1998-08-24 08:39:39 +00:00
|
|
|
/*-
|
|
|
|
* Copyright (c) 1998 Doug Rabson
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*
|
1999-08-28 01:08:13 +00:00
|
|
|
* $FreeBSD$
|
1998-08-24 08:39:39 +00:00
|
|
|
*/
|
|
|
|
#ifndef _MACHINE_ATOMIC_H_
|
2005-07-09 12:38:53 +00:00
|
|
|
#define _MACHINE_ATOMIC_H_
|
1998-08-24 08:39:39 +00:00
|
|
|
|
2005-03-02 21:33:29 +00:00
|
|
|
#ifndef _SYS_CDEFS_H_
|
|
|
|
#error this file needs sys/cdefs.h as a prerequisite
|
|
|
|
#endif
|
|
|
|
|
1998-08-24 08:39:39 +00:00
|
|
|
/*
|
2006-12-29 15:29:49 +00:00
|
|
|
* Various simple operations on memory, each of which is atomic in the
|
|
|
|
* presence of interrupts and multiple processors.
|
1998-08-24 08:39:39 +00:00
|
|
|
*
|
2006-12-29 14:28:23 +00:00
|
|
|
* atomic_set_char(P, V) (*(u_char *)(P) |= (V))
|
|
|
|
* atomic_clear_char(P, V) (*(u_char *)(P) &= ~(V))
|
|
|
|
* atomic_add_char(P, V) (*(u_char *)(P) += (V))
|
|
|
|
* atomic_subtract_char(P, V) (*(u_char *)(P) -= (V))
|
1999-07-13 06:35:25 +00:00
|
|
|
*
|
2006-12-29 14:28:23 +00:00
|
|
|
* atomic_set_short(P, V) (*(u_short *)(P) |= (V))
|
|
|
|
* atomic_clear_short(P, V) (*(u_short *)(P) &= ~(V))
|
|
|
|
* atomic_add_short(P, V) (*(u_short *)(P) += (V))
|
|
|
|
* atomic_subtract_short(P, V) (*(u_short *)(P) -= (V))
|
1999-07-13 06:35:25 +00:00
|
|
|
*
|
2006-12-29 14:28:23 +00:00
|
|
|
* atomic_set_int(P, V) (*(u_int *)(P) |= (V))
|
|
|
|
* atomic_clear_int(P, V) (*(u_int *)(P) &= ~(V))
|
|
|
|
* atomic_add_int(P, V) (*(u_int *)(P) += (V))
|
|
|
|
* atomic_subtract_int(P, V) (*(u_int *)(P) -= (V))
|
2006-12-29 15:29:49 +00:00
|
|
|
* atomic_readandclear_int(P) (return (*(u_int *)(P)); *(u_int *)(P) = 0;)
|
1999-07-13 06:35:25 +00:00
|
|
|
*
|
2006-12-29 14:28:23 +00:00
|
|
|
* atomic_set_long(P, V) (*(u_long *)(P) |= (V))
|
|
|
|
* atomic_clear_long(P, V) (*(u_long *)(P) &= ~(V))
|
|
|
|
* atomic_add_long(P, V) (*(u_long *)(P) += (V))
|
|
|
|
* atomic_subtract_long(P, V) (*(u_long *)(P) -= (V))
|
2006-12-29 15:29:49 +00:00
|
|
|
* atomic_readandclear_long(P) (return (*(u_long *)(P)); *(u_long *)(P) = 0;)
|
1998-08-24 08:39:39 +00:00
|
|
|
*/
|
|
|
|
|
1999-07-13 06:35:25 +00:00
|
|
|
/*
|
1999-08-18 04:08:31 +00:00
|
|
|
* The above functions are expanded inline in the statically-linked
|
|
|
|
* kernel. Lock prefixes are generated if an SMP kernel is being
|
|
|
|
* built.
|
|
|
|
*
|
|
|
|
* Kernel modules call real functions which are built into the kernel.
|
|
|
|
* This allows kernel modules to be portable between UP and SMP systems.
|
1999-07-13 06:35:25 +00:00
|
|
|
*/
|
2005-07-09 12:38:53 +00:00
|
|
|
#if defined(KLD_MODULE) || !defined(__GNUCLIKE_ASM)
|
|
|
|
#define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
|
2002-07-17 16:19:37 +00:00
|
|
|
void atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
|
1999-08-18 04:08:31 +00:00
|
|
|
|
2006-12-29 14:28:23 +00:00
|
|
|
int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
|
|
|
|
u_int atomic_fetchadd_int(volatile u_int *p, u_int v);
|
2000-09-06 11:21:14 +00:00
|
|
|
|
2001-01-16 00:18:36 +00:00
|
|
|
#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
|
|
|
|
u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \
|
2002-07-17 16:19:37 +00:00
|
|
|
void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
|
2001-01-16 00:18:36 +00:00
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
#else /* !KLD_MODULE && __GNUCLIKE_ASM */
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2001-10-08 20:58:24 +00:00
|
|
|
/*
|
2006-12-29 15:29:49 +00:00
|
|
|
* For userland, always use lock prefixes so that the binaries will run
|
|
|
|
* on both SMP and !SMP systems.
|
2001-10-08 20:58:24 +00:00
|
|
|
*/
|
|
|
|
#if defined(SMP) || !defined(_KERNEL)
|
2006-12-29 13:36:26 +00:00
|
|
|
#define MPLOCKED "lock ; "
|
2002-02-11 03:41:59 +00:00
|
|
|
#else
|
2005-07-09 12:38:53 +00:00
|
|
|
#define MPLOCKED
|
2002-02-11 03:41:59 +00:00
|
|
|
#endif
|
1999-07-13 03:32:17 +00:00
|
|
|
|
1999-07-13 06:35:25 +00:00
|
|
|
/*
|
|
|
|
* The assembly is volatilized to demark potential before-and-after side
|
|
|
|
* effects if an interrupt or SMP collision were to occur.
|
|
|
|
*/
|
2005-07-09 12:38:53 +00:00
|
|
|
#define ATOMIC_ASM(NAME, TYPE, OP, CONS, V) \
|
1999-07-13 06:35:25 +00:00
|
|
|
static __inline void \
|
1999-07-23 23:45:50 +00:00
|
|
|
atomic_##NAME##_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
|
1999-07-13 06:35:25 +00:00
|
|
|
{ \
|
2006-12-29 13:36:26 +00:00
|
|
|
__asm __volatile(MPLOCKED OP \
|
2006-12-29 14:28:23 +00:00
|
|
|
: "=m" (*p) \
|
|
|
|
: CONS (V), "m" (*p)); \
|
2003-11-17 02:55:25 +00:00
|
|
|
} \
|
|
|
|
struct __hack
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2000-09-06 11:21:14 +00:00
|
|
|
/*
|
|
|
|
* Atomic compare and set, used by the mutex functions
|
|
|
|
*
|
|
|
|
* if (*dst == exp) *dst = src (all 32 bit words)
|
|
|
|
*
|
|
|
|
* Returns 0 on failure, non-zero on success
|
|
|
|
*/
|
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#ifdef CPU_DISABLE_CMPXCHG
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2000-09-06 11:21:14 +00:00
|
|
|
static __inline int
|
|
|
|
atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
|
|
|
|
{
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
u_char res;
|
2000-09-06 11:21:14 +00:00
|
|
|
|
|
|
|
__asm __volatile(
|
|
|
|
" pushfl ; "
|
|
|
|
" cli ; "
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
" cmpl %3,%4 ; "
|
2000-09-06 11:21:14 +00:00
|
|
|
" jne 1f ; "
|
2005-09-15 19:31:22 +00:00
|
|
|
" movl %2,%1 ; "
|
2000-09-06 11:21:14 +00:00
|
|
|
"1: "
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
" sete %0 ; "
|
2000-09-06 11:21:14 +00:00
|
|
|
" popfl ; "
|
|
|
|
"# atomic_cmpset_int"
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
: "=q" (res), /* 0 */
|
2005-09-15 19:31:22 +00:00
|
|
|
"=m" (*dst) /* 1 */
|
|
|
|
: "r" (src), /* 2 */
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
"r" (exp), /* 3 */
|
|
|
|
"m" (*dst) /* 4 */
|
2000-09-06 11:21:14 +00:00
|
|
|
: "memory");
|
|
|
|
|
|
|
|
return (res);
|
|
|
|
}
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#else /* !CPU_DISABLE_CMPXCHG */
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2000-09-06 11:21:14 +00:00
|
|
|
static __inline int
|
|
|
|
atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
|
|
|
|
{
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
u_char res;
|
2000-09-06 11:21:14 +00:00
|
|
|
|
2006-12-29 14:28:23 +00:00
|
|
|
__asm __volatile(
|
2006-12-29 13:36:26 +00:00
|
|
|
" " MPLOCKED " "
|
2005-09-15 19:31:22 +00:00
|
|
|
" cmpxchgl %2,%1 ; "
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
" sete %0 ; "
|
2000-09-06 11:21:14 +00:00
|
|
|
"1: "
|
|
|
|
"# atomic_cmpset_int"
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
: "=a" (res), /* 0 */
|
2005-09-15 19:31:22 +00:00
|
|
|
"=m" (*dst) /* 1 */
|
|
|
|
: "r" (src), /* 2 */
|
Avoid an instruction in atomic_cmpset_{int_long)() in most cases.
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
2006-12-27 20:26:00 +00:00
|
|
|
"a" (exp), /* 3 */
|
|
|
|
"m" (*dst) /* 4 */
|
2005-09-15 19:31:22 +00:00
|
|
|
: "memory");
|
2000-09-06 11:21:14 +00:00
|
|
|
|
|
|
|
return (res);
|
|
|
|
}
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#endif /* CPU_DISABLE_CMPXCHG */
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2005-09-27 17:39:11 +00:00
|
|
|
/*
|
|
|
|
* Atomically add the value of v to the integer pointed to by p and return
|
|
|
|
* the previous value of *p.
|
|
|
|
*/
|
|
|
|
static __inline u_int
|
|
|
|
atomic_fetchadd_int(volatile u_int *p, u_int v)
|
|
|
|
{
|
|
|
|
|
2006-12-29 14:28:23 +00:00
|
|
|
__asm __volatile(
|
2006-12-29 13:36:26 +00:00
|
|
|
" " MPLOCKED " "
|
2005-09-27 17:39:11 +00:00
|
|
|
" xaddl %0, %1 ; "
|
|
|
|
"# atomic_fetchadd_int"
|
|
|
|
: "+r" (v), /* 0 (result) */
|
|
|
|
"=m" (*p) /* 1 */
|
|
|
|
: "m" (*p)); /* 2 */
|
|
|
|
|
|
|
|
return (v);
|
|
|
|
}
|
|
|
|
|
2004-11-12 19:18:46 +00:00
|
|
|
#if defined(_KERNEL) && !defined(SMP)
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2000-10-20 07:00:48 +00:00
|
|
|
/*
|
2004-11-11 22:42:25 +00:00
|
|
|
* We assume that a = b will do atomic loads and stores. However, on a
|
|
|
|
* PentiumPro or higher, reads may pass writes, so for that case we have
|
|
|
|
* to use a serializing instruction (i.e. with LOCK) to do the load in
|
|
|
|
* SMP kernels. For UP kernels, however, the cache of the single processor
|
|
|
|
* is always consistent, so we don't need any memory barriers.
|
2000-10-20 07:00:48 +00:00
|
|
|
*/
|
2005-07-09 12:38:53 +00:00
|
|
|
#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
|
2000-10-20 07:00:48 +00:00
|
|
|
static __inline u_##TYPE \
|
|
|
|
atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
|
|
|
|
{ \
|
|
|
|
return (*p); \
|
|
|
|
} \
|
|
|
|
\
|
|
|
|
static __inline void \
|
|
|
|
atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
|
|
|
|
{ \
|
2004-11-11 22:42:25 +00:00
|
|
|
*p = v; \
|
2003-11-17 02:55:25 +00:00
|
|
|
} \
|
|
|
|
struct __hack
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#else /* !(_KERNEL && !SMP) */
|
2001-01-14 09:55:21 +00:00
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
|
2001-01-14 09:55:21 +00:00
|
|
|
static __inline u_##TYPE \
|
|
|
|
atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
|
|
|
|
{ \
|
|
|
|
u_##TYPE res; \
|
|
|
|
\
|
2006-12-29 13:36:26 +00:00
|
|
|
__asm __volatile(MPLOCKED LOP \
|
2006-12-29 15:29:49 +00:00
|
|
|
: "=a" (res), /* 0 */ \
|
2005-09-15 19:31:22 +00:00
|
|
|
"=m" (*p) /* 1 */ \
|
|
|
|
: "m" (*p) /* 2 */ \
|
|
|
|
: "memory"); \
|
2001-01-14 09:55:21 +00:00
|
|
|
\
|
|
|
|
return (res); \
|
|
|
|
} \
|
|
|
|
\
|
|
|
|
/* \
|
|
|
|
* The XCHG instruction asserts LOCK automagically. \
|
|
|
|
*/ \
|
|
|
|
static __inline void \
|
|
|
|
atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
|
|
|
|
{ \
|
|
|
|
__asm __volatile(SOP \
|
2005-09-15 19:31:22 +00:00
|
|
|
: "=m" (*p), /* 0 */ \
|
2001-01-14 09:55:21 +00:00
|
|
|
"+r" (v) /* 1 */ \
|
2005-09-15 19:31:22 +00:00
|
|
|
: "m" (*p)); /* 2 */ \
|
2003-11-17 02:55:25 +00:00
|
|
|
} \
|
|
|
|
struct __hack
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#endif /* _KERNEL && !SMP */
|
2002-07-18 15:56:46 +00:00
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
#endif /* KLD_MODULE || !__GNUCLIKE_ASM */
|
2001-01-16 00:18:36 +00:00
|
|
|
|
2002-07-17 16:19:37 +00:00
|
|
|
ATOMIC_ASM(set, char, "orb %b1,%0", "iq", v);
|
|
|
|
ATOMIC_ASM(clear, char, "andb %b1,%0", "iq", ~v);
|
|
|
|
ATOMIC_ASM(add, char, "addb %b1,%0", "iq", v);
|
|
|
|
ATOMIC_ASM(subtract, char, "subb %b1,%0", "iq", v);
|
2001-12-18 08:51:34 +00:00
|
|
|
|
2002-07-17 16:19:37 +00:00
|
|
|
ATOMIC_ASM(set, short, "orw %w1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(clear, short, "andw %w1,%0", "ir", ~v);
|
|
|
|
ATOMIC_ASM(add, short, "addw %w1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(subtract, short, "subw %w1,%0", "ir", v);
|
2001-12-18 08:51:34 +00:00
|
|
|
|
2002-07-17 16:19:37 +00:00
|
|
|
ATOMIC_ASM(set, int, "orl %1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(clear, int, "andl %1,%0", "ir", ~v);
|
|
|
|
ATOMIC_ASM(add, int, "addl %1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(subtract, int, "subl %1,%0", "ir", v);
|
2001-12-18 08:51:34 +00:00
|
|
|
|
2002-07-17 16:19:37 +00:00
|
|
|
ATOMIC_ASM(set, long, "orl %1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(clear, long, "andl %1,%0", "ir", ~v);
|
|
|
|
ATOMIC_ASM(add, long, "addl %1,%0", "ir", v);
|
|
|
|
ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v);
|
2000-10-20 07:00:48 +00:00
|
|
|
|
2002-07-17 16:19:37 +00:00
|
|
|
ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0");
|
|
|
|
ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
|
|
|
|
ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0");
|
|
|
|
ATOMIC_STORE_LOAD(long, "cmpxchgl %0,%1", "xchgl %1,%0");
|
2000-10-20 07:00:48 +00:00
|
|
|
|
2001-01-16 00:18:36 +00:00
|
|
|
#undef ATOMIC_ASM
|
2000-10-20 07:00:48 +00:00
|
|
|
#undef ATOMIC_STORE_LOAD
|
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#ifndef WANT_FUNCTIONS
|
2005-07-09 12:38:53 +00:00
|
|
|
|
|
|
|
static __inline int
|
|
|
|
atomic_cmpset_long(volatile u_long *dst, u_long exp, u_long src)
|
|
|
|
{
|
|
|
|
|
|
|
|
return (atomic_cmpset_int((volatile u_int *)dst, (u_int)exp,
|
|
|
|
(u_int)src));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Read the current value and store a zero in the destination. */
|
|
|
|
#ifdef __GNUCLIKE_ASM
|
|
|
|
|
|
|
|
static __inline u_int
|
|
|
|
atomic_readandclear_int(volatile u_int *addr)
|
|
|
|
{
|
2006-12-29 15:29:49 +00:00
|
|
|
u_int res;
|
2005-07-09 12:38:53 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
res = 0;
|
2006-12-29 14:28:23 +00:00
|
|
|
__asm __volatile(
|
2005-07-09 12:38:53 +00:00
|
|
|
" xchgl %1,%0 ; "
|
|
|
|
"# atomic_readandclear_int"
|
2006-12-29 15:29:49 +00:00
|
|
|
: "+r" (res), /* 0 */
|
|
|
|
"=m" (*addr) /* 1 */
|
2005-09-15 19:31:22 +00:00
|
|
|
: "m" (*addr));
|
2005-07-09 12:38:53 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
return (res);
|
2005-07-09 12:38:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static __inline u_long
|
|
|
|
atomic_readandclear_long(volatile u_long *addr)
|
|
|
|
{
|
2006-12-29 15:29:49 +00:00
|
|
|
u_long res;
|
2005-07-09 12:38:53 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
res = 0;
|
2006-12-29 14:28:23 +00:00
|
|
|
__asm __volatile(
|
2005-07-09 12:38:53 +00:00
|
|
|
" xchgl %1,%0 ; "
|
|
|
|
"# atomic_readandclear_long"
|
2006-12-29 15:48:18 +00:00
|
|
|
: "+r" (res), /* 0 */
|
2006-12-29 15:29:49 +00:00
|
|
|
"=m" (*addr) /* 1 */
|
2005-09-15 19:31:22 +00:00
|
|
|
: "m" (*addr));
|
2005-07-09 12:38:53 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
return (res);
|
2005-07-09 12:38:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
#else /* !__GNUCLIKE_ASM */
|
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
u_int atomic_readandclear_int(volatile u_int *addr);
|
|
|
|
u_long atomic_readandclear_long(volatile u_long *addr);
|
2005-07-09 12:38:53 +00:00
|
|
|
|
|
|
|
#endif /* __GNUCLIKE_ASM */
|
|
|
|
|
|
|
|
/* Acquire and release variants are identical to the normal ones. */
|
2001-01-16 00:18:36 +00:00
|
|
|
#define atomic_set_acq_char atomic_set_char
|
|
|
|
#define atomic_set_rel_char atomic_set_char
|
|
|
|
#define atomic_clear_acq_char atomic_clear_char
|
|
|
|
#define atomic_clear_rel_char atomic_clear_char
|
|
|
|
#define atomic_add_acq_char atomic_add_char
|
|
|
|
#define atomic_add_rel_char atomic_add_char
|
|
|
|
#define atomic_subtract_acq_char atomic_subtract_char
|
|
|
|
#define atomic_subtract_rel_char atomic_subtract_char
|
|
|
|
|
|
|
|
#define atomic_set_acq_short atomic_set_short
|
|
|
|
#define atomic_set_rel_short atomic_set_short
|
|
|
|
#define atomic_clear_acq_short atomic_clear_short
|
|
|
|
#define atomic_clear_rel_short atomic_clear_short
|
|
|
|
#define atomic_add_acq_short atomic_add_short
|
|
|
|
#define atomic_add_rel_short atomic_add_short
|
|
|
|
#define atomic_subtract_acq_short atomic_subtract_short
|
|
|
|
#define atomic_subtract_rel_short atomic_subtract_short
|
|
|
|
|
|
|
|
#define atomic_set_acq_int atomic_set_int
|
|
|
|
#define atomic_set_rel_int atomic_set_int
|
|
|
|
#define atomic_clear_acq_int atomic_clear_int
|
|
|
|
#define atomic_clear_rel_int atomic_clear_int
|
|
|
|
#define atomic_add_acq_int atomic_add_int
|
|
|
|
#define atomic_add_rel_int atomic_add_int
|
|
|
|
#define atomic_subtract_acq_int atomic_subtract_int
|
|
|
|
#define atomic_subtract_rel_int atomic_subtract_int
|
2005-07-09 12:38:53 +00:00
|
|
|
#define atomic_cmpset_acq_int atomic_cmpset_int
|
|
|
|
#define atomic_cmpset_rel_int atomic_cmpset_int
|
2001-01-16 00:18:36 +00:00
|
|
|
|
|
|
|
#define atomic_set_acq_long atomic_set_long
|
|
|
|
#define atomic_set_rel_long atomic_set_long
|
|
|
|
#define atomic_clear_acq_long atomic_clear_long
|
|
|
|
#define atomic_clear_rel_long atomic_clear_long
|
|
|
|
#define atomic_add_acq_long atomic_add_long
|
|
|
|
#define atomic_add_rel_long atomic_add_long
|
|
|
|
#define atomic_subtract_acq_long atomic_subtract_long
|
|
|
|
#define atomic_subtract_rel_long atomic_subtract_long
|
2005-07-09 12:38:53 +00:00
|
|
|
#define atomic_cmpset_acq_long atomic_cmpset_long
|
|
|
|
#define atomic_cmpset_rel_long atomic_cmpset_long
|
2001-01-16 00:18:36 +00:00
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
/* Operations on 8-bit bytes. */
|
2001-01-16 00:18:36 +00:00
|
|
|
#define atomic_set_8 atomic_set_char
|
|
|
|
#define atomic_set_acq_8 atomic_set_acq_char
|
|
|
|
#define atomic_set_rel_8 atomic_set_rel_char
|
|
|
|
#define atomic_clear_8 atomic_clear_char
|
|
|
|
#define atomic_clear_acq_8 atomic_clear_acq_char
|
|
|
|
#define atomic_clear_rel_8 atomic_clear_rel_char
|
|
|
|
#define atomic_add_8 atomic_add_char
|
|
|
|
#define atomic_add_acq_8 atomic_add_acq_char
|
|
|
|
#define atomic_add_rel_8 atomic_add_rel_char
|
|
|
|
#define atomic_subtract_8 atomic_subtract_char
|
|
|
|
#define atomic_subtract_acq_8 atomic_subtract_acq_char
|
|
|
|
#define atomic_subtract_rel_8 atomic_subtract_rel_char
|
|
|
|
#define atomic_load_acq_8 atomic_load_acq_char
|
|
|
|
#define atomic_store_rel_8 atomic_store_rel_char
|
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
/* Operations on 16-bit words. */
|
2001-01-16 00:18:36 +00:00
|
|
|
#define atomic_set_16 atomic_set_short
|
|
|
|
#define atomic_set_acq_16 atomic_set_acq_short
|
|
|
|
#define atomic_set_rel_16 atomic_set_rel_short
|
|
|
|
#define atomic_clear_16 atomic_clear_short
|
|
|
|
#define atomic_clear_acq_16 atomic_clear_acq_short
|
|
|
|
#define atomic_clear_rel_16 atomic_clear_rel_short
|
|
|
|
#define atomic_add_16 atomic_add_short
|
|
|
|
#define atomic_add_acq_16 atomic_add_acq_short
|
|
|
|
#define atomic_add_rel_16 atomic_add_rel_short
|
|
|
|
#define atomic_subtract_16 atomic_subtract_short
|
|
|
|
#define atomic_subtract_acq_16 atomic_subtract_acq_short
|
|
|
|
#define atomic_subtract_rel_16 atomic_subtract_rel_short
|
|
|
|
#define atomic_load_acq_16 atomic_load_acq_short
|
|
|
|
#define atomic_store_rel_16 atomic_store_rel_short
|
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
/* Operations on 32-bit double words. */
|
2001-01-16 00:18:36 +00:00
|
|
|
#define atomic_set_32 atomic_set_int
|
|
|
|
#define atomic_set_acq_32 atomic_set_acq_int
|
|
|
|
#define atomic_set_rel_32 atomic_set_rel_int
|
|
|
|
#define atomic_clear_32 atomic_clear_int
|
|
|
|
#define atomic_clear_acq_32 atomic_clear_acq_int
|
|
|
|
#define atomic_clear_rel_32 atomic_clear_rel_int
|
|
|
|
#define atomic_add_32 atomic_add_int
|
|
|
|
#define atomic_add_acq_32 atomic_add_acq_int
|
|
|
|
#define atomic_add_rel_32 atomic_add_rel_int
|
|
|
|
#define atomic_subtract_32 atomic_subtract_int
|
|
|
|
#define atomic_subtract_acq_32 atomic_subtract_acq_int
|
|
|
|
#define atomic_subtract_rel_32 atomic_subtract_rel_int
|
|
|
|
#define atomic_load_acq_32 atomic_load_acq_int
|
|
|
|
#define atomic_store_rel_32 atomic_store_rel_int
|
|
|
|
#define atomic_cmpset_32 atomic_cmpset_int
|
|
|
|
#define atomic_cmpset_acq_32 atomic_cmpset_acq_int
|
|
|
|
#define atomic_cmpset_rel_32 atomic_cmpset_rel_int
|
|
|
|
#define atomic_readandclear_32 atomic_readandclear_int
|
2005-09-27 17:39:11 +00:00
|
|
|
#define atomic_fetchadd_32 atomic_fetchadd_int
|
2001-01-16 00:18:36 +00:00
|
|
|
|
2005-07-09 12:38:53 +00:00
|
|
|
/* Operations on pointers. */
|
2006-03-28 14:34:48 +00:00
|
|
|
#define atomic_set_ptr(p, v) \
|
|
|
|
atomic_set_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_set_acq_ptr(p, v) \
|
|
|
|
atomic_set_acq_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_set_rel_ptr(p, v) \
|
|
|
|
atomic_set_rel_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_clear_ptr(p, v) \
|
|
|
|
atomic_clear_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_clear_acq_ptr(p, v) \
|
|
|
|
atomic_clear_acq_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_clear_rel_ptr(p, v) \
|
|
|
|
atomic_clear_rel_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_add_ptr(p, v) \
|
|
|
|
atomic_add_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_add_acq_ptr(p, v) \
|
|
|
|
atomic_add_acq_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_add_rel_ptr(p, v) \
|
|
|
|
atomic_add_rel_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_subtract_ptr(p, v) \
|
|
|
|
atomic_subtract_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_subtract_acq_ptr(p, v) \
|
|
|
|
atomic_subtract_acq_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_subtract_rel_ptr(p, v) \
|
|
|
|
atomic_subtract_rel_int((volatile u_int *)(p), (u_int)(v))
|
|
|
|
#define atomic_load_acq_ptr(p) \
|
|
|
|
atomic_load_acq_int((volatile u_int *)(p))
|
|
|
|
#define atomic_store_rel_ptr(p, v) \
|
|
|
|
atomic_store_rel_int((volatile u_int *)(p), (v))
|
|
|
|
#define atomic_cmpset_ptr(dst, old, new) \
|
|
|
|
atomic_cmpset_int((volatile u_int *)(dst), (u_int)(old), (u_int)(new))
|
|
|
|
#define atomic_cmpset_acq_ptr(dst, old, new) \
|
2006-12-29 14:28:23 +00:00
|
|
|
atomic_cmpset_acq_int((volatile u_int *)(dst), (u_int)(old), \
|
|
|
|
(u_int)(new))
|
2006-03-28 14:34:48 +00:00
|
|
|
#define atomic_cmpset_rel_ptr(dst, old, new) \
|
2006-12-29 14:28:23 +00:00
|
|
|
atomic_cmpset_rel_int((volatile u_int *)(dst), (u_int)(old), \
|
|
|
|
(u_int)(new))
|
2006-03-28 14:34:48 +00:00
|
|
|
#define atomic_readandclear_ptr(p) \
|
|
|
|
atomic_readandclear_int((volatile u_int *)(p))
|
2000-10-20 07:00:48 +00:00
|
|
|
|
2006-12-29 15:29:49 +00:00
|
|
|
#endif /* !WANT_FUNCTIONS */
|
2006-12-29 14:28:23 +00:00
|
|
|
|
|
|
|
#endif /* !_MACHINE_ATOMIC_H_ */
|