aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/include/asm
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/include/asm')
-rw-r--r--arch/x86/include/asm/local.h37
-rw-r--r--arch/x86/include/asm/percpu.h119
-rw-r--r--arch/x86/include/asm/system.h8
3 files changed, 91 insertions, 73 deletions
diff --git a/arch/x86/include/asm/local.h b/arch/x86/include/asm/local.h
index 47b9b6f19057..2e9972468a5d 100644
--- a/arch/x86/include/asm/local.h
+++ b/arch/x86/include/asm/local.h
@@ -195,41 +195,4 @@ static inline long local_sub_return(long i, local_t *l)
195#define __local_add(i, l) local_add((i), (l)) 195#define __local_add(i, l) local_add((i), (l))
196#define __local_sub(i, l) local_sub((i), (l)) 196#define __local_sub(i, l) local_sub((i), (l))
197 197
198/* Use these for per-cpu local_t variables: on some archs they are
199 * much more efficient than these naive implementations. Note they take
200 * a variable, not an address.
201 *
202 * X86_64: This could be done better if we moved the per cpu data directly
203 * after GS.
204 */
205
206/* Need to disable preemption for the cpu local counters otherwise we could
207 still access a variable of a previous CPU in a non atomic way. */
208#define cpu_local_wrap_v(l) \
209({ \
210 local_t res__; \
211 preempt_disable(); \
212 res__ = (l); \
213 preempt_enable(); \
214 res__; \
215})
216#define cpu_local_wrap(l) \
217({ \
218 preempt_disable(); \
219 (l); \
220 preempt_enable(); \
221}) \
222
223#define cpu_local_read(l) cpu_local_wrap_v(local_read(&__get_cpu_var((l))))
224#define cpu_local_set(l, i) cpu_local_wrap(local_set(&__get_cpu_var((l)), (i)))
225#define cpu_local_inc(l) cpu_local_wrap(local_inc(&__get_cpu_var((l))))
226#define cpu_local_dec(l) cpu_local_wrap(local_dec(&__get_cpu_var((l))))
227#define cpu_local_add(i, l) cpu_local_wrap(local_add((i), &__get_cpu_var((l))))
228#define cpu_local_sub(i, l) cpu_local_wrap(local_sub((i), &__get_cpu_var((l))))
229
230#define __cpu_local_inc(l) cpu_local_inc((l))
231#define __cpu_local_dec(l) cpu_local_dec((l))
232#define __cpu_local_add(i, l) cpu_local_add((i), (l))
233#define __cpu_local_sub(i, l) cpu_local_sub((i), (l))
234
235#endif /* _ASM_X86_LOCAL_H */ 198#endif /* _ASM_X86_LOCAL_H */
diff --git a/arch/x86/include/asm/percpu.h b/arch/x86/include/asm/percpu.h
index 0c44196b78ac..66a272dfd8b8 100644
--- a/arch/x86/include/asm/percpu.h
+++ b/arch/x86/include/asm/percpu.h
@@ -25,19 +25,18 @@
25 */ 25 */
26#ifdef CONFIG_SMP 26#ifdef CONFIG_SMP
27#define PER_CPU(var, reg) \ 27#define PER_CPU(var, reg) \
28 __percpu_mov_op %__percpu_seg:per_cpu__this_cpu_off, reg; \ 28 __percpu_mov_op %__percpu_seg:this_cpu_off, reg; \
29 lea per_cpu__##var(reg), reg 29 lea var(reg), reg
30#define PER_CPU_VAR(var) %__percpu_seg:per_cpu__##var 30#define PER_CPU_VAR(var) %__percpu_seg:var
31#else /* ! SMP */ 31#else /* ! SMP */
32#define PER_CPU(var, reg) \ 32#define PER_CPU(var, reg) __percpu_mov_op $var, reg
33 __percpu_mov_op $per_cpu__##var, reg 33#define PER_CPU_VAR(var) var
34#define PER_CPU_VAR(var) per_cpu__##var
35#endif /* SMP */ 34#endif /* SMP */
36 35
37#ifdef CONFIG_X86_64_SMP 36#ifdef CONFIG_X86_64_SMP
38#define INIT_PER_CPU_VAR(var) init_per_cpu__##var 37#define INIT_PER_CPU_VAR(var) init_per_cpu__##var
39#else 38#else
40#define INIT_PER_CPU_VAR(var) per_cpu__##var 39#define INIT_PER_CPU_VAR(var) var
41#endif 40#endif
42 41
43#else /* ...!ASSEMBLY */ 42#else /* ...!ASSEMBLY */
@@ -60,12 +59,12 @@
60 * There also must be an entry in vmlinux_64.lds.S 59 * There also must be an entry in vmlinux_64.lds.S
61 */ 60 */
62#define DECLARE_INIT_PER_CPU(var) \ 61#define DECLARE_INIT_PER_CPU(var) \
63 extern typeof(per_cpu_var(var)) init_per_cpu_var(var) 62 extern typeof(var) init_per_cpu_var(var)
64 63
65#ifdef CONFIG_X86_64_SMP 64#ifdef CONFIG_X86_64_SMP
66#define init_per_cpu_var(var) init_per_cpu__##var 65#define init_per_cpu_var(var) init_per_cpu__##var
67#else 66#else
68#define init_per_cpu_var(var) per_cpu_var(var) 67#define init_per_cpu_var(var) var
69#endif 68#endif
70 69
71/* For arch-specific code, we can use direct single-insn ops (they 70/* For arch-specific code, we can use direct single-insn ops (they
@@ -104,6 +103,64 @@ do { \
104 } \ 103 } \
105} while (0) 104} while (0)
106 105
106/*
107 * Generate a percpu add to memory instruction and optimize code
108 * if a one is added or subtracted.
109 */
110#define percpu_add_op(var, val) \
111do { \
112 typedef typeof(var) pao_T__; \
113 const int pao_ID__ = (__builtin_constant_p(val) && \
114 ((val) == 1 || (val) == -1)) ? (val) : 0; \
115 if (0) { \
116 pao_T__ pao_tmp__; \
117 pao_tmp__ = (val); \
118 } \
119 switch (sizeof(var)) { \
120 case 1: \
121 if (pao_ID__ == 1) \
122 asm("incb "__percpu_arg(0) : "+m" (var)); \
123 else if (pao_ID__ == -1) \
124 asm("decb "__percpu_arg(0) : "+m" (var)); \
125 else \
126 asm("addb %1, "__percpu_arg(0) \
127 : "+m" (var) \
128 : "qi" ((pao_T__)(val))); \
129 break; \
130 case 2: \
131 if (pao_ID__ == 1) \
132 asm("incw "__percpu_arg(0) : "+m" (var)); \
133 else if (pao_ID__ == -1) \
134 asm("decw "__percpu_arg(0) : "+m" (var)); \
135 else \
136 asm("addw %1, "__percpu_arg(0) \
137 : "+m" (var) \
138 : "ri" ((pao_T__)(val))); \
139 break; \
140 case 4: \
141 if (pao_ID__ == 1) \
142 asm("incl "__percpu_arg(0) : "+m" (var)); \
143 else if (pao_ID__ == -1) \
144 asm("decl "__percpu_arg(0) : "+m" (var)); \
145 else \
146 asm("addl %1, "__percpu_arg(0) \
147 : "+m" (var) \
148 : "ri" ((pao_T__)(val))); \
149 break; \
150 case 8: \
151 if (pao_ID__ == 1) \
152 asm("incq "__percpu_arg(0) : "+m" (var)); \
153 else if (pao_ID__ == -1) \
154 asm("decq "__percpu_arg(0) : "+m" (var)); \
155 else \
156 asm("addq %1, "__percpu_arg(0) \
157 : "+m" (var) \
158 : "re" ((pao_T__)(val))); \
159 break; \
160 default: __bad_percpu_size(); \
161 } \
162} while (0)
163
107#define percpu_from_op(op, var, constraint) \ 164#define percpu_from_op(op, var, constraint) \
108({ \ 165({ \
109 typeof(var) pfo_ret__; \ 166 typeof(var) pfo_ret__; \
@@ -142,16 +199,14 @@ do { \
142 * per-thread variables implemented as per-cpu variables and thus 199 * per-thread variables implemented as per-cpu variables and thus
143 * stable for the duration of the respective task. 200 * stable for the duration of the respective task.
144 */ 201 */
145#define percpu_read(var) percpu_from_op("mov", per_cpu__##var, \ 202#define percpu_read(var) percpu_from_op("mov", var, "m" (var))
146 "m" (per_cpu__##var)) 203#define percpu_read_stable(var) percpu_from_op("mov", var, "p" (&(var)))
147#define percpu_read_stable(var) percpu_from_op("mov", per_cpu__##var, \ 204#define percpu_write(var, val) percpu_to_op("mov", var, val)
148 "p" (&per_cpu__##var)) 205#define percpu_add(var, val) percpu_add_op(var, val)
149#define percpu_write(var, val) percpu_to_op("mov", per_cpu__##var, val) 206#define percpu_sub(var, val) percpu_add_op(var, -(val))
150#define percpu_add(var, val) percpu_to_op("add", per_cpu__##var, val) 207#define percpu_and(var, val) percpu_to_op("and", var, val)
151#define percpu_sub(var, val) percpu_to_op("sub", per_cpu__##var, val) 208#define percpu_or(var, val) percpu_to_op("or", var, val)
152#define percpu_and(var, val) percpu_to_op("and", per_cpu__##var, val) 209#define percpu_xor(var, val) percpu_to_op("xor", var, val)
153#define percpu_or(var, val) percpu_to_op("or", per_cpu__##var, val)
154#define percpu_xor(var, val) percpu_to_op("xor", per_cpu__##var, val)
155 210
156#define __this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 211#define __this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
157#define __this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 212#define __this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
@@ -160,9 +215,9 @@ do { \
160#define __this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val) 215#define __this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
161#define __this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val) 216#define __this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
162#define __this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val) 217#define __this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
163#define __this_cpu_add_1(pcp, val) percpu_to_op("add", (pcp), val) 218#define __this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
164#define __this_cpu_add_2(pcp, val) percpu_to_op("add", (pcp), val) 219#define __this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
165#define __this_cpu_add_4(pcp, val) percpu_to_op("add", (pcp), val) 220#define __this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
166#define __this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 221#define __this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
167#define __this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 222#define __this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
168#define __this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 223#define __this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
@@ -179,9 +234,9 @@ do { \
179#define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val) 234#define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
180#define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val) 235#define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
181#define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val) 236#define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
182#define this_cpu_add_1(pcp, val) percpu_to_op("add", (pcp), val) 237#define this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
183#define this_cpu_add_2(pcp, val) percpu_to_op("add", (pcp), val) 238#define this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
184#define this_cpu_add_4(pcp, val) percpu_to_op("add", (pcp), val) 239#define this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
185#define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 240#define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
186#define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 241#define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
187#define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 242#define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
@@ -192,9 +247,9 @@ do { \
192#define this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val) 247#define this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val)
193#define this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val) 248#define this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val)
194 249
195#define irqsafe_cpu_add_1(pcp, val) percpu_to_op("add", (pcp), val) 250#define irqsafe_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
196#define irqsafe_cpu_add_2(pcp, val) percpu_to_op("add", (pcp), val) 251#define irqsafe_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
197#define irqsafe_cpu_add_4(pcp, val) percpu_to_op("add", (pcp), val) 252#define irqsafe_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
198#define irqsafe_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val) 253#define irqsafe_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
199#define irqsafe_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val) 254#define irqsafe_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
200#define irqsafe_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val) 255#define irqsafe_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
@@ -212,19 +267,19 @@ do { \
212#ifdef CONFIG_X86_64 267#ifdef CONFIG_X86_64
213#define __this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 268#define __this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
214#define __this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val) 269#define __this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
215#define __this_cpu_add_8(pcp, val) percpu_to_op("add", (pcp), val) 270#define __this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
216#define __this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 271#define __this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
217#define __this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 272#define __this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
218#define __this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 273#define __this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
219 274
220#define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp)) 275#define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
221#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val) 276#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
222#define this_cpu_add_8(pcp, val) percpu_to_op("add", (pcp), val) 277#define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
223#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 278#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
224#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 279#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
225#define this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 280#define this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
226 281
227#define irqsafe_cpu_add_8(pcp, val) percpu_to_op("add", (pcp), val) 282#define irqsafe_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
228#define irqsafe_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val) 283#define irqsafe_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
229#define irqsafe_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val) 284#define irqsafe_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
230#define irqsafe_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val) 285#define irqsafe_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
@@ -236,7 +291,7 @@ do { \
236({ \ 291({ \
237 int old__; \ 292 int old__; \
238 asm volatile("btr %2,"__percpu_arg(1)"\n\tsbbl %0,%0" \ 293 asm volatile("btr %2,"__percpu_arg(1)"\n\tsbbl %0,%0" \
239 : "=r" (old__), "+m" (per_cpu__##var) \ 294 : "=r" (old__), "+m" (var) \
240 : "dIr" (bit)); \ 295 : "dIr" (bit)); \
241 old__; \ 296 old__; \
242}) 297})
diff --git a/arch/x86/include/asm/system.h b/arch/x86/include/asm/system.h
index e04740f7a0bb..b8fe48ee2ed9 100644
--- a/arch/x86/include/asm/system.h
+++ b/arch/x86/include/asm/system.h
@@ -32,7 +32,7 @@ extern void show_regs_common(void);
32 "movl %P[task_canary](%[next]), %%ebx\n\t" \ 32 "movl %P[task_canary](%[next]), %%ebx\n\t" \
33 "movl %%ebx, "__percpu_arg([stack_canary])"\n\t" 33 "movl %%ebx, "__percpu_arg([stack_canary])"\n\t"
34#define __switch_canary_oparam \ 34#define __switch_canary_oparam \
35 , [stack_canary] "=m" (per_cpu_var(stack_canary.canary)) 35 , [stack_canary] "=m" (stack_canary.canary)
36#define __switch_canary_iparam \ 36#define __switch_canary_iparam \
37 , [task_canary] "i" (offsetof(struct task_struct, stack_canary)) 37 , [task_canary] "i" (offsetof(struct task_struct, stack_canary))
38#else /* CC_STACKPROTECTOR */ 38#else /* CC_STACKPROTECTOR */
@@ -114,7 +114,7 @@ do { \
114 "movq %P[task_canary](%%rsi),%%r8\n\t" \ 114 "movq %P[task_canary](%%rsi),%%r8\n\t" \
115 "movq %%r8,"__percpu_arg([gs_canary])"\n\t" 115 "movq %%r8,"__percpu_arg([gs_canary])"\n\t"
116#define __switch_canary_oparam \ 116#define __switch_canary_oparam \
117 , [gs_canary] "=m" (per_cpu_var(irq_stack_union.stack_canary)) 117 , [gs_canary] "=m" (irq_stack_union.stack_canary)
118#define __switch_canary_iparam \ 118#define __switch_canary_iparam \
119 , [task_canary] "i" (offsetof(struct task_struct, stack_canary)) 119 , [task_canary] "i" (offsetof(struct task_struct, stack_canary))
120#else /* CC_STACKPROTECTOR */ 120#else /* CC_STACKPROTECTOR */
@@ -133,7 +133,7 @@ do { \
133 __switch_canary \ 133 __switch_canary \
134 "movq %P[thread_info](%%rsi),%%r8\n\t" \ 134 "movq %P[thread_info](%%rsi),%%r8\n\t" \
135 "movq %%rax,%%rdi\n\t" \ 135 "movq %%rax,%%rdi\n\t" \
136 "testl %[_tif_fork],%P[ti_flags](%%r8)\n\t" \ 136 "testl %[_tif_fork],%P[ti_flags](%%r8)\n\t" \
137 "jnz ret_from_fork\n\t" \ 137 "jnz ret_from_fork\n\t" \
138 RESTORE_CONTEXT \ 138 RESTORE_CONTEXT \
139 : "=a" (last) \ 139 : "=a" (last) \
@@ -143,7 +143,7 @@ do { \
143 [ti_flags] "i" (offsetof(struct thread_info, flags)), \ 143 [ti_flags] "i" (offsetof(struct thread_info, flags)), \
144 [_tif_fork] "i" (_TIF_FORK), \ 144 [_tif_fork] "i" (_TIF_FORK), \
145 [thread_info] "i" (offsetof(struct task_struct, stack)), \ 145 [thread_info] "i" (offsetof(struct task_struct, stack)), \
146 [current_task] "m" (per_cpu_var(current_task)) \ 146 [current_task] "m" (current_task) \
147 __switch_canary_iparam \ 147 __switch_canary_iparam \
148 : "memory", "cc" __EXTRA_CLOBBER) 148 : "memory", "cc" __EXTRA_CLOBBER)
149#endif 149#endif
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-13 16:46:21 -0500