |
2197 | - |
2198 | /* Reserve SMP trampoline */ |
2199 | - reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE); |
2200 | + reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, 2*PAGE_SIZE); |
2201 | #endif |
2202 | #endif |
2203 | |
2204 | @@ -573,8 +568,6 @@ |
2205 | early_quirks(); |
2206 | #endif |
2207 | |
2208 | - zap_low_mappings(0); |
2209 | - |
2210 | /* |
2211 | * set this early, so we dont allocate cpu0 |
2212 | * if MADT list doesnt list BSP first |
2213 | @@ -877,6 +870,10 @@ |
2214 | |
2215 | /* RDTSC can be speculated around */ |
2216 | clear_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability); |
2217 | + |
2218 | + /* Family 10 doesn't support C states in MWAIT so don't use it */ |
2219 | + if (c->x86 == 0x10 && !force_mwait) |
2220 | + clear_bit(X86_FEATURE_MWAIT, &c->x86_capability); |
2221 | } |
2222 | |
2223 | static void __cpuinit detect_ht(struct cpuinfo_x86 *c) |
2224 | @@ -1159,9 +1156,7 @@ |
2225 | #ifdef CONFIG_X86_MCE |
2226 | mcheck_init(c); |
2227 | #endif |
2228 | - if (c == &boot_cpu_data) |
2229 | - mtrr_bp_init(); |
2230 | - else |
2231 | + if (c != &boot_cpu_data) |
2232 | mtrr_ap_init(); |
2233 | #ifdef CONFIG_NUMA |
2234 | numa_add_cpu(smp_processor_id()); |
2235 | @@ -1252,9 +1247,8 @@ |
2236 | "stc", |
2237 | "100mhzsteps", |
2238 | "hwpstate", |
2239 | - NULL, /* tsc invariant mapped to constant_tsc */ |
2240 | - NULL, |
2241 | - /* nothing */ /* constant_tsc - moved to flags */ |
2242 | + "", /* tsc invariant mapped to constant_tsc */ |
2243 | + /* nothing */ |
2244 | }; |
2245 | |
2246 | |
2247 | --- a/arch/x86/kernel/smp_32-xen.c |
2248 | +++ b/arch/x86/kernel/smp_32-xen.c |
2249 | @@ -13,7 +13,6 @@ |
2250 | #include <linux/mm.h> |
2251 | #include <linux/delay.h> |
2252 | #include <linux/spinlock.h> |
2253 | -#include <linux/smp_lock.h> |
2254 | #include <linux/kernel_stat.h> |
2255 | #include <linux/mc146818rtc.h> |
2256 | #include <linux/cache.h> |
2257 | @@ -216,7 +215,6 @@ |
2258 | static struct mm_struct * flush_mm; |
2259 | static unsigned long flush_va; |
2260 | static DEFINE_SPINLOCK(tlbstate_lock); |
2261 | -#define FLUSH_ALL 0xffffffff |
2262 | |
2263 | /* |
2264 | * We cannot call mmdrop() because we are in interrupt context, |
2265 | @@ -298,7 +296,7 @@ |
2266 | |
2267 | if (flush_mm == per_cpu(cpu_tlbstate, cpu).active_mm) { |
2268 | if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK) { |
2269 | - if (flush_va == FLUSH_ALL) |
2270 | + if (flush_va == TLB_FLUSH_ALL) |
2271 | local_flush_tlb(); |
2272 | else |
2273 | __flush_tlb_one(flush_va); |
2274 | @@ -314,9 +312,11 @@ |
2275 | return IRQ_HANDLED; |
2276 | } |
2277 | |
2278 | -static void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm, |
2279 | - unsigned long va) |
2280 | +void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm, |
2281 | + unsigned long va) |
2282 | { |
2283 | + cpumask_t cpumask = *cpumaskp; |
2284 | + |
2285 | /* |
2286 | * A couple of (to be removed) sanity checks: |
2287 | * |
2288 | @@ -327,10 +327,12 @@ |
2289 | BUG_ON(cpu_isset(smp_processor_id(), cpumask)); |
2290 | BUG_ON(!mm); |
2291 | |
2292 | +#ifdef CONFIG_HOTPLUG_CPU |
2293 | /* If a CPU which we ran on has gone down, OK. */ |
2294 | cpus_and(cpumask, cpumask, cpu_online_map); |
2295 | - if (cpus_empty(cpumask)) |
2296 | + if (unlikely(cpus_empty(cpumask))) |
2297 | return; |
2298 | +#endif |
2299 | |
2300 | /* |
2301 | * i'm not happy about this global shared spinlock in the |
2302 | @@ -341,17 +343,7 @@ |
2303 | |
2304 | flush_mm = mm; |
2305 | flush_va = va; |
2306 | -#if NR_CPUS <= BITS_PER_LONG |
2307 | - atomic_set_mask(cpumask, &flush_cpumask); |
2308 | -#else |
2309 | - { |
2310 | - int k; |
2311 | - unsigned long *flush_mask = (unsigned long *)&flush_cpumask; |
2312 | - unsigned long *cpu_mask = (unsigned long *)&cpumask; |
2313 | - for (k = 0; k < BITS_TO_LONGS(NR_CPUS); ++k) |
2314 | - atomic_set_mask(cpu_mask[k], &flush_mask[k]); |
2315 | - } |
2316 | -#endif |
2317 | + cpus_or(flush_cpumask, cpumask, flush_cpumask); |
2318 | /* |
2319 | * We have to send the IPI only to |
2320 | * CPUs affected. |
2321 | @@ -378,7 +370,7 @@ |
2322 | |
2323 | local_flush_tlb(); |
2324 | if (!cpus_empty(cpu_mask)) |
2325 | - flush_tlb_others(cpu_mask, mm, FLUSH_ALL); |
2326 | + flush_tlb_others(cpu_mask, mm, TLB_FLUSH_ALL); |
2327 | preempt_enable(); |
2328 | } |
2329 | |
2330 | @@ -397,7 +389,7 @@ |
2331 | leave_mm(smp_processor_id()); |
2332 | } |
2333 | if (!cpus_empty(cpu_mask)) |
2334 | - flush_tlb_others(cpu_mask, mm, FLUSH_ALL); |
2335 | + flush_tlb_others(cpu_mask, mm, TLB_FLUSH_ALL); |
2336 | |
2337 | preempt_enable(); |
2338 | } |
2339 | @@ -446,7 +438,7 @@ |
2340 | * it goes straight through and wastes no time serializing |
2341 | * anything. Worst case is that we lose a reschedule ... |
2342 | */ |
2343 | -void smp_send_reschedule(int cpu) |
2344 | +void xen_smp_send_reschedule(int cpu) |
2345 | { |
2346 | WARN_ON(cpu_is_offline(cpu)); |
2347 | send_IPI_mask(cpumask_of_cpu(cpu), RESCHEDULE_VECTOR); |
2348 | @@ -478,36 +470,79 @@ |
2349 | |
2350 | static struct call_data_struct *call_data; |
2351 | |
2352 | +static void __smp_call_function(void (*func) (void *info), void *info, |
2353 | + int nonatomic, int wait) |
2354 | +{ |
2355 | + struct call_data_struct data; |
2356 | + int cpus = num_online_cpus() - 1; |
2357 | + |
2358 | + if (!cpus) |
2359 | + return; |
2360 | + |
2361 | + data.func = func; |
2362 | + data.info = info; |
2363 | + atomic_set(&data.started, 0); |
2364 | + data.wait = wait; |
2365 | + if (wait) |
2366 | + atomic_set(&data.finished, 0); |
2367 | + |
2368 | + call_data = &data; |
2369 | + mb(); |
2370 | + |
2371 | + /* Send a message to all other CPUs and wait for them to respond */ |
2372 | + send_IPI_allbutself(CALL_FUNCTION_VECTOR); |
2373 | + |
2374 | + /* Wait for response */ |
2375 | + while (atomic_read(&data.started) != cpus) |
2376 | + cpu_relax(); |
2377 | + |
2378 | + if (wait) |
2379 | + while (atomic_read(&data.finished) != cpus) |
2380 | + cpu_relax(); |
2381 | +} |
2382 | + |
2383 | + |
2384 | /** |
2385 | - * smp_call_function(): Run a function on all other CPUs. |
2386 | + * smp_call_function_mask(): Run a function on a set of other CPUs. |
2387 | + * @mask: The set of cpus to run on. Must not include the current cpu. |
2388 | * @func: The function to run. This must be fast and non-blocking. |
2389 | * @info: An arbitrary pointer to pass to the function. |
2390 | - * @nonatomic: currently unused. |
2391 | * @wait: If true, wait (atomically) until function has completed on other CPUs. |
2392 | * |
2393 | - * Returns 0 on success, else a negative status code. Does not return until |
2394 | - * remote CPUs are nearly ready to execute <<func>> or are or have executed. |
2395 | + * Returns 0 on success, else a negative status code. |
2396 | + * |
2397 | + * If @wait is true, then returns once @func has returned; otherwise |
2398 | + * it returns just before the target cpu calls @func. |
2399 | * |
2400 | * You must not call this function with disabled interrupts or from a |
2401 | * hardware interrupt handler or from a bottom half handler. |
2402 | */ |
2403 | -int smp_call_function (void (*func) (void *info), void *info, int nonatomic, |
2404 | - int wait) |
2405 | +int |
2406 | +xen_smp_call_function_mask(cpumask_t mask, |
2407 | + void (*func)(void *), void *info, |
2408 | + int wait) |
2409 | { |
2410 | struct call_data_struct data; |
2411 | + cpumask_t allbutself; |
2412 | int cpus; |
2413 | |
2414 | + /* Can deadlock when called with interrupts disabled */ |
2415 | + WARN_ON(irqs_disabled()); |
2416 | + |
2417 | /* Holding any lock stops cpus from going down. */ |
2418 | spin_lock(&call_lock); |
2419 | - cpus = num_online_cpus() - 1; |
2420 | + |
2421 | + allbutself = cpu_online_map; |
2422 | + cpu_clear(smp_processor_id(), allbutself); |
2423 | + |
2424 | + cpus_and(mask, mask, allbutself); |
2425 | + cpus = cpus_weight(mask); |
2426 | + |
2427 | if (!cpus) { |
2428 | spin_unlock(&call_lock); |
2429 | return 0; |
2430 | } |
2431 | |
2432 | - /* Can deadlock when called with interrupts disabled */ |
2433 | - WARN_ON(irqs_disabled()); |
2434 | - |
2435 | data.func = func; |
2436 | data.info = info; |
2437 | atomic_set(&data.started, 0); |
2438 | @@ -517,9 +552,12 @@ |
2439 | |
2440 | call_data = &data; |
2441 | mb(); |
2442 | - |
2443 | - /* Send a message to all other CPUs and wait for them to respond */ |
2444 | - send_IPI_allbutself(CALL_FUNCTION_VECTOR); |
2445 | + |
2446 | + /* Send a message to other CPUs */ |
2447 | + if (cpus_equal(mask, allbutself)) |
2448 | + send_IPI_allbutself(CALL_FUNCTION_VECTOR); |
2449 | + else |
2450 | + send_IPI_mask(mask, CALL_FUNCTION_VECTOR); |
2451 | |
2452 | /* Wait for response */ |
2453 | while (atomic_read(&data.started) != cpus) |
2454 | @@ -532,15 +570,14 @@ |
2455 | |
2456 | return 0; |
2457 | } |
2458 | -EXPORT_SYMBOL(smp_call_function); |
2459 | |
2460 | static void stop_this_cpu (void * dummy) |
2461 | { |
2462 | + local_irq_disable(); |
2463 | /* |
2464 | * Remove this CPU: |
2465 | */ |
2466 | cpu_clear(smp_processor_id(), cpu_online_map); |
2467 | - local_irq_disable(); |
2468 | disable_all_local_evtchn(); |
2469 | if (cpu_data[smp_processor_id()].hlt_works_ok) |
2470 | for(;;) halt(); |
2471 | @@ -551,13 +588,18 @@ |
2472 | * this function calls the 'stop' function on all other CPUs in the system. |
2473 | */ |
2474 | |
2475 | -void smp_send_stop(void) |
2476 | +void xen_smp_send_stop(void) |
2477 | { |
2478 | - smp_call_function(stop_this_cpu, NULL, 1, 0); |
2479 | + /* Don't deadlock on the call lock in panic */ |
2480 | + int nolock = !spin_trylock(&call_lock); |
2481 | + unsigned long flags; |
2482 | |
2483 | - local_irq_disable(); |
2484 | + local_irq_save(flags); |
2485 | + __smp_call_function(stop_this_cpu, NULL, 0, 0); |
2486 | + if (!nolock) |
2487 | + spin_unlock(&call_lock); |
2488 | disable_all_local_evtchn(); |
2489 | - local_irq_enable(); |
2490 | + local_irq_restore(flags); |
2491 | } |
2492 | |
2493 | /* |
2494 | @@ -598,74 +640,3 @@ |
2495 | |
2496 | return IRQ_HANDLED; |
2497 | } |
2498 | - |
2499 | -/* |
2500 | - * this function sends a 'generic call function' IPI to one other CPU |
2501 | - * in the system. |
2502 | - * |
2503 | - * cpu is a standard Linux logical CPU number. |
2504 | - */ |
2505 | -static void |
2506 | -__smp_call_function_single(int cpu, void (*func) (void *info), void *info, |
2507 | - int nonatomic, int wait) |
2508 | -{ |
2509 | - struct call_data_struct data; |
2510 | - int cpus = 1; |
2511 | - |
2512 | - data.func = func; |
2513 | - data.info = info; |
2514 | - atomic_set(&data.started, 0); |
2515 | - data.wait = wait; |
2516 | - if (wait) |
2517 | - atomic_set(&data.finished, 0); |
2518 | - |
2519 | - call_data = &data; |
2520 | - wmb(); |
2521 | - /* Send a message to all other CPUs and wait for them to respond */ |
2522 | - send_IPI_mask(cpumask_of_cpu(cpu), CALL_FUNCTION_VECTOR); |
2523 | - |
2524 | - /* Wait for response */ |
2525 | - while (atomic_read(&data.started) != cpus) |
2526 | - cpu_relax(); |
2527 | - |
2528 | - if (!wait) |
2529 | - return; |
2530 | - |
2531 | - while (atomic_read(&data.finished) != cpus) |
2532 | - cpu_relax(); |
2533 | -} |
2534 | - |
2535 | -/* |
2536 | - * smp_call_function_single - Run a function on another CPU |
2537 | - * @func: The function to run. This must be fast and non-blocking. |
2538 | - * @info: An arbitrary pointer to pass to the function. |
2539 | - * @nonatomic: Currently unused. |
2540 | - * @wait: If true, wait until function has completed on other CPUs. |
2541 | - * |
2542 | - * Retrurns 0 on success, else a negative status code. |
2543 | - * |
2544 | - * Does not return until the remote CPU is nearly ready to execute <func> |
2545 | - * or is or has executed. |
2546 | - */ |
2547 | - |
2548 | -int smp_call_function_single(int cpu, void (*func) (void *info), void *info, |
2549 | - int nonatomic, int wait) |
2550 | -{ |
2551 | - /* prevent preemption and reschedule on another processor */ |
2552 | - int me = get_cpu(); |
2553 | - if (cpu == me) { |
2554 | - WARN_ON(1); |
2555 | - put_cpu(); |
2556 | - return -EBUSY; |
2557 | - } |
2558 | - |
2559 | - /* Can deadlock when called with interrupts disabled */ |
2560 | - WARN_ON(irqs_disabled()); |
2561 | - |
2562 | - spin_lock_bh(&call_lock); |
2563 | - __smp_call_function_single(cpu, func, info, nonatomic, wait); |
2564 | - spin_unlock_bh(&call_lock); |
2565 | - put_cpu(); |
2566 | - return 0; |
2567 | -} |
2568 | -EXPORT_SYMBOL(smp_call_function_single); |
2569 | --- a/arch/x86/kernel/smp_64-xen.c |
2570 | +++ b/arch/x86/kernel/smp_64-xen.c |
2571 | @@ -14,7 +14,6 @@ |
2572 | #include <linux/mm.h> |
2573 | #include <linux/delay.h> |
2574 | #include <linux/spinlock.h> |
2575 | -#include <linux/smp_lock.h> |
2576 | #include <linux/smp.h> |
2577 | #include <linux/kernel_stat.h> |
2578 | #include <linux/mc146818rtc.h> |
2579 | @@ -457,44 +456,36 @@ |
2580 | } |
2581 | EXPORT_SYMBOL(smp_call_function); |
2582 | |
2583 | -void smp_stop_cpu(void) |
2584 | +static void stop_this_cpu(void *dummy) |
2585 | { |
2586 | - unsigned long flags; |
2587 | + local_irq_disable(); |
2588 | /* |
2589 | * Remove this CPU: |
2590 | */ |
2591 | cpu_clear(smp_processor_id(), cpu_online_map); |
2592 | - local_irq_save(flags); |
2593 | disable_all_local_evtchn(); |
2594 | - local_irq_restore(flags); |
2595 | -} |
2596 | - |
2597 | -static void smp_really_stop_cpu(void *dummy) |
2598 | -{ |
2599 | - smp_stop_cpu(); |
2600 | for (;;) |
2601 | halt(); |
2602 | } |
2603 | |
2604 | void smp_send_stop(void) |
2605 | { |
2606 | - int nolock = 0; |
2607 | + int nolock; |
2608 | + unsigned long flags; |
2609 | + |
2610 | #ifndef CONFIG_XEN |
2611 | if (reboot_force) |
2612 | return; |
2613 | #endif |
2614 | + |
2615 | /* Don't deadlock on the call lock in panic */ |
2616 | - if (!spin_trylock(&call_lock)) { |
2617 | - /* ignore locking because we have panicked anyways */ |
2618 | - nolock = 1; |
2619 | - } |
2620 | - __smp_call_function(smp_really_stop_cpu, NULL, 0, 0); |
2621 | + nolock = !spin_trylock(&call_lock); |
2622 | + local_irq_save(flags); |
2623 | + __smp_call_function(stop_this_cpu, NULL, 0, 0); |
2624 | if (!nolock) |
2625 | spin_unlock(&call_lock); |
2626 | - |
2627 | - local_irq_disable(); |
2628 | disable_all_local_evtchn(); |
2629 | - local_irq_enable(); |
2630 | + local_irq_restore(flags); |
2631 | } |
2632 | |
2633 | /* |
2634 | --- a/arch/x86/kernel/time_32-xen.c |
2635 | +++ b/arch/x86/kernel/time_32-xen.c |
2636 | @@ -80,7 +80,6 @@ |
2637 | #include <asm/i8253.h> |
2638 | DEFINE_SPINLOCK(i8253_lock); |
2639 | EXPORT_SYMBOL(i8253_lock); |
2640 | -int pit_latch_buggy; /* extern */ |
2641 | #else |
2642 | volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES; |
2643 | #endif |
2644 | @@ -589,7 +588,7 @@ |
2645 | return IRQ_HANDLED; |
2646 | } |
2647 | |
2648 | -void mark_tsc_unstable(void) |
2649 | +void mark_tsc_unstable(char *reason) |
2650 | { |
2651 | #ifndef CONFIG_XEN /* XXX Should tell the hypervisor about this fact. */ |
2652 | tsc_unstable = 1; |
2653 | @@ -597,17 +596,18 @@ |
2654 | } |
2655 | EXPORT_SYMBOL_GPL(mark_tsc_unstable); |
2656 | |
2657 | +static cycle_t cs_last; |
2658 | + |
2659 | static cycle_t xen_clocksource_read(void) |
2660 | { |
2661 | cycle_t ret = sched_clock(); |
2662 | |
2663 | #ifdef CONFIG_SMP |
2664 | for (;;) { |
2665 | - static cycle_t last_ret; |
2666 | #ifndef CONFIG_64BIT |
2667 | - cycle_t last = cmpxchg64(&last_ret, 0, 0); |
2668 | + cycle_t last = cmpxchg64(&cs_last, 0, 0); |
2669 | #else |
2670 | - cycle_t last = last_ret; |
2671 | + cycle_t last = cs_last; |
2672 | #define cmpxchg64 cmpxchg |
2673 | #endif |
2674 | |
2675 | @@ -627,7 +627,7 @@ |
2676 | } |
2677 | ret = last; |
2678 | } |
2679 | - if (cmpxchg64(&last_ret, last, ret) == last) |
2680 | + if (cmpxchg64(&cs_last, last, ret) == last) |
2681 | break; |
2682 | } |
2683 | #endif |
2684 | @@ -635,6 +635,14 @@ |
2685 | return ret; |
2686 | } |
2687 | |
2688 | +static void xen_clocksource_resume(void) |
2689 | +{ |
2690 | + extern void time_resume(void); |
2691 | + |
2692 | + time_resume(); |
2693 | + cs_last = sched_clock(); |
2694 | +} |
2695 | + |
2696 | static struct clocksource clocksource_xen = { |
2697 | .name = "xen", |
2698 | .rating = 400, |
2699 | @@ -643,6 +651,7 @@ |
2700 | .mult = 1 << XEN_SHIFT, /* time directly in nanoseconds */ |
2701 | .shift = XEN_SHIFT, |
2702 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
2703 | + .resume = xen_clocksource_resume, |
2704 | }; |
2705 | |
2706 | static void init_missing_ticks_accounting(unsigned int cpu) |
2707 | @@ -731,35 +740,6 @@ |
2708 | mod_timer(&sync_xen_wallclock_timer, jiffies + 1); |
2709 | } |
2710 | |
2711 | -static int timer_resume(struct sys_device *dev) |
2712 | -{ |
2713 | - extern void time_resume(void); |
2714 | - time_resume(); |
2715 | - return 0; |
2716 | -} |
2717 | - |
2718 | -static struct sysdev_class timer_sysclass = { |
2719 | - .resume = timer_resume, |
2720 | - set_kset_name("timer"), |
2721 | -}; |
2722 | - |
2723 | - |
2724 | -/* XXX this driverfs stuff should probably go elsewhere later -john */ |
2725 | -static struct sys_device device_timer = { |
2726 | - .id = 0, |
2727 | - .cls = &timer_sysclass, |
2728 | -}; |
2729 | - |
2730 | -static int time_init_device(void) |
2731 | -{ |
2732 | - int error = sysdev_class_register(&timer_sysclass); |
2733 | - if (!error) |
2734 | - error = sysdev_register(&device_timer); |
2735 | - return error; |
2736 | -} |
2737 | - |
2738 | -device_initcall(time_init_device); |
2739 | - |
2740 | extern void (*late_time_init)(void); |
2741 | |
2742 | /* Dynamically-mapped IRQ. */ |
2743 | @@ -772,7 +752,7 @@ |
2744 | VIRQ_TIMER, |
2745 | 0, |
2746 | timer_interrupt, |
2747 | - SA_INTERRUPT, |
2748 | + IRQF_DISABLED, |
2749 | "timer0", |
2750 | NULL); |
2751 | BUG_ON(per_cpu(timer_irq, 0) < 0); |
2752 | @@ -890,21 +870,21 @@ |
2753 | cpu_clear(smp_processor_id(), nohz_cpu_mask); |
2754 | } |
2755 | |
2756 | -void raw_safe_halt(void) |
2757 | +void xen_safe_halt(void) |
2758 | { |
2759 | stop_hz_timer(); |
2760 | /* Blocking includes an implicit local_irq_enable(). */ |
2761 | HYPERVISOR_block(); |
2762 | start_hz_timer(); |
2763 | } |
2764 | -EXPORT_SYMBOL(raw_safe_halt); |
2765 | +EXPORT_SYMBOL(xen_safe_halt); |
2766 | |
2767 | -void halt(void) |
2768 | +void xen_halt(void) |
2769 | { |
2770 | if (irqs_disabled()) |
2771 | VOID(HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL)); |
2772 | } |
2773 | -EXPORT_SYMBOL(halt); |
2774 | +EXPORT_SYMBOL(xen_halt); |
2775 | |
2776 | /* No locking required. Interrupts are disabled on all CPUs. */ |
2777 | void time_resume(void) |
2778 | @@ -967,7 +947,7 @@ |
2779 | irq = bind_virq_to_irqhandler(VIRQ_TIMER, |
2780 | cpu, |
2781 | timer_interrupt, |
2782 | - SA_INTERRUPT, |
2783 | + IRQF_DISABLED, |
2784 | timer_name[cpu], |
2785 | NULL); |
2786 | if (irq < 0) |
2787 | --- a/arch/x86/kernel/traps_32-xen.c |
2788 | +++ b/arch/x86/kernel/traps_32-xen.c |
2789 | @@ -52,7 +52,7 @@ |
2790 | #include <asm/unwind.h> |
2791 | #include <asm/smp.h> |
2792 | #include <asm/arch_hooks.h> |
2793 | -#include <asm/kdebug.h> |
2794 | +#include <linux/kdebug.h> |
2795 | #include <asm/stacktrace.h> |
2796 | |
2797 | #include <linux/module.h> |
2798 | @@ -101,20 +101,6 @@ |
2799 | |
2800 | int kstack_depth_to_print = 24; |
2801 | static unsigned int code_bytes = 64; |
2802 | -ATOMIC_NOTIFIER_HEAD(i386die_chain); |
2803 | - |
2804 | -int register_die_notifier(struct notifier_block *nb) |
2805 | -{ |
2806 | - vmalloc_sync_all(); |
2807 | - return atomic_notifier_chain_register(&i386die_chain, nb); |
2808 | -} |
2809 | -EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */ |
2810 | - |
2811 | -int unregister_die_notifier(struct notifier_block *nb) |
2812 | -{ |
2813 | - return atomic_notifier_chain_unregister(&i386die_chain, nb); |
2814 | -} |
2815 | -EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */ |
2816 | |
2817 | static inline int valid_stack_ptr(struct thread_info *tinfo, void *p) |
2818 | { |
2819 | @@ -325,7 +311,7 @@ |
2820 | regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, ss); |
2821 | printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)", |
2822 | TASK_COMM_LEN, current->comm, current->pid, |
2823 | - current_thread_info(), current, current->thread_info); |
2824 | + current_thread_info(), current, task_thread_info(current)); |
2825 | /* |
2826 | * When in-kernel, we also print out the stack and code at the |
2827 | * time of the fault.. |
2828 | @@ -482,8 +468,6 @@ |
2829 | siginfo_t *info) |
2830 | { |
2831 | struct task_struct *tsk = current; |
2832 | - tsk->thread.error_code = error_code; |
2833 | - tsk->thread.trap_no = trapnr; |
2834 | |
2835 | if (regs->eflags & VM_MASK) { |
2836 | if (vm86) |
2837 | @@ -495,6 +479,18 @@ |
2838 | goto kernel_trap; |
2839 | |
2840 | trap_signal: { |
2841 | + /* |
2842 | + * We want error_code and trap_no set for userspace faults and |
2843 | + * kernelspace faults which result in die(), but not |
2844 | + * kernelspace faults which are fixed up. die() gives the |
2845 | + * process no chance to handle the signal and notice the |
2846 | + * kernel fault information, so that won't result in polluting |
2847 | + * the information about previously queued, but not yet |
2848 | + * delivered, faults. See also do_general_protection below. |
2849 | + */ |
2850 | + tsk->thread.error_code = error_code; |
2851 | + tsk->thread.trap_no = trapnr; |
2852 | + |
2853 | if (info) |
2854 | force_sig_info(signr, info, tsk); |
2855 | else |
2856 | @@ -503,8 +499,11 @@ |
2857 | } |
2858 | |
2859 | kernel_trap: { |
2860 | - if (!fixup_exception(regs)) |
2861 | + if (!fixup_exception(regs)) { |
2862 | + tsk->thread.error_code = error_code; |
2863 | + tsk->thread.trap_no = trapnr; |
2864 | die(str, regs, error_code); |
2865 | + } |
2866 | return; |
2867 | } |
2868 | |
2869 | @@ -578,9 +577,6 @@ |
2870 | fastcall void __kprobes do_general_protection(struct pt_regs * regs, |
2871 | long error_code) |
2872 | { |
2873 | - current->thread.error_code = error_code; |
2874 | - current->thread.trap_no = 13; |
2875 | - |
2876 | if (regs->eflags & VM_MASK) |
2877 | goto gp_in_vm86; |
2878 | |
2879 | @@ -599,6 +595,8 @@ |
2880 | |
2881 | gp_in_kernel: |
2882 | if (!fixup_exception(regs)) { |
2883 | + current->thread.error_code = error_code; |
2884 | + current->thread.trap_no = 13; |
2885 | if (notify_die(DIE_GPF, "general protection fault", regs, |
2886 | error_code, 13, SIGSEGV) == NOTIFY_STOP) |
2887 | return; |
2888 | @@ -987,9 +985,7 @@ |
2889 | fastcall unsigned long patch_espfix_desc(unsigned long uesp, |
2890 | unsigned long kesp) |
2891 | { |
2892 | - int cpu = smp_processor_id(); |
2893 | - struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
2894 | - struct desc_struct *gdt = (struct desc_struct *)cpu_gdt_descr->address; |
2895 | + struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt; |
2896 | unsigned long base = (kesp - uesp) & -THREAD_SIZE; |
2897 | unsigned long new_kesp = kesp - base; |
2898 | unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT; |
2899 | --- a/arch/x86/kernel/traps_64-xen.c |
2900 | +++ b/arch/x86/kernel/traps_64-xen.c |
2901 | @@ -32,6 +32,7 @@ |
2902 | #include <linux/unwind.h> |
2903 | #include <linux/uaccess.h> |
2904 | #include <linux/bug.h> |
2905 | +#include <linux/kdebug.h> |
2906 | |
2907 | #include <asm/system.h> |
2908 | #include <asm/io.h> |
2909 | @@ -39,7 +40,6 @@ |
2910 | #include <asm/debugreg.h> |
2911 | #include <asm/desc.h> |
2912 | #include <asm/i387.h> |
2913 | -#include <asm/kdebug.h> |
2914 | #include <asm/processor.h> |
2915 | #include <asm/unwind.h> |
2916 | #include <asm/smp.h> |
2917 | @@ -71,22 +71,6 @@ |
2918 | asmlinkage void machine_check(void); |
2919 | asmlinkage void spurious_interrupt_bug(void); |
2920 | |
2921 | -ATOMIC_NOTIFIER_HEAD(die_chain); |
2922 | -EXPORT_SYMBOL(die_chain); |
2923 | - |
2924 | -int register_die_notifier(struct notifier_block *nb) |
2925 | -{ |
2926 | - vmalloc_sync_all(); |
2927 | - return atomic_notifier_chain_register(&die_chain, nb); |
2928 | -} |
2929 | -EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */ |
2930 | - |
2931 | -int unregister_die_notifier(struct notifier_block *nb) |
2932 | -{ |
2933 | - return atomic_notifier_chain_unregister(&die_chain, nb); |
2934 | -} |
2935 | -EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */ |
2936 | - |
2937 | static inline void conditional_sti(struct pt_regs *regs) |
2938 | { |
2939 | if (regs->eflags & X86_EFLAGS_IF) |
2940 | @@ -428,8 +412,7 @@ |
2941 | const int cpu = smp_processor_id(); |
2942 | struct task_struct *cur = cpu_pda(cpu)->pcurrent; |
2943 | |
2944 | - rsp = regs->rsp; |
2945 | - |
2946 | + rsp = regs->rsp; |
2947 | printk("CPU %d ", cpu); |
2948 | __show_regs(regs); |
2949 | printk("Process %s (pid: %d, threadinfo %p, task %p)\n", |
2950 | @@ -440,7 +423,6 @@ |
2951 | * time of the fault.. |
2952 | */ |
2953 | if (in_kernel) { |
2954 | - |
2955 | printk("Stack: "); |
2956 | _show_stack(NULL, regs, (unsigned long*)rsp); |
2957 | |
2958 | @@ -485,13 +467,14 @@ |
2959 | |
2960 | unsigned __kprobes long oops_begin(void) |
2961 | { |
2962 | - int cpu = smp_processor_id(); |
2963 | + int cpu; |
2964 | unsigned long flags; |
2965 | |
2966 | oops_enter(); |
2967 | |
2968 | /* racy, but better than risking deadlock. */ |
2969 | local_irq_save(flags); |
2970 | + cpu = smp_processor_id(); |
2971 | if (!spin_trylock(&die_lock)) { |
2972 | if (cpu == die_owner) |
2973 | /* nested oops. should stop eventually */; |
2974 | @@ -585,10 +568,20 @@ |
2975 | { |
2976 | struct task_struct *tsk = current; |
2977 | |
2978 | - tsk->thread.error_code = error_code; |
2979 | - tsk->thread.trap_no = trapnr; |
2980 | - |
2981 | if (user_mode(regs)) { |
2982 | + /* |
2983 | + * We want error_code and trap_no set for userspace |
2984 | + * faults and kernelspace faults which result in |
2985 | + * die(), but not kernelspace faults which are fixed |
2986 | + * up. die() gives the process no chance to handle |
2987 | + * the signal and notice the kernel fault information, |
2988 | + * so that won't result in polluting the information |
2989 | + * about previously queued, but not yet delivered, |
2990 | + * faults. See also do_general_protection below. |
2991 | + */ |
2992 | + tsk->thread.error_code = error_code; |
2993 | + tsk->thread.trap_no = trapnr; |
2994 | + |
2995 | if (exception_trace && unhandled_signal(tsk, signr)) |
2996 | printk(KERN_INFO |
2997 | "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n", |
2998 | @@ -609,8 +602,11 @@ |
2999 | fixup = search_exception_tables(regs->rip); |
3000 | if (fixup) |
3001 | regs->rip = fixup->fixup; |
3002 | - else |
3003 | + else { |
3004 | + tsk->thread.error_code = error_code; |
3005 | + tsk->thread.trap_no = trapnr; |
3006 | die(str, regs, error_code); |
3007 | + } |
3008 | return; |
3009 | } |
3010 | } |
3011 | @@ -686,10 +682,10 @@ |
3012 | |
3013 | conditional_sti(regs); |
3014 | |
3015 | - tsk->thread.error_code = error_code; |
3016 | - tsk->thread.trap_no = 13; |
3017 | - |
3018 | if (user_mode(regs)) { |
3019 | + tsk->thread.error_code = error_code; |
3020 | + tsk->thread.trap_no = 13; |
3021 | + |
3022 | if (exception_trace && unhandled_signal(tsk, SIGSEGV)) |
3023 | printk(KERN_INFO |
3024 | "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n", |
3025 | @@ -708,6 +704,9 @@ |
3026 | regs->rip = fixup->fixup; |
3027 | return; |
3028 | } |
3029 | + |
3030 | + tsk->thread.error_code = error_code; |
3031 | + tsk->thread.trap_no = 13; |
3032 | if (notify_die(DIE_GPF, "general protection fault", regs, |
3033 | error_code, 13, SIGSEGV) == NOTIFY_STOP) |
3034 | return; |
3035 | --- a/arch/x86/kernel/vsyscall_64-xen.c |
3036 | +++ b/arch/x86/kernel/vsyscall_64-xen.c |
3037 | @@ -45,14 +45,34 @@ |
3038 | |
3039 | #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr))) |
3040 | #define __syscall_clobber "r11","rcx","memory" |
3041 | +#define __pa_vsymbol(x) \ |
3042 | + ({unsigned long v; \ |
3043 | + extern char __vsyscall_0; \ |
3044 | + asm("" : "=r" (v) : "0" (x)); \ |
3045 | + ((v - VSYSCALL_FIRST_PAGE) + __pa_symbol(&__vsyscall_0)); }) |
3046 | |
3047 | +/* |
3048 | + * vsyscall_gtod_data contains data that is : |
3049 | + * - readonly from vsyscalls |
3050 | + * - writen by timer interrupt or systcl (/proc/sys/kernel/vsyscall64) |
3051 | + * Try to keep this structure as small as possible to avoid cache line ping pongs |
3052 | + */ |
3053 | struct vsyscall_gtod_data_t { |
3054 | - seqlock_t lock; |
3055 | - int sysctl_enabled; |
3056 | - struct timeval wall_time_tv; |
3057 | + seqlock_t lock; |
3058 | + |
3059 | + /* open coded 'struct timespec' */ |
3060 | + time_t wall_time_sec; |
3061 | + u32 wall_time_nsec; |
3062 | + |
3063 | + int sysctl_enabled; |
3064 | struct timezone sys_tz; |
3065 | - cycle_t offset_base; |
3066 | - struct clocksource clock; |
3067 | + struct { /* extract of a clocksource struct */ |
3068 | + cycle_t (*vread)(void); |
3069 | + cycle_t cycle_last; |
3070 | + cycle_t mask; |
3071 | + u32 mult; |
3072 | + u32 shift; |
3073 | + } clock; |
3074 | }; |
3075 | int __vgetcpu_mode __section_vgetcpu_mode; |
3076 | |
3077 | @@ -68,9 +88,13 @@ |
3078 | |
3079 | write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags); |
3080 | /* copy vsyscall data */ |
3081 | - vsyscall_gtod_data.clock = *clock; |
3082 | - vsyscall_gtod_data.wall_time_tv.tv_sec = wall_time->tv_sec; |
3083 | - vsyscall_gtod_data.wall_time_tv.tv_usec = wall_time->tv_nsec/1000; |
3084 | + vsyscall_gtod_data.clock.vread = clock->vread; |
3085 | + vsyscall_gtod_data.clock.cycle_last = clock->cycle_last; |
3086 | + vsyscall_gtod_data.clock.mask = clock->mask; |
3087 | + vsyscall_gtod_data.clock.mult = clock->mult; |
3088 | + vsyscall_gtod_data.clock.shift = clock->shift; |
3089 | + vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec; |
3090 | + vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec; |
3091 | vsyscall_gtod_data.sys_tz = sys_tz; |
3092 | write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags); |
3093 | } |
3094 | @@ -105,7 +129,8 @@ |
3095 | static __always_inline void do_vgettimeofday(struct timeval * tv) |
3096 | { |
3097 | cycle_t now, base, mask, cycle_delta; |
3098 | - unsigned long seq, mult, shift, nsec_delta; |
3099 | + unsigned seq; |
3100 | + unsigned long mult, shift, nsec; |
3101 | cycle_t (*vread)(void); |
3102 | do { |
3103 | seq = read_seqbegin(&__vsyscall_gtod_data.lock); |
3104 | @@ -121,21 +146,20 @@ |
3105 | mult = __vsyscall_gtod_data.clock.mult; |
3106 | shift = __vsyscall_gtod_data.clock.shift; |
3107 | |
3108 | - *tv = __vsyscall_gtod_data.wall_time_tv; |
3109 | - |
3110 | + tv->tv_sec = __vsyscall_gtod_data.wall_time_sec; |
3111 | + nsec = __vsyscall_gtod_data.wall_time_nsec; |
3112 | } while (read_seqretry(&__vsyscall_gtod_data.lock, seq)); |
3113 | |
3114 | /* calculate interval: */ |
3115 | cycle_delta = (now - base) & mask; |
3116 | /* convert to nsecs: */ |
3117 | - nsec_delta = (cycle_delta * mult) >> shift; |
3118 | + nsec += (cycle_delta * mult) >> shift; |
3119 | |
3120 | - /* convert to usecs and add to timespec: */ |
3121 | - tv->tv_usec += nsec_delta / NSEC_PER_USEC; |
3122 | - while (tv->tv_usec > USEC_PER_SEC) { |
3123 | + while (nsec >= NSEC_PER_SEC) { |
3124 | tv->tv_sec += 1; |
3125 | - tv->tv_usec -= USEC_PER_SEC; |
3126 | + nsec -= NSEC_PER_SEC; |
3127 | } |
3128 | + tv->tv_usec = nsec / NSEC_PER_USEC; |
3129 | } |
3130 | |
3131 | int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz) |
3132 | @@ -151,11 +175,16 @@ |
3133 | * unlikely */ |
3134 | time_t __vsyscall(1) vtime(time_t *t) |
3135 | { |
3136 | + struct timeval tv; |
3137 | + time_t result; |
3138 | if (unlikely(!__vsyscall_gtod_data.sysctl_enabled)) |
3139 | return time_syscall(t); |
3140 | - else if (t) |
3141 | - *t = __vsyscall_gtod_data.wall_time_tv.tv_sec; |
3142 | - return __vsyscall_gtod_data.wall_time_tv.tv_sec; |
3143 | + |
3144 | + vgettimeofday(&tv, 0); |
3145 | + result = tv.tv_sec; |
3146 | + if (t) |
3147 | + *t = result; |
3148 | + return result; |
3149 | } |
3150 | |
3151 | /* Fast way to get current CPU and node. |
3152 | @@ -224,10 +253,10 @@ |
3153 | return ret; |
3154 | /* gcc has some trouble with __va(__pa()), so just do it this |
3155 | way. */ |
3156 | - map1 = ioremap(__pa_symbol(&vsysc1), 2); |
3157 | + map1 = ioremap(__pa_vsymbol(&vsysc1), 2); |
3158 | if (!map1) |
3159 | return -ENOMEM; |
3160 | - map2 = ioremap(__pa_symbol(&vsysc2), 2); |
3161 | + map2 = ioremap(__pa_vsymbol(&vsysc2), 2); |
3162 | if (!map2) { |
3163 | ret = -ENOMEM; |
3164 | goto out; |
3165 | @@ -304,7 +333,7 @@ |
3166 | cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg) |
3167 | { |
3168 | long cpu = (long)arg; |
3169 | - if (action == CPU_ONLINE) |
3170 | + if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) |
3171 | smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 0, 1); |
3172 | return NOTIFY_DONE; |
3173 | } |
3174 | --- a/arch/x86/mm/fault_32-xen.c |
3175 | +++ b/arch/x86/mm/fault_32-xen.c |
3176 | @@ -14,19 +14,20 @@ |
3177 | #include <linux/mman.h> |
3178 | #include <linux/mm.h> |
3179 | #include <linux/smp.h> |
3180 | -#include <linux/smp_lock.h> |
3181 | #include <linux/interrupt.h> |
3182 | #include <linux/init.h> |
3183 | #include <linux/tty.h> |
3184 | #include <linux/vt_kern.h> /* For unblank_screen() */ |
3185 | #include <linux/highmem.h> |
3186 | +#include <linux/bootmem.h> /* for max_low_pfn */ |
3187 | +#include <linux/vmalloc.h> |
3188 | #include <linux/module.h> |
3189 | #include <linux/kprobes.h> |
3190 | #include <linux/uaccess.h> |
3191 | +#include <linux/kdebug.h> |
3192 | |
3193 | #include <asm/system.h> |
3194 | #include <asm/desc.h> |
3195 | -#include <asm/kdebug.h> |
3196 | #include <asm/segment.h> |
3197 | |
3198 | extern void die(const char *,struct pt_regs *,long); |
3199 | @@ -259,25 +260,20 @@ |
3200 | unsigned long page; |
3201 | |
3202 | page = read_cr3(); |
3203 | - page = ((unsigned long *) __va(page))[address >> 22]; |
3204 | - if (oops_may_print()) |
3205 | - printk(KERN_ALERT "*pde = ma %08lx pa %08lx\n", page, |
3206 | - machine_to_phys(page)); |
3207 | + page = ((unsigned long *) __va(page))[address >> PGDIR_SHIFT]; |
3208 | + printk(KERN_ALERT "*pde = ma %08lx pa %08lx\n", page, |
3209 | + machine_to_phys(page)); |
3210 | /* |
3211 | * We must not directly access the pte in the highpte |
3212 | * case if the page table is located in highmem. |
3213 | * And lets rather not kmap-atomic the pte, just in case |
3214 | * it's allocated already. |
3215 | */ |
3216 | -#ifdef CONFIG_HIGHPTE |
3217 | - if ((page >> PAGE_SHIFT) >= highstart_pfn) |
3218 | - return; |
3219 | -#endif |
3220 | - if ((page & 1) && oops_may_print()) { |
3221 | - page &= PAGE_MASK; |
3222 | - address &= 0x003ff000; |
3223 | - page = machine_to_phys(page); |
3224 | - page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT]; |
3225 | + if ((machine_to_phys(page) >> PAGE_SHIFT) < max_low_pfn |
3226 | + && (page & _PAGE_PRESENT)) { |
3227 | + page = machine_to_phys(page & PAGE_MASK); |
3228 | + page = ((unsigned long *) __va(page))[(address >> PAGE_SHIFT) |
3229 | + & (PTRS_PER_PTE - 1)]; |
3230 | printk(KERN_ALERT "*pte = ma %08lx pa %08lx\n", page, |
3231 | machine_to_phys(page)); |
3232 | } |
3233 | @@ -581,6 +577,11 @@ |
3234 | bad_area_nosemaphore: |
3235 | /* User mode accesses just cause a SIGSEGV */ |
3236 | if (error_code & 4) { |
3237 | + /* |
3238 | + * It's possible to have interrupts off here. |
3239 | + */ |
3240 | + local_irq_enable(); |
3241 | + |
3242 | /* |
3243 | * Valid to do another page fault here because this one came |
3244 | * from user space. |
3245 | @@ -633,7 +634,7 @@ |
3246 | bust_spinlocks(1); |
3247 | |
3248 | if (oops_may_print()) { |
3249 | - #ifdef CONFIG_X86_PAE |
3250 | +#ifdef CONFIG_X86_PAE |
3251 | if (error_code & 16) { |
3252 | pte_t *pte = lookup_address(address); |
3253 | |
3254 | @@ -642,7 +643,7 @@ |
3255 | "NX-protected page - exploit attempt? " |
3256 | "(uid: %d)\n", current->uid); |
3257 | } |
3258 | - #endif |
3259 | +#endif |
3260 | if (address < PAGE_SIZE) |
3261 | printk(KERN_ALERT "BUG: unable to handle kernel NULL " |
3262 | "pointer dereference"); |
3263 | @@ -652,8 +653,8 @@ |
3264 | printk(" at virtual address %08lx\n",address); |
3265 | printk(KERN_ALERT " printing eip:\n"); |
3266 | printk("%08lx\n", regs->eip); |
3267 | + dump_fault_path(address); |
3268 | } |
3269 | - dump_fault_path(address); |
3270 | tsk->thread.cr2 = address; |
3271 | tsk->thread.trap_no = 14; |
3272 | tsk->thread.error_code = error_code; |
3273 | @@ -694,7 +695,6 @@ |
3274 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
3275 | } |
3276 | |
3277 | -#if !HAVE_SHARED_KERNEL_PMD |
3278 | void vmalloc_sync_all(void) |
3279 | { |
3280 | /* |
3281 | @@ -710,6 +710,9 @@ |
3282 | static unsigned long start = TASK_SIZE; |
3283 | unsigned long address; |
3284 | |
3285 | + if (SHARED_KERNEL_PMD) |
3286 | + return; |
3287 | + |
3288 | BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK); |
3289 | for (address = start; |
3290 | address >= TASK_SIZE && address < hypervisor_virt_start; |
3291 | @@ -739,4 +742,3 @@ |
3292 | start = address + (1UL << PMD_SHIFT); |
3293 | } |
3294 | } |
3295 | -#endif |
3296 | --- a/arch/x86/mm/fault_64-xen.c |
3297 | +++ b/arch/x86/mm/fault_64-xen.c |
3298 | @@ -15,22 +15,22 @@ |
3299 | #include <linux/mman.h> |
3300 | #include <linux/mm.h> |
3301 | #include <linux/smp.h> |
3302 | -#include <linux/smp_lock.h> |
3303 | #include <linux/interrupt.h> |
3304 | #include <linux/init.h> |
3305 | #include <linux/tty.h> |
3306 | #include <linux/vt_kern.h> /* For unblank_screen() */ |
3307 | #include <linux/compiler.h> |
3308 | +#include <linux/vmalloc.h> |
3309 | #include <linux/module.h> |
3310 | #include <linux/kprobes.h> |
3311 | #include <linux/uaccess.h> |
3312 | +#include <linux/kdebug.h> |
3313 | |
3314 | #include <asm/system.h> |
3315 | #include <asm/pgalloc.h> |
3316 | #include <asm/smp.h> |
3317 | #include <asm/tlbflush.h> |
3318 | #include <asm/proto.h> |
3319 | -#include <asm/kdebug.h> |
3320 | #include <asm-generic/sections.h> |
3321 | |
3322 | /* Page fault error code bits */ |
3323 | @@ -537,6 +537,12 @@ |
3324 | bad_area_nosemaphore: |
3325 | /* User mode accesses just cause a SIGSEGV */ |
3326 | if (error_code & PF_USER) { |
3327 | + |
3328 | + /* |
3329 | + * It's possible to have interrupts off here. |
3330 | + */ |
3331 | + local_irq_enable(); |
3332 | + |
3333 | if (is_prefetch(regs, address, error_code)) |
3334 | return; |
3335 | |
3336 | @@ -646,7 +652,7 @@ |
3337 | } |
3338 | |
3339 | DEFINE_SPINLOCK(pgd_lock); |
3340 | -struct page *pgd_list; |
3341 | +LIST_HEAD(pgd_list); |
3342 | |
3343 | void vmalloc_sync_all(void) |
3344 | { |
3345 | @@ -666,8 +672,7 @@ |
3346 | if (pgd_none(*pgd_ref)) |
3347 | continue; |
3348 | spin_lock(&pgd_lock); |
3349 | - for (page = pgd_list; page; |
3350 | - page = (struct page *)page->index) { |
3351 | + list_for_each_entry(page, &pgd_list, lru) { |
3352 | pgd_t *pgd; |
3353 | pgd = (pgd_t *)page_address(page) + pgd_index(address); |
3354 | if (pgd_none(*pgd)) |
3355 | --- a/arch/x86/mm/highmem_32-xen.c |
3356 | +++ b/arch/x86/mm/highmem_32-xen.c |
3357 | @@ -26,7 +26,7 @@ |
3358 | * However when holding an atomic kmap is is not legal to sleep, so atomic |
3359 | * kmaps are appropriate for short, tight code paths only. |
3360 | */ |
3361 | -static void *__kmap_atomic(struct page *page, enum km_type type, pgprot_t prot) |
3362 | +void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot) |
3363 | { |
3364 | enum fixed_addresses idx; |
3365 | unsigned long vaddr; |
3366 | @@ -49,15 +49,7 @@ |
3367 | |
3368 | void *kmap_atomic(struct page *page, enum km_type type) |
3369 | { |
3370 | - return __kmap_atomic(page, type, kmap_prot); |
3371 | -} |
3372 | - |
3373 | -/* Same as kmap_atomic but with PAGE_KERNEL_RO page protection. */ |
3374 | -void *kmap_atomic_pte(struct page *page, enum km_type type) |
3375 | -{ |
3376 | - return __kmap_atomic(page, type, |
3377 | - test_bit(PG_pinned, &page->flags) |
3378 | - ? PAGE_KERNEL_RO : kmap_prot); |
3379 | + return kmap_atomic_prot(page, type, kmap_prot); |
3380 | } |
3381 | |
3382 | void kunmap_atomic(void *kvaddr, enum km_type type) |
3383 | @@ -80,6 +72,7 @@ |
3384 | #endif |
3385 | } |
3386 | |
3387 | + arch_flush_lazy_mmu_mode(); |
3388 | pagefault_enable(); |
3389 | } |
3390 | |
3391 | @@ -117,6 +110,5 @@ |
3392 | EXPORT_SYMBOL(kmap); |
3393 | EXPORT_SYMBOL(kunmap); |
3394 | EXPORT_SYMBOL(kmap_atomic); |
3395 | -EXPORT_SYMBOL(kmap_atomic_pte); |
3396 | EXPORT_SYMBOL(kunmap_atomic); |
3397 | EXPORT_SYMBOL(kmap_atomic_to_page); |
3398 | --- a/arch/x86/mm/init_32-xen.c |
3399 | +++ b/arch/x86/mm/init_32-xen.c |
3400 | @@ -22,6 +22,7 @@ |
3401 | #include <linux/init.h> |
3402 | #include <linux/highmem.h> |
3403 | #include <linux/pagemap.h> |
3404 | +#include <linux/pfn.h> |
3405 | #include <linux/poison.h> |
3406 | #include <linux/bootmem.h> |
3407 | #include <linux/slab.h> |
3408 | @@ -67,17 +68,19 @@ |
3409 | pmd_t *pmd_table; |
3410 | |
3411 | #ifdef CONFIG_X86_PAE |
3412 | - pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE); |
3413 | - paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT); |
3414 | - make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables); |
3415 | - set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
3416 | - pud = pud_offset(pgd, 0); |
3417 | - if (pmd_table != pmd_offset(pud, 0)) |
3418 | - BUG(); |
3419 | -#else |
3420 | + if (!(__pgd_val(*pgd) & _PAGE_PRESENT)) { |
3421 | + pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE); |
3422 | + |
3423 | + paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT); |
3424 | + make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables); |
3425 | + set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
3426 | + pud = pud_offset(pgd, 0); |
3427 | + if (pmd_table != pmd_offset(pud, 0)) |
3428 | + BUG(); |
3429 | + } |
3430 | +#endif |
3431 | pud = pud_offset(pgd, 0); |
3432 | pmd_table = pmd_offset(pud, 0); |
3433 | -#endif |
3434 | |
3435 | return pmd_table; |
3436 | } |
3437 | @@ -88,16 +91,18 @@ |
3438 | */ |
3439 | static pte_t * __init one_page_table_init(pmd_t *pmd) |
3440 | { |
3441 | +#if CONFIG_XEN_COMPAT <= 0x030002 |
3442 | if (pmd_none(*pmd)) { |
3443 | +#else |
3444 | + if (!(__pmd_val(*pmd) & _PAGE_PRESENT)) { |
3445 | +#endif |
3446 | pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE); |
3447 | + |
3448 | paravirt_alloc_pt(__pa(page_table) >> PAGE_SHIFT); |
3449 | make_lowmem_page_readonly(page_table, |
3450 | XENFEAT_writable_page_tables); |
3451 | set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); |
3452 | - if (page_table != pte_offset_kernel(pmd, 0)) |
3453 | - BUG(); |
3454 | - |
3455 | - return page_table; |
3456 | + BUG_ON(page_table != pte_offset_kernel(pmd, 0)); |
3457 | } |
3458 | |
3459 | return pte_offset_kernel(pmd, 0); |
3460 | @@ -117,7 +122,6 @@ |
3461 | static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base) |
3462 | { |
3463 | pgd_t *pgd; |
3464 | - pud_t *pud; |
3465 | pmd_t *pmd; |
3466 | int pgd_idx, pmd_idx; |
3467 | unsigned long vaddr; |
3468 | @@ -128,12 +132,10 @@ |
3469 | pgd = pgd_base + pgd_idx; |
3470 | |
3471 | for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { |
3472 | - if (pgd_none(*pgd)) |
3473 | - one_md_table_init(pgd); |
3474 | - pud = pud_offset(pgd, vaddr); |
3475 | - pmd = pmd_offset(pud, vaddr); |
3476 | + pmd = one_md_table_init(pgd); |
3477 | + pmd = pmd + pmd_index(vaddr); |
3478 | for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) { |
3479 | - if (vaddr < hypervisor_virt_start && pmd_none(*pmd)) |
3480 | + if (vaddr < hypervisor_virt_start) |
3481 | one_page_table_init(pmd); |
3482 | |
3483 | vaddr += PMD_SIZE; |
3484 | @@ -196,24 +198,25 @@ |
3485 | /* Map with big pages if possible, otherwise create normal page tables. */ |
3486 | if (cpu_has_pse) { |
3487 | unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1; |
3488 | - |
3489 | if (is_kernel_text(address) || is_kernel_text(address2)) |
3490 | set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC)); |
3491 | else |
3492 | set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE)); |
3493 | + |
3494 | pfn += PTRS_PER_PTE; |
3495 | } else { |
3496 | pte = one_page_table_init(pmd); |
3497 | |
3498 | - pte += pte_ofs; |
3499 | - for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) { |
3500 | - /* XEN: Only map initial RAM allocation. */ |
3501 | - if ((pfn >= max_ram_pfn) || pte_present(*pte)) |
3502 | - continue; |
3503 | - if (is_kernel_text(address)) |
3504 | - set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC)); |
3505 | - else |
3506 | - set_pte(pte, pfn_pte(pfn, PAGE_KERNEL)); |
3507 | + for (pte += pte_ofs; |
3508 | + pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; |
3509 | + pte++, pfn++, pte_ofs++, address += PAGE_SIZE) { |
3510 | + /* XEN: Only map initial RAM allocation. */ |
3511 | + if ((pfn >= max_ram_pfn) || pte_present(*pte)) |
3512 | + continue; |
3513 | + if (is_kernel_text(address)) |
3514 | + set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC)); |
3515 | + else |
3516 | + set_pte(pte, pfn_pte(pfn, PAGE_KERNEL)); |
3517 | } |
3518 | pte_ofs = 0; |
3519 | } |
3520 | @@ -383,15 +386,44 @@ |
3521 | |
3522 | pgd_t *swapper_pg_dir; |
3523 | |
3524 | +static void __init xen_pagetable_setup_start(pgd_t *base) |
3525 | +{ |
3526 | +} |
3527 | + |
3528 | +static void __init xen_pagetable_setup_done(pgd_t *base) |
3529 | +{ |
3530 | +} |
3531 | + |
3532 | +/* |
3533 | + * Build a proper pagetable for the kernel mappings. Up until this |
3534 | + * point, we've been running on some set of pagetables constructed by |
3535 | + * the boot process. |
3536 | + * |
3537 | + * If we're booting on native hardware, this will be a pagetable |
3538 | + * constructed in arch/i386/kernel/head.S, and not running in PAE mode |
3539 | + * (even if we'll end up running in PAE). The root of the pagetable |
3540 | + * will be swapper_pg_dir. |
3541 | + * |
3542 | + * If we're booting paravirtualized under a hypervisor, then there are |
3543 | + * more options: we may already be running PAE, and the pagetable may |
3544 | + * or may not be based in swapper_pg_dir. In any case, |
3545 | + * paravirt_pagetable_setup_start() will set up swapper_pg_dir |
3546 | + * appropriately for the rest of the initialization to work. |
3547 | + * |
3548 | + * In general, pagetable_init() assumes that the pagetable may already |
3549 | + * be partially populated, and so it avoids stomping on any existing |
3550 | + * mappings. |
3551 | + */ |
3552 | static void __init pagetable_init (void) |
3553 | { |
3554 | - unsigned long vaddr; |
3555 | + unsigned long vaddr, end; |
3556 | pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base; |
3557 | |
3558 | + xen_pagetable_setup_start(pgd_base); |
3559 | + |
3560 | /* Enable PSE if available */ |
3561 | - if (cpu_has_pse) { |
3562 | + if (cpu_has_pse) |
3563 | set_in_cr4(X86_CR4_PSE); |
3564 | - } |
3565 | |
3566 | /* Enable PGE if available */ |
3567 | if (cpu_has_pge) { |
3568 | @@ -408,9 +440,12 @@ |
3569 | * created - mappings will be set by set_fixmap(): |
3570 | */ |
3571 | vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; |
3572 | - page_table_range_init(vaddr, hypervisor_virt_start, pgd_base); |
3573 | + end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; |
3574 | + page_table_range_init(vaddr, end, pgd_base); |
3575 | |
3576 | permanent_kmaps_init(pgd_base); |
3577 | + |
3578 | + xen_pagetable_setup_done(pgd_base); |
3579 | } |
3580 | |
3581 | #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP) |
3582 | @@ -757,34 +792,29 @@ |
3583 | EXPORT_SYMBOL_GPL(remove_memory); |
3584 | #endif |
3585 | |
3586 | -struct kmem_cache *pgd_cache; |
3587 | struct kmem_cache *pmd_cache; |
3588 | |
3589 | void __init pgtable_cache_init(void) |
3590 | { |
3591 | + size_t pgd_size = PTRS_PER_PGD*sizeof(pgd_t); |
3592 | + |
3593 | if (PTRS_PER_PMD > 1) { |
3594 | pmd_cache = kmem_cache_create("pmd", |
3595 | PTRS_PER_PMD*sizeof(pmd_t), |
3596 | PTRS_PER_PMD*sizeof(pmd_t), |
3597 | - 0, |
3598 | + SLAB_PANIC, |
3599 | pmd_ctor, |
3600 | NULL); |
3601 | - if (!pmd_cache) |
3602 | - panic("pgtable_cache_init(): cannot create pmd cache"); |
3603 | + if (!SHARED_KERNEL_PMD) { |
3604 | + /* If we're in PAE mode and have a non-shared |
3605 | + kernel pmd, then the pgd size must be a |
3606 | + page size. This is because the pgd_list |
3607 | + links through the page structure, so there |
3608 | + can only be one pgd per page for this to |
3609 | + work. */ |
3610 | + pgd_size = PAGE_SIZE; |
3611 | + } |
3612 | } |
3613 | - pgd_cache = kmem_cache_create("pgd", |
3614 | -#ifndef CONFIG_XEN |
3615 | - PTRS_PER_PGD*sizeof(pgd_t), |
3616 | - PTRS_PER_PGD*sizeof(pgd_t), |
3617 | -#else |
3618 | - PAGE_SIZE, |
3619 | - PAGE_SIZE, |
3620 | -#endif |
3621 | - 0, |
3622 | - pgd_ctor, |
3623 | - PTRS_PER_PMD == 1 ? pgd_dtor : NULL); |
3624 | - if (!pgd_cache) |
3625 | - panic("pgtable_cache_init(): Cannot create pgd cache"); |
3626 | } |
3627 | |
3628 | /* |
3629 | @@ -818,13 +848,26 @@ |
3630 | |
3631 | void mark_rodata_ro(void) |
3632 | { |
3633 | - unsigned long addr = (unsigned long)__start_rodata; |
3634 | - |
3635 | - for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE) |
3636 | - change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO); |
3637 | + unsigned long start = PFN_ALIGN(_text); |
3638 | + unsigned long size = PFN_ALIGN(_etext) - start; |
3639 | |
3640 | - printk("Write protecting the kernel read-only data: %uk\n", |
3641 | - (__end_rodata - __start_rodata) >> 10); |
3642 | +#ifndef CONFIG_KPROBES |
3643 | +#ifdef CONFIG_HOTPLUG_CPU |
3644 | + /* It must still be possible to apply SMP alternatives. */ |
3645 | + if (num_possible_cpus() <= 1) |
3646 | +#endif |
3647 | + { |
3648 | + change_page_attr(virt_to_page(start), |
3649 | + size >> PAGE_SHIFT, PAGE_KERNEL_RX); |
3650 | + printk("Write protecting the kernel text: %luk\n", size >> 10); |
3651 | + } |
3652 | +#endif |
3653 | + start += size; |
3654 | + size = (unsigned long)__end_rodata - start; |
3655 | + change_page_attr(virt_to_page(start), |
3656 | + size >> PAGE_SHIFT, PAGE_KERNEL_RO); |
3657 | + printk("Write protecting the kernel read-only data: %luk\n", |
3658 | + size >> 10); |
3659 | |
3660 | /* |
3661 | * change_page_attr() requires a global_flush_tlb() call after it. |
3662 | @@ -847,7 +890,7 @@ |
3663 | free_page(addr); |
3664 | totalram_pages++; |
3665 | } |
3666 | - printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10); |
3667 | + printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
3668 | } |
3669 | |
3670 | void free_initmem(void) |
3671 | --- a/arch/x86/mm/init_64-xen.c |
3672 | +++ b/arch/x86/mm/init_64-xen.c |
3673 | @@ -25,10 +25,12 @@ |
3674 | #include <linux/bootmem.h> |
3675 | #include <linux/proc_fs.h> |
3676 | #include <linux/pci.h> |
3677 | +#include <linux/pfn.h> |
3678 | #include <linux/poison.h> |
3679 | #include <linux/dma-mapping.h> |
3680 | #include <linux/module.h> |
3681 | #include <linux/memory_hotplug.h> |
3682 | +#include <linux/nmi.h> |
3683 | |
3684 | #include <asm/processor.h> |
3685 | #include <asm/system.h> |
3686 | @@ -51,7 +53,7 @@ |
3687 | #define Dprintk(x...) |
3688 | #endif |
3689 | |
3690 | -struct dma_mapping_ops* dma_ops; |
3691 | +const struct dma_mapping_ops* dma_ops; |
3692 | EXPORT_SYMBOL(dma_ops); |
3693 | |
3694 | #if CONFIG_XEN_COMPAT <= 0x030002 |
3695 | @@ -189,6 +191,13 @@ |
3696 | |
3697 | for_each_online_pgdat(pgdat) { |
3698 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
3699 | + /* this loop can take a while with 256 GB and 4k pages |
3700 | + so update the NMI watchdog */ |
3701 | + if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) { |
3702 | + touch_nmi_watchdog(); |
3703 | + } |
3704 | + if (!pfn_valid(pgdat->node_start_pfn + i)) |
3705 | + continue; |
3706 | page = pfn_to_page(pgdat->node_start_pfn + i); |
3707 | total++; |
3708 | if (PageReserved(page)) |
3709 | @@ -350,7 +359,7 @@ |
3710 | } |
3711 | } |
3712 | |
3713 | -unsigned long __initdata table_start, table_end; |
3714 | +unsigned long __meminitdata table_start, table_end; |
3715 | |
3716 | static __meminit void *alloc_static_page(unsigned long *phys) |
3717 | { |
3718 | @@ -367,7 +376,7 @@ |
3719 | start_pfn++; |
3720 | memset((void *)va, 0, PAGE_SIZE); |
3721 | return (void *)va; |
3722 | -} |
3723 | +} |
3724 | |
3725 | #define PTE_SIZE PAGE_SIZE |
3726 | |
3727 | @@ -408,28 +417,46 @@ |
3728 | |
3729 | #ifndef CONFIG_XEN |
3730 | /* Must run before zap_low_mappings */ |
3731 | -__init void *early_ioremap(unsigned long addr, unsigned long size) |
3732 | +__meminit void *early_ioremap(unsigned long addr, unsigned long size) |
3733 | { |
3734 | - unsigned long map = round_down(addr, LARGE_PAGE_SIZE); |
3735 | - |
3736 | - /* actually usually some more */ |
3737 | - if (size >= LARGE_PAGE_SIZE) { |
3738 | - return NULL; |
3739 | + unsigned long vaddr; |
3740 | + pmd_t *pmd, *last_pmd; |
3741 | + int i, pmds; |
3742 | + |
3743 | + pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; |
3744 | + vaddr = __START_KERNEL_map; |
3745 | + pmd = level2_kernel_pgt; |
3746 | + last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; |
3747 | + for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { |
3748 | + for (i = 0; i < pmds; i++) { |
3749 | + if (pmd_present(pmd[i])) |
3750 | + goto next; |
3751 | + } |
3752 | + vaddr += addr & ~PMD_MASK; |
3753 | + addr &= PMD_MASK; |
3754 | + for (i = 0; i < pmds; i++, addr += PMD_SIZE) |
3755 | + set_pmd(pmd + i,__pmd(addr | _KERNPG_TABLE | _PAGE_PSE)); |
3756 | + __flush_tlb(); |
3757 | + return (void *)vaddr; |
3758 | + next: |
3759 | + ; |
3760 | } |
3761 | - set_pmd(temp_mappings[0].pmd, __pmd(map | _KERNPG_TABLE | _PAGE_PSE)); |
3762 | - map += LARGE_PAGE_SIZE; |
3763 | - set_pmd(temp_mappings[1].pmd, __pmd(map | _KERNPG_TABLE | _PAGE_PSE)); |
3764 | - __flush_tlb(); |
3765 | - return temp_mappings[0].address + (addr & (LARGE_PAGE_SIZE-1)); |
3766 | + printk("early_ioremap(0x%lx, %lu) failed\n", addr, size); |
3767 | + return NULL; |
3768 | } |
3769 | |
3770 | /* To avoid virtual aliases later */ |
3771 | -__init void early_iounmap(void *addr, unsigned long size) |
3772 | +__meminit void early_iounmap(void *addr, unsigned long size) |
3773 | { |
3774 | - if ((void *)round_down((unsigned long)addr, LARGE_PAGE_SIZE) != temp_mappings[0].address) |
3775 | - printk("early_iounmap: bad address %p\n", addr); |
3776 | - set_pmd(temp_mappings[0].pmd, __pmd(0)); |
3777 | - set_pmd(temp_mappings[1].pmd, __pmd(0)); |
3778 | + unsigned long vaddr; |
3779 | + pmd_t *pmd; |
3780 | + int i, pmds; |
3781 | + |
3782 | + vaddr = (unsigned long)addr; |
3783 | + pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; |
3784 | + pmd = level2_kernel_pgt + pmd_index(vaddr); |
3785 | + for (i = 0; i < pmds; i++) |
3786 | + pmd_clear(pmd + i); |
3787 | __flush_tlb(); |
3788 | } |
3789 | #endif |
3790 | @@ -763,14 +790,6 @@ |
3791 | __flush_tlb_all(); |
3792 | } |
3793 | |
3794 | -void __cpuinit zap_low_mappings(int cpu) |
3795 | -{ |
3796 | - /* this is not required for Xen */ |
3797 | -#if 0 |
3798 | - swap_low_mappings(); |
3799 | -#endif |
3800 | -} |
3801 | - |
3802 | #ifndef CONFIG_NUMA |
3803 | void __init paging_init(void) |
3804 | { |
3805 | @@ -961,17 +980,6 @@ |
3806 | reservedpages << (PAGE_SHIFT-10), |
3807 | datasize >> 10, |
3808 | initsize >> 10); |
3809 | - |
3810 | -#ifndef CONFIG_XEN |
3811 | -#ifdef CONFIG_SMP |
3812 | - /* |
3813 | - * Sync boot_level4_pgt mappings with the init_level4_pgt |
3814 | - * except for the low identity mappings which are already zapped |
3815 | - * in init_level4_pgt. This sync-up is essential for AP's bringup |
3816 | - */ |
3817 | - memcpy(boot_level4_pgt+1, init_level4_pgt+1, (PTRS_PER_PGD-1)*sizeof(pgd_t)); |
3818 | -#endif |
3819 | -#endif |
3820 | } |
3821 | |
3822 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
3823 | @@ -981,7 +989,7 @@ |
3824 | if (begin >= end) |
3825 | return; |
3826 | |
3827 | - printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10); |
3828 | + printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
3829 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
3830 | ClearPageReserved(virt_to_page(addr)); |
3831 | init_page_count(virt_to_page(addr)); |
3832 | @@ -990,24 +998,17 @@ |
3833 | if (addr >= __START_KERNEL_map) { |
3834 | /* make_readonly() reports all kernel addresses. */ |
3835 | __make_page_writable(__va(__pa(addr))); |
3836 | - if (HYPERVISOR_update_va_mapping(addr, __pte(0), 0)) { |
3837 | - pgd_t *pgd = pgd_offset_k(addr); |
3838 | - pud_t *pud = pud_offset(pgd, addr); |
3839 | - pmd_t *pmd = pmd_offset(pud, addr); |
3840 | - pte_t *pte = pte_offset_kernel(pmd, addr); |
3841 | - |
3842 | - xen_l1_entry_update(pte, __pte(0)); /* fallback */ |
3843 | - } |
3844 | + change_page_attr_addr(addr, 1, __pgprot(0)); |
3845 | } |
3846 | free_page(addr); |
3847 | totalram_pages++; |
3848 | } |
3849 | + if (addr > __START_KERNEL_map) |
3850 | + global_flush_tlb(); |
3851 | } |
3852 | |
3853 | void free_initmem(void) |
3854 | { |
3855 | - memset(__initdata_begin, POISON_FREE_INITDATA, |
3856 | - __initdata_end - __initdata_begin); |
3857 | free_init_pages("unused kernel memory", |
3858 | (unsigned long)(&__init_begin), |
3859 | (unsigned long)(&__init_end)); |
3860 | @@ -1017,13 +1018,28 @@ |
3861 | |
3862 | void mark_rodata_ro(void) |
3863 | { |
3864 | - unsigned long addr = (unsigned long)__start_rodata; |
3865 | + unsigned long start = (unsigned long)_stext, end; |
3866 | + |
3867 | +#ifdef CONFIG_HOTPLUG_CPU |
3868 | + /* It must still be possible to apply SMP alternatives. */ |
3869 | + if (num_possible_cpus() > 1) |
3870 | + start = (unsigned long)_etext; |
3871 | +#endif |
3872 | + |
3873 | +#ifdef CONFIG_KPROBES |
3874 | + start = (unsigned long)__start_rodata; |
3875 | +#endif |
3876 | + |
3877 | + end = (unsigned long)__end_rodata; |
3878 | + start = (start + PAGE_SIZE - 1) & PAGE_MASK; |
3879 | + end &= PAGE_MASK; |
3880 | + if (end <= start) |
3881 | + return; |
3882 | |
3883 | - for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE) |
3884 | - change_page_attr_addr(addr, 1, PAGE_KERNEL_RO); |
3885 | + change_page_attr_addr(start, (end - start) >> PAGE_SHIFT, PAGE_KERNEL_RO); |
3886 | |
3887 | - printk ("Write protecting the kernel read-only data: %luk\n", |
3888 | - (__end_rodata - __start_rodata) >> 10); |
3889 | + printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
3890 | + (end - start) >> 10); |
3891 | |
3892 | /* |
3893 | * change_page_attr_addr() requires a global_flush_tlb() call after it. |
3894 | @@ -1176,3 +1192,11 @@ |
3895 | { |
3896 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
3897 | } |
3898 | + |
3899 | +#ifndef CONFIG_XEN |
3900 | +void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size) |
3901 | +{ |
3902 | + return __alloc_bootmem_core(pgdat->bdata, size, |
3903 | + SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0); |
3904 | +} |
3905 | +#endif |
3906 | --- a/arch/x86/mm/ioremap_32-xen.c |
3907 | +++ b/arch/x86/mm/ioremap_32-xen.c |
3908 | @@ -13,6 +13,7 @@ |
3909 | #include <linux/slab.h> |
3910 | #include <linux/module.h> |
3911 | #include <linux/io.h> |
3912 | +#include <linux/sched.h> |
3913 | #include <asm/fixmap.h> |
3914 | #include <asm/cacheflush.h> |
3915 | #include <asm/tlbflush.h> |
3916 | --- a/arch/x86/mm/pageattr_64-xen.c |
3917 | +++ b/arch/x86/mm/pageattr_64-xen.c |
3918 | @@ -215,13 +215,13 @@ |
3919 | preempt_enable(); |
3920 | } |
3921 | |
3922 | -void _arch_dup_mmap(struct mm_struct *mm) |
3923 | +void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) |
3924 | { |
3925 | if (!mm->context.pinned) |
3926 | mm_pin(mm); |
3927 | } |
3928 | |
3929 | -void _arch_exit_mmap(struct mm_struct *mm) |
3930 | +void arch_exit_mmap(struct mm_struct *mm) |
3931 | { |
3932 | struct task_struct *tsk = current; |
3933 | |
3934 | @@ -337,10 +337,11 @@ |
3935 | struct page *pg; |
3936 | |
3937 | /* When clflush is available always use it because it is |
3938 | - much cheaper than WBINVD */ |
3939 | - if (!cpu_has_clflush) |
3940 | + much cheaper than WBINVD. Disable clflush for now because |
3941 | + the high level code is not ready yet */ |
3942 | + if (1 || !cpu_has_clflush) |
3943 | asm volatile("wbinvd" ::: "memory"); |
3944 | - list_for_each_entry(pg, l, lru) { |
3945 | + else list_for_each_entry(pg, l, lru) { |
3946 | void *adr = page_address(pg); |
3947 | if (cpu_has_clflush) |
3948 | cache_flush_page(adr); |
3949 | @@ -454,16 +455,24 @@ |
3950 | */ |
3951 | int change_page_attr_addr(unsigned long address, int numpages, pgprot_t prot) |
3952 | { |
3953 | - int err = 0; |
3954 | + int err = 0, kernel_map = 0; |
3955 | int i; |
3956 | |
3957 | + if (address >= __START_KERNEL_map |
3958 | + && address < __START_KERNEL_map + KERNEL_TEXT_SIZE) { |
3959 | + address = (unsigned long)__va(__pa(address)); |
3960 | + kernel_map = 1; |
3961 | + } |
3962 | + |
3963 | down_write(&init_mm.mmap_sem); |
3964 | for (i = 0; i < numpages; i++, address += PAGE_SIZE) { |
3965 | unsigned long pfn = __pa(address) >> PAGE_SHIFT; |
3966 | |
3967 | - err = __change_page_attr(address, pfn, prot, PAGE_KERNEL); |
3968 | - if (err) |
3969 | - break; |
3970 | + if (!kernel_map || pte_present(pfn_pte(0, prot))) { |
3971 | + err = __change_page_attr(address, pfn, prot, PAGE_KERNEL); |
3972 | + if (err) |
3973 | + break; |
3974 | + } |
3975 | /* Handle kernel mapping too which aliases part of the |
3976 | * lowmem */ |
3977 | if (__pa(address) < KERNEL_TEXT_SIZE) { |
3978 | --- a/arch/x86/mm/pgtable_32-xen.c |
3979 | +++ b/arch/x86/mm/pgtable_32-xen.c |
3980 | @@ -13,6 +13,7 @@ |
3981 | #include <linux/pagemap.h> |
3982 | #include <linux/spinlock.h> |
3983 | #include <linux/module.h> |
3984 | +#include <linux/quicklist.h> |
3985 | |
3986 | #include <asm/system.h> |
3987 | #include <asm/pgtable.h> |
3988 | @@ -212,8 +213,6 @@ |
3989 | * against pageattr.c; it is the unique case in which a valid change |
3990 | * of kernel pagetables can't be lazily synchronized by vmalloc faults. |
3991 | * vmalloc faults work because attached pagetables are never freed. |
3992 | - * The locking scheme was chosen on the basis of manfred's |
3993 | - * recommendations and having no core impact whatsoever. |
3994 | * -- wli |
3995 | */ |
3996 | DEFINE_SPINLOCK(pgd_lock); |
3997 | @@ -239,37 +238,59 @@ |
3998 | set_page_private(next, (unsigned long)pprev); |
3999 | } |
4000 | |
4001 | -void pgd_ctor(void *pgd, struct kmem_cache *cache, unsigned long unused) |
4002 | + |
4003 | + |
4004 | +#if (PTRS_PER_PMD == 1) |
4005 | +/* Non-PAE pgd constructor */ |
4006 | +void pgd_ctor(void *pgd) |
4007 | { |
4008 | unsigned long flags; |
4009 | |
4010 | - if (PTRS_PER_PMD > 1) { |
4011 | - if (HAVE_SHARED_KERNEL_PMD) |
4012 | - clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD, |
4013 | - swapper_pg_dir + USER_PTRS_PER_PGD, |
4014 | - KERNEL_PGD_PTRS); |
4015 | - } else { |
4016 | - spin_lock_irqsave(&pgd_lock, flags); |
4017 | + /* !PAE, no pagetable sharing */ |
4018 | + memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t)); |
4019 | + |
4020 | + spin_lock_irqsave(&pgd_lock, flags); |
4021 | + |
4022 | + /* must happen under lock */ |
4023 | + clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD, |
4024 | + swapper_pg_dir + USER_PTRS_PER_PGD, |
4025 | + KERNEL_PGD_PTRS); |
4026 | + |
4027 | + paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT, |
4028 | + __pa(swapper_pg_dir) >> PAGE_SHIFT, |
4029 | + USER_PTRS_PER_PGD, |
4030 | + KERNEL_PGD_PTRS); |
4031 | + pgd_list_add(pgd); |
4032 | + spin_unlock_irqrestore(&pgd_lock, flags); |
4033 | +} |
4034 | +#else /* PTRS_PER_PMD > 1 */ |
4035 | +/* PAE pgd constructor */ |
4036 | +void pgd_ctor(void *pgd) |
4037 | +{ |
4038 | + /* PAE, kernel PMD may be shared */ |
4039 | + |
4040 | + if (SHARED_KERNEL_PMD) { |
4041 | clone_pgd_range((pgd_t *)pgd + USER_PTRS_PER_PGD, |
4042 | swapper_pg_dir + USER_PTRS_PER_PGD, |
4043 | KERNEL_PGD_PTRS); |
4044 | - memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t)); |
4045 | - |
4046 | - /* must happen under lock */ |
4047 | - paravirt_alloc_pd_clone(__pa(pgd) >> PAGE_SHIFT, |
4048 | - __pa(swapper_pg_dir) >> PAGE_SHIFT, |
4049 | - USER_PTRS_PER_PGD, PTRS_PER_PGD - USER_PTRS_PER_PGD); |
4050 | + } else { |
4051 | + unsigned long flags; |
4052 | |
4053 | + memset(pgd, 0, USER_PTRS_PER_PGD*sizeof(pgd_t)); |
4054 | + spin_lock_irqsave(&pgd_lock, flags); |
4055 | pgd_list_add(pgd); |
4056 | spin_unlock_irqrestore(&pgd_lock, flags); |
4057 | } |
4058 | } |
4059 | +#endif /* PTRS_PER_PMD */ |
4060 | |
4061 | -/* never called when PTRS_PER_PMD > 1 */ |
4062 | -void pgd_dtor(void *pgd, struct kmem_cache *cache, unsigned long unused) |
4063 | +void pgd_dtor(void *pgd) |
4064 | { |
4065 | unsigned long flags; /* can be called from interrupt context */ |
4066 | |
4067 | + if (SHARED_KERNEL_PMD) |
4068 | + return; |
4069 | + |
4070 | paravirt_release_pd(__pa(pgd) >> PAGE_SHIFT); |
4071 | spin_lock_irqsave(&pgd_lock, flags); |
4072 | pgd_list_del(pgd); |
4073 | @@ -278,11 +299,46 @@ |
4074 | pgd_test_and_unpin(pgd); |
4075 | } |
4076 | |
4077 | +#define UNSHARED_PTRS_PER_PGD \ |
4078 | + (SHARED_KERNEL_PMD ? USER_PTRS_PER_PGD : PTRS_PER_PGD) |
4079 | + |
4080 | +/* If we allocate a pmd for part of the kernel address space, then |
4081 | + make sure its initialized with the appropriate kernel mappings. |
4082 | + Otherwise use a cached zeroed pmd. */ |
4083 | +static pmd_t *pmd_cache_alloc(int idx) |
4084 | +{ |
4085 | + pmd_t *pmd; |
4086 | + |
4087 | + if (idx >= USER_PTRS_PER_PGD) { |
4088 | + pmd = (pmd_t *)__get_free_page(GFP_KERNEL); |
4089 | + |
4090 | +#ifndef CONFIG_XEN |
4091 | + if (pmd) |
4092 | + memcpy(pmd, |
4093 | + (void *)pgd_page_vaddr(swapper_pg_dir[idx]), |
4094 | + sizeof(pmd_t) * PTRS_PER_PMD); |
4095 | +#endif |
4096 | + } else |
4097 | + pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL); |
4098 | + |
4099 | + return pmd; |
4100 | +} |
4101 | + |
4102 | +static void pmd_cache_free(pmd_t *pmd, int idx) |
4103 | +{ |
4104 | + if (idx >= USER_PTRS_PER_PGD) { |
4105 | + make_lowmem_page_writable(pmd, XENFEAT_writable_page_tables); |
4106 | + memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t)); |
4107 | + free_page((unsigned long)pmd); |
4108 | + } else |
4109 | + kmem_cache_free(pmd_cache, pmd); |
4110 | +} |
4111 | + |
4112 | pgd_t *pgd_alloc(struct mm_struct *mm) |
4113 | { |
4114 | int i; |
4115 | - pgd_t *pgd = kmem_cache_alloc(pgd_cache, GFP_KERNEL); |
4116 | - pmd_t **pmd; |
4117 | + pgd_t *pgd = quicklist_alloc(0, GFP_KERNEL, pgd_ctor); |
4118 | + pmd_t **pmds = NULL; |
4119 | unsigned long flags; |
4120 | |
4121 | pgd_test_and_unpin(pgd); |
4122 | @@ -290,37 +346,40 @@ |
4123 | if (PTRS_PER_PMD == 1 || !pgd) |
4124 | return pgd; |
4125 | |
4126 | - if (HAVE_SHARED_KERNEL_PMD) { |
4127 | - for (i = 0; i < USER_PTRS_PER_PGD; ++i) { |
4128 | - pmd_t *pmd = kmem_cache_alloc(pmd_cache, GFP_KERNEL); |
4129 | - if (!pmd) |
4130 | - goto out_oom; |
4131 | - paravirt_alloc_pd(__pa(pmd) >> PAGE_SHIFT); |
4132 | - set_pgd(&pgd[i], __pgd(1 + __pa(pmd))); |
4133 | +#ifdef CONFIG_XEN |
4134 | + if (!SHARED_KERNEL_PMD) { |
4135 | + /* |
4136 | + * We can race save/restore (if we sleep during a GFP_KERNEL memory |
4137 | + * allocation). We therefore store virtual addresses of pmds as they |
4138 | + * do not change across save/restore, and poke the machine addresses |
4139 | + * into the pgdir under the pgd_lock. |
4140 | + */ |
4141 | + pmds = kmalloc(PTRS_PER_PGD * sizeof(pmd_t *), GFP_KERNEL); |
4142 | + if (!pmds) { |
4143 | + quicklist_free(0, pgd_dtor, pgd); |
4144 | + return NULL; |
4145 | } |
4146 | - return pgd; |
4147 | - } |
4148 | - |
4149 | - /* |
4150 | - * We can race save/restore (if we sleep during a GFP_KERNEL memory |
4151 | - * allocation). We therefore store virtual addresses of pmds as they |
4152 | - * do not change across save/restore, and poke the machine addresses |
4153 | - * into the pgdir under the pgd_lock. |
4154 | - */ |
4155 | - pmd = kmalloc(PTRS_PER_PGD * sizeof(pmd_t *), GFP_KERNEL); |
4156 | - if (!pmd) { |
4157 | - kmem_cache_free(pgd_cache, pgd); |
4158 | - return NULL; |
4159 | } |
4160 | +#endif |
4161 | |
4162 | /* Allocate pmds, remember virtual addresses. */ |
4163 | - for (i = 0; i < PTRS_PER_PGD; ++i) { |
4164 | - pmd[i] = kmem_cache_alloc(pmd_cache, GFP_KERNEL); |
4165 | - if (!pmd[i]) |
4166 | + for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) { |
4167 | + pmd_t *pmd = pmd_cache_alloc(i); |
4168 | + |
4169 | + if (!pmd) |
4170 | goto out_oom; |
4171 | + |
4172 | paravirt_alloc_pd(__pa(pmd) >> PAGE_SHIFT); |
4173 | + if (pmds) |
4174 | + pmds[i] = pmd; |
4175 | + else |
4176 | + set_pgd(&pgd[i], __pgd(1 + __pa(pmd))); |
4177 | } |
4178 | |
4179 | +#ifdef CONFIG_XEN |
4180 | + if (SHARED_KERNEL_PMD) |
4181 | + return pgd; |
4182 | + |
4183 | spin_lock_irqsave(&pgd_lock, flags); |
4184 | |
4185 | /* Protect against save/restore: move below 4GB under pgd_lock. */ |
4186 | @@ -335,44 +394,40 @@ |
4187 | |
4188 | /* Copy kernel pmd contents and write-protect the new pmds. */ |
4189 | for (i = USER_PTRS_PER_PGD; i < PTRS_PER_PGD; i++) { |
4190 | - unsigned long v = (unsigned long)i << PGDIR_SHIFT; |
4191 | - pgd_t *kpgd = pgd_offset_k(v); |
4192 | - pud_t *kpud = pud_offset(kpgd, v); |
4193 | - pmd_t *kpmd = pmd_offset(kpud, v); |
4194 | - memcpy(pmd[i], kpmd, PAGE_SIZE); |
4195 | + memcpy(pmds[i], |
4196 | + (void *)pgd_page_vaddr(swapper_pg_dir[i]), |
4197 | + sizeof(pmd_t) * PTRS_PER_PMD); |
4198 | make_lowmem_page_readonly( |
4199 | - pmd[i], XENFEAT_writable_page_tables); |
4200 | + pmds[i], XENFEAT_writable_page_tables); |
4201 | } |
4202 | |
4203 | /* It is safe to poke machine addresses of pmds under the pmd_lock. */ |
4204 | for (i = 0; i < PTRS_PER_PGD; i++) |
4205 | - set_pgd(&pgd[i], __pgd(1 + __pa(pmd[i]))); |
4206 | - |
4207 | - /* Ensure this pgd gets picked up and pinned on save/restore. */ |
4208 | - pgd_list_add(pgd); |
4209 | + set_pgd(&pgd[i], __pgd(1 + __pa(pmds[i]))); |
4210 | |
4211 | spin_unlock_irqrestore(&pgd_lock, flags); |
4212 | |
4213 | - kfree(pmd); |
4214 | + kfree(pmds); |
4215 | +#endif |
4216 | |
4217 | return pgd; |
4218 | |
4219 | out_oom: |
4220 | - if (HAVE_SHARED_KERNEL_PMD) { |
4221 | + if (!pmds) { |
4222 | for (i--; i >= 0; i--) { |
4223 | pgd_t pgdent = pgd[i]; |
4224 | void* pmd = (void *)__va(pgd_val(pgdent)-1); |
4225 | paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT); |
4226 | - kmem_cache_free(pmd_cache, pmd); |
4227 | + pmd_cache_free(pmd, i); |
4228 | } |
4229 | } else { |
4230 | for (i--; i >= 0; i--) { |
4231 | - paravirt_release_pd(__pa(pmd[i]) >> PAGE_SHIFT); |
4232 | - kmem_cache_free(pmd_cache, pmd[i]); |
4233 | + paravirt_release_pd(__pa(pmds[i]) >> PAGE_SHIFT); |
4234 | + pmd_cache_free(pmds[i], i); |
4235 | } |
4236 | - kfree(pmd); |
4237 | + kfree(pmds); |
4238 | } |
4239 | - kmem_cache_free(pgd_cache, pgd); |
4240 | + quicklist_free(0, pgd_dtor, pgd); |
4241 | return NULL; |
4242 | } |
4243 | |
4244 | @@ -392,35 +447,24 @@ |
4245 | |
4246 | /* in the PAE case user pgd entries are overwritten before usage */ |
4247 | if (PTRS_PER_PMD > 1) { |
4248 | - for (i = 0; i < USER_PTRS_PER_PGD; ++i) { |
4249 | + for (i = 0; i < UNSHARED_PTRS_PER_PGD; ++i) { |
4250 | pgd_t pgdent = pgd[i]; |
4251 | void* pmd = (void *)__va(pgd_val(pgdent)-1); |
4252 | paravirt_release_pd(__pa(pmd) >> PAGE_SHIFT); |
4253 | - kmem_cache_free(pmd_cache, pmd); |
4254 | + pmd_cache_free(pmd, i); |
4255 | } |
4256 | |
4257 | - if (!HAVE_SHARED_KERNEL_PMD) { |
4258 | - unsigned long flags; |
4259 | - spin_lock_irqsave(&pgd_lock, flags); |
4260 | - pgd_list_del(pgd); |
4261 | - spin_unlock_irqrestore(&pgd_lock, flags); |
4262 | - |
4263 | - for (i = USER_PTRS_PER_PGD; i < PTRS_PER_PGD; i++) { |
4264 | - pmd_t *pmd = (void *)__va(pgd_val(pgd[i])-1); |
4265 | - make_lowmem_page_writable( |
4266 | - pmd, XENFEAT_writable_page_tables); |
4267 | - memset(pmd, 0, PTRS_PER_PMD*sizeof(pmd_t)); |
4268 | - kmem_cache_free(pmd_cache, pmd); |
4269 | - } |
4270 | - |
4271 | - if (!xen_feature(XENFEAT_pae_pgdir_above_4gb)) |
4272 | - xen_destroy_contiguous_region( |
4273 | - (unsigned long)pgd, 0); |
4274 | - } |
4275 | + if (!xen_feature(XENFEAT_pae_pgdir_above_4gb)) |
4276 | + xen_destroy_contiguous_region((unsigned long)pgd, 0); |
4277 | } |
4278 | |
4279 | /* in the non-PAE case, free_pgtables() clears user pgd entries */ |
4280 | - kmem_cache_free(pgd_cache, pgd); |
4281 | + quicklist_free(0, pgd_dtor, pgd); |
4282 | +} |
4283 | + |
4284 | +void check_pgt_cache(void) |
4285 | +{ |
4286 | + quicklist_trim(0, pgd_dtor, 25, 16); |
4287 | } |
4288 | |
4289 | void make_lowmem_page_readonly(void *va, unsigned int feature) |
4290 | @@ -717,13 +761,13 @@ |
4291 | spin_unlock_irqrestore(&pgd_lock, flags); |
4292 | } |
4293 | |
4294 | -void _arch_dup_mmap(struct mm_struct *mm) |
4295 | +void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) |
4296 | { |
4297 | if (!test_bit(PG_pinned, &virt_to_page(mm->pgd)->flags)) |
4298 | mm_pin(mm); |
4299 | } |
4300 | |
4301 | -void _arch_exit_mmap(struct mm_struct *mm) |
4302 | +void arch_exit_mmap(struct mm_struct *mm) |
4303 | { |
4304 | struct task_struct *tsk = current; |
4305 | |
4306 | --- a/drivers/char/tpm/tpm_xen.c |
4307 | +++ b/drivers/char/tpm/tpm_xen.c |
4308 | @@ -463,7 +463,7 @@ |
4309 | tp->backend_id = domid; |
4310 | |
4311 | err = bind_listening_port_to_irqhandler( |
4312 | - domid, tpmif_int, SA_SAMPLE_RANDOM, "tpmif", tp); |
4313 | + domid, tpmif_int, IRQF_SAMPLE_RANDOM, "tpmif", tp); |
4314 | if (err <= 0) { |
4315 | WPRINTK("bind_listening_port_to_irqhandler failed " |
4316 | "(err=%d)\n", err); |
4317 | --- a/drivers/xen/blkfront/blkfront.c |
4318 | +++ b/drivers/xen/blkfront/blkfront.c |
4319 | @@ -236,7 +236,7 @@ |
4320 | info->ring_ref = err; |
4321 | |
4322 | err = bind_listening_port_to_irqhandler( |
4323 | - dev->otherend_id, blkif_int, SA_SAMPLE_RANDOM, "blkif", info); |
4324 | + dev->otherend_id, blkif_int, IRQF_SAMPLE_RANDOM, "blkif", info); |
4325 | if (err <= 0) { |
4326 | xenbus_dev_fatal(dev, err, |
4327 | "bind_listening_port_to_irqhandler"); |
4328 | --- a/drivers/xen/char/mem.c |
4329 | +++ b/drivers/xen/char/mem.c |
4330 | @@ -18,7 +18,6 @@ |
4331 | #include <linux/raw.h> |
4332 | #include <linux/tty.h> |
4333 | #include <linux/capability.h> |
4334 | -#include <linux/smp_lock.h> |
4335 | #include <linux/ptrace.h> |
4336 | #include <linux/device.h> |
4337 | #include <asm/pgalloc.h> |
4338 | --- a/drivers/xen/core/hypervisor_sysfs.c |
4339 | +++ b/drivers/xen/core/hypervisor_sysfs.c |
4340 | @@ -50,7 +50,7 @@ |
4341 | if (!is_running_on_xen()) |
4342 | return -ENODEV; |
4343 | |
4344 | - hypervisor_subsys.kset.kobj.ktype = &hyp_sysfs_kobj_type; |
4345 | + hypervisor_subsys.kobj.ktype = &hyp_sysfs_kobj_type; |
4346 | return 0; |
4347 | } |
4348 | |
4349 | --- a/drivers/xen/core/smpboot.c |
4350 | +++ b/drivers/xen/core/smpboot.c |
4351 | @@ -121,7 +121,7 @@ |
4352 | rc = bind_ipi_to_irqhandler(RESCHEDULE_VECTOR, |
4353 | cpu, |
4354 | smp_reschedule_interrupt, |
4355 | - SA_INTERRUPT, |
4356 | + IRQF_DISABLED, |
4357 | resched_name[cpu], |
4358 | NULL); |
4359 | if (rc < 0) |
4360 | @@ -132,7 +132,7 @@ |
4361 | rc = bind_ipi_to_irqhandler(CALL_FUNCTION_VECTOR, |
4362 | cpu, |
4363 | smp_call_function_interrupt, |
4364 | - SA_INTERRUPT, |
4365 | + IRQF_DISABLED, |
4366 | callfunc_name[cpu], |
4367 | NULL); |
4368 | if (rc < 0) |
4369 | @@ -165,13 +165,12 @@ |
4370 | |
4371 | void __cpuinit cpu_bringup(void) |
4372 | { |
4373 | + cpu_init(); |
4374 | #ifdef __i386__ |
4375 | - cpu_set_gdt(current_thread_info()->cpu); |
4376 | - secondary_cpu_init(); |
4377 | + identify_secondary_cpu(cpu_data + smp_processor_id()); |
4378 | #else |
4379 | - cpu_init(); |
4380 | -#endif |
4381 | identify_cpu(cpu_data + smp_processor_id()); |
4382 | +#endif |
4383 | touch_softlockup_watchdog(); |
4384 | preempt_disable(); |
4385 | local_irq_enable(); |
4386 | @@ -191,11 +190,6 @@ |
4387 | static DEFINE_SPINLOCK(ctxt_lock); |
4388 | |
4389 | struct task_struct *idle = idle_task(cpu); |
4390 | -#ifdef __x86_64__ |
4391 | - struct desc_ptr *gdt_descr = &cpu_gdt_descr[cpu]; |
4392 | -#else |
4393 | - struct Xgt_desc_struct *gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
4394 | -#endif |
4395 | |
4396 | if (cpu_test_and_set(cpu, cpu_initialized_map)) |
4397 | return; |
4398 | @@ -218,11 +212,11 @@ |
4399 | smp_trap_init(ctxt.trap_ctxt); |
4400 | |
4401 | ctxt.ldt_ents = 0; |
4402 | - |
4403 | - ctxt.gdt_frames[0] = virt_to_mfn(gdt_descr->address); |
4404 | - ctxt.gdt_ents = gdt_descr->size / 8; |
4405 | + ctxt.gdt_ents = GDT_SIZE / 8; |
4406 | |
4407 | #ifdef __i386__ |
4408 | + ctxt.gdt_frames[0] = virt_to_mfn(get_cpu_gdt_table(cpu)); |
4409 | + |
4410 | ctxt.user_regs.cs = __KERNEL_CS; |
4411 | ctxt.user_regs.esp = idle->thread.esp0 - sizeof(struct pt_regs); |
4412 | |
4413 | @@ -235,7 +229,11 @@ |
4414 | ctxt.failsafe_callback_eip = (unsigned long)failsafe_callback; |
4415 | |
4416 | ctxt.ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir)); |
4417 | + |
4418 | + ctxt.user_regs.fs = __KERNEL_PERCPU; |
4419 | #else /* __x86_64__ */ |
4420 | + ctxt.gdt_frames[0] = virt_to_mfn(cpu_gdt_descr[cpu].address); |
4421 | + |
4422 | ctxt.user_regs.cs = __KERNEL_CS; |
4423 | ctxt.user_regs.esp = idle->thread.rsp0 - sizeof(struct pt_regs); |
4424 | |
4425 | @@ -265,9 +263,8 @@ |
4426 | struct vcpu_get_physid cpu_id; |
4427 | #ifdef __x86_64__ |
4428 | struct desc_ptr *gdt_descr; |
4429 | -#else |
4430 | - struct Xgt_desc_struct *gdt_descr; |
4431 | #endif |
4432 | + void *gdt_addr; |
4433 | |
4434 | apicid = 0; |
4435 | if (HYPERVISOR_vcpu_op(VCPUOP_get_physid, 0, &cpu_id) == 0) |
4436 | @@ -317,14 +314,12 @@ |
4437 | } |
4438 | gdt_descr->size = GDT_SIZE; |
4439 | memcpy((void *)gdt_descr->address, cpu_gdt_table, GDT_SIZE); |
4440 | + gdt_addr = (void *)gdt_descr->address; |
4441 | #else |
4442 | - if (unlikely(!init_gdt(cpu, idle))) |
4443 | - continue; |
4444 | - gdt_descr = &per_cpu(cpu_gdt_descr, cpu); |
4445 | + init_gdt(cpu); |
4446 | + gdt_addr = get_cpu_gdt_table(cpu); |
4447 | #endif |
4448 | - make_page_readonly( |
4449 | - (void *)gdt_descr->address, |
4450 | - XENFEAT_writable_descriptor_tables); |
4451 | + make_page_readonly(gdt_addr, XENFEAT_writable_descriptor_tables); |
4452 | |
4453 | apicid = cpu; |
4454 | if (HYPERVISOR_vcpu_op(VCPUOP_get_physid, cpu, &cpu_id) == 0) |
4455 | @@ -338,7 +333,9 @@ |
4456 | #ifdef __x86_64__ |
4457 | cpu_pda(cpu)->pcurrent = idle; |
4458 | cpu_pda(cpu)->cpunumber = cpu; |
4459 | - clear_ti_thread_flag(idle->thread_info, TIF_FORK); |
4460 | + clear_ti_thread_flag(task_thread_info(idle), TIF_FORK); |
4461 | +#else |
4462 | + per_cpu(current_task, cpu) = idle; |
4463 | #endif |
4464 | |
4465 | irq_ctx_init(cpu); |
4466 | @@ -363,8 +360,12 @@ |
4467 | #endif |
4468 | } |
4469 | |
4470 | -void __devinit smp_prepare_boot_cpu(void) |
4471 | +void __init smp_prepare_boot_cpu(void) |
4472 | { |
4473 | +#ifdef __i386__ |
4474 | + init_gdt(smp_processor_id()); |
4475 | + switch_to_new_gdt(); |
4476 | +#endif |
4477 | prefill_possible_map(); |
4478 | } |
4479 | |
4480 | --- a/drivers/xen/core/xen_sysfs.c |
4481 | +++ b/drivers/xen/core/xen_sysfs.c |
4482 | @@ -28,12 +28,12 @@ |
4483 | |
4484 | static int __init xen_sysfs_type_init(void) |
4485 | { |
4486 | - return sysfs_create_file(&hypervisor_subsys.kset.kobj, &type_attr.attr); |
4487 | + return sysfs_create_file(&hypervisor_subsys.kobj, &type_attr.attr); |
4488 | } |
4489 | |
4490 | static void xen_sysfs_type_destroy(void) |
4491 | { |
4492 | - sysfs_remove_file(&hypervisor_subsys.kset.kobj, &type_attr.attr); |
4493 | + sysfs_remove_file(&hypervisor_subsys.kobj, &type_attr.attr); |
4494 | } |
4495 | |
4496 | /* xen version attributes */ |
4497 | @@ -89,13 +89,13 @@ |
4498 | |
4499 | static int __init xen_sysfs_version_init(void) |
4500 | { |
4501 | - return sysfs_create_group(&hypervisor_subsys.kset.kobj, |
4502 | + return sysfs_create_group(&hypervisor_subsys.kobj, |
4503 | &version_group); |
4504 | } |
4505 | |
4506 | static void xen_sysfs_version_destroy(void) |
4507 | { |
4508 | - sysfs_remove_group(&hypervisor_subsys.kset.kobj, &version_group); |
4509 | + sysfs_remove_group(&hypervisor_subsys.kobj, &version_group); |
4510 | } |
4511 | |
4512 | /* UUID */ |
4513 | @@ -125,12 +125,12 @@ |
4514 | |
4515 | static int __init xen_sysfs_uuid_init(void) |
4516 | { |
4517 | - return sysfs_create_file(&hypervisor_subsys.kset.kobj, &uuid_attr.attr); |
4518 | + return sysfs_create_file(&hypervisor_subsys.kobj, &uuid_attr.attr); |
4519 | } |
4520 | |
4521 | static void xen_sysfs_uuid_destroy(void) |
4522 | { |
4523 | - sysfs_remove_file(&hypervisor_subsys.kset.kobj, &uuid_attr.attr); |
4524 | + sysfs_remove_file(&hypervisor_subsys.kobj, &uuid_attr.attr); |
4525 | } |
4526 | |
4527 | /* xen compilation attributes */ |
4528 | @@ -203,13 +203,13 @@ |
4529 | |
4530 | int __init static xen_compilation_init(void) |
4531 | { |
4532 | - return sysfs_create_group(&hypervisor_subsys.kset.kobj, |
4533 | + return sysfs_create_group(&hypervisor_subsys.kobj, |
4534 | &xen_compilation_group); |
4535 | } |
4536 | |
4537 | static void xen_compilation_destroy(void) |
4538 | { |
4539 | - sysfs_remove_group(&hypervisor_subsys.kset.kobj, |
4540 | + sysfs_remove_group(&hypervisor_subsys.kobj, |
4541 | &xen_compilation_group); |
4542 | } |
4543 | |
4544 | @@ -324,13 +324,13 @@ |
4545 | |
4546 | static int __init xen_properties_init(void) |
4547 | { |
4548 | - return sysfs_create_group(&hypervisor_subsys.kset.kobj, |
4549 | + return sysfs_create_group(&hypervisor_subsys.kobj, |
4550 | &xen_properties_group); |
4551 | } |
4552 | |
4553 | static void xen_properties_destroy(void) |
4554 | { |
4555 | - sysfs_remove_group(&hypervisor_subsys.kset.kobj, |
4556 | + sysfs_remove_group(&hypervisor_subsys.kobj, |
4557 | &xen_properties_group); |
4558 | } |
4559 | |
4560 | --- a/drivers/xen/netback/netback.c |
4561 | +++ b/drivers/xen/netback/netback.c |
4562 | @@ -180,7 +180,7 @@ |
4563 | goto err; |
4564 | |
4565 | skb_reserve(nskb, 16 + NET_IP_ALIGN); |
4566 | - headlen = nskb->end - nskb->data; |
4567 | + headlen = skb_end_pointer(nskb) - nskb->data; |
4568 | if (headlen > skb_headlen(skb)) |
4569 | headlen = skb_headlen(skb); |
4570 | ret = skb_copy_bits(skb, 0, __skb_put(nskb, headlen), headlen); |
4571 | @@ -226,11 +226,15 @@ |
4572 | len -= copy; |
4573 | } |
4574 | |
4575 | +#ifdef NET_SKBUFF_DATA_USES_OFFSET |
4576 | + offset = 0; |
4577 | +#else |
4578 | offset = nskb->data - skb->data; |
4579 | +#endif |
4580 | |
4581 | - nskb->h.raw = skb->h.raw + offset; |
4582 | - nskb->nh.raw = skb->nh.raw + offset; |
4583 | - nskb->mac.raw = skb->mac.raw + offset; |
4584 | + nskb->transport_header = skb->transport_header + offset; |
4585 | + nskb->network_header = skb->network_header + offset; |
4586 | + nskb->mac_header = skb->mac_header + offset; |
4587 | |
4588 | return nskb; |
4589 | |
4590 | @@ -1601,7 +1605,7 @@ |
4591 | (void)bind_virq_to_irqhandler(VIRQ_DEBUG, |
4592 | 0, |
4593 | netif_be_dbg, |
4594 | - SA_SHIRQ, |
4595 | + IRQF_SHARED, |
4596 | "net-be-dbg", |
4597 | &netif_be_dbg); |
4598 | #endif |
4599 | --- a/drivers/xen/netfront/netfront.c |
4600 | +++ b/drivers/xen/netfront/netfront.c |
4601 | @@ -513,7 +513,7 @@ |
4602 | memcpy(netdev->dev_addr, info->mac, ETH_ALEN); |
4603 | |
4604 | err = bind_listening_port_to_irqhandler( |
4605 | - dev->otherend_id, netif_int, SA_SAMPLE_RANDOM, netdev->name, |
4606 | + dev->otherend_id, netif_int, IRQF_SAMPLE_RANDOM, netdev->name, |
4607 | netdev); |
4608 | if (err < 0) |
4609 | goto fail; |
4610 | --- a/drivers/xen/pciback/xenbus.c |
4611 | +++ b/drivers/xen/pciback/xenbus.c |
4612 | @@ -86,7 +86,7 @@ |
4613 | |
4614 | err = bind_interdomain_evtchn_to_irqhandler( |
4615 | pdev->xdev->otherend_id, remote_evtchn, pciback_handle_event, |
4616 | - SA_SAMPLE_RANDOM, "pciback", pdev); |
4617 | + IRQF_SAMPLE_RANDOM, "pciback", pdev); |
4618 | if (err < 0) { |
4619 | xenbus_dev_fatal(pdev->xdev, err, |
4620 | "Error binding event channel to IRQ"); |
4621 | --- a/drivers/xen/pcifront/xenbus.c |
4622 | +++ b/drivers/xen/pcifront/xenbus.c |
4623 | @@ -10,10 +10,6 @@ |
4624 | #include <xen/gnttab.h> |
4625 | #include "pcifront.h" |
4626 | |
4627 | -#ifndef __init_refok |
4628 | -#define __init_refok |
4629 | -#endif |
4630 | - |
4631 | #define INVALID_GRANT_REF (0) |
4632 | #define INVALID_EVTCHN (-1) |
4633 | |
4634 | --- a/drivers/xen/sfc_netback/accel_fwd.c |
4635 | +++ b/drivers/xen/sfc_netback/accel_fwd.c |
4636 | @@ -308,7 +308,7 @@ |
4637 | static inline int packet_is_arp_reply(struct sk_buff *skb) |
4638 | { |
4639 | return skb->protocol == ntohs(ETH_P_ARP) |
4640 | - && skb->nh.arph->ar_op == ntohs(ARPOP_REPLY); |
4641 | + && arp_hdr(skb)->ar_op == ntohs(ARPOP_REPLY); |
4642 | } |
4643 | |
4644 | |
4645 | @@ -392,12 +392,13 @@ |
4646 | |
4647 | BUG_ON(fwd_priv == NULL); |
4648 | |
4649 | - if (is_broadcast_ether_addr(skb->mac.raw) && packet_is_arp_reply(skb)) { |
4650 | + if (is_broadcast_ether_addr(skb_mac_header(skb)) |
4651 | + && packet_is_arp_reply(skb)) { |
4652 | /* |
4653 | * update our fast path forwarding to reflect this |
4654 | * gratuitous ARP |
4655 | */ |
4656 | - mac = skb->mac.raw+ETH_ALEN; |
4657 | + mac = skb_mac_header(skb)+ETH_ALEN; |
4658 | |
4659 | DPRINTK("%s: found gratuitous ARP for " MAC_FMT "\n", |
4660 | __FUNCTION__, MAC_ARG(mac)); |
4661 | --- a/drivers/xen/sfc_netback/accel_solarflare.c |
4662 | +++ b/drivers/xen/sfc_netback/accel_solarflare.c |
4663 | @@ -114,7 +114,7 @@ |
4664 | BUG_ON(port == NULL); |
4665 | |
4666 | NETBACK_ACCEL_STATS_OP(global_stats.dl_tx_packets++); |
4667 | - if (skb->mac.raw != NULL) |
4668 | + if (skb_mac_header_was_set(skb)) |
4669 | netback_accel_tx_packet(skb, port->fwd_priv); |
4670 | else { |
4671 | DPRINTK("Ignoring packet with missing mac address\n"); |
4672 | --- a/drivers/xen/sfc_netfront/accel_tso.c |
4673 | +++ b/drivers/xen/sfc_netfront/accel_tso.c |
4674 | @@ -33,10 +33,9 @@ |
4675 | |
4676 | #include "accel_tso.h" |
4677 | |
4678 | -#define PTR_DIFF(p1, p2) ((u8*)(p1) - (u8*)(p2)) |
4679 | -#define ETH_HDR_LEN(skb) ((skb)->nh.raw - (skb)->data) |
4680 | -#define SKB_TCP_OFF(skb) PTR_DIFF ((skb)->h.th, (skb)->data) |
4681 | -#define SKB_IP_OFF(skb) PTR_DIFF ((skb)->nh.iph, (skb)->data) |
4682 | +#define ETH_HDR_LEN(skb) skb_network_offset(skb) |
4683 | +#define SKB_TCP_OFF(skb) skb_transport_offset(skb) |
4684 | +#define SKB_IP_OFF(skb) skb_network_offset(skb) |
4685 | |
4686 | /* |
4687 | * Set a maximum number of buffers in each output packet to make life |
4688 | @@ -114,9 +113,8 @@ |
4689 | static inline void tso_check_safe(struct sk_buff *skb) { |
4690 | EPRINTK_ON(skb->protocol != htons (ETH_P_IP)); |
4691 | EPRINTK_ON(((struct ethhdr*) skb->data)->h_proto != htons (ETH_P_IP)); |
4692 | - EPRINTK_ON(skb->nh.iph->protocol != IPPROTO_TCP); |
4693 | - EPRINTK_ON((SKB_TCP_OFF(skb) |
4694 | - + (skb->h.th->doff << 2u)) > skb_headlen(skb)); |
4695 | + EPRINTK_ON(ip_hdr(skb)->protocol != IPPROTO_TCP); |
4696 | + EPRINTK_ON((SKB_TCP_OFF(skb) + tcp_hdrlen(skb)) > skb_headlen(skb)); |
4697 | } |
4698 | |
4699 | |
4700 | @@ -129,17 +127,17 @@ |
4701 | * All ethernet/IP/TCP headers combined size is TCP header size |
4702 | * plus offset of TCP header relative to start of packet. |
4703 | */ |
4704 | - st->p.header_length = (skb->h.th->doff << 2u) + SKB_TCP_OFF(skb); |
4705 | + st->p.header_length = tcp_hdrlen(skb) + SKB_TCP_OFF(skb); |
4706 | st->p.full_packet_size = (st->p.header_length |
4707 | + skb_shinfo(skb)->gso_size); |
4708 | st->p.gso_size = skb_shinfo(skb)->gso_size; |
4709 | |
4710 | - st->p.ip_id = htons(skb->nh.iph->id); |
4711 | - st->seqnum = ntohl(skb->h.th->seq); |
4712 | + st->p.ip_id = htons(ip_hdr(skb)->id); |
4713 | + st->seqnum = ntohl(tcp_hdr(skb)->seq); |
4714 | |
4715 | - EPRINTK_ON(skb->h.th->urg); |
4716 | - EPRINTK_ON(skb->h.th->syn); |
4717 | - EPRINTK_ON(skb->h.th->rst); |
4718 | + EPRINTK_ON(tcp_hdr(skb)->urg); |
4719 | + EPRINTK_ON(tcp_hdr(skb)->syn); |
4720 | + EPRINTK_ON(tcp_hdr(skb)->rst); |
4721 | |
4722 | st->remaining_len = skb->len - st->p.header_length; |
4723 | |
4724 | @@ -258,8 +256,8 @@ |
4725 | /* This packet will be the last in the TSO burst. */ |
4726 | ip_length = (st->p.header_length - ETH_HDR_LEN(skb) |
4727 | + st->remaining_len); |
4728 | - tsoh_th->fin = skb->h.th->fin; |
4729 | - tsoh_th->psh = skb->h.th->psh; |
4730 | + tsoh_th->fin = tcp_hdr(skb)->fin; |
4731 | + tsoh_th->psh = tcp_hdr(skb)->psh; |
4732 | } |
4733 | |
4734 | tsoh_iph->tot_len = htons(ip_length); |
4735 | --- a/drivers/xen/sfc_netfront/accel_vi.c |
4736 | +++ b/drivers/xen/sfc_netfront/accel_vi.c |
4737 | @@ -463,7 +463,7 @@ |
4738 | |
4739 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
4740 | /* Set to zero to encourage falcon to work it out for us */ |
4741 | - *(u16*)(skb->h.raw + skb->csum_offset) = 0; |
4742 | + *(u16*)(skb->head + skb->csum_start + skb->csum_offset) = 0; |
4743 | } |
4744 | |
4745 | if (multi_post_start_new_buffer(vnic, &state)) { |
4746 | @@ -582,7 +582,7 @@ |
4747 | |
4748 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
4749 | /* Set to zero to encourage falcon to work it out for us */ |
4750 | - *(u16*)(skb->h.raw + skb->csum_offset) = 0; |
4751 | + *(u16*)(skb->head + skb->csum_start + skb->csum_offset) = 0; |
4752 | } |
4753 | NETFRONT_ACCEL_PKTBUFF_FOR_EACH_FRAGMENT |
4754 | (skb, idx, frag_data, frag_len, { |
4755 | --- a/drivers/xen/sfc_netfront/accel_xenbus.c |
4756 | +++ b/drivers/xen/sfc_netfront/accel_xenbus.c |
4757 | @@ -356,7 +356,7 @@ |
4758 | /* Create xenbus msg event channel */ |
4759 | err = bind_listening_port_to_irqhandler |
4760 | (dev->otherend_id, netfront_accel_msg_channel_irq_from_bend, |
4761 | - SA_SAMPLE_RANDOM, "vnicctrl", vnic); |
4762 | + IRQF_SAMPLE_RANDOM, "vnicctrl", vnic); |
4763 | if (err < 0) { |
4764 | EPRINTK("Couldn't bind msg event channel\n"); |
4765 | goto fail_msg_irq; |
4766 | @@ -367,7 +367,7 @@ |
4767 | /* Create xenbus net event channel */ |
4768 | err = bind_listening_port_to_irqhandler |
4769 | (dev->otherend_id, netfront_accel_net_channel_irq_from_bend, |
4770 | - SA_SAMPLE_RANDOM, "vnicfront", vnic); |
4771 | + IRQF_SAMPLE_RANDOM, "vnicfront", vnic); |
4772 | if (err < 0) { |
4773 | EPRINTK("Couldn't bind net event channel\n"); |
4774 | goto fail_net_irq; |
4775 | --- a/drivers/xen/xenoprof/xenoprofile.c |
4776 | +++ b/drivers/xen/xenoprof/xenoprofile.c |
4777 | @@ -236,7 +236,7 @@ |
4778 | result = bind_virq_to_irqhandler(VIRQ_XENOPROF, |
4779 | i, |
4780 | xenoprof_ovf_interrupt, |
4781 | - SA_INTERRUPT, |
4782 | + IRQF_DISABLED, |
4783 | "xenoprof", |
4784 | NULL); |
4785 | |
4786 | --- a/fs/aio.c |
4787 | +++ b/fs/aio.c |
4788 | @@ -38,7 +38,7 @@ |
4789 | |
4790 | #ifdef CONFIG_EPOLL |
4791 | #include <linux/poll.h> |
4792 | -#include <linux/eventpoll.h> |
4793 | +#include <linux/anon_inodes.h> |
4794 | #endif |
4795 | |
4796 | #if DEBUG > 1 |
4797 | @@ -1308,7 +1308,7 @@ |
4798 | |
4799 | /* make_aio_fd: |
4800 | * Create a file descriptor that can be used to poll the event queue. |
4801 | - * Based and piggybacked on the excellent epoll code. |
4802 | + * Based on the excellent epoll code. |
4803 | */ |
4804 | |
4805 | static int make_aio_fd(struct kioctx *ioctx) |
4806 | @@ -1317,7 +1317,8 @@ |
4807 | struct inode *inode; |
4808 | struct file *file; |
4809 | |
4810 | - error = ep_getfd(&fd, &inode, &file, NULL, &aioq_fops); |
4811 | + error = anon_inode_getfd(&fd, &inode, &file, "[aioq]", |
4812 | + &aioq_fops, ioctx); |
4813 | if (error) |
4814 | return error; |
4815 | |
4816 | --- a/include/asm-x86/mach-xen/asm/desc_32.h |
4817 | +++ b/include/asm-x86/mach-xen/asm/desc_32.h |
4818 | @@ -11,23 +11,24 @@ |
4819 | |
4820 | #include <asm/mmu.h> |
4821 | |
4822 | -extern struct desc_struct cpu_gdt_table[GDT_ENTRIES]; |
4823 | - |
4824 | struct Xgt_desc_struct { |
4825 | unsigned short size; |
4826 | unsigned long address __attribute__((packed)); |
4827 | unsigned short pad; |
4828 | } __attribute__ ((packed)); |
4829 | |
4830 | -extern struct Xgt_desc_struct idt_descr; |
4831 | -DECLARE_PER_CPU(struct Xgt_desc_struct, cpu_gdt_descr); |
4832 | -extern struct Xgt_desc_struct early_gdt_descr; |
4833 | +struct gdt_page |
4834 | +{ |
4835 | + struct desc_struct gdt[GDT_ENTRIES]; |
4836 | +} __attribute__((aligned(PAGE_SIZE))); |
4837 | +DECLARE_PER_CPU(struct gdt_page, gdt_page); |
4838 | |
4839 | static inline struct desc_struct *get_cpu_gdt_table(unsigned int cpu) |
4840 | { |
4841 | - return (struct desc_struct *)per_cpu(cpu_gdt_descr, cpu).address; |
4842 | + return per_cpu(gdt_page, cpu).gdt; |
4843 | } |
4844 | |
4845 | +extern struct Xgt_desc_struct idt_descr; |
4846 | extern struct desc_struct idt_table[]; |
4847 | extern void set_intr_gate(unsigned int irq, void * addr); |
4848 | |
4849 | @@ -55,53 +56,32 @@ |
4850 | #define DESCTYPE_S 0x10 /* !system */ |
4851 | |
4852 | #ifndef CONFIG_XEN |
4853 | -#define load_TR_desc() __asm__ __volatile__("ltr %w0"::"q" (GDT_ENTRY_TSS*8)) |
4854 | - |
4855 | -#define load_gdt(dtr) __asm__ __volatile("lgdt %0"::"m" (*dtr)) |
4856 | -#define load_idt(dtr) __asm__ __volatile("lidt %0"::"m" (*dtr)) |
4857 | +#define load_TR_desc() native_load_tr_desc() |
4858 | +#define load_gdt(dtr) native_load_gdt(dtr) |
4859 | +#define load_idt(dtr) native_load_idt(dtr) |
4860 | #define load_tr(tr) __asm__ __volatile("ltr %0"::"m" (tr)) |
4861 | #define load_ldt(ldt) __asm__ __volatile("lldt %0"::"m" (ldt)) |
4862 | |
4863 | -#define store_gdt(dtr) __asm__ ("sgdt %0":"=m" (*dtr)) |
4864 | -#define store_idt(dtr) __asm__ ("sidt %0":"=m" (*dtr)) |
4865 | -#define store_tr(tr) __asm__ ("str %0":"=m" (tr)) |
4866 | +#define store_gdt(dtr) native_store_gdt(dtr) |
4867 | +#define store_idt(dtr) native_store_idt(dtr) |
4868 | +#define store_tr(tr) (tr = native_store_tr()) |
4869 | #define store_ldt(ldt) __asm__ ("sldt %0":"=m" (ldt)) |
4870 | -#endif |
4871 | |
4872 | -#if TLS_SIZE != 24 |
4873 | -# error update this code. |
4874 | -#endif |
4875 | - |
4876 | -static inline void load_TLS(struct thread_struct *t, unsigned int cpu) |
4877 | -{ |
4878 | -#define C(i) if (HYPERVISOR_update_descriptor(virt_to_machine(&get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN + i]), \ |
4879 | - *(u64 *)&t->tls_array[i]) \ |
4880 | - BUG() |
4881 | - C(0); C(1); C(2); |
4882 | -#undef C |
4883 | -} |
4884 | +#define load_TLS(t, cpu) native_load_tls(t, cpu) |
4885 | +#define set_ldt native_set_ldt |
4886 | |
4887 | -#ifndef CONFIG_XEN |
4888 | #define write_ldt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b) |
4889 | #define write_gdt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b) |
4890 | #define write_idt_entry(dt, entry, a, b) write_dt_entry(dt, entry, a, b) |
4891 | |
4892 | -static inline void write_dt_entry(void *dt, int entry, __u32 entry_a, __u32 entry_b) |
4893 | +static inline void write_dt_entry(struct desc_struct *dt, |
4894 | + int entry, u32 entry_low, u32 entry_high) |
4895 | { |
4896 | - __u32 *lp = (__u32 *)((char *)dt + entry*8); |
4897 | - *lp = entry_a; |
4898 | - *(lp+1) = entry_b; |
4899 | + dt[entry].a = entry_low; |
4900 | + dt[entry].b = entry_high; |
4901 | } |
4902 | -#define set_ldt native_set_ldt |
4903 | -#else |
4904 | -extern int write_ldt_entry(void *ldt, int entry, __u32 entry_a, __u32 entry_b); |
4905 | -extern int write_gdt_entry(void *gdt, int entry, __u32 entry_a, __u32 entry_b); |
4906 | -#define set_ldt xen_set_ldt |
4907 | -#endif |
4908 | |
4909 | -#ifndef CONFIG_XEN |
4910 | -static inline fastcall void native_set_ldt(const void *addr, |
4911 | - unsigned int entries) |
4912 | +static inline void native_set_ldt(const void *addr, unsigned int entries) |
4913 | { |
4914 | if (likely(entries == 0)) |
4915 | __asm__ __volatile__("lldt %w0"::"q" (0)); |
4916 | @@ -116,6 +96,65 @@ |
4917 | __asm__ __volatile__("lldt %w0"::"q" (GDT_ENTRY_LDT*8)); |
4918 | } |
4919 | } |
4920 | + |
4921 | + |
4922 | +static inline void native_load_tr_desc(void) |
4923 | +{ |
4924 | + asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8)); |
4925 | +} |
4926 | + |
4927 | +static inline void native_load_gdt(const struct Xgt_desc_struct *dtr) |
4928 | +{ |
4929 | + asm volatile("lgdt %0"::"m" (*dtr)); |
4930 | +} |
4931 | + |
4932 | +static inline void native_load_idt(const struct Xgt_desc_struct *dtr) |
4933 | +{ |
4934 | + asm volatile("lidt %0"::"m" (*dtr)); |
4935 | +} |
4936 | + |
4937 | +static inline void native_store_gdt(struct Xgt_desc_struct *dtr) |
4938 | +{ |
4939 | + asm ("sgdt %0":"=m" (*dtr)); |
4940 | +} |
4941 | + |
4942 | +static inline void native_store_idt(struct Xgt_desc_struct *dtr) |
4943 | +{ |
4944 | + asm ("sidt %0":"=m" (*dtr)); |
4945 | +} |
4946 | + |
4947 | +static inline unsigned long native_store_tr(void) |
4948 | +{ |
4949 | + unsigned long tr; |
4950 | + asm ("str %0":"=r" (tr)); |
4951 | + return tr; |
4952 | +} |
4953 | + |
4954 | +static inline void native_load_tls(struct thread_struct *t, unsigned int cpu) |
4955 | +{ |
4956 | + unsigned int i; |
4957 | + struct desc_struct *gdt = get_cpu_gdt_table(cpu); |
4958 | + |
4959 | + for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++) |
4960 | + gdt[GDT_ENTRY_TLS_MIN + i] = t->tls_array[i]; |
4961 | +} |
4962 | +#else |
4963 | +#define load_TLS(t, cpu) xen_load_tls(t, cpu) |
4964 | +#define set_ldt xen_set_ldt |
4965 | + |
4966 | +extern int write_ldt_entry(void *ldt, int entry, __u32 entry_a, __u32 entry_b); |
4967 | +extern int write_gdt_entry(void *gdt, int entry, __u32 entry_a, __u32 entry_b); |
4968 | + |
4969 | +static inline void xen_load_tls(struct thread_struct *t, unsigned int cpu) |
4970 | +{ |
4971 | + unsigned int i; |
4972 | + struct desc_struct *gdt = get_cpu_gdt_table(cpu) + GDT_ENTRY_TLS_MIN; |
4973 | + |
4974 | + for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++) |
4975 | + if (HYPERVISOR_update_descriptor(virt_to_machine(&gdt[i]), |
4976 | + *(u64 *)&t->tls_array[i])) |
4977 | + BUG(); |
4978 | +} |
4979 | #endif |
4980 | |
4981 | #ifndef CONFIG_X86_NO_IDT |
4982 | --- a/include/asm-x86/mach-xen/asm/desc_64.h |
4983 | +++ b/include/asm-x86/mach-xen/asm/desc_64.h |
4984 | @@ -127,16 +127,6 @@ |
4985 | DESC_LDT, size * 8 - 1); |
4986 | } |
4987 | |
4988 | -static inline void set_seg_base(unsigned cpu, int entry, void *base) |
4989 | -{ |
4990 | - struct desc_struct *d = &cpu_gdt(cpu)[entry]; |
4991 | - u32 addr = (u32)(u64)base; |
4992 | - BUG_ON((u64)base >> 32); |
4993 | - d->base0 = addr & 0xffff; |
4994 | - d->base1 = (addr >> 16) & 0xff; |
4995 | - d->base2 = (addr >> 24) & 0xff; |
4996 | -} |
4997 | - |
4998 | #define LDT_entry_a(info) \ |
4999 | ((((info)->base_addr & 0x0000ffff) << 16) | ((info)->limit & 0x0ffff)) |
5000 | /* Don't allow setting of the lm bit. It is useless anyways because |
5001 | @@ -165,25 +155,15 @@ |
5002 | (info)->useable == 0 && \ |
5003 | (info)->lm == 0) |
5004 | |
5005 | -#if TLS_SIZE != 24 |
5006 | -# error update this code. |
5007 | -#endif |
5008 | - |
5009 | static inline void load_TLS(struct thread_struct *t, unsigned int cpu) |
5010 | { |
5011 | -#if 0 |
5012 | + unsigned int i; |
5013 | u64 *gdt = (u64 *)(cpu_gdt(cpu) + GDT_ENTRY_TLS_MIN); |
5014 | - gdt[0] = t->tls_array[0]; |
5015 | - gdt[1] = t->tls_array[1]; |
5016 | - gdt[2] = t->tls_array[2]; |
5017 | -#endif |
5018 | -#define C(i) \ |
5019 | - if (HYPERVISOR_update_descriptor(virt_to_machine(&cpu_gdt(cpu)[GDT_ENTRY_TLS_MIN + i]), \ |
5020 | - t->tls_array[i])) \ |
5021 | - BUG(); |
5022 | |
5023 | - C(0); C(1); C(2); |
5024 | -#undef C |
5025 | + for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++) |
5026 | + if (HYPERVISOR_update_descriptor(virt_to_machine(&gdt[i]), |
5027 | + t->tls_array[i])) |
5028 | + BUG(); |
5029 | } |
5030 | |
5031 | /* |
5032 | --- a/include/asm-x86/mach-xen/asm/dma-mapping_64.h |
5033 | +++ b/include/asm-x86/mach-xen/asm/dma-mapping_64.h |
5034 | @@ -51,7 +51,7 @@ |
5035 | }; |
5036 | |
5037 | extern dma_addr_t bad_dma_address; |
5038 | -extern struct dma_mapping_ops* dma_ops; |
5039 | +extern const struct dma_mapping_ops* dma_ops; |
5040 | extern int iommu_merge; |
5041 | |
5042 | #if 0 |
5043 | --- a/include/asm-x86/mach-xen/asm/fixmap_32.h |
5044 | +++ b/include/asm-x86/mach-xen/asm/fixmap_32.h |
5045 | @@ -19,10 +19,8 @@ |
5046 | * the start of the fixmap. |
5047 | */ |
5048 | extern unsigned long __FIXADDR_TOP; |
5049 | -#ifdef CONFIG_COMPAT_VDSO |
5050 | -#define FIXADDR_USER_START __fix_to_virt(FIX_VDSO) |
5051 | -#define FIXADDR_USER_END __fix_to_virt(FIX_VDSO - 1) |
5052 | -#endif |
5053 | +#define FIXADDR_USER_START __fix_to_virt(FIX_VDSO) |
5054 | +#define FIXADDR_USER_END __fix_to_virt(FIX_VDSO - 1) |
5055 | |
5056 | #ifndef __ASSEMBLY__ |
5057 | #include <linux/kernel.h> |
5058 | @@ -85,6 +83,9 @@ |
5059 | #ifdef CONFIG_PCI_MMCONFIG |
5060 | FIX_PCIE_MCFG, |
5061 | #endif |
5062 | +#ifdef CONFIG_PARAVIRT |
5063 | + FIX_PARAVIRT_BOOTMAP, |
5064 | +#endif |
5065 | FIX_SHARED_INFO, |
5066 | #define NR_FIX_ISAMAPS 256 |
5067 | FIX_ISAMAP_END, |
5068 | --- a/include/asm-x86/mach-xen/asm/fixmap_64.h |
5069 | +++ b/include/asm-x86/mach-xen/asm/fixmap_64.h |
5070 | @@ -15,7 +15,6 @@ |
5071 | #include <asm/apicdef.h> |
5072 | #include <asm/page.h> |
5073 | #include <asm/vsyscall.h> |
5074 | -#include <asm/vsyscall32.h> |
5075 | #include <asm/acpi.h> |
5076 | |
5077 | /* |
5078 | --- a/include/asm-x86/mach-xen/asm/highmem.h |
5079 | +++ b/include/asm-x86/mach-xen/asm/highmem.h |
5080 | @@ -67,12 +67,18 @@ |
5081 | |
5082 | void *kmap(struct page *page); |
5083 | void kunmap(struct page *page); |
5084 | +void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot); |
5085 | void *kmap_atomic(struct page *page, enum km_type type); |
5086 | void *kmap_atomic_pte(struct page *page, enum km_type type); |
5087 | void kunmap_atomic(void *kvaddr, enum km_type type); |
5088 | void *kmap_atomic_pfn(unsigned long pfn, enum km_type type); |
5089 | struct page *kmap_atomic_to_page(void *ptr); |
5090 | |
5091 | +#define kmap_atomic_pte(page, type) \ |
5092 | + kmap_atomic_prot(page, type, \ |
5093 | + test_bit(PG_pinned, &(page)->flags) \ |
5094 | + ? PAGE_KERNEL_RO : kmap_prot) |
5095 | + |
5096 | #define flush_cache_kmaps() do { } while (0) |
5097 | |
5098 | #endif /* __KERNEL__ */ |
5099 | --- a/include/asm-x86/mach-xen/asm/io_32.h |
5100 | +++ b/include/asm-x86/mach-xen/asm/io_32.h |
5101 | @@ -263,15 +263,18 @@ |
5102 | |
5103 | #endif /* __KERNEL__ */ |
5104 | |
5105 | -#define __SLOW_DOWN_IO "outb %%al,$0x80;" |
5106 | +static inline void xen_io_delay(void) |
5107 | +{ |
5108 | + asm volatile("outb %%al,$0x80" : : : "memory"); |
5109 | +} |
5110 | |
5111 | static inline void slow_down_io(void) { |
5112 | - __asm__ __volatile__( |
5113 | - __SLOW_DOWN_IO |
5114 | + xen_io_delay(); |
5115 | #ifdef REALLY_SLOW_IO |
5116 | - __SLOW_DOWN_IO __SLOW_DOWN_IO __SLOW_DOWN_IO |
5117 | + xen_io_delay(); |
5118 | + xen_io_delay(); |
5119 | + xen_io_delay(); |
5120 | #endif |
5121 | - : : ); |
5122 | } |
5123 | |
5124 | #ifdef CONFIG_X86_NUMAQ |
5125 | --- a/include/asm-x86/mach-xen/asm/irqflags_32.h |
5126 | +++ b/include/asm-x86/mach-xen/asm/irqflags_32.h |
5127 | @@ -11,6 +11,43 @@ |
5128 | #define _ASM_IRQFLAGS_H |
5129 | |
5130 | #ifndef __ASSEMBLY__ |
5131 | +#define xen_save_fl(void) (current_vcpu_info()->evtchn_upcall_mask) |
5132 | + |
5133 | +#define xen_restore_fl(f) \ |
5134 | +do { \ |
5135 | + vcpu_info_t *_vcpu; \ |
5136 | + barrier(); \ |
5137 | + _vcpu = current_vcpu_info(); \ |
5138 | + if ((_vcpu->evtchn_upcall_mask = (f)) == 0) { \ |
5139 | + barrier(); /* unmask then check (avoid races) */\ |
5140 | + if (unlikely(_vcpu->evtchn_upcall_pending)) \ |
5141 | + force_evtchn_callback(); \ |
5142 | + } \ |
5143 | +} while (0) |
5144 | + |
5145 | +#define xen_irq_disable() \ |
5146 | +do { \ |
5147 | + current_vcpu_info()->evtchn_upcall_mask = 1; \ |
5148 | + barrier(); \ |
5149 | +} while (0) |
5150 | + |
5151 | +#define xen_irq_enable() \ |
5152 | +do { \ |
5153 | + vcpu_info_t *_vcpu; \ |
5154 | + barrier(); \ |
5155 | + _vcpu = current_vcpu_info(); \ |
5156 | + _vcpu->evtchn_upcall_mask = 0; \ |
5157 | + barrier(); /* unmask then check (avoid races) */ \ |
5158 | + if (unlikely(_vcpu->evtchn_upcall_pending)) \ |
5159 | + force_evtchn_callback(); \ |
5160 | +} while (0) |
5161 | + |
5162 | +void xen_safe_halt(void); |
5163 | + |
5164 | +void xen_halt(void); |
5165 | +#endif /* __ASSEMBLY__ */ |
5166 | + |
5167 | +#ifndef __ASSEMBLY__ |
5168 | |
5169 | /* |
5170 | * The use of 'barrier' in the following reflects their use as local-lock |
5171 | @@ -20,48 +57,31 @@ |
5172 | * includes these barriers, for example. |
5173 | */ |
5174 | |
5175 | -#define __raw_local_save_flags() (current_vcpu_info()->evtchn_upcall_mask) |
5176 | +#define __raw_local_save_flags(void) xen_save_fl() |
5177 | |
5178 | -#define raw_local_irq_restore(x) \ |
5179 | -do { \ |
5180 | - vcpu_info_t *_vcpu; \ |
5181 | - barrier(); \ |
5182 | - _vcpu = current_vcpu_info(); \ |
5183 | - if ((_vcpu->evtchn_upcall_mask = (x)) == 0) { \ |
5184 | - barrier(); /* unmask then check (avoid races) */ \ |
5185 | - if (unlikely(_vcpu->evtchn_upcall_pending)) \ |
5186 | - force_evtchn_callback(); \ |
5187 | - } \ |
5188 | -} while (0) |
5189 | +#define raw_local_irq_restore(flags) xen_restore_fl(flags) |
5190 | |
5191 | -#define raw_local_irq_disable() \ |
5192 | -do { \ |
5193 | - current_vcpu_info()->evtchn_upcall_mask = 1; \ |
5194 | - barrier(); \ |
5195 | -} while (0) |
5196 | +#define raw_local_irq_disable() xen_irq_disable() |
5197 | |
5198 | -#define raw_local_irq_enable() \ |
5199 | -do { \ |
5200 | - vcpu_info_t *_vcpu; \ |
5201 | - barrier(); \ |
5202 | - _vcpu = current_vcpu_info(); \ |
5203 | - _vcpu->evtchn_upcall_mask = 0; \ |
5204 | - barrier(); /* unmask then check (avoid races) */ \ |
5205 | - if (unlikely(_vcpu->evtchn_upcall_pending)) \ |
5206 | - force_evtchn_callback(); \ |
5207 | -} while (0) |
5208 | +#define raw_local_irq_enable() xen_irq_enable() |
5209 | |
5210 | /* |
5211 | * Used in the idle loop; sti takes one instruction cycle |
5212 | * to complete: |
5213 | */ |
5214 | -void raw_safe_halt(void); |
5215 | +static inline void raw_safe_halt(void) |
5216 | +{ |
5217 | + xen_safe_halt(); |
5218 | +} |
5219 | |
5220 | /* |
5221 | * Used when interrupts are already enabled or to |
5222 | * shutdown the processor: |
5223 | */ |
5224 | -void halt(void); |
5225 | +static inline void halt(void) |
5226 | +{ |
5227 | + xen_halt(); |
5228 | +} |
5229 | |
5230 | /* |
5231 | * For spinlocks, etc: |
5232 | --- a/include/asm-x86/mach-xen/asm/irqflags_64.h |
5233 | +++ b/include/asm-x86/mach-xen/asm/irqflags_64.h |
5234 | @@ -9,6 +9,7 @@ |
5235 | */ |
5236 | #ifndef _ASM_IRQFLAGS_H |
5237 | #define _ASM_IRQFLAGS_H |
5238 | +#include <asm/processor-flags.h> |
5239 | |
5240 | #ifndef __ASSEMBLY__ |
5241 | /* |
5242 | @@ -50,19 +51,19 @@ |
5243 | { |
5244 | unsigned long flags = __raw_local_save_flags(); |
5245 | |
5246 | - raw_local_irq_restore((flags & ~(1 << 9)) | (1 << 18)); |
5247 | + raw_local_irq_restore((flags & ~X86_EFLAGS_IF) | X86_EFLAGS_AC); |
5248 | } |
5249 | |
5250 | static inline void raw_local_irq_enable(void) |
5251 | { |
5252 | unsigned long flags = __raw_local_save_flags(); |
5253 | |
5254 | - raw_local_irq_restore((flags | (1 << 9)) & ~(1 << 18)); |
5255 | + raw_local_irq_restore((flags | X86_EFLAGS_IF) & (~X86_EFLAGS_AC)); |
5256 | } |
5257 | |
5258 | static inline int raw_irqs_disabled_flags(unsigned long flags) |
5259 | { |
5260 | - return !(flags & (1<<9)) || (flags & (1 << 18)); |
5261 | + return !(flags & X86_EFLAGS_IF) || (flags & X86_EFLAGS_AC); |
5262 | } |
5263 | |
5264 | #else /* CONFIG_X86_VSMP */ |
5265 | @@ -118,13 +119,21 @@ |
5266 | * Used in the idle loop; sti takes one instruction cycle |
5267 | * to complete: |
5268 | */ |
5269 | -void raw_safe_halt(void); |
5270 | +void xen_safe_halt(void); |
5271 | +static inline void raw_safe_halt(void) |
5272 | +{ |
5273 | + xen_safe_halt(); |
5274 | +} |
5275 | |
5276 | /* |
5277 | * Used when interrupts are already enabled or to |
5278 | * shutdown the processor: |
5279 | */ |
5280 | -void halt(void); |
5281 | +void xen_halt(void); |
5282 | +static inline void halt(void) |
5283 | +{ |
5284 | + xen_halt(); |
5285 | +} |
5286 | |
5287 | #else /* __ASSEMBLY__: */ |
5288 | # ifdef CONFIG_TRACE_IRQFLAGS |
5289 | --- a/include/asm-x86/mach-xen/asm/mmu.h |
5290 | +++ b/include/asm-x86/mach-xen/asm/mmu.h |
5291 | @@ -18,12 +18,4 @@ |
5292 | #endif |
5293 | } mm_context_t; |
5294 | |
5295 | -/* mm/memory.c:exit_mmap hook */ |
5296 | -extern void _arch_exit_mmap(struct mm_struct *mm); |
5297 | -#define arch_exit_mmap(_mm) _arch_exit_mmap(_mm) |
5298 | - |
5299 | -/* kernel/fork.c:dup_mmap hook */ |
5300 | -extern void _arch_dup_mmap(struct mm_struct *mm); |
5301 | -#define arch_dup_mmap(mm, oldmm) ((void)(oldmm), _arch_dup_mmap(mm)) |
5302 | - |
5303 | #endif |
5304 | --- a/include/asm-x86/mach-xen/asm/mmu_64.h |
5305 | +++ b/include/asm-x86/mach-xen/asm/mmu_64.h |
5306 | @@ -25,14 +25,6 @@ |
5307 | #ifdef CONFIG_XEN |
5308 | extern struct list_head mm_unpinned; |
5309 | extern spinlock_t mm_unpinned_lock; |
5310 | - |
5311 | -/* mm/memory.c:exit_mmap hook */ |
5312 | -extern void _arch_exit_mmap(struct mm_struct *mm); |
5313 | -#define arch_exit_mmap(_mm) _arch_exit_mmap(_mm) |
5314 | - |
5315 | -/* kernel/fork.c:dup_mmap hook */ |
5316 | -extern void _arch_dup_mmap(struct mm_struct *mm); |
5317 | -#define arch_dup_mmap(mm, oldmm) ((void)(oldmm), _arch_dup_mmap(mm)) |
5318 | #endif |
5319 | |
5320 | #endif |
5321 | --- a/include/asm-x86/mach-xen/asm/mmu_context_32.h |
5322 | +++ b/include/asm-x86/mach-xen/asm/mmu_context_32.h |
5323 | @@ -6,6 +6,20 @@ |
5324 | #include <asm/pgalloc.h> |
5325 | #include <asm/tlbflush.h> |
5326 | |
5327 | +void arch_exit_mmap(struct mm_struct *mm); |
5328 | +void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm); |
5329 | + |
5330 | +void mm_pin(struct mm_struct *mm); |
5331 | +void mm_unpin(struct mm_struct *mm); |
5332 | +void mm_pin_all(void); |
5333 | + |
5334 | +static inline void xen_activate_mm(struct mm_struct *prev, |
5335 | + struct mm_struct *next) |
5336 | +{ |
5337 | + if (!test_bit(PG_pinned, &virt_to_page(next->pgd)->flags)) |
5338 | + mm_pin(next); |
5339 | +} |
5340 | + |
5341 | /* |
5342 | * Used for LDT copy/destruction. |
5343 | */ |
5344 | @@ -37,10 +51,6 @@ |
5345 | : : "r" (0) ); |
5346 | } |
5347 | |
5348 | -extern void mm_pin(struct mm_struct *mm); |
5349 | -extern void mm_unpin(struct mm_struct *mm); |
5350 | -void mm_pin_all(void); |
5351 | - |
5352 | static inline void switch_mm(struct mm_struct *prev, |
5353 | struct mm_struct *next, |
5354 | struct task_struct *tsk) |
5355 | @@ -97,11 +107,10 @@ |
5356 | #define deactivate_mm(tsk, mm) \ |
5357 | asm("movl %0,%%gs": :"r" (0)); |
5358 | |
5359 | -static inline void activate_mm(struct mm_struct *prev, struct mm_struct *next) |
5360 | -{ |
5361 | - if (!test_bit(PG_pinned, &virt_to_page(next->pgd)->flags)) |
5362 | - mm_pin(next); |
5363 | - switch_mm(prev, next, NULL); |
5364 | -} |
5365 | +#define activate_mm(prev, next) \ |
5366 | + do { \ |
5367 | + xen_activate_mm(prev, next); \ |
5368 | + switch_mm((prev),(next),NULL); \ |
5369 | + } while(0) |
5370 | |
5371 | #endif |
5372 | --- a/include/asm-x86/mach-xen/asm/mmu_context_64.h |
5373 | +++ b/include/asm-x86/mach-xen/asm/mmu_context_64.h |
5374 | @@ -9,6 +9,9 @@ |
5375 | #include <asm/pgtable.h> |
5376 | #include <asm/tlbflush.h> |
5377 | |
5378 | +void arch_exit_mmap(struct mm_struct *mm); |
5379 | +void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm); |
5380 | + |
5381 | /* |
5382 | * possibly do the LDT unload here? |
5383 | */ |
5384 | --- a/include/asm-x86/mach-xen/asm/page_64.h |
5385 | +++ b/include/asm-x86/mach-xen/asm/page_64.h |
5386 | @@ -7,6 +7,7 @@ |
5387 | #include <linux/types.h> |
5388 | #include <asm/bug.h> |
5389 | #endif |
5390 | +#include <linux/const.h> |
5391 | #include <xen/interface/xen.h> |
5392 | |
5393 | /* |
5394 | @@ -19,18 +20,14 @@ |
5395 | |
5396 | /* PAGE_SHIFT determines the page size */ |
5397 | #define PAGE_SHIFT 12 |
5398 | -#ifdef __ASSEMBLY__ |
5399 | -#define PAGE_SIZE (0x1 << PAGE_SHIFT) |
5400 | -#else |
5401 | -#define PAGE_SIZE (1UL << PAGE_SHIFT) |
5402 | -#endif |
5403 | +#define PAGE_SIZE (_AC(1,UL) << PAGE_SHIFT) |
5404 | #define PAGE_MASK (~(PAGE_SIZE-1)) |
5405 | |
5406 | /* See Documentation/x86_64/mm.txt for a description of the memory map. */ |
5407 | #define __PHYSICAL_MASK_SHIFT 46 |
5408 | -#define __PHYSICAL_MASK ((1UL << __PHYSICAL_MASK_SHIFT) - 1) |
5409 | +#define __PHYSICAL_MASK ((_AC(1,UL) << __PHYSICAL_MASK_SHIFT) - 1) |
5410 | #define __VIRTUAL_MASK_SHIFT 48 |
5411 | -#define __VIRTUAL_MASK ((1UL << __VIRTUAL_MASK_SHIFT) - 1) |
5412 | +#define __VIRTUAL_MASK ((_AC(1,UL) << __VIRTUAL_MASK_SHIFT) - 1) |
5413 | |
5414 | #define PHYSICAL_PAGE_MASK (~(PAGE_SIZE-1) & __PHYSICAL_MASK) |
5415 | |
5416 | @@ -55,10 +52,10 @@ |
5417 | #define N_EXCEPTION_STACKS 5 /* hw limit: 7 */ |
5418 | |
5419 | #define LARGE_PAGE_MASK (~(LARGE_PAGE_SIZE-1)) |
5420 | -#define LARGE_PAGE_SIZE (1UL << PMD_SHIFT) |
5421 | +#define LARGE_PAGE_SIZE (_AC(1,UL) << PMD_SHIFT) |
5422 | |
5423 | #define HPAGE_SHIFT PMD_SHIFT |
5424 | -#define HPAGE_SIZE ((1UL) << HPAGE_SHIFT) |
5425 | +#define HPAGE_SIZE (_AC(1,UL) << HPAGE_SHIFT) |
5426 | #define HPAGE_MASK (~(HPAGE_SIZE - 1)) |
5427 | #define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT) |
5428 | |
5429 | @@ -152,17 +149,23 @@ |
5430 | |
5431 | #define __pgprot(x) ((pgprot_t) { (x) } ) |
5432 | |
5433 | -#define __PHYSICAL_START ((unsigned long)CONFIG_PHYSICAL_START) |
5434 | -#define __START_KERNEL (__START_KERNEL_map + __PHYSICAL_START) |
5435 | -#define __START_KERNEL_map 0xffffffff80000000UL |
5436 | -#define __PAGE_OFFSET 0xffff880000000000UL |
5437 | +#endif /* !__ASSEMBLY__ */ |
5438 | |
5439 | -#else |
5440 | #define __PHYSICAL_START CONFIG_PHYSICAL_START |
5441 | +#define __KERNEL_ALIGN 0x200000 |
5442 | + |
5443 | +/* |
5444 | + * Make sure kernel is aligned to 2MB address. Catching it at compile |
5445 | + * time is better. Change your config file and compile the kernel |
5446 | + * for a 2MB aligned address (CONFIG_PHYSICAL_START) |
5447 | + */ |
5448 | +#if (CONFIG_PHYSICAL_START % __KERNEL_ALIGN) != 0 |
5449 | +#error "CONFIG_PHYSICAL_START must be a multiple of 2MB" |
5450 | +#endif |
5451 | + |
5452 | #define __START_KERNEL (__START_KERNEL_map + __PHYSICAL_START) |
5453 | -#define __START_KERNEL_map 0xffffffff80000000 |
5454 | -#define __PAGE_OFFSET 0xffff880000000000 |
5455 | -#endif /* !__ASSEMBLY__ */ |
5456 | +#define __START_KERNEL_map _AC(0xffffffff80000000, UL) |
5457 | +#define __PAGE_OFFSET _AC(0xffff880000000000, UL) |
5458 | |
5459 | #if CONFIG_XEN_COMPAT <= 0x030002 |
5460 | #undef LOAD_OFFSET |
5461 | @@ -172,20 +175,20 @@ |
5462 | /* to align the pointer to the (next) page boundary */ |
5463 | #define PAGE_ALIGN(addr) (((addr)+PAGE_SIZE-1)&PAGE_MASK) |
5464 | |
5465 | -#define KERNEL_TEXT_SIZE (40UL*1024*1024) |
5466 | -#define KERNEL_TEXT_START 0xffffffff80000000UL |
5467 | +#define KERNEL_TEXT_SIZE (40*1024*1024) |
5468 | +#define KERNEL_TEXT_START _AC(0xffffffff80000000, UL) |
5469 | + |
5470 | +#define PAGE_OFFSET __PAGE_OFFSET |
5471 | |
5472 | -#define PAGE_OFFSET ((unsigned long)__PAGE_OFFSET) |
5473 | +#ifndef __ASSEMBLY__ |
5474 | +static inline unsigned long __phys_addr(unsigned long x) |
5475 | +{ |
5476 | + return x - (x >= __START_KERNEL_map ? __START_KERNEL_map : PAGE_OFFSET); |
5477 | +} |
5478 | +#endif |
5479 | |
5480 | -/* Note: __pa(&symbol_visible_to_c) should be always replaced with __pa_symbol. |
5481 | - Otherwise you risk miscompilation. */ |
5482 | -#define __pa(x) (((unsigned long)(x)>=__START_KERNEL_map)?(unsigned long)(x) - (unsigned long)__START_KERNEL_map:(unsigned long)(x) - PAGE_OFFSET) |
5483 | -/* __pa_symbol should be used for C visible symbols. |
5484 | - This seems to be the official gcc blessed way to do such arithmetic. */ |
5485 | -#define __pa_symbol(x) \ |
5486 | - ({unsigned long v; \ |
5487 | - asm("" : "=r" (v) : "0" (x)); \ |
5488 | - __pa(v); }) |
5489 | +#define __pa(x) __phys_addr((unsigned long)(x)) |
5490 | +#define __pa_symbol(x) __phys_addr((unsigned long)(x)) |
5491 | |
5492 | #define __va(x) ((void *)((unsigned long)(x)+PAGE_OFFSET)) |
5493 | #define __boot_va(x) __va(x) |
5494 | --- a/include/asm-x86/mach-xen/asm/pgalloc_32.h |
5495 | +++ b/include/asm-x86/mach-xen/asm/pgalloc_32.h |
5496 | @@ -1,7 +1,6 @@ |
5497 | #ifndef _I386_PGALLOC_H |
5498 | #define _I386_PGALLOC_H |
5499 | |
5500 | -#include <asm/fixmap.h> |
5501 | #include <linux/threads.h> |
5502 | #include <linux/mm.h> /* for struct page */ |
5503 | #include <asm/io.h> /* for phys_to_virt and page_to_pseudophys */ |
5504 | @@ -69,6 +68,4 @@ |
5505 | #define pud_populate(mm, pmd, pte) BUG() |
5506 | #endif |
5507 | |
5508 | -#define check_pgt_cache() do { } while (0) |
5509 | - |
5510 | #endif /* _I386_PGALLOC_H */ |
5511 | --- a/include/asm-x86/mach-xen/asm/pgalloc_64.h |
5512 | +++ b/include/asm-x86/mach-xen/asm/pgalloc_64.h |
5513 | @@ -1,7 +1,6 @@ |
5514 | #ifndef _X86_64_PGALLOC_H |
5515 | #define _X86_64_PGALLOC_H |
5516 | |
5517 | -#include <asm/fixmap.h> |
5518 | #include <asm/pda.h> |
5519 | #include <linux/threads.h> |
5520 | #include <linux/mm.h> |
5521 | @@ -100,24 +99,16 @@ |
5522 | struct page *page = virt_to_page(pgd); |
5523 | |
5524 | spin_lock(&pgd_lock); |
5525 | - page->index = (pgoff_t)pgd_list; |
5526 | - if (pgd_list) |
5527 | - pgd_list->private = (unsigned long)&page->index; |
5528 | - pgd_list = page; |
5529 | - page->private = (unsigned long)&pgd_list; |
5530 | + list_add(&page->lru, &pgd_list); |
5531 | spin_unlock(&pgd_lock); |
5532 | } |
5533 | |
5534 | static inline void pgd_list_del(pgd_t *pgd) |
5535 | { |
5536 | - struct page *next, **pprev, *page = virt_to_page(pgd); |
5537 | + struct page *page = virt_to_page(pgd); |
5538 | |
5539 | spin_lock(&pgd_lock); |
5540 | - next = (struct page *)page->index; |
5541 | - pprev = (struct page **)page->private; |
5542 | - *pprev = next; |
5543 | - if (next) |
5544 | - next->private = (unsigned long)pprev; |
5545 | + list_del(&page->lru); |
5546 | spin_unlock(&pgd_lock); |
5547 | } |
5548 | |
5549 | --- a/include/asm-x86/mach-xen/asm/pgtable-2level.h |
5550 | +++ b/include/asm-x86/mach-xen/asm/pgtable-2level.h |
5551 | @@ -13,22 +13,43 @@ |
5552 | * within a page table are directly modified. Thus, the following |
5553 | * hook is made available. |
5554 | */ |
5555 | -#define set_pte(pteptr, pteval) (*(pteptr) = pteval) |
5556 | - |
5557 | -#define set_pte_at(_mm,addr,ptep,pteval) do { \ |
5558 | - if (((_mm) != current->mm && (_mm) != &init_mm) || \ |
5559 | - HYPERVISOR_update_va_mapping((addr), (pteval), 0)) \ |
5560 | - set_pte((ptep), (pteval)); \ |
5561 | -} while (0) |
5562 | - |
5563 | -#define set_pmd(pmdptr, pmdval) xen_l2_entry_update((pmdptr), (pmdval)) |
5564 | +static inline void xen_set_pte(pte_t *ptep , pte_t pte) |
5565 | +{ |
5566 | + *ptep = pte; |
5567 | +} |
5568 | +static inline void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, |
5569 | + pte_t *ptep , pte_t pte) |
5570 | +{ |
5571 | + if ((mm != current->mm && mm != &init_mm) || |
5572 | + HYPERVISOR_update_va_mapping(addr, pte, 0)) |
5573 | + xen_set_pte(ptep, pte); |
5574 | +} |
5575 | +static inline void xen_set_pmd(pmd_t *pmdp, pmd_t pmd) |
5576 | +{ |
5577 | + xen_l2_entry_update(pmdp, pmd); |
5578 | +} |
5579 | +#define set_pte(pteptr, pteval) xen_set_pte(pteptr, pteval) |
5580 | +#define set_pte_at(mm,addr,ptep,pteval) xen_set_pte_at(mm, addr, ptep, pteval) |
5581 | +#define set_pmd(pmdptr, pmdval) xen_set_pmd(pmdptr, pmdval) |
5582 | |
5583 | #define set_pte_atomic(pteptr, pteval) set_pte(pteptr,pteval) |
5584 | |
5585 | #define pte_clear(mm,addr,xp) do { set_pte_at(mm, addr, xp, __pte(0)); } while (0) |
5586 | #define pmd_clear(xp) do { set_pmd(xp, __pmd(0)); } while (0) |
5587 | |
5588 | -#define raw_ptep_get_and_clear(xp, pte) __pte_ma(xchg(&(xp)->pte_low, 0)) |
5589 | +static inline void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *xp) |
5590 | +{ |
5591 | + xen_set_pte_at(mm, addr, xp, __pte(0)); |
5592 | +} |
5593 | + |
5594 | +#ifdef CONFIG_SMP |
5595 | +static inline pte_t xen_ptep_get_and_clear(pte_t *xp, pte_t res) |
5596 | +{ |
5597 | + return __pte_ma(xchg(&xp->pte_low, 0)); |
5598 | +} |
5599 | +#else |
5600 | +#define xen_ptep_get_and_clear(xp, res) xen_local_ptep_get_and_clear(xp, res) |
5601 | +#endif |
5602 | |
5603 | #define __HAVE_ARCH_PTEP_CLEAR_FLUSH |
5604 | #define ptep_clear_flush(vma, addr, ptep) \ |
5605 | @@ -95,6 +116,4 @@ |
5606 | #define __pte_to_swp_entry(pte) ((swp_entry_t) { (pte).pte_low }) |
5607 | #define __swp_entry_to_pte(x) ((pte_t) { (x).val }) |
5608 | |
5609 | -void vmalloc_sync_all(void); |
5610 | - |
5611 | #endif /* _I386_PGTABLE_2LEVEL_H */ |
5612 | --- a/include/asm-x86/mach-xen/asm/pgtable-3level-defs.h |
5613 | +++ b/include/asm-x86/mach-xen/asm/pgtable-3level-defs.h |
5614 | @@ -1,7 +1,7 @@ |
5615 | #ifndef _I386_PGTABLE_3LEVEL_DEFS_H |
5616 | #define _I386_PGTABLE_3LEVEL_DEFS_H |
5617 | |
5618 | -#define HAVE_SHARED_KERNEL_PMD 0 |
5619 | +#define SHARED_KERNEL_PMD 0 |
5620 | |
5621 | /* |
5622 | * PGDIR_SHIFT determines what a top-level page table entry can map |
5623 | --- a/include/asm-x86/mach-xen/asm/pgtable-3level.h |
5624 | +++ b/include/asm-x86/mach-xen/asm/pgtable-3level.h |
5625 | @@ -52,32 +52,40 @@ |
5626 | * value and then use set_pte to update it. -ben |
5627 | */ |
5628 | |
5629 | -static inline void set_pte(pte_t *ptep, pte_t pte) |
5630 | +static inline void xen_set_pte(pte_t *ptep, pte_t pte) |
5631 | { |
5632 | ptep->pte_high = pte.pte_high; |
5633 | smp_wmb(); |
5634 | ptep->pte_low = pte.pte_low; |
5635 | } |
5636 | -#define set_pte_atomic(pteptr,pteval) \ |
5637 | - set_64bit((unsigned long long *)(pteptr),__pte_val(pteval)) |
5638 | |
5639 | -#define set_pte_at(_mm,addr,ptep,pteval) do { \ |
5640 | - if (((_mm) != current->mm && (_mm) != &init_mm) || \ |
5641 | - HYPERVISOR_update_va_mapping((addr), (pteval), 0)) \ |
5642 | - set_pte((ptep), (pteval)); \ |
5643 | -} while (0) |
5644 | - |
5645 | -#define set_pmd(pmdptr,pmdval) \ |
5646 | - xen_l2_entry_update((pmdptr), (pmdval)) |
5647 | -#define set_pud(pudptr,pudval) \ |
5648 | - xen_l3_entry_update((pudptr), (pudval)) |
5649 | +static inline void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, |
5650 | + pte_t *ptep , pte_t pte) |
5651 | +{ |
5652 | + if ((mm != current->mm && mm != &init_mm) || |
5653 | + HYPERVISOR_update_va_mapping(addr, pte, 0)) |
5654 | + xen_set_pte(ptep, pte); |
5655 | +} |
5656 | + |
5657 | +static inline void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
5658 | +{ |
5659 | + set_64bit((unsigned long long *)(ptep),__pte_val(pte)); |
5660 | +} |
5661 | +static inline void xen_set_pmd(pmd_t *pmdp, pmd_t pmd) |
5662 | +{ |
5663 | + xen_l2_entry_update(pmdp, pmd); |
5664 | +} |
5665 | +static inline void xen_set_pud(pud_t *pudp, pud_t pud) |
5666 | +{ |
5667 | + xen_l3_entry_update(pudp, pud); |
5668 | +} |
5669 | |
5670 | /* |
5671 | * For PTEs and PDEs, we must clear the P-bit first when clearing a page table |
5672 | * entry, so clear the bottom half first and enforce ordering with a compiler |
5673 | * barrier. |
5674 | */ |
5675 | -static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
5676 | +static inline void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
5677 | { |
5678 | if ((mm != current->mm && mm != &init_mm) |
5679 | || HYPERVISOR_update_va_mapping(addr, __pte(0), 0)) { |
5680 | @@ -87,7 +95,18 @@ |
5681 | } |
5682 | } |
5683 | |
5684 | -#define pmd_clear(xp) do { set_pmd(xp, __pmd(0)); } while (0) |
5685 | +static inline void xen_pmd_clear(pmd_t *pmd) |
5686 | +{ |
5687 | + xen_l2_entry_update(pmd, __pmd(0)); |
5688 | +} |
5689 | + |
5690 | +#define set_pte(ptep, pte) xen_set_pte(ptep, pte) |
5691 | +#define set_pte_at(mm, addr, ptep, pte) xen_set_pte_at(mm, addr, ptep, pte) |
5692 | +#define set_pte_atomic(ptep, pte) xen_set_pte_atomic(ptep, pte) |
5693 | +#define set_pmd(pmdp, pmd) xen_set_pmd(pmdp, pmd) |
5694 | +#define set_pud(pudp, pud) xen_set_pud(pudp, pud) |
5695 | +#define pte_clear(mm, addr, ptep) xen_pte_clear(mm, addr, ptep) |
5696 | +#define pmd_clear(pmd) xen_pmd_clear(pmd) |
5697 | |
5698 | /* |
5699 | * Pentium-II erratum A13: in PAE mode we explicitly have to flush |
5700 | @@ -108,7 +127,8 @@ |
5701 | #define pmd_offset(pud, address) ((pmd_t *) pud_page(*(pud)) + \ |
5702 | pmd_index(address)) |
5703 | |
5704 | -static inline pte_t raw_ptep_get_and_clear(pte_t *ptep, pte_t res) |
5705 | +#ifdef CONFIG_SMP |
5706 | +static inline pte_t xen_ptep_get_and_clear(pte_t *ptep, pte_t res) |
5707 | { |
5708 | uint64_t val = __pte_val(res); |
5709 | if (__cmpxchg64(ptep, val, 0) != val) { |
5710 | @@ -119,6 +139,9 @@ |
5711 | } |
5712 | return res; |
5713 | } |
5714 | +#else |
5715 | +#define xen_ptep_get_and_clear(xp, pte) xen_local_ptep_get_and_clear(xp, pte) |
5716 | +#endif |
5717 | |
5718 | #define __HAVE_ARCH_PTEP_CLEAR_FLUSH |
5719 | #define ptep_clear_flush(vma, addr, ptep) \ |
5720 | @@ -165,13 +188,13 @@ |
5721 | static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) |
5722 | { |
5723 | return __pte((((unsigned long long)page_nr << PAGE_SHIFT) | |
5724 | - pgprot_val(pgprot)) & __supported_pte_mask); |
5725 | + pgprot_val(pgprot)) & __supported_pte_mask); |
5726 | } |
5727 | |
5728 | static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot) |
5729 | { |
5730 | return __pmd((((unsigned long long)page_nr << PAGE_SHIFT) | |
5731 | - pgprot_val(pgprot)) & __supported_pte_mask); |
5732 | + pgprot_val(pgprot)) & __supported_pte_mask); |
5733 | } |
5734 | |
5735 | /* |
5736 | @@ -191,6 +214,4 @@ |
5737 | |
5738 | #define __pmd_free_tlb(tlb, x) do { } while (0) |
5739 | |
5740 | -void vmalloc_sync_all(void); |
5741 | - |
5742 | #endif /* _I386_PGTABLE_3LEVEL_H */ |
5743 | --- a/include/asm-x86/mach-xen/asm/pgtable_32.h |
5744 | +++ b/include/asm-x86/mach-xen/asm/pgtable_32.h |
5745 | @@ -24,11 +24,11 @@ |
5746 | #include <linux/slab.h> |
5747 | #include <linux/list.h> |
5748 | #include <linux/spinlock.h> |
5749 | +#include <linux/sched.h> |
5750 | |
5751 | /* Is this pagetable pinned? */ |
5752 | #define PG_pinned PG_arch_1 |
5753 | |
5754 | -struct mm_struct; |
5755 | struct vm_area_struct; |
5756 | |
5757 | /* |
5758 | @@ -38,17 +38,16 @@ |
5759 | #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page)) |
5760 | extern unsigned long empty_zero_page[1024]; |
5761 | extern pgd_t *swapper_pg_dir; |
5762 | -extern struct kmem_cache *pgd_cache; |
5763 | extern struct kmem_cache *pmd_cache; |
5764 | extern spinlock_t pgd_lock; |
5765 | extern struct page *pgd_list; |
5766 | +void check_pgt_cache(void); |
5767 | |
5768 | void pmd_ctor(void *, struct kmem_cache *, unsigned long); |
5769 | -void pgd_ctor(void *, struct kmem_cache *, unsigned long); |
5770 | -void pgd_dtor(void *, struct kmem_cache *, unsigned long); |
5771 | void pgtable_cache_init(void); |
5772 | void paging_init(void); |
5773 | |
5774 | + |
5775 | /* |
5776 | * The Linux x86 paging architecture is 'compile-time dual-mode', it |
5777 | * implements both the traditional 2-level x86 page tables and the |
5778 | @@ -165,6 +164,7 @@ |
5779 | |
5780 | extern unsigned long long __PAGE_KERNEL, __PAGE_KERNEL_EXEC; |
5781 | #define __PAGE_KERNEL_RO (__PAGE_KERNEL & ~_PAGE_RW) |
5782 | +#define __PAGE_KERNEL_RX (__PAGE_KERNEL_EXEC & ~_PAGE_RW) |
5783 | #define __PAGE_KERNEL_NOCACHE (__PAGE_KERNEL | _PAGE_PCD) |
5784 | #define __PAGE_KERNEL_LARGE (__PAGE_KERNEL | _PAGE_PSE) |
5785 | #define __PAGE_KERNEL_LARGE_EXEC (__PAGE_KERNEL_EXEC | _PAGE_PSE) |
5786 | @@ -172,6 +172,7 @@ |
5787 | #define PAGE_KERNEL __pgprot(__PAGE_KERNEL) |
5788 | #define PAGE_KERNEL_RO __pgprot(__PAGE_KERNEL_RO) |
5789 | #define PAGE_KERNEL_EXEC __pgprot(__PAGE_KERNEL_EXEC) |
5790 | +#define PAGE_KERNEL_RX __pgprot(__PAGE_KERNEL_RX) |
5791 | #define PAGE_KERNEL_NOCACHE __pgprot(__PAGE_KERNEL_NOCACHE) |
5792 | #define PAGE_KERNEL_LARGE __pgprot(__PAGE_KERNEL_LARGE) |
5793 | #define PAGE_KERNEL_LARGE_EXEC __pgprot(__PAGE_KERNEL_LARGE_EXEC) |
5794 | @@ -275,7 +276,13 @@ |
5795 | */ |
5796 | #define pte_update(mm, addr, ptep) do { } while (0) |
5797 | #define pte_update_defer(mm, addr, ptep) do { } while (0) |
5798 | -#define paravirt_map_pt_hook(slot, va, pfn) do { } while (0) |
5799 | + |
5800 | +/* local pte updates need not use xchg for locking */ |
5801 | +static inline pte_t xen_local_ptep_get_and_clear(pte_t *ptep, pte_t res) |
5802 | +{ |
5803 | + xen_set_pte(ptep, __pte(0)); |
5804 | + return res; |
5805 | +} |
5806 | |
5807 | /* |
5808 | * We only update the dirty/accessed state if we set |
5809 | @@ -286,17 +293,34 @@ |
5810 | */ |
5811 | #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS |
5812 | #define ptep_set_access_flags(vma, address, ptep, entry, dirty) \ |
5813 | -do { \ |
5814 | - if (dirty) \ |
5815 | +({ \ |
5816 | + int __changed = !pte_same(*(ptep), entry); \ |
5817 | + if (__changed && (dirty)) \ |
5818 | ptep_establish(vma, address, ptep, entry); \ |
5819 | -} while (0) |
5820 | + __changed; \ |
5821 | +}) |
5822 | |
5823 | -/* |
5824 | - * We don't actually have these, but we want to advertise them so that |
5825 | - * we can encompass the flush here. |
5826 | - */ |
5827 | #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY |
5828 | +#define ptep_test_and_clear_dirty(vma, addr, ptep) ({ \ |
5829 | + int __ret = 0; \ |
5830 | + if (pte_dirty(*(ptep))) \ |
5831 | + __ret = test_and_clear_bit(_PAGE_BIT_DIRTY, \ |
5832 | + &(ptep)->pte_low); \ |
5833 | + if (__ret) \ |
5834 | + pte_update((vma)->vm_mm, addr, ptep); \ |
5835 | + __ret; \ |
5836 | +}) |
5837 | + |
5838 | #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG |
5839 | +#define ptep_test_and_clear_young(vma, addr, ptep) ({ \ |
5840 | + int __ret = 0; \ |
5841 | + if (pte_young(*(ptep))) \ |
5842 | + __ret = test_and_clear_bit(_PAGE_BIT_ACCESSED, \ |
5843 | + &(ptep)->pte_low); \ |
5844 | + if (__ret) \ |
5845 | + pte_update((vma)->vm_mm, addr, ptep); \ |
5846 | + __ret; \ |
5847 | +}) |
5848 | |
5849 | /* |
5850 | * Rules for using ptep_establish: the pte MUST be a user pte, and |
5851 | @@ -323,7 +347,7 @@ |
5852 | int __dirty = pte_dirty(__pte); \ |
5853 | __pte = pte_mkclean(__pte); \ |
5854 | if (test_bit(PG_pinned, &virt_to_page((vma)->vm_mm->pgd)->flags)) \ |
5855 | - ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \ |
5856 | + (void)ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \ |
5857 | else if (__dirty) \ |
5858 | (ptep)->pte_low = __pte.pte_low; \ |
5859 | __dirty; \ |
5860 | @@ -336,7 +360,7 @@ |
5861 | int __young = pte_young(__pte); \ |
5862 | __pte = pte_mkold(__pte); \ |
5863 | if (test_bit(PG_pinned, &virt_to_page((vma)->vm_mm->pgd)->flags)) \ |
5864 | - ptep_set_access_flags(vma, address, ptep, __pte, __young); \ |
5865 | + (void)ptep_set_access_flags(vma, address, ptep, __pte, __young); \ |
5866 | else if (__young) \ |
5867 | (ptep)->pte_low = __pte.pte_low; \ |
5868 | __young; \ |
5869 | @@ -349,7 +373,7 @@ |
5870 | if (!pte_none(pte) |
5871 | && (mm != &init_mm |
5872 | || HYPERVISOR_update_va_mapping(addr, __pte(0), 0))) { |
5873 | - pte = raw_ptep_get_and_clear(ptep, pte); |
5874 | + pte = xen_ptep_get_and_clear(ptep, pte); |
5875 | pte_update(mm, addr, ptep); |
5876 | } |
5877 | return pte; |
5878 | @@ -491,24 +515,10 @@ |
5879 | #endif |
5880 | |
5881 | #if defined(CONFIG_HIGHPTE) |
5882 | -#define pte_offset_map(dir, address) \ |
5883 | -({ \ |
5884 | - pte_t *__ptep; \ |
5885 | - unsigned pfn = pmd_val(*(dir)) >> PAGE_SHIFT; \ |
5886 | - __ptep = (pte_t *)kmap_atomic_pte(pfn_to_page(pfn),KM_PTE0); \ |
5887 | - paravirt_map_pt_hook(KM_PTE0,__ptep, pfn); \ |
5888 | - __ptep = __ptep + pte_index(address); \ |
5889 | - __ptep; \ |
5890 | -}) |
5891 | -#define pte_offset_map_nested(dir, address) \ |
5892 | -({ \ |
5893 | - pte_t *__ptep; \ |
5894 | - unsigned pfn = pmd_val(*(dir)) >> PAGE_SHIFT; \ |
5895 | - __ptep = (pte_t *)kmap_atomic_pte(pfn_to_page(pfn),KM_PTE1); \ |
5896 | - paravirt_map_pt_hook(KM_PTE1,__ptep, pfn); \ |
5897 | - __ptep = __ptep + pte_index(address); \ |
5898 | - __ptep; \ |
5899 | -}) |
5900 | +#define pte_offset_map(dir, address) \ |
5901 | + ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE0) + pte_index(address)) |
5902 | +#define pte_offset_map_nested(dir, address) \ |
5903 | + ((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE1) + pte_index(address)) |
5904 | #define pte_unmap(pte) kunmap_atomic(pte, KM_PTE0) |
5905 | #define pte_unmap_nested(pte) kunmap_atomic(pte, KM_PTE1) |
5906 | #else |
5907 | @@ -587,10 +597,6 @@ |
5908 | #define io_remap_pfn_range(vma,from,pfn,size,prot) \ |
5909 | direct_remap_pfn_range(vma,from,pfn,size,prot,DOMID_IO) |
5910 | |
5911 | -#define MK_IOSPACE_PFN(space, pfn) (pfn) |
5912 | -#define GET_IOSPACE(pfn) 0 |
5913 | -#define GET_PFN(pfn) (pfn) |
5914 | - |
5915 | #include <asm-generic/pgtable.h> |
5916 | |
5917 | #endif /* _I386_PGTABLE_H */ |
5918 | --- a/include/asm-x86/mach-xen/asm/pgtable_64.h |
5919 | +++ b/include/asm-x86/mach-xen/asm/pgtable_64.h |
5920 | @@ -1,12 +1,14 @@ |
5921 | #ifndef _X86_64_PGTABLE_H |
5922 | #define _X86_64_PGTABLE_H |
5923 | |
5924 | +#include <linux/const.h> |
5925 | +#ifndef __ASSEMBLY__ |
5926 | + |
5927 | /* |
5928 | * This file contains the functions and defines necessary to modify and use |
5929 | * the x86-64 page table tree. |
5930 | */ |
5931 | #include <asm/processor.h> |
5932 | -#include <asm/fixmap.h> |
5933 | #include <asm/bitops.h> |
5934 | #include <linux/threads.h> |
5935 | #include <linux/sched.h> |
5936 | @@ -34,11 +36,9 @@ |
5937 | #endif |
5938 | |
5939 | extern pud_t level3_kernel_pgt[512]; |
5940 | -extern pud_t level3_physmem_pgt[512]; |
5941 | extern pud_t level3_ident_pgt[512]; |
5942 | extern pmd_t level2_kernel_pgt[512]; |
5943 | extern pgd_t init_level4_pgt[]; |
5944 | -extern pgd_t boot_level4_pgt[]; |
5945 | extern unsigned long __supported_pte_mask; |
5946 | |
5947 | #define swapper_pg_dir init_level4_pgt |
5948 | @@ -53,6 +53,8 @@ |
5949 | extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)]; |
5950 | #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page)) |
5951 | |
5952 | +#endif /* !__ASSEMBLY__ */ |
5953 | + |
5954 | /* |
5955 | * PGDIR_SHIFT determines what a top-level page table entry can map |
5956 | */ |
5957 | @@ -77,6 +79,8 @@ |
5958 | */ |
5959 | #define PTRS_PER_PTE 512 |
5960 | |
5961 | +#ifndef __ASSEMBLY__ |
5962 | + |
5963 | #define pte_ERROR(e) \ |
5964 | printk("%s:%d: bad pte %p(%016lx pfn %010lx).\n", __FILE__, __LINE__, \ |
5965 | &(e), __pte_val(e), pte_pfn(e)) |
5966 | @@ -119,22 +123,23 @@ |
5967 | |
5968 | #define pte_pgprot(a) (__pgprot((a).pte & ~PHYSICAL_PAGE_MASK)) |
5969 | |
5970 | -#define PMD_SIZE (1UL << PMD_SHIFT) |
5971 | +#endif /* !__ASSEMBLY__ */ |
5972 | + |
5973 | +#define PMD_SIZE (_AC(1,UL) << PMD_SHIFT) |
5974 | #define PMD_MASK (~(PMD_SIZE-1)) |
5975 | -#define PUD_SIZE (1UL << PUD_SHIFT) |
5976 | +#define PUD_SIZE (_AC(1,UL) << PUD_SHIFT) |
5977 | #define PUD_MASK (~(PUD_SIZE-1)) |
5978 | -#define PGDIR_SIZE (1UL << PGDIR_SHIFT) |
5979 | +#define PGDIR_SIZE (_AC(1,UL) << PGDIR_SHIFT) |
5980 | #define PGDIR_MASK (~(PGDIR_SIZE-1)) |
5981 | |
5982 | #define USER_PTRS_PER_PGD ((TASK_SIZE-1)/PGDIR_SIZE+1) |
5983 | #define FIRST_USER_ADDRESS 0 |
5984 | |
5985 | -#ifndef __ASSEMBLY__ |
5986 | -#define MAXMEM 0x3fffffffffffUL |
5987 | -#define VMALLOC_START 0xffffc20000000000UL |
5988 | -#define VMALLOC_END 0xffffe1ffffffffffUL |
5989 | -#define MODULES_VADDR 0xffffffff88000000UL |
5990 | -#define MODULES_END 0xfffffffffff00000UL |
5991 | +#define MAXMEM _AC(0x3fffffffffff, UL) |
5992 | +#define VMALLOC_START _AC(0xffffc20000000000, UL) |
5993 | +#define VMALLOC_END _AC(0xffffe1ffffffffff, UL) |
5994 | +#define MODULES_VADDR _AC(0xffffffff88000000, UL) |
5995 | +#define MODULES_END _AC(0xfffffffffff00000, UL) |
5996 | #define MODULES_LEN (MODULES_END - MODULES_VADDR) |
5997 | |
5998 | #define _PAGE_BIT_PRESENT 0 |
5999 | @@ -160,16 +165,18 @@ |
6000 | #define _PAGE_GLOBAL 0x100 /* Global TLB entry */ |
6001 | |
6002 | #define _PAGE_PROTNONE 0x080 /* If not present */ |
6003 | -#define _PAGE_NX (1UL<<_PAGE_BIT_NX) |
6004 | +#define _PAGE_NX (_AC(1,UL)<<_PAGE_BIT_NX) |
6005 | |
6006 | /* Mapped page is I/O or foreign and has no associated page struct. */ |
6007 | #define _PAGE_IO 0x200 |
6008 | |
6009 | +#ifndef __ASSEMBLY__ |
6010 | #if CONFIG_XEN_COMPAT <= 0x030002 |
6011 | extern unsigned int __kernel_page_user; |
6012 | #else |
6013 | #define __kernel_page_user 0 |
6014 | #endif |
6015 | +#endif |
6016 | |
6017 | #define _PAGE_TABLE (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_DIRTY) |
6018 | #define _KERNPG_TABLE (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY | __kernel_page_user) |
6019 | @@ -234,6 +241,8 @@ |
6020 | #define __S110 PAGE_SHARED_EXEC |
6021 | #define __S111 PAGE_SHARED_EXEC |
6022 | |
6023 | +#ifndef __ASSEMBLY__ |
6024 | + |
6025 | static inline unsigned long pgd_bad(pgd_t pgd) |
6026 | { |
6027 | return __pgd_val(pgd) & ~(PTE_MASK | _KERNPG_TABLE | _PAGE_USER); |
6028 | @@ -345,6 +354,20 @@ |
6029 | static inline pte_t pte_mkhuge(pte_t pte) { __pte_val(pte) |= _PAGE_PSE; return pte; } |
6030 | static inline pte_t pte_clrhuge(pte_t pte) { __pte_val(pte) &= ~_PAGE_PSE; return pte; } |
6031 | |
6032 | +static inline int ptep_test_and_clear_dirty(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep) |
6033 | +{ |
6034 | + if (!pte_dirty(*ptep)) |
6035 | + return 0; |
6036 | + return test_and_clear_bit(_PAGE_BIT_DIRTY, &ptep->pte); |
6037 | +} |
6038 | + |
6039 | +static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep) |
6040 | +{ |
6041 | + if (!pte_young(*ptep)) |
6042 | + return 0; |
6043 | + return test_and_clear_bit(_PAGE_BIT_ACCESSED, &ptep->pte); |
6044 | +} |
6045 | + |
6046 | static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
6047 | { |
6048 | pte_t pte = *ptep; |
6049 | @@ -470,18 +493,12 @@ |
6050 | * bit at the same time. */ |
6051 | #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS |
6052 | #define ptep_set_access_flags(vma, address, ptep, entry, dirty) \ |
6053 | - do { \ |
6054 | - if (dirty) \ |
6055 | - ptep_establish(vma, address, ptep, entry); \ |
6056 | - } while (0) |
6057 | - |
6058 | - |
6059 | -/* |
6060 | - * i386 says: We don't actually have these, but we want to advertise |
6061 | - * them so that we can encompass the flush here. |
6062 | - */ |
6063 | -#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY |
6064 | -#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG |
6065 | +({ \ |
6066 | + int __changed = !pte_same(*(ptep), entry); \ |
6067 | + if (__changed && (dirty)) \ |
6068 | + ptep_establish(vma, address, ptep, entry); \ |
6069 | + __changed; \ |
6070 | +}) |
6071 | |
6072 | #define __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH |
6073 | #define ptep_clear_flush_dirty(vma, address, ptep) \ |
6074 | @@ -490,7 +507,7 @@ |
6075 | int __dirty = pte_dirty(__pte); \ |
6076 | __pte = pte_mkclean(__pte); \ |
6077 | if ((vma)->vm_mm->context.pinned) \ |
6078 | - ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \ |
6079 | + (void)ptep_set_access_flags(vma, address, ptep, __pte, __dirty); \ |
6080 | else if (__dirty) \ |
6081 | set_pte(ptep, __pte); \ |
6082 | __dirty; \ |
6083 | @@ -503,7 +520,7 @@ |
6084 | int __young = pte_young(__pte); \ |
6085 | __pte = pte_mkold(__pte); \ |
6086 | if ((vma)->vm_mm->context.pinned) \ |
6087 | - ptep_set_access_flags(vma, address, ptep, __pte, __young); \ |
6088 | + (void)ptep_set_access_flags(vma, address, ptep, __pte, __young); \ |
6089 | else if (__young) \ |
6090 | set_pte(ptep, __pte); \ |
6091 | __young; \ |
6092 | @@ -517,10 +534,7 @@ |
6093 | #define __swp_entry_to_pte(x) ((pte_t) { (x).val }) |
6094 | |
6095 | extern spinlock_t pgd_lock; |
6096 | -extern struct page *pgd_list; |
6097 | -void vmalloc_sync_all(void); |
6098 | - |
6099 | -#endif /* !__ASSEMBLY__ */ |
6100 | +extern struct list_head pgd_list; |
6101 | |
6102 | extern int kern_addr_valid(unsigned long addr); |
6103 | |
6104 | @@ -559,10 +573,6 @@ |
6105 | #define io_remap_pfn_range(vma, vaddr, pfn, size, prot) \ |
6106 | direct_remap_pfn_range(vma,vaddr,pfn,size,prot,DOMID_IO) |
6107 | |
6108 | -#define MK_IOSPACE_PFN(space, pfn) (pfn) |
6109 | -#define GET_IOSPACE(pfn) 0 |
6110 | -#define GET_PFN(pfn) (pfn) |
6111 | - |
6112 | #define HAVE_ARCH_UNMAPPED_AREA |
6113 | |
6114 | #define pgtable_cache_init() do { } while (0) |
6115 | @@ -576,11 +586,14 @@ |
6116 | #define kc_offset_to_vaddr(o) \ |
6117 | (((o) & (1UL << (__VIRTUAL_MASK_SHIFT-1))) ? ((o) | (~__VIRTUAL_MASK)) : (o)) |
6118 | |
6119 | +#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG |
6120 | +#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY |
6121 | #define __HAVE_ARCH_PTEP_GET_AND_CLEAR |
6122 | #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL |
6123 | #define __HAVE_ARCH_PTEP_CLEAR_FLUSH |
6124 | #define __HAVE_ARCH_PTEP_SET_WRPROTECT |
6125 | #define __HAVE_ARCH_PTE_SAME |
6126 | #include <asm-generic/pgtable.h> |
6127 | +#endif /* !__ASSEMBLY__ */ |
6128 | |
6129 | #endif /* _X86_64_PGTABLE_H */ |
6130 | --- a/include/asm-x86/mach-xen/asm/processor_32.h |
6131 | +++ b/include/asm-x86/mach-xen/asm/processor_32.h |
6132 | @@ -21,6 +21,7 @@ |
6133 | #include <asm/percpu.h> |
6134 | #include <linux/cpumask.h> |
6135 | #include <linux/init.h> |
6136 | +#include <asm/processor-flags.h> |
6137 | #include <xen/interface/physdev.h> |
6138 | |
6139 | /* flag for disabling the tsc */ |
6140 | @@ -118,7 +119,8 @@ |
6141 | |
6142 | void __init cpu_detect(struct cpuinfo_x86 *c); |
6143 | |
6144 | -extern void identify_cpu(struct cpuinfo_x86 *); |
6145 | +extern void identify_boot_cpu(void); |
6146 | +extern void identify_secondary_cpu(struct cpuinfo_x86 *); |
6147 | extern void print_cpu_info(struct cpuinfo_x86 *); |
6148 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); |
6149 | extern unsigned short num_cache_leaves; |
6150 | @@ -129,29 +131,8 @@ |
6151 | static inline void detect_ht(struct cpuinfo_x86 *c) {} |
6152 | #endif |
6153 | |
6154 | -/* |
6155 | - * EFLAGS bits |
6156 | - */ |
6157 | -#define X86_EFLAGS_CF 0x00000001 /* Carry Flag */ |
6158 | -#define X86_EFLAGS_PF 0x00000004 /* Parity Flag */ |
6159 | -#define X86_EFLAGS_AF 0x00000010 /* Auxillary carry Flag */ |
6160 | -#define X86_EFLAGS_ZF 0x00000040 /* Zero Flag */ |
6161 | -#define X86_EFLAGS_SF 0x00000080 /* Sign Flag */ |
6162 | -#define X86_EFLAGS_TF 0x00000100 /* Trap Flag */ |
6163 | -#define X86_EFLAGS_IF 0x00000200 /* Interrupt Flag */ |
6164 | -#define X86_EFLAGS_DF 0x00000400 /* Direction Flag */ |
6165 | -#define X86_EFLAGS_OF 0x00000800 /* Overflow Flag */ |
6166 | -#define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */ |
6167 | -#define X86_EFLAGS_NT 0x00004000 /* Nested Task */ |
6168 | -#define X86_EFLAGS_RF 0x00010000 /* Resume Flag */ |
6169 | -#define X86_EFLAGS_VM 0x00020000 /* Virtual Mode */ |
6170 | -#define X86_EFLAGS_AC 0x00040000 /* Alignment Check */ |
6171 | -#define X86_EFLAGS_VIF 0x00080000 /* Virtual Interrupt Flag */ |
6172 | -#define X86_EFLAGS_VIP 0x00100000 /* Virtual Interrupt Pending */ |
6173 | -#define X86_EFLAGS_ID 0x00200000 /* CPUID detection flag */ |
6174 | - |
6175 | -static inline fastcall void xen_cpuid(unsigned int *eax, unsigned int *ebx, |
6176 | - unsigned int *ecx, unsigned int *edx) |
6177 | +static inline void xen_cpuid(unsigned int *eax, unsigned int *ebx, |
6178 | + unsigned int *ecx, unsigned int *edx) |
6179 | { |
6180 | /* ecx is often an input as well as an output. */ |
6181 | __asm__(XEN_CPUID |
6182 | @@ -165,21 +146,6 @@ |
6183 | #define load_cr3(pgdir) write_cr3(__pa(pgdir)) |
6184 | |
6185 | /* |
6186 | - * Intel CPU features in CR4 |
6187 | - */ |
6188 | -#define X86_CR4_VME 0x0001 /* enable vm86 extensions */ |
6189 | -#define X86_CR4_PVI 0x0002 /* virtual interrupts flag enable */ |
6190 | -#define X86_CR4_TSD 0x0004 /* disable time stamp at ipl 3 */ |
6191 | -#define X86_CR4_DE 0x0008 /* enable debugging extensions */ |
6192 | -#define X86_CR4_PSE 0x0010 /* enable page size extensions */ |
6193 | -#define X86_CR4_PAE 0x0020 /* enable physical address extensions */ |
6194 | -#define X86_CR4_MCE 0x0040 /* Machine check enable */ |
6195 | -#define X86_CR4_PGE 0x0080 /* enable global pages */ |
6196 | -#define X86_CR4_PCE 0x0100 /* enable performance counters at ipl 3 */ |
6197 | -#define X86_CR4_OSFXSR 0x0200 /* enable fast FPU save and restore */ |
6198 | -#define X86_CR4_OSXMMEXCPT 0x0400 /* enable unmasked SSE exceptions */ |
6199 | - |
6200 | -/* |
6201 | * Save the cr4 feature set we're using (ie |
6202 | * Pentium 4MB enable and PPro Global page |
6203 | * enable), so that any CPU's that boot up |
6204 | @@ -206,26 +172,6 @@ |
6205 | } |
6206 | |
6207 | /* |
6208 | - * NSC/Cyrix CPU configuration register indexes |
6209 | - */ |
6210 | - |
6211 | -#define CX86_PCR0 0x20 |
6212 | -#define CX86_GCR 0xb8 |
6213 | -#define CX86_CCR0 0xc0 |
6214 | -#define CX86_CCR1 0xc1 |
6215 | -#define CX86_CCR2 0xc2 |
6216 | -#define CX86_CCR3 0xc3 |
6217 | -#define CX86_CCR4 0xe8 |
6218 | -#define CX86_CCR5 0xe9 |
6219 | -#define CX86_CCR6 0xea |
6220 | -#define CX86_CCR7 0xeb |
6221 | -#define CX86_PCR1 0xf0 |
6222 | -#define CX86_DIR0 0xfe |
6223 | -#define CX86_DIR1 0xff |
6224 | -#define CX86_ARR_BASE 0xc4 |
6225 | -#define CX86_RCR_BASE 0xdc |
6226 | - |
6227 | -/* |
6228 | * NSC/Cyrix CPU indexed register access macros |
6229 | */ |
6230 | |
6231 | @@ -351,7 +297,8 @@ |
6232 | struct thread_struct; |
6233 | |
6234 | #ifndef CONFIG_X86_NO_TSS |
6235 | -struct tss_struct { |
6236 | +/* This is the TSS defined by the hardware. */ |
6237 | +struct i386_hw_tss { |
6238 | unsigned short back_link,__blh; |
6239 | unsigned long esp0; |
6240 | unsigned short ss0,__ss0h; |
6241 | @@ -375,6 +322,11 @@ |
6242 | unsigned short gs, __gsh; |
6243 | unsigned short ldt, __ldth; |
6244 | unsigned short trace, io_bitmap_base; |
6245 | +} __attribute__((packed)); |
6246 | + |
6247 | +struct tss_struct { |
6248 | + struct i386_hw_tss x86_tss; |
6249 | + |
6250 | /* |
6251 | * The extra 1 is there because the CPU will access an |
6252 | * additional byte beyond the end of the IO permission |
6253 | @@ -428,10 +380,11 @@ |
6254 | }; |
6255 | |
6256 | #define INIT_THREAD { \ |
6257 | + .esp0 = sizeof(init_stack) + (long)&init_stack, \ |
6258 | .vm86_info = NULL, \ |
6259 | .sysenter_cs = __KERNEL_CS, \ |
6260 | .io_bitmap_ptr = NULL, \ |
6261 | - .fs = __KERNEL_PDA, \ |
6262 | + .fs = __KERNEL_PERCPU, \ |
6263 | } |
6264 | |
6265 | /* |
6266 | @@ -441,10 +394,12 @@ |
6267 | * be within the limit. |
6268 | */ |
6269 | #define INIT_TSS { \ |
6270 | - .esp0 = sizeof(init_stack) + (long)&init_stack, \ |
6271 | - .ss0 = __KERNEL_DS, \ |
6272 | - .ss1 = __KERNEL_CS, \ |
6273 | - .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \ |
6274 | + .x86_tss = { \ |
6275 | + .esp0 = sizeof(init_stack) + (long)&init_stack, \ |
6276 | + .ss0 = __KERNEL_DS, \ |
6277 | + .ss1 = __KERNEL_CS, \ |
6278 | + .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \ |
6279 | + }, \ |
6280 | .io_bitmap = { [ 0 ... IO_BITMAP_LONGS] = ~0 }, \ |
6281 | } |
6282 | |
6283 | @@ -551,38 +506,33 @@ |
6284 | |
6285 | #define cpu_relax() rep_nop() |
6286 | |
6287 | -#define paravirt_enabled() 0 |
6288 | -#define __cpuid xen_cpuid |
6289 | - |
6290 | #ifndef CONFIG_X86_NO_TSS |
6291 | -static inline void __load_esp0(struct tss_struct *tss, struct thread_struct *thread) |
6292 | +static inline void native_load_esp0(struct tss_struct *tss, struct thread_struct *thread) |
6293 | { |
6294 | - tss->esp0 = thread->esp0; |
6295 | + tss->x86_tss.esp0 = thread->esp0; |
6296 | /* This can only happen when SEP is enabled, no need to test "SEP"arately */ |
6297 | - if (unlikely(tss->ss1 != thread->sysenter_cs)) { |
6298 | - tss->ss1 = thread->sysenter_cs; |
6299 | + if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
6300 | + tss->x86_tss.ss1 = thread->sysenter_cs; |
6301 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); |
6302 | } |
6303 | } |
6304 | -#define load_esp0(tss, thread) \ |
6305 | - __load_esp0(tss, thread) |
6306 | #else |
6307 | -#define load_esp0(tss, thread) do { \ |
6308 | +#define xen_load_esp0(tss, thread) do { \ |
6309 | if (HYPERVISOR_stack_switch(__KERNEL_DS, (thread)->esp0)) \ |
6310 | BUG(); \ |
6311 | } while (0) |
6312 | #endif |
6313 | |
6314 | |
6315 | -/* |
6316 | - * These special macros can be used to get or set a debugging register |
6317 | - */ |
6318 | -#define get_debugreg(var, register) \ |
6319 | - (var) = HYPERVISOR_get_debugreg(register) |
6320 | -#define set_debugreg(value, register) \ |
6321 | - WARN_ON(HYPERVISOR_set_debugreg(register, value)) |
6322 | +static inline unsigned long xen_get_debugreg(int regno) |
6323 | +{ |
6324 | + return HYPERVISOR_get_debugreg(regno); |
6325 | +} |
6326 | |
6327 | -#define set_iopl_mask xen_set_iopl_mask |
6328 | +static inline void xen_set_debugreg(int regno, unsigned long value) |
6329 | +{ |
6330 | + WARN_ON(HYPERVISOR_set_debugreg(regno, value)); |
6331 | +} |
6332 | |
6333 | /* |
6334 | * Set IOPL bits in EFLAGS from given mask |
6335 | @@ -597,6 +547,21 @@ |
6336 | } |
6337 | |
6338 | |
6339 | +#define paravirt_enabled() 0 |
6340 | +#define __cpuid xen_cpuid |
6341 | + |
6342 | +#define load_esp0 xen_load_esp0 |
6343 | + |
6344 | +/* |
6345 | + * These special macros can be used to get or set a debugging register |
6346 | + */ |
6347 | +#define get_debugreg(var, register) \ |
6348 | + (var) = xen_get_debugreg(register) |
6349 | +#define set_debugreg(value, register) \ |
6350 | + xen_set_debugreg(register, value) |
6351 | + |
6352 | +#define set_iopl_mask xen_set_iopl_mask |
6353 | + |
6354 | /* |
6355 | * Generic CPUID function |
6356 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx |
6357 | @@ -749,8 +714,14 @@ |
6358 | extern void enable_sep_cpu(void); |
6359 | extern int sysenter_setup(void); |
6360 | |
6361 | -extern int init_gdt(int cpu, struct task_struct *idle); |
6362 | +/* Defined in head.S */ |
6363 | +extern struct Xgt_desc_struct early_gdt_descr; |
6364 | + |
6365 | extern void cpu_set_gdt(int); |
6366 | -extern void secondary_cpu_init(void); |
6367 | +extern void switch_to_new_gdt(void); |
6368 | +extern void cpu_init(void); |
6369 | +extern void init_gdt(int cpu); |
6370 | + |
6371 | +extern int force_mwait; |
6372 | |
6373 | #endif /* __ASM_I386_PROCESSOR_H */ |
6374 | --- a/include/asm-x86/mach-xen/asm/processor_64.h |
6375 | +++ b/include/asm-x86/mach-xen/asm/processor_64.h |
6376 | @@ -20,6 +20,7 @@ |
6377 | #include <asm/percpu.h> |
6378 | #include <linux/personality.h> |
6379 | #include <linux/cpumask.h> |
6380 | +#include <asm/processor-flags.h> |
6381 | |
6382 | #define TF_MASK 0x00000100 |
6383 | #define IF_MASK 0x00000200 |
6384 | @@ -103,42 +104,6 @@ |
6385 | extern unsigned short num_cache_leaves; |
6386 | |
6387 | /* |
6388 | - * EFLAGS bits |
6389 | - */ |
6390 | -#define X86_EFLAGS_CF 0x00000001 /* Carry Flag */ |
6391 | -#define X86_EFLAGS_PF 0x00000004 /* Parity Flag */ |
6392 | -#define X86_EFLAGS_AF 0x00000010 /* Auxillary carry Flag */ |
6393 | -#define X86_EFLAGS_ZF 0x00000040 /* Zero Flag */ |
6394 | -#define X86_EFLAGS_SF 0x00000080 /* Sign Flag */ |
6395 | -#define X86_EFLAGS_TF 0x00000100 /* Trap Flag */ |
6396 | -#define X86_EFLAGS_IF 0x00000200 /* Interrupt Flag */ |
6397 | -#define X86_EFLAGS_DF 0x00000400 /* Direction Flag */ |
6398 | -#define X86_EFLAGS_OF 0x00000800 /* Overflow Flag */ |
6399 | -#define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */ |
6400 | -#define X86_EFLAGS_NT 0x00004000 /* Nested Task */ |
6401 | -#define X86_EFLAGS_RF 0x00010000 /* Resume Flag */ |
6402 | -#define X86_EFLAGS_VM 0x00020000 /* Virtual Mode */ |
6403 | -#define X86_EFLAGS_AC 0x00040000 /* Alignment Check */ |
6404 | -#define X86_EFLAGS_VIF 0x00080000 /* Virtual Interrupt Flag */ |
6405 | -#define X86_EFLAGS_VIP 0x00100000 /* Virtual Interrupt Pending */ |
6406 | -#define X86_EFLAGS_ID 0x00200000 /* CPUID detection flag */ |
6407 | - |
6408 | -/* |
6409 | - * Intel CPU features in CR4 |
6410 | - */ |
6411 | -#define X86_CR4_VME 0x0001 /* enable vm86 extensions */ |
6412 | -#define X86_CR4_PVI 0x0002 /* virtual interrupts flag enable */ |
6413 | -#define X86_CR4_TSD 0x0004 /* disable time stamp at ipl 3 */ |
6414 | -#define X86_CR4_DE 0x0008 /* enable debugging extensions */ |
6415 | -#define X86_CR4_PSE 0x0010 /* enable page size extensions */ |
6416 | -#define X86_CR4_PAE 0x0020 /* enable physical address extensions */ |
6417 | -#define X86_CR4_MCE 0x0040 /* Machine check enable */ |
6418 | -#define X86_CR4_PGE 0x0080 /* enable global pages */ |
6419 | -#define X86_CR4_PCE 0x0100 /* enable performance counters at ipl 3 */ |
6420 | -#define X86_CR4_OSFXSR 0x0200 /* enable fast FPU save and restore */ |
6421 | -#define X86_CR4_OSXMMEXCPT 0x0400 /* enable unmasked SSE exceptions */ |
6422 | - |
6423 | -/* |
6424 | * Save the cr4 feature set we're using (ie |
6425 | * Pentium 4MB enable and PPro Global page |
6426 | * enable), so that any CPU's that boot up |
6427 | @@ -203,7 +168,7 @@ |
6428 | u32 mxcsr; |
6429 | u32 mxcsr_mask; |
6430 | u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */ |
6431 | - u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 128 bytes */ |
6432 | + u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */ |
6433 | u32 padding[24]; |
6434 | } __attribute__ ((aligned (16))); |
6435 | |
6436 | @@ -436,22 +401,6 @@ |
6437 | #define cpu_relax() rep_nop() |
6438 | |
6439 | /* |
6440 | - * NSC/Cyrix CPU configuration register indexes |
6441 | - */ |
6442 | -#define CX86_CCR0 0xc0 |
6443 | -#define CX86_CCR1 0xc1 |
6444 | -#define CX86_CCR2 0xc2 |
6445 | -#define CX86_CCR3 0xc3 |
6446 | -#define CX86_CCR4 0xe8 |
6447 | -#define CX86_CCR5 0xe9 |
6448 | -#define CX86_CCR6 0xea |
6449 | -#define CX86_CCR7 0xeb |
6450 | -#define CX86_DIR0 0xfe |
6451 | -#define CX86_DIR1 0xff |
6452 | -#define CX86_ARR_BASE 0xc4 |
6453 | -#define CX86_RCR_BASE 0xdc |
6454 | - |
6455 | -/* |
6456 | * NSC/Cyrix CPU indexed register access macros |
6457 | */ |
6458 | |
6459 | --- a/include/asm-x86/mach-xen/asm/scatterlist_32.h |
6460 | +++ b/include/asm-x86/mach-xen/asm/scatterlist_32.h |
6461 | @@ -1,6 +1,8 @@ |
6462 | #ifndef _I386_SCATTERLIST_H |
6463 | #define _I386_SCATTERLIST_H |
6464 | |
6465 | +#include <asm/types.h> |
6466 | + |
6467 | struct scatterlist { |
6468 | struct page *page; |
6469 | unsigned int offset; |
6470 | --- a/include/asm-x86/mach-xen/asm/segment_32.h |
6471 | +++ b/include/asm-x86/mach-xen/asm/segment_32.h |
6472 | @@ -39,7 +39,7 @@ |
6473 | * 25 - APM BIOS support |
6474 | * |
6475 | * 26 - ESPFIX small SS |
6476 | - * 27 - PDA [ per-cpu private data area ] |
6477 | + * 27 - per-cpu [ offset to per-cpu data area ] |
6478 | * 28 - unused |
6479 | * 29 - unused |
6480 | * 30 - unused |
6481 | @@ -74,8 +74,12 @@ |
6482 | #define GDT_ENTRY_ESPFIX_SS (GDT_ENTRY_KERNEL_BASE + 14) |
6483 | #define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS * 8) |
6484 | |
6485 | -#define GDT_ENTRY_PDA (GDT_ENTRY_KERNEL_BASE + 15) |
6486 | -#define __KERNEL_PDA (GDT_ENTRY_PDA * 8) |
6487 | +#define GDT_ENTRY_PERCPU (GDT_ENTRY_KERNEL_BASE + 15) |
6488 | +#ifdef CONFIG_SMP |
6489 | +#define __KERNEL_PERCPU (GDT_ENTRY_PERCPU * 8) |
6490 | +#else |
6491 | +#define __KERNEL_PERCPU 0 |
6492 | +#endif |
6493 | |
6494 | #define GDT_ENTRY_DOUBLEFAULT_TSS 31 |
6495 | |
6496 | --- a/include/asm-x86/mach-xen/asm/smp_32.h |
6497 | +++ b/include/asm-x86/mach-xen/asm/smp_32.h |
6498 | @@ -8,19 +8,15 @@ |
6499 | #include <linux/kernel.h> |
6500 | #include <linux/threads.h> |
6501 | #include <linux/cpumask.h> |
6502 | -#include <asm/pda.h> |
6503 | #endif |
6504 | |
6505 | -#ifdef CONFIG_X86_LOCAL_APIC |
6506 | -#ifndef __ASSEMBLY__ |
6507 | -#include <asm/fixmap.h> |
6508 | +#if defined(CONFIG_X86_LOCAL_APIC) && !defined(__ASSEMBLY__) |
6509 | #include <asm/bitops.h> |
6510 | #include <asm/mpspec.h> |
6511 | +#include <asm/apic.h> |
6512 | #ifdef CONFIG_X86_IO_APIC |
6513 | #include <asm/io_apic.h> |
6514 | #endif |
6515 | -#include <asm/apic.h> |
6516 | -#endif |
6517 | #endif |
6518 | |
6519 | #define BAD_APICID 0xFFu |
6520 | @@ -52,9 +48,76 @@ |
6521 | extern void cpu_uninit(void); |
6522 | #endif |
6523 | |
6524 | -#ifndef CONFIG_PARAVIRT |
6525 | +#ifndef CONFIG_XEN |
6526 | +struct smp_ops |
6527 | +{ |
6528 | + void (*smp_prepare_boot_cpu)(void); |
6529 | + void (*smp_prepare_cpus)(unsigned max_cpus); |
6530 | + int (*cpu_up)(unsigned cpu); |
6531 | + void (*smp_cpus_done)(unsigned max_cpus); |
6532 | + |
6533 | + void (*smp_send_stop)(void); |
6534 | + void (*smp_send_reschedule)(int cpu); |
6535 | + int (*smp_call_function_mask)(cpumask_t mask, |
6536 | + void (*func)(void *info), void *info, |
6537 | + int wait); |
6538 | +}; |
6539 | + |
6540 | +extern struct smp_ops smp_ops; |
6541 | + |
6542 | +static inline void smp_prepare_boot_cpu(void) |
6543 | +{ |
6544 | + smp_ops.smp_prepare_boot_cpu(); |
6545 | +} |
6546 | +static inline void smp_prepare_cpus(unsigned int max_cpus) |
6547 | +{ |
6548 | + smp_ops.smp_prepare_cpus(max_cpus); |
6549 | +} |
6550 | +static inline int __cpu_up(unsigned int cpu) |
6551 | +{ |
6552 | + return smp_ops.cpu_up(cpu); |
6553 | +} |
6554 | +static inline void smp_cpus_done(unsigned int max_cpus) |
6555 | +{ |
6556 | + smp_ops.smp_cpus_done(max_cpus); |
6557 | +} |
6558 | + |
6559 | +static inline void smp_send_stop(void) |
6560 | +{ |
6561 | + smp_ops.smp_send_stop(); |
6562 | +} |
6563 | +static inline void smp_send_reschedule(int cpu) |
6564 | +{ |
6565 | + smp_ops.smp_send_reschedule(cpu); |
6566 | +} |
6567 | +static inline int smp_call_function_mask(cpumask_t mask, |
6568 | + void (*func) (void *info), void *info, |
6569 | + int wait) |
6570 | +{ |
6571 | + return smp_ops.smp_call_function_mask(mask, func, info, wait); |
6572 | +} |
6573 | + |
6574 | +void native_smp_prepare_boot_cpu(void); |
6575 | +void native_smp_prepare_cpus(unsigned int max_cpus); |
6576 | +int native_cpu_up(unsigned int cpunum); |
6577 | +void native_smp_cpus_done(unsigned int max_cpus); |
6578 | + |
6579 | #define startup_ipi_hook(phys_apicid, start_eip, start_esp) \ |
6580 | do { } while (0) |
6581 | + |
6582 | +#else |
6583 | + |
6584 | + |
6585 | +void xen_smp_send_stop(void); |
6586 | +void xen_smp_send_reschedule(int cpu); |
6587 | +int xen_smp_call_function_mask(cpumask_t mask, |
6588 | + void (*func) (void *info), void *info, |
6589 | + int wait); |
6590 | + |
6591 | +#define smp_send_stop xen_smp_send_stop |
6592 | +#define smp_send_reschedule xen_smp_send_reschedule |
6593 | +#define smp_call_function_mask xen_smp_call_function_mask |
6594 | + |
6595 | #endif |
6596 | |
6597 | /* |
6598 | @@ -62,7 +125,8 @@ |
6599 | * from the initial startup. We map APIC_BASE very early in page_setup(), |
6600 | * so this is correct in the x86 case. |
6601 | */ |
6602 | -#define raw_smp_processor_id() (read_pda(cpu_number)) |
6603 | +DECLARE_PER_CPU(int, cpu_number); |
6604 | +#define raw_smp_processor_id() (x86_read_percpu(cpu_number)) |
6605 | |
6606 | extern cpumask_t cpu_possible_map; |
6607 | #define cpu_callin_map cpu_possible_map |
6608 | @@ -73,20 +137,6 @@ |
6609 | return cpus_weight(cpu_possible_map); |
6610 | } |
6611 | |
6612 | -#ifdef CONFIG_X86_LOCAL_APIC |
6613 | - |
6614 | -#ifdef APIC_DEFINITION |
6615 | -extern int hard_smp_processor_id(void); |
6616 | -#else |
6617 | -#include <mach_apicdef.h> |
6618 | -static inline int hard_smp_processor_id(void) |
6619 | -{ |
6620 | - /* we don't want to mark this access volatile - bad code generation */ |
6621 | - return GET_APIC_ID(*(unsigned long *)(APIC_BASE+APIC_ID)); |
6622 | -} |
6623 | -#endif |
6624 | -#endif |
6625 | - |
6626 | #define safe_smp_processor_id() smp_processor_id() |
6627 | extern int __cpu_disable(void); |
6628 | extern void __cpu_die(unsigned int cpu); |
6629 | @@ -102,10 +152,31 @@ |
6630 | |
6631 | #define NO_PROC_ID 0xFF /* No processor magic marker */ |
6632 | |
6633 | -#endif |
6634 | +#endif /* CONFIG_SMP */ |
6635 | |
6636 | #ifndef __ASSEMBLY__ |
6637 | |
6638 | +#ifdef CONFIG_X86_LOCAL_APIC |
6639 | + |
6640 | +#ifdef APIC_DEFINITION |
6641 | +extern int hard_smp_processor_id(void); |
6642 | +#else |
6643 | +#include <mach_apicdef.h> |
6644 | +static inline int hard_smp_processor_id(void) |
6645 | +{ |
6646 | + /* we don't want to mark this access volatile - bad code generation */ |
6647 | + return GET_APIC_ID(*(unsigned long *)(APIC_BASE+APIC_ID)); |
6648 | +} |
6649 | +#endif /* APIC_DEFINITION */ |
6650 | + |
6651 | +#else /* CONFIG_X86_LOCAL_APIC */ |
6652 | + |
6653 | +#ifndef CONFIG_SMP |
6654 | +#define hard_smp_processor_id() 0 |
6655 | +#endif |
6656 | + |
6657 | +#endif /* CONFIG_X86_LOCAL_APIC */ |
6658 | + |
6659 | extern u8 apicid_2_node[]; |
6660 | |
6661 | #ifdef CONFIG_X86_LOCAL_APIC |
6662 | --- a/include/asm-x86/mach-xen/asm/smp_64.h |
6663 | +++ b/include/asm-x86/mach-xen/asm/smp_64.h |
6664 | @@ -11,12 +11,11 @@ |
6665 | extern int disable_apic; |
6666 | |
6667 | #ifdef CONFIG_X86_LOCAL_APIC |
6668 | -#include <asm/fixmap.h> |
6669 | #include <asm/mpspec.h> |
6670 | +#include <asm/apic.h> |
6671 | #ifdef CONFIG_X86_IO_APIC |
6672 | #include <asm/io_apic.h> |
6673 | #endif |
6674 | -#include <asm/apic.h> |
6675 | #include <asm/thread_info.h> |
6676 | #endif |
6677 | |
6678 | @@ -41,7 +40,6 @@ |
6679 | extern void unlock_ipi_call_lock(void); |
6680 | extern int smp_num_siblings; |
6681 | extern void smp_send_reschedule(int cpu); |
6682 | -void smp_stop_cpu(void); |
6683 | |
6684 | extern cpumask_t cpu_sibling_map[NR_CPUS]; |
6685 | extern cpumask_t cpu_core_map[NR_CPUS]; |
6686 | @@ -62,14 +60,6 @@ |
6687 | |
6688 | #define raw_smp_processor_id() read_pda(cpunumber) |
6689 | |
6690 | -#ifdef CONFIG_X86_LOCAL_APIC |
6691 | -static inline int hard_smp_processor_id(void) |
6692 | -{ |
6693 | - /* we don't want to mark this access volatile - bad code generation */ |
6694 | - return GET_APIC_ID(*(unsigned int *)(APIC_BASE+APIC_ID)); |
6695 | -} |
6696 | -#endif |
6697 | - |
6698 | extern int __cpu_disable(void); |
6699 | extern void __cpu_die(unsigned int cpu); |
6700 | extern void prefill_possible_map(void); |
6701 | @@ -78,6 +68,14 @@ |
6702 | |
6703 | #define NO_PROC_ID 0xFF /* No processor magic marker */ |
6704 | |
6705 | +#endif /* CONFIG_SMP */ |
6706 | + |
6707 | +#ifdef CONFIG_X86_LOCAL_APIC |
6708 | +static inline int hard_smp_processor_id(void) |
6709 | +{ |
6710 | + /* we don't want to mark this access volatile - bad code generation */ |
6711 | + return GET_APIC_ID(*(unsigned int *)(APIC_BASE+APIC_ID)); |
6712 | +} |
6713 | #endif |
6714 | |
6715 | /* |
6716 | --- a/include/asm-x86/mach-xen/asm/system_32.h |
6717 | +++ b/include/asm-x86/mach-xen/asm/system_32.h |
6718 | @@ -4,7 +4,7 @@ |
6719 | #include <linux/kernel.h> |
6720 | #include <asm/segment.h> |
6721 | #include <asm/cpufeature.h> |
6722 | -#include <linux/bitops.h> /* for LOCK_PREFIX */ |
6723 | +#include <asm/cmpxchg.h> |
6724 | #include <asm/synch_bitops.h> |
6725 | #include <asm/hypervisor.h> |
6726 | |
6727 | @@ -90,308 +90,102 @@ |
6728 | #define savesegment(seg, value) \ |
6729 | asm volatile("mov %%" #seg ",%0":"=rm" (value)) |
6730 | |
6731 | -#define read_cr0() ({ \ |
6732 | - unsigned int __dummy; \ |
6733 | - __asm__ __volatile__( \ |
6734 | - "movl %%cr0,%0\n\t" \ |
6735 | - :"=r" (__dummy)); \ |
6736 | - __dummy; \ |
6737 | -}) |
6738 | -#define write_cr0(x) \ |
6739 | - __asm__ __volatile__("movl %0,%%cr0": :"r" (x)) |
6740 | - |
6741 | -#define read_cr2() (current_vcpu_info()->arch.cr2) |
6742 | -#define write_cr2(x) \ |
6743 | - __asm__ __volatile__("movl %0,%%cr2": :"r" (x)) |
6744 | - |
6745 | -#define read_cr3() ({ \ |
6746 | - unsigned int __dummy; \ |
6747 | - __asm__ ( \ |
6748 | - "movl %%cr3,%0\n\t" \ |
6749 | - :"=r" (__dummy)); \ |
6750 | - __dummy = xen_cr3_to_pfn(__dummy); \ |
6751 | - mfn_to_pfn(__dummy) << PAGE_SHIFT; \ |
6752 | -}) |
6753 | -#define write_cr3(x) ({ \ |
6754 | - unsigned int __dummy = pfn_to_mfn((x) >> PAGE_SHIFT); \ |
6755 | - __dummy = xen_pfn_to_cr3(__dummy); \ |
6756 | - __asm__ __volatile__("movl %0,%%cr3": :"r" (__dummy)); \ |
6757 | -}) |
6758 | -#define read_cr4() ({ \ |
6759 | - unsigned int __dummy; \ |
6760 | - __asm__( \ |
6761 | - "movl %%cr4,%0\n\t" \ |
6762 | - :"=r" (__dummy)); \ |
6763 | - __dummy; \ |
6764 | -}) |
6765 | -#define read_cr4_safe() ({ \ |
6766 | - unsigned int __dummy; \ |
6767 | - /* This could fault if %cr4 does not exist */ \ |
6768 | - __asm__("1: movl %%cr4, %0 \n" \ |
6769 | - "2: \n" \ |
6770 | - ".section __ex_table,\"a\" \n" \ |
6771 | - ".long 1b,2b \n" \ |
6772 | - ".previous \n" \ |
6773 | - : "=r" (__dummy): "0" (0)); \ |
6774 | - __dummy; \ |
6775 | -}) |
6776 | - |
6777 | -#define write_cr4(x) \ |
6778 | - __asm__ __volatile__("movl %0,%%cr4": :"r" (x)) |
6779 | - |
6780 | -#define wbinvd() \ |
6781 | - __asm__ __volatile__ ("wbinvd": : :"memory") |
6782 | - |
6783 | -/* Clear the 'TS' bit */ |
6784 | -#define clts() (HYPERVISOR_fpu_taskswitch(0)) |
6785 | - |
6786 | -/* Set the 'TS' bit */ |
6787 | -#define stts() (HYPERVISOR_fpu_taskswitch(1)) |
6788 | - |
6789 | -#endif /* __KERNEL__ */ |
6790 | - |
6791 | -static inline unsigned long get_limit(unsigned long segment) |
6792 | +static inline void xen_clts(void) |
6793 | { |
6794 | - unsigned long __limit; |
6795 | - __asm__("lsll %1,%0" |
6796 | - :"=r" (__limit):"r" (segment)); |
6797 | - return __limit+1; |
6798 | + HYPERVISOR_fpu_taskswitch(0); |
6799 | } |
6800 | |
6801 | -#define nop() __asm__ __volatile__ ("nop") |
6802 | - |
6803 | -#define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr)))) |
6804 | - |
6805 | -#define tas(ptr) (xchg((ptr),1)) |
6806 | - |
6807 | -struct __xchg_dummy { unsigned long a[100]; }; |
6808 | -#define __xg(x) ((struct __xchg_dummy *)(x)) |
6809 | +static inline unsigned long xen_read_cr0(void) |
6810 | +{ |
6811 | + unsigned long val; |
6812 | + asm volatile("movl %%cr0,%0\n\t" :"=r" (val)); |
6813 | + return val; |
6814 | +} |
6815 | |
6816 | +static inline void xen_write_cr0(unsigned long val) |
6817 | +{ |
6818 | + asm volatile("movl %0,%%cr0": :"r" (val)); |
6819 | +} |
6820 | |
6821 | -#ifdef CONFIG_X86_CMPXCHG64 |
6822 | +#define xen_read_cr2() (current_vcpu_info()->arch.cr2) |
6823 | |
6824 | -/* |
6825 | - * The semantics of XCHGCMP8B are a bit strange, this is why |
6826 | - * there is a loop and the loading of %%eax and %%edx has to |
6827 | - * be inside. This inlines well in most cases, the cached |
6828 | - * cost is around ~38 cycles. (in the future we might want |
6829 | - * to do an SIMD/3DNOW!/MMX/FPU 64-bit store here, but that |
6830 | - * might have an implicit FPU-save as a cost, so it's not |
6831 | - * clear which path to go.) |
6832 | - * |
6833 | - * cmpxchg8b must be used with the lock prefix here to allow |
6834 | - * the instruction to be executed atomically, see page 3-102 |
6835 | - * of the instruction set reference 24319102.pdf. We need |
6836 | - * the reader side to see the coherent 64bit value. |
6837 | - */ |
6838 | -static inline void __set_64bit (unsigned long long * ptr, |
6839 | - unsigned int low, unsigned int high) |
6840 | +static inline void xen_write_cr2(unsigned long val) |
6841 | { |
6842 | - __asm__ __volatile__ ( |
6843 | - "\n1:\t" |
6844 | - "movl (%0), %%eax\n\t" |
6845 | - "movl 4(%0), %%edx\n\t" |
6846 | - "lock cmpxchg8b (%0)\n\t" |
6847 | - "jnz 1b" |
6848 | - : /* no outputs */ |
6849 | - : "D"(ptr), |
6850 | - "b"(low), |
6851 | - "c"(high) |
6852 | - : "ax","dx","memory"); |
6853 | + asm volatile("movl %0,%%cr2": :"r" (val)); |
6854 | } |
6855 | |
6856 | -static inline void __set_64bit_constant (unsigned long long *ptr, |
6857 | - unsigned long long value) |
6858 | +static inline unsigned long xen_read_cr3(void) |
6859 | { |
6860 | - __set_64bit(ptr,(unsigned int)(value), (unsigned int)((value)>>32ULL)); |
6861 | + unsigned long val; |
6862 | + asm volatile("movl %%cr3,%0\n\t" :"=r" (val)); |
6863 | + return mfn_to_pfn(xen_cr3_to_pfn(val)) << PAGE_SHIFT; |
6864 | } |
6865 | -#define ll_low(x) *(((unsigned int*)&(x))+0) |
6866 | -#define ll_high(x) *(((unsigned int*)&(x))+1) |
6867 | |
6868 | -static inline void __set_64bit_var (unsigned long long *ptr, |
6869 | - unsigned long long value) |
6870 | +static inline void xen_write_cr3(unsigned long val) |
6871 | { |
6872 | - __set_64bit(ptr,ll_low(value), ll_high(value)); |
6873 | + val = xen_pfn_to_cr3(pfn_to_mfn(val >> PAGE_SHIFT)); |
6874 | + asm volatile("movl %0,%%cr3": :"r" (val)); |
6875 | } |
6876 | |
6877 | -#define set_64bit(ptr,value) \ |
6878 | -(__builtin_constant_p(value) ? \ |
6879 | - __set_64bit_constant(ptr, value) : \ |
6880 | - __set_64bit_var(ptr, value) ) |
6881 | +static inline unsigned long xen_read_cr4(void) |
6882 | +{ |
6883 | + unsigned long val; |
6884 | + asm volatile("movl %%cr4,%0\n\t" :"=r" (val)); |
6885 | + return val; |
6886 | +} |
6887 | |
6888 | -#define _set_64bit(ptr,value) \ |
6889 | -(__builtin_constant_p(value) ? \ |
6890 | - __set_64bit(ptr, (unsigned int)(value), (unsigned int)((value)>>32ULL) ) : \ |
6891 | - __set_64bit(ptr, ll_low(value), ll_high(value)) ) |
6892 | +static inline unsigned long xen_read_cr4_safe(void) |
6893 | +{ |
6894 | + unsigned long val; |
6895 | + /* This could fault if %cr4 does not exist */ |
6896 | + asm("1: movl %%cr4, %0 \n" |
6897 | + "2: \n" |
6898 | + ".section __ex_table,\"a\" \n" |
6899 | + ".long 1b,2b \n" |
6900 | + ".previous \n" |
6901 | + : "=r" (val): "0" (0)); |
6902 | + return val; |
6903 | +} |
6904 | |
6905 | -#endif |
6906 | +static inline void xen_write_cr4(unsigned long val) |
6907 | +{ |
6908 | + asm volatile("movl %0,%%cr4": :"r" (val)); |
6909 | +} |
6910 | |
6911 | -/* |
6912 | - * Note: no "lock" prefix even on SMP: xchg always implies lock anyway |
6913 | - * Note 2: xchg has side effect, so that attribute volatile is necessary, |
6914 | - * but generally the primitive is invalid, *ptr is output argument. --ANK |
6915 | - */ |
6916 | -static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size) |
6917 | +static inline void xen_wbinvd(void) |
6918 | { |
6919 | - switch (size) { |
6920 | - case 1: |
6921 | - __asm__ __volatile__("xchgb %b0,%1" |
6922 | - :"=q" (x) |
6923 | - :"m" (*__xg(ptr)), "0" (x) |
6924 | - :"memory"); |
6925 | - break; |
6926 | - case 2: |
6927 | - __asm__ __volatile__("xchgw %w0,%1" |
6928 | - :"=r" (x) |
6929 | - :"m" (*__xg(ptr)), "0" (x) |
6930 | - :"memory"); |
6931 | - break; |
6932 | - case 4: |
6933 | - __asm__ __volatile__("xchgl %0,%1" |
6934 | - :"=r" (x) |
6935 | - :"m" (*__xg(ptr)), "0" (x) |
6936 | - :"memory"); |
6937 | - break; |
6938 | - } |
6939 | - return x; |
6940 | + asm volatile("wbinvd": : :"memory"); |
6941 | } |
6942 | |
6943 | -/* |
6944 | - * Atomic compare and exchange. Compare OLD with MEM, if identical, |
6945 | - * store NEW in MEM. Return the initial value in MEM. Success is |
6946 | - * indicated by comparing RETURN with OLD. |
6947 | - */ |
6948 | +#define read_cr0() (xen_read_cr0()) |
6949 | +#define write_cr0(x) (xen_write_cr0(x)) |
6950 | +#define read_cr2() (xen_read_cr2()) |
6951 | +#define write_cr2(x) (xen_write_cr2(x)) |
6952 | +#define read_cr3() (xen_read_cr3()) |
6953 | +#define write_cr3(x) (xen_write_cr3(x)) |
6954 | +#define read_cr4() (xen_read_cr4()) |
6955 | +#define read_cr4_safe() (xen_read_cr4_safe()) |
6956 | +#define write_cr4(x) (xen_write_cr4(x)) |
6957 | +#define wbinvd() (xen_wbinvd()) |
6958 | |
6959 | -#ifdef CONFIG_X86_CMPXCHG |
6960 | -#define __HAVE_ARCH_CMPXCHG 1 |
6961 | -#define cmpxchg(ptr,o,n)\ |
6962 | - ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\ |
6963 | - (unsigned long)(n),sizeof(*(ptr)))) |
6964 | -#define sync_cmpxchg(ptr,o,n)\ |
6965 | - ((__typeof__(*(ptr)))__sync_cmpxchg((ptr),(unsigned long)(o),\ |
6966 | - (unsigned long)(n),sizeof(*(ptr)))) |
6967 | -#endif |
6968 | - |
6969 | -static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old, |
6970 | - unsigned long new, int size) |
6971 | -{ |
6972 | - unsigned long prev; |
6973 | - switch (size) { |
6974 | - case 1: |
6975 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2" |
6976 | - : "=a"(prev) |
6977 | - : "q"(new), "m"(*__xg(ptr)), "0"(old) |
6978 | - : "memory"); |
6979 | - return prev; |
6980 | - case 2: |
6981 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2" |
6982 | - : "=a"(prev) |
6983 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
6984 | - : "memory"); |
6985 | - return prev; |
6986 | - case 4: |
6987 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %1,%2" |
6988 | - : "=a"(prev) |
6989 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
6990 | - : "memory"); |
6991 | - return prev; |
6992 | - } |
6993 | - return old; |
6994 | -} |
6995 | +/* Clear the 'TS' bit */ |
6996 | +#define clts() (xen_clts()) |
6997 | |
6998 | -/* |
6999 | - * Always use locked operations when touching memory shared with a |
7000 | - * hypervisor, since the system may be SMP even if the guest kernel |
7001 | - * isn't. |
7002 | - */ |
7003 | -static inline unsigned long __sync_cmpxchg(volatile void *ptr, |
7004 | - unsigned long old, |
7005 | - unsigned long new, int size) |
7006 | -{ |
7007 | - unsigned long prev; |
7008 | - switch (size) { |
7009 | - case 1: |
7010 | - __asm__ __volatile__("lock; cmpxchgb %b1,%2" |
7011 | - : "=a"(prev) |
7012 | - : "q"(new), "m"(*__xg(ptr)), "0"(old) |
7013 | - : "memory"); |
7014 | - return prev; |
7015 | - case 2: |
7016 | - __asm__ __volatile__("lock; cmpxchgw %w1,%2" |
7017 | - : "=a"(prev) |
7018 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
7019 | - : "memory"); |
7020 | - return prev; |
7021 | - case 4: |
7022 | - __asm__ __volatile__("lock; cmpxchgl %1,%2" |
7023 | - : "=a"(prev) |
7024 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
7025 | - : "memory"); |
7026 | - return prev; |
7027 | - } |
7028 | - return old; |
7029 | -} |
7030 | +/* Set the 'TS' bit */ |
7031 | +#define stts() (HYPERVISOR_fpu_taskswitch(1)) |
7032 | |
7033 | -#ifndef CONFIG_X86_CMPXCHG |
7034 | -/* |
7035 | - * Building a kernel capable running on 80386. It may be necessary to |
7036 | - * simulate the cmpxchg on the 80386 CPU. For that purpose we define |
7037 | - * a function for each of the sizes we support. |
7038 | - */ |
7039 | +#endif /* __KERNEL__ */ |
7040 | |
7041 | -extern unsigned long cmpxchg_386_u8(volatile void *, u8, u8); |
7042 | -extern unsigned long cmpxchg_386_u16(volatile void *, u16, u16); |
7043 | -extern unsigned long cmpxchg_386_u32(volatile void *, u32, u32); |
7044 | - |
7045 | -static inline unsigned long cmpxchg_386(volatile void *ptr, unsigned long old, |
7046 | - unsigned long new, int size) |
7047 | -{ |
7048 | - switch (size) { |
7049 | - case 1: |
7050 | - return cmpxchg_386_u8(ptr, old, new); |
7051 | - case 2: |
7052 | - return cmpxchg_386_u16(ptr, old, new); |
7053 | - case 4: |
7054 | - return cmpxchg_386_u32(ptr, old, new); |
7055 | - } |
7056 | - return old; |
7057 | -} |
7058 | - |
7059 | -#define cmpxchg(ptr,o,n) \ |
7060 | -({ \ |
7061 | - __typeof__(*(ptr)) __ret; \ |
7062 | - if (likely(boot_cpu_data.x86 > 3)) \ |
7063 | - __ret = __cmpxchg((ptr), (unsigned long)(o), \ |
7064 | - (unsigned long)(n), sizeof(*(ptr))); \ |
7065 | - else \ |
7066 | - __ret = cmpxchg_386((ptr), (unsigned long)(o), \ |
7067 | - (unsigned long)(n), sizeof(*(ptr))); \ |
7068 | - __ret; \ |
7069 | -}) |
7070 | -#endif |
7071 | - |
7072 | -#ifdef CONFIG_X86_CMPXCHG64 |
7073 | - |
7074 | -static inline unsigned long long __cmpxchg64(volatile void *ptr, unsigned long long old, |
7075 | - unsigned long long new) |
7076 | -{ |
7077 | - unsigned long long prev; |
7078 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchg8b %3" |
7079 | - : "=A"(prev) |
7080 | - : "b"((unsigned long)new), |
7081 | - "c"((unsigned long)(new >> 32)), |
7082 | - "m"(*__xg(ptr)), |
7083 | - "0"(old) |
7084 | - : "memory"); |
7085 | - return prev; |
7086 | -} |
7087 | - |
7088 | -#define cmpxchg64(ptr,o,n)\ |
7089 | - ((__typeof__(*(ptr)))__cmpxchg64((ptr),(unsigned long long)(o),\ |
7090 | - (unsigned long long)(n))) |
7091 | +static inline unsigned long get_limit(unsigned long segment) |
7092 | +{ |
7093 | + unsigned long __limit; |
7094 | + __asm__("lsll %1,%0" |
7095 | + :"=r" (__limit):"r" (segment)); |
7096 | + return __limit+1; |
7097 | +} |
7098 | + |
7099 | +#define nop() __asm__ __volatile__ ("nop") |
7100 | |
7101 | -#endif |
7102 | - |
7103 | /* |
7104 | * Force strict CPU ordering. |
7105 | * And yes, this is required on UP too when we're talking |
7106 | --- a/include/asm-x86/mach-xen/asm/system_64.h |
7107 | +++ b/include/asm-x86/mach-xen/asm/system_64.h |
7108 | @@ -3,7 +3,7 @@ |
7109 | |
7110 | #include <linux/kernel.h> |
7111 | #include <asm/segment.h> |
7112 | -#include <asm/alternative.h> |
7113 | +#include <asm/cmpxchg.h> |
7114 | |
7115 | #include <asm/synch_bitops.h> |
7116 | #include <asm/hypervisor.h> |
7117 | @@ -43,7 +43,7 @@ |
7118 | [threadrsp] "i" (offsetof(struct task_struct, thread.rsp)), \ |
7119 | [ti_flags] "i" (offsetof(struct thread_info, flags)),\ |
7120 | [tif_fork] "i" (TIF_FORK), \ |
7121 | - [thread_info] "i" (offsetof(struct task_struct, thread_info)), \ |
7122 | + [thread_info] "i" (offsetof(struct task_struct, stack)), \ |
7123 | [pda_pcurrent] "i" (offsetof(struct x8664_pda, pcurrent)) \ |
7124 | : "memory", "cc" __EXTRA_CLOBBER) |
7125 | |
7126 | @@ -92,6 +92,12 @@ |
7127 | machine_to_phys(__dummy); \ |
7128 | }) |
7129 | |
7130 | +static inline void write_cr3(unsigned long val) |
7131 | +{ |
7132 | + val = phys_to_machine(val); |
7133 | + asm volatile("movq %0,%%cr3" :: "r" (val) : "memory"); |
7134 | +} |
7135 | + |
7136 | static inline unsigned long read_cr4(void) |
7137 | { |
7138 | unsigned long cr4; |
7139 | @@ -101,7 +107,7 @@ |
7140 | |
7141 | static inline void write_cr4(unsigned long val) |
7142 | { |
7143 | - asm volatile("movq %0,%%cr4" :: "r" (val)); |
7144 | + asm volatile("movq %0,%%cr4" :: "r" (val) : "memory"); |
7145 | } |
7146 | |
7147 | #define stts() (HYPERVISOR_fpu_taskswitch(1)) |
7148 | @@ -122,100 +128,6 @@ |
7149 | |
7150 | #define nop() __asm__ __volatile__ ("nop") |
7151 | |
7152 | -#define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr)))) |
7153 | - |
7154 | -#define tas(ptr) (xchg((ptr),1)) |
7155 | - |
7156 | -#define __xg(x) ((volatile long *)(x)) |
7157 | - |
7158 | -static inline void set_64bit(volatile unsigned long *ptr, unsigned long val) |
7159 | -{ |
7160 | - *ptr = val; |
7161 | -} |
7162 | - |
7163 | -#define _set_64bit set_64bit |
7164 | - |
7165 | -/* |
7166 | - * Note: no "lock" prefix even on SMP: xchg always implies lock anyway |
7167 | - * Note 2: xchg has side effect, so that attribute volatile is necessary, |
7168 | - * but generally the primitive is invalid, *ptr is output argument. --ANK |
7169 | - */ |
7170 | -static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size) |
7171 | -{ |
7172 | - switch (size) { |
7173 | - case 1: |
7174 | - __asm__ __volatile__("xchgb %b0,%1" |
7175 | - :"=q" (x) |
7176 | - :"m" (*__xg(ptr)), "0" (x) |
7177 | - :"memory"); |
7178 | - break; |
7179 | - case 2: |
7180 | - __asm__ __volatile__("xchgw %w0,%1" |
7181 | - :"=r" (x) |
7182 | - :"m" (*__xg(ptr)), "0" (x) |
7183 | - :"memory"); |
7184 | - break; |
7185 | - case 4: |
7186 | - __asm__ __volatile__("xchgl %k0,%1" |
7187 | - :"=r" (x) |
7188 | - :"m" (*__xg(ptr)), "0" (x) |
7189 | - :"memory"); |
7190 | - break; |
7191 | - case 8: |
7192 | - __asm__ __volatile__("xchgq %0,%1" |
7193 | - :"=r" (x) |
7194 | - :"m" (*__xg(ptr)), "0" (x) |
7195 | - :"memory"); |
7196 | - break; |
7197 | - } |
7198 | - return x; |
7199 | -} |
7200 | - |
7201 | -/* |
7202 | - * Atomic compare and exchange. Compare OLD with MEM, if identical, |
7203 | - * store NEW in MEM. Return the initial value in MEM. Success is |
7204 | - * indicated by comparing RETURN with OLD. |
7205 | - */ |
7206 | - |
7207 | -#define __HAVE_ARCH_CMPXCHG 1 |
7208 | - |
7209 | -static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old, |
7210 | - unsigned long new, int size) |
7211 | -{ |
7212 | - unsigned long prev; |
7213 | - switch (size) { |
7214 | - case 1: |
7215 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2" |
7216 | - : "=a"(prev) |
7217 | - : "q"(new), "m"(*__xg(ptr)), "0"(old) |
7218 | - : "memory"); |
7219 | - return prev; |
7220 | - case 2: |
7221 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2" |
7222 | - : "=a"(prev) |
7223 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
7224 | - : "memory"); |
7225 | - return prev; |
7226 | - case 4: |
7227 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %k1,%2" |
7228 | - : "=a"(prev) |
7229 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
7230 | - : "memory"); |
7231 | - return prev; |
7232 | - case 8: |
7233 | - __asm__ __volatile__(LOCK_PREFIX "cmpxchgq %1,%2" |
7234 | - : "=a"(prev) |
7235 | - : "r"(new), "m"(*__xg(ptr)), "0"(old) |
7236 | - : "memory"); |
7237 | - return prev; |
7238 | - } |
7239 | - return old; |
7240 | -} |
7241 | - |
7242 | -#define cmpxchg(ptr,o,n)\ |
7243 | - ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\ |
7244 | - (unsigned long)(n),sizeof(*(ptr)))) |
7245 | - |
7246 | #ifdef CONFIG_SMP |
7247 | #define smp_mb() mb() |
7248 | #define smp_rmb() rmb() |
7249 | --- a/include/asm-x86/mach-xen/asm/tlbflush_32.h |
7250 | +++ b/include/asm-x86/mach-xen/asm/tlbflush_32.h |
7251 | @@ -29,8 +29,13 @@ |
7252 | * and page-granular flushes are available only on i486 and up. |
7253 | */ |
7254 | |
7255 | +#define TLB_FLUSH_ALL 0xffffffff |
7256 | + |
7257 | + |
7258 | #ifndef CONFIG_SMP |
7259 | |
7260 | +#include <linux/sched.h> |
7261 | + |
7262 | #define flush_tlb() __flush_tlb() |
7263 | #define flush_tlb_all() __flush_tlb_all() |
7264 | #define local_flush_tlb() __flush_tlb() |
7265 | @@ -55,7 +60,7 @@ |
7266 | __flush_tlb(); |
7267 | } |
7268 | |
7269 | -#else |
7270 | +#else /* SMP */ |
7271 | |
7272 | #include <asm/smp.h> |
7273 | |
7274 | @@ -84,9 +89,7 @@ |
7275 | char __cacheline_padding[L1_CACHE_BYTES-8]; |
7276 | }; |
7277 | DECLARE_PER_CPU(struct tlb_state, cpu_tlbstate); |
7278 | - |
7279 | - |
7280 | -#endif |
7281 | +#endif /* SMP */ |
7282 | |
7283 | #define flush_tlb_kernel_range(start, end) flush_tlb_all() |
7284 | |
7285 | --- a/include/asm-x86/mach-xen/asm/tlbflush_64.h |
7286 | +++ b/include/asm-x86/mach-xen/asm/tlbflush_64.h |
7287 | @@ -2,7 +2,9 @@ |
7288 | #define _X8664_TLBFLUSH_H |
7289 | |
7290 | #include <linux/mm.h> |
7291 | +#include <linux/sched.h> |
7292 | #include <asm/processor.h> |
7293 | +#include <asm/system.h> |
7294 | |
7295 | #define __flush_tlb() xen_tlb_flush() |
7296 | |
7297 | --- a/lib/swiotlb-xen.c |
7298 | +++ b/lib/swiotlb-xen.c |
7299 | @@ -729,7 +729,6 @@ |
7300 | return (mask >= ((1UL << dma_bits) - 1)); |
7301 | } |
7302 | |
7303 | -EXPORT_SYMBOL(swiotlb_init); |
7304 | EXPORT_SYMBOL(swiotlb_map_single); |
7305 | EXPORT_SYMBOL(swiotlb_unmap_single); |
7306 | EXPORT_SYMBOL(swiotlb_map_sg); |
7307 | --- a/net/core/dev.c |
7308 | +++ b/net/core/dev.c |
7309 | @@ -1590,12 +1590,17 @@ |
7310 | inline int skb_checksum_setup(struct sk_buff *skb) |
7311 | { |
7312 | if (skb->proto_csum_blank) { |
7313 | + struct iphdr *iph; |
7314 | + unsigned char *th; |
7315 | + |
7316 | if (skb->protocol != htons(ETH_P_IP)) |
7317 | goto out; |
7318 | - skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl; |
7319 | - if (skb->h.raw >= skb->tail) |
7320 | + iph = ip_hdr(skb); |
7321 | + th = skb_network_header(skb) + 4 * iph->ihl; |
7322 | + if (th >= skb_tail_pointer(skb)) |
7323 | goto out; |
7324 | - switch (skb->nh.iph->protocol) { |
7325 | + skb->csum_start = th - skb->head; |
7326 | + switch (iph->protocol) { |
7327 | case IPPROTO_TCP: |
7328 | skb->csum_offset = offsetof(struct tcphdr, check); |
7329 | break; |
7330 | @@ -1606,10 +1611,10 @@ |
7331 | if (net_ratelimit()) |
7332 | printk(KERN_ERR "Attempting to checksum a non-" |
7333 | "TCP/UDP packet, dropping a protocol" |
7334 | - " %d packet", skb->nh.iph->protocol); |
7335 | + " %d packet", iph->protocol); |
7336 | goto out; |
7337 | } |
7338 | - if ((skb->h.raw + skb->csum_offset + 2) > skb->tail) |
7339 | + if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) |
7340 | goto out; |
7341 | skb->ip_summed = CHECKSUM_PARTIAL; |
7342 | skb->proto_csum_blank = 0; |
7343 | --- a/scripts/Makefile.xen.awk |
7344 | +++ b/scripts/Makefile.xen.awk |
7345 | @@ -13,7 +13,7 @@ |
7346 | next |
7347 | } |
7348 | |
7349 | -/:[[:space:]]*%\.[cS][[:space:]]/ { |
7350 | +/:[[:space:]]*\$\(src\)\/%\.[cS][[:space:]]/ { |
7351 | line = gensub(/%.([cS])/, "%-xen.\\1", "g", $0) |
7352 | line = gensub(/(single-used-m)/, "xen-\\1", "g", line) |
7353 | print line |