Contents of /trunk/kernel-alx-legacy/patches-4.9/0151-4.9.52-all-fixes.patch
Parent Directory | Revision Log
Revision 3608 -
(show annotations)
(download)
Fri Aug 14 07:34:29 2020 UTC (4 years, 1 month ago) by niro
File size: 123220 byte(s)
Fri Aug 14 07:34:29 2020 UTC (4 years, 1 month ago) by niro
File size: 123220 byte(s)
-added kerenl-alx-legacy pkg
1 | diff --git a/Makefile b/Makefile |
2 | index b48aebbe187f..c53de1e38c6a 100644 |
3 | --- a/Makefile |
4 | +++ b/Makefile |
5 | @@ -1,6 +1,6 @@ |
6 | VERSION = 4 |
7 | PATCHLEVEL = 9 |
8 | -SUBLEVEL = 51 |
9 | +SUBLEVEL = 52 |
10 | EXTRAVERSION = |
11 | NAME = Roaring Lionus |
12 | |
13 | diff --git a/arch/arc/kernel/entry.S b/arch/arc/kernel/entry.S |
14 | index 1eea99beecc3..85d9ea4a0acc 100644 |
15 | --- a/arch/arc/kernel/entry.S |
16 | +++ b/arch/arc/kernel/entry.S |
17 | @@ -92,6 +92,12 @@ ENTRY(EV_MachineCheck) |
18 | lr r0, [efa] |
19 | mov r1, sp |
20 | |
21 | + ; hardware auto-disables MMU, re-enable it to allow kernel vaddr |
22 | + ; access for say stack unwinding of modules for crash dumps |
23 | + lr r3, [ARC_REG_PID] |
24 | + or r3, r3, MMU_ENABLE |
25 | + sr r3, [ARC_REG_PID] |
26 | + |
27 | lsr r3, r2, 8 |
28 | bmsk r3, r3, 7 |
29 | brne r3, ECR_C_MCHK_DUP_TLB, 1f |
30 | diff --git a/arch/arc/mm/tlb.c b/arch/arc/mm/tlb.c |
31 | index bdb295e09160..a4dc881da277 100644 |
32 | --- a/arch/arc/mm/tlb.c |
33 | +++ b/arch/arc/mm/tlb.c |
34 | @@ -896,9 +896,6 @@ void do_tlb_overlap_fault(unsigned long cause, unsigned long address, |
35 | |
36 | local_irq_save(flags); |
37 | |
38 | - /* re-enable the MMU */ |
39 | - write_aux_reg(ARC_REG_PID, MMU_ENABLE | read_aux_reg(ARC_REG_PID)); |
40 | - |
41 | /* loop thru all sets of TLB */ |
42 | for (set = 0; set < mmu->sets; set++) { |
43 | |
44 | diff --git a/arch/mips/math-emu/dp_fmax.c b/arch/mips/math-emu/dp_fmax.c |
45 | index fd71b8daaaf2..5bec64f2884e 100644 |
46 | --- a/arch/mips/math-emu/dp_fmax.c |
47 | +++ b/arch/mips/math-emu/dp_fmax.c |
48 | @@ -47,14 +47,26 @@ union ieee754dp ieee754dp_fmax(union ieee754dp x, union ieee754dp y) |
49 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
50 | return ieee754dp_nanxcpt(x); |
51 | |
52 | - /* numbers are preferred to NaNs */ |
53 | + /* |
54 | + * Quiet NaN handling |
55 | + */ |
56 | + |
57 | + /* |
58 | + * The case of both inputs quiet NaNs |
59 | + */ |
60 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
61 | + return x; |
62 | + |
63 | + /* |
64 | + * The cases of exactly one input quiet NaN (numbers |
65 | + * are here preferred as returned values to NaNs) |
66 | + */ |
67 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
68 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
69 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
70 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
71 | return x; |
72 | |
73 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
74 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
75 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
76 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
77 | @@ -80,9 +92,7 @@ union ieee754dp ieee754dp_fmax(union ieee754dp x, union ieee754dp y) |
78 | return ys ? x : y; |
79 | |
80 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
81 | - if (xs == ys) |
82 | - return x; |
83 | - return ieee754dp_zero(1); |
84 | + return ieee754dp_zero(xs & ys); |
85 | |
86 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
87 | DPDNORMX; |
88 | @@ -106,16 +116,32 @@ union ieee754dp ieee754dp_fmax(union ieee754dp x, union ieee754dp y) |
89 | else if (xs < ys) |
90 | return x; |
91 | |
92 | - /* Compare exponent */ |
93 | - if (xe > ye) |
94 | - return x; |
95 | - else if (xe < ye) |
96 | - return y; |
97 | + /* Signs of inputs are equal, let's compare exponents */ |
98 | + if (xs == 0) { |
99 | + /* Inputs are both positive */ |
100 | + if (xe > ye) |
101 | + return x; |
102 | + else if (xe < ye) |
103 | + return y; |
104 | + } else { |
105 | + /* Inputs are both negative */ |
106 | + if (xe > ye) |
107 | + return y; |
108 | + else if (xe < ye) |
109 | + return x; |
110 | + } |
111 | |
112 | - /* Compare mantissa */ |
113 | + /* Signs and exponents of inputs are equal, let's compare mantissas */ |
114 | + if (xs == 0) { |
115 | + /* Inputs are both positive, with equal signs and exponents */ |
116 | + if (xm <= ym) |
117 | + return y; |
118 | + return x; |
119 | + } |
120 | + /* Inputs are both negative, with equal signs and exponents */ |
121 | if (xm <= ym) |
122 | - return y; |
123 | - return x; |
124 | + return x; |
125 | + return y; |
126 | } |
127 | |
128 | union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
129 | @@ -147,14 +173,26 @@ union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
130 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
131 | return ieee754dp_nanxcpt(x); |
132 | |
133 | - /* numbers are preferred to NaNs */ |
134 | + /* |
135 | + * Quiet NaN handling |
136 | + */ |
137 | + |
138 | + /* |
139 | + * The case of both inputs quiet NaNs |
140 | + */ |
141 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
142 | + return x; |
143 | + |
144 | + /* |
145 | + * The cases of exactly one input quiet NaN (numbers |
146 | + * are here preferred as returned values to NaNs) |
147 | + */ |
148 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
149 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
150 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
151 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
152 | return x; |
153 | |
154 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
155 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
156 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
157 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
158 | @@ -164,6 +202,9 @@ union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
159 | /* |
160 | * Infinity and zero handling |
161 | */ |
162 | + case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
163 | + return ieee754dp_inf(xs & ys); |
164 | + |
165 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
166 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
167 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
168 | @@ -171,7 +212,6 @@ union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
169 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
170 | return x; |
171 | |
172 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
173 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF): |
174 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF): |
175 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
176 | @@ -180,9 +220,7 @@ union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
177 | return y; |
178 | |
179 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
180 | - if (xs == ys) |
181 | - return x; |
182 | - return ieee754dp_zero(1); |
183 | + return ieee754dp_zero(xs & ys); |
184 | |
185 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
186 | DPDNORMX; |
187 | @@ -207,7 +245,11 @@ union ieee754dp ieee754dp_fmaxa(union ieee754dp x, union ieee754dp y) |
188 | return y; |
189 | |
190 | /* Compare mantissa */ |
191 | - if (xm <= ym) |
192 | + if (xm < ym) |
193 | return y; |
194 | - return x; |
195 | + else if (xm > ym) |
196 | + return x; |
197 | + else if (xs == 0) |
198 | + return x; |
199 | + return y; |
200 | } |
201 | diff --git a/arch/mips/math-emu/dp_fmin.c b/arch/mips/math-emu/dp_fmin.c |
202 | index c1072b0dfb95..a287b23818d8 100644 |
203 | --- a/arch/mips/math-emu/dp_fmin.c |
204 | +++ b/arch/mips/math-emu/dp_fmin.c |
205 | @@ -47,14 +47,26 @@ union ieee754dp ieee754dp_fmin(union ieee754dp x, union ieee754dp y) |
206 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
207 | return ieee754dp_nanxcpt(x); |
208 | |
209 | - /* numbers are preferred to NaNs */ |
210 | + /* |
211 | + * Quiet NaN handling |
212 | + */ |
213 | + |
214 | + /* |
215 | + * The case of both inputs quiet NaNs |
216 | + */ |
217 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
218 | + return x; |
219 | + |
220 | + /* |
221 | + * The cases of exactly one input quiet NaN (numbers |
222 | + * are here preferred as returned values to NaNs) |
223 | + */ |
224 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
225 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
226 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
227 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
228 | return x; |
229 | |
230 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
231 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
232 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
233 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
234 | @@ -80,9 +92,7 @@ union ieee754dp ieee754dp_fmin(union ieee754dp x, union ieee754dp y) |
235 | return ys ? y : x; |
236 | |
237 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
238 | - if (xs == ys) |
239 | - return x; |
240 | - return ieee754dp_zero(1); |
241 | + return ieee754dp_zero(xs | ys); |
242 | |
243 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
244 | DPDNORMX; |
245 | @@ -106,16 +116,32 @@ union ieee754dp ieee754dp_fmin(union ieee754dp x, union ieee754dp y) |
246 | else if (xs < ys) |
247 | return y; |
248 | |
249 | - /* Compare exponent */ |
250 | - if (xe > ye) |
251 | - return y; |
252 | - else if (xe < ye) |
253 | - return x; |
254 | + /* Signs of inputs are the same, let's compare exponents */ |
255 | + if (xs == 0) { |
256 | + /* Inputs are both positive */ |
257 | + if (xe > ye) |
258 | + return y; |
259 | + else if (xe < ye) |
260 | + return x; |
261 | + } else { |
262 | + /* Inputs are both negative */ |
263 | + if (xe > ye) |
264 | + return x; |
265 | + else if (xe < ye) |
266 | + return y; |
267 | + } |
268 | |
269 | - /* Compare mantissa */ |
270 | + /* Signs and exponents of inputs are equal, let's compare mantissas */ |
271 | + if (xs == 0) { |
272 | + /* Inputs are both positive, with equal signs and exponents */ |
273 | + if (xm <= ym) |
274 | + return x; |
275 | + return y; |
276 | + } |
277 | + /* Inputs are both negative, with equal signs and exponents */ |
278 | if (xm <= ym) |
279 | - return x; |
280 | - return y; |
281 | + return y; |
282 | + return x; |
283 | } |
284 | |
285 | union ieee754dp ieee754dp_fmina(union ieee754dp x, union ieee754dp y) |
286 | @@ -147,14 +173,26 @@ union ieee754dp ieee754dp_fmina(union ieee754dp x, union ieee754dp y) |
287 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
288 | return ieee754dp_nanxcpt(x); |
289 | |
290 | - /* numbers are preferred to NaNs */ |
291 | + /* |
292 | + * Quiet NaN handling |
293 | + */ |
294 | + |
295 | + /* |
296 | + * The case of both inputs quiet NaNs |
297 | + */ |
298 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
299 | + return x; |
300 | + |
301 | + /* |
302 | + * The cases of exactly one input quiet NaN (numbers |
303 | + * are here preferred as returned values to NaNs) |
304 | + */ |
305 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
306 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
307 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
308 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
309 | return x; |
310 | |
311 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
312 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
313 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
314 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
315 | @@ -164,25 +202,25 @@ union ieee754dp ieee754dp_fmina(union ieee754dp x, union ieee754dp y) |
316 | /* |
317 | * Infinity and zero handling |
318 | */ |
319 | + case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
320 | + return ieee754dp_inf(xs | ys); |
321 | + |
322 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
323 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
324 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
325 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO): |
326 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
327 | - return x; |
328 | + return y; |
329 | |
330 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
331 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF): |
332 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF): |
333 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
334 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM): |
335 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM): |
336 | - return y; |
337 | + return x; |
338 | |
339 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
340 | - if (xs == ys) |
341 | - return x; |
342 | - return ieee754dp_zero(1); |
343 | + return ieee754dp_zero(xs | ys); |
344 | |
345 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
346 | DPDNORMX; |
347 | @@ -207,7 +245,11 @@ union ieee754dp ieee754dp_fmina(union ieee754dp x, union ieee754dp y) |
348 | return x; |
349 | |
350 | /* Compare mantissa */ |
351 | - if (xm <= ym) |
352 | + if (xm < ym) |
353 | + return x; |
354 | + else if (xm > ym) |
355 | + return y; |
356 | + else if (xs == 1) |
357 | return x; |
358 | return y; |
359 | } |
360 | diff --git a/arch/mips/math-emu/dp_maddf.c b/arch/mips/math-emu/dp_maddf.c |
361 | index 4a2d03c72959..e0d9be5fbf4c 100644 |
362 | --- a/arch/mips/math-emu/dp_maddf.c |
363 | +++ b/arch/mips/math-emu/dp_maddf.c |
364 | @@ -14,22 +14,45 @@ |
365 | |
366 | #include "ieee754dp.h" |
367 | |
368 | -enum maddf_flags { |
369 | - maddf_negate_product = 1 << 0, |
370 | -}; |
371 | + |
372 | +/* 128 bits shift right logical with rounding. */ |
373 | +void srl128(u64 *hptr, u64 *lptr, int count) |
374 | +{ |
375 | + u64 low; |
376 | + |
377 | + if (count >= 128) { |
378 | + *lptr = *hptr != 0 || *lptr != 0; |
379 | + *hptr = 0; |
380 | + } else if (count >= 64) { |
381 | + if (count == 64) { |
382 | + *lptr = *hptr | (*lptr != 0); |
383 | + } else { |
384 | + low = *lptr; |
385 | + *lptr = *hptr >> (count - 64); |
386 | + *lptr |= (*hptr << (128 - count)) != 0 || low != 0; |
387 | + } |
388 | + *hptr = 0; |
389 | + } else { |
390 | + low = *lptr; |
391 | + *lptr = low >> count | *hptr << (64 - count); |
392 | + *lptr |= (low << (64 - count)) != 0; |
393 | + *hptr = *hptr >> count; |
394 | + } |
395 | +} |
396 | |
397 | static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
398 | union ieee754dp y, enum maddf_flags flags) |
399 | { |
400 | int re; |
401 | int rs; |
402 | - u64 rm; |
403 | unsigned lxm; |
404 | unsigned hxm; |
405 | unsigned lym; |
406 | unsigned hym; |
407 | u64 lrm; |
408 | u64 hrm; |
409 | + u64 lzm; |
410 | + u64 hzm; |
411 | u64 t; |
412 | u64 at; |
413 | int s; |
414 | @@ -48,52 +71,34 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
415 | |
416 | ieee754_clearcx(); |
417 | |
418 | - switch (zc) { |
419 | - case IEEE754_CLASS_SNAN: |
420 | - ieee754_setcx(IEEE754_INVALID_OPERATION); |
421 | + /* |
422 | + * Handle the cases when at least one of x, y or z is a NaN. |
423 | + * Order of precedence is sNaN, qNaN and z, x, y. |
424 | + */ |
425 | + if (zc == IEEE754_CLASS_SNAN) |
426 | return ieee754dp_nanxcpt(z); |
427 | - case IEEE754_CLASS_DNORM: |
428 | - DPDNORMZ; |
429 | - /* QNAN is handled separately below */ |
430 | - } |
431 | - |
432 | - switch (CLPAIR(xc, yc)) { |
433 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN): |
434 | - case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN): |
435 | - case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN): |
436 | - case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN): |
437 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN): |
438 | - return ieee754dp_nanxcpt(y); |
439 | - |
440 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN): |
441 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN): |
442 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO): |
443 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM): |
444 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM): |
445 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
446 | + if (xc == IEEE754_CLASS_SNAN) |
447 | return ieee754dp_nanxcpt(x); |
448 | - |
449 | - case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
450 | - case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
451 | - case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
452 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
453 | + if (yc == IEEE754_CLASS_SNAN) |
454 | + return ieee754dp_nanxcpt(y); |
455 | + if (zc == IEEE754_CLASS_QNAN) |
456 | + return z; |
457 | + if (xc == IEEE754_CLASS_QNAN) |
458 | + return x; |
459 | + if (yc == IEEE754_CLASS_QNAN) |
460 | return y; |
461 | |
462 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
463 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
464 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
465 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
466 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF): |
467 | - return x; |
468 | + if (zc == IEEE754_CLASS_DNORM) |
469 | + DPDNORMZ; |
470 | + /* ZERO z cases are handled separately below */ |
471 | |
472 | + switch (CLPAIR(xc, yc)) { |
473 | |
474 | /* |
475 | * Infinity handling |
476 | */ |
477 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
478 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
479 | - if (zc == IEEE754_CLASS_QNAN) |
480 | - return z; |
481 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
482 | return ieee754dp_indef(); |
483 | |
484 | @@ -102,9 +107,27 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
485 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
486 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
487 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
488 | - if (zc == IEEE754_CLASS_QNAN) |
489 | - return z; |
490 | - return ieee754dp_inf(xs ^ ys); |
491 | + if ((zc == IEEE754_CLASS_INF) && |
492 | + ((!(flags & MADDF_NEGATE_PRODUCT) && (zs != (xs ^ ys))) || |
493 | + ((flags & MADDF_NEGATE_PRODUCT) && (zs == (xs ^ ys))))) { |
494 | + /* |
495 | + * Cases of addition of infinities with opposite signs |
496 | + * or subtraction of infinities with same signs. |
497 | + */ |
498 | + ieee754_setcx(IEEE754_INVALID_OPERATION); |
499 | + return ieee754dp_indef(); |
500 | + } |
501 | + /* |
502 | + * z is here either not an infinity, or an infinity having the |
503 | + * same sign as product (x*y) (in case of MADDF.D instruction) |
504 | + * or product -(x*y) (in MSUBF.D case). The result must be an |
505 | + * infinity, and its sign is determined only by the value of |
506 | + * (flags & MADDF_NEGATE_PRODUCT) and the signs of x and y. |
507 | + */ |
508 | + if (flags & MADDF_NEGATE_PRODUCT) |
509 | + return ieee754dp_inf(1 ^ (xs ^ ys)); |
510 | + else |
511 | + return ieee754dp_inf(xs ^ ys); |
512 | |
513 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
514 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM): |
515 | @@ -113,32 +136,42 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
516 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
517 | if (zc == IEEE754_CLASS_INF) |
518 | return ieee754dp_inf(zs); |
519 | - /* Multiplication is 0 so just return z */ |
520 | + if (zc == IEEE754_CLASS_ZERO) { |
521 | + /* Handle cases +0 + (-0) and similar ones. */ |
522 | + if ((!(flags & MADDF_NEGATE_PRODUCT) |
523 | + && (zs == (xs ^ ys))) || |
524 | + ((flags & MADDF_NEGATE_PRODUCT) |
525 | + && (zs != (xs ^ ys)))) |
526 | + /* |
527 | + * Cases of addition of zeros of equal signs |
528 | + * or subtraction of zeroes of opposite signs. |
529 | + * The sign of the resulting zero is in any |
530 | + * such case determined only by the sign of z. |
531 | + */ |
532 | + return z; |
533 | + |
534 | + return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD); |
535 | + } |
536 | + /* x*y is here 0, and z is not 0, so just return z */ |
537 | return z; |
538 | |
539 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
540 | DPDNORMX; |
541 | |
542 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM): |
543 | - if (zc == IEEE754_CLASS_QNAN) |
544 | - return z; |
545 | - else if (zc == IEEE754_CLASS_INF) |
546 | + if (zc == IEEE754_CLASS_INF) |
547 | return ieee754dp_inf(zs); |
548 | DPDNORMY; |
549 | break; |
550 | |
551 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM): |
552 | - if (zc == IEEE754_CLASS_QNAN) |
553 | - return z; |
554 | - else if (zc == IEEE754_CLASS_INF) |
555 | + if (zc == IEEE754_CLASS_INF) |
556 | return ieee754dp_inf(zs); |
557 | DPDNORMX; |
558 | break; |
559 | |
560 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM): |
561 | - if (zc == IEEE754_CLASS_QNAN) |
562 | - return z; |
563 | - else if (zc == IEEE754_CLASS_INF) |
564 | + if (zc == IEEE754_CLASS_INF) |
565 | return ieee754dp_inf(zs); |
566 | /* fall through to real computations */ |
567 | } |
568 | @@ -157,7 +190,7 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
569 | |
570 | re = xe + ye; |
571 | rs = xs ^ ys; |
572 | - if (flags & maddf_negate_product) |
573 | + if (flags & MADDF_NEGATE_PRODUCT) |
574 | rs ^= 1; |
575 | |
576 | /* shunt to top of word */ |
577 | @@ -165,7 +198,7 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
578 | ym <<= 64 - (DP_FBITS + 1); |
579 | |
580 | /* |
581 | - * Multiply 64 bits xm, ym to give high 64 bits rm with stickness. |
582 | + * Multiply 64 bits xm and ym to give 128 bits result in hrm:lrm. |
583 | */ |
584 | |
585 | /* 32 * 32 => 64 */ |
586 | @@ -195,78 +228,110 @@ static union ieee754dp _dp_maddf(union ieee754dp z, union ieee754dp x, |
587 | |
588 | hrm = hrm + (t >> 32); |
589 | |
590 | - rm = hrm | (lrm != 0); |
591 | - |
592 | - /* |
593 | - * Sticky shift down to normal rounding precision. |
594 | - */ |
595 | - if ((s64) rm < 0) { |
596 | - rm = (rm >> (64 - (DP_FBITS + 1 + 3))) | |
597 | - ((rm << (DP_FBITS + 1 + 3)) != 0); |
598 | + /* Put explicit bit at bit 126 if necessary */ |
599 | + if ((int64_t)hrm < 0) { |
600 | + lrm = (hrm << 63) | (lrm >> 1); |
601 | + hrm = hrm >> 1; |
602 | re++; |
603 | - } else { |
604 | - rm = (rm >> (64 - (DP_FBITS + 1 + 3 + 1))) | |
605 | - ((rm << (DP_FBITS + 1 + 3 + 1)) != 0); |
606 | } |
607 | - assert(rm & (DP_HIDDEN_BIT << 3)); |
608 | |
609 | - /* And now the addition */ |
610 | - assert(zm & DP_HIDDEN_BIT); |
611 | + assert(hrm & (1 << 62)); |
612 | |
613 | - /* |
614 | - * Provide guard,round and stick bit space. |
615 | - */ |
616 | - zm <<= 3; |
617 | + if (zc == IEEE754_CLASS_ZERO) { |
618 | + /* |
619 | + * Move explicit bit from bit 126 to bit 55 since the |
620 | + * ieee754dp_format code expects the mantissa to be |
621 | + * 56 bits wide (53 + 3 rounding bits). |
622 | + */ |
623 | + srl128(&hrm, &lrm, (126 - 55)); |
624 | + return ieee754dp_format(rs, re, lrm); |
625 | + } |
626 | + |
627 | + /* Move explicit bit from bit 52 to bit 126 */ |
628 | + lzm = 0; |
629 | + hzm = zm << 10; |
630 | + assert(hzm & (1 << 62)); |
631 | |
632 | + /* Make the exponents the same */ |
633 | if (ze > re) { |
634 | /* |
635 | * Have to shift y fraction right to align. |
636 | */ |
637 | s = ze - re; |
638 | - rm = XDPSRS(rm, s); |
639 | + srl128(&hrm, &lrm, s); |
640 | re += s; |
641 | } else if (re > ze) { |
642 | /* |
643 | * Have to shift x fraction right to align. |
644 | */ |
645 | s = re - ze; |
646 | - zm = XDPSRS(zm, s); |
647 | + srl128(&hzm, &lzm, s); |
648 | ze += s; |
649 | } |
650 | assert(ze == re); |
651 | assert(ze <= DP_EMAX); |
652 | |
653 | + /* Do the addition */ |
654 | if (zs == rs) { |
655 | /* |
656 | - * Generate 28 bit result of adding two 27 bit numbers |
657 | - * leaving result in xm, xs and xe. |
658 | + * Generate 128 bit result by adding two 127 bit numbers |
659 | + * leaving result in hzm:lzm, zs and ze. |
660 | */ |
661 | - zm = zm + rm; |
662 | - |
663 | - if (zm >> (DP_FBITS + 1 + 3)) { /* carry out */ |
664 | - zm = XDPSRS1(zm); |
665 | + hzm = hzm + hrm + (lzm > (lzm + lrm)); |
666 | + lzm = lzm + lrm; |
667 | + if ((int64_t)hzm < 0) { /* carry out */ |
668 | + srl128(&hzm, &lzm, 1); |
669 | ze++; |
670 | } |
671 | } else { |
672 | - if (zm >= rm) { |
673 | - zm = zm - rm; |
674 | + if (hzm > hrm || (hzm == hrm && lzm >= lrm)) { |
675 | + hzm = hzm - hrm - (lzm < lrm); |
676 | + lzm = lzm - lrm; |
677 | } else { |
678 | - zm = rm - zm; |
679 | + hzm = hrm - hzm - (lrm < lzm); |
680 | + lzm = lrm - lzm; |
681 | zs = rs; |
682 | } |
683 | - if (zm == 0) |
684 | + if (lzm == 0 && hzm == 0) |
685 | return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD); |
686 | |
687 | /* |
688 | - * Normalize to rounding precision. |
689 | + * Put explicit bit at bit 126 if necessary. |
690 | */ |
691 | - while ((zm >> (DP_FBITS + 3)) == 0) { |
692 | - zm <<= 1; |
693 | - ze--; |
694 | + if (hzm == 0) { |
695 | + /* left shift by 63 or 64 bits */ |
696 | + if ((int64_t)lzm < 0) { |
697 | + /* MSB of lzm is the explicit bit */ |
698 | + hzm = lzm >> 1; |
699 | + lzm = lzm << 63; |
700 | + ze -= 63; |
701 | + } else { |
702 | + hzm = lzm; |
703 | + lzm = 0; |
704 | + ze -= 64; |
705 | + } |
706 | + } |
707 | + |
708 | + t = 0; |
709 | + while ((hzm >> (62 - t)) == 0) |
710 | + t++; |
711 | + |
712 | + assert(t <= 62); |
713 | + if (t) { |
714 | + hzm = hzm << t | lzm >> (64 - t); |
715 | + lzm = lzm << t; |
716 | + ze -= t; |
717 | } |
718 | } |
719 | |
720 | - return ieee754dp_format(zs, ze, zm); |
721 | + /* |
722 | + * Move explicit bit from bit 126 to bit 55 since the |
723 | + * ieee754dp_format code expects the mantissa to be |
724 | + * 56 bits wide (53 + 3 rounding bits). |
725 | + */ |
726 | + srl128(&hzm, &lzm, (126 - 55)); |
727 | + |
728 | + return ieee754dp_format(zs, ze, lzm); |
729 | } |
730 | |
731 | union ieee754dp ieee754dp_maddf(union ieee754dp z, union ieee754dp x, |
732 | @@ -278,5 +343,5 @@ union ieee754dp ieee754dp_maddf(union ieee754dp z, union ieee754dp x, |
733 | union ieee754dp ieee754dp_msubf(union ieee754dp z, union ieee754dp x, |
734 | union ieee754dp y) |
735 | { |
736 | - return _dp_maddf(z, x, y, maddf_negate_product); |
737 | + return _dp_maddf(z, x, y, MADDF_NEGATE_PRODUCT); |
738 | } |
739 | diff --git a/arch/mips/math-emu/ieee754int.h b/arch/mips/math-emu/ieee754int.h |
740 | index 8bc2f6963324..dd2071f430e0 100644 |
741 | --- a/arch/mips/math-emu/ieee754int.h |
742 | +++ b/arch/mips/math-emu/ieee754int.h |
743 | @@ -26,6 +26,10 @@ |
744 | |
745 | #define CLPAIR(x, y) ((x)*6+(y)) |
746 | |
747 | +enum maddf_flags { |
748 | + MADDF_NEGATE_PRODUCT = 1 << 0, |
749 | +}; |
750 | + |
751 | static inline void ieee754_clearcx(void) |
752 | { |
753 | ieee754_csr.cx = 0; |
754 | diff --git a/arch/mips/math-emu/ieee754sp.h b/arch/mips/math-emu/ieee754sp.h |
755 | index 8476067075fe..0f63e4202cff 100644 |
756 | --- a/arch/mips/math-emu/ieee754sp.h |
757 | +++ b/arch/mips/math-emu/ieee754sp.h |
758 | @@ -45,6 +45,10 @@ static inline int ieee754sp_finite(union ieee754sp x) |
759 | return SPBEXP(x) != SP_EMAX + 1 + SP_EBIAS; |
760 | } |
761 | |
762 | +/* 64 bit right shift with rounding */ |
763 | +#define XSPSRS64(v, rs) \ |
764 | + (((rs) >= 64) ? ((v) != 0) : ((v) >> (rs)) | ((v) << (64-(rs)) != 0)) |
765 | + |
766 | /* 3bit extended single precision sticky right shift */ |
767 | #define XSPSRS(v, rs) \ |
768 | ((rs > (SP_FBITS+3))?1:((v) >> (rs)) | ((v) << (32-(rs)) != 0)) |
769 | diff --git a/arch/mips/math-emu/sp_fmax.c b/arch/mips/math-emu/sp_fmax.c |
770 | index 4d000844e48e..74a5a00d2f22 100644 |
771 | --- a/arch/mips/math-emu/sp_fmax.c |
772 | +++ b/arch/mips/math-emu/sp_fmax.c |
773 | @@ -47,14 +47,26 @@ union ieee754sp ieee754sp_fmax(union ieee754sp x, union ieee754sp y) |
774 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
775 | return ieee754sp_nanxcpt(x); |
776 | |
777 | - /* numbers are preferred to NaNs */ |
778 | + /* |
779 | + * Quiet NaN handling |
780 | + */ |
781 | + |
782 | + /* |
783 | + * The case of both inputs quiet NaNs |
784 | + */ |
785 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
786 | + return x; |
787 | + |
788 | + /* |
789 | + * The cases of exactly one input quiet NaN (numbers |
790 | + * are here preferred as returned values to NaNs) |
791 | + */ |
792 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
793 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
794 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
795 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
796 | return x; |
797 | |
798 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
799 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
800 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
801 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
802 | @@ -80,9 +92,7 @@ union ieee754sp ieee754sp_fmax(union ieee754sp x, union ieee754sp y) |
803 | return ys ? x : y; |
804 | |
805 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
806 | - if (xs == ys) |
807 | - return x; |
808 | - return ieee754sp_zero(1); |
809 | + return ieee754sp_zero(xs & ys); |
810 | |
811 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
812 | SPDNORMX; |
813 | @@ -106,16 +116,32 @@ union ieee754sp ieee754sp_fmax(union ieee754sp x, union ieee754sp y) |
814 | else if (xs < ys) |
815 | return x; |
816 | |
817 | - /* Compare exponent */ |
818 | - if (xe > ye) |
819 | - return x; |
820 | - else if (xe < ye) |
821 | - return y; |
822 | + /* Signs of inputs are equal, let's compare exponents */ |
823 | + if (xs == 0) { |
824 | + /* Inputs are both positive */ |
825 | + if (xe > ye) |
826 | + return x; |
827 | + else if (xe < ye) |
828 | + return y; |
829 | + } else { |
830 | + /* Inputs are both negative */ |
831 | + if (xe > ye) |
832 | + return y; |
833 | + else if (xe < ye) |
834 | + return x; |
835 | + } |
836 | |
837 | - /* Compare mantissa */ |
838 | + /* Signs and exponents of inputs are equal, let's compare mantissas */ |
839 | + if (xs == 0) { |
840 | + /* Inputs are both positive, with equal signs and exponents */ |
841 | + if (xm <= ym) |
842 | + return y; |
843 | + return x; |
844 | + } |
845 | + /* Inputs are both negative, with equal signs and exponents */ |
846 | if (xm <= ym) |
847 | - return y; |
848 | - return x; |
849 | + return x; |
850 | + return y; |
851 | } |
852 | |
853 | union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
854 | @@ -147,14 +173,26 @@ union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
855 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
856 | return ieee754sp_nanxcpt(x); |
857 | |
858 | - /* numbers are preferred to NaNs */ |
859 | + /* |
860 | + * Quiet NaN handling |
861 | + */ |
862 | + |
863 | + /* |
864 | + * The case of both inputs quiet NaNs |
865 | + */ |
866 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
867 | + return x; |
868 | + |
869 | + /* |
870 | + * The cases of exactly one input quiet NaN (numbers |
871 | + * are here preferred as returned values to NaNs) |
872 | + */ |
873 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
874 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
875 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
876 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
877 | return x; |
878 | |
879 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
880 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
881 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
882 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
883 | @@ -164,6 +202,9 @@ union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
884 | /* |
885 | * Infinity and zero handling |
886 | */ |
887 | + case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
888 | + return ieee754sp_inf(xs & ys); |
889 | + |
890 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
891 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
892 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
893 | @@ -171,7 +212,6 @@ union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
894 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
895 | return x; |
896 | |
897 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
898 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF): |
899 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF): |
900 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
901 | @@ -180,9 +220,7 @@ union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
902 | return y; |
903 | |
904 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
905 | - if (xs == ys) |
906 | - return x; |
907 | - return ieee754sp_zero(1); |
908 | + return ieee754sp_zero(xs & ys); |
909 | |
910 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
911 | SPDNORMX; |
912 | @@ -207,7 +245,11 @@ union ieee754sp ieee754sp_fmaxa(union ieee754sp x, union ieee754sp y) |
913 | return y; |
914 | |
915 | /* Compare mantissa */ |
916 | - if (xm <= ym) |
917 | + if (xm < ym) |
918 | return y; |
919 | - return x; |
920 | + else if (xm > ym) |
921 | + return x; |
922 | + else if (xs == 0) |
923 | + return x; |
924 | + return y; |
925 | } |
926 | diff --git a/arch/mips/math-emu/sp_fmin.c b/arch/mips/math-emu/sp_fmin.c |
927 | index 4eb1bb9e9dec..c51385f46b09 100644 |
928 | --- a/arch/mips/math-emu/sp_fmin.c |
929 | +++ b/arch/mips/math-emu/sp_fmin.c |
930 | @@ -47,14 +47,26 @@ union ieee754sp ieee754sp_fmin(union ieee754sp x, union ieee754sp y) |
931 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
932 | return ieee754sp_nanxcpt(x); |
933 | |
934 | - /* numbers are preferred to NaNs */ |
935 | + /* |
936 | + * Quiet NaN handling |
937 | + */ |
938 | + |
939 | + /* |
940 | + * The case of both inputs quiet NaNs |
941 | + */ |
942 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
943 | + return x; |
944 | + |
945 | + /* |
946 | + * The cases of exactly one input quiet NaN (numbers |
947 | + * are here preferred as returned values to NaNs) |
948 | + */ |
949 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
950 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
951 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
952 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
953 | return x; |
954 | |
955 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
956 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
957 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
958 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
959 | @@ -80,9 +92,7 @@ union ieee754sp ieee754sp_fmin(union ieee754sp x, union ieee754sp y) |
960 | return ys ? y : x; |
961 | |
962 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
963 | - if (xs == ys) |
964 | - return x; |
965 | - return ieee754sp_zero(1); |
966 | + return ieee754sp_zero(xs | ys); |
967 | |
968 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
969 | SPDNORMX; |
970 | @@ -106,16 +116,32 @@ union ieee754sp ieee754sp_fmin(union ieee754sp x, union ieee754sp y) |
971 | else if (xs < ys) |
972 | return y; |
973 | |
974 | - /* Compare exponent */ |
975 | - if (xe > ye) |
976 | - return y; |
977 | - else if (xe < ye) |
978 | - return x; |
979 | + /* Signs of inputs are the same, let's compare exponents */ |
980 | + if (xs == 0) { |
981 | + /* Inputs are both positive */ |
982 | + if (xe > ye) |
983 | + return y; |
984 | + else if (xe < ye) |
985 | + return x; |
986 | + } else { |
987 | + /* Inputs are both negative */ |
988 | + if (xe > ye) |
989 | + return x; |
990 | + else if (xe < ye) |
991 | + return y; |
992 | + } |
993 | |
994 | - /* Compare mantissa */ |
995 | + /* Signs and exponents of inputs are equal, let's compare mantissas */ |
996 | + if (xs == 0) { |
997 | + /* Inputs are both positive, with equal signs and exponents */ |
998 | + if (xm <= ym) |
999 | + return x; |
1000 | + return y; |
1001 | + } |
1002 | + /* Inputs are both negative, with equal signs and exponents */ |
1003 | if (xm <= ym) |
1004 | - return x; |
1005 | - return y; |
1006 | + return y; |
1007 | + return x; |
1008 | } |
1009 | |
1010 | union ieee754sp ieee754sp_fmina(union ieee754sp x, union ieee754sp y) |
1011 | @@ -147,14 +173,26 @@ union ieee754sp ieee754sp_fmina(union ieee754sp x, union ieee754sp y) |
1012 | case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
1013 | return ieee754sp_nanxcpt(x); |
1014 | |
1015 | - /* numbers are preferred to NaNs */ |
1016 | + /* |
1017 | + * Quiet NaN handling |
1018 | + */ |
1019 | + |
1020 | + /* |
1021 | + * The case of both inputs quiet NaNs |
1022 | + */ |
1023 | + case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
1024 | + return x; |
1025 | + |
1026 | + /* |
1027 | + * The cases of exactly one input quiet NaN (numbers |
1028 | + * are here preferred as returned values to NaNs) |
1029 | + */ |
1030 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
1031 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
1032 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
1033 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
1034 | return x; |
1035 | |
1036 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
1037 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
1038 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
1039 | case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
1040 | @@ -164,25 +202,25 @@ union ieee754sp ieee754sp_fmina(union ieee754sp x, union ieee754sp y) |
1041 | /* |
1042 | * Infinity and zero handling |
1043 | */ |
1044 | + case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
1045 | + return ieee754sp_inf(xs | ys); |
1046 | + |
1047 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
1048 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
1049 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
1050 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO): |
1051 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
1052 | - return x; |
1053 | + return y; |
1054 | |
1055 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
1056 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF): |
1057 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF): |
1058 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
1059 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM): |
1060 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM): |
1061 | - return y; |
1062 | + return x; |
1063 | |
1064 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
1065 | - if (xs == ys) |
1066 | - return x; |
1067 | - return ieee754sp_zero(1); |
1068 | + return ieee754sp_zero(xs | ys); |
1069 | |
1070 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
1071 | SPDNORMX; |
1072 | @@ -207,7 +245,11 @@ union ieee754sp ieee754sp_fmina(union ieee754sp x, union ieee754sp y) |
1073 | return x; |
1074 | |
1075 | /* Compare mantissa */ |
1076 | - if (xm <= ym) |
1077 | + if (xm < ym) |
1078 | + return x; |
1079 | + else if (xm > ym) |
1080 | + return y; |
1081 | + else if (xs == 1) |
1082 | return x; |
1083 | return y; |
1084 | } |
1085 | diff --git a/arch/mips/math-emu/sp_maddf.c b/arch/mips/math-emu/sp_maddf.c |
1086 | index a8cd8b4f235e..7195fe785d81 100644 |
1087 | --- a/arch/mips/math-emu/sp_maddf.c |
1088 | +++ b/arch/mips/math-emu/sp_maddf.c |
1089 | @@ -14,9 +14,6 @@ |
1090 | |
1091 | #include "ieee754sp.h" |
1092 | |
1093 | -enum maddf_flags { |
1094 | - maddf_negate_product = 1 << 0, |
1095 | -}; |
1096 | |
1097 | static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1098 | union ieee754sp y, enum maddf_flags flags) |
1099 | @@ -24,14 +21,8 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1100 | int re; |
1101 | int rs; |
1102 | unsigned rm; |
1103 | - unsigned short lxm; |
1104 | - unsigned short hxm; |
1105 | - unsigned short lym; |
1106 | - unsigned short hym; |
1107 | - unsigned lrm; |
1108 | - unsigned hrm; |
1109 | - unsigned t; |
1110 | - unsigned at; |
1111 | + uint64_t rm64; |
1112 | + uint64_t zm64; |
1113 | int s; |
1114 | |
1115 | COMPXSP; |
1116 | @@ -48,51 +39,35 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1117 | |
1118 | ieee754_clearcx(); |
1119 | |
1120 | - switch (zc) { |
1121 | - case IEEE754_CLASS_SNAN: |
1122 | - ieee754_setcx(IEEE754_INVALID_OPERATION); |
1123 | + /* |
1124 | + * Handle the cases when at least one of x, y or z is a NaN. |
1125 | + * Order of precedence is sNaN, qNaN and z, x, y. |
1126 | + */ |
1127 | + if (zc == IEEE754_CLASS_SNAN) |
1128 | return ieee754sp_nanxcpt(z); |
1129 | - case IEEE754_CLASS_DNORM: |
1130 | - SPDNORMZ; |
1131 | - /* QNAN is handled separately below */ |
1132 | - } |
1133 | - |
1134 | - switch (CLPAIR(xc, yc)) { |
1135 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN): |
1136 | - case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN): |
1137 | - case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN): |
1138 | - case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN): |
1139 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN): |
1140 | + if (xc == IEEE754_CLASS_SNAN) |
1141 | + return ieee754sp_nanxcpt(x); |
1142 | + if (yc == IEEE754_CLASS_SNAN) |
1143 | return ieee754sp_nanxcpt(y); |
1144 | + if (zc == IEEE754_CLASS_QNAN) |
1145 | + return z; |
1146 | + if (xc == IEEE754_CLASS_QNAN) |
1147 | + return x; |
1148 | + if (yc == IEEE754_CLASS_QNAN) |
1149 | + return y; |
1150 | |
1151 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN): |
1152 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN): |
1153 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO): |
1154 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM): |
1155 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM): |
1156 | - case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF): |
1157 | - return ieee754sp_nanxcpt(x); |
1158 | + if (zc == IEEE754_CLASS_DNORM) |
1159 | + SPDNORMZ; |
1160 | + /* ZERO z cases are handled separately below */ |
1161 | |
1162 | - case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN): |
1163 | - case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN): |
1164 | - case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN): |
1165 | - case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN): |
1166 | - return y; |
1167 | + switch (CLPAIR(xc, yc)) { |
1168 | |
1169 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN): |
1170 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO): |
1171 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM): |
1172 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM): |
1173 | - case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF): |
1174 | - return x; |
1175 | |
1176 | /* |
1177 | * Infinity handling |
1178 | */ |
1179 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
1180 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
1181 | - if (zc == IEEE754_CLASS_QNAN) |
1182 | - return z; |
1183 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
1184 | return ieee754sp_indef(); |
1185 | |
1186 | @@ -101,9 +76,27 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1187 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
1188 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
1189 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
1190 | - if (zc == IEEE754_CLASS_QNAN) |
1191 | - return z; |
1192 | - return ieee754sp_inf(xs ^ ys); |
1193 | + if ((zc == IEEE754_CLASS_INF) && |
1194 | + ((!(flags & MADDF_NEGATE_PRODUCT) && (zs != (xs ^ ys))) || |
1195 | + ((flags & MADDF_NEGATE_PRODUCT) && (zs == (xs ^ ys))))) { |
1196 | + /* |
1197 | + * Cases of addition of infinities with opposite signs |
1198 | + * or subtraction of infinities with same signs. |
1199 | + */ |
1200 | + ieee754_setcx(IEEE754_INVALID_OPERATION); |
1201 | + return ieee754sp_indef(); |
1202 | + } |
1203 | + /* |
1204 | + * z is here either not an infinity, or an infinity having the |
1205 | + * same sign as product (x*y) (in case of MADDF.D instruction) |
1206 | + * or product -(x*y) (in MSUBF.D case). The result must be an |
1207 | + * infinity, and its sign is determined only by the value of |
1208 | + * (flags & MADDF_NEGATE_PRODUCT) and the signs of x and y. |
1209 | + */ |
1210 | + if (flags & MADDF_NEGATE_PRODUCT) |
1211 | + return ieee754sp_inf(1 ^ (xs ^ ys)); |
1212 | + else |
1213 | + return ieee754sp_inf(xs ^ ys); |
1214 | |
1215 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
1216 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM): |
1217 | @@ -112,32 +105,42 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1218 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
1219 | if (zc == IEEE754_CLASS_INF) |
1220 | return ieee754sp_inf(zs); |
1221 | - /* Multiplication is 0 so just return z */ |
1222 | + if (zc == IEEE754_CLASS_ZERO) { |
1223 | + /* Handle cases +0 + (-0) and similar ones. */ |
1224 | + if ((!(flags & MADDF_NEGATE_PRODUCT) |
1225 | + && (zs == (xs ^ ys))) || |
1226 | + ((flags & MADDF_NEGATE_PRODUCT) |
1227 | + && (zs != (xs ^ ys)))) |
1228 | + /* |
1229 | + * Cases of addition of zeros of equal signs |
1230 | + * or subtraction of zeroes of opposite signs. |
1231 | + * The sign of the resulting zero is in any |
1232 | + * such case determined only by the sign of z. |
1233 | + */ |
1234 | + return z; |
1235 | + |
1236 | + return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD); |
1237 | + } |
1238 | + /* x*y is here 0, and z is not 0, so just return z */ |
1239 | return z; |
1240 | |
1241 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
1242 | SPDNORMX; |
1243 | |
1244 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM): |
1245 | - if (zc == IEEE754_CLASS_QNAN) |
1246 | - return z; |
1247 | - else if (zc == IEEE754_CLASS_INF) |
1248 | + if (zc == IEEE754_CLASS_INF) |
1249 | return ieee754sp_inf(zs); |
1250 | SPDNORMY; |
1251 | break; |
1252 | |
1253 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM): |
1254 | - if (zc == IEEE754_CLASS_QNAN) |
1255 | - return z; |
1256 | - else if (zc == IEEE754_CLASS_INF) |
1257 | + if (zc == IEEE754_CLASS_INF) |
1258 | return ieee754sp_inf(zs); |
1259 | SPDNORMX; |
1260 | break; |
1261 | |
1262 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM): |
1263 | - if (zc == IEEE754_CLASS_QNAN) |
1264 | - return z; |
1265 | - else if (zc == IEEE754_CLASS_INF) |
1266 | + if (zc == IEEE754_CLASS_INF) |
1267 | return ieee754sp_inf(zs); |
1268 | /* fall through to real computations */ |
1269 | } |
1270 | @@ -158,108 +161,93 @@ static union ieee754sp _sp_maddf(union ieee754sp z, union ieee754sp x, |
1271 | |
1272 | re = xe + ye; |
1273 | rs = xs ^ ys; |
1274 | - if (flags & maddf_negate_product) |
1275 | + if (flags & MADDF_NEGATE_PRODUCT) |
1276 | rs ^= 1; |
1277 | |
1278 | - /* shunt to top of word */ |
1279 | - xm <<= 32 - (SP_FBITS + 1); |
1280 | - ym <<= 32 - (SP_FBITS + 1); |
1281 | - |
1282 | - /* |
1283 | - * Multiply 32 bits xm, ym to give high 32 bits rm with stickness. |
1284 | - */ |
1285 | - lxm = xm & 0xffff; |
1286 | - hxm = xm >> 16; |
1287 | - lym = ym & 0xffff; |
1288 | - hym = ym >> 16; |
1289 | - |
1290 | - lrm = lxm * lym; /* 16 * 16 => 32 */ |
1291 | - hrm = hxm * hym; /* 16 * 16 => 32 */ |
1292 | + /* Multiple 24 bit xm and ym to give 48 bit results */ |
1293 | + rm64 = (uint64_t)xm * ym; |
1294 | |
1295 | - t = lxm * hym; /* 16 * 16 => 32 */ |
1296 | - at = lrm + (t << 16); |
1297 | - hrm += at < lrm; |
1298 | - lrm = at; |
1299 | - hrm = hrm + (t >> 16); |
1300 | + /* Shunt to top of word */ |
1301 | + rm64 = rm64 << 16; |
1302 | |
1303 | - t = hxm * lym; /* 16 * 16 => 32 */ |
1304 | - at = lrm + (t << 16); |
1305 | - hrm += at < lrm; |
1306 | - lrm = at; |
1307 | - hrm = hrm + (t >> 16); |
1308 | - |
1309 | - rm = hrm | (lrm != 0); |
1310 | - |
1311 | - /* |
1312 | - * Sticky shift down to normal rounding precision. |
1313 | - */ |
1314 | - if ((int) rm < 0) { |
1315 | - rm = (rm >> (32 - (SP_FBITS + 1 + 3))) | |
1316 | - ((rm << (SP_FBITS + 1 + 3)) != 0); |
1317 | + /* Put explicit bit at bit 62 if necessary */ |
1318 | + if ((int64_t) rm64 < 0) { |
1319 | + rm64 = rm64 >> 1; |
1320 | re++; |
1321 | - } else { |
1322 | - rm = (rm >> (32 - (SP_FBITS + 1 + 3 + 1))) | |
1323 | - ((rm << (SP_FBITS + 1 + 3 + 1)) != 0); |
1324 | } |
1325 | - assert(rm & (SP_HIDDEN_BIT << 3)); |
1326 | |
1327 | - /* And now the addition */ |
1328 | + assert(rm64 & (1 << 62)); |
1329 | |
1330 | - assert(zm & SP_HIDDEN_BIT); |
1331 | + if (zc == IEEE754_CLASS_ZERO) { |
1332 | + /* |
1333 | + * Move explicit bit from bit 62 to bit 26 since the |
1334 | + * ieee754sp_format code expects the mantissa to be |
1335 | + * 27 bits wide (24 + 3 rounding bits). |
1336 | + */ |
1337 | + rm = XSPSRS64(rm64, (62 - 26)); |
1338 | + return ieee754sp_format(rs, re, rm); |
1339 | + } |
1340 | |
1341 | - /* |
1342 | - * Provide guard,round and stick bit space. |
1343 | - */ |
1344 | - zm <<= 3; |
1345 | + /* Move explicit bit from bit 23 to bit 62 */ |
1346 | + zm64 = (uint64_t)zm << (62 - 23); |
1347 | + assert(zm64 & (1 << 62)); |
1348 | |
1349 | + /* Make the exponents the same */ |
1350 | if (ze > re) { |
1351 | /* |
1352 | * Have to shift r fraction right to align. |
1353 | */ |
1354 | s = ze - re; |
1355 | - rm = XSPSRS(rm, s); |
1356 | + rm64 = XSPSRS64(rm64, s); |
1357 | re += s; |
1358 | } else if (re > ze) { |
1359 | /* |
1360 | * Have to shift z fraction right to align. |
1361 | */ |
1362 | s = re - ze; |
1363 | - zm = XSPSRS(zm, s); |
1364 | + zm64 = XSPSRS64(zm64, s); |
1365 | ze += s; |
1366 | } |
1367 | assert(ze == re); |
1368 | assert(ze <= SP_EMAX); |
1369 | |
1370 | + /* Do the addition */ |
1371 | if (zs == rs) { |
1372 | /* |
1373 | - * Generate 28 bit result of adding two 27 bit numbers |
1374 | - * leaving result in zm, zs and ze. |
1375 | + * Generate 64 bit result by adding two 63 bit numbers |
1376 | + * leaving result in zm64, zs and ze. |
1377 | */ |
1378 | - zm = zm + rm; |
1379 | - |
1380 | - if (zm >> (SP_FBITS + 1 + 3)) { /* carry out */ |
1381 | - zm = XSPSRS1(zm); |
1382 | + zm64 = zm64 + rm64; |
1383 | + if ((int64_t)zm64 < 0) { /* carry out */ |
1384 | + zm64 = XSPSRS1(zm64); |
1385 | ze++; |
1386 | } |
1387 | } else { |
1388 | - if (zm >= rm) { |
1389 | - zm = zm - rm; |
1390 | + if (zm64 >= rm64) { |
1391 | + zm64 = zm64 - rm64; |
1392 | } else { |
1393 | - zm = rm - zm; |
1394 | + zm64 = rm64 - zm64; |
1395 | zs = rs; |
1396 | } |
1397 | - if (zm == 0) |
1398 | + if (zm64 == 0) |
1399 | return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD); |
1400 | |
1401 | /* |
1402 | - * Normalize in extended single precision |
1403 | + * Put explicit bit at bit 62 if necessary. |
1404 | */ |
1405 | - while ((zm >> (SP_MBITS + 3)) == 0) { |
1406 | - zm <<= 1; |
1407 | + while ((zm64 >> 62) == 0) { |
1408 | + zm64 <<= 1; |
1409 | ze--; |
1410 | } |
1411 | - |
1412 | } |
1413 | + |
1414 | + /* |
1415 | + * Move explicit bit from bit 62 to bit 26 since the |
1416 | + * ieee754sp_format code expects the mantissa to be |
1417 | + * 27 bits wide (24 + 3 rounding bits). |
1418 | + */ |
1419 | + zm = XSPSRS64(zm64, (62 - 26)); |
1420 | + |
1421 | return ieee754sp_format(zs, ze, zm); |
1422 | } |
1423 | |
1424 | @@ -272,5 +260,5 @@ union ieee754sp ieee754sp_maddf(union ieee754sp z, union ieee754sp x, |
1425 | union ieee754sp ieee754sp_msubf(union ieee754sp z, union ieee754sp x, |
1426 | union ieee754sp y) |
1427 | { |
1428 | - return _sp_maddf(z, x, y, maddf_negate_product); |
1429 | + return _sp_maddf(z, x, y, MADDF_NEGATE_PRODUCT); |
1430 | } |
1431 | diff --git a/arch/powerpc/kernel/align.c b/arch/powerpc/kernel/align.c |
1432 | index b2da7c8baed7..292458b694fb 100644 |
1433 | --- a/arch/powerpc/kernel/align.c |
1434 | +++ b/arch/powerpc/kernel/align.c |
1435 | @@ -235,6 +235,28 @@ static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr) |
1436 | |
1437 | #define SWIZ_PTR(p) ((unsigned char __user *)((p) ^ swiz)) |
1438 | |
1439 | +#define __get_user_or_set_dar(_regs, _dest, _addr) \ |
1440 | + ({ \ |
1441 | + int rc = 0; \ |
1442 | + typeof(_addr) __addr = (_addr); \ |
1443 | + if (__get_user_inatomic(_dest, __addr)) { \ |
1444 | + _regs->dar = (unsigned long)__addr; \ |
1445 | + rc = -EFAULT; \ |
1446 | + } \ |
1447 | + rc; \ |
1448 | + }) |
1449 | + |
1450 | +#define __put_user_or_set_dar(_regs, _src, _addr) \ |
1451 | + ({ \ |
1452 | + int rc = 0; \ |
1453 | + typeof(_addr) __addr = (_addr); \ |
1454 | + if (__put_user_inatomic(_src, __addr)) { \ |
1455 | + _regs->dar = (unsigned long)__addr; \ |
1456 | + rc = -EFAULT; \ |
1457 | + } \ |
1458 | + rc; \ |
1459 | + }) |
1460 | + |
1461 | static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr, |
1462 | unsigned int reg, unsigned int nb, |
1463 | unsigned int flags, unsigned int instr, |
1464 | @@ -263,9 +285,10 @@ static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr, |
1465 | } else { |
1466 | unsigned long pc = regs->nip ^ (swiz & 4); |
1467 | |
1468 | - if (__get_user_inatomic(instr, |
1469 | - (unsigned int __user *)pc)) |
1470 | + if (__get_user_or_set_dar(regs, instr, |
1471 | + (unsigned int __user *)pc)) |
1472 | return -EFAULT; |
1473 | + |
1474 | if (swiz == 0 && (flags & SW)) |
1475 | instr = cpu_to_le32(instr); |
1476 | nb = (instr >> 11) & 0x1f; |
1477 | @@ -309,31 +332,31 @@ static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr, |
1478 | ((nb0 + 3) / 4) * sizeof(unsigned long)); |
1479 | |
1480 | for (i = 0; i < nb; ++i, ++p) |
1481 | - if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz), |
1482 | - SWIZ_PTR(p))) |
1483 | + if (__get_user_or_set_dar(regs, REG_BYTE(rptr, i ^ bswiz), |
1484 | + SWIZ_PTR(p))) |
1485 | return -EFAULT; |
1486 | if (nb0 > 0) { |
1487 | rptr = ®s->gpr[0]; |
1488 | addr += nb; |
1489 | for (i = 0; i < nb0; ++i, ++p) |
1490 | - if (__get_user_inatomic(REG_BYTE(rptr, |
1491 | - i ^ bswiz), |
1492 | - SWIZ_PTR(p))) |
1493 | + if (__get_user_or_set_dar(regs, |
1494 | + REG_BYTE(rptr, i ^ bswiz), |
1495 | + SWIZ_PTR(p))) |
1496 | return -EFAULT; |
1497 | } |
1498 | |
1499 | } else { |
1500 | for (i = 0; i < nb; ++i, ++p) |
1501 | - if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz), |
1502 | - SWIZ_PTR(p))) |
1503 | + if (__put_user_or_set_dar(regs, REG_BYTE(rptr, i ^ bswiz), |
1504 | + SWIZ_PTR(p))) |
1505 | return -EFAULT; |
1506 | if (nb0 > 0) { |
1507 | rptr = ®s->gpr[0]; |
1508 | addr += nb; |
1509 | for (i = 0; i < nb0; ++i, ++p) |
1510 | - if (__put_user_inatomic(REG_BYTE(rptr, |
1511 | - i ^ bswiz), |
1512 | - SWIZ_PTR(p))) |
1513 | + if (__put_user_or_set_dar(regs, |
1514 | + REG_BYTE(rptr, i ^ bswiz), |
1515 | + SWIZ_PTR(p))) |
1516 | return -EFAULT; |
1517 | } |
1518 | } |
1519 | @@ -345,29 +368,32 @@ static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr, |
1520 | * Only POWER6 has these instructions, and it does true little-endian, |
1521 | * so we don't need the address swizzling. |
1522 | */ |
1523 | -static int emulate_fp_pair(unsigned char __user *addr, unsigned int reg, |
1524 | - unsigned int flags) |
1525 | +static int emulate_fp_pair(struct pt_regs *regs, unsigned char __user *addr, |
1526 | + unsigned int reg, unsigned int flags) |
1527 | { |
1528 | char *ptr0 = (char *) ¤t->thread.TS_FPR(reg); |
1529 | char *ptr1 = (char *) ¤t->thread.TS_FPR(reg+1); |
1530 | - int i, ret, sw = 0; |
1531 | + int i, sw = 0; |
1532 | |
1533 | if (reg & 1) |
1534 | return 0; /* invalid form: FRS/FRT must be even */ |
1535 | if (flags & SW) |
1536 | sw = 7; |
1537 | - ret = 0; |
1538 | + |
1539 | for (i = 0; i < 8; ++i) { |
1540 | if (!(flags & ST)) { |
1541 | - ret |= __get_user(ptr0[i^sw], addr + i); |
1542 | - ret |= __get_user(ptr1[i^sw], addr + i + 8); |
1543 | + if (__get_user_or_set_dar(regs, ptr0[i^sw], addr + i)) |
1544 | + return -EFAULT; |
1545 | + if (__get_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8)) |
1546 | + return -EFAULT; |
1547 | } else { |
1548 | - ret |= __put_user(ptr0[i^sw], addr + i); |
1549 | - ret |= __put_user(ptr1[i^sw], addr + i + 8); |
1550 | + if (__put_user_or_set_dar(regs, ptr0[i^sw], addr + i)) |
1551 | + return -EFAULT; |
1552 | + if (__put_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8)) |
1553 | + return -EFAULT; |
1554 | } |
1555 | } |
1556 | - if (ret) |
1557 | - return -EFAULT; |
1558 | + |
1559 | return 1; /* exception handled and fixed up */ |
1560 | } |
1561 | |
1562 | @@ -377,24 +403,27 @@ static int emulate_lq_stq(struct pt_regs *regs, unsigned char __user *addr, |
1563 | { |
1564 | char *ptr0 = (char *)®s->gpr[reg]; |
1565 | char *ptr1 = (char *)®s->gpr[reg+1]; |
1566 | - int i, ret, sw = 0; |
1567 | + int i, sw = 0; |
1568 | |
1569 | if (reg & 1) |
1570 | return 0; /* invalid form: GPR must be even */ |
1571 | if (flags & SW) |
1572 | sw = 7; |
1573 | - ret = 0; |
1574 | + |
1575 | for (i = 0; i < 8; ++i) { |
1576 | if (!(flags & ST)) { |
1577 | - ret |= __get_user(ptr0[i^sw], addr + i); |
1578 | - ret |= __get_user(ptr1[i^sw], addr + i + 8); |
1579 | + if (__get_user_or_set_dar(regs, ptr0[i^sw], addr + i)) |
1580 | + return -EFAULT; |
1581 | + if (__get_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8)) |
1582 | + return -EFAULT; |
1583 | } else { |
1584 | - ret |= __put_user(ptr0[i^sw], addr + i); |
1585 | - ret |= __put_user(ptr1[i^sw], addr + i + 8); |
1586 | + if (__put_user_or_set_dar(regs, ptr0[i^sw], addr + i)) |
1587 | + return -EFAULT; |
1588 | + if (__put_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8)) |
1589 | + return -EFAULT; |
1590 | } |
1591 | } |
1592 | - if (ret) |
1593 | - return -EFAULT; |
1594 | + |
1595 | return 1; /* exception handled and fixed up */ |
1596 | } |
1597 | #endif /* CONFIG_PPC64 */ |
1598 | @@ -687,9 +716,14 @@ static int emulate_vsx(unsigned char __user *addr, unsigned int reg, |
1599 | for (j = 0; j < length; j += elsize) { |
1600 | for (i = 0; i < elsize; ++i) { |
1601 | if (flags & ST) |
1602 | - ret |= __put_user(ptr[i^sw], addr + i); |
1603 | + ret = __put_user_or_set_dar(regs, ptr[i^sw], |
1604 | + addr + i); |
1605 | else |
1606 | - ret |= __get_user(ptr[i^sw], addr + i); |
1607 | + ret = __get_user_or_set_dar(regs, ptr[i^sw], |
1608 | + addr + i); |
1609 | + |
1610 | + if (ret) |
1611 | + return ret; |
1612 | } |
1613 | ptr += elsize; |
1614 | #ifdef __LITTLE_ENDIAN__ |
1615 | @@ -739,7 +773,7 @@ int fix_alignment(struct pt_regs *regs) |
1616 | unsigned int dsisr; |
1617 | unsigned char __user *addr; |
1618 | unsigned long p, swiz; |
1619 | - int ret, i; |
1620 | + int i; |
1621 | union data { |
1622 | u64 ll; |
1623 | double dd; |
1624 | @@ -936,7 +970,7 @@ int fix_alignment(struct pt_regs *regs) |
1625 | if (flags & F) { |
1626 | /* Special case for 16-byte FP loads and stores */ |
1627 | PPC_WARN_ALIGNMENT(fp_pair, regs); |
1628 | - return emulate_fp_pair(addr, reg, flags); |
1629 | + return emulate_fp_pair(regs, addr, reg, flags); |
1630 | } else { |
1631 | #ifdef CONFIG_PPC64 |
1632 | /* Special case for 16-byte loads and stores */ |
1633 | @@ -966,15 +1000,12 @@ int fix_alignment(struct pt_regs *regs) |
1634 | } |
1635 | |
1636 | data.ll = 0; |
1637 | - ret = 0; |
1638 | p = (unsigned long)addr; |
1639 | |
1640 | for (i = 0; i < nb; i++) |
1641 | - ret |= __get_user_inatomic(data.v[start + i], |
1642 | - SWIZ_PTR(p++)); |
1643 | - |
1644 | - if (unlikely(ret)) |
1645 | - return -EFAULT; |
1646 | + if (__get_user_or_set_dar(regs, data.v[start + i], |
1647 | + SWIZ_PTR(p++))) |
1648 | + return -EFAULT; |
1649 | |
1650 | } else if (flags & F) { |
1651 | data.ll = current->thread.TS_FPR(reg); |
1652 | @@ -1046,15 +1077,13 @@ int fix_alignment(struct pt_regs *regs) |
1653 | break; |
1654 | } |
1655 | |
1656 | - ret = 0; |
1657 | p = (unsigned long)addr; |
1658 | |
1659 | for (i = 0; i < nb; i++) |
1660 | - ret |= __put_user_inatomic(data.v[start + i], |
1661 | - SWIZ_PTR(p++)); |
1662 | + if (__put_user_or_set_dar(regs, data.v[start + i], |
1663 | + SWIZ_PTR(p++))) |
1664 | + return -EFAULT; |
1665 | |
1666 | - if (unlikely(ret)) |
1667 | - return -EFAULT; |
1668 | } else if (flags & F) |
1669 | current->thread.TS_FPR(reg) = data.ll; |
1670 | else |
1671 | diff --git a/arch/s390/include/asm/mmu.h b/arch/s390/include/asm/mmu.h |
1672 | index bea785d7f853..af85d6b12028 100644 |
1673 | --- a/arch/s390/include/asm/mmu.h |
1674 | +++ b/arch/s390/include/asm/mmu.h |
1675 | @@ -5,6 +5,7 @@ |
1676 | #include <linux/errno.h> |
1677 | |
1678 | typedef struct { |
1679 | + spinlock_t lock; |
1680 | cpumask_t cpu_attach_mask; |
1681 | atomic_t flush_count; |
1682 | unsigned int flush_mm; |
1683 | @@ -25,6 +26,7 @@ typedef struct { |
1684 | } mm_context_t; |
1685 | |
1686 | #define INIT_MM_CONTEXT(name) \ |
1687 | + .context.lock = __SPIN_LOCK_UNLOCKED(name.context.lock), \ |
1688 | .context.pgtable_lock = \ |
1689 | __SPIN_LOCK_UNLOCKED(name.context.pgtable_lock), \ |
1690 | .context.pgtable_list = LIST_HEAD_INIT(name.context.pgtable_list), \ |
1691 | diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h |
1692 | index 515fea5a3fc4..f65a708ac395 100644 |
1693 | --- a/arch/s390/include/asm/mmu_context.h |
1694 | +++ b/arch/s390/include/asm/mmu_context.h |
1695 | @@ -15,6 +15,7 @@ |
1696 | static inline int init_new_context(struct task_struct *tsk, |
1697 | struct mm_struct *mm) |
1698 | { |
1699 | + spin_lock_init(&mm->context.lock); |
1700 | spin_lock_init(&mm->context.pgtable_lock); |
1701 | INIT_LIST_HEAD(&mm->context.pgtable_list); |
1702 | spin_lock_init(&mm->context.gmap_lock); |
1703 | @@ -93,7 +94,6 @@ static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next, |
1704 | if (prev == next) |
1705 | return; |
1706 | cpumask_set_cpu(cpu, &next->context.cpu_attach_mask); |
1707 | - cpumask_set_cpu(cpu, mm_cpumask(next)); |
1708 | /* Clear old ASCE by loading the kernel ASCE. */ |
1709 | __ctl_load(S390_lowcore.kernel_asce, 1, 1); |
1710 | __ctl_load(S390_lowcore.kernel_asce, 7, 7); |
1711 | @@ -111,9 +111,8 @@ static inline void finish_arch_post_lock_switch(void) |
1712 | preempt_disable(); |
1713 | while (atomic_read(&mm->context.flush_count)) |
1714 | cpu_relax(); |
1715 | - |
1716 | - if (mm->context.flush_mm) |
1717 | - __tlb_flush_mm(mm); |
1718 | + cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm)); |
1719 | + __tlb_flush_mm_lazy(mm); |
1720 | preempt_enable(); |
1721 | } |
1722 | set_fs(current->thread.mm_segment); |
1723 | @@ -126,6 +125,7 @@ static inline void activate_mm(struct mm_struct *prev, |
1724 | struct mm_struct *next) |
1725 | { |
1726 | switch_mm(prev, next, current); |
1727 | + cpumask_set_cpu(smp_processor_id(), mm_cpumask(next)); |
1728 | set_user_asce(next); |
1729 | } |
1730 | |
1731 | diff --git a/arch/s390/include/asm/tlbflush.h b/arch/s390/include/asm/tlbflush.h |
1732 | index 39846100682a..eed927aeb08f 100644 |
1733 | --- a/arch/s390/include/asm/tlbflush.h |
1734 | +++ b/arch/s390/include/asm/tlbflush.h |
1735 | @@ -43,23 +43,6 @@ static inline void __tlb_flush_global(void) |
1736 | * Flush TLB entries for a specific mm on all CPUs (in case gmap is used |
1737 | * this implicates multiple ASCEs!). |
1738 | */ |
1739 | -static inline void __tlb_flush_full(struct mm_struct *mm) |
1740 | -{ |
1741 | - preempt_disable(); |
1742 | - atomic_inc(&mm->context.flush_count); |
1743 | - if (cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id()))) { |
1744 | - /* Local TLB flush */ |
1745 | - __tlb_flush_local(); |
1746 | - } else { |
1747 | - /* Global TLB flush */ |
1748 | - __tlb_flush_global(); |
1749 | - /* Reset TLB flush mask */ |
1750 | - cpumask_copy(mm_cpumask(mm), &mm->context.cpu_attach_mask); |
1751 | - } |
1752 | - atomic_dec(&mm->context.flush_count); |
1753 | - preempt_enable(); |
1754 | -} |
1755 | - |
1756 | static inline void __tlb_flush_mm(struct mm_struct *mm) |
1757 | { |
1758 | unsigned long gmap_asce; |
1759 | @@ -71,16 +54,18 @@ static inline void __tlb_flush_mm(struct mm_struct *mm) |
1760 | */ |
1761 | preempt_disable(); |
1762 | atomic_inc(&mm->context.flush_count); |
1763 | + /* Reset TLB flush mask */ |
1764 | + cpumask_copy(mm_cpumask(mm), &mm->context.cpu_attach_mask); |
1765 | + barrier(); |
1766 | gmap_asce = READ_ONCE(mm->context.gmap_asce); |
1767 | if (MACHINE_HAS_IDTE && gmap_asce != -1UL) { |
1768 | if (gmap_asce) |
1769 | __tlb_flush_idte(gmap_asce); |
1770 | __tlb_flush_idte(mm->context.asce); |
1771 | } else { |
1772 | - __tlb_flush_full(mm); |
1773 | + /* Global TLB flush */ |
1774 | + __tlb_flush_global(); |
1775 | } |
1776 | - /* Reset TLB flush mask */ |
1777 | - cpumask_copy(mm_cpumask(mm), &mm->context.cpu_attach_mask); |
1778 | atomic_dec(&mm->context.flush_count); |
1779 | preempt_enable(); |
1780 | } |
1781 | @@ -94,7 +79,6 @@ static inline void __tlb_flush_kernel(void) |
1782 | } |
1783 | #else |
1784 | #define __tlb_flush_global() __tlb_flush_local() |
1785 | -#define __tlb_flush_full(mm) __tlb_flush_local() |
1786 | |
1787 | /* |
1788 | * Flush TLB entries for a specific ASCE on all CPUs. |
1789 | @@ -112,10 +96,12 @@ static inline void __tlb_flush_kernel(void) |
1790 | |
1791 | static inline void __tlb_flush_mm_lazy(struct mm_struct * mm) |
1792 | { |
1793 | + spin_lock(&mm->context.lock); |
1794 | if (mm->context.flush_mm) { |
1795 | - __tlb_flush_mm(mm); |
1796 | mm->context.flush_mm = 0; |
1797 | + __tlb_flush_mm(mm); |
1798 | } |
1799 | + spin_unlock(&mm->context.lock); |
1800 | } |
1801 | |
1802 | /* |
1803 | diff --git a/block/blk-core.c b/block/blk-core.c |
1804 | index d1f2801ce836..95379fc83805 100644 |
1805 | --- a/block/blk-core.c |
1806 | +++ b/block/blk-core.c |
1807 | @@ -233,7 +233,7 @@ EXPORT_SYMBOL(blk_start_queue_async); |
1808 | **/ |
1809 | void blk_start_queue(struct request_queue *q) |
1810 | { |
1811 | - WARN_ON(!irqs_disabled()); |
1812 | + WARN_ON(!in_interrupt() && !irqs_disabled()); |
1813 | |
1814 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
1815 | __blk_run_queue(q); |
1816 | diff --git a/crypto/algif_skcipher.c b/crypto/algif_skcipher.c |
1817 | index 45af0fe00f33..aaf2f810d170 100644 |
1818 | --- a/crypto/algif_skcipher.c |
1819 | +++ b/crypto/algif_skcipher.c |
1820 | @@ -143,8 +143,10 @@ static int skcipher_alloc_sgl(struct sock *sk) |
1821 | sg_init_table(sgl->sg, MAX_SGL_ENTS + 1); |
1822 | sgl->cur = 0; |
1823 | |
1824 | - if (sg) |
1825 | + if (sg) { |
1826 | sg_chain(sg, MAX_SGL_ENTS + 1, sgl->sg); |
1827 | + sg_unmark_end(sg + (MAX_SGL_ENTS - 1)); |
1828 | + } |
1829 | |
1830 | list_add_tail(&sgl->list, &ctx->tsgl); |
1831 | } |
1832 | diff --git a/drivers/block/skd_main.c b/drivers/block/skd_main.c |
1833 | index 3822eae102db..6f78cea75103 100644 |
1834 | --- a/drivers/block/skd_main.c |
1835 | +++ b/drivers/block/skd_main.c |
1836 | @@ -2163,6 +2163,9 @@ static void skd_send_fitmsg(struct skd_device *skdev, |
1837 | */ |
1838 | qcmd |= FIT_QCMD_MSGSIZE_64; |
1839 | |
1840 | + /* Make sure skd_msg_buf is written before the doorbell is triggered. */ |
1841 | + smp_wmb(); |
1842 | + |
1843 | SKD_WRITEQ(skdev, qcmd, FIT_Q_COMMAND); |
1844 | |
1845 | } |
1846 | @@ -2209,6 +2212,9 @@ static void skd_send_special_fitmsg(struct skd_device *skdev, |
1847 | qcmd = skspcl->mb_dma_address; |
1848 | qcmd |= FIT_QCMD_QID_NORMAL + FIT_QCMD_MSGSIZE_128; |
1849 | |
1850 | + /* Make sure skd_msg_buf is written before the doorbell is triggered. */ |
1851 | + smp_wmb(); |
1852 | + |
1853 | SKD_WRITEQ(skdev, qcmd, FIT_Q_COMMAND); |
1854 | } |
1855 | |
1856 | @@ -4622,15 +4628,16 @@ static void skd_free_disk(struct skd_device *skdev) |
1857 | { |
1858 | struct gendisk *disk = skdev->disk; |
1859 | |
1860 | - if (disk != NULL) { |
1861 | - struct request_queue *q = disk->queue; |
1862 | + if (disk && (disk->flags & GENHD_FL_UP)) |
1863 | + del_gendisk(disk); |
1864 | |
1865 | - if (disk->flags & GENHD_FL_UP) |
1866 | - del_gendisk(disk); |
1867 | - if (q) |
1868 | - blk_cleanup_queue(q); |
1869 | - put_disk(disk); |
1870 | + if (skdev->queue) { |
1871 | + blk_cleanup_queue(skdev->queue); |
1872 | + skdev->queue = NULL; |
1873 | + disk->queue = NULL; |
1874 | } |
1875 | + |
1876 | + put_disk(disk); |
1877 | skdev->disk = NULL; |
1878 | } |
1879 | |
1880 | diff --git a/drivers/crypto/ccp/ccp-crypto-aes-xts.c b/drivers/crypto/ccp/ccp-crypto-aes-xts.c |
1881 | index 58a4244b4752..3f26a415ef44 100644 |
1882 | --- a/drivers/crypto/ccp/ccp-crypto-aes-xts.c |
1883 | +++ b/drivers/crypto/ccp/ccp-crypto-aes-xts.c |
1884 | @@ -1,8 +1,9 @@ |
1885 | /* |
1886 | * AMD Cryptographic Coprocessor (CCP) AES XTS crypto API support |
1887 | * |
1888 | - * Copyright (C) 2013 Advanced Micro Devices, Inc. |
1889 | + * Copyright (C) 2013,2017 Advanced Micro Devices, Inc. |
1890 | * |
1891 | + * Author: Gary R Hook <gary.hook@amd.com> |
1892 | * Author: Tom Lendacky <thomas.lendacky@amd.com> |
1893 | * |
1894 | * This program is free software; you can redistribute it and/or modify |
1895 | @@ -164,6 +165,7 @@ static int ccp_aes_xts_crypt(struct ablkcipher_request *req, |
1896 | memset(&rctx->cmd, 0, sizeof(rctx->cmd)); |
1897 | INIT_LIST_HEAD(&rctx->cmd.entry); |
1898 | rctx->cmd.engine = CCP_ENGINE_XTS_AES_128; |
1899 | + rctx->cmd.u.xts.type = CCP_AES_TYPE_128; |
1900 | rctx->cmd.u.xts.action = (encrypt) ? CCP_AES_ACTION_ENCRYPT |
1901 | : CCP_AES_ACTION_DECRYPT; |
1902 | rctx->cmd.u.xts.unit_size = unit_size; |
1903 | diff --git a/drivers/crypto/ccp/ccp-dev-v5.c b/drivers/crypto/ccp/ccp-dev-v5.c |
1904 | index 2c0ce5f605b3..17b19a68e269 100644 |
1905 | --- a/drivers/crypto/ccp/ccp-dev-v5.c |
1906 | +++ b/drivers/crypto/ccp/ccp-dev-v5.c |
1907 | @@ -131,6 +131,7 @@ union ccp_function { |
1908 | #define CCP_AES_MODE(p) ((p)->aes.mode) |
1909 | #define CCP_AES_TYPE(p) ((p)->aes.type) |
1910 | #define CCP_XTS_SIZE(p) ((p)->aes_xts.size) |
1911 | +#define CCP_XTS_TYPE(p) ((p)->aes_xts.type) |
1912 | #define CCP_XTS_ENCRYPT(p) ((p)->aes_xts.encrypt) |
1913 | #define CCP_SHA_TYPE(p) ((p)->sha.type) |
1914 | #define CCP_RSA_SIZE(p) ((p)->rsa.size) |
1915 | @@ -318,6 +319,7 @@ static int ccp5_perform_xts_aes(struct ccp_op *op) |
1916 | CCP5_CMD_PROT(&desc) = 0; |
1917 | |
1918 | function.raw = 0; |
1919 | + CCP_XTS_TYPE(&function) = op->u.xts.type; |
1920 | CCP_XTS_ENCRYPT(&function) = op->u.xts.action; |
1921 | CCP_XTS_SIZE(&function) = op->u.xts.unit_size; |
1922 | CCP5_CMD_FUNCTION(&desc) = function.raw; |
1923 | diff --git a/drivers/crypto/ccp/ccp-dev.h b/drivers/crypto/ccp/ccp-dev.h |
1924 | index 8ac7ae17e1f4..e23c36c7691c 100644 |
1925 | --- a/drivers/crypto/ccp/ccp-dev.h |
1926 | +++ b/drivers/crypto/ccp/ccp-dev.h |
1927 | @@ -187,6 +187,7 @@ |
1928 | #define CCP_AES_CTX_SB_COUNT 1 |
1929 | |
1930 | #define CCP_XTS_AES_KEY_SB_COUNT 1 |
1931 | +#define CCP5_XTS_AES_KEY_SB_COUNT 2 |
1932 | #define CCP_XTS_AES_CTX_SB_COUNT 1 |
1933 | |
1934 | #define CCP_SHA_SB_COUNT 1 |
1935 | @@ -472,6 +473,7 @@ struct ccp_aes_op { |
1936 | }; |
1937 | |
1938 | struct ccp_xts_aes_op { |
1939 | + enum ccp_aes_type type; |
1940 | enum ccp_aes_action action; |
1941 | enum ccp_xts_aes_unit_size unit_size; |
1942 | }; |
1943 | diff --git a/drivers/crypto/ccp/ccp-ops.c b/drivers/crypto/ccp/ccp-ops.c |
1944 | index 50fae4442801..64deb006c3be 100644 |
1945 | --- a/drivers/crypto/ccp/ccp-ops.c |
1946 | +++ b/drivers/crypto/ccp/ccp-ops.c |
1947 | @@ -779,6 +779,8 @@ static int ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, |
1948 | struct ccp_op op; |
1949 | unsigned int unit_size, dm_offset; |
1950 | bool in_place = false; |
1951 | + unsigned int sb_count; |
1952 | + enum ccp_aes_type aestype; |
1953 | int ret; |
1954 | |
1955 | switch (xts->unit_size) { |
1956 | @@ -802,7 +804,9 @@ static int ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, |
1957 | return -EINVAL; |
1958 | } |
1959 | |
1960 | - if (xts->key_len != AES_KEYSIZE_128) |
1961 | + if (xts->key_len == AES_KEYSIZE_128) |
1962 | + aestype = CCP_AES_TYPE_128; |
1963 | + else |
1964 | return -EINVAL; |
1965 | |
1966 | if (!xts->final && (xts->src_len & (AES_BLOCK_SIZE - 1))) |
1967 | @@ -824,23 +828,44 @@ static int ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, |
1968 | op.sb_key = cmd_q->sb_key; |
1969 | op.sb_ctx = cmd_q->sb_ctx; |
1970 | op.init = 1; |
1971 | + op.u.xts.type = aestype; |
1972 | op.u.xts.action = xts->action; |
1973 | op.u.xts.unit_size = xts->unit_size; |
1974 | |
1975 | - /* All supported key sizes fit in a single (32-byte) SB entry |
1976 | - * and must be in little endian format. Use the 256-bit byte |
1977 | - * swap passthru option to convert from big endian to little |
1978 | - * endian. |
1979 | + /* A version 3 device only supports 128-bit keys, which fits into a |
1980 | + * single SB entry. A version 5 device uses a 512-bit vector, so two |
1981 | + * SB entries. |
1982 | */ |
1983 | + if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) |
1984 | + sb_count = CCP_XTS_AES_KEY_SB_COUNT; |
1985 | + else |
1986 | + sb_count = CCP5_XTS_AES_KEY_SB_COUNT; |
1987 | ret = ccp_init_dm_workarea(&key, cmd_q, |
1988 | - CCP_XTS_AES_KEY_SB_COUNT * CCP_SB_BYTES, |
1989 | + sb_count * CCP_SB_BYTES, |
1990 | DMA_TO_DEVICE); |
1991 | if (ret) |
1992 | return ret; |
1993 | |
1994 | - dm_offset = CCP_SB_BYTES - AES_KEYSIZE_128; |
1995 | - ccp_set_dm_area(&key, dm_offset, xts->key, 0, xts->key_len); |
1996 | - ccp_set_dm_area(&key, 0, xts->key, dm_offset, xts->key_len); |
1997 | + if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) { |
1998 | + /* All supported key sizes must be in little endian format. |
1999 | + * Use the 256-bit byte swap passthru option to convert from |
2000 | + * big endian to little endian. |
2001 | + */ |
2002 | + dm_offset = CCP_SB_BYTES - AES_KEYSIZE_128; |
2003 | + ccp_set_dm_area(&key, dm_offset, xts->key, 0, xts->key_len); |
2004 | + ccp_set_dm_area(&key, 0, xts->key, xts->key_len, xts->key_len); |
2005 | + } else { |
2006 | + /* Version 5 CCPs use a 512-bit space for the key: each portion |
2007 | + * occupies 256 bits, or one entire slot, and is zero-padded. |
2008 | + */ |
2009 | + unsigned int pad; |
2010 | + |
2011 | + dm_offset = CCP_SB_BYTES; |
2012 | + pad = dm_offset - xts->key_len; |
2013 | + ccp_set_dm_area(&key, pad, xts->key, 0, xts->key_len); |
2014 | + ccp_set_dm_area(&key, dm_offset + pad, xts->key, xts->key_len, |
2015 | + xts->key_len); |
2016 | + } |
2017 | ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, |
2018 | CCP_PASSTHRU_BYTESWAP_256BIT); |
2019 | if (ret) { |
2020 | diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c |
2021 | index 7309c0824887..a2449d77af07 100644 |
2022 | --- a/drivers/devfreq/devfreq.c |
2023 | +++ b/drivers/devfreq/devfreq.c |
2024 | @@ -574,7 +574,7 @@ struct devfreq *devfreq_add_device(struct device *dev, |
2025 | err = device_register(&devfreq->dev); |
2026 | if (err) { |
2027 | mutex_unlock(&devfreq->lock); |
2028 | - goto err_out; |
2029 | + goto err_dev; |
2030 | } |
2031 | |
2032 | devfreq->trans_table = devm_kzalloc(&devfreq->dev, sizeof(unsigned int) * |
2033 | @@ -618,6 +618,9 @@ struct devfreq *devfreq_add_device(struct device *dev, |
2034 | mutex_unlock(&devfreq_list_lock); |
2035 | |
2036 | device_unregister(&devfreq->dev); |
2037 | +err_dev: |
2038 | + if (devfreq) |
2039 | + kfree(devfreq); |
2040 | err_out: |
2041 | return ERR_PTR(err); |
2042 | } |
2043 | diff --git a/drivers/gpu/drm/sun4i/sun4i_drv.c b/drivers/gpu/drm/sun4i/sun4i_drv.c |
2044 | index c3b21865443e..1feec34ca9dd 100644 |
2045 | --- a/drivers/gpu/drm/sun4i/sun4i_drv.c |
2046 | +++ b/drivers/gpu/drm/sun4i/sun4i_drv.c |
2047 | @@ -47,6 +47,13 @@ static void sun4i_drv_disable_vblank(struct drm_device *drm, unsigned int pipe) |
2048 | sun4i_tcon_enable_vblank(tcon, false); |
2049 | } |
2050 | |
2051 | +static void sun4i_drv_lastclose(struct drm_device *dev) |
2052 | +{ |
2053 | + struct sun4i_drv *drv = dev->dev_private; |
2054 | + |
2055 | + drm_fbdev_cma_restore_mode(drv->fbdev); |
2056 | +} |
2057 | + |
2058 | static const struct file_operations sun4i_drv_fops = { |
2059 | .owner = THIS_MODULE, |
2060 | .open = drm_open, |
2061 | @@ -65,6 +72,7 @@ static struct drm_driver sun4i_drv_driver = { |
2062 | .driver_features = DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_ATOMIC, |
2063 | |
2064 | /* Generic Operations */ |
2065 | + .lastclose = sun4i_drv_lastclose, |
2066 | .fops = &sun4i_drv_fops, |
2067 | .name = "sun4i-drm", |
2068 | .desc = "Allwinner sun4i Display Engine", |
2069 | diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c |
2070 | index 63e82f8e8308..fb4ce0394ac7 100644 |
2071 | --- a/drivers/infiniband/core/addr.c |
2072 | +++ b/drivers/infiniband/core/addr.c |
2073 | @@ -446,15 +446,10 @@ static int addr6_resolve(struct sockaddr_in6 *src_in, |
2074 | |
2075 | ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6); |
2076 | if (ret < 0) |
2077 | - goto put; |
2078 | + return ret; |
2079 | |
2080 | rt = (struct rt6_info *)dst; |
2081 | - if (ipv6_addr_any(&fl6.saddr)) { |
2082 | - ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev, |
2083 | - &fl6.daddr, 0, &fl6.saddr); |
2084 | - if (ret) |
2085 | - goto put; |
2086 | - |
2087 | + if (ipv6_addr_any(&src_in->sin6_addr)) { |
2088 | src_in->sin6_family = AF_INET6; |
2089 | src_in->sin6_addr = fl6.saddr; |
2090 | } |
2091 | @@ -471,9 +466,6 @@ static int addr6_resolve(struct sockaddr_in6 *src_in, |
2092 | |
2093 | *pdst = dst; |
2094 | return 0; |
2095 | -put: |
2096 | - dst_release(dst); |
2097 | - return ret; |
2098 | } |
2099 | #else |
2100 | static int addr6_resolve(struct sockaddr_in6 *src_in, |
2101 | diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c |
2102 | index 4bd5b5caa243..613074e963bb 100644 |
2103 | --- a/drivers/infiniband/hw/hfi1/rc.c |
2104 | +++ b/drivers/infiniband/hw/hfi1/rc.c |
2105 | @@ -551,7 +551,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) |
2106 | case IB_WR_RDMA_WRITE: |
2107 | if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) |
2108 | qp->s_lsn++; |
2109 | - /* FALLTHROUGH */ |
2110 | + goto no_flow_control; |
2111 | case IB_WR_RDMA_WRITE_WITH_IMM: |
2112 | /* If no credit, return. */ |
2113 | if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT) && |
2114 | @@ -559,6 +559,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) |
2115 | qp->s_flags |= RVT_S_WAIT_SSN_CREDIT; |
2116 | goto bail; |
2117 | } |
2118 | +no_flow_control: |
2119 | put_ib_reth_vaddr( |
2120 | wqe->rdma_wr.remote_addr, |
2121 | &ohdr->u.rc.reth); |
2122 | diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c |
2123 | index f3fe787c9426..c1523f9a3c12 100644 |
2124 | --- a/drivers/infiniband/hw/qib/qib_rc.c |
2125 | +++ b/drivers/infiniband/hw/qib/qib_rc.c |
2126 | @@ -357,7 +357,7 @@ int qib_make_rc_req(struct rvt_qp *qp, unsigned long *flags) |
2127 | case IB_WR_RDMA_WRITE: |
2128 | if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) |
2129 | qp->s_lsn++; |
2130 | - /* FALLTHROUGH */ |
2131 | + goto no_flow_control; |
2132 | case IB_WR_RDMA_WRITE_WITH_IMM: |
2133 | /* If no credit, return. */ |
2134 | if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT) && |
2135 | @@ -365,7 +365,7 @@ int qib_make_rc_req(struct rvt_qp *qp, unsigned long *flags) |
2136 | qp->s_flags |= RVT_S_WAIT_SSN_CREDIT; |
2137 | goto bail; |
2138 | } |
2139 | - |
2140 | +no_flow_control: |
2141 | ohdr->u.rc.reth.vaddr = |
2142 | cpu_to_be64(wqe->rdma_wr.remote_addr); |
2143 | ohdr->u.rc.reth.rkey = |
2144 | diff --git a/drivers/input/serio/i8042-x86ia64io.h b/drivers/input/serio/i8042-x86ia64io.h |
2145 | index 5be14ad29d46..dbf09836ff30 100644 |
2146 | --- a/drivers/input/serio/i8042-x86ia64io.h |
2147 | +++ b/drivers/input/serio/i8042-x86ia64io.h |
2148 | @@ -904,6 +904,13 @@ static const struct dmi_system_id __initconst i8042_dmi_kbdreset_table[] = { |
2149 | DMI_MATCH(DMI_PRODUCT_NAME, "P34"), |
2150 | }, |
2151 | }, |
2152 | + { |
2153 | + /* Gigabyte P57 - Elantech touchpad */ |
2154 | + .matches = { |
2155 | + DMI_MATCH(DMI_SYS_VENDOR, "GIGABYTE"), |
2156 | + DMI_MATCH(DMI_PRODUCT_NAME, "P57"), |
2157 | + }, |
2158 | + }, |
2159 | { |
2160 | /* Schenker XMG C504 - Elantech touchpad */ |
2161 | .matches = { |
2162 | diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h |
2163 | index c3ea03c9a1a8..02619cabda8b 100644 |
2164 | --- a/drivers/md/bcache/bcache.h |
2165 | +++ b/drivers/md/bcache/bcache.h |
2166 | @@ -333,6 +333,7 @@ struct cached_dev { |
2167 | /* Limit number of writeback bios in flight */ |
2168 | struct semaphore in_flight; |
2169 | struct task_struct *writeback_thread; |
2170 | + struct workqueue_struct *writeback_write_wq; |
2171 | |
2172 | struct keybuf writeback_keys; |
2173 | |
2174 | diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c |
2175 | index a37c1776f2e3..e0f1c6d534fe 100644 |
2176 | --- a/drivers/md/bcache/request.c |
2177 | +++ b/drivers/md/bcache/request.c |
2178 | @@ -196,12 +196,12 @@ static void bch_data_insert_start(struct closure *cl) |
2179 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
2180 | struct bio *bio = op->bio, *n; |
2181 | |
2182 | - if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0) |
2183 | - wake_up_gc(op->c); |
2184 | - |
2185 | if (op->bypass) |
2186 | return bch_data_invalidate(cl); |
2187 | |
2188 | + if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0) |
2189 | + wake_up_gc(op->c); |
2190 | + |
2191 | /* |
2192 | * Journal writes are marked REQ_PREFLUSH; if the original write was a |
2193 | * flush, it'll wait on the journal write. |
2194 | diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c |
2195 | index 66669c8f4161..f4557f558b24 100644 |
2196 | --- a/drivers/md/bcache/super.c |
2197 | +++ b/drivers/md/bcache/super.c |
2198 | @@ -1025,7 +1025,7 @@ int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c) |
2199 | } |
2200 | |
2201 | if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) { |
2202 | - bch_sectors_dirty_init(dc); |
2203 | + bch_sectors_dirty_init(&dc->disk); |
2204 | atomic_set(&dc->has_dirty, 1); |
2205 | atomic_inc(&dc->count); |
2206 | bch_writeback_queue(dc); |
2207 | @@ -1058,6 +1058,8 @@ static void cached_dev_free(struct closure *cl) |
2208 | cancel_delayed_work_sync(&dc->writeback_rate_update); |
2209 | if (!IS_ERR_OR_NULL(dc->writeback_thread)) |
2210 | kthread_stop(dc->writeback_thread); |
2211 | + if (dc->writeback_write_wq) |
2212 | + destroy_workqueue(dc->writeback_write_wq); |
2213 | |
2214 | mutex_lock(&bch_register_lock); |
2215 | |
2216 | @@ -1229,6 +1231,7 @@ static int flash_dev_run(struct cache_set *c, struct uuid_entry *u) |
2217 | goto err; |
2218 | |
2219 | bcache_device_attach(d, c, u - c->uuids); |
2220 | + bch_sectors_dirty_init(d); |
2221 | bch_flash_dev_request_init(d); |
2222 | add_disk(d->disk); |
2223 | |
2224 | @@ -1967,6 +1970,8 @@ static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr, |
2225 | else |
2226 | err = "device busy"; |
2227 | mutex_unlock(&bch_register_lock); |
2228 | + if (!IS_ERR(bdev)) |
2229 | + bdput(bdev); |
2230 | if (attr == &ksysfs_register_quiet) |
2231 | goto out; |
2232 | } |
2233 | diff --git a/drivers/md/bcache/sysfs.c b/drivers/md/bcache/sysfs.c |
2234 | index b3ff57d61dde..4fbb5532f24c 100644 |
2235 | --- a/drivers/md/bcache/sysfs.c |
2236 | +++ b/drivers/md/bcache/sysfs.c |
2237 | @@ -191,7 +191,7 @@ STORE(__cached_dev) |
2238 | { |
2239 | struct cached_dev *dc = container_of(kobj, struct cached_dev, |
2240 | disk.kobj); |
2241 | - unsigned v = size; |
2242 | + ssize_t v = size; |
2243 | struct cache_set *c; |
2244 | struct kobj_uevent_env *env; |
2245 | |
2246 | @@ -226,7 +226,7 @@ STORE(__cached_dev) |
2247 | bch_cached_dev_run(dc); |
2248 | |
2249 | if (attr == &sysfs_cache_mode) { |
2250 | - ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1); |
2251 | + v = bch_read_string_list(buf, bch_cache_modes + 1); |
2252 | |
2253 | if (v < 0) |
2254 | return v; |
2255 | diff --git a/drivers/md/bcache/util.c b/drivers/md/bcache/util.c |
2256 | index dde6172f3f10..eb70f6894f05 100644 |
2257 | --- a/drivers/md/bcache/util.c |
2258 | +++ b/drivers/md/bcache/util.c |
2259 | @@ -73,24 +73,44 @@ STRTO_H(strtouint, unsigned int) |
2260 | STRTO_H(strtoll, long long) |
2261 | STRTO_H(strtoull, unsigned long long) |
2262 | |
2263 | +/** |
2264 | + * bch_hprint() - formats @v to human readable string for sysfs. |
2265 | + * |
2266 | + * @v - signed 64 bit integer |
2267 | + * @buf - the (at least 8 byte) buffer to format the result into. |
2268 | + * |
2269 | + * Returns the number of bytes used by format. |
2270 | + */ |
2271 | ssize_t bch_hprint(char *buf, int64_t v) |
2272 | { |
2273 | static const char units[] = "?kMGTPEZY"; |
2274 | - char dec[4] = ""; |
2275 | - int u, t = 0; |
2276 | - |
2277 | - for (u = 0; v >= 1024 || v <= -1024; u++) { |
2278 | - t = v & ~(~0 << 10); |
2279 | - v >>= 10; |
2280 | - } |
2281 | - |
2282 | - if (!u) |
2283 | - return sprintf(buf, "%llu", v); |
2284 | - |
2285 | - if (v < 100 && v > -100) |
2286 | - snprintf(dec, sizeof(dec), ".%i", t / 100); |
2287 | - |
2288 | - return sprintf(buf, "%lli%s%c", v, dec, units[u]); |
2289 | + int u = 0, t; |
2290 | + |
2291 | + uint64_t q; |
2292 | + |
2293 | + if (v < 0) |
2294 | + q = -v; |
2295 | + else |
2296 | + q = v; |
2297 | + |
2298 | + /* For as long as the number is more than 3 digits, but at least |
2299 | + * once, shift right / divide by 1024. Keep the remainder for |
2300 | + * a digit after the decimal point. |
2301 | + */ |
2302 | + do { |
2303 | + u++; |
2304 | + |
2305 | + t = q & ~(~0 << 10); |
2306 | + q >>= 10; |
2307 | + } while (q >= 1000); |
2308 | + |
2309 | + if (v < 0) |
2310 | + /* '-', up to 3 digits, '.', 1 digit, 1 character, null; |
2311 | + * yields 8 bytes. |
2312 | + */ |
2313 | + return sprintf(buf, "-%llu.%i%c", q, t * 10 / 1024, units[u]); |
2314 | + else |
2315 | + return sprintf(buf, "%llu.%i%c", q, t * 10 / 1024, units[u]); |
2316 | } |
2317 | |
2318 | ssize_t bch_snprint_string_list(char *buf, size_t size, const char * const list[], |
2319 | diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c |
2320 | index e51644e503a5..4ce2b19fe120 100644 |
2321 | --- a/drivers/md/bcache/writeback.c |
2322 | +++ b/drivers/md/bcache/writeback.c |
2323 | @@ -20,7 +20,8 @@ |
2324 | static void __update_writeback_rate(struct cached_dev *dc) |
2325 | { |
2326 | struct cache_set *c = dc->disk.c; |
2327 | - uint64_t cache_sectors = c->nbuckets * c->sb.bucket_size; |
2328 | + uint64_t cache_sectors = c->nbuckets * c->sb.bucket_size - |
2329 | + bcache_flash_devs_sectors_dirty(c); |
2330 | uint64_t cache_dirty_target = |
2331 | div_u64(cache_sectors * dc->writeback_percent, 100); |
2332 | |
2333 | @@ -186,7 +187,7 @@ static void write_dirty(struct closure *cl) |
2334 | |
2335 | closure_bio_submit(&io->bio, cl); |
2336 | |
2337 | - continue_at(cl, write_dirty_finish, system_wq); |
2338 | + continue_at(cl, write_dirty_finish, io->dc->writeback_write_wq); |
2339 | } |
2340 | |
2341 | static void read_dirty_endio(struct bio *bio) |
2342 | @@ -206,7 +207,7 @@ static void read_dirty_submit(struct closure *cl) |
2343 | |
2344 | closure_bio_submit(&io->bio, cl); |
2345 | |
2346 | - continue_at(cl, write_dirty, system_wq); |
2347 | + continue_at(cl, write_dirty, io->dc->writeback_write_wq); |
2348 | } |
2349 | |
2350 | static void read_dirty(struct cached_dev *dc) |
2351 | @@ -482,17 +483,17 @@ static int sectors_dirty_init_fn(struct btree_op *_op, struct btree *b, |
2352 | return MAP_CONTINUE; |
2353 | } |
2354 | |
2355 | -void bch_sectors_dirty_init(struct cached_dev *dc) |
2356 | +void bch_sectors_dirty_init(struct bcache_device *d) |
2357 | { |
2358 | struct sectors_dirty_init op; |
2359 | |
2360 | bch_btree_op_init(&op.op, -1); |
2361 | - op.inode = dc->disk.id; |
2362 | + op.inode = d->id; |
2363 | |
2364 | - bch_btree_map_keys(&op.op, dc->disk.c, &KEY(op.inode, 0, 0), |
2365 | + bch_btree_map_keys(&op.op, d->c, &KEY(op.inode, 0, 0), |
2366 | sectors_dirty_init_fn, 0); |
2367 | |
2368 | - dc->disk.sectors_dirty_last = bcache_dev_sectors_dirty(&dc->disk); |
2369 | + d->sectors_dirty_last = bcache_dev_sectors_dirty(d); |
2370 | } |
2371 | |
2372 | void bch_cached_dev_writeback_init(struct cached_dev *dc) |
2373 | @@ -516,6 +517,11 @@ void bch_cached_dev_writeback_init(struct cached_dev *dc) |
2374 | |
2375 | int bch_cached_dev_writeback_start(struct cached_dev *dc) |
2376 | { |
2377 | + dc->writeback_write_wq = alloc_workqueue("bcache_writeback_wq", |
2378 | + WQ_MEM_RECLAIM, 0); |
2379 | + if (!dc->writeback_write_wq) |
2380 | + return -ENOMEM; |
2381 | + |
2382 | dc->writeback_thread = kthread_create(bch_writeback_thread, dc, |
2383 | "bcache_writeback"); |
2384 | if (IS_ERR(dc->writeback_thread)) |
2385 | diff --git a/drivers/md/bcache/writeback.h b/drivers/md/bcache/writeback.h |
2386 | index 301eaf565167..cdf8d253209e 100644 |
2387 | --- a/drivers/md/bcache/writeback.h |
2388 | +++ b/drivers/md/bcache/writeback.h |
2389 | @@ -14,6 +14,25 @@ static inline uint64_t bcache_dev_sectors_dirty(struct bcache_device *d) |
2390 | return ret; |
2391 | } |
2392 | |
2393 | +static inline uint64_t bcache_flash_devs_sectors_dirty(struct cache_set *c) |
2394 | +{ |
2395 | + uint64_t i, ret = 0; |
2396 | + |
2397 | + mutex_lock(&bch_register_lock); |
2398 | + |
2399 | + for (i = 0; i < c->nr_uuids; i++) { |
2400 | + struct bcache_device *d = c->devices[i]; |
2401 | + |
2402 | + if (!d || !UUID_FLASH_ONLY(&c->uuids[i])) |
2403 | + continue; |
2404 | + ret += bcache_dev_sectors_dirty(d); |
2405 | + } |
2406 | + |
2407 | + mutex_unlock(&bch_register_lock); |
2408 | + |
2409 | + return ret; |
2410 | +} |
2411 | + |
2412 | static inline unsigned offset_to_stripe(struct bcache_device *d, |
2413 | uint64_t offset) |
2414 | { |
2415 | @@ -85,7 +104,7 @@ static inline void bch_writeback_add(struct cached_dev *dc) |
2416 | |
2417 | void bcache_dev_sectors_dirty_add(struct cache_set *, unsigned, uint64_t, int); |
2418 | |
2419 | -void bch_sectors_dirty_init(struct cached_dev *dc); |
2420 | +void bch_sectors_dirty_init(struct bcache_device *); |
2421 | void bch_cached_dev_writeback_init(struct cached_dev *); |
2422 | int bch_cached_dev_writeback_start(struct cached_dev *); |
2423 | |
2424 | diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c |
2425 | index 2d826927a3bf..fb02c3979bf4 100644 |
2426 | --- a/drivers/md/bitmap.c |
2427 | +++ b/drivers/md/bitmap.c |
2428 | @@ -1992,6 +1992,11 @@ int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
2429 | long pages; |
2430 | struct bitmap_page *new_bp; |
2431 | |
2432 | + if (bitmap->storage.file && !init) { |
2433 | + pr_info("md: cannot resize file-based bitmap\n"); |
2434 | + return -EINVAL; |
2435 | + } |
2436 | + |
2437 | if (chunksize == 0) { |
2438 | /* If there is enough space, leave the chunk size unchanged, |
2439 | * else increase by factor of two until there is enough space. |
2440 | diff --git a/drivers/media/usb/uvc/uvc_ctrl.c b/drivers/media/usb/uvc/uvc_ctrl.c |
2441 | index c2ee6e39fd0c..20397aba6849 100644 |
2442 | --- a/drivers/media/usb/uvc/uvc_ctrl.c |
2443 | +++ b/drivers/media/usb/uvc/uvc_ctrl.c |
2444 | @@ -2002,6 +2002,13 @@ int uvc_ctrl_add_mapping(struct uvc_video_chain *chain, |
2445 | goto done; |
2446 | } |
2447 | |
2448 | + /* Validate the user-provided bit-size and offset */ |
2449 | + if (mapping->size > 32 || |
2450 | + mapping->offset + mapping->size > ctrl->info.size * 8) { |
2451 | + ret = -EINVAL; |
2452 | + goto done; |
2453 | + } |
2454 | + |
2455 | list_for_each_entry(map, &ctrl->info.mappings, list) { |
2456 | if (mapping->id == map->id) { |
2457 | uvc_trace(UVC_TRACE_CONTROL, "Can't add mapping '%s', " |
2458 | diff --git a/drivers/media/v4l2-core/v4l2-compat-ioctl32.c b/drivers/media/v4l2-core/v4l2-compat-ioctl32.c |
2459 | index bacecbd68a6d..dc51dd86377d 100644 |
2460 | --- a/drivers/media/v4l2-core/v4l2-compat-ioctl32.c |
2461 | +++ b/drivers/media/v4l2-core/v4l2-compat-ioctl32.c |
2462 | @@ -773,7 +773,8 @@ static int put_v4l2_event32(struct v4l2_event *kp, struct v4l2_event32 __user *u |
2463 | copy_to_user(&up->u, &kp->u, sizeof(kp->u)) || |
2464 | put_user(kp->pending, &up->pending) || |
2465 | put_user(kp->sequence, &up->sequence) || |
2466 | - compat_put_timespec(&kp->timestamp, &up->timestamp) || |
2467 | + put_user(kp->timestamp.tv_sec, &up->timestamp.tv_sec) || |
2468 | + put_user(kp->timestamp.tv_nsec, &up->timestamp.tv_nsec) || |
2469 | put_user(kp->id, &up->id) || |
2470 | copy_to_user(up->reserved, kp->reserved, 8 * sizeof(__u32))) |
2471 | return -EFAULT; |
2472 | diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c |
2473 | index b57fc6d6e28a..d08dfc8b9ba9 100644 |
2474 | --- a/drivers/pci/hotplug/pciehp_hpc.c |
2475 | +++ b/drivers/pci/hotplug/pciehp_hpc.c |
2476 | @@ -586,6 +586,14 @@ static irqreturn_t pciehp_isr(int irq, void *dev_id) |
2477 | events = status & (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | |
2478 | PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_CC | |
2479 | PCI_EXP_SLTSTA_DLLSC); |
2480 | + |
2481 | + /* |
2482 | + * If we've already reported a power fault, don't report it again |
2483 | + * until we've done something to handle it. |
2484 | + */ |
2485 | + if (ctrl->power_fault_detected) |
2486 | + events &= ~PCI_EXP_SLTSTA_PFD; |
2487 | + |
2488 | if (!events) |
2489 | return IRQ_NONE; |
2490 | |
2491 | diff --git a/drivers/pci/hotplug/shpchp_hpc.c b/drivers/pci/hotplug/shpchp_hpc.c |
2492 | index de0ea474fb73..e5824c7b7b6b 100644 |
2493 | --- a/drivers/pci/hotplug/shpchp_hpc.c |
2494 | +++ b/drivers/pci/hotplug/shpchp_hpc.c |
2495 | @@ -1062,6 +1062,8 @@ int shpc_init(struct controller *ctrl, struct pci_dev *pdev) |
2496 | if (rc) { |
2497 | ctrl_info(ctrl, "Can't get msi for the hotplug controller\n"); |
2498 | ctrl_info(ctrl, "Use INTx for the hotplug controller\n"); |
2499 | + } else { |
2500 | + pci_set_master(pdev); |
2501 | } |
2502 | |
2503 | rc = request_irq(ctrl->pci_dev->irq, shpc_isr, IRQF_SHARED, |
2504 | diff --git a/drivers/pinctrl/pinctrl-amd.c b/drivers/pinctrl/pinctrl-amd.c |
2505 | index c9a146948192..a5b7bd3c9bac 100644 |
2506 | --- a/drivers/pinctrl/pinctrl-amd.c |
2507 | +++ b/drivers/pinctrl/pinctrl-amd.c |
2508 | @@ -32,6 +32,7 @@ |
2509 | #include <linux/pinctrl/pinconf.h> |
2510 | #include <linux/pinctrl/pinconf-generic.h> |
2511 | |
2512 | +#include "core.h" |
2513 | #include "pinctrl-utils.h" |
2514 | #include "pinctrl-amd.h" |
2515 | |
2516 | @@ -712,6 +713,69 @@ static const struct pinconf_ops amd_pinconf_ops = { |
2517 | .pin_config_group_set = amd_pinconf_group_set, |
2518 | }; |
2519 | |
2520 | +#ifdef CONFIG_PM_SLEEP |
2521 | +static bool amd_gpio_should_save(struct amd_gpio *gpio_dev, unsigned int pin) |
2522 | +{ |
2523 | + const struct pin_desc *pd = pin_desc_get(gpio_dev->pctrl, pin); |
2524 | + |
2525 | + if (!pd) |
2526 | + return false; |
2527 | + |
2528 | + /* |
2529 | + * Only restore the pin if it is actually in use by the kernel (or |
2530 | + * by userspace). |
2531 | + */ |
2532 | + if (pd->mux_owner || pd->gpio_owner || |
2533 | + gpiochip_line_is_irq(&gpio_dev->gc, pin)) |
2534 | + return true; |
2535 | + |
2536 | + return false; |
2537 | +} |
2538 | + |
2539 | +int amd_gpio_suspend(struct device *dev) |
2540 | +{ |
2541 | + struct platform_device *pdev = to_platform_device(dev); |
2542 | + struct amd_gpio *gpio_dev = platform_get_drvdata(pdev); |
2543 | + struct pinctrl_desc *desc = gpio_dev->pctrl->desc; |
2544 | + int i; |
2545 | + |
2546 | + for (i = 0; i < desc->npins; i++) { |
2547 | + int pin = desc->pins[i].number; |
2548 | + |
2549 | + if (!amd_gpio_should_save(gpio_dev, pin)) |
2550 | + continue; |
2551 | + |
2552 | + gpio_dev->saved_regs[i] = readl(gpio_dev->base + pin*4); |
2553 | + } |
2554 | + |
2555 | + return 0; |
2556 | +} |
2557 | + |
2558 | +int amd_gpio_resume(struct device *dev) |
2559 | +{ |
2560 | + struct platform_device *pdev = to_platform_device(dev); |
2561 | + struct amd_gpio *gpio_dev = platform_get_drvdata(pdev); |
2562 | + struct pinctrl_desc *desc = gpio_dev->pctrl->desc; |
2563 | + int i; |
2564 | + |
2565 | + for (i = 0; i < desc->npins; i++) { |
2566 | + int pin = desc->pins[i].number; |
2567 | + |
2568 | + if (!amd_gpio_should_save(gpio_dev, pin)) |
2569 | + continue; |
2570 | + |
2571 | + writel(gpio_dev->saved_regs[i], gpio_dev->base + pin*4); |
2572 | + } |
2573 | + |
2574 | + return 0; |
2575 | +} |
2576 | + |
2577 | +static const struct dev_pm_ops amd_gpio_pm_ops = { |
2578 | + SET_LATE_SYSTEM_SLEEP_PM_OPS(amd_gpio_suspend, |
2579 | + amd_gpio_resume) |
2580 | +}; |
2581 | +#endif |
2582 | + |
2583 | static struct pinctrl_desc amd_pinctrl_desc = { |
2584 | .pins = kerncz_pins, |
2585 | .npins = ARRAY_SIZE(kerncz_pins), |
2586 | @@ -751,6 +815,14 @@ static int amd_gpio_probe(struct platform_device *pdev) |
2587 | return -EINVAL; |
2588 | } |
2589 | |
2590 | +#ifdef CONFIG_PM_SLEEP |
2591 | + gpio_dev->saved_regs = devm_kcalloc(&pdev->dev, amd_pinctrl_desc.npins, |
2592 | + sizeof(*gpio_dev->saved_regs), |
2593 | + GFP_KERNEL); |
2594 | + if (!gpio_dev->saved_regs) |
2595 | + return -ENOMEM; |
2596 | +#endif |
2597 | + |
2598 | gpio_dev->pdev = pdev; |
2599 | gpio_dev->gc.direction_input = amd_gpio_direction_input; |
2600 | gpio_dev->gc.direction_output = amd_gpio_direction_output; |
2601 | @@ -839,6 +911,9 @@ static struct platform_driver amd_gpio_driver = { |
2602 | .driver = { |
2603 | .name = "amd_gpio", |
2604 | .acpi_match_table = ACPI_PTR(amd_gpio_acpi_match), |
2605 | +#ifdef CONFIG_PM_SLEEP |
2606 | + .pm = &amd_gpio_pm_ops, |
2607 | +#endif |
2608 | }, |
2609 | .probe = amd_gpio_probe, |
2610 | .remove = amd_gpio_remove, |
2611 | diff --git a/drivers/pinctrl/pinctrl-amd.h b/drivers/pinctrl/pinctrl-amd.h |
2612 | index 7bfea47dbb47..e8bbb20779d0 100644 |
2613 | --- a/drivers/pinctrl/pinctrl-amd.h |
2614 | +++ b/drivers/pinctrl/pinctrl-amd.h |
2615 | @@ -95,6 +95,7 @@ struct amd_gpio { |
2616 | struct gpio_chip gc; |
2617 | struct resource *res; |
2618 | struct platform_device *pdev; |
2619 | + u32 *saved_regs; |
2620 | }; |
2621 | |
2622 | /* KERNCZ configuration*/ |
2623 | diff --git a/drivers/s390/scsi/zfcp_dbf.c b/drivers/s390/scsi/zfcp_dbf.c |
2624 | index d5bf36ec8a75..34367d172961 100644 |
2625 | --- a/drivers/s390/scsi/zfcp_dbf.c |
2626 | +++ b/drivers/s390/scsi/zfcp_dbf.c |
2627 | @@ -3,7 +3,7 @@ |
2628 | * |
2629 | * Debug traces for zfcp. |
2630 | * |
2631 | - * Copyright IBM Corp. 2002, 2016 |
2632 | + * Copyright IBM Corp. 2002, 2017 |
2633 | */ |
2634 | |
2635 | #define KMSG_COMPONENT "zfcp" |
2636 | @@ -447,6 +447,7 @@ static u16 zfcp_dbf_san_res_cap_len_if_gpn_ft(char *tag, |
2637 | struct fc_ct_hdr *reqh = sg_virt(ct_els->req); |
2638 | struct fc_ns_gid_ft *reqn = (struct fc_ns_gid_ft *)(reqh + 1); |
2639 | struct scatterlist *resp_entry = ct_els->resp; |
2640 | + struct fc_ct_hdr *resph; |
2641 | struct fc_gpn_ft_resp *acc; |
2642 | int max_entries, x, last = 0; |
2643 | |
2644 | @@ -473,6 +474,13 @@ static u16 zfcp_dbf_san_res_cap_len_if_gpn_ft(char *tag, |
2645 | return len; /* not GPN_FT response so do not cap */ |
2646 | |
2647 | acc = sg_virt(resp_entry); |
2648 | + |
2649 | + /* cap all but accept CT responses to at least the CT header */ |
2650 | + resph = (struct fc_ct_hdr *)acc; |
2651 | + if ((ct_els->status) || |
2652 | + (resph->ct_cmd != cpu_to_be16(FC_FS_ACC))) |
2653 | + return max(FC_CT_HDR_LEN, ZFCP_DBF_SAN_MAX_PAYLOAD); |
2654 | + |
2655 | max_entries = (reqh->ct_mr_size * 4 / sizeof(struct fc_gpn_ft_resp)) |
2656 | + 1 /* zfcp_fc_scan_ports: bytes correct, entries off-by-one |
2657 | * to account for header as 1st pseudo "entry" */; |
2658 | @@ -555,8 +563,8 @@ void zfcp_dbf_scsi(char *tag, int level, struct scsi_cmnd *sc, |
2659 | rec->scsi_retries = sc->retries; |
2660 | rec->scsi_allowed = sc->allowed; |
2661 | rec->scsi_id = sc->device->id; |
2662 | - /* struct zfcp_dbf_scsi needs to be updated to handle 64bit LUNs */ |
2663 | rec->scsi_lun = (u32)sc->device->lun; |
2664 | + rec->scsi_lun_64_hi = (u32)(sc->device->lun >> 32); |
2665 | rec->host_scribble = (unsigned long)sc->host_scribble; |
2666 | |
2667 | memcpy(rec->scsi_opcode, sc->cmnd, |
2668 | @@ -564,19 +572,32 @@ void zfcp_dbf_scsi(char *tag, int level, struct scsi_cmnd *sc, |
2669 | |
2670 | if (fsf) { |
2671 | rec->fsf_req_id = fsf->req_id; |
2672 | + rec->pl_len = FCP_RESP_WITH_EXT; |
2673 | fcp_rsp = (struct fcp_resp_with_ext *) |
2674 | &(fsf->qtcb->bottom.io.fcp_rsp); |
2675 | + /* mandatory parts of FCP_RSP IU in this SCSI record */ |
2676 | memcpy(&rec->fcp_rsp, fcp_rsp, FCP_RESP_WITH_EXT); |
2677 | if (fcp_rsp->resp.fr_flags & FCP_RSP_LEN_VAL) { |
2678 | fcp_rsp_info = (struct fcp_resp_rsp_info *) &fcp_rsp[1]; |
2679 | rec->fcp_rsp_info = fcp_rsp_info->rsp_code; |
2680 | + rec->pl_len += be32_to_cpu(fcp_rsp->ext.fr_rsp_len); |
2681 | } |
2682 | if (fcp_rsp->resp.fr_flags & FCP_SNS_LEN_VAL) { |
2683 | - rec->pl_len = min((u16)SCSI_SENSE_BUFFERSIZE, |
2684 | - (u16)ZFCP_DBF_PAY_MAX_REC); |
2685 | - zfcp_dbf_pl_write(dbf, sc->sense_buffer, rec->pl_len, |
2686 | - "fcp_sns", fsf->req_id); |
2687 | + rec->pl_len += be32_to_cpu(fcp_rsp->ext.fr_sns_len); |
2688 | } |
2689 | + /* complete FCP_RSP IU in associated PAYload record |
2690 | + * but only if there are optional parts |
2691 | + */ |
2692 | + if (fcp_rsp->resp.fr_flags != 0) |
2693 | + zfcp_dbf_pl_write( |
2694 | + dbf, fcp_rsp, |
2695 | + /* at least one full PAY record |
2696 | + * but not beyond hardware response field |
2697 | + */ |
2698 | + min_t(u16, max_t(u16, rec->pl_len, |
2699 | + ZFCP_DBF_PAY_MAX_REC), |
2700 | + FSF_FCP_RSP_SIZE), |
2701 | + "fcp_riu", fsf->req_id); |
2702 | } |
2703 | |
2704 | debug_event(dbf->scsi, level, rec, sizeof(*rec)); |
2705 | diff --git a/drivers/s390/scsi/zfcp_dbf.h b/drivers/s390/scsi/zfcp_dbf.h |
2706 | index db186d44cfaf..b60667c145fd 100644 |
2707 | --- a/drivers/s390/scsi/zfcp_dbf.h |
2708 | +++ b/drivers/s390/scsi/zfcp_dbf.h |
2709 | @@ -2,7 +2,7 @@ |
2710 | * zfcp device driver |
2711 | * debug feature declarations |
2712 | * |
2713 | - * Copyright IBM Corp. 2008, 2016 |
2714 | + * Copyright IBM Corp. 2008, 2017 |
2715 | */ |
2716 | |
2717 | #ifndef ZFCP_DBF_H |
2718 | @@ -204,7 +204,7 @@ enum zfcp_dbf_scsi_id { |
2719 | * @id: unique number of recovery record type |
2720 | * @tag: identifier string specifying the location of initiation |
2721 | * @scsi_id: scsi device id |
2722 | - * @scsi_lun: scsi device logical unit number |
2723 | + * @scsi_lun: scsi device logical unit number, low part of 64 bit, old 32 bit |
2724 | * @scsi_result: scsi result |
2725 | * @scsi_retries: current retry number of scsi request |
2726 | * @scsi_allowed: allowed retries |
2727 | @@ -214,6 +214,7 @@ enum zfcp_dbf_scsi_id { |
2728 | * @host_scribble: LLD specific data attached to SCSI request |
2729 | * @pl_len: length of paload stored as zfcp_dbf_pay |
2730 | * @fsf_rsp: response for fsf request |
2731 | + * @scsi_lun_64_hi: scsi device logical unit number, high part of 64 bit |
2732 | */ |
2733 | struct zfcp_dbf_scsi { |
2734 | u8 id; |
2735 | @@ -230,6 +231,7 @@ struct zfcp_dbf_scsi { |
2736 | u64 host_scribble; |
2737 | u16 pl_len; |
2738 | struct fcp_resp_with_ext fcp_rsp; |
2739 | + u32 scsi_lun_64_hi; |
2740 | } __packed; |
2741 | |
2742 | /** |
2743 | @@ -323,7 +325,11 @@ void zfcp_dbf_hba_fsf_response(struct zfcp_fsf_req *req) |
2744 | { |
2745 | struct fsf_qtcb *qtcb = req->qtcb; |
2746 | |
2747 | - if ((qtcb->prefix.prot_status != FSF_PROT_GOOD) && |
2748 | + if (unlikely(req->status & (ZFCP_STATUS_FSFREQ_DISMISSED | |
2749 | + ZFCP_STATUS_FSFREQ_ERROR))) { |
2750 | + zfcp_dbf_hba_fsf_resp("fs_rerr", 3, req); |
2751 | + |
2752 | + } else if ((qtcb->prefix.prot_status != FSF_PROT_GOOD) && |
2753 | (qtcb->prefix.prot_status != FSF_PROT_FSF_STATUS_PRESENTED)) { |
2754 | zfcp_dbf_hba_fsf_resp("fs_perr", 1, req); |
2755 | |
2756 | @@ -401,7 +407,8 @@ void zfcp_dbf_scsi_abort(char *tag, struct scsi_cmnd *scmd, |
2757 | * @flag: indicates type of reset (Target Reset, Logical Unit Reset) |
2758 | */ |
2759 | static inline |
2760 | -void zfcp_dbf_scsi_devreset(char *tag, struct scsi_cmnd *scmnd, u8 flag) |
2761 | +void zfcp_dbf_scsi_devreset(char *tag, struct scsi_cmnd *scmnd, u8 flag, |
2762 | + struct zfcp_fsf_req *fsf_req) |
2763 | { |
2764 | char tmp_tag[ZFCP_DBF_TAG_LEN]; |
2765 | |
2766 | @@ -411,7 +418,7 @@ void zfcp_dbf_scsi_devreset(char *tag, struct scsi_cmnd *scmnd, u8 flag) |
2767 | memcpy(tmp_tag, "lr_", 3); |
2768 | |
2769 | memcpy(&tmp_tag[3], tag, 4); |
2770 | - _zfcp_dbf_scsi(tmp_tag, 1, scmnd, NULL); |
2771 | + _zfcp_dbf_scsi(tmp_tag, 1, scmnd, fsf_req); |
2772 | } |
2773 | |
2774 | /** |
2775 | diff --git a/drivers/s390/scsi/zfcp_fc.h b/drivers/s390/scsi/zfcp_fc.h |
2776 | index df2b541c8287..a2275825186f 100644 |
2777 | --- a/drivers/s390/scsi/zfcp_fc.h |
2778 | +++ b/drivers/s390/scsi/zfcp_fc.h |
2779 | @@ -4,7 +4,7 @@ |
2780 | * Fibre Channel related definitions and inline functions for the zfcp |
2781 | * device driver |
2782 | * |
2783 | - * Copyright IBM Corp. 2009 |
2784 | + * Copyright IBM Corp. 2009, 2017 |
2785 | */ |
2786 | |
2787 | #ifndef ZFCP_FC_H |
2788 | @@ -279,6 +279,10 @@ void zfcp_fc_eval_fcp_rsp(struct fcp_resp_with_ext *fcp_rsp, |
2789 | !(rsp_flags & FCP_SNS_LEN_VAL) && |
2790 | fcp_rsp->resp.fr_status == SAM_STAT_GOOD) |
2791 | set_host_byte(scsi, DID_ERROR); |
2792 | + } else if (unlikely(rsp_flags & FCP_RESID_OVER)) { |
2793 | + /* FCP_DL was not sufficient for SCSI data length */ |
2794 | + if (fcp_rsp->resp.fr_status == SAM_STAT_GOOD) |
2795 | + set_host_byte(scsi, DID_ERROR); |
2796 | } |
2797 | } |
2798 | |
2799 | diff --git a/drivers/s390/scsi/zfcp_fsf.c b/drivers/s390/scsi/zfcp_fsf.c |
2800 | index 27ff38f839fc..1964391db904 100644 |
2801 | --- a/drivers/s390/scsi/zfcp_fsf.c |
2802 | +++ b/drivers/s390/scsi/zfcp_fsf.c |
2803 | @@ -928,8 +928,8 @@ static void zfcp_fsf_send_ct_handler(struct zfcp_fsf_req *req) |
2804 | |
2805 | switch (header->fsf_status) { |
2806 | case FSF_GOOD: |
2807 | - zfcp_dbf_san_res("fsscth2", req); |
2808 | ct->status = 0; |
2809 | + zfcp_dbf_san_res("fsscth2", req); |
2810 | break; |
2811 | case FSF_SERVICE_CLASS_NOT_SUPPORTED: |
2812 | zfcp_fsf_class_not_supp(req); |
2813 | @@ -1109,8 +1109,8 @@ static void zfcp_fsf_send_els_handler(struct zfcp_fsf_req *req) |
2814 | |
2815 | switch (header->fsf_status) { |
2816 | case FSF_GOOD: |
2817 | - zfcp_dbf_san_res("fsselh1", req); |
2818 | send_els->status = 0; |
2819 | + zfcp_dbf_san_res("fsselh1", req); |
2820 | break; |
2821 | case FSF_SERVICE_CLASS_NOT_SUPPORTED: |
2822 | zfcp_fsf_class_not_supp(req); |
2823 | @@ -2258,7 +2258,8 @@ int zfcp_fsf_fcp_cmnd(struct scsi_cmnd *scsi_cmnd) |
2824 | fcp_cmnd = (struct fcp_cmnd *) &req->qtcb->bottom.io.fcp_cmnd; |
2825 | zfcp_fc_scsi_to_fcp(fcp_cmnd, scsi_cmnd, 0); |
2826 | |
2827 | - if (scsi_prot_sg_count(scsi_cmnd)) { |
2828 | + if ((scsi_get_prot_op(scsi_cmnd) != SCSI_PROT_NORMAL) && |
2829 | + scsi_prot_sg_count(scsi_cmnd)) { |
2830 | zfcp_qdio_set_data_div(qdio, &req->qdio_req, |
2831 | scsi_prot_sg_count(scsi_cmnd)); |
2832 | retval = zfcp_qdio_sbals_from_sg(qdio, &req->qdio_req, |
2833 | diff --git a/drivers/s390/scsi/zfcp_scsi.c b/drivers/s390/scsi/zfcp_scsi.c |
2834 | index 07ffdbb5107f..9bd9b9a29dfc 100644 |
2835 | --- a/drivers/s390/scsi/zfcp_scsi.c |
2836 | +++ b/drivers/s390/scsi/zfcp_scsi.c |
2837 | @@ -3,7 +3,7 @@ |
2838 | * |
2839 | * Interface to Linux SCSI midlayer. |
2840 | * |
2841 | - * Copyright IBM Corp. 2002, 2016 |
2842 | + * Copyright IBM Corp. 2002, 2017 |
2843 | */ |
2844 | |
2845 | #define KMSG_COMPONENT "zfcp" |
2846 | @@ -273,25 +273,29 @@ static int zfcp_task_mgmt_function(struct scsi_cmnd *scpnt, u8 tm_flags) |
2847 | |
2848 | zfcp_erp_wait(adapter); |
2849 | ret = fc_block_scsi_eh(scpnt); |
2850 | - if (ret) |
2851 | + if (ret) { |
2852 | + zfcp_dbf_scsi_devreset("fiof", scpnt, tm_flags, NULL); |
2853 | return ret; |
2854 | + } |
2855 | |
2856 | if (!(atomic_read(&adapter->status) & |
2857 | ZFCP_STATUS_COMMON_RUNNING)) { |
2858 | - zfcp_dbf_scsi_devreset("nres", scpnt, tm_flags); |
2859 | + zfcp_dbf_scsi_devreset("nres", scpnt, tm_flags, NULL); |
2860 | return SUCCESS; |
2861 | } |
2862 | } |
2863 | - if (!fsf_req) |
2864 | + if (!fsf_req) { |
2865 | + zfcp_dbf_scsi_devreset("reqf", scpnt, tm_flags, NULL); |
2866 | return FAILED; |
2867 | + } |
2868 | |
2869 | wait_for_completion(&fsf_req->completion); |
2870 | |
2871 | if (fsf_req->status & ZFCP_STATUS_FSFREQ_TMFUNCFAILED) { |
2872 | - zfcp_dbf_scsi_devreset("fail", scpnt, tm_flags); |
2873 | + zfcp_dbf_scsi_devreset("fail", scpnt, tm_flags, fsf_req); |
2874 | retval = FAILED; |
2875 | } else { |
2876 | - zfcp_dbf_scsi_devreset("okay", scpnt, tm_flags); |
2877 | + zfcp_dbf_scsi_devreset("okay", scpnt, tm_flags, fsf_req); |
2878 | zfcp_scsi_forget_cmnds(zfcp_sdev, tm_flags); |
2879 | } |
2880 | |
2881 | diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c |
2882 | index d8b1fbd4c8aa..35cbd36f8d3b 100644 |
2883 | --- a/drivers/scsi/megaraid/megaraid_sas_base.c |
2884 | +++ b/drivers/scsi/megaraid/megaraid_sas_base.c |
2885 | @@ -1901,9 +1901,12 @@ static void megasas_complete_outstanding_ioctls(struct megasas_instance *instanc |
2886 | if (cmd_fusion->sync_cmd_idx != (u32)ULONG_MAX) { |
2887 | cmd_mfi = instance->cmd_list[cmd_fusion->sync_cmd_idx]; |
2888 | if (cmd_mfi->sync_cmd && |
2889 | - cmd_mfi->frame->hdr.cmd != MFI_CMD_ABORT) |
2890 | + (cmd_mfi->frame->hdr.cmd != MFI_CMD_ABORT)) { |
2891 | + cmd_mfi->frame->hdr.cmd_status = |
2892 | + MFI_STAT_WRONG_STATE; |
2893 | megasas_complete_cmd(instance, |
2894 | cmd_mfi, DID_OK); |
2895 | + } |
2896 | } |
2897 | } |
2898 | } else { |
2899 | @@ -5290,7 +5293,8 @@ static int megasas_init_fw(struct megasas_instance *instance) |
2900 | instance->throttlequeuedepth = |
2901 | MEGASAS_THROTTLE_QUEUE_DEPTH; |
2902 | |
2903 | - if (resetwaittime > MEGASAS_RESET_WAIT_TIME) |
2904 | + if ((resetwaittime < 1) || |
2905 | + (resetwaittime > MEGASAS_RESET_WAIT_TIME)) |
2906 | resetwaittime = MEGASAS_RESET_WAIT_TIME; |
2907 | |
2908 | if ((scmd_timeout < 10) || (scmd_timeout > MEGASAS_DEFAULT_CMD_TIMEOUT)) |
2909 | @@ -5459,6 +5463,14 @@ megasas_register_aen(struct megasas_instance *instance, u32 seq_num, |
2910 | prev_aen.word = |
2911 | le32_to_cpu(instance->aen_cmd->frame->dcmd.mbox.w[1]); |
2912 | |
2913 | + if ((curr_aen.members.class < MFI_EVT_CLASS_DEBUG) || |
2914 | + (curr_aen.members.class > MFI_EVT_CLASS_DEAD)) { |
2915 | + dev_info(&instance->pdev->dev, |
2916 | + "%s %d out of range class %d send by application\n", |
2917 | + __func__, __LINE__, curr_aen.members.class); |
2918 | + return 0; |
2919 | + } |
2920 | + |
2921 | /* |
2922 | * A class whose enum value is smaller is inclusive of all |
2923 | * higher values. If a PROGRESS (= -1) was previously |
2924 | diff --git a/drivers/scsi/qla2xxx/qla_attr.c b/drivers/scsi/qla2xxx/qla_attr.c |
2925 | index 8c4641b518b5..9a34afcb1c4c 100644 |
2926 | --- a/drivers/scsi/qla2xxx/qla_attr.c |
2927 | +++ b/drivers/scsi/qla2xxx/qla_attr.c |
2928 | @@ -318,6 +318,8 @@ qla2x00_sysfs_write_optrom_ctl(struct file *filp, struct kobject *kobj, |
2929 | return -EINVAL; |
2930 | if (start > ha->optrom_size) |
2931 | return -EINVAL; |
2932 | + if (size > ha->optrom_size - start) |
2933 | + size = ha->optrom_size - start; |
2934 | |
2935 | mutex_lock(&ha->optrom_mutex); |
2936 | switch (val) { |
2937 | @@ -343,8 +345,7 @@ qla2x00_sysfs_write_optrom_ctl(struct file *filp, struct kobject *kobj, |
2938 | } |
2939 | |
2940 | ha->optrom_region_start = start; |
2941 | - ha->optrom_region_size = start + size > ha->optrom_size ? |
2942 | - ha->optrom_size - start : size; |
2943 | + ha->optrom_region_size = start + size; |
2944 | |
2945 | ha->optrom_state = QLA_SREADING; |
2946 | ha->optrom_buffer = vmalloc(ha->optrom_region_size); |
2947 | @@ -417,8 +418,7 @@ qla2x00_sysfs_write_optrom_ctl(struct file *filp, struct kobject *kobj, |
2948 | } |
2949 | |
2950 | ha->optrom_region_start = start; |
2951 | - ha->optrom_region_size = start + size > ha->optrom_size ? |
2952 | - ha->optrom_size - start : size; |
2953 | + ha->optrom_region_size = start + size; |
2954 | |
2955 | ha->optrom_state = QLA_SWRITING; |
2956 | ha->optrom_buffer = vmalloc(ha->optrom_region_size); |
2957 | diff --git a/drivers/scsi/qla2xxx/qla_mid.c b/drivers/scsi/qla2xxx/qla_mid.c |
2958 | index 3dfb54abc874..f8ae70476b3a 100644 |
2959 | --- a/drivers/scsi/qla2xxx/qla_mid.c |
2960 | +++ b/drivers/scsi/qla2xxx/qla_mid.c |
2961 | @@ -74,7 +74,7 @@ qla24xx_deallocate_vp_id(scsi_qla_host_t *vha) |
2962 | * ensures no active vp_list traversal while the vport is removed |
2963 | * from the queue) |
2964 | */ |
2965 | - wait_event_timeout(vha->vref_waitq, atomic_read(&vha->vref_count), |
2966 | + wait_event_timeout(vha->vref_waitq, !atomic_read(&vha->vref_count), |
2967 | 10*HZ); |
2968 | |
2969 | spin_lock_irqsave(&ha->vport_slock, flags); |
2970 | diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c |
2971 | index 9236a13d5d2a..02dfbc1373e3 100644 |
2972 | --- a/drivers/scsi/sg.c |
2973 | +++ b/drivers/scsi/sg.c |
2974 | @@ -122,7 +122,7 @@ struct sg_device; /* forward declarations */ |
2975 | struct sg_fd; |
2976 | |
2977 | typedef struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */ |
2978 | - struct sg_request *nextrp; /* NULL -> tail request (slist) */ |
2979 | + struct list_head entry; /* list entry */ |
2980 | struct sg_fd *parentfp; /* NULL -> not in use */ |
2981 | Sg_scatter_hold data; /* hold buffer, perhaps scatter list */ |
2982 | sg_io_hdr_t header; /* scsi command+info, see <scsi/sg.h> */ |
2983 | @@ -146,8 +146,7 @@ typedef struct sg_fd { /* holds the state of a file descriptor */ |
2984 | int timeout; /* defaults to SG_DEFAULT_TIMEOUT */ |
2985 | int timeout_user; /* defaults to SG_DEFAULT_TIMEOUT_USER */ |
2986 | Sg_scatter_hold reserve; /* buffer held for this file descriptor */ |
2987 | - unsigned save_scat_len; /* original length of trunc. scat. element */ |
2988 | - Sg_request *headrp; /* head of request slist, NULL->empty */ |
2989 | + struct list_head rq_list; /* head of request list */ |
2990 | struct fasync_struct *async_qp; /* used by asynchronous notification */ |
2991 | Sg_request req_arr[SG_MAX_QUEUE]; /* used as singly-linked list */ |
2992 | char low_dma; /* as in parent but possibly overridden to 1 */ |
2993 | @@ -829,6 +828,39 @@ static int max_sectors_bytes(struct request_queue *q) |
2994 | return max_sectors << 9; |
2995 | } |
2996 | |
2997 | +static void |
2998 | +sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo) |
2999 | +{ |
3000 | + Sg_request *srp; |
3001 | + int val; |
3002 | + unsigned int ms; |
3003 | + |
3004 | + val = 0; |
3005 | + list_for_each_entry(srp, &sfp->rq_list, entry) { |
3006 | + if (val > SG_MAX_QUEUE) |
3007 | + break; |
3008 | + rinfo[val].req_state = srp->done + 1; |
3009 | + rinfo[val].problem = |
3010 | + srp->header.masked_status & |
3011 | + srp->header.host_status & |
3012 | + srp->header.driver_status; |
3013 | + if (srp->done) |
3014 | + rinfo[val].duration = |
3015 | + srp->header.duration; |
3016 | + else { |
3017 | + ms = jiffies_to_msecs(jiffies); |
3018 | + rinfo[val].duration = |
3019 | + (ms > srp->header.duration) ? |
3020 | + (ms - srp->header.duration) : 0; |
3021 | + } |
3022 | + rinfo[val].orphan = srp->orphan; |
3023 | + rinfo[val].sg_io_owned = srp->sg_io_owned; |
3024 | + rinfo[val].pack_id = srp->header.pack_id; |
3025 | + rinfo[val].usr_ptr = srp->header.usr_ptr; |
3026 | + val++; |
3027 | + } |
3028 | +} |
3029 | + |
3030 | static long |
3031 | sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) |
3032 | { |
3033 | @@ -941,7 +973,7 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) |
3034 | if (!access_ok(VERIFY_WRITE, ip, sizeof (int))) |
3035 | return -EFAULT; |
3036 | read_lock_irqsave(&sfp->rq_list_lock, iflags); |
3037 | - for (srp = sfp->headrp; srp; srp = srp->nextrp) { |
3038 | + list_for_each_entry(srp, &sfp->rq_list, entry) { |
3039 | if ((1 == srp->done) && (!srp->sg_io_owned)) { |
3040 | read_unlock_irqrestore(&sfp->rq_list_lock, |
3041 | iflags); |
3042 | @@ -954,7 +986,8 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) |
3043 | return 0; |
3044 | case SG_GET_NUM_WAITING: |
3045 | read_lock_irqsave(&sfp->rq_list_lock, iflags); |
3046 | - for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) { |
3047 | + val = 0; |
3048 | + list_for_each_entry(srp, &sfp->rq_list, entry) { |
3049 | if ((1 == srp->done) && (!srp->sg_io_owned)) |
3050 | ++val; |
3051 | } |
3052 | @@ -1022,42 +1055,15 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) |
3053 | return -EFAULT; |
3054 | else { |
3055 | sg_req_info_t *rinfo; |
3056 | - unsigned int ms; |
3057 | |
3058 | - rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE, |
3059 | - GFP_KERNEL); |
3060 | + rinfo = kzalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE, |
3061 | + GFP_KERNEL); |
3062 | if (!rinfo) |
3063 | return -ENOMEM; |
3064 | read_lock_irqsave(&sfp->rq_list_lock, iflags); |
3065 | - for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE; |
3066 | - ++val, srp = srp ? srp->nextrp : srp) { |
3067 | - memset(&rinfo[val], 0, SZ_SG_REQ_INFO); |
3068 | - if (srp) { |
3069 | - rinfo[val].req_state = srp->done + 1; |
3070 | - rinfo[val].problem = |
3071 | - srp->header.masked_status & |
3072 | - srp->header.host_status & |
3073 | - srp->header.driver_status; |
3074 | - if (srp->done) |
3075 | - rinfo[val].duration = |
3076 | - srp->header.duration; |
3077 | - else { |
3078 | - ms = jiffies_to_msecs(jiffies); |
3079 | - rinfo[val].duration = |
3080 | - (ms > srp->header.duration) ? |
3081 | - (ms - srp->header.duration) : 0; |
3082 | - } |
3083 | - rinfo[val].orphan = srp->orphan; |
3084 | - rinfo[val].sg_io_owned = |
3085 | - srp->sg_io_owned; |
3086 | - rinfo[val].pack_id = |
3087 | - srp->header.pack_id; |
3088 | - rinfo[val].usr_ptr = |
3089 | - srp->header.usr_ptr; |
3090 | - } |
3091 | - } |
3092 | + sg_fill_request_table(sfp, rinfo); |
3093 | read_unlock_irqrestore(&sfp->rq_list_lock, iflags); |
3094 | - result = __copy_to_user(p, rinfo, |
3095 | + result = __copy_to_user(p, rinfo, |
3096 | SZ_SG_REQ_INFO * SG_MAX_QUEUE); |
3097 | result = result ? -EFAULT : 0; |
3098 | kfree(rinfo); |
3099 | @@ -1163,7 +1169,7 @@ sg_poll(struct file *filp, poll_table * wait) |
3100 | return POLLERR; |
3101 | poll_wait(filp, &sfp->read_wait, wait); |
3102 | read_lock_irqsave(&sfp->rq_list_lock, iflags); |
3103 | - for (srp = sfp->headrp; srp; srp = srp->nextrp) { |
3104 | + list_for_each_entry(srp, &sfp->rq_list, entry) { |
3105 | /* if any read waiting, flag it */ |
3106 | if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned)) |
3107 | res = POLLIN | POLLRDNORM; |
3108 | @@ -2049,7 +2055,6 @@ sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp) |
3109 | req_schp->pages = NULL; |
3110 | req_schp->page_order = 0; |
3111 | req_schp->sglist_len = 0; |
3112 | - sfp->save_scat_len = 0; |
3113 | srp->res_used = 0; |
3114 | /* Called without mutex lock to avoid deadlock */ |
3115 | sfp->res_in_use = 0; |
3116 | @@ -2062,7 +2067,7 @@ sg_get_rq_mark(Sg_fd * sfp, int pack_id) |
3117 | unsigned long iflags; |
3118 | |
3119 | write_lock_irqsave(&sfp->rq_list_lock, iflags); |
3120 | - for (resp = sfp->headrp; resp; resp = resp->nextrp) { |
3121 | + list_for_each_entry(resp, &sfp->rq_list, entry) { |
3122 | /* look for requests that are ready + not SG_IO owned */ |
3123 | if ((1 == resp->done) && (!resp->sg_io_owned) && |
3124 | ((-1 == pack_id) || (resp->header.pack_id == pack_id))) { |
3125 | @@ -2080,70 +2085,45 @@ sg_add_request(Sg_fd * sfp) |
3126 | { |
3127 | int k; |
3128 | unsigned long iflags; |
3129 | - Sg_request *resp; |
3130 | Sg_request *rp = sfp->req_arr; |
3131 | |
3132 | write_lock_irqsave(&sfp->rq_list_lock, iflags); |
3133 | - resp = sfp->headrp; |
3134 | - if (!resp) { |
3135 | - memset(rp, 0, sizeof (Sg_request)); |
3136 | - rp->parentfp = sfp; |
3137 | - resp = rp; |
3138 | - sfp->headrp = resp; |
3139 | - } else { |
3140 | - if (0 == sfp->cmd_q) |
3141 | - resp = NULL; /* command queuing disallowed */ |
3142 | - else { |
3143 | - for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) { |
3144 | - if (!rp->parentfp) |
3145 | - break; |
3146 | - } |
3147 | - if (k < SG_MAX_QUEUE) { |
3148 | - memset(rp, 0, sizeof (Sg_request)); |
3149 | - rp->parentfp = sfp; |
3150 | - while (resp->nextrp) |
3151 | - resp = resp->nextrp; |
3152 | - resp->nextrp = rp; |
3153 | - resp = rp; |
3154 | - } else |
3155 | - resp = NULL; |
3156 | + if (!list_empty(&sfp->rq_list)) { |
3157 | + if (!sfp->cmd_q) |
3158 | + goto out_unlock; |
3159 | + |
3160 | + for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) { |
3161 | + if (!rp->parentfp) |
3162 | + break; |
3163 | } |
3164 | + if (k >= SG_MAX_QUEUE) |
3165 | + goto out_unlock; |
3166 | } |
3167 | - if (resp) { |
3168 | - resp->nextrp = NULL; |
3169 | - resp->header.duration = jiffies_to_msecs(jiffies); |
3170 | - } |
3171 | + memset(rp, 0, sizeof (Sg_request)); |
3172 | + rp->parentfp = sfp; |
3173 | + rp->header.duration = jiffies_to_msecs(jiffies); |
3174 | + list_add_tail(&rp->entry, &sfp->rq_list); |
3175 | write_unlock_irqrestore(&sfp->rq_list_lock, iflags); |
3176 | - return resp; |
3177 | + return rp; |
3178 | +out_unlock: |
3179 | + write_unlock_irqrestore(&sfp->rq_list_lock, iflags); |
3180 | + return NULL; |
3181 | } |
3182 | |
3183 | /* Return of 1 for found; 0 for not found */ |
3184 | static int |
3185 | sg_remove_request(Sg_fd * sfp, Sg_request * srp) |
3186 | { |
3187 | - Sg_request *prev_rp; |
3188 | - Sg_request *rp; |
3189 | unsigned long iflags; |
3190 | int res = 0; |
3191 | |
3192 | - if ((!sfp) || (!srp) || (!sfp->headrp)) |
3193 | + if (!sfp || !srp || list_empty(&sfp->rq_list)) |
3194 | return res; |
3195 | write_lock_irqsave(&sfp->rq_list_lock, iflags); |
3196 | - prev_rp = sfp->headrp; |
3197 | - if (srp == prev_rp) { |
3198 | - sfp->headrp = prev_rp->nextrp; |
3199 | - prev_rp->parentfp = NULL; |
3200 | + if (!list_empty(&srp->entry)) { |
3201 | + list_del(&srp->entry); |
3202 | + srp->parentfp = NULL; |
3203 | res = 1; |
3204 | - } else { |
3205 | - while ((rp = prev_rp->nextrp)) { |
3206 | - if (srp == rp) { |
3207 | - prev_rp->nextrp = rp->nextrp; |
3208 | - rp->parentfp = NULL; |
3209 | - res = 1; |
3210 | - break; |
3211 | - } |
3212 | - prev_rp = rp; |
3213 | - } |
3214 | } |
3215 | write_unlock_irqrestore(&sfp->rq_list_lock, iflags); |
3216 | return res; |
3217 | @@ -2162,7 +2142,7 @@ sg_add_sfp(Sg_device * sdp) |
3218 | |
3219 | init_waitqueue_head(&sfp->read_wait); |
3220 | rwlock_init(&sfp->rq_list_lock); |
3221 | - |
3222 | + INIT_LIST_HEAD(&sfp->rq_list); |
3223 | kref_init(&sfp->f_ref); |
3224 | mutex_init(&sfp->f_mutex); |
3225 | sfp->timeout = SG_DEFAULT_TIMEOUT; |
3226 | @@ -2203,10 +2183,13 @@ sg_remove_sfp_usercontext(struct work_struct *work) |
3227 | { |
3228 | struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work); |
3229 | struct sg_device *sdp = sfp->parentdp; |
3230 | + Sg_request *srp; |
3231 | |
3232 | /* Cleanup any responses which were never read(). */ |
3233 | - while (sfp->headrp) |
3234 | - sg_finish_rem_req(sfp->headrp); |
3235 | + while (!list_empty(&sfp->rq_list)) { |
3236 | + srp = list_first_entry(&sfp->rq_list, Sg_request, entry); |
3237 | + sg_finish_rem_req(srp); |
3238 | + } |
3239 | |
3240 | if (sfp->reserve.bufflen > 0) { |
3241 | SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp, |
3242 | @@ -2609,7 +2592,7 @@ static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v) |
3243 | /* must be called while holding sg_index_lock */ |
3244 | static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp) |
3245 | { |
3246 | - int k, m, new_interface, blen, usg; |
3247 | + int k, new_interface, blen, usg; |
3248 | Sg_request *srp; |
3249 | Sg_fd *fp; |
3250 | const sg_io_hdr_t *hp; |
3251 | @@ -2629,13 +2612,11 @@ static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp) |
3252 | seq_printf(s, " cmd_q=%d f_packid=%d k_orphan=%d closed=0\n", |
3253 | (int) fp->cmd_q, (int) fp->force_packid, |
3254 | (int) fp->keep_orphan); |
3255 | - for (m = 0, srp = fp->headrp; |
3256 | - srp != NULL; |
3257 | - ++m, srp = srp->nextrp) { |
3258 | + list_for_each_entry(srp, &fp->rq_list, entry) { |
3259 | hp = &srp->header; |
3260 | new_interface = (hp->interface_id == '\0') ? 0 : 1; |
3261 | if (srp->res_used) { |
3262 | - if (new_interface && |
3263 | + if (new_interface && |
3264 | (SG_FLAG_MMAP_IO & hp->flags)) |
3265 | cp = " mmap>> "; |
3266 | else |
3267 | @@ -2666,7 +2647,7 @@ static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp) |
3268 | seq_printf(s, "ms sgat=%d op=0x%02x\n", usg, |
3269 | (int) srp->data.cmd_opcode); |
3270 | } |
3271 | - if (0 == m) |
3272 | + if (list_empty(&fp->rq_list)) |
3273 | seq_puts(s, " No requests active\n"); |
3274 | read_unlock(&fp->rq_list_lock); |
3275 | } |
3276 | diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c |
3277 | index c5ab1b0037fc..2bf96d33428a 100644 |
3278 | --- a/drivers/scsi/storvsc_drv.c |
3279 | +++ b/drivers/scsi/storvsc_drv.c |
3280 | @@ -1559,6 +1559,8 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) |
3281 | ret = storvsc_do_io(dev, cmd_request); |
3282 | |
3283 | if (ret == -EAGAIN) { |
3284 | + if (payload_sz > sizeof(cmd_request->mpb)) |
3285 | + kfree(payload); |
3286 | /* no more space */ |
3287 | return SCSI_MLQUEUE_DEVICE_BUSY; |
3288 | } |
3289 | diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c |
3290 | index aa80dc94ddc2..c220c2c0893f 100644 |
3291 | --- a/drivers/tty/tty_buffer.c |
3292 | +++ b/drivers/tty/tty_buffer.c |
3293 | @@ -361,6 +361,32 @@ int tty_insert_flip_string_flags(struct tty_port *port, |
3294 | } |
3295 | EXPORT_SYMBOL(tty_insert_flip_string_flags); |
3296 | |
3297 | +/** |
3298 | + * __tty_insert_flip_char - Add one character to the tty buffer |
3299 | + * @port: tty port |
3300 | + * @ch: character |
3301 | + * @flag: flag byte |
3302 | + * |
3303 | + * Queue a single byte to the tty buffering, with an optional flag. |
3304 | + * This is the slow path of tty_insert_flip_char. |
3305 | + */ |
3306 | +int __tty_insert_flip_char(struct tty_port *port, unsigned char ch, char flag) |
3307 | +{ |
3308 | + struct tty_buffer *tb; |
3309 | + int flags = (flag == TTY_NORMAL) ? TTYB_NORMAL : 0; |
3310 | + |
3311 | + if (!__tty_buffer_request_room(port, 1, flags)) |
3312 | + return 0; |
3313 | + |
3314 | + tb = port->buf.tail; |
3315 | + if (~tb->flags & TTYB_NORMAL) |
3316 | + *flag_buf_ptr(tb, tb->used) = flag; |
3317 | + *char_buf_ptr(tb, tb->used++) = ch; |
3318 | + |
3319 | + return 1; |
3320 | +} |
3321 | +EXPORT_SYMBOL(__tty_insert_flip_char); |
3322 | + |
3323 | /** |
3324 | * tty_schedule_flip - push characters to ldisc |
3325 | * @port: tty port to push from |
3326 | diff --git a/fs/ext4/super.c b/fs/ext4/super.c |
3327 | index 5fa9ba1de429..f72535e1898f 100644 |
3328 | --- a/fs/ext4/super.c |
3329 | +++ b/fs/ext4/super.c |
3330 | @@ -2334,6 +2334,7 @@ static void ext4_orphan_cleanup(struct super_block *sb, |
3331 | unsigned int s_flags = sb->s_flags; |
3332 | int nr_orphans = 0, nr_truncates = 0; |
3333 | #ifdef CONFIG_QUOTA |
3334 | + int quota_update = 0; |
3335 | int i; |
3336 | #endif |
3337 | if (!es->s_last_orphan) { |
3338 | @@ -2372,14 +2373,32 @@ static void ext4_orphan_cleanup(struct super_block *sb, |
3339 | #ifdef CONFIG_QUOTA |
3340 | /* Needed for iput() to work correctly and not trash data */ |
3341 | sb->s_flags |= MS_ACTIVE; |
3342 | - /* Turn on quotas so that they are updated correctly */ |
3343 | + |
3344 | + /* |
3345 | + * Turn on quotas which were not enabled for read-only mounts if |
3346 | + * filesystem has quota feature, so that they are updated correctly. |
3347 | + */ |
3348 | + if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) { |
3349 | + int ret = ext4_enable_quotas(sb); |
3350 | + |
3351 | + if (!ret) |
3352 | + quota_update = 1; |
3353 | + else |
3354 | + ext4_msg(sb, KERN_ERR, |
3355 | + "Cannot turn on quotas: error %d", ret); |
3356 | + } |
3357 | + |
3358 | + /* Turn on journaled quotas used for old sytle */ |
3359 | for (i = 0; i < EXT4_MAXQUOTAS; i++) { |
3360 | if (EXT4_SB(sb)->s_qf_names[i]) { |
3361 | int ret = ext4_quota_on_mount(sb, i); |
3362 | - if (ret < 0) |
3363 | + |
3364 | + if (!ret) |
3365 | + quota_update = 1; |
3366 | + else |
3367 | ext4_msg(sb, KERN_ERR, |
3368 | "Cannot turn on journaled " |
3369 | - "quota: error %d", ret); |
3370 | + "quota: type %d: error %d", i, ret); |
3371 | } |
3372 | } |
3373 | #endif |
3374 | @@ -2438,10 +2457,12 @@ static void ext4_orphan_cleanup(struct super_block *sb, |
3375 | ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up", |
3376 | PLURAL(nr_truncates)); |
3377 | #ifdef CONFIG_QUOTA |
3378 | - /* Turn quotas off */ |
3379 | - for (i = 0; i < EXT4_MAXQUOTAS; i++) { |
3380 | - if (sb_dqopt(sb)->files[i]) |
3381 | - dquot_quota_off(sb, i); |
3382 | + /* Turn off quotas if they were enabled for orphan cleanup */ |
3383 | + if (quota_update) { |
3384 | + for (i = 0; i < EXT4_MAXQUOTAS; i++) { |
3385 | + if (sb_dqopt(sb)->files[i]) |
3386 | + dquot_quota_off(sb, i); |
3387 | + } |
3388 | } |
3389 | #endif |
3390 | sb->s_flags = s_flags; /* Restore MS_RDONLY status */ |
3391 | @@ -5365,6 +5386,9 @@ static int ext4_enable_quotas(struct super_block *sb) |
3392 | DQUOT_USAGE_ENABLED | |
3393 | (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0)); |
3394 | if (err) { |
3395 | + for (type--; type >= 0; type--) |
3396 | + dquot_quota_off(sb, type); |
3397 | + |
3398 | ext4_warning(sb, |
3399 | "Failed to enable quota tracking " |
3400 | "(type=%d, err=%d). Please run " |
3401 | diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c |
3402 | index 0a2115084c3f..582bfee40345 100644 |
3403 | --- a/fs/nfs/callback.c |
3404 | +++ b/fs/nfs/callback.c |
3405 | @@ -75,7 +75,10 @@ nfs4_callback_svc(void *vrqstp) |
3406 | |
3407 | set_freezable(); |
3408 | |
3409 | - while (!kthread_should_stop()) { |
3410 | + while (!kthread_freezable_should_stop(NULL)) { |
3411 | + |
3412 | + if (signal_pending(current)) |
3413 | + flush_signals(current); |
3414 | /* |
3415 | * Listen for a request on the socket |
3416 | */ |
3417 | @@ -84,6 +87,8 @@ nfs4_callback_svc(void *vrqstp) |
3418 | continue; |
3419 | svc_process(rqstp); |
3420 | } |
3421 | + svc_exit_thread(rqstp); |
3422 | + module_put_and_exit(0); |
3423 | return 0; |
3424 | } |
3425 | |
3426 | @@ -102,9 +107,10 @@ nfs41_callback_svc(void *vrqstp) |
3427 | |
3428 | set_freezable(); |
3429 | |
3430 | - while (!kthread_should_stop()) { |
3431 | - if (try_to_freeze()) |
3432 | - continue; |
3433 | + while (!kthread_freezable_should_stop(NULL)) { |
3434 | + |
3435 | + if (signal_pending(current)) |
3436 | + flush_signals(current); |
3437 | |
3438 | prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); |
3439 | spin_lock_bh(&serv->sv_cb_lock); |
3440 | @@ -120,11 +126,13 @@ nfs41_callback_svc(void *vrqstp) |
3441 | error); |
3442 | } else { |
3443 | spin_unlock_bh(&serv->sv_cb_lock); |
3444 | - schedule(); |
3445 | + if (!kthread_should_stop()) |
3446 | + schedule(); |
3447 | finish_wait(&serv->sv_cb_waitq, &wq); |
3448 | } |
3449 | - flush_signals(current); |
3450 | } |
3451 | + svc_exit_thread(rqstp); |
3452 | + module_put_and_exit(0); |
3453 | return 0; |
3454 | } |
3455 | |
3456 | @@ -220,14 +228,14 @@ static int nfs_callback_up_net(int minorversion, struct svc_serv *serv, |
3457 | static struct svc_serv_ops nfs40_cb_sv_ops = { |
3458 | .svo_function = nfs4_callback_svc, |
3459 | .svo_enqueue_xprt = svc_xprt_do_enqueue, |
3460 | - .svo_setup = svc_set_num_threads, |
3461 | + .svo_setup = svc_set_num_threads_sync, |
3462 | .svo_module = THIS_MODULE, |
3463 | }; |
3464 | #if defined(CONFIG_NFS_V4_1) |
3465 | static struct svc_serv_ops nfs41_cb_sv_ops = { |
3466 | .svo_function = nfs41_callback_svc, |
3467 | .svo_enqueue_xprt = svc_xprt_do_enqueue, |
3468 | - .svo_setup = svc_set_num_threads, |
3469 | + .svo_setup = svc_set_num_threads_sync, |
3470 | .svo_module = THIS_MODULE, |
3471 | }; |
3472 | |
3473 | diff --git a/fs/orangefs/acl.c b/fs/orangefs/acl.c |
3474 | index 7a3754488312..9409aac232f7 100644 |
3475 | --- a/fs/orangefs/acl.c |
3476 | +++ b/fs/orangefs/acl.c |
3477 | @@ -61,9 +61,9 @@ struct posix_acl *orangefs_get_acl(struct inode *inode, int type) |
3478 | return acl; |
3479 | } |
3480 | |
3481 | -int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type) |
3482 | +static int __orangefs_set_acl(struct inode *inode, struct posix_acl *acl, |
3483 | + int type) |
3484 | { |
3485 | - struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
3486 | int error = 0; |
3487 | void *value = NULL; |
3488 | size_t size = 0; |
3489 | @@ -72,22 +72,6 @@ int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type) |
3490 | switch (type) { |
3491 | case ACL_TYPE_ACCESS: |
3492 | name = XATTR_NAME_POSIX_ACL_ACCESS; |
3493 | - if (acl) { |
3494 | - umode_t mode; |
3495 | - |
3496 | - error = posix_acl_update_mode(inode, &mode, &acl); |
3497 | - if (error) { |
3498 | - gossip_err("%s: posix_acl_update_mode err: %d\n", |
3499 | - __func__, |
3500 | - error); |
3501 | - return error; |
3502 | - } |
3503 | - |
3504 | - if (inode->i_mode != mode) |
3505 | - SetModeFlag(orangefs_inode); |
3506 | - inode->i_mode = mode; |
3507 | - mark_inode_dirty_sync(inode); |
3508 | - } |
3509 | break; |
3510 | case ACL_TYPE_DEFAULT: |
3511 | name = XATTR_NAME_POSIX_ACL_DEFAULT; |
3512 | @@ -132,6 +116,29 @@ int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type) |
3513 | return error; |
3514 | } |
3515 | |
3516 | +int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type) |
3517 | +{ |
3518 | + int error; |
3519 | + |
3520 | + if (type == ACL_TYPE_ACCESS && acl) { |
3521 | + umode_t mode; |
3522 | + |
3523 | + error = posix_acl_update_mode(inode, &mode, &acl); |
3524 | + if (error) { |
3525 | + gossip_err("%s: posix_acl_update_mode err: %d\n", |
3526 | + __func__, |
3527 | + error); |
3528 | + return error; |
3529 | + } |
3530 | + |
3531 | + if (inode->i_mode != mode) |
3532 | + SetModeFlag(ORANGEFS_I(inode)); |
3533 | + inode->i_mode = mode; |
3534 | + mark_inode_dirty_sync(inode); |
3535 | + } |
3536 | + return __orangefs_set_acl(inode, acl, type); |
3537 | +} |
3538 | + |
3539 | int orangefs_init_acl(struct inode *inode, struct inode *dir) |
3540 | { |
3541 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
3542 | @@ -146,13 +153,14 @@ int orangefs_init_acl(struct inode *inode, struct inode *dir) |
3543 | return error; |
3544 | |
3545 | if (default_acl) { |
3546 | - error = orangefs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); |
3547 | + error = __orangefs_set_acl(inode, default_acl, |
3548 | + ACL_TYPE_DEFAULT); |
3549 | posix_acl_release(default_acl); |
3550 | } |
3551 | |
3552 | if (acl) { |
3553 | if (!error) |
3554 | - error = orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS); |
3555 | + error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS); |
3556 | posix_acl_release(acl); |
3557 | } |
3558 | |
3559 | diff --git a/include/linux/ccp.h b/include/linux/ccp.h |
3560 | index edc5d04b9632..1cfe5ef3060b 100644 |
3561 | --- a/include/linux/ccp.h |
3562 | +++ b/include/linux/ccp.h |
3563 | @@ -1,7 +1,7 @@ |
3564 | /* |
3565 | * AMD Cryptographic Coprocessor (CCP) driver |
3566 | * |
3567 | - * Copyright (C) 2013,2016 Advanced Micro Devices, Inc. |
3568 | + * Copyright (C) 2013,2017 Advanced Micro Devices, Inc. |
3569 | * |
3570 | * Author: Tom Lendacky <thomas.lendacky@amd.com> |
3571 | * Author: Gary R Hook <gary.hook@amd.com> |
3572 | @@ -222,6 +222,7 @@ enum ccp_xts_aes_unit_size { |
3573 | * AES operation the new IV overwrites the old IV. |
3574 | */ |
3575 | struct ccp_xts_aes_engine { |
3576 | + enum ccp_aes_type type; |
3577 | enum ccp_aes_action action; |
3578 | enum ccp_xts_aes_unit_size unit_size; |
3579 | |
3580 | diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h |
3581 | index 7321ae933867..102c84dcc11a 100644 |
3582 | --- a/include/linux/sunrpc/svc.h |
3583 | +++ b/include/linux/sunrpc/svc.h |
3584 | @@ -470,6 +470,7 @@ void svc_pool_map_put(void); |
3585 | struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int, |
3586 | struct svc_serv_ops *); |
3587 | int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int); |
3588 | +int svc_set_num_threads_sync(struct svc_serv *, struct svc_pool *, int); |
3589 | int svc_pool_stats_open(struct svc_serv *serv, struct file *file); |
3590 | void svc_destroy(struct svc_serv *); |
3591 | void svc_shutdown_net(struct svc_serv *, struct net *); |
3592 | diff --git a/include/linux/tty_flip.h b/include/linux/tty_flip.h |
3593 | index c28dd523f96e..d43837f2ce3a 100644 |
3594 | --- a/include/linux/tty_flip.h |
3595 | +++ b/include/linux/tty_flip.h |
3596 | @@ -12,6 +12,7 @@ extern int tty_prepare_flip_string(struct tty_port *port, |
3597 | unsigned char **chars, size_t size); |
3598 | extern void tty_flip_buffer_push(struct tty_port *port); |
3599 | void tty_schedule_flip(struct tty_port *port); |
3600 | +int __tty_insert_flip_char(struct tty_port *port, unsigned char ch, char flag); |
3601 | |
3602 | static inline int tty_insert_flip_char(struct tty_port *port, |
3603 | unsigned char ch, char flag) |
3604 | @@ -26,7 +27,7 @@ static inline int tty_insert_flip_char(struct tty_port *port, |
3605 | *char_buf_ptr(tb, tb->used++) = ch; |
3606 | return 1; |
3607 | } |
3608 | - return tty_insert_flip_string_flags(port, &ch, &flag, 1); |
3609 | + return __tty_insert_flip_char(port, ch, flag); |
3610 | } |
3611 | |
3612 | static inline int tty_insert_flip_string(struct tty_port *port, |
3613 | diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c |
3614 | index 6e432ed7d0fe..53ed8ae5de1c 100644 |
3615 | --- a/kernel/trace/ftrace.c |
3616 | +++ b/kernel/trace/ftrace.c |
3617 | @@ -2747,13 +2747,14 @@ static int ftrace_shutdown(struct ftrace_ops *ops, int command) |
3618 | |
3619 | if (!command || !ftrace_enabled) { |
3620 | /* |
3621 | - * If these are per_cpu ops, they still need their |
3622 | - * per_cpu field freed. Since, function tracing is |
3623 | + * If these are dynamic or per_cpu ops, they still |
3624 | + * need their data freed. Since, function tracing is |
3625 | * not currently active, we can just free them |
3626 | * without synchronizing all CPUs. |
3627 | */ |
3628 | - if (ops->flags & FTRACE_OPS_FL_PER_CPU) |
3629 | - per_cpu_ops_free(ops); |
3630 | + if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) |
3631 | + goto free_ops; |
3632 | + |
3633 | return 0; |
3634 | } |
3635 | |
3636 | @@ -2808,6 +2809,7 @@ static int ftrace_shutdown(struct ftrace_ops *ops, int command) |
3637 | if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) { |
3638 | schedule_on_each_cpu(ftrace_sync); |
3639 | |
3640 | + free_ops: |
3641 | arch_ftrace_trampoline_free(ops); |
3642 | |
3643 | if (ops->flags & FTRACE_OPS_FL_PER_CPU) |
3644 | diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c |
3645 | index 7379f735a9f4..f95bf81529f5 100644 |
3646 | --- a/kernel/trace/trace.c |
3647 | +++ b/kernel/trace/trace.c |
3648 | @@ -2369,11 +2369,17 @@ static char *get_trace_buf(void) |
3649 | if (!buffer || buffer->nesting >= 4) |
3650 | return NULL; |
3651 | |
3652 | - return &buffer->buffer[buffer->nesting++][0]; |
3653 | + buffer->nesting++; |
3654 | + |
3655 | + /* Interrupts must see nesting incremented before we use the buffer */ |
3656 | + barrier(); |
3657 | + return &buffer->buffer[buffer->nesting][0]; |
3658 | } |
3659 | |
3660 | static void put_trace_buf(void) |
3661 | { |
3662 | + /* Don't let the decrement of nesting leak before this */ |
3663 | + barrier(); |
3664 | this_cpu_dec(trace_percpu_buffer->nesting); |
3665 | } |
3666 | |
3667 | @@ -5658,7 +5664,7 @@ static int tracing_set_clock(struct trace_array *tr, const char *clockstr) |
3668 | tracing_reset_online_cpus(&tr->trace_buffer); |
3669 | |
3670 | #ifdef CONFIG_TRACER_MAX_TRACE |
3671 | - if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer) |
3672 | + if (tr->max_buffer.buffer) |
3673 | ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func); |
3674 | tracing_reset_online_cpus(&tr->max_buffer); |
3675 | #endif |
3676 | diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c |
3677 | index b0f86ea77881..ca70d11b8aa7 100644 |
3678 | --- a/kernel/trace/trace_selftest.c |
3679 | +++ b/kernel/trace/trace_selftest.c |
3680 | @@ -272,7 +272,7 @@ static int trace_selftest_ops(struct trace_array *tr, int cnt) |
3681 | goto out_free; |
3682 | if (cnt > 1) { |
3683 | if (trace_selftest_test_global_cnt == 0) |
3684 | - goto out; |
3685 | + goto out_free; |
3686 | } |
3687 | if (trace_selftest_test_dyn_cnt == 0) |
3688 | goto out_free; |
3689 | diff --git a/mm/page_alloc.c b/mm/page_alloc.c |
3690 | index 2abf8d5f0ad4..7064aae8ded7 100644 |
3691 | --- a/mm/page_alloc.c |
3692 | +++ b/mm/page_alloc.c |
3693 | @@ -2100,13 +2100,25 @@ static void unreserve_highatomic_pageblock(const struct alloc_context *ac) |
3694 | continue; |
3695 | |
3696 | /* |
3697 | - * It should never happen but changes to locking could |
3698 | - * inadvertently allow a per-cpu drain to add pages |
3699 | - * to MIGRATE_HIGHATOMIC while unreserving so be safe |
3700 | - * and watch for underflows. |
3701 | + * In page freeing path, migratetype change is racy so |
3702 | + * we can counter several free pages in a pageblock |
3703 | + * in this loop althoug we changed the pageblock type |
3704 | + * from highatomic to ac->migratetype. So we should |
3705 | + * adjust the count once. |
3706 | */ |
3707 | - zone->nr_reserved_highatomic -= min(pageblock_nr_pages, |
3708 | - zone->nr_reserved_highatomic); |
3709 | + if (get_pageblock_migratetype(page) == |
3710 | + MIGRATE_HIGHATOMIC) { |
3711 | + /* |
3712 | + * It should never happen but changes to |
3713 | + * locking could inadvertently allow a per-cpu |
3714 | + * drain to add pages to MIGRATE_HIGHATOMIC |
3715 | + * while unreserving so be safe and watch for |
3716 | + * underflows. |
3717 | + */ |
3718 | + zone->nr_reserved_highatomic -= min( |
3719 | + pageblock_nr_pages, |
3720 | + zone->nr_reserved_highatomic); |
3721 | + } |
3722 | |
3723 | /* |
3724 | * Convert to ac->migratetype and avoid the normal |
3725 | diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c |
3726 | index 6bd150882ba4..ed9ce7c63252 100644 |
3727 | --- a/net/netfilter/nf_conntrack_core.c |
3728 | +++ b/net/netfilter/nf_conntrack_core.c |
3729 | @@ -95,19 +95,26 @@ static struct conntrack_gc_work conntrack_gc_work; |
3730 | |
3731 | void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) |
3732 | { |
3733 | + /* 1) Acquire the lock */ |
3734 | spin_lock(lock); |
3735 | - while (unlikely(nf_conntrack_locks_all)) { |
3736 | - spin_unlock(lock); |
3737 | |
3738 | - /* |
3739 | - * Order the 'nf_conntrack_locks_all' load vs. the |
3740 | - * spin_unlock_wait() loads below, to ensure |
3741 | - * that 'nf_conntrack_locks_all_lock' is indeed held: |
3742 | - */ |
3743 | - smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ |
3744 | - spin_unlock_wait(&nf_conntrack_locks_all_lock); |
3745 | - spin_lock(lock); |
3746 | - } |
3747 | + /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics |
3748 | + * It pairs with the smp_store_release() in nf_conntrack_all_unlock() |
3749 | + */ |
3750 | + if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false)) |
3751 | + return; |
3752 | + |
3753 | + /* fast path failed, unlock */ |
3754 | + spin_unlock(lock); |
3755 | + |
3756 | + /* Slow path 1) get global lock */ |
3757 | + spin_lock(&nf_conntrack_locks_all_lock); |
3758 | + |
3759 | + /* Slow path 2) get the lock we want */ |
3760 | + spin_lock(lock); |
3761 | + |
3762 | + /* Slow path 3) release the global lock */ |
3763 | + spin_unlock(&nf_conntrack_locks_all_lock); |
3764 | } |
3765 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
3766 | |
3767 | @@ -148,28 +155,27 @@ static void nf_conntrack_all_lock(void) |
3768 | int i; |
3769 | |
3770 | spin_lock(&nf_conntrack_locks_all_lock); |
3771 | - nf_conntrack_locks_all = true; |
3772 | |
3773 | - /* |
3774 | - * Order the above store of 'nf_conntrack_locks_all' against |
3775 | - * the spin_unlock_wait() loads below, such that if |
3776 | - * nf_conntrack_lock() observes 'nf_conntrack_locks_all' |
3777 | - * we must observe nf_conntrack_locks[] held: |
3778 | - */ |
3779 | - smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ |
3780 | + nf_conntrack_locks_all = true; |
3781 | |
3782 | for (i = 0; i < CONNTRACK_LOCKS; i++) { |
3783 | - spin_unlock_wait(&nf_conntrack_locks[i]); |
3784 | + spin_lock(&nf_conntrack_locks[i]); |
3785 | + |
3786 | + /* This spin_unlock provides the "release" to ensure that |
3787 | + * nf_conntrack_locks_all==true is visible to everyone that |
3788 | + * acquired spin_lock(&nf_conntrack_locks[]). |
3789 | + */ |
3790 | + spin_unlock(&nf_conntrack_locks[i]); |
3791 | } |
3792 | } |
3793 | |
3794 | static void nf_conntrack_all_unlock(void) |
3795 | { |
3796 | - /* |
3797 | - * All prior stores must be complete before we clear |
3798 | + /* All prior stores must be complete before we clear |
3799 | * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock() |
3800 | * might observe the false value but not the entire |
3801 | - * critical section: |
3802 | + * critical section. |
3803 | + * It pairs with the smp_load_acquire() in nf_conntrack_lock() |
3804 | */ |
3805 | smp_store_release(&nf_conntrack_locks_all, false); |
3806 | spin_unlock(&nf_conntrack_locks_all_lock); |
3807 | diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c |
3808 | index 75f290bddca1..272c34551979 100644 |
3809 | --- a/net/sunrpc/svc.c |
3810 | +++ b/net/sunrpc/svc.c |
3811 | @@ -702,59 +702,32 @@ choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) |
3812 | return task; |
3813 | } |
3814 | |
3815 | -/* |
3816 | - * Create or destroy enough new threads to make the number |
3817 | - * of threads the given number. If `pool' is non-NULL, applies |
3818 | - * only to threads in that pool, otherwise round-robins between |
3819 | - * all pools. Caller must ensure that mutual exclusion between this and |
3820 | - * server startup or shutdown. |
3821 | - * |
3822 | - * Destroying threads relies on the service threads filling in |
3823 | - * rqstp->rq_task, which only the nfs ones do. Assumes the serv |
3824 | - * has been created using svc_create_pooled(). |
3825 | - * |
3826 | - * Based on code that used to be in nfsd_svc() but tweaked |
3827 | - * to be pool-aware. |
3828 | - */ |
3829 | -int |
3830 | -svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3831 | +/* create new threads */ |
3832 | +static int |
3833 | +svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3834 | { |
3835 | struct svc_rqst *rqstp; |
3836 | struct task_struct *task; |
3837 | struct svc_pool *chosen_pool; |
3838 | - int error = 0; |
3839 | unsigned int state = serv->sv_nrthreads-1; |
3840 | int node; |
3841 | |
3842 | - if (pool == NULL) { |
3843 | - /* The -1 assumes caller has done a svc_get() */ |
3844 | - nrservs -= (serv->sv_nrthreads-1); |
3845 | - } else { |
3846 | - spin_lock_bh(&pool->sp_lock); |
3847 | - nrservs -= pool->sp_nrthreads; |
3848 | - spin_unlock_bh(&pool->sp_lock); |
3849 | - } |
3850 | - |
3851 | - /* create new threads */ |
3852 | - while (nrservs > 0) { |
3853 | + do { |
3854 | nrservs--; |
3855 | chosen_pool = choose_pool(serv, pool, &state); |
3856 | |
3857 | node = svc_pool_map_get_node(chosen_pool->sp_id); |
3858 | rqstp = svc_prepare_thread(serv, chosen_pool, node); |
3859 | - if (IS_ERR(rqstp)) { |
3860 | - error = PTR_ERR(rqstp); |
3861 | - break; |
3862 | - } |
3863 | + if (IS_ERR(rqstp)) |
3864 | + return PTR_ERR(rqstp); |
3865 | |
3866 | __module_get(serv->sv_ops->svo_module); |
3867 | task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp, |
3868 | node, "%s", serv->sv_name); |
3869 | if (IS_ERR(task)) { |
3870 | - error = PTR_ERR(task); |
3871 | module_put(serv->sv_ops->svo_module); |
3872 | svc_exit_thread(rqstp); |
3873 | - break; |
3874 | + return PTR_ERR(task); |
3875 | } |
3876 | |
3877 | rqstp->rq_task = task; |
3878 | @@ -763,18 +736,103 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3879 | |
3880 | svc_sock_update_bufs(serv); |
3881 | wake_up_process(task); |
3882 | - } |
3883 | + } while (nrservs > 0); |
3884 | + |
3885 | + return 0; |
3886 | +} |
3887 | + |
3888 | + |
3889 | +/* destroy old threads */ |
3890 | +static int |
3891 | +svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3892 | +{ |
3893 | + struct task_struct *task; |
3894 | + unsigned int state = serv->sv_nrthreads-1; |
3895 | + |
3896 | /* destroy old threads */ |
3897 | - while (nrservs < 0 && |
3898 | - (task = choose_victim(serv, pool, &state)) != NULL) { |
3899 | + do { |
3900 | + task = choose_victim(serv, pool, &state); |
3901 | + if (task == NULL) |
3902 | + break; |
3903 | send_sig(SIGINT, task, 1); |
3904 | nrservs++; |
3905 | + } while (nrservs < 0); |
3906 | + |
3907 | + return 0; |
3908 | +} |
3909 | + |
3910 | +/* |
3911 | + * Create or destroy enough new threads to make the number |
3912 | + * of threads the given number. If `pool' is non-NULL, applies |
3913 | + * only to threads in that pool, otherwise round-robins between |
3914 | + * all pools. Caller must ensure that mutual exclusion between this and |
3915 | + * server startup or shutdown. |
3916 | + * |
3917 | + * Destroying threads relies on the service threads filling in |
3918 | + * rqstp->rq_task, which only the nfs ones do. Assumes the serv |
3919 | + * has been created using svc_create_pooled(). |
3920 | + * |
3921 | + * Based on code that used to be in nfsd_svc() but tweaked |
3922 | + * to be pool-aware. |
3923 | + */ |
3924 | +int |
3925 | +svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3926 | +{ |
3927 | + if (pool == NULL) { |
3928 | + /* The -1 assumes caller has done a svc_get() */ |
3929 | + nrservs -= (serv->sv_nrthreads-1); |
3930 | + } else { |
3931 | + spin_lock_bh(&pool->sp_lock); |
3932 | + nrservs -= pool->sp_nrthreads; |
3933 | + spin_unlock_bh(&pool->sp_lock); |
3934 | } |
3935 | |
3936 | - return error; |
3937 | + if (nrservs > 0) |
3938 | + return svc_start_kthreads(serv, pool, nrservs); |
3939 | + if (nrservs < 0) |
3940 | + return svc_signal_kthreads(serv, pool, nrservs); |
3941 | + return 0; |
3942 | } |
3943 | EXPORT_SYMBOL_GPL(svc_set_num_threads); |
3944 | |
3945 | +/* destroy old threads */ |
3946 | +static int |
3947 | +svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3948 | +{ |
3949 | + struct task_struct *task; |
3950 | + unsigned int state = serv->sv_nrthreads-1; |
3951 | + |
3952 | + /* destroy old threads */ |
3953 | + do { |
3954 | + task = choose_victim(serv, pool, &state); |
3955 | + if (task == NULL) |
3956 | + break; |
3957 | + kthread_stop(task); |
3958 | + nrservs++; |
3959 | + } while (nrservs < 0); |
3960 | + return 0; |
3961 | +} |
3962 | + |
3963 | +int |
3964 | +svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
3965 | +{ |
3966 | + if (pool == NULL) { |
3967 | + /* The -1 assumes caller has done a svc_get() */ |
3968 | + nrservs -= (serv->sv_nrthreads-1); |
3969 | + } else { |
3970 | + spin_lock_bh(&pool->sp_lock); |
3971 | + nrservs -= pool->sp_nrthreads; |
3972 | + spin_unlock_bh(&pool->sp_lock); |
3973 | + } |
3974 | + |
3975 | + if (nrservs > 0) |
3976 | + return svc_start_kthreads(serv, pool, nrservs); |
3977 | + if (nrservs < 0) |
3978 | + return svc_stop_kthreads(serv, pool, nrservs); |
3979 | + return 0; |
3980 | +} |
3981 | +EXPORT_SYMBOL_GPL(svc_set_num_threads_sync); |
3982 | + |
3983 | /* |
3984 | * Called from a server thread as it's exiting. Caller must hold the "service |
3985 | * mutex" for the service. |