aboutsummaryrefslogtreecommitdiffstats
path: root/include/asm-i386
diff options
context:
space:
mode:
authorGerd Hoffmann <kraxel@suse.de>2006-03-23 05:59:32 -0500
committerLinus Torvalds <torvalds@g5.osdl.org>2006-03-23 10:38:04 -0500
commit9a0b5817ad97bb718ab85322759d19a238712b47 (patch)
tree39bd21eb69c4001b99096d96a76a2e5d37904108 /include/asm-i386
parent4d7d8c82c181711d28c8336108330a9121f5ef07 (diff)
[PATCH] x86: SMP alternatives
Implement SMP alternatives, i.e. switching at runtime between different code versions for UP and SMP. The code can patch both SMP->UP and UP->SMP. The UP->SMP case is useful for CPU hotplug. With CONFIG_CPU_HOTPLUG enabled the code switches to UP at boot time and when the number of CPUs goes down to 1, and switches to SMP when the number of CPUs goes up to 2. Without CONFIG_CPU_HOTPLUG or on non-SMP-capable systems the code is patched once at boot time (if needed) and the tables are released afterwards. The changes in detail: * The current alternatives bits are moved to a separate file, the SMP alternatives code is added there. * The patch adds some new elf sections to the kernel: .smp_altinstructions like .altinstructions, also contains a list of alt_instr structs. .smp_altinstr_replacement like .altinstr_replacement, but also has some space to save original instruction before replaving it. .smp_locks list of pointers to lock prefixes which can be nop'ed out on UP. The first two are used to replace more complex instruction sequences such as spinlocks and semaphores. It would be possible to deal with the lock prefixes with that as well, but by handling them as special case the table sizes become much smaller. * The sections are page-aligned and padded up to page size, so they can be free if they are not needed. * Splitted the code to release init pages to a separate function and use it to release the elf sections if they are unused. Signed-off-by: Gerd Hoffmann <kraxel@suse.de> Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'include/asm-i386')
-rw-r--r--include/asm-i386/alternative.h129
-rw-r--r--include/asm-i386/atomic.h28
-rw-r--r--include/asm-i386/bitops.h7
-rw-r--r--include/asm-i386/cpufeature.h1
-rw-r--r--include/asm-i386/mutex.h6
-rw-r--r--include/asm-i386/rwlock.h56
-rw-r--r--include/asm-i386/semaphore.h8
-rw-r--r--include/asm-i386/spinlock.h21
-rw-r--r--include/asm-i386/system.h62
9 files changed, 194 insertions, 124 deletions
diff --git a/include/asm-i386/alternative.h b/include/asm-i386/alternative.h
new file mode 100644
index 000000000000..e201decea0c9
--- /dev/null
+++ b/include/asm-i386/alternative.h
@@ -0,0 +1,129 @@
1#ifndef _I386_ALTERNATIVE_H
2#define _I386_ALTERNATIVE_H
3
4#ifdef __KERNEL__
5
6struct alt_instr {
7 u8 *instr; /* original instruction */
8 u8 *replacement;
9 u8 cpuid; /* cpuid bit set for replacement */
10 u8 instrlen; /* length of original instruction */
11 u8 replacementlen; /* length of new instruction, <= instrlen */
12 u8 pad;
13};
14
15extern void apply_alternatives(struct alt_instr *start, struct alt_instr *end);
16
17struct module;
18extern void alternatives_smp_module_add(struct module *mod, char *name,
19 void *locks, void *locks_end,
20 void *text, void *text_end);
21extern void alternatives_smp_module_del(struct module *mod);
22extern void alternatives_smp_switch(int smp);
23
24#endif
25
26/*
27 * Alternative instructions for different CPU types or capabilities.
28 *
29 * This allows to use optimized instructions even on generic binary
30 * kernels.
31 *
32 * length of oldinstr must be longer or equal the length of newinstr
33 * It can be padded with nops as needed.
34 *
35 * For non barrier like inlines please define new variants
36 * without volatile and memory clobber.
37 */
38#define alternative(oldinstr, newinstr, feature) \
39 asm volatile ("661:\n\t" oldinstr "\n662:\n" \
40 ".section .altinstructions,\"a\"\n" \
41 " .align 4\n" \
42 " .long 661b\n" /* label */ \
43 " .long 663f\n" /* new instruction */ \
44 " .byte %c0\n" /* feature bit */ \
45 " .byte 662b-661b\n" /* sourcelen */ \
46 " .byte 664f-663f\n" /* replacementlen */ \
47 ".previous\n" \
48 ".section .altinstr_replacement,\"ax\"\n" \
49 "663:\n\t" newinstr "\n664:\n" /* replacement */\
50 ".previous" :: "i" (feature) : "memory")
51
52/*
53 * Alternative inline assembly with input.
54 *
55 * Pecularities:
56 * No memory clobber here.
57 * Argument numbers start with 1.
58 * Best is to use constraints that are fixed size (like (%1) ... "r")
59 * If you use variable sized constraints like "m" or "g" in the
60 * replacement maake sure to pad to the worst case length.
61 */
62#define alternative_input(oldinstr, newinstr, feature, input...) \
63 asm volatile ("661:\n\t" oldinstr "\n662:\n" \
64 ".section .altinstructions,\"a\"\n" \
65 " .align 4\n" \
66 " .long 661b\n" /* label */ \
67 " .long 663f\n" /* new instruction */ \
68 " .byte %c0\n" /* feature bit */ \
69 " .byte 662b-661b\n" /* sourcelen */ \
70 " .byte 664f-663f\n" /* replacementlen */ \
71 ".previous\n" \
72 ".section .altinstr_replacement,\"ax\"\n" \
73 "663:\n\t" newinstr "\n664:\n" /* replacement */\
74 ".previous" :: "i" (feature), ##input)
75
76/*
77 * Alternative inline assembly for SMP.
78 *
79 * alternative_smp() takes two versions (SMP first, UP second) and is
80 * for more complex stuff such as spinlocks.
81 *
82 * The LOCK_PREFIX macro defined here replaces the LOCK and
83 * LOCK_PREFIX macros used everywhere in the source tree.
84 *
85 * SMP alternatives use the same data structures as the other
86 * alternatives and the X86_FEATURE_UP flag to indicate the case of a
87 * UP system running a SMP kernel. The existing apply_alternatives()
88 * works fine for patching a SMP kernel for UP.
89 *
90 * The SMP alternative tables can be kept after boot and contain both
91 * UP and SMP versions of the instructions to allow switching back to
92 * SMP at runtime, when hotplugging in a new CPU, which is especially
93 * useful in virtualized environments.
94 *
95 * The very common lock prefix is handled as special case in a
96 * separate table which is a pure address list without replacement ptr
97 * and size information. That keeps the table sizes small.
98 */
99
100#ifdef CONFIG_SMP
101#define alternative_smp(smpinstr, upinstr, args...) \
102 asm volatile ("661:\n\t" smpinstr "\n662:\n" \
103 ".section .smp_altinstructions,\"a\"\n" \
104 " .align 4\n" \
105 " .long 661b\n" /* label */ \
106 " .long 663f\n" /* new instruction */ \
107 " .byte 0x68\n" /* X86_FEATURE_UP */ \
108 " .byte 662b-661b\n" /* sourcelen */ \
109 " .byte 664f-663f\n" /* replacementlen */ \
110 ".previous\n" \
111 ".section .smp_altinstr_replacement,\"awx\"\n" \
112 "663:\n\t" upinstr "\n" /* replacement */ \
113 "664:\n\t.fill 662b-661b,1,0x42\n" /* space for original */ \
114 ".previous" : args)
115
116#define LOCK_PREFIX \
117 ".section .smp_locks,\"a\"\n" \
118 " .align 4\n" \
119 " .long 661f\n" /* address */ \
120 ".previous\n" \
121 "661:\n\tlock; "
122
123#else /* ! CONFIG_SMP */
124#define alternative_smp(smpinstr, upinstr, args...) \
125 asm volatile (upinstr : args)
126#define LOCK_PREFIX ""
127#endif
128
129#endif /* _I386_ALTERNATIVE_H */
diff --git a/include/asm-i386/atomic.h b/include/asm-i386/atomic.h
index de649d3aa2d4..78b0032d1f29 100644
--- a/include/asm-i386/atomic.h
+++ b/include/asm-i386/atomic.h
@@ -10,12 +10,6 @@
10 * resource counting etc.. 10 * resource counting etc..
11 */ 11 */
12 12
13#ifdef CONFIG_SMP
14#define LOCK "lock ; "
15#else
16#define LOCK ""
17#endif
18
19/* 13/*
20 * Make sure gcc doesn't try to be clever and move things around 14 * Make sure gcc doesn't try to be clever and move things around
21 * on us. We need to use _exactly_ the address the user gave us, 15 * on us. We need to use _exactly_ the address the user gave us,
@@ -52,7 +46,7 @@ typedef struct { volatile int counter; } atomic_t;
52static __inline__ void atomic_add(int i, atomic_t *v) 46static __inline__ void atomic_add(int i, atomic_t *v)
53{ 47{
54 __asm__ __volatile__( 48 __asm__ __volatile__(
55 LOCK "addl %1,%0" 49 LOCK_PREFIX "addl %1,%0"
56 :"=m" (v->counter) 50 :"=m" (v->counter)
57 :"ir" (i), "m" (v->counter)); 51 :"ir" (i), "m" (v->counter));
58} 52}
@@ -67,7 +61,7 @@ static __inline__ void atomic_add(int i, atomic_t *v)
67static __inline__ void atomic_sub(int i, atomic_t *v) 61static __inline__ void atomic_sub(int i, atomic_t *v)
68{ 62{
69 __asm__ __volatile__( 63 __asm__ __volatile__(
70 LOCK "subl %1,%0" 64 LOCK_PREFIX "subl %1,%0"
71 :"=m" (v->counter) 65 :"=m" (v->counter)
72 :"ir" (i), "m" (v->counter)); 66 :"ir" (i), "m" (v->counter));
73} 67}
@@ -86,7 +80,7 @@ static __inline__ int atomic_sub_and_test(int i, atomic_t *v)
86 unsigned char c; 80 unsigned char c;
87 81
88 __asm__ __volatile__( 82 __asm__ __volatile__(
89 LOCK "subl %2,%0; sete %1" 83 LOCK_PREFIX "subl %2,%0; sete %1"
90 :"=m" (v->counter), "=qm" (c) 84 :"=m" (v->counter), "=qm" (c)
91 :"ir" (i), "m" (v->counter) : "memory"); 85 :"ir" (i), "m" (v->counter) : "memory");
92 return c; 86 return c;
@@ -101,7 +95,7 @@ static __inline__ int atomic_sub_and_test(int i, atomic_t *v)
101static __inline__ void atomic_inc(atomic_t *v) 95static __inline__ void atomic_inc(atomic_t *v)
102{ 96{
103 __asm__ __volatile__( 97 __asm__ __volatile__(
104 LOCK "incl %0" 98 LOCK_PREFIX "incl %0"
105 :"=m" (v->counter) 99 :"=m" (v->counter)
106 :"m" (v->counter)); 100 :"m" (v->counter));
107} 101}
@@ -115,7 +109,7 @@ static __inline__ void atomic_inc(atomic_t *v)
115static __inline__ void atomic_dec(atomic_t *v) 109static __inline__ void atomic_dec(atomic_t *v)
116{ 110{
117 __asm__ __volatile__( 111 __asm__ __volatile__(
118 LOCK "decl %0" 112 LOCK_PREFIX "decl %0"
119 :"=m" (v->counter) 113 :"=m" (v->counter)
120 :"m" (v->counter)); 114 :"m" (v->counter));
121} 115}
@@ -133,7 +127,7 @@ static __inline__ int atomic_dec_and_test(atomic_t *v)
133 unsigned char c; 127 unsigned char c;
134 128
135 __asm__ __volatile__( 129 __asm__ __volatile__(
136 LOCK "decl %0; sete %1" 130 LOCK_PREFIX "decl %0; sete %1"
137 :"=m" (v->counter), "=qm" (c) 131 :"=m" (v->counter), "=qm" (c)
138 :"m" (v->counter) : "memory"); 132 :"m" (v->counter) : "memory");
139 return c != 0; 133 return c != 0;
@@ -152,7 +146,7 @@ static __inline__ int atomic_inc_and_test(atomic_t *v)
152 unsigned char c; 146 unsigned char c;
153 147
154 __asm__ __volatile__( 148 __asm__ __volatile__(
155 LOCK "incl %0; sete %1" 149 LOCK_PREFIX "incl %0; sete %1"
156 :"=m" (v->counter), "=qm" (c) 150 :"=m" (v->counter), "=qm" (c)
157 :"m" (v->counter) : "memory"); 151 :"m" (v->counter) : "memory");
158 return c != 0; 152 return c != 0;
@@ -172,7 +166,7 @@ static __inline__ int atomic_add_negative(int i, atomic_t *v)
172 unsigned char c; 166 unsigned char c;
173 167
174 __asm__ __volatile__( 168 __asm__ __volatile__(
175 LOCK "addl %2,%0; sets %1" 169 LOCK_PREFIX "addl %2,%0; sets %1"
176 :"=m" (v->counter), "=qm" (c) 170 :"=m" (v->counter), "=qm" (c)
177 :"ir" (i), "m" (v->counter) : "memory"); 171 :"ir" (i), "m" (v->counter) : "memory");
178 return c; 172 return c;
@@ -195,7 +189,7 @@ static __inline__ int atomic_add_return(int i, atomic_t *v)
195 /* Modern 486+ processor */ 189 /* Modern 486+ processor */
196 __i = i; 190 __i = i;
197 __asm__ __volatile__( 191 __asm__ __volatile__(
198 LOCK "xaddl %0, %1;" 192 LOCK_PREFIX "xaddl %0, %1;"
199 :"=r"(i) 193 :"=r"(i)
200 :"m"(v->counter), "0"(i)); 194 :"m"(v->counter), "0"(i));
201 return i + __i; 195 return i + __i;
@@ -242,11 +236,11 @@ static __inline__ int atomic_sub_return(int i, atomic_t *v)
242 236
243/* These are x86-specific, used by some header files */ 237/* These are x86-specific, used by some header files */
244#define atomic_clear_mask(mask, addr) \ 238#define atomic_clear_mask(mask, addr) \
245__asm__ __volatile__(LOCK "andl %0,%1" \ 239__asm__ __volatile__(LOCK_PREFIX "andl %0,%1" \
246: : "r" (~(mask)),"m" (*addr) : "memory") 240: : "r" (~(mask)),"m" (*addr) : "memory")
247 241
248#define atomic_set_mask(mask, addr) \ 242#define atomic_set_mask(mask, addr) \
249__asm__ __volatile__(LOCK "orl %0,%1" \ 243__asm__ __volatile__(LOCK_PREFIX "orl %0,%1" \
250: : "r" (mask),"m" (*(addr)) : "memory") 244: : "r" (mask),"m" (*(addr)) : "memory")
251 245
252/* Atomic operations are already serializing on x86 */ 246/* Atomic operations are already serializing on x86 */
diff --git a/include/asm-i386/bitops.h b/include/asm-i386/bitops.h
index 88e6ca248cd7..7d20b95edb3b 100644
--- a/include/asm-i386/bitops.h
+++ b/include/asm-i386/bitops.h
@@ -7,6 +7,7 @@
7 7
8#include <linux/config.h> 8#include <linux/config.h>
9#include <linux/compiler.h> 9#include <linux/compiler.h>
10#include <asm/alternative.h>
10 11
11/* 12/*
12 * These have to be done with inline assembly: that way the bit-setting 13 * These have to be done with inline assembly: that way the bit-setting
@@ -16,12 +17,6 @@
16 * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). 17 * bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1).
17 */ 18 */
18 19
19#ifdef CONFIG_SMP
20#define LOCK_PREFIX "lock ; "
21#else
22#define LOCK_PREFIX ""
23#endif
24
25#define ADDR (*(volatile long *) addr) 20#define ADDR (*(volatile long *) addr)
26 21
27/** 22/**
diff --git a/include/asm-i386/cpufeature.h b/include/asm-i386/cpufeature.h
index c4ec2a4d8fdf..5c0b5876b931 100644
--- a/include/asm-i386/cpufeature.h
+++ b/include/asm-i386/cpufeature.h
@@ -70,6 +70,7 @@
70#define X86_FEATURE_P3 (3*32+ 6) /* P3 */ 70#define X86_FEATURE_P3 (3*32+ 6) /* P3 */
71#define X86_FEATURE_P4 (3*32+ 7) /* P4 */ 71#define X86_FEATURE_P4 (3*32+ 7) /* P4 */
72#define X86_FEATURE_CONSTANT_TSC (3*32+ 8) /* TSC ticks at a constant rate */ 72#define X86_FEATURE_CONSTANT_TSC (3*32+ 8) /* TSC ticks at a constant rate */
73#define X86_FEATURE_UP (3*32+ 9) /* smp kernel running on up */
73 74
74/* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */ 75/* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */
75#define X86_FEATURE_XMM3 (4*32+ 0) /* Streaming SIMD Extensions-3 */ 76#define X86_FEATURE_XMM3 (4*32+ 0) /* Streaming SIMD Extensions-3 */
diff --git a/include/asm-i386/mutex.h b/include/asm-i386/mutex.h
index 9b2199e829f3..05a538531229 100644
--- a/include/asm-i386/mutex.h
+++ b/include/asm-i386/mutex.h
@@ -9,6 +9,8 @@
9#ifndef _ASM_MUTEX_H 9#ifndef _ASM_MUTEX_H
10#define _ASM_MUTEX_H 10#define _ASM_MUTEX_H
11 11
12#include "asm/alternative.h"
13
12/** 14/**
13 * __mutex_fastpath_lock - try to take the lock by moving the count 15 * __mutex_fastpath_lock - try to take the lock by moving the count
14 * from 1 to a 0 value 16 * from 1 to a 0 value
@@ -27,7 +29,7 @@ do { \
27 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \ 29 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \
28 \ 30 \
29 __asm__ __volatile__( \ 31 __asm__ __volatile__( \
30 LOCK " decl (%%eax) \n" \ 32 LOCK_PREFIX " decl (%%eax) \n" \
31 " js 2f \n" \ 33 " js 2f \n" \
32 "1: \n" \ 34 "1: \n" \
33 \ 35 \
@@ -83,7 +85,7 @@ do { \
83 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \ 85 typecheck_fn(fastcall void (*)(atomic_t *), fail_fn); \
84 \ 86 \
85 __asm__ __volatile__( \ 87 __asm__ __volatile__( \
86 LOCK " incl (%%eax) \n" \ 88 LOCK_PREFIX " incl (%%eax) \n" \
87 " jle 2f \n" \ 89 " jle 2f \n" \
88 "1: \n" \ 90 "1: \n" \
89 \ 91 \
diff --git a/include/asm-i386/rwlock.h b/include/asm-i386/rwlock.h
index b57cc7afdf7e..94f00195d543 100644
--- a/include/asm-i386/rwlock.h
+++ b/include/asm-i386/rwlock.h
@@ -21,21 +21,23 @@
21#define RW_LOCK_BIAS_STR "0x01000000" 21#define RW_LOCK_BIAS_STR "0x01000000"
22 22
23#define __build_read_lock_ptr(rw, helper) \ 23#define __build_read_lock_ptr(rw, helper) \
24 asm volatile(LOCK "subl $1,(%0)\n\t" \ 24 alternative_smp("lock; subl $1,(%0)\n\t" \
25 "jns 1f\n" \ 25 "jns 1f\n" \
26 "call " helper "\n\t" \ 26 "call " helper "\n\t" \
27 "1:\n" \ 27 "1:\n", \
28 ::"a" (rw) : "memory") 28 "subl $1,(%0)\n\t", \
29 :"a" (rw) : "memory")
29 30
30#define __build_read_lock_const(rw, helper) \ 31#define __build_read_lock_const(rw, helper) \
31 asm volatile(LOCK "subl $1,%0\n\t" \ 32 alternative_smp("lock; subl $1,%0\n\t" \
32 "jns 1f\n" \ 33 "jns 1f\n" \
33 "pushl %%eax\n\t" \ 34 "pushl %%eax\n\t" \
34 "leal %0,%%eax\n\t" \ 35 "leal %0,%%eax\n\t" \
35 "call " helper "\n\t" \ 36 "call " helper "\n\t" \
36 "popl %%eax\n\t" \ 37 "popl %%eax\n\t" \
37 "1:\n" \ 38 "1:\n", \
38 :"=m" (*(volatile int *)rw) : : "memory") 39 "subl $1,%0\n\t", \
40 "=m" (*(volatile int *)rw) : : "memory")
39 41
40#define __build_read_lock(rw, helper) do { \ 42#define __build_read_lock(rw, helper) do { \
41 if (__builtin_constant_p(rw)) \ 43 if (__builtin_constant_p(rw)) \
@@ -45,21 +47,23 @@
45 } while (0) 47 } while (0)
46 48
47#define __build_write_lock_ptr(rw, helper) \ 49#define __build_write_lock_ptr(rw, helper) \
48 asm volatile(LOCK "subl $" RW_LOCK_BIAS_STR ",(%0)\n\t" \ 50 alternative_smp("lock; subl $" RW_LOCK_BIAS_STR ",(%0)\n\t" \
49 "jz 1f\n" \ 51 "jz 1f\n" \
50 "call " helper "\n\t" \ 52 "call " helper "\n\t" \
51 "1:\n" \ 53 "1:\n", \
52 ::"a" (rw) : "memory") 54 "subl $" RW_LOCK_BIAS_STR ",(%0)\n\t", \
55 :"a" (rw) : "memory")
53 56
54#define __build_write_lock_const(rw, helper) \ 57#define __build_write_lock_const(rw, helper) \
55 asm volatile(LOCK "subl $" RW_LOCK_BIAS_STR ",%0\n\t" \ 58 alternative_smp("lock; subl $" RW_LOCK_BIAS_STR ",%0\n\t" \
56 "jz 1f\n" \ 59 "jz 1f\n" \
57 "pushl %%eax\n\t" \ 60 "pushl %%eax\n\t" \
58 "leal %0,%%eax\n\t" \ 61 "leal %0,%%eax\n\t" \
59 "call " helper "\n\t" \ 62 "call " helper "\n\t" \
60 "popl %%eax\n\t" \ 63 "popl %%eax\n\t" \
61 "1:\n" \ 64 "1:\n", \
62 :"=m" (*(volatile int *)rw) : : "memory") 65 "subl $" RW_LOCK_BIAS_STR ",%0\n\t", \
66 "=m" (*(volatile int *)rw) : : "memory")
63 67
64#define __build_write_lock(rw, helper) do { \ 68#define __build_write_lock(rw, helper) do { \
65 if (__builtin_constant_p(rw)) \ 69 if (__builtin_constant_p(rw)) \
diff --git a/include/asm-i386/semaphore.h b/include/asm-i386/semaphore.h
index 6a42b2142fd6..f7a0f310c524 100644
--- a/include/asm-i386/semaphore.h
+++ b/include/asm-i386/semaphore.h
@@ -99,7 +99,7 @@ static inline void down(struct semaphore * sem)
99 might_sleep(); 99 might_sleep();
100 __asm__ __volatile__( 100 __asm__ __volatile__(
101 "# atomic down operation\n\t" 101 "# atomic down operation\n\t"
102 LOCK "decl %0\n\t" /* --sem->count */ 102 LOCK_PREFIX "decl %0\n\t" /* --sem->count */
103 "js 2f\n" 103 "js 2f\n"
104 "1:\n" 104 "1:\n"
105 LOCK_SECTION_START("") 105 LOCK_SECTION_START("")
@@ -123,7 +123,7 @@ static inline int down_interruptible(struct semaphore * sem)
123 might_sleep(); 123 might_sleep();
124 __asm__ __volatile__( 124 __asm__ __volatile__(
125 "# atomic interruptible down operation\n\t" 125 "# atomic interruptible down operation\n\t"
126 LOCK "decl %1\n\t" /* --sem->count */ 126 LOCK_PREFIX "decl %1\n\t" /* --sem->count */
127 "js 2f\n\t" 127 "js 2f\n\t"
128 "xorl %0,%0\n" 128 "xorl %0,%0\n"
129 "1:\n" 129 "1:\n"
@@ -148,7 +148,7 @@ static inline int down_trylock(struct semaphore * sem)
148 148
149 __asm__ __volatile__( 149 __asm__ __volatile__(
150 "# atomic interruptible down operation\n\t" 150 "# atomic interruptible down operation\n\t"
151 LOCK "decl %1\n\t" /* --sem->count */ 151 LOCK_PREFIX "decl %1\n\t" /* --sem->count */
152 "js 2f\n\t" 152 "js 2f\n\t"
153 "xorl %0,%0\n" 153 "xorl %0,%0\n"
154 "1:\n" 154 "1:\n"
@@ -173,7 +173,7 @@ static inline void up(struct semaphore * sem)
173{ 173{
174 __asm__ __volatile__( 174 __asm__ __volatile__(
175 "# atomic up operation\n\t" 175 "# atomic up operation\n\t"
176 LOCK "incl %0\n\t" /* ++sem->count */ 176 LOCK_PREFIX "incl %0\n\t" /* ++sem->count */
177 "jle 2f\n" 177 "jle 2f\n"
178 "1:\n" 178 "1:\n"
179 LOCK_SECTION_START("") 179 LOCK_SECTION_START("")
diff --git a/include/asm-i386/spinlock.h b/include/asm-i386/spinlock.h
index 23604350cdf4..a1b8a8a30e21 100644
--- a/include/asm-i386/spinlock.h
+++ b/include/asm-i386/spinlock.h
@@ -48,18 +48,23 @@
48 "jmp 1b\n" \ 48 "jmp 1b\n" \
49 "4:\n\t" 49 "4:\n\t"
50 50
51#define __raw_spin_lock_string_up \
52 "\n\tdecb %0"
53
51static inline void __raw_spin_lock(raw_spinlock_t *lock) 54static inline void __raw_spin_lock(raw_spinlock_t *lock)
52{ 55{
53 __asm__ __volatile__( 56 alternative_smp(
54 __raw_spin_lock_string 57 __raw_spin_lock_string,
55 :"=m" (lock->slock) : : "memory"); 58 __raw_spin_lock_string_up,
59 "=m" (lock->slock) : : "memory");
56} 60}
57 61
58static inline void __raw_spin_lock_flags(raw_spinlock_t *lock, unsigned long flags) 62static inline void __raw_spin_lock_flags(raw_spinlock_t *lock, unsigned long flags)
59{ 63{
60 __asm__ __volatile__( 64 alternative_smp(
61 __raw_spin_lock_string_flags 65 __raw_spin_lock_string_flags,
62 :"=m" (lock->slock) : "r" (flags) : "memory"); 66 __raw_spin_lock_string_up,
67 "=m" (lock->slock) : "r" (flags) : "memory");
63} 68}
64 69
65static inline int __raw_spin_trylock(raw_spinlock_t *lock) 70static inline int __raw_spin_trylock(raw_spinlock_t *lock)
@@ -178,12 +183,12 @@ static inline int __raw_write_trylock(raw_rwlock_t *lock)
178 183
179static inline void __raw_read_unlock(raw_rwlock_t *rw) 184static inline void __raw_read_unlock(raw_rwlock_t *rw)
180{ 185{
181 asm volatile("lock ; incl %0" :"=m" (rw->lock) : : "memory"); 186 asm volatile(LOCK_PREFIX "incl %0" :"=m" (rw->lock) : : "memory");
182} 187}
183 188
184static inline void __raw_write_unlock(raw_rwlock_t *rw) 189static inline void __raw_write_unlock(raw_rwlock_t *rw)
185{ 190{
186 asm volatile("lock ; addl $" RW_LOCK_BIAS_STR ", %0" 191 asm volatile(LOCK_PREFIX "addl $" RW_LOCK_BIAS_STR ", %0"
187 : "=m" (rw->lock) : : "memory"); 192 : "=m" (rw->lock) : : "memory");
188} 193}
189 194
diff --git a/include/asm-i386/system.h b/include/asm-i386/system.h
index 399145a247f2..d0d8d7448d88 100644
--- a/include/asm-i386/system.h
+++ b/include/asm-i386/system.h
@@ -352,67 +352,6 @@ static inline unsigned long long __cmpxchg64(volatile void *ptr, unsigned long l
352 352
353#endif 353#endif
354 354
355#ifdef __KERNEL__
356struct alt_instr {
357 __u8 *instr; /* original instruction */
358 __u8 *replacement;
359 __u8 cpuid; /* cpuid bit set for replacement */
360 __u8 instrlen; /* length of original instruction */
361 __u8 replacementlen; /* length of new instruction, <= instrlen */
362 __u8 pad;
363};
364#endif
365
366/*
367 * Alternative instructions for different CPU types or capabilities.
368 *
369 * This allows to use optimized instructions even on generic binary
370 * kernels.
371 *
372 * length of oldinstr must be longer or equal the length of newinstr
373 * It can be padded with nops as needed.
374 *
375 * For non barrier like inlines please define new variants
376 * without volatile and memory clobber.
377 */
378#define alternative(oldinstr, newinstr, feature) \
379 asm volatile ("661:\n\t" oldinstr "\n662:\n" \
380 ".section .altinstructions,\"a\"\n" \
381 " .align 4\n" \
382 " .long 661b\n" /* label */ \
383 " .long 663f\n" /* new instruction */ \
384 " .byte %c0\n" /* feature bit */ \
385 " .byte 662b-661b\n" /* sourcelen */ \
386 " .byte 664f-663f\n" /* replacementlen */ \
387 ".previous\n" \
388 ".section .altinstr_replacement,\"ax\"\n" \
389 "663:\n\t" newinstr "\n664:\n" /* replacement */ \
390 ".previous" :: "i" (feature) : "memory")
391
392/*
393 * Alternative inline assembly with input.
394 *
395 * Pecularities:
396 * No memory clobber here.
397 * Argument numbers start with 1.
398 * Best is to use constraints that are fixed size (like (%1) ... "r")
399 * If you use variable sized constraints like "m" or "g" in the
400 * replacement maake sure to pad to the worst case length.
401 */
402#define alternative_input(oldinstr, newinstr, feature, input...) \
403 asm volatile ("661:\n\t" oldinstr "\n662:\n" \
404 ".section .altinstructions,\"a\"\n" \
405 " .align 4\n" \
406 " .long 661b\n" /* label */ \
407 " .long 663f\n" /* new instruction */ \
408 " .byte %c0\n" /* feature bit */ \
409 " .byte 662b-661b\n" /* sourcelen */ \
410 " .byte 664f-663f\n" /* replacementlen */ \
411 ".previous\n" \
412 ".section .altinstr_replacement,\"ax\"\n" \
413 "663:\n\t" newinstr "\n664:\n" /* replacement */ \
414 ".previous" :: "i" (feature), ##input)
415
416/* 355/*
417 * Force strict CPU ordering. 356 * Force strict CPU ordering.
418 * And yes, this is required on UP too when we're talking 357 * And yes, this is required on UP too when we're talking
@@ -558,5 +497,6 @@ static inline void sched_cacheflush(void)
558} 497}
559 498
560extern unsigned long arch_align_stack(unsigned long sp); 499extern unsigned long arch_align_stack(unsigned long sp);
500extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
561 501
562#endif 502#endif
generated by cgit v1.2.2 (git 2.25.0) at 2024-09-21 13:01:20 -0400