2 * Copyright (c) 1998 Doug Rabson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #ifndef _MACHINE_ATOMIC_H_
29 #define _MACHINE_ATOMIC_H_
32 #error this file needs sys/cdefs.h as a prerequisite
36 #include <machine/md_var.h>
37 #include <machine/specialreg.h>
40 #define mb() __asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc")
41 #define wmb() __asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc")
42 #define rmb() __asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc")
45 * Various simple operations on memory, each of which is atomic in the
46 * presence of interrupts and multiple processors.
48 * atomic_set_char(P, V) (*(u_char *)(P) |= (V))
49 * atomic_clear_char(P, V) (*(u_char *)(P) &= ~(V))
50 * atomic_add_char(P, V) (*(u_char *)(P) += (V))
51 * atomic_subtract_char(P, V) (*(u_char *)(P) -= (V))
53 * atomic_set_short(P, V) (*(u_short *)(P) |= (V))
54 * atomic_clear_short(P, V) (*(u_short *)(P) &= ~(V))
55 * atomic_add_short(P, V) (*(u_short *)(P) += (V))
56 * atomic_subtract_short(P, V) (*(u_short *)(P) -= (V))
58 * atomic_set_int(P, V) (*(u_int *)(P) |= (V))
59 * atomic_clear_int(P, V) (*(u_int *)(P) &= ~(V))
60 * atomic_add_int(P, V) (*(u_int *)(P) += (V))
61 * atomic_subtract_int(P, V) (*(u_int *)(P) -= (V))
62 * atomic_swap_int(P, V) (return (*(u_int *)(P)); *(u_int *)(P) = (V);)
63 * atomic_readandclear_int(P) (return (*(u_int *)(P)); *(u_int *)(P) = 0;)
65 * atomic_set_long(P, V) (*(u_long *)(P) |= (V))
66 * atomic_clear_long(P, V) (*(u_long *)(P) &= ~(V))
67 * atomic_add_long(P, V) (*(u_long *)(P) += (V))
68 * atomic_subtract_long(P, V) (*(u_long *)(P) -= (V))
69 * atomic_swap_long(P, V) (return (*(u_long *)(P)); *(u_long *)(P) = (V);)
70 * atomic_readandclear_long(P) (return (*(u_long *)(P)); *(u_long *)(P) = 0;)
74 * The above functions are expanded inline in the statically-linked
75 * kernel. Lock prefixes are generated if an SMP kernel is being
78 * Kernel modules call real functions which are built into the kernel.
79 * This allows kernel modules to be portable between UP and SMP systems.
81 #if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
82 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
83 void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v); \
84 void atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
86 int atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src);
87 u_int atomic_fetchadd_int(volatile u_int *p, u_int v);
88 int atomic_testandset_int(volatile u_int *p, u_int v);
90 #define ATOMIC_LOAD(TYPE, LOP) \
91 u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p)
92 #define ATOMIC_STORE(TYPE) \
93 void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
95 int atomic_cmpset_64(volatile uint64_t *, uint64_t, uint64_t);
96 uint64_t atomic_load_acq_64(volatile uint64_t *);
97 void atomic_store_rel_64(volatile uint64_t *, uint64_t);
98 uint64_t atomic_swap_64(volatile uint64_t *, uint64_t);
100 #else /* !KLD_MODULE && __GNUCLIKE_ASM */
103 * For userland, always use lock prefixes so that the binaries will run
104 * on both SMP and !SMP systems.
106 #if defined(SMP) || !defined(_KERNEL)
107 #define MPLOCKED "lock ; "
113 * The assembly is volatilized to avoid code chunk removal by the compiler.
114 * GCC aggressively reorders operations and memory clobbering is necessary
115 * in order to avoid that for memory barriers.
117 #define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
118 static __inline void \
119 atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
121 __asm __volatile(MPLOCKED OP \
127 static __inline void \
128 atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
130 __asm __volatile(MPLOCKED OP \
138 * Atomic compare and set, used by the mutex functions
140 * if (*dst == expect) *dst = src (all 32 bit words)
142 * Returns 0 on failure, non-zero on success
145 #ifdef CPU_DISABLE_CMPXCHG
148 atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src)
161 "# atomic_cmpset_int"
162 : "=q" (res), /* 0 */
170 #else /* !CPU_DISABLE_CMPXCHG */
173 atomic_cmpset_int(volatile u_int *dst, u_int expect, u_int src)
181 "# atomic_cmpset_int"
182 : "=q" (res), /* 0 */
184 "+a" (expect) /* 2 */
190 #endif /* CPU_DISABLE_CMPXCHG */
193 * Atomically add the value of v to the integer pointed to by p and return
194 * the previous value of *p.
196 static __inline u_int
197 atomic_fetchadd_int(volatile u_int *p, u_int v)
203 "# atomic_fetchadd_int"
211 atomic_testandset_int(volatile u_int *p, u_int v)
219 "# atomic_testandset_int"
220 : "=q" (res), /* 0 */
222 : "Ir" (v & 0x1f) /* 2 */
228 * We assume that a = b will do atomic loads and stores. Due to the
229 * IA32 memory model, a simple store guarantees release semantics.
231 * However, loads may pass stores, so for atomic_load_acq we have to
232 * ensure a Store/Load barrier to do the load in SMP kernels. We use
233 * "lock cmpxchg" as recommended by the AMD Software Optimization
234 * Guide, and not mfence. For UP kernels, however, the cache of the
235 * single processor is always consistent, so we only need to take care
238 #define ATOMIC_STORE(TYPE) \
239 static __inline void \
240 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
242 __compiler_membar(); \
247 #if defined(_KERNEL) && !defined(SMP)
249 #define ATOMIC_LOAD(TYPE, LOP) \
250 static __inline u_##TYPE \
251 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
256 __compiler_membar(); \
261 #else /* !(_KERNEL && !SMP) */
263 #define ATOMIC_LOAD(TYPE, LOP) \
264 static __inline u_##TYPE \
265 atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
269 __asm __volatile(MPLOCKED LOP \
270 : "=a" (res), /* 0 */ \
272 : : "memory", "cc"); \
277 #endif /* _KERNEL && !SMP */
281 #ifdef WANT_FUNCTIONS
282 int atomic_cmpset_64_i386(volatile uint64_t *, uint64_t, uint64_t);
283 int atomic_cmpset_64_i586(volatile uint64_t *, uint64_t, uint64_t);
284 uint64_t atomic_load_acq_64_i386(volatile uint64_t *);
285 uint64_t atomic_load_acq_64_i586(volatile uint64_t *);
286 void atomic_store_rel_64_i386(volatile uint64_t *, uint64_t);
287 void atomic_store_rel_64_i586(volatile uint64_t *, uint64_t);
288 uint64_t atomic_swap_64_i386(volatile uint64_t *, uint64_t);
289 uint64_t atomic_swap_64_i586(volatile uint64_t *, uint64_t);
292 /* I486 does not support SMP or CMPXCHG8B. */
294 atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src)
296 volatile uint32_t *p;
299 p = (volatile uint32_t *)dst;
305 " orl %%edx,%%eax ; "
312 : "+A" (expect), /* 0 */
314 "+m" (*(p + 1)), /* 2 */
316 : "r" ((uint32_t)src), /* 4 */
317 "r" ((uint32_t)(src >> 32)) /* 5 */
322 static __inline uint64_t
323 atomic_load_acq_64_i386(volatile uint64_t *p)
325 volatile uint32_t *q;
328 q = (volatile uint32_t *)p;
335 : "=&A" (res) /* 0 */
337 "m" (*(q + 1)) /* 2 */
343 atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v)
345 volatile uint32_t *q;
347 q = (volatile uint32_t *)p;
355 "=m" (*(q + 1)) /* 1 */
360 static __inline uint64_t
361 atomic_swap_64_i386(volatile uint64_t *p, uint64_t v)
363 volatile uint32_t *q;
366 q = (volatile uint32_t *)p;
375 : "=&A" (res), /* 0 */
377 "+m" (*(q + 1)) /* 2 */
378 : "r" ((uint32_t)v), /* 3 */
379 "r" ((uint32_t)(v >> 32))); /* 4 */
384 atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src)
392 : "=q" (res), /* 0 */
394 "+A" (expect) /* 2 */
395 : "b" ((uint32_t)src), /* 3 */
396 "c" ((uint32_t)(src >> 32)) /* 4 */
401 static __inline uint64_t
402 atomic_load_acq_64_i586(volatile uint64_t *p)
407 " movl %%ebx,%%eax ; "
408 " movl %%ecx,%%edx ; "
411 : "=&A" (res), /* 0 */
418 atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v)
422 " movl %%eax,%%ebx ; "
423 " movl %%edx,%%ecx ; "
430 : : "ebx", "ecx", "memory", "cc");
433 static __inline uint64_t
434 atomic_swap_64_i586(volatile uint64_t *p, uint64_t v)
438 " movl %%eax,%%ebx ; "
439 " movl %%edx,%%ecx ; "
446 : : "ebx", "ecx", "memory", "cc");
451 atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src)
454 if ((cpu_feature & CPUID_CX8) == 0)
455 return (atomic_cmpset_64_i386(dst, expect, src));
457 return (atomic_cmpset_64_i586(dst, expect, src));
460 static __inline uint64_t
461 atomic_load_acq_64(volatile uint64_t *p)
464 if ((cpu_feature & CPUID_CX8) == 0)
465 return (atomic_load_acq_64_i386(p));
467 return (atomic_load_acq_64_i586(p));
471 atomic_store_rel_64(volatile uint64_t *p, uint64_t v)
474 if ((cpu_feature & CPUID_CX8) == 0)
475 atomic_store_rel_64_i386(p, v);
477 atomic_store_rel_64_i586(p, v);
480 static __inline uint64_t
481 atomic_swap_64(volatile uint64_t *p, uint64_t v)
484 if ((cpu_feature & CPUID_CX8) == 0)
485 return (atomic_swap_64_i386(p, v));
487 return (atomic_swap_64_i586(p, v));
492 #endif /* KLD_MODULE || !__GNUCLIKE_ASM */
494 ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v);
495 ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v);
496 ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v);
497 ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v);
499 ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v);
500 ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v);
501 ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v);
502 ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v);
504 ATOMIC_ASM(set, int, "orl %1,%0", "ir", v);
505 ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v);
506 ATOMIC_ASM(add, int, "addl %1,%0", "ir", v);
507 ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v);
509 ATOMIC_ASM(set, long, "orl %1,%0", "ir", v);
510 ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v);
511 ATOMIC_ASM(add, long, "addl %1,%0", "ir", v);
512 ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v);
514 ATOMIC_LOAD(char, "cmpxchgb %b0,%1");
515 ATOMIC_LOAD(short, "cmpxchgw %w0,%1");
516 ATOMIC_LOAD(int, "cmpxchgl %0,%1");
517 ATOMIC_LOAD(long, "cmpxchgl %0,%1");
528 #ifndef WANT_FUNCTIONS
531 atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src)
534 return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect,
538 static __inline u_long
539 atomic_fetchadd_long(volatile u_long *p, u_long v)
542 return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v));
546 atomic_testandset_long(volatile u_long *p, u_int v)
549 return (atomic_testandset_int((volatile u_int *)p, v));
552 /* Read the current value and store a new value in the destination. */
553 #ifdef __GNUCLIKE_ASM
555 static __inline u_int
556 atomic_swap_int(volatile u_int *p, u_int v)
567 static __inline u_long
568 atomic_swap_long(volatile u_long *p, u_long v)
571 return (atomic_swap_int((volatile u_int *)p, (u_int)v));
574 #else /* !__GNUCLIKE_ASM */
576 u_int atomic_swap_int(volatile u_int *p, u_int v);
577 u_long atomic_swap_long(volatile u_long *p, u_long v);
579 #endif /* __GNUCLIKE_ASM */
581 #define atomic_set_acq_char atomic_set_barr_char
582 #define atomic_set_rel_char atomic_set_barr_char
583 #define atomic_clear_acq_char atomic_clear_barr_char
584 #define atomic_clear_rel_char atomic_clear_barr_char
585 #define atomic_add_acq_char atomic_add_barr_char
586 #define atomic_add_rel_char atomic_add_barr_char
587 #define atomic_subtract_acq_char atomic_subtract_barr_char
588 #define atomic_subtract_rel_char atomic_subtract_barr_char
590 #define atomic_set_acq_short atomic_set_barr_short
591 #define atomic_set_rel_short atomic_set_barr_short
592 #define atomic_clear_acq_short atomic_clear_barr_short
593 #define atomic_clear_rel_short atomic_clear_barr_short
594 #define atomic_add_acq_short atomic_add_barr_short
595 #define atomic_add_rel_short atomic_add_barr_short
596 #define atomic_subtract_acq_short atomic_subtract_barr_short
597 #define atomic_subtract_rel_short atomic_subtract_barr_short
599 #define atomic_set_acq_int atomic_set_barr_int
600 #define atomic_set_rel_int atomic_set_barr_int
601 #define atomic_clear_acq_int atomic_clear_barr_int
602 #define atomic_clear_rel_int atomic_clear_barr_int
603 #define atomic_add_acq_int atomic_add_barr_int
604 #define atomic_add_rel_int atomic_add_barr_int
605 #define atomic_subtract_acq_int atomic_subtract_barr_int
606 #define atomic_subtract_rel_int atomic_subtract_barr_int
607 #define atomic_cmpset_acq_int atomic_cmpset_int
608 #define atomic_cmpset_rel_int atomic_cmpset_int
610 #define atomic_set_acq_long atomic_set_barr_long
611 #define atomic_set_rel_long atomic_set_barr_long
612 #define atomic_clear_acq_long atomic_clear_barr_long
613 #define atomic_clear_rel_long atomic_clear_barr_long
614 #define atomic_add_acq_long atomic_add_barr_long
615 #define atomic_add_rel_long atomic_add_barr_long
616 #define atomic_subtract_acq_long atomic_subtract_barr_long
617 #define atomic_subtract_rel_long atomic_subtract_barr_long
618 #define atomic_cmpset_acq_long atomic_cmpset_long
619 #define atomic_cmpset_rel_long atomic_cmpset_long
621 #define atomic_readandclear_int(p) atomic_swap_int(p, 0)
622 #define atomic_readandclear_long(p) atomic_swap_long(p, 0)
624 /* Operations on 8-bit bytes. */
625 #define atomic_set_8 atomic_set_char
626 #define atomic_set_acq_8 atomic_set_acq_char
627 #define atomic_set_rel_8 atomic_set_rel_char
628 #define atomic_clear_8 atomic_clear_char
629 #define atomic_clear_acq_8 atomic_clear_acq_char
630 #define atomic_clear_rel_8 atomic_clear_rel_char
631 #define atomic_add_8 atomic_add_char
632 #define atomic_add_acq_8 atomic_add_acq_char
633 #define atomic_add_rel_8 atomic_add_rel_char
634 #define atomic_subtract_8 atomic_subtract_char
635 #define atomic_subtract_acq_8 atomic_subtract_acq_char
636 #define atomic_subtract_rel_8 atomic_subtract_rel_char
637 #define atomic_load_acq_8 atomic_load_acq_char
638 #define atomic_store_rel_8 atomic_store_rel_char
640 /* Operations on 16-bit words. */
641 #define atomic_set_16 atomic_set_short
642 #define atomic_set_acq_16 atomic_set_acq_short
643 #define atomic_set_rel_16 atomic_set_rel_short
644 #define atomic_clear_16 atomic_clear_short
645 #define atomic_clear_acq_16 atomic_clear_acq_short
646 #define atomic_clear_rel_16 atomic_clear_rel_short
647 #define atomic_add_16 atomic_add_short
648 #define atomic_add_acq_16 atomic_add_acq_short
649 #define atomic_add_rel_16 atomic_add_rel_short
650 #define atomic_subtract_16 atomic_subtract_short
651 #define atomic_subtract_acq_16 atomic_subtract_acq_short
652 #define atomic_subtract_rel_16 atomic_subtract_rel_short
653 #define atomic_load_acq_16 atomic_load_acq_short
654 #define atomic_store_rel_16 atomic_store_rel_short
656 /* Operations on 32-bit double words. */
657 #define atomic_set_32 atomic_set_int
658 #define atomic_set_acq_32 atomic_set_acq_int
659 #define atomic_set_rel_32 atomic_set_rel_int
660 #define atomic_clear_32 atomic_clear_int
661 #define atomic_clear_acq_32 atomic_clear_acq_int
662 #define atomic_clear_rel_32 atomic_clear_rel_int
663 #define atomic_add_32 atomic_add_int
664 #define atomic_add_acq_32 atomic_add_acq_int
665 #define atomic_add_rel_32 atomic_add_rel_int
666 #define atomic_subtract_32 atomic_subtract_int
667 #define atomic_subtract_acq_32 atomic_subtract_acq_int
668 #define atomic_subtract_rel_32 atomic_subtract_rel_int
669 #define atomic_load_acq_32 atomic_load_acq_int
670 #define atomic_store_rel_32 atomic_store_rel_int
671 #define atomic_cmpset_32 atomic_cmpset_int
672 #define atomic_cmpset_acq_32 atomic_cmpset_acq_int
673 #define atomic_cmpset_rel_32 atomic_cmpset_rel_int
674 #define atomic_swap_32 atomic_swap_int
675 #define atomic_readandclear_32 atomic_readandclear_int
676 #define atomic_fetchadd_32 atomic_fetchadd_int
677 #define atomic_testandset_32 atomic_testandset_int
679 /* Operations on pointers. */
680 #define atomic_set_ptr(p, v) \
681 atomic_set_int((volatile u_int *)(p), (u_int)(v))
682 #define atomic_set_acq_ptr(p, v) \
683 atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
684 #define atomic_set_rel_ptr(p, v) \
685 atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
686 #define atomic_clear_ptr(p, v) \
687 atomic_clear_int((volatile u_int *)(p), (u_int)(v))
688 #define atomic_clear_acq_ptr(p, v) \
689 atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
690 #define atomic_clear_rel_ptr(p, v) \
691 atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
692 #define atomic_add_ptr(p, v) \
693 atomic_add_int((volatile u_int *)(p), (u_int)(v))
694 #define atomic_add_acq_ptr(p, v) \
695 atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
696 #define atomic_add_rel_ptr(p, v) \
697 atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
698 #define atomic_subtract_ptr(p, v) \
699 atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
700 #define atomic_subtract_acq_ptr(p, v) \
701 atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
702 #define atomic_subtract_rel_ptr(p, v) \
703 atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
704 #define atomic_load_acq_ptr(p) \
705 atomic_load_acq_int((volatile u_int *)(p))
706 #define atomic_store_rel_ptr(p, v) \
707 atomic_store_rel_int((volatile u_int *)(p), (v))
708 #define atomic_cmpset_ptr(dst, old, new) \
709 atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
710 #define atomic_cmpset_acq_ptr(dst, old, new) \
711 atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
713 #define atomic_cmpset_rel_ptr(dst, old, new) \
714 atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
716 #define atomic_swap_ptr(p, v) \
717 atomic_swap_int((volatile u_int *)(p), (u_int)(v))
718 #define atomic_readandclear_ptr(p) \
719 atomic_readandclear_int((volatile u_int *)(p))
721 #endif /* !WANT_FUNCTIONS */
723 #endif /* !_MACHINE_ATOMIC_H_ */