| 1 | /*- |
| 2 | * SPDX-License-Identifier: BSD-2-Clause |
| 3 | * |
| 4 | * Copyright (c) 1998 Doug Rabson |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * 1. Redistributions of source code must retain the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer in the |
| 14 | * documentation and/or other materials provided with the distribution. |
| 15 | * |
| 16 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
| 17 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
| 20 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 21 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 22 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 23 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 24 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 25 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 26 | * SUCH DAMAGE. |
| 27 | */ |
| 28 | #ifndef _MACHINE_ATOMIC_H_ |
| 29 | #define	_MACHINE_ATOMIC_H_ |
| 30 | |
| 31 | #include <sys/atomic_common.h> |
| 32 | |
| 33 | #ifdef _KERNEL |
| 34 | #include <machine/md_var.h> |
| 35 | #include <machine/specialreg.h> |
| 36 | #endif |
| 37 | |
| 38 | #ifndef __OFFSETOF_MONITORBUF |
| 39 | /* |
| 40 | * __OFFSETOF_MONITORBUF == __pcpu_offset(pc_monitorbuf). |
| 41 | * |
| 42 | * The open-coded number is used instead of the symbolic expression to |
| 43 | * avoid a dependency on sys/pcpu.h in machine/atomic.h consumers. |
| 44 | * An assertion in i386/vm_machdep.c ensures that the value is correct. |
| 45 | */ |
| 46 | #define	__OFFSETOF_MONITORBUF	0x80 |
| 47 | |
| 48 | static __inline void |
| 49 | __mbk(void) |
| 50 | { |
| 51 | |
| 52 | 	__asm __volatile("lock; addl $0,%%fs:%c0" |
| 53 | 	 : : "i" (__OFFSETOF_MONITORBUF) : "memory", "cc"); |
| 54 | } |
| 55 | |
| 56 | static __inline void |
| 57 | __mbu(void) |
| 58 | { |
| 59 | |
| 60 | 	__asm __volatile("lock; addl $0,(%%esp)" : : : "memory", "cc"); |
| 61 | } |
| 62 | #endif |
| 63 | |
| 64 | /* |
| 65 | * Various simple operations on memory, each of which is atomic in the |
| 66 | * presence of interrupts and multiple processors. |
| 67 | * |
| 68 | * atomic_set_char(P, V)	(*(u_char *)(P) |= (V)) |
| 69 | * atomic_clear_char(P, V)	(*(u_char *)(P) &= ~(V)) |
| 70 | * atomic_add_char(P, V)	(*(u_char *)(P) += (V)) |
| 71 | * atomic_subtract_char(P, V)	(*(u_char *)(P) -= (V)) |
| 72 | * |
| 73 | * atomic_set_short(P, V)	(*(u_short *)(P) |= (V)) |
| 74 | * atomic_clear_short(P, V)	(*(u_short *)(P) &= ~(V)) |
| 75 | * atomic_add_short(P, V)	(*(u_short *)(P) += (V)) |
| 76 | * atomic_subtract_short(P, V)	(*(u_short *)(P) -= (V)) |
| 77 | * |
| 78 | * atomic_set_int(P, V)		(*(u_int *)(P) |= (V)) |
| 79 | * atomic_clear_int(P, V)	(*(u_int *)(P) &= ~(V)) |
| 80 | * atomic_add_int(P, V)		(*(u_int *)(P) += (V)) |
| 81 | * atomic_subtract_int(P, V)	(*(u_int *)(P) -= (V)) |
| 82 | * atomic_swap_int(P, V)	(return (*(u_int *)(P)); *(u_int *)(P) = (V);) |
| 83 | * atomic_readandclear_int(P)	(return (*(u_int *)(P)); *(u_int *)(P) = 0;) |
| 84 | * |
| 85 | * atomic_set_long(P, V)	(*(u_long *)(P) |= (V)) |
| 86 | * atomic_clear_long(P, V)	(*(u_long *)(P) &= ~(V)) |
| 87 | * atomic_add_long(P, V)	(*(u_long *)(P) += (V)) |
| 88 | * atomic_subtract_long(P, V)	(*(u_long *)(P) -= (V)) |
| 89 | * atomic_swap_long(P, V)	(return (*(u_long *)(P)); *(u_long *)(P) = (V);) |
| 90 | * atomic_readandclear_long(P)	(return (*(u_long *)(P)); *(u_long *)(P) = 0;) |
| 91 | */ |
| 92 | |
| 93 | /* |
| 94 | * Always use lock prefixes. The result is slightly less optimal for |
| 95 | * UP systems, but it matters less now, and sometimes UP is emulated |
| 96 | * over SMP. |
| 97 | * |
| 98 | * The assembly is volatilized to avoid code chunk removal by the compiler. |
| 99 | * GCC aggressively reorders operations and memory clobbering is necessary |
| 100 | * in order to avoid that for memory barriers. |
| 101 | */ |
| 102 | #define	ATOMIC_ASM(NAME, TYPE, OP, CONS, V)		\ |
| 103 | static __inline void					\ |
| 104 | atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ |
| 105 | {							\ |
| 106 | 	__asm __volatile("lock; " OP			\ |
| 107 | 	: "+m" (*p)					\ |
| 108 | 	: CONS (V)					\ |
| 109 | 	: "cc");					\ |
| 110 | }							\ |
| 111 | 							\ |
| 112 | static __inline void					\ |
| 113 | atomic_##NAME##_barr_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\ |
| 114 | {							\ |
| 115 | 	__asm __volatile("lock; " OP			\ |
| 116 | 	: "+m" (*p)					\ |
| 117 | 	: CONS (V)					\ |
| 118 | 	: "memory", "cc");				\ |
| 119 | }							\ |
| 120 | struct __hack |
| 121 | |
| 122 | /* |
| 123 | * Atomic compare and set, used by the mutex functions. |
| 124 | * |
| 125 | * cmpset: |
| 126 | *	if (*dst == expect) |
| 127 | *		*dst = src |
| 128 | * |
| 129 | * fcmpset: |
| 130 | *	if (*dst == *expect) |
| 131 | *		*dst = src |
| 132 | *	else |
| 133 | *		*expect = *dst |
| 134 | * |
| 135 | * Returns 0 on failure, non-zero on success. |
| 136 | */ |
| 137 | #define	ATOMIC_CMPSET(TYPE, CONS)			\ |
| 138 | static __inline int					\ |
| 139 | atomic_cmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE expect, u_##TYPE src) \ |
| 140 | {							\ |
| 141 | 	u_char res;					\ |
| 142 | 							\ |
| 143 | 	__asm __volatile(				\ |
| 144 | 	"	lock; cmpxchg	%3,%1 ;	"		\ |
| 145 | 	"	sete	%0 ;		"		\ |
| 146 | 	"# atomic_cmpset_" #TYPE "	"		\ |
| 147 | 	: "=q" (res),			/* 0 */		\ |
| 148 | 	 "+m" (*dst),			/* 1 */		\ |
| 149 | 	 "+a" (expect)			/* 2 */		\ |
| 150 | 	: CONS (src)			/* 3 */		\ |
| 151 | 	: "memory", "cc");				\ |
| 152 | 	return (res);					\ |
| 153 | }							\ |
| 154 | 							\ |
| 155 | static __inline int					\ |
| 156 | atomic_fcmpset_##TYPE(volatile u_##TYPE *dst, u_##TYPE *expect, u_##TYPE src) \ |
| 157 | {							\ |
| 158 | 	u_char res;					\ |
| 159 | 							\ |
| 160 | 	__asm __volatile(				\ |
| 161 | 	"	lock; cmpxchg	%3,%1 ;	"		\ |
| 162 | 	"	sete	%0 ;		"		\ |
| 163 | 	"# atomic_fcmpset_" #TYPE "	"		\ |
| 164 | 	: "=q" (res),			/* 0 */		\ |
| 165 | 	 "+m" (*dst),			/* 1 */		\ |
| 166 | 	 "+a" (*expect)		/* 2 */		\ |
| 167 | 	: CONS (src)			/* 3 */		\ |
| 168 | 	: "memory", "cc");				\ |
| 169 | 	return (res);					\ |
| 170 | } |
| 171 | |
| 172 | ATOMIC_CMPSET(char, "q"); |
| 173 | ATOMIC_CMPSET(short, "r"); |
| 174 | ATOMIC_CMPSET(int, "r"); |
| 175 | |
| 176 | /* |
| 177 | * Atomically add the value of v to the integer pointed to by p and return |
| 178 | * the previous value of *p. |
| 179 | */ |
| 180 | static __inline u_int |
| 181 | atomic_fetchadd_int(volatile u_int *p, u_int v) |
| 182 | { |
| 183 | |
| 184 | 	__asm __volatile( |
| 185 | 	"	lock; xaddl	%0,%1 ;	" |
| 186 | 	"# atomic_fetchadd_int" |
| 187 | 	: "+r" (v),			/* 0 */ |
| 188 | 	 "+m" (*p)			/* 1 */ |
| 189 | 	: : "cc"); |
| 190 | 	return (v); |
| 191 | } |
| 192 | |
| 193 | static __inline int |
| 194 | atomic_testandset_int(volatile u_int *p, u_int v) |
| 195 | { |
| 196 | 	u_char res; |
| 197 | |
| 198 | 	__asm __volatile( |
| 199 | 	"	lock; btsl	%2,%1 ;	" |
| 200 | 	"	setc	%0 ;		" |
| 201 | 	"# atomic_testandset_int" |
| 202 | 	: "=q" (res),			/* 0 */ |
| 203 | 	 "+m" (*p)			/* 1 */ |
| 204 | 	: "Ir" (v & 0x1f)		/* 2 */ |
| 205 | 	: "cc"); |
| 206 | 	return (res); |
| 207 | } |
| 208 | |
| 209 | static __inline int |
| 210 | atomic_testandclear_int(volatile u_int *p, u_int v) |
| 211 | { |
| 212 | 	u_char res; |
| 213 | |
| 214 | 	__asm __volatile( |
| 215 | 	"	lock; btrl	%2,%1 ;	" |
| 216 | 	"	setc	%0 ;		" |
| 217 | 	"# atomic_testandclear_int" |
| 218 | 	: "=q" (res),			/* 0 */ |
| 219 | 	 "+m" (*p)			/* 1 */ |
| 220 | 	: "Ir" (v & 0x1f)		/* 2 */ |
| 221 | 	: "cc"); |
| 222 | 	return (res); |
| 223 | } |
| 224 | |
| 225 | /* |
| 226 | * We assume that a = b will do atomic loads and stores. Due to the |
| 227 | * IA32 memory model, a simple store guarantees release semantics. |
| 228 | * |
| 229 | * However, a load may pass a store if they are performed on distinct |
| 230 | * addresses, so we need Store/Load barrier for sequentially |
| 231 | * consistent fences in SMP kernels. We use "lock addl $0,mem" for a |
| 232 | * Store/Load barrier, as recommended by the AMD Software Optimization |
| 233 | * Guide, and not mfence. In the kernel, we use a private per-cpu |
| 234 | * cache line for "mem", to avoid introducing false data |
| 235 | * dependencies. In user space, we use the word at the top of the |
| 236 | * stack. |
| 237 | * |
| 238 | * For UP kernels, however, the memory of the single processor is |
| 239 | * always consistent, so we only need to stop the compiler from |
| 240 | * reordering accesses in a way that violates the semantics of acquire |
| 241 | * and release. |
| 242 | */ |
| 243 | |
| 244 | #if defined(_KERNEL) |
| 245 | #define	__storeload_barrier()	__mbk() |
| 246 | #else /* !_KERNEL */ |
| 247 | #define	__storeload_barrier()	__mbu() |
| 248 | #endif /* _KERNEL*/ |
| 249 | |
| 250 | #define	ATOMIC_LOAD(TYPE)					\ |
| 251 | static __inline u_##TYPE					\ |
| 252 | atomic_load_acq_##TYPE(const volatile u_##TYPE *p)		\ |
| 253 | {								\ |
| 254 | 	u_##TYPE res;						\ |
| 255 | 								\ |
| 256 | 	res = *p;						\ |
| 257 | 	__compiler_membar();					\ |
| 258 | 	return (res);						\ |
| 259 | }								\ |
| 260 | struct __hack |
| 261 | |
| 262 | #define	ATOMIC_STORE(TYPE)					\ |
| 263 | static __inline void						\ |
| 264 | atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)	\ |
| 265 | {								\ |
| 266 | 								\ |
| 267 | 	__compiler_membar();					\ |
| 268 | 	*p = v;							\ |
| 269 | }								\ |
| 270 | struct __hack |
| 271 | |
| 272 | static __inline void |
| 273 | atomic_thread_fence_acq(void) |
| 274 | { |
| 275 | |
| 276 | 	__compiler_membar(); |
| 277 | } |
| 278 | |
| 279 | static __inline void |
| 280 | atomic_thread_fence_rel(void) |
| 281 | { |
| 282 | |
| 283 | 	__compiler_membar(); |
| 284 | } |
| 285 | |
| 286 | static __inline void |
| 287 | atomic_thread_fence_acq_rel(void) |
| 288 | { |
| 289 | |
| 290 | 	__compiler_membar(); |
| 291 | } |
| 292 | |
| 293 | static __inline void |
| 294 | atomic_thread_fence_seq_cst(void) |
| 295 | { |
| 296 | |
| 297 | 	__storeload_barrier(); |
| 298 | } |
| 299 | |
| 300 | #ifdef _KERNEL |
| 301 | |
| 302 | /* I486 does not support SMP or CMPXCHG8B. */ |
| 303 | static __inline int |
| 304 | atomic_cmpset_64_i386(volatile uint64_t *dst, uint64_t expect, uint64_t src) |
| 305 | { |
| 306 | 	volatile uint32_t *p; |
| 307 | 	u_char res; |
| 308 | |
| 309 | 	p = (volatile uint32_t *)dst; |
| 310 | 	__asm __volatile( |
| 311 | 	"	pushfl ;		" |
| 312 | 	"	cli ;			" |
| 313 | 	"	xorl	%1,%%eax ;	" |
| 314 | 	"	xorl	%2,%%edx ;	" |
| 315 | 	"	orl	%%edx,%%eax ;	" |
| 316 | 	"	jne	1f ;		" |
| 317 | 	"	movl	%4,%1 ;		" |
| 318 | 	"	movl	%5,%2 ;		" |
| 319 | 	"1:				" |
| 320 | 	"	sete	%3 ;		" |
| 321 | 	"	popfl" |
| 322 | 	: "+A" (expect),		/* 0 */ |
| 323 | 	 "+m" (*p),			/* 1 */ |
| 324 | 	 "+m" (*(p + 1)),		/* 2 */ |
| 325 | 	 "=q" (res)			/* 3 */ |
| 326 | 	: "r" ((uint32_t)src),		/* 4 */ |
| 327 | 	 "r" ((uint32_t)(src >> 32))	/* 5 */ |
| 328 | 	: "memory", "cc"); |
| 329 | 	return (res); |
| 330 | } |
| 331 | |
| 332 | static __inline int |
| 333 | atomic_fcmpset_64_i386(volatile uint64_t *dst, uint64_t *expect, uint64_t src) |
| 334 | { |
| 335 | |
| 336 | 	if (atomic_cmpset_64_i386(dst, *expect, src)) { |
| 337 | 		return (1); |
| 338 | 	} else { |
| 339 | 		*expect = *dst; |
| 340 | 		return (0); |
| 341 | 	} |
| 342 | } |
| 343 | |
| 344 | static __inline uint64_t |
| 345 | atomic_load_acq_64_i386(const volatile uint64_t *p) |
| 346 | { |
| 347 | 	const volatile uint32_t *q; |
| 348 | 	uint64_t res; |
| 349 | |
| 350 | 	q = (const volatile uint32_t *)p; |
| 351 | 	__asm __volatile( |
| 352 | 	"	pushfl ;		" |
| 353 | 	"	cli ;			" |
| 354 | 	"	movl	%1,%%eax ;	" |
| 355 | 	"	movl	%2,%%edx ;	" |
| 356 | 	"	popfl" |
| 357 | 	: "=&A" (res)			/* 0 */ |
| 358 | 	: "m" (*q),			/* 1 */ |
| 359 | 	 "m" (*(q + 1))		/* 2 */ |
| 360 | 	: "memory"); |
| 361 | 	return (res); |
| 362 | } |
| 363 | |
| 364 | static __inline void |
| 365 | atomic_store_rel_64_i386(volatile uint64_t *p, uint64_t v) |
| 366 | { |
| 367 | 	volatile uint32_t *q; |
| 368 | |
| 369 | 	q = (volatile uint32_t *)p; |
| 370 | 	__asm __volatile( |
| 371 | 	"	pushfl ;		" |
| 372 | 	"	cli ;			" |
| 373 | 	"	movl	%%eax,%0 ;	" |
| 374 | 	"	movl	%%edx,%1 ;	" |
| 375 | 	"	popfl" |
| 376 | 	: "=m" (*q),			/* 0 */ |
| 377 | 	 "=m" (*(q + 1))		/* 1 */ |
| 378 | 	: "A" (v)			/* 2 */ |
| 379 | 	: "memory"); |
| 380 | } |
| 381 | |
| 382 | static __inline uint64_t |
| 383 | atomic_swap_64_i386(volatile uint64_t *p, uint64_t v) |
| 384 | { |
| 385 | 	volatile uint32_t *q; |
| 386 | 	uint64_t res; |
| 387 | |
| 388 | 	q = (volatile uint32_t *)p; |
| 389 | 	__asm __volatile( |
| 390 | 	"	pushfl ;		" |
| 391 | 	"	cli ;			" |
| 392 | 	"	movl	%1,%%eax ;	" |
| 393 | 	"	movl	%2,%%edx ;	" |
| 394 | 	"	movl	%4,%2 ;		" |
| 395 | 	"	movl	%3,%1 ;		" |
| 396 | 	"	popfl" |
| 397 | 	: "=&A" (res),			/* 0 */ |
| 398 | 	 "+m" (*q),			/* 1 */ |
| 399 | 	 "+m" (*(q + 1))		/* 2 */ |
| 400 | 	: "r" ((uint32_t)v),		/* 3 */ |
| 401 | 	 "r" ((uint32_t)(v >> 32)));	/* 4 */ |
| 402 | 	return (res); |
| 403 | } |
| 404 | |
| 405 | static __inline int |
| 406 | atomic_cmpset_64_i586(volatile uint64_t *dst, uint64_t expect, uint64_t src) |
| 407 | { |
| 408 | 	u_char res; |
| 409 | |
| 410 | 	__asm __volatile( |
| 411 | 	"	lock; cmpxchg8b %1 ;	" |
| 412 | 	"	sete	%0" |
| 413 | 	: "=q" (res),			/* 0 */ |
| 414 | 	 "+m" (*dst),			/* 1 */ |
| 415 | 	 "+A" (expect)			/* 2 */ |
| 416 | 	: "b" ((uint32_t)src),		/* 3 */ |
| 417 | 	 "c" ((uint32_t)(src >> 32))	/* 4 */ |
| 418 | 	: "memory", "cc"); |
| 419 | 	return (res); |
| 420 | } |
| 421 | |
| 422 | static __inline int |
| 423 | atomic_fcmpset_64_i586(volatile uint64_t *dst, uint64_t *expect, uint64_t src) |
| 424 | { |
| 425 | 	u_char res; |
| 426 | |
| 427 | 	__asm __volatile( |
| 428 | 	"	lock; cmpxchg8b %1 ;	" |
| 429 | 	"	sete	%0" |
| 430 | 	: "=q" (res),			/* 0 */ |
| 431 | 	 "+m" (*dst),			/* 1 */ |
| 432 | 	 "+A" (*expect)		/* 2 */ |
| 433 | 	: "b" ((uint32_t)src),		/* 3 */ |
| 434 | 	 "c" ((uint32_t)(src >> 32))	/* 4 */ |
| 435 | 	: "memory", "cc"); |
| 436 | 	return (res); |
| 437 | } |
| 438 | |
| 439 | /* |
| 440 | * Architecturally always writes back some value to '*p' so will trigger |
| 441 | * a #GP(0) on read-only mappings. |
| 442 | */ |
| 443 | static __inline uint64_t |
| 444 | atomic_load_acq_64_i586(const volatile uint64_t *p) |
| 445 | { |
| 446 | 	uint64_t res; |
| 447 | |
| 448 | 	__asm __volatile( |
| 449 | 	"	movl	%%ebx,%%eax ;	" |
| 450 | 	"	movl	%%ecx,%%edx ;	" |
| 451 | 	"	lock; cmpxchg8b %1" |
| 452 | 	: "=&A" (res)			/* 0 */ |
| 453 | 	: "m" (*p)			/* 1 */ |
| 454 | 	: "memory", "cc"); |
| 455 | 	return (res); |
| 456 | } |
| 457 | |
| 458 | static __inline void |
| 459 | atomic_store_rel_64_i586(volatile uint64_t *p, uint64_t v) |
| 460 | { |
| 461 | |
| 462 | 	__asm __volatile( |
| 463 | 	"	movl	%%eax,%%ebx ;	" |
| 464 | 	"	movl	%%edx,%%ecx ;	" |
| 465 | 	"1:				" |
| 466 | 	"	lock; cmpxchg8b %0 ;	" |
| 467 | 	"	jne	1b" |
| 468 | 	: "+m" (*p),			/* 0 */ |
| 469 | 	 "+A" (v)			/* 1 */ |
| 470 | 	: : "ebx", "ecx", "memory", "cc"); |
| 471 | } |
| 472 | |
| 473 | static __inline uint64_t |
| 474 | atomic_swap_64_i586(volatile uint64_t *p, uint64_t v) |
| 475 | { |
| 476 | |
| 477 | 	__asm __volatile( |
| 478 | 	"	movl	%%eax,%%ebx ;	" |
| 479 | 	"	movl	%%edx,%%ecx ;	" |
| 480 | 	"1:				" |
| 481 | 	"	lock; cmpxchg8b %0 ;	" |
| 482 | 	"	jne	1b" |
| 483 | 	: "+m" (*p),			/* 0 */ |
| 484 | 	 "+A" (v)			/* 1 */ |
| 485 | 	: : "ebx", "ecx", "memory", "cc"); |
| 486 | 	return (v); |
| 487 | } |
| 488 | |
| 489 | static __inline int |
| 490 | atomic_cmpset_64(volatile uint64_t *dst, uint64_t expect, uint64_t src) |
| 491 | { |
| 492 | |
| 493 | 	if ((cpu_feature & CPUID_CX8) == 0) |
| 494 | 		return (atomic_cmpset_64_i386(dst, expect, src)); |
| 495 | 	else |
| 496 | 		return (atomic_cmpset_64_i586(dst, expect, src)); |
| 497 | } |
| 498 | |
| 499 | static __inline int |
| 500 | atomic_fcmpset_64(volatile uint64_t *dst, uint64_t *expect, uint64_t src) |
| 501 | { |
| 502 | |
| 503 | 	if ((cpu_feature & CPUID_CX8) == 0) |
| 504 | 		return (atomic_fcmpset_64_i386(dst, expect, src)); |
| 505 | 	else |
| 506 | 		return (atomic_fcmpset_64_i586(dst, expect, src)); |
| 507 | } |
| 508 | |
| 509 | static __inline uint64_t |
| 510 | atomic_load_acq_64(const volatile uint64_t *p) |
| 511 | { |
| 512 | |
| 513 | 	if ((cpu_feature & CPUID_CX8) == 0) |
| 514 | 		return (atomic_load_acq_64_i386(p)); |
| 515 | 	else |
| 516 | 		return (atomic_load_acq_64_i586(p)); |
| 517 | } |
| 518 | |
| 519 | static __inline void |
| 520 | atomic_store_rel_64(volatile uint64_t *p, uint64_t v) |
| 521 | { |
| 522 | |
| 523 | 	if ((cpu_feature & CPUID_CX8) == 0) |
| 524 | 		atomic_store_rel_64_i386(p, v); |
| 525 | 	else |
| 526 | 		atomic_store_rel_64_i586(p, v); |
| 527 | } |
| 528 | |
| 529 | static __inline uint64_t |
| 530 | atomic_swap_64(volatile uint64_t *p, uint64_t v) |
| 531 | { |
| 532 | |
| 533 | 	if ((cpu_feature & CPUID_CX8) == 0) |
| 534 | 		return (atomic_swap_64_i386(p, v)); |
| 535 | 	else |
| 536 | 		return (atomic_swap_64_i586(p, v)); |
| 537 | } |
| 538 | |
| 539 | static __inline uint64_t |
| 540 | atomic_fetchadd_64(volatile uint64_t *p, uint64_t v) |
| 541 | { |
| 542 | |
| 543 | 	for (;;) { |
| 544 | 		uint64_t t = *p; |
| 545 | 		if (atomic_cmpset_64(p, t, t + v)) |
| 546 | 			return (t); |
| 547 | 	} |
| 548 | } |
| 549 | |
| 550 | static __inline void |
| 551 | atomic_add_64(volatile uint64_t *p, uint64_t v) |
| 552 | { |
| 553 | 	uint64_t t; |
| 554 | |
| 555 | 	for (;;) { |
| 556 | 		t = *p; |
| 557 | 		if (atomic_cmpset_64(p, t, t + v)) |
| 558 | 			break; |
| 559 | 	} |
| 560 | } |
| 561 | |
| 562 | static __inline void |
| 563 | atomic_subtract_64(volatile uint64_t *p, uint64_t v) |
| 564 | { |
| 565 | 	uint64_t t; |
| 566 | |
| 567 | 	for (;;) { |
| 568 | 		t = *p; |
| 569 | 		if (atomic_cmpset_64(p, t, t - v)) |
| 570 | 			break; |
| 571 | 	} |
| 572 | } |
| 573 | |
| 574 | #endif /* _KERNEL */ |
| 575 | |
| 576 | ATOMIC_ASM(set,	 char, "orb %b1,%0", "iq", v); |
| 577 | ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v); |
| 578 | ATOMIC_ASM(add,	 char, "addb %b1,%0", "iq", v); |
| 579 | ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v); |
| 580 | |
| 581 | ATOMIC_ASM(set,	 short, "orw %w1,%0", "ir", v); |
| 582 | ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v); |
| 583 | ATOMIC_ASM(add,	 short, "addw %w1,%0", "ir", v); |
| 584 | ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v); |
| 585 | |
| 586 | ATOMIC_ASM(set,	 int, "orl %1,%0", "ir", v); |
| 587 | ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v); |
| 588 | ATOMIC_ASM(add,	 int, "addl %1,%0", "ir", v); |
| 589 | ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v); |
| 590 | |
| 591 | ATOMIC_ASM(set,	 long, "orl %1,%0", "ir", v); |
| 592 | ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v); |
| 593 | ATOMIC_ASM(add,	 long, "addl %1,%0", "ir", v); |
| 594 | ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v); |
| 595 | |
| 596 | #define	ATOMIC_LOADSTORE(TYPE)				\ |
| 597 | 	ATOMIC_LOAD(TYPE);				\ |
| 598 | 	ATOMIC_STORE(TYPE) |
| 599 | |
| 600 | ATOMIC_LOADSTORE(char); |
| 601 | ATOMIC_LOADSTORE(short); |
| 602 | ATOMIC_LOADSTORE(int); |
| 603 | ATOMIC_LOADSTORE(long); |
| 604 | |
| 605 | #undef ATOMIC_ASM |
| 606 | #undef ATOMIC_LOAD |
| 607 | #undef ATOMIC_STORE |
| 608 | #undef ATOMIC_LOADSTORE |
| 609 | |
| 610 | static __inline int |
| 611 | atomic_cmpset_long(volatile u_long *dst, u_long expect, u_long src) |
| 612 | { |
| 613 | |
| 614 | 	return (atomic_cmpset_int((volatile u_int *)dst, (u_int)expect, |
| 615 | 	 (u_int)src)); |
| 616 | } |
| 617 | |
| 618 | static __inline int |
| 619 | atomic_fcmpset_long(volatile u_long *dst, u_long *expect, u_long src) |
| 620 | { |
| 621 | |
| 622 | 	return (atomic_fcmpset_int((volatile u_int *)dst, (u_int *)expect, |
| 623 | 	 (u_int)src)); |
| 624 | } |
| 625 | |
| 626 | static __inline u_long |
| 627 | atomic_fetchadd_long(volatile u_long *p, u_long v) |
| 628 | { |
| 629 | |
| 630 | 	return (atomic_fetchadd_int((volatile u_int *)p, (u_int)v)); |
| 631 | } |
| 632 | |
| 633 | static __inline int |
| 634 | atomic_testandset_long(volatile u_long *p, u_int v) |
| 635 | { |
| 636 | |
| 637 | 	return (atomic_testandset_int((volatile u_int *)p, v)); |
| 638 | } |
| 639 | |
| 640 | static __inline int |
| 641 | atomic_testandclear_long(volatile u_long *p, u_int v) |
| 642 | { |
| 643 | |
| 644 | 	return (atomic_testandclear_int((volatile u_int *)p, v)); |
| 645 | } |
| 646 | |
| 647 | /* Read the current value and store a new value in the destination. */ |
| 648 | static __inline u_int |
| 649 | atomic_swap_int(volatile u_int *p, u_int v) |
| 650 | { |
| 651 | |
| 652 | 	__asm __volatile( |
| 653 | 	"	xchgl	%1,%0 ;		" |
| 654 | 	"# atomic_swap_int" |
| 655 | 	: "+r" (v),			/* 0 */ |
| 656 | 	 "+m" (*p));			/* 1 */ |
| 657 | 	return (v); |
| 658 | } |
| 659 | |
| 660 | static __inline u_long |
| 661 | atomic_swap_long(volatile u_long *p, u_long v) |
| 662 | { |
| 663 | |
| 664 | 	return (atomic_swap_int((volatile u_int *)p, (u_int)v)); |
| 665 | } |
| 666 | |
| 667 | #define	atomic_set_acq_char		atomic_set_barr_char |
| 668 | #define	atomic_set_rel_char		atomic_set_barr_char |
| 669 | #define	atomic_clear_acq_char		atomic_clear_barr_char |
| 670 | #define	atomic_clear_rel_char		atomic_clear_barr_char |
| 671 | #define	atomic_add_acq_char		atomic_add_barr_char |
| 672 | #define	atomic_add_rel_char		atomic_add_barr_char |
| 673 | #define	atomic_subtract_acq_char	atomic_subtract_barr_char |
| 674 | #define	atomic_subtract_rel_char	atomic_subtract_barr_char |
| 675 | #define	atomic_cmpset_acq_char		atomic_cmpset_char |
| 676 | #define	atomic_cmpset_rel_char		atomic_cmpset_char |
| 677 | #define	atomic_fcmpset_acq_char		atomic_fcmpset_char |
| 678 | #define	atomic_fcmpset_rel_char		atomic_fcmpset_char |
| 679 | |
| 680 | #define	atomic_set_acq_short		atomic_set_barr_short |
| 681 | #define	atomic_set_rel_short		atomic_set_barr_short |
| 682 | #define	atomic_clear_acq_short		atomic_clear_barr_short |
| 683 | #define	atomic_clear_rel_short		atomic_clear_barr_short |
| 684 | #define	atomic_add_acq_short		atomic_add_barr_short |
| 685 | #define	atomic_add_rel_short		atomic_add_barr_short |
| 686 | #define	atomic_subtract_acq_short	atomic_subtract_barr_short |
| 687 | #define	atomic_subtract_rel_short	atomic_subtract_barr_short |
| 688 | #define	atomic_cmpset_acq_short		atomic_cmpset_short |
| 689 | #define	atomic_cmpset_rel_short		atomic_cmpset_short |
| 690 | #define	atomic_fcmpset_acq_short	atomic_fcmpset_short |
| 691 | #define	atomic_fcmpset_rel_short	atomic_fcmpset_short |
| 692 | |
| 693 | #define	atomic_set_acq_int		atomic_set_barr_int |
| 694 | #define	atomic_set_rel_int		atomic_set_barr_int |
| 695 | #define	atomic_clear_acq_int		atomic_clear_barr_int |
| 696 | #define	atomic_clear_rel_int		atomic_clear_barr_int |
| 697 | #define	atomic_add_acq_int		atomic_add_barr_int |
| 698 | #define	atomic_add_rel_int		atomic_add_barr_int |
| 699 | #define	atomic_subtract_acq_int		atomic_subtract_barr_int |
| 700 | #define	atomic_subtract_rel_int		atomic_subtract_barr_int |
| 701 | #define	atomic_cmpset_acq_int		atomic_cmpset_int |
| 702 | #define	atomic_cmpset_rel_int		atomic_cmpset_int |
| 703 | #define	atomic_fcmpset_acq_int		atomic_fcmpset_int |
| 704 | #define	atomic_fcmpset_rel_int		atomic_fcmpset_int |
| 705 | |
| 706 | #define	atomic_set_acq_long		atomic_set_barr_long |
| 707 | #define	atomic_set_rel_long		atomic_set_barr_long |
| 708 | #define	atomic_clear_acq_long		atomic_clear_barr_long |
| 709 | #define	atomic_clear_rel_long		atomic_clear_barr_long |
| 710 | #define	atomic_add_acq_long		atomic_add_barr_long |
| 711 | #define	atomic_add_rel_long		atomic_add_barr_long |
| 712 | #define	atomic_subtract_acq_long	atomic_subtract_barr_long |
| 713 | #define	atomic_subtract_rel_long	atomic_subtract_barr_long |
| 714 | #define	atomic_cmpset_acq_long		atomic_cmpset_long |
| 715 | #define	atomic_cmpset_rel_long		atomic_cmpset_long |
| 716 | #define	atomic_fcmpset_acq_long		atomic_fcmpset_long |
| 717 | #define	atomic_fcmpset_rel_long		atomic_fcmpset_long |
| 718 | |
| 719 | #define	atomic_readandclear_int(p)	atomic_swap_int(p, 0) |
| 720 | #define	atomic_readandclear_long(p)	atomic_swap_long(p, 0) |
| 721 | #define	atomic_testandset_acq_long	atomic_testandset_long |
| 722 | |
| 723 | /* Operations on 8-bit bytes. */ |
| 724 | #define	atomic_set_8		atomic_set_char |
| 725 | #define	atomic_set_acq_8	atomic_set_acq_char |
| 726 | #define	atomic_set_rel_8	atomic_set_rel_char |
| 727 | #define	atomic_clear_8		atomic_clear_char |
| 728 | #define	atomic_clear_acq_8	atomic_clear_acq_char |
| 729 | #define	atomic_clear_rel_8	atomic_clear_rel_char |
| 730 | #define	atomic_add_8		atomic_add_char |
| 731 | #define	atomic_add_acq_8	atomic_add_acq_char |
| 732 | #define	atomic_add_rel_8	atomic_add_rel_char |
| 733 | #define	atomic_subtract_8	atomic_subtract_char |
| 734 | #define	atomic_subtract_acq_8	atomic_subtract_acq_char |
| 735 | #define	atomic_subtract_rel_8	atomic_subtract_rel_char |
| 736 | #define	atomic_load_acq_8	atomic_load_acq_char |
| 737 | #define	atomic_store_rel_8	atomic_store_rel_char |
| 738 | #define	atomic_cmpset_8		atomic_cmpset_char |
| 739 | #define	atomic_cmpset_acq_8	atomic_cmpset_acq_char |
| 740 | #define	atomic_cmpset_rel_8	atomic_cmpset_rel_char |
| 741 | #define	atomic_fcmpset_8	atomic_fcmpset_char |
| 742 | #define	atomic_fcmpset_acq_8	atomic_fcmpset_acq_char |
| 743 | #define	atomic_fcmpset_rel_8	atomic_fcmpset_rel_char |
| 744 | |
| 745 | /* Operations on 16-bit words. */ |
| 746 | #define	atomic_set_16		atomic_set_short |
| 747 | #define	atomic_set_acq_16	atomic_set_acq_short |
| 748 | #define	atomic_set_rel_16	atomic_set_rel_short |
| 749 | #define	atomic_clear_16		atomic_clear_short |
| 750 | #define	atomic_clear_acq_16	atomic_clear_acq_short |
| 751 | #define	atomic_clear_rel_16	atomic_clear_rel_short |
| 752 | #define	atomic_add_16		atomic_add_short |
| 753 | #define	atomic_add_acq_16	atomic_add_acq_short |
| 754 | #define	atomic_add_rel_16	atomic_add_rel_short |
| 755 | #define	atomic_subtract_16	atomic_subtract_short |
| 756 | #define	atomic_subtract_acq_16	atomic_subtract_acq_short |
| 757 | #define	atomic_subtract_rel_16	atomic_subtract_rel_short |
| 758 | #define	atomic_load_acq_16	atomic_load_acq_short |
| 759 | #define	atomic_store_rel_16	atomic_store_rel_short |
| 760 | #define	atomic_cmpset_16	atomic_cmpset_short |
| 761 | #define	atomic_cmpset_acq_16	atomic_cmpset_acq_short |
| 762 | #define	atomic_cmpset_rel_16	atomic_cmpset_rel_short |
| 763 | #define	atomic_fcmpset_16	atomic_fcmpset_short |
| 764 | #define	atomic_fcmpset_acq_16	atomic_fcmpset_acq_short |
| 765 | #define	atomic_fcmpset_rel_16	atomic_fcmpset_rel_short |
| 766 | |
| 767 | /* Operations on 32-bit double words. */ |
| 768 | #define	atomic_set_32		atomic_set_int |
| 769 | #define	atomic_set_acq_32	atomic_set_acq_int |
| 770 | #define	atomic_set_rel_32	atomic_set_rel_int |
| 771 | #define	atomic_clear_32		atomic_clear_int |
| 772 | #define	atomic_clear_acq_32	atomic_clear_acq_int |
| 773 | #define	atomic_clear_rel_32	atomic_clear_rel_int |
| 774 | #define	atomic_add_32		atomic_add_int |
| 775 | #define	atomic_add_acq_32	atomic_add_acq_int |
| 776 | #define	atomic_add_rel_32	atomic_add_rel_int |
| 777 | #define	atomic_subtract_32	atomic_subtract_int |
| 778 | #define	atomic_subtract_acq_32	atomic_subtract_acq_int |
| 779 | #define	atomic_subtract_rel_32	atomic_subtract_rel_int |
| 780 | #define	atomic_load_acq_32	atomic_load_acq_int |
| 781 | #define	atomic_store_rel_32	atomic_store_rel_int |
| 782 | #define	atomic_cmpset_32	atomic_cmpset_int |
| 783 | #define	atomic_cmpset_acq_32	atomic_cmpset_acq_int |
| 784 | #define	atomic_cmpset_rel_32	atomic_cmpset_rel_int |
| 785 | #define	atomic_fcmpset_32	atomic_fcmpset_int |
| 786 | #define	atomic_fcmpset_acq_32	atomic_fcmpset_acq_int |
| 787 | #define	atomic_fcmpset_rel_32	atomic_fcmpset_rel_int |
| 788 | #define	atomic_swap_32		atomic_swap_int |
| 789 | #define	atomic_readandclear_32	atomic_readandclear_int |
| 790 | #define	atomic_fetchadd_32	atomic_fetchadd_int |
| 791 | #define	atomic_testandset_32	atomic_testandset_int |
| 792 | #define	atomic_testandclear_32	atomic_testandclear_int |
| 793 | |
| 794 | #ifdef _KERNEL |
| 795 | /* Operations on 64-bit quad words. */ |
| 796 | #define	atomic_cmpset_acq_64 atomic_cmpset_64 |
| 797 | #define	atomic_cmpset_rel_64 atomic_cmpset_64 |
| 798 | #define	atomic_fcmpset_acq_64 atomic_fcmpset_64 |
| 799 | #define	atomic_fcmpset_rel_64 atomic_fcmpset_64 |
| 800 | #define	atomic_fetchadd_acq_64	atomic_fetchadd_64 |
| 801 | #define	atomic_fetchadd_rel_64	atomic_fetchadd_64 |
| 802 | #define	atomic_add_acq_64 atomic_add_64 |
| 803 | #define	atomic_add_rel_64 atomic_add_64 |
| 804 | #define	atomic_subtract_acq_64 atomic_subtract_64 |
| 805 | #define	atomic_subtract_rel_64 atomic_subtract_64 |
| 806 | #define	atomic_load_64 atomic_load_acq_64 |
| 807 | #define	atomic_store_64 atomic_store_rel_64 |
| 808 | #endif |
| 809 | |
| 810 | /* Operations on pointers. */ |
| 811 | #define	atomic_set_ptr(p, v) \ |
| 812 | 	atomic_set_int((volatile u_int *)(p), (u_int)(v)) |
| 813 | #define	atomic_set_acq_ptr(p, v) \ |
| 814 | 	atomic_set_acq_int((volatile u_int *)(p), (u_int)(v)) |
| 815 | #define	atomic_set_rel_ptr(p, v) \ |
| 816 | 	atomic_set_rel_int((volatile u_int *)(p), (u_int)(v)) |
| 817 | #define	atomic_clear_ptr(p, v) \ |
| 818 | 	atomic_clear_int((volatile u_int *)(p), (u_int)(v)) |
| 819 | #define	atomic_clear_acq_ptr(p, v) \ |
| 820 | 	atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v)) |
| 821 | #define	atomic_clear_rel_ptr(p, v) \ |
| 822 | 	atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v)) |
| 823 | #define	atomic_add_ptr(p, v) \ |
| 824 | 	atomic_add_int((volatile u_int *)(p), (u_int)(v)) |
| 825 | #define	atomic_add_acq_ptr(p, v) \ |
| 826 | 	atomic_add_acq_int((volatile u_int *)(p), (u_int)(v)) |
| 827 | #define	atomic_add_rel_ptr(p, v) \ |
| 828 | 	atomic_add_rel_int((volatile u_int *)(p), (u_int)(v)) |
| 829 | #define	atomic_subtract_ptr(p, v) \ |
| 830 | 	atomic_subtract_int((volatile u_int *)(p), (u_int)(v)) |
| 831 | #define	atomic_subtract_acq_ptr(p, v) \ |
| 832 | 	atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v)) |
| 833 | #define	atomic_subtract_rel_ptr(p, v) \ |
| 834 | 	atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v)) |
| 835 | #define	atomic_load_acq_ptr(p) \ |
| 836 | 	atomic_load_acq_int((const volatile u_int *)(p)) |
| 837 | #define	atomic_store_rel_ptr(p, v) \ |
| 838 | 	atomic_store_rel_int((volatile u_int *)(p), (v)) |
| 839 | #define	atomic_cmpset_ptr(dst, old, new) \ |
| 840 | 	atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new)) |
| 841 | #define	atomic_cmpset_acq_ptr(dst, old, new) \ |
| 842 | 	atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \ |
| 843 | 	 (u_int)(new)) |
| 844 | #define	atomic_cmpset_rel_ptr(dst, old, new) \ |
| 845 | 	atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \ |
| 846 | 	 (u_int)(new)) |
| 847 | #define	atomic_fcmpset_ptr(dst, old, new) \ |
| 848 | 	atomic_fcmpset_int((volatile u_int *)(dst), (u_int *)(old), (u_int)(new)) |
| 849 | #define	atomic_fcmpset_acq_ptr(dst, old, new) \ |
| 850 | 	atomic_fcmpset_acq_int((volatile u_int *)(dst), (u_int *)(old), \ |
| 851 | 	 (u_int)(new)) |
| 852 | #define	atomic_fcmpset_rel_ptr(dst, old, new) \ |
| 853 | 	atomic_fcmpset_rel_int((volatile u_int *)(dst), (u_int *)(old), \ |
| 854 | 	 (u_int)(new)) |
| 855 | #define	atomic_swap_ptr(p, v) \ |
| 856 | 	atomic_swap_int((volatile u_int *)(p), (u_int)(v)) |
| 857 | #define	atomic_readandclear_ptr(p) \ |
| 858 | 	atomic_readandclear_int((volatile u_int *)(p)) |
| 859 | #define	atomic_testandclear_ptr(p, val) \ |
| 860 | 	atomic_testandclear_int((volatile u_int *)(p), (val)) |
| 861 | #define	atomic_testandset_ptr(p, val) \ |
| 862 | 	atomic_testandset_int((volatile u_int *)(p), (val)) |
| 863 | |
| 864 | #if defined(_KERNEL) |
| 865 | #define	mb()	__mbk() |
| 866 | #define	wmb()	__mbk() |
| 867 | #define	rmb()	__mbk() |
| 868 | #else |
| 869 | #define	mb()	__mbu() |
| 870 | #define	wmb()	__mbu() |
| 871 | #define	rmb()	__mbu() |
| 872 | #endif |
| 873 | |
| 874 | #endif /* !_MACHINE_ATOMIC_H_ */ |