diff options
author | Ralf Baechle <ralf@linux-mips.org> | 2006-10-30 22:45:07 -0500 |
---|---|---|
committer | Ralf Baechle <ralf@linux-mips.org> | 2006-12-04 17:43:14 -0500 |
commit | 0004a9dfeaa709a7f853487aba19932c9b1a87c8 (patch) | |
tree | e9f1f4b1ca897e57f46778cef283617ba83fc855 | |
parent | 08f57f7ffe5819e537301b1f1109fa4fc670bfff (diff) |
[MIPS] Cleanup memory barriers for weakly ordered systems.
Also the R4000 / R4600 LL/SC instructions imply a sync so no explicit sync
needed.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
-rw-r--r-- | arch/mips/Kconfig | 4 | ||||
-rw-r--r-- | arch/mips/kernel/smp.c | 6 | ||||
-rw-r--r-- | include/asm-mips/atomic.h | 37 | ||||
-rw-r--r-- | include/asm-mips/barrier.h | 132 | ||||
-rw-r--r-- | include/asm-mips/bitops.h | 27 | ||||
-rw-r--r-- | include/asm-mips/futex.h | 22 | ||||
-rw-r--r-- | include/asm-mips/spinlock.h | 53 | ||||
-rw-r--r-- | include/asm-mips/system.h | 156 |
8 files changed, 225 insertions, 212 deletions
diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig index 3947e5daea39..4d64960be035 100644 --- a/arch/mips/Kconfig +++ b/arch/mips/Kconfig | |||
@@ -1277,6 +1277,7 @@ config CPU_RM9000 | |||
1277 | select CPU_SUPPORTS_32BIT_KERNEL | 1277 | select CPU_SUPPORTS_32BIT_KERNEL |
1278 | select CPU_SUPPORTS_64BIT_KERNEL | 1278 | select CPU_SUPPORTS_64BIT_KERNEL |
1279 | select CPU_SUPPORTS_HIGHMEM | 1279 | select CPU_SUPPORTS_HIGHMEM |
1280 | select WEAK_ORDERING | ||
1280 | 1281 | ||
1281 | config CPU_SB1 | 1282 | config CPU_SB1 |
1282 | bool "SB1" | 1283 | bool "SB1" |
@@ -1285,6 +1286,7 @@ config CPU_SB1 | |||
1285 | select CPU_SUPPORTS_32BIT_KERNEL | 1286 | select CPU_SUPPORTS_32BIT_KERNEL |
1286 | select CPU_SUPPORTS_64BIT_KERNEL | 1287 | select CPU_SUPPORTS_64BIT_KERNEL |
1287 | select CPU_SUPPORTS_HIGHMEM | 1288 | select CPU_SUPPORTS_HIGHMEM |
1289 | select WEAK_ORDERING | ||
1288 | 1290 | ||
1289 | endchoice | 1291 | endchoice |
1290 | 1292 | ||
@@ -1345,6 +1347,8 @@ config SYS_HAS_CPU_RM9000 | |||
1345 | config SYS_HAS_CPU_SB1 | 1347 | config SYS_HAS_CPU_SB1 |
1346 | bool | 1348 | bool |
1347 | 1349 | ||
1350 | config WEAK_ORDERING | ||
1351 | bool | ||
1348 | endmenu | 1352 | endmenu |
1349 | 1353 | ||
1350 | # | 1354 | # |
diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c index 49db516789e0..f2a8701e414d 100644 --- a/arch/mips/kernel/smp.c +++ b/arch/mips/kernel/smp.c | |||
@@ -172,7 +172,7 @@ int smp_call_function (void (*func) (void *info), void *info, int retry, | |||
172 | 172 | ||
173 | spin_lock(&smp_call_lock); | 173 | spin_lock(&smp_call_lock); |
174 | call_data = &data; | 174 | call_data = &data; |
175 | mb(); | 175 | smp_mb(); |
176 | 176 | ||
177 | /* Send a message to all other CPUs and wait for them to respond */ | 177 | /* Send a message to all other CPUs and wait for them to respond */ |
178 | for_each_online_cpu(i) | 178 | for_each_online_cpu(i) |
@@ -204,7 +204,7 @@ void smp_call_function_interrupt(void) | |||
204 | * Notify initiating CPU that I've grabbed the data and am | 204 | * Notify initiating CPU that I've grabbed the data and am |
205 | * about to execute the function. | 205 | * about to execute the function. |
206 | */ | 206 | */ |
207 | mb(); | 207 | smp_mb(); |
208 | atomic_inc(&call_data->started); | 208 | atomic_inc(&call_data->started); |
209 | 209 | ||
210 | /* | 210 | /* |
@@ -215,7 +215,7 @@ void smp_call_function_interrupt(void) | |||
215 | irq_exit(); | 215 | irq_exit(); |
216 | 216 | ||
217 | if (wait) { | 217 | if (wait) { |
218 | mb(); | 218 | smp_mb(); |
219 | atomic_inc(&call_data->finished); | 219 | atomic_inc(&call_data->finished); |
220 | } | 220 | } |
221 | } | 221 | } |
diff --git a/include/asm-mips/atomic.h b/include/asm-mips/atomic.h index 365767074c79..c1a2409bb52a 100644 --- a/include/asm-mips/atomic.h +++ b/include/asm-mips/atomic.h | |||
@@ -15,6 +15,7 @@ | |||
15 | #define _ASM_ATOMIC_H | 15 | #define _ASM_ATOMIC_H |
16 | 16 | ||
17 | #include <linux/irqflags.h> | 17 | #include <linux/irqflags.h> |
18 | #include <asm/barrier.h> | ||
18 | #include <asm/cpu-features.h> | 19 | #include <asm/cpu-features.h> |
19 | #include <asm/war.h> | 20 | #include <asm/war.h> |
20 | 21 | ||
@@ -130,6 +131,8 @@ static __inline__ int atomic_add_return(int i, atomic_t * v) | |||
130 | { | 131 | { |
131 | unsigned long result; | 132 | unsigned long result; |
132 | 133 | ||
134 | smp_mb(); | ||
135 | |||
133 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 136 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
134 | unsigned long temp; | 137 | unsigned long temp; |
135 | 138 | ||
@@ -140,7 +143,6 @@ static __inline__ int atomic_add_return(int i, atomic_t * v) | |||
140 | " sc %0, %2 \n" | 143 | " sc %0, %2 \n" |
141 | " beqzl %0, 1b \n" | 144 | " beqzl %0, 1b \n" |
142 | " addu %0, %1, %3 \n" | 145 | " addu %0, %1, %3 \n" |
143 | " sync \n" | ||
144 | " .set mips0 \n" | 146 | " .set mips0 \n" |
145 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 147 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
146 | : "Ir" (i), "m" (v->counter) | 148 | : "Ir" (i), "m" (v->counter) |
@@ -155,7 +157,6 @@ static __inline__ int atomic_add_return(int i, atomic_t * v) | |||
155 | " sc %0, %2 \n" | 157 | " sc %0, %2 \n" |
156 | " beqz %0, 1b \n" | 158 | " beqz %0, 1b \n" |
157 | " addu %0, %1, %3 \n" | 159 | " addu %0, %1, %3 \n" |
158 | " sync \n" | ||
159 | " .set mips0 \n" | 160 | " .set mips0 \n" |
160 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 161 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
161 | : "Ir" (i), "m" (v->counter) | 162 | : "Ir" (i), "m" (v->counter) |
@@ -170,6 +171,8 @@ static __inline__ int atomic_add_return(int i, atomic_t * v) | |||
170 | local_irq_restore(flags); | 171 | local_irq_restore(flags); |
171 | } | 172 | } |
172 | 173 | ||
174 | smp_mb(); | ||
175 | |||
173 | return result; | 176 | return result; |
174 | } | 177 | } |
175 | 178 | ||
@@ -177,6 +180,8 @@ static __inline__ int atomic_sub_return(int i, atomic_t * v) | |||
177 | { | 180 | { |
178 | unsigned long result; | 181 | unsigned long result; |
179 | 182 | ||
183 | smp_mb(); | ||
184 | |||
180 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 185 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
181 | unsigned long temp; | 186 | unsigned long temp; |
182 | 187 | ||
@@ -187,7 +192,6 @@ static __inline__ int atomic_sub_return(int i, atomic_t * v) | |||
187 | " sc %0, %2 \n" | 192 | " sc %0, %2 \n" |
188 | " beqzl %0, 1b \n" | 193 | " beqzl %0, 1b \n" |
189 | " subu %0, %1, %3 \n" | 194 | " subu %0, %1, %3 \n" |
190 | " sync \n" | ||
191 | " .set mips0 \n" | 195 | " .set mips0 \n" |
192 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 196 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
193 | : "Ir" (i), "m" (v->counter) | 197 | : "Ir" (i), "m" (v->counter) |
@@ -202,7 +206,6 @@ static __inline__ int atomic_sub_return(int i, atomic_t * v) | |||
202 | " sc %0, %2 \n" | 206 | " sc %0, %2 \n" |
203 | " beqz %0, 1b \n" | 207 | " beqz %0, 1b \n" |
204 | " subu %0, %1, %3 \n" | 208 | " subu %0, %1, %3 \n" |
205 | " sync \n" | ||
206 | " .set mips0 \n" | 209 | " .set mips0 \n" |
207 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 210 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
208 | : "Ir" (i), "m" (v->counter) | 211 | : "Ir" (i), "m" (v->counter) |
@@ -217,6 +220,8 @@ static __inline__ int atomic_sub_return(int i, atomic_t * v) | |||
217 | local_irq_restore(flags); | 220 | local_irq_restore(flags); |
218 | } | 221 | } |
219 | 222 | ||
223 | smp_mb(); | ||
224 | |||
220 | return result; | 225 | return result; |
221 | } | 226 | } |
222 | 227 | ||
@@ -232,6 +237,8 @@ static __inline__ int atomic_sub_if_positive(int i, atomic_t * v) | |||
232 | { | 237 | { |
233 | unsigned long result; | 238 | unsigned long result; |
234 | 239 | ||
240 | smp_mb(); | ||
241 | |||
235 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 242 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
236 | unsigned long temp; | 243 | unsigned long temp; |
237 | 244 | ||
@@ -245,7 +252,6 @@ static __inline__ int atomic_sub_if_positive(int i, atomic_t * v) | |||
245 | " beqzl %0, 1b \n" | 252 | " beqzl %0, 1b \n" |
246 | " subu %0, %1, %3 \n" | 253 | " subu %0, %1, %3 \n" |
247 | " .set reorder \n" | 254 | " .set reorder \n" |
248 | " sync \n" | ||
249 | "1: \n" | 255 | "1: \n" |
250 | " .set mips0 \n" | 256 | " .set mips0 \n" |
251 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 257 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
@@ -264,7 +270,6 @@ static __inline__ int atomic_sub_if_positive(int i, atomic_t * v) | |||
264 | " beqz %0, 1b \n" | 270 | " beqz %0, 1b \n" |
265 | " subu %0, %1, %3 \n" | 271 | " subu %0, %1, %3 \n" |
266 | " .set reorder \n" | 272 | " .set reorder \n" |
267 | " sync \n" | ||
268 | "1: \n" | 273 | "1: \n" |
269 | " .set mips0 \n" | 274 | " .set mips0 \n" |
270 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 275 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
@@ -281,6 +286,8 @@ static __inline__ int atomic_sub_if_positive(int i, atomic_t * v) | |||
281 | local_irq_restore(flags); | 286 | local_irq_restore(flags); |
282 | } | 287 | } |
283 | 288 | ||
289 | smp_mb(); | ||
290 | |||
284 | return result; | 291 | return result; |
285 | } | 292 | } |
286 | 293 | ||
@@ -484,6 +491,8 @@ static __inline__ long atomic64_add_return(long i, atomic64_t * v) | |||
484 | { | 491 | { |
485 | unsigned long result; | 492 | unsigned long result; |
486 | 493 | ||
494 | smp_mb(); | ||
495 | |||
487 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 496 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
488 | unsigned long temp; | 497 | unsigned long temp; |
489 | 498 | ||
@@ -494,7 +503,6 @@ static __inline__ long atomic64_add_return(long i, atomic64_t * v) | |||
494 | " scd %0, %2 \n" | 503 | " scd %0, %2 \n" |
495 | " beqzl %0, 1b \n" | 504 | " beqzl %0, 1b \n" |
496 | " addu %0, %1, %3 \n" | 505 | " addu %0, %1, %3 \n" |
497 | " sync \n" | ||
498 | " .set mips0 \n" | 506 | " .set mips0 \n" |
499 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 507 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
500 | : "Ir" (i), "m" (v->counter) | 508 | : "Ir" (i), "m" (v->counter) |
@@ -509,7 +517,6 @@ static __inline__ long atomic64_add_return(long i, atomic64_t * v) | |||
509 | " scd %0, %2 \n" | 517 | " scd %0, %2 \n" |
510 | " beqz %0, 1b \n" | 518 | " beqz %0, 1b \n" |
511 | " addu %0, %1, %3 \n" | 519 | " addu %0, %1, %3 \n" |
512 | " sync \n" | ||
513 | " .set mips0 \n" | 520 | " .set mips0 \n" |
514 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 521 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
515 | : "Ir" (i), "m" (v->counter) | 522 | : "Ir" (i), "m" (v->counter) |
@@ -524,6 +531,8 @@ static __inline__ long atomic64_add_return(long i, atomic64_t * v) | |||
524 | local_irq_restore(flags); | 531 | local_irq_restore(flags); |
525 | } | 532 | } |
526 | 533 | ||
534 | smp_mb(); | ||
535 | |||
527 | return result; | 536 | return result; |
528 | } | 537 | } |
529 | 538 | ||
@@ -531,6 +540,8 @@ static __inline__ long atomic64_sub_return(long i, atomic64_t * v) | |||
531 | { | 540 | { |
532 | unsigned long result; | 541 | unsigned long result; |
533 | 542 | ||
543 | smp_mb(); | ||
544 | |||
534 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 545 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
535 | unsigned long temp; | 546 | unsigned long temp; |
536 | 547 | ||
@@ -541,7 +552,6 @@ static __inline__ long atomic64_sub_return(long i, atomic64_t * v) | |||
541 | " scd %0, %2 \n" | 552 | " scd %0, %2 \n" |
542 | " beqzl %0, 1b \n" | 553 | " beqzl %0, 1b \n" |
543 | " subu %0, %1, %3 \n" | 554 | " subu %0, %1, %3 \n" |
544 | " sync \n" | ||
545 | " .set mips0 \n" | 555 | " .set mips0 \n" |
546 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 556 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
547 | : "Ir" (i), "m" (v->counter) | 557 | : "Ir" (i), "m" (v->counter) |
@@ -556,7 +566,6 @@ static __inline__ long atomic64_sub_return(long i, atomic64_t * v) | |||
556 | " scd %0, %2 \n" | 566 | " scd %0, %2 \n" |
557 | " beqz %0, 1b \n" | 567 | " beqz %0, 1b \n" |
558 | " subu %0, %1, %3 \n" | 568 | " subu %0, %1, %3 \n" |
559 | " sync \n" | ||
560 | " .set mips0 \n" | 569 | " .set mips0 \n" |
561 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 570 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
562 | : "Ir" (i), "m" (v->counter) | 571 | : "Ir" (i), "m" (v->counter) |
@@ -571,6 +580,8 @@ static __inline__ long atomic64_sub_return(long i, atomic64_t * v) | |||
571 | local_irq_restore(flags); | 580 | local_irq_restore(flags); |
572 | } | 581 | } |
573 | 582 | ||
583 | smp_mb(); | ||
584 | |||
574 | return result; | 585 | return result; |
575 | } | 586 | } |
576 | 587 | ||
@@ -586,6 +597,8 @@ static __inline__ long atomic64_sub_if_positive(long i, atomic64_t * v) | |||
586 | { | 597 | { |
587 | unsigned long result; | 598 | unsigned long result; |
588 | 599 | ||
600 | smp_mb(); | ||
601 | |||
589 | if (cpu_has_llsc && R10000_LLSC_WAR) { | 602 | if (cpu_has_llsc && R10000_LLSC_WAR) { |
590 | unsigned long temp; | 603 | unsigned long temp; |
591 | 604 | ||
@@ -599,7 +612,6 @@ static __inline__ long atomic64_sub_if_positive(long i, atomic64_t * v) | |||
599 | " beqzl %0, 1b \n" | 612 | " beqzl %0, 1b \n" |
600 | " dsubu %0, %1, %3 \n" | 613 | " dsubu %0, %1, %3 \n" |
601 | " .set reorder \n" | 614 | " .set reorder \n" |
602 | " sync \n" | ||
603 | "1: \n" | 615 | "1: \n" |
604 | " .set mips0 \n" | 616 | " .set mips0 \n" |
605 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 617 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
@@ -618,7 +630,6 @@ static __inline__ long atomic64_sub_if_positive(long i, atomic64_t * v) | |||
618 | " beqz %0, 1b \n" | 630 | " beqz %0, 1b \n" |
619 | " dsubu %0, %1, %3 \n" | 631 | " dsubu %0, %1, %3 \n" |
620 | " .set reorder \n" | 632 | " .set reorder \n" |
621 | " sync \n" | ||
622 | "1: \n" | 633 | "1: \n" |
623 | " .set mips0 \n" | 634 | " .set mips0 \n" |
624 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) | 635 | : "=&r" (result), "=&r" (temp), "=m" (v->counter) |
@@ -635,6 +646,8 @@ static __inline__ long atomic64_sub_if_positive(long i, atomic64_t * v) | |||
635 | local_irq_restore(flags); | 646 | local_irq_restore(flags); |
636 | } | 647 | } |
637 | 648 | ||
649 | smp_mb(); | ||
650 | |||
638 | return result; | 651 | return result; |
639 | } | 652 | } |
640 | 653 | ||
diff --git a/include/asm-mips/barrier.h b/include/asm-mips/barrier.h new file mode 100644 index 000000000000..ed82631b0017 --- /dev/null +++ b/include/asm-mips/barrier.h | |||
@@ -0,0 +1,132 @@ | |||
1 | /* | ||
2 | * This file is subject to the terms and conditions of the GNU General Public | ||
3 | * License. See the file "COPYING" in the main directory of this archive | ||
4 | * for more details. | ||
5 | * | ||
6 | * Copyright (C) 2006 by Ralf Baechle (ralf@linux-mips.org) | ||
7 | */ | ||
8 | #ifndef __ASM_BARRIER_H | ||
9 | #define __ASM_BARRIER_H | ||
10 | |||
11 | /* | ||
12 | * read_barrier_depends - Flush all pending reads that subsequents reads | ||
13 | * depend on. | ||
14 | * | ||
15 | * No data-dependent reads from memory-like regions are ever reordered | ||
16 | * over this barrier. All reads preceding this primitive are guaranteed | ||
17 | * to access memory (but not necessarily other CPUs' caches) before any | ||
18 | * reads following this primitive that depend on the data return by | ||
19 | * any of the preceding reads. This primitive is much lighter weight than | ||
20 | * rmb() on most CPUs, and is never heavier weight than is | ||
21 | * rmb(). | ||
22 | * | ||
23 | * These ordering constraints are respected by both the local CPU | ||
24 | * and the compiler. | ||
25 | * | ||
26 | * Ordering is not guaranteed by anything other than these primitives, | ||
27 | * not even by data dependencies. See the documentation for | ||
28 | * memory_barrier() for examples and URLs to more information. | ||
29 | * | ||
30 | * For example, the following code would force ordering (the initial | ||
31 | * value of "a" is zero, "b" is one, and "p" is "&a"): | ||
32 | * | ||
33 | * <programlisting> | ||
34 | * CPU 0 CPU 1 | ||
35 | * | ||
36 | * b = 2; | ||
37 | * memory_barrier(); | ||
38 | * p = &b; q = p; | ||
39 | * read_barrier_depends(); | ||
40 | * d = *q; | ||
41 | * </programlisting> | ||
42 | * | ||
43 | * because the read of "*q" depends on the read of "p" and these | ||
44 | * two reads are separated by a read_barrier_depends(). However, | ||
45 | * the following code, with the same initial values for "a" and "b": | ||
46 | * | ||
47 | * <programlisting> | ||
48 | * CPU 0 CPU 1 | ||
49 | * | ||
50 | * a = 2; | ||
51 | * memory_barrier(); | ||
52 | * b = 3; y = b; | ||
53 | * read_barrier_depends(); | ||
54 | * x = a; | ||
55 | * </programlisting> | ||
56 | * | ||
57 | * does not enforce ordering, since there is no data dependency between | ||
58 | * the read of "a" and the read of "b". Therefore, on some CPUs, such | ||
59 | * as Alpha, "y" could be set to 3 and "x" to 0. Use rmb() | ||
60 | * in cases like this where there are no data dependencies. | ||
61 | */ | ||
62 | |||
63 | #define read_barrier_depends() do { } while(0) | ||
64 | #define smp_read_barrier_depends() do { } while(0) | ||
65 | |||
66 | #ifdef CONFIG_CPU_HAS_SYNC | ||
67 | #define __sync() \ | ||
68 | __asm__ __volatile__( \ | ||
69 | ".set push\n\t" \ | ||
70 | ".set noreorder\n\t" \ | ||
71 | ".set mips2\n\t" \ | ||
72 | "sync\n\t" \ | ||
73 | ".set pop" \ | ||
74 | : /* no output */ \ | ||
75 | : /* no input */ \ | ||
76 | : "memory") | ||
77 | #else | ||
78 | #define __sync() do { } while(0) | ||
79 | #endif | ||
80 | |||
81 | #define __fast_iob() \ | ||
82 | __asm__ __volatile__( \ | ||
83 | ".set push\n\t" \ | ||
84 | ".set noreorder\n\t" \ | ||
85 | "lw $0,%0\n\t" \ | ||
86 | "nop\n\t" \ | ||
87 | ".set pop" \ | ||
88 | : /* no output */ \ | ||
89 | : "m" (*(int *)CKSEG1) \ | ||
90 | : "memory") | ||
91 | |||
92 | #define fast_wmb() __sync() | ||
93 | #define fast_rmb() __sync() | ||
94 | #define fast_mb() __sync() | ||
95 | #define fast_iob() \ | ||
96 | do { \ | ||
97 | __sync(); \ | ||
98 | __fast_iob(); \ | ||
99 | } while (0) | ||
100 | |||
101 | #ifdef CONFIG_CPU_HAS_WB | ||
102 | |||
103 | #include <asm/wbflush.h> | ||
104 | |||
105 | #define wmb() fast_wmb() | ||
106 | #define rmb() fast_rmb() | ||
107 | #define mb() wbflush() | ||
108 | #define iob() wbflush() | ||
109 | |||
110 | #else /* !CONFIG_CPU_HAS_WB */ | ||
111 | |||
112 | #define wmb() fast_wmb() | ||
113 | #define rmb() fast_rmb() | ||
114 | #define mb() fast_mb() | ||
115 | #define iob() fast_iob() | ||
116 | |||
117 | #endif /* !CONFIG_CPU_HAS_WB */ | ||
118 | |||
119 | #if defined(CONFIG_WEAK_ORDERING) && defined(CONFIG_SMP) | ||
120 | #define __WEAK_ORDERING_MB " sync \n" | ||
121 | #else | ||
122 | #define __WEAK_ORDERING_MB " \n" | ||
123 | #endif | ||
124 | |||
125 | #define smp_mb() __asm__ __volatile__(__WEAK_ORDERING_MB : : :"memory") | ||
126 | #define smp_rmb() __asm__ __volatile__(__WEAK_ORDERING_MB : : :"memory") | ||
127 | #define smp_wmb() __asm__ __volatile__(__WEAK_ORDERING_MB : : :"memory") | ||
128 | |||
129 | #define set_mb(var, value) \ | ||
130 | do { var = value; smp_mb(); } while (0) | ||
131 | |||
132 | #endif /* __ASM_BARRIER_H */ | ||
diff --git a/include/asm-mips/bitops.h b/include/asm-mips/bitops.h index b9007411b60f..06445de1324b 100644 --- a/include/asm-mips/bitops.h +++ b/include/asm-mips/bitops.h | |||
@@ -3,7 +3,7 @@ | |||
3 | * License. See the file "COPYING" in the main directory of this archive | 3 | * License. See the file "COPYING" in the main directory of this archive |
4 | * for more details. | 4 | * for more details. |
5 | * | 5 | * |
6 | * Copyright (c) 1994 - 1997, 1999, 2000 Ralf Baechle (ralf@gnu.org) | 6 | * Copyright (c) 1994 - 1997, 1999, 2000, 06 Ralf Baechle (ralf@linux-mips.org) |
7 | * Copyright (c) 1999, 2000 Silicon Graphics, Inc. | 7 | * Copyright (c) 1999, 2000 Silicon Graphics, Inc. |
8 | */ | 8 | */ |
9 | #ifndef _ASM_BITOPS_H | 9 | #ifndef _ASM_BITOPS_H |
@@ -12,6 +12,7 @@ | |||
12 | #include <linux/compiler.h> | 12 | #include <linux/compiler.h> |
13 | #include <linux/irqflags.h> | 13 | #include <linux/irqflags.h> |
14 | #include <linux/types.h> | 14 | #include <linux/types.h> |
15 | #include <asm/barrier.h> | ||
15 | #include <asm/bug.h> | 16 | #include <asm/bug.h> |
16 | #include <asm/byteorder.h> /* sigh ... */ | 17 | #include <asm/byteorder.h> /* sigh ... */ |
17 | #include <asm/cpu-features.h> | 18 | #include <asm/cpu-features.h> |
@@ -204,9 +205,6 @@ static inline int test_and_set_bit(unsigned long nr, | |||
204 | " " __SC "%2, %1 \n" | 205 | " " __SC "%2, %1 \n" |
205 | " beqzl %2, 1b \n" | 206 | " beqzl %2, 1b \n" |
206 | " and %2, %0, %3 \n" | 207 | " and %2, %0, %3 \n" |
207 | #ifdef CONFIG_SMP | ||
208 | " sync \n" | ||
209 | #endif | ||
210 | " .set mips0 \n" | 208 | " .set mips0 \n" |
211 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 209 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
212 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 210 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -226,9 +224,6 @@ static inline int test_and_set_bit(unsigned long nr, | |||
226 | " " __SC "%2, %1 \n" | 224 | " " __SC "%2, %1 \n" |
227 | " beqz %2, 1b \n" | 225 | " beqz %2, 1b \n" |
228 | " and %2, %0, %3 \n" | 226 | " and %2, %0, %3 \n" |
229 | #ifdef CONFIG_SMP | ||
230 | " sync \n" | ||
231 | #endif | ||
232 | " .set pop \n" | 227 | " .set pop \n" |
233 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 228 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
234 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 229 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -250,6 +245,8 @@ static inline int test_and_set_bit(unsigned long nr, | |||
250 | 245 | ||
251 | return retval; | 246 | return retval; |
252 | } | 247 | } |
248 | |||
249 | smp_mb(); | ||
253 | } | 250 | } |
254 | 251 | ||
255 | /* | 252 | /* |
@@ -275,9 +272,6 @@ static inline int test_and_clear_bit(unsigned long nr, | |||
275 | " " __SC "%2, %1 \n" | 272 | " " __SC "%2, %1 \n" |
276 | " beqzl %2, 1b \n" | 273 | " beqzl %2, 1b \n" |
277 | " and %2, %0, %3 \n" | 274 | " and %2, %0, %3 \n" |
278 | #ifdef CONFIG_SMP | ||
279 | " sync \n" | ||
280 | #endif | ||
281 | " .set mips0 \n" | 275 | " .set mips0 \n" |
282 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 276 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
283 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 277 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -298,9 +292,6 @@ static inline int test_and_clear_bit(unsigned long nr, | |||
298 | " " __SC "%2, %1 \n" | 292 | " " __SC "%2, %1 \n" |
299 | " beqz %2, 1b \n" | 293 | " beqz %2, 1b \n" |
300 | " and %2, %0, %3 \n" | 294 | " and %2, %0, %3 \n" |
301 | #ifdef CONFIG_SMP | ||
302 | " sync \n" | ||
303 | #endif | ||
304 | " .set pop \n" | 295 | " .set pop \n" |
305 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 296 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
306 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 297 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -322,6 +313,8 @@ static inline int test_and_clear_bit(unsigned long nr, | |||
322 | 313 | ||
323 | return retval; | 314 | return retval; |
324 | } | 315 | } |
316 | |||
317 | smp_mb(); | ||
325 | } | 318 | } |
326 | 319 | ||
327 | /* | 320 | /* |
@@ -346,9 +339,6 @@ static inline int test_and_change_bit(unsigned long nr, | |||
346 | " " __SC "%2, %1 \n" | 339 | " " __SC "%2, %1 \n" |
347 | " beqzl %2, 1b \n" | 340 | " beqzl %2, 1b \n" |
348 | " and %2, %0, %3 \n" | 341 | " and %2, %0, %3 \n" |
349 | #ifdef CONFIG_SMP | ||
350 | " sync \n" | ||
351 | #endif | ||
352 | " .set mips0 \n" | 342 | " .set mips0 \n" |
353 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 343 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
354 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 344 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -368,9 +358,6 @@ static inline int test_and_change_bit(unsigned long nr, | |||
368 | " " __SC "\t%2, %1 \n" | 358 | " " __SC "\t%2, %1 \n" |
369 | " beqz %2, 1b \n" | 359 | " beqz %2, 1b \n" |
370 | " and %2, %0, %3 \n" | 360 | " and %2, %0, %3 \n" |
371 | #ifdef CONFIG_SMP | ||
372 | " sync \n" | ||
373 | #endif | ||
374 | " .set pop \n" | 361 | " .set pop \n" |
375 | : "=&r" (temp), "=m" (*m), "=&r" (res) | 362 | : "=&r" (temp), "=m" (*m), "=&r" (res) |
376 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) | 363 | : "r" (1UL << (nr & SZLONG_MASK)), "m" (*m) |
@@ -391,6 +378,8 @@ static inline int test_and_change_bit(unsigned long nr, | |||
391 | 378 | ||
392 | return retval; | 379 | return retval; |
393 | } | 380 | } |
381 | |||
382 | smp_mb(); | ||
394 | } | 383 | } |
395 | 384 | ||
396 | #include <asm-generic/bitops/non-atomic.h> | 385 | #include <asm-generic/bitops/non-atomic.h> |
diff --git a/include/asm-mips/futex.h b/include/asm-mips/futex.h index ed023eae0674..927a216bd530 100644 --- a/include/asm-mips/futex.h +++ b/include/asm-mips/futex.h | |||
@@ -1,19 +1,21 @@ | |||
1 | /* | ||
2 | * This file is subject to the terms and conditions of the GNU General Public | ||
3 | * License. See the file "COPYING" in the main directory of this archive | ||
4 | * for more details. | ||
5 | * | ||
6 | * Copyright (c) 2006 Ralf Baechle (ralf@linux-mips.org) | ||
7 | */ | ||
1 | #ifndef _ASM_FUTEX_H | 8 | #ifndef _ASM_FUTEX_H |
2 | #define _ASM_FUTEX_H | 9 | #define _ASM_FUTEX_H |
3 | 10 | ||
4 | #ifdef __KERNEL__ | 11 | #ifdef __KERNEL__ |
5 | 12 | ||
6 | #include <linux/futex.h> | 13 | #include <linux/futex.h> |
14 | #include <asm/barrier.h> | ||
7 | #include <asm/errno.h> | 15 | #include <asm/errno.h> |
8 | #include <asm/uaccess.h> | 16 | #include <asm/uaccess.h> |
9 | #include <asm/war.h> | 17 | #include <asm/war.h> |
10 | 18 | ||
11 | #ifdef CONFIG_SMP | ||
12 | #define __FUTEX_SMP_SYNC " sync \n" | ||
13 | #else | ||
14 | #define __FUTEX_SMP_SYNC | ||
15 | #endif | ||
16 | |||
17 | #define __futex_atomic_op(insn, ret, oldval, uaddr, oparg) \ | 19 | #define __futex_atomic_op(insn, ret, oldval, uaddr, oparg) \ |
18 | { \ | 20 | { \ |
19 | if (cpu_has_llsc && R10000_LLSC_WAR) { \ | 21 | if (cpu_has_llsc && R10000_LLSC_WAR) { \ |
@@ -27,7 +29,7 @@ | |||
27 | " .set mips3 \n" \ | 29 | " .set mips3 \n" \ |
28 | "2: sc $1, %2 \n" \ | 30 | "2: sc $1, %2 \n" \ |
29 | " beqzl $1, 1b \n" \ | 31 | " beqzl $1, 1b \n" \ |
30 | __FUTEX_SMP_SYNC \ | 32 | __WEAK_ORDERING_MB \ |
31 | "3: \n" \ | 33 | "3: \n" \ |
32 | " .set pop \n" \ | 34 | " .set pop \n" \ |
33 | " .set mips0 \n" \ | 35 | " .set mips0 \n" \ |
@@ -53,7 +55,7 @@ | |||
53 | " .set mips3 \n" \ | 55 | " .set mips3 \n" \ |
54 | "2: sc $1, %2 \n" \ | 56 | "2: sc $1, %2 \n" \ |
55 | " beqz $1, 1b \n" \ | 57 | " beqz $1, 1b \n" \ |
56 | __FUTEX_SMP_SYNC \ | 58 | __WEAK_ORDERING_MB \ |
57 | "3: \n" \ | 59 | "3: \n" \ |
58 | " .set pop \n" \ | 60 | " .set pop \n" \ |
59 | " .set mips0 \n" \ | 61 | " .set mips0 \n" \ |
@@ -150,7 +152,7 @@ futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval, int newval) | |||
150 | " .set mips3 \n" | 152 | " .set mips3 \n" |
151 | "2: sc $1, %1 \n" | 153 | "2: sc $1, %1 \n" |
152 | " beqzl $1, 1b \n" | 154 | " beqzl $1, 1b \n" |
153 | __FUTEX_SMP_SYNC | 155 | __WEAK_ORDERING_MB |
154 | "3: \n" | 156 | "3: \n" |
155 | " .set pop \n" | 157 | " .set pop \n" |
156 | " .section .fixup,\"ax\" \n" | 158 | " .section .fixup,\"ax\" \n" |
@@ -177,7 +179,7 @@ futex_atomic_cmpxchg_inatomic(int __user *uaddr, int oldval, int newval) | |||
177 | " .set mips3 \n" | 179 | " .set mips3 \n" |
178 | "2: sc $1, %1 \n" | 180 | "2: sc $1, %1 \n" |
179 | " beqz $1, 1b \n" | 181 | " beqz $1, 1b \n" |
180 | __FUTEX_SMP_SYNC | 182 | __WEAK_ORDERING_MB |
181 | "3: \n" | 183 | "3: \n" |
182 | " .set pop \n" | 184 | " .set pop \n" |
183 | " .section .fixup,\"ax\" \n" | 185 | " .section .fixup,\"ax\" \n" |
diff --git a/include/asm-mips/spinlock.h b/include/asm-mips/spinlock.h index c8d5587467bb..fc3217fc1118 100644 --- a/include/asm-mips/spinlock.h +++ b/include/asm-mips/spinlock.h | |||
@@ -3,12 +3,13 @@ | |||
3 | * License. See the file "COPYING" in the main directory of this archive | 3 | * License. See the file "COPYING" in the main directory of this archive |
4 | * for more details. | 4 | * for more details. |
5 | * | 5 | * |
6 | * Copyright (C) 1999, 2000 by Ralf Baechle | 6 | * Copyright (C) 1999, 2000, 06 by Ralf Baechle |
7 | * Copyright (C) 1999, 2000 Silicon Graphics, Inc. | 7 | * Copyright (C) 1999, 2000 Silicon Graphics, Inc. |
8 | */ | 8 | */ |
9 | #ifndef _ASM_SPINLOCK_H | 9 | #ifndef _ASM_SPINLOCK_H |
10 | #define _ASM_SPINLOCK_H | 10 | #define _ASM_SPINLOCK_H |
11 | 11 | ||
12 | #include <asm/barrier.h> | ||
12 | #include <asm/war.h> | 13 | #include <asm/war.h> |
13 | 14 | ||
14 | /* | 15 | /* |
@@ -40,7 +41,6 @@ static inline void __raw_spin_lock(raw_spinlock_t *lock) | |||
40 | " sc %1, %0 \n" | 41 | " sc %1, %0 \n" |
41 | " beqzl %1, 1b \n" | 42 | " beqzl %1, 1b \n" |
42 | " nop \n" | 43 | " nop \n" |
43 | " sync \n" | ||
44 | " .set reorder \n" | 44 | " .set reorder \n" |
45 | : "=m" (lock->lock), "=&r" (tmp) | 45 | : "=m" (lock->lock), "=&r" (tmp) |
46 | : "m" (lock->lock) | 46 | : "m" (lock->lock) |
@@ -53,19 +53,22 @@ static inline void __raw_spin_lock(raw_spinlock_t *lock) | |||
53 | " li %1, 1 \n" | 53 | " li %1, 1 \n" |
54 | " sc %1, %0 \n" | 54 | " sc %1, %0 \n" |
55 | " beqz %1, 1b \n" | 55 | " beqz %1, 1b \n" |
56 | " sync \n" | 56 | " nop \n" |
57 | " .set reorder \n" | 57 | " .set reorder \n" |
58 | : "=m" (lock->lock), "=&r" (tmp) | 58 | : "=m" (lock->lock), "=&r" (tmp) |
59 | : "m" (lock->lock) | 59 | : "m" (lock->lock) |
60 | : "memory"); | 60 | : "memory"); |
61 | } | 61 | } |
62 | |||
63 | smp_mb(); | ||
62 | } | 64 | } |
63 | 65 | ||
64 | static inline void __raw_spin_unlock(raw_spinlock_t *lock) | 66 | static inline void __raw_spin_unlock(raw_spinlock_t *lock) |
65 | { | 67 | { |
68 | smp_mb(); | ||
69 | |||
66 | __asm__ __volatile__( | 70 | __asm__ __volatile__( |
67 | " .set noreorder # __raw_spin_unlock \n" | 71 | " .set noreorder # __raw_spin_unlock \n" |
68 | " sync \n" | ||
69 | " sw $0, %0 \n" | 72 | " sw $0, %0 \n" |
70 | " .set\treorder \n" | 73 | " .set\treorder \n" |
71 | : "=m" (lock->lock) | 74 | : "=m" (lock->lock) |
@@ -86,7 +89,6 @@ static inline unsigned int __raw_spin_trylock(raw_spinlock_t *lock) | |||
86 | " beqzl %2, 1b \n" | 89 | " beqzl %2, 1b \n" |
87 | " nop \n" | 90 | " nop \n" |
88 | " andi %2, %0, 1 \n" | 91 | " andi %2, %0, 1 \n" |
89 | " sync \n" | ||
90 | " .set reorder" | 92 | " .set reorder" |
91 | : "=&r" (temp), "=m" (lock->lock), "=&r" (res) | 93 | : "=&r" (temp), "=m" (lock->lock), "=&r" (res) |
92 | : "m" (lock->lock) | 94 | : "m" (lock->lock) |
@@ -99,13 +101,14 @@ static inline unsigned int __raw_spin_trylock(raw_spinlock_t *lock) | |||
99 | " sc %2, %1 \n" | 101 | " sc %2, %1 \n" |
100 | " beqz %2, 1b \n" | 102 | " beqz %2, 1b \n" |
101 | " andi %2, %0, 1 \n" | 103 | " andi %2, %0, 1 \n" |
102 | " sync \n" | ||
103 | " .set reorder" | 104 | " .set reorder" |
104 | : "=&r" (temp), "=m" (lock->lock), "=&r" (res) | 105 | : "=&r" (temp), "=m" (lock->lock), "=&r" (res) |
105 | : "m" (lock->lock) | 106 | : "m" (lock->lock) |
106 | : "memory"); | 107 | : "memory"); |
107 | } | 108 | } |
108 | 109 | ||
110 | smp_mb(); | ||
111 | |||
109 | return res == 0; | 112 | return res == 0; |
110 | } | 113 | } |
111 | 114 | ||
@@ -143,7 +146,6 @@ static inline void __raw_read_lock(raw_rwlock_t *rw) | |||
143 | " sc %1, %0 \n" | 146 | " sc %1, %0 \n" |
144 | " beqzl %1, 1b \n" | 147 | " beqzl %1, 1b \n" |
145 | " nop \n" | 148 | " nop \n" |
146 | " sync \n" | ||
147 | " .set reorder \n" | 149 | " .set reorder \n" |
148 | : "=m" (rw->lock), "=&r" (tmp) | 150 | : "=m" (rw->lock), "=&r" (tmp) |
149 | : "m" (rw->lock) | 151 | : "m" (rw->lock) |
@@ -156,12 +158,14 @@ static inline void __raw_read_lock(raw_rwlock_t *rw) | |||
156 | " addu %1, 1 \n" | 158 | " addu %1, 1 \n" |
157 | " sc %1, %0 \n" | 159 | " sc %1, %0 \n" |
158 | " beqz %1, 1b \n" | 160 | " beqz %1, 1b \n" |
159 | " sync \n" | 161 | " nop \n" |
160 | " .set reorder \n" | 162 | " .set reorder \n" |
161 | : "=m" (rw->lock), "=&r" (tmp) | 163 | : "=m" (rw->lock), "=&r" (tmp) |
162 | : "m" (rw->lock) | 164 | : "m" (rw->lock) |
163 | : "memory"); | 165 | : "memory"); |
164 | } | 166 | } |
167 | |||
168 | smp_mb(); | ||
165 | } | 169 | } |
166 | 170 | ||
167 | /* Note the use of sub, not subu which will make the kernel die with an | 171 | /* Note the use of sub, not subu which will make the kernel die with an |
@@ -171,13 +175,14 @@ static inline void __raw_read_unlock(raw_rwlock_t *rw) | |||
171 | { | 175 | { |
172 | unsigned int tmp; | 176 | unsigned int tmp; |
173 | 177 | ||
178 | smp_mb(); | ||
179 | |||
174 | if (R10000_LLSC_WAR) { | 180 | if (R10000_LLSC_WAR) { |
175 | __asm__ __volatile__( | 181 | __asm__ __volatile__( |
176 | "1: ll %1, %2 # __raw_read_unlock \n" | 182 | "1: ll %1, %2 # __raw_read_unlock \n" |
177 | " sub %1, 1 \n" | 183 | " sub %1, 1 \n" |
178 | " sc %1, %0 \n" | 184 | " sc %1, %0 \n" |
179 | " beqzl %1, 1b \n" | 185 | " beqzl %1, 1b \n" |
180 | " sync \n" | ||
181 | : "=m" (rw->lock), "=&r" (tmp) | 186 | : "=m" (rw->lock), "=&r" (tmp) |
182 | : "m" (rw->lock) | 187 | : "m" (rw->lock) |
183 | : "memory"); | 188 | : "memory"); |
@@ -188,7 +193,7 @@ static inline void __raw_read_unlock(raw_rwlock_t *rw) | |||
188 | " sub %1, 1 \n" | 193 | " sub %1, 1 \n" |
189 | " sc %1, %0 \n" | 194 | " sc %1, %0 \n" |
190 | " beqz %1, 1b \n" | 195 | " beqz %1, 1b \n" |
191 | " sync \n" | 196 | " nop \n" |
192 | " .set reorder \n" | 197 | " .set reorder \n" |
193 | : "=m" (rw->lock), "=&r" (tmp) | 198 | : "=m" (rw->lock), "=&r" (tmp) |
194 | : "m" (rw->lock) | 199 | : "m" (rw->lock) |
@@ -208,7 +213,7 @@ static inline void __raw_write_lock(raw_rwlock_t *rw) | |||
208 | " lui %1, 0x8000 \n" | 213 | " lui %1, 0x8000 \n" |
209 | " sc %1, %0 \n" | 214 | " sc %1, %0 \n" |
210 | " beqzl %1, 1b \n" | 215 | " beqzl %1, 1b \n" |
211 | " sync \n" | 216 | " nop \n" |
212 | " .set reorder \n" | 217 | " .set reorder \n" |
213 | : "=m" (rw->lock), "=&r" (tmp) | 218 | : "=m" (rw->lock), "=&r" (tmp) |
214 | : "m" (rw->lock) | 219 | : "m" (rw->lock) |
@@ -221,18 +226,22 @@ static inline void __raw_write_lock(raw_rwlock_t *rw) | |||
221 | " lui %1, 0x8000 \n" | 226 | " lui %1, 0x8000 \n" |
222 | " sc %1, %0 \n" | 227 | " sc %1, %0 \n" |
223 | " beqz %1, 1b \n" | 228 | " beqz %1, 1b \n" |
224 | " sync \n" | 229 | " nop \n" |
225 | " .set reorder \n" | 230 | " .set reorder \n" |
226 | : "=m" (rw->lock), "=&r" (tmp) | 231 | : "=m" (rw->lock), "=&r" (tmp) |
227 | : "m" (rw->lock) | 232 | : "m" (rw->lock) |
228 | : "memory"); | 233 | : "memory"); |
229 | } | 234 | } |
235 | |||
236 | smp_mb(); | ||
230 | } | 237 | } |
231 | 238 | ||
232 | static inline void __raw_write_unlock(raw_rwlock_t *rw) | 239 | static inline void __raw_write_unlock(raw_rwlock_t *rw) |
233 | { | 240 | { |
241 | smp_mb(); | ||
242 | |||
234 | __asm__ __volatile__( | 243 | __asm__ __volatile__( |
235 | " sync # __raw_write_unlock \n" | 244 | " # __raw_write_unlock \n" |
236 | " sw $0, %0 \n" | 245 | " sw $0, %0 \n" |
237 | : "=m" (rw->lock) | 246 | : "=m" (rw->lock) |
238 | : "m" (rw->lock) | 247 | : "m" (rw->lock) |
@@ -252,11 +261,10 @@ static inline int __raw_read_trylock(raw_rwlock_t *rw) | |||
252 | " bnez %1, 2f \n" | 261 | " bnez %1, 2f \n" |
253 | " addu %1, 1 \n" | 262 | " addu %1, 1 \n" |
254 | " sc %1, %0 \n" | 263 | " sc %1, %0 \n" |
255 | " beqzl %1, 1b \n" | ||
256 | " .set reorder \n" | 264 | " .set reorder \n" |
257 | #ifdef CONFIG_SMP | 265 | " beqzl %1, 1b \n" |
258 | " sync \n" | 266 | " nop \n" |
259 | #endif | 267 | __WEAK_ORDERING_MB |
260 | " li %2, 1 \n" | 268 | " li %2, 1 \n" |
261 | "2: \n" | 269 | "2: \n" |
262 | : "=m" (rw->lock), "=&r" (tmp), "=&r" (ret) | 270 | : "=m" (rw->lock), "=&r" (tmp), "=&r" (ret) |
@@ -271,10 +279,9 @@ static inline int __raw_read_trylock(raw_rwlock_t *rw) | |||
271 | " addu %1, 1 \n" | 279 | " addu %1, 1 \n" |
272 | " sc %1, %0 \n" | 280 | " sc %1, %0 \n" |
273 | " beqz %1, 1b \n" | 281 | " beqz %1, 1b \n" |
282 | " nop \n" | ||
274 | " .set reorder \n" | 283 | " .set reorder \n" |
275 | #ifdef CONFIG_SMP | 284 | __WEAK_ORDERING_MB |
276 | " sync \n" | ||
277 | #endif | ||
278 | " li %2, 1 \n" | 285 | " li %2, 1 \n" |
279 | "2: \n" | 286 | "2: \n" |
280 | : "=m" (rw->lock), "=&r" (tmp), "=&r" (ret) | 287 | : "=m" (rw->lock), "=&r" (tmp), "=&r" (ret) |
@@ -299,7 +306,8 @@ static inline int __raw_write_trylock(raw_rwlock_t *rw) | |||
299 | " lui %1, 0x8000 \n" | 306 | " lui %1, 0x8000 \n" |
300 | " sc %1, %0 \n" | 307 | " sc %1, %0 \n" |
301 | " beqzl %1, 1b \n" | 308 | " beqzl %1, 1b \n" |
302 | " sync \n" | 309 | " nop \n" |
310 | __WEAK_ORDERING_MB | ||
303 | " li %2, 1 \n" | 311 | " li %2, 1 \n" |
304 | " .set reorder \n" | 312 | " .set reorder \n" |
305 | "2: \n" | 313 | "2: \n" |
@@ -315,7 +323,8 @@ static inline int __raw_write_trylock(raw_rwlock_t *rw) | |||
315 | " lui %1, 0x8000 \n" | 323 | " lui %1, 0x8000 \n" |
316 | " sc %1, %0 \n" | 324 | " sc %1, %0 \n" |
317 | " beqz %1, 1b \n" | 325 | " beqz %1, 1b \n" |
318 | " sync \n" | 326 | " nop \n" |
327 | __WEAK_ORDERING_MB | ||
319 | " li %2, 1 \n" | 328 | " li %2, 1 \n" |
320 | " .set reorder \n" | 329 | " .set reorder \n" |
321 | "2: \n" | 330 | "2: \n" |
diff --git a/include/asm-mips/system.h b/include/asm-mips/system.h index 3056feed5a36..9428057a50cf 100644 --- a/include/asm-mips/system.h +++ b/include/asm-mips/system.h | |||
@@ -3,7 +3,7 @@ | |||
3 | * License. See the file "COPYING" in the main directory of this archive | 3 | * License. See the file "COPYING" in the main directory of this archive |
4 | * for more details. | 4 | * for more details. |
5 | * | 5 | * |
6 | * Copyright (C) 1994, 95, 96, 97, 98, 99, 2003 by Ralf Baechle | 6 | * Copyright (C) 1994, 95, 96, 97, 98, 99, 2003, 06 by Ralf Baechle |
7 | * Copyright (C) 1996 by Paul M. Antoine | 7 | * Copyright (C) 1996 by Paul M. Antoine |
8 | * Copyright (C) 1999 Silicon Graphics | 8 | * Copyright (C) 1999 Silicon Graphics |
9 | * Kevin D. Kissell, kevink@mips.org and Carsten Langgaard, carstenl@mips.com | 9 | * Kevin D. Kissell, kevink@mips.org and Carsten Langgaard, carstenl@mips.com |
@@ -16,132 +16,12 @@ | |||
16 | #include <linux/irqflags.h> | 16 | #include <linux/irqflags.h> |
17 | 17 | ||
18 | #include <asm/addrspace.h> | 18 | #include <asm/addrspace.h> |
19 | #include <asm/barrier.h> | ||
19 | #include <asm/cpu-features.h> | 20 | #include <asm/cpu-features.h> |
20 | #include <asm/dsp.h> | 21 | #include <asm/dsp.h> |
21 | #include <asm/ptrace.h> | 22 | #include <asm/ptrace.h> |
22 | #include <asm/war.h> | 23 | #include <asm/war.h> |
23 | 24 | ||
24 | /* | ||
25 | * read_barrier_depends - Flush all pending reads that subsequents reads | ||
26 | * depend on. | ||
27 | * | ||
28 | * No data-dependent reads from memory-like regions are ever reordered | ||
29 | * over this barrier. All reads preceding this primitive are guaranteed | ||
30 | * to access memory (but not necessarily other CPUs' caches) before any | ||
31 | * reads following this primitive that depend on the data return by | ||
32 | * any of the preceding reads. This primitive is much lighter weight than | ||
33 | * rmb() on most CPUs, and is never heavier weight than is | ||
34 | * rmb(). | ||
35 | * | ||
36 | * These ordering constraints are respected by both the local CPU | ||
37 | * and the compiler. | ||
38 | * | ||
39 | * Ordering is not guaranteed by anything other than these primitives, | ||
40 | * not even by data dependencies. See the documentation for | ||
41 | * memory_barrier() for examples and URLs to more information. | ||
42 | * | ||
43 | * For example, the following code would force ordering (the initial | ||
44 | * value of "a" is zero, "b" is one, and "p" is "&a"): | ||
45 | * | ||
46 | * <programlisting> | ||
47 | * CPU 0 CPU 1 | ||
48 | * | ||
49 | * b = 2; | ||
50 | * memory_barrier(); | ||
51 | * p = &b; q = p; | ||
52 | * read_barrier_depends(); | ||
53 | * d = *q; | ||
54 | * </programlisting> | ||
55 | * | ||
56 | * because the read of "*q" depends on the read of "p" and these | ||
57 | * two reads are separated by a read_barrier_depends(). However, | ||
58 | * the following code, with the same initial values for "a" and "b": | ||
59 | * | ||
60 | * <programlisting> | ||
61 | * CPU 0 CPU 1 | ||
62 | * | ||
63 | * a = 2; | ||
64 | * memory_barrier(); | ||
65 | * b = 3; y = b; | ||
66 | * read_barrier_depends(); | ||
67 | * x = a; | ||
68 | * </programlisting> | ||
69 | * | ||
70 | * does not enforce ordering, since there is no data dependency between | ||
71 | * the read of "a" and the read of "b". Therefore, on some CPUs, such | ||
72 | * as Alpha, "y" could be set to 3 and "x" to 0. Use rmb() | ||
73 | * in cases like this where there are no data dependencies. | ||
74 | */ | ||
75 | |||
76 | #define read_barrier_depends() do { } while(0) | ||
77 | |||
78 | #ifdef CONFIG_CPU_HAS_SYNC | ||
79 | #define __sync() \ | ||
80 | __asm__ __volatile__( \ | ||
81 | ".set push\n\t" \ | ||
82 | ".set noreorder\n\t" \ | ||
83 | ".set mips2\n\t" \ | ||
84 | "sync\n\t" \ | ||
85 | ".set pop" \ | ||
86 | : /* no output */ \ | ||
87 | : /* no input */ \ | ||
88 | : "memory") | ||
89 | #else | ||
90 | #define __sync() do { } while(0) | ||
91 | #endif | ||
92 | |||
93 | #define __fast_iob() \ | ||
94 | __asm__ __volatile__( \ | ||
95 | ".set push\n\t" \ | ||
96 | ".set noreorder\n\t" \ | ||
97 | "lw $0,%0\n\t" \ | ||
98 | "nop\n\t" \ | ||
99 | ".set pop" \ | ||
100 | : /* no output */ \ | ||
101 | : "m" (*(int *)CKSEG1) \ | ||
102 | : "memory") | ||
103 | |||
104 | #define fast_wmb() __sync() | ||
105 | #define fast_rmb() __sync() | ||
106 | #define fast_mb() __sync() | ||
107 | #define fast_iob() \ | ||
108 | do { \ | ||
109 | __sync(); \ | ||
110 | __fast_iob(); \ | ||
111 | } while (0) | ||
112 | |||
113 | #ifdef CONFIG_CPU_HAS_WB | ||
114 | |||
115 | #include <asm/wbflush.h> | ||
116 | |||
117 | #define wmb() fast_wmb() | ||
118 | #define rmb() fast_rmb() | ||
119 | #define mb() wbflush() | ||
120 | #define iob() wbflush() | ||
121 | |||
122 | #else /* !CONFIG_CPU_HAS_WB */ | ||
123 | |||
124 | #define wmb() fast_wmb() | ||
125 | #define rmb() fast_rmb() | ||
126 | #define mb() fast_mb() | ||
127 | #define iob() fast_iob() | ||
128 | |||
129 | #endif /* !CONFIG_CPU_HAS_WB */ | ||
130 | |||
131 | #ifdef CONFIG_SMP | ||
132 | #define smp_mb() mb() | ||
133 | #define smp_rmb() rmb() | ||
134 | #define smp_wmb() wmb() | ||
135 | #define smp_read_barrier_depends() read_barrier_depends() | ||
136 | #else | ||
137 | #define smp_mb() barrier() | ||
138 | #define smp_rmb() barrier() | ||
139 | #define smp_wmb() barrier() | ||
140 | #define smp_read_barrier_depends() do { } while(0) | ||
141 | #endif | ||
142 | |||
143 | #define set_mb(var, value) \ | ||
144 | do { var = value; mb(); } while (0) | ||
145 | 25 | ||
146 | /* | 26 | /* |
147 | * switch_to(n) should switch tasks to task nr n, first | 27 | * switch_to(n) should switch tasks to task nr n, first |
@@ -217,9 +97,6 @@ static inline unsigned long __xchg_u32(volatile int * m, unsigned int val) | |||
217 | " .set mips3 \n" | 97 | " .set mips3 \n" |
218 | " sc %2, %1 \n" | 98 | " sc %2, %1 \n" |
219 | " beqzl %2, 1b \n" | 99 | " beqzl %2, 1b \n" |
220 | #ifdef CONFIG_SMP | ||
221 | " sync \n" | ||
222 | #endif | ||
223 | " .set mips0 \n" | 100 | " .set mips0 \n" |
224 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) | 101 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) |
225 | : "R" (*m), "Jr" (val) | 102 | : "R" (*m), "Jr" (val) |
@@ -235,9 +112,6 @@ static inline unsigned long __xchg_u32(volatile int * m, unsigned int val) | |||
235 | " .set mips3 \n" | 112 | " .set mips3 \n" |
236 | " sc %2, %1 \n" | 113 | " sc %2, %1 \n" |
237 | " beqz %2, 1b \n" | 114 | " beqz %2, 1b \n" |
238 | #ifdef CONFIG_SMP | ||
239 | " sync \n" | ||
240 | #endif | ||
241 | " .set mips0 \n" | 115 | " .set mips0 \n" |
242 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) | 116 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) |
243 | : "R" (*m), "Jr" (val) | 117 | : "R" (*m), "Jr" (val) |
@@ -251,6 +125,8 @@ static inline unsigned long __xchg_u32(volatile int * m, unsigned int val) | |||
251 | local_irq_restore(flags); /* implies memory barrier */ | 125 | local_irq_restore(flags); /* implies memory barrier */ |
252 | } | 126 | } |
253 | 127 | ||
128 | smp_mb(); | ||
129 | |||
254 | return retval; | 130 | return retval; |
255 | } | 131 | } |
256 | 132 | ||
@@ -268,9 +144,6 @@ static inline __u64 __xchg_u64(volatile __u64 * m, __u64 val) | |||
268 | " move %2, %z4 \n" | 144 | " move %2, %z4 \n" |
269 | " scd %2, %1 \n" | 145 | " scd %2, %1 \n" |
270 | " beqzl %2, 1b \n" | 146 | " beqzl %2, 1b \n" |
271 | #ifdef CONFIG_SMP | ||
272 | " sync \n" | ||
273 | #endif | ||
274 | " .set mips0 \n" | 147 | " .set mips0 \n" |
275 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) | 148 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) |
276 | : "R" (*m), "Jr" (val) | 149 | : "R" (*m), "Jr" (val) |
@@ -284,9 +157,6 @@ static inline __u64 __xchg_u64(volatile __u64 * m, __u64 val) | |||
284 | " move %2, %z4 \n" | 157 | " move %2, %z4 \n" |
285 | " scd %2, %1 \n" | 158 | " scd %2, %1 \n" |
286 | " beqz %2, 1b \n" | 159 | " beqz %2, 1b \n" |
287 | #ifdef CONFIG_SMP | ||
288 | " sync \n" | ||
289 | #endif | ||
290 | " .set mips0 \n" | 160 | " .set mips0 \n" |
291 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) | 161 | : "=&r" (retval), "=m" (*m), "=&r" (dummy) |
292 | : "R" (*m), "Jr" (val) | 162 | : "R" (*m), "Jr" (val) |
@@ -300,6 +170,8 @@ static inline __u64 __xchg_u64(volatile __u64 * m, __u64 val) | |||
300 | local_irq_restore(flags); /* implies memory barrier */ | 170 | local_irq_restore(flags); /* implies memory barrier */ |
301 | } | 171 | } |
302 | 172 | ||
173 | smp_mb(); | ||
174 | |||
303 | return retval; | 175 | return retval; |
304 | } | 176 | } |
305 | #else | 177 | #else |
@@ -345,9 +217,6 @@ static inline unsigned long __cmpxchg_u32(volatile int * m, unsigned long old, | |||
345 | " .set mips3 \n" | 217 | " .set mips3 \n" |
346 | " sc $1, %1 \n" | 218 | " sc $1, %1 \n" |
347 | " beqzl $1, 1b \n" | 219 | " beqzl $1, 1b \n" |
348 | #ifdef CONFIG_SMP | ||
349 | " sync \n" | ||
350 | #endif | ||
351 | "2: \n" | 220 | "2: \n" |
352 | " .set pop \n" | 221 | " .set pop \n" |
353 | : "=&r" (retval), "=R" (*m) | 222 | : "=&r" (retval), "=R" (*m) |
@@ -365,9 +234,6 @@ static inline unsigned long __cmpxchg_u32(volatile int * m, unsigned long old, | |||
365 | " .set mips3 \n" | 234 | " .set mips3 \n" |
366 | " sc $1, %1 \n" | 235 | " sc $1, %1 \n" |
367 | " beqz $1, 1b \n" | 236 | " beqz $1, 1b \n" |
368 | #ifdef CONFIG_SMP | ||
369 | " sync \n" | ||
370 | #endif | ||
371 | "2: \n" | 237 | "2: \n" |
372 | " .set pop \n" | 238 | " .set pop \n" |
373 | : "=&r" (retval), "=R" (*m) | 239 | : "=&r" (retval), "=R" (*m) |
@@ -383,6 +249,8 @@ static inline unsigned long __cmpxchg_u32(volatile int * m, unsigned long old, | |||
383 | local_irq_restore(flags); /* implies memory barrier */ | 249 | local_irq_restore(flags); /* implies memory barrier */ |
384 | } | 250 | } |
385 | 251 | ||
252 | smp_mb(); | ||
253 | |||
386 | return retval; | 254 | return retval; |
387 | } | 255 | } |
388 | 256 | ||
@@ -402,9 +270,6 @@ static inline unsigned long __cmpxchg_u64(volatile int * m, unsigned long old, | |||
402 | " move $1, %z4 \n" | 270 | " move $1, %z4 \n" |
403 | " scd $1, %1 \n" | 271 | " scd $1, %1 \n" |
404 | " beqzl $1, 1b \n" | 272 | " beqzl $1, 1b \n" |
405 | #ifdef CONFIG_SMP | ||
406 | " sync \n" | ||
407 | #endif | ||
408 | "2: \n" | 273 | "2: \n" |
409 | " .set pop \n" | 274 | " .set pop \n" |
410 | : "=&r" (retval), "=R" (*m) | 275 | : "=&r" (retval), "=R" (*m) |
@@ -420,9 +285,6 @@ static inline unsigned long __cmpxchg_u64(volatile int * m, unsigned long old, | |||
420 | " move $1, %z4 \n" | 285 | " move $1, %z4 \n" |
421 | " scd $1, %1 \n" | 286 | " scd $1, %1 \n" |
422 | " beqz $1, 1b \n" | 287 | " beqz $1, 1b \n" |
423 | #ifdef CONFIG_SMP | ||
424 | " sync \n" | ||
425 | #endif | ||
426 | "2: \n" | 288 | "2: \n" |
427 | " .set pop \n" | 289 | " .set pop \n" |
428 | : "=&r" (retval), "=R" (*m) | 290 | : "=&r" (retval), "=R" (*m) |
@@ -438,6 +300,8 @@ static inline unsigned long __cmpxchg_u64(volatile int * m, unsigned long old, | |||
438 | local_irq_restore(flags); /* implies memory barrier */ | 300 | local_irq_restore(flags); /* implies memory barrier */ |
439 | } | 301 | } |
440 | 302 | ||
303 | smp_mb(); | ||
304 | |||
441 | return retval; | 305 | return retval; |
442 | } | 306 | } |
443 | #else | 307 | #else |