Commit | Line | Data |
---|---|---|
c767a54b JP |
1 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
2 | ||
61c4628b SS |
3 | #include <linux/errno.h> |
4 | #include <linux/kernel.h> | |
5 | #include <linux/mm.h> | |
6 | #include <linux/smp.h> | |
389d1fb1 | 7 | #include <linux/prctl.h> |
61c4628b SS |
8 | #include <linux/slab.h> |
9 | #include <linux/sched.h> | |
7f424a8b PZ |
10 | #include <linux/module.h> |
11 | #include <linux/pm.h> | |
aa276e1c | 12 | #include <linux/clockchips.h> |
9d62dcdf | 13 | #include <linux/random.h> |
7c68af6e | 14 | #include <linux/user-return-notifier.h> |
814e2c84 AI |
15 | #include <linux/dmi.h> |
16 | #include <linux/utsname.h> | |
90e24014 RW |
17 | #include <linux/stackprotector.h> |
18 | #include <linux/tick.h> | |
19 | #include <linux/cpuidle.h> | |
61613521 | 20 | #include <trace/events/power.h> |
24f1e32c | 21 | #include <linux/hw_breakpoint.h> |
93789b32 | 22 | #include <asm/cpu.h> |
d3ec5cae | 23 | #include <asm/apic.h> |
2c1b284e | 24 | #include <asm/syscalls.h> |
389d1fb1 JF |
25 | #include <asm/idle.h> |
26 | #include <asm/uaccess.h> | |
27 | #include <asm/i387.h> | |
1361b83a | 28 | #include <asm/fpu-internal.h> |
66cb5917 | 29 | #include <asm/debugreg.h> |
90e24014 RW |
30 | #include <asm/nmi.h> |
31 | ||
45046892 TG |
32 | /* |
33 | * per-CPU TSS segments. Threads are completely 'soft' on Linux, | |
34 | * no more per-task TSS's. The TSS size is kept cacheline-aligned | |
35 | * so they are allowed to end up in the .data..cacheline_aligned | |
36 | * section. Since TSS's are completely CPU-local, we want them | |
37 | * on exact cacheline boundaries, to eliminate cacheline ping-pong. | |
38 | */ | |
277d5b40 | 39 | __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss) = INIT_TSS; |
45046892 | 40 | |
90e24014 RW |
41 | #ifdef CONFIG_X86_64 |
42 | static DEFINE_PER_CPU(unsigned char, is_idle); | |
43 | static ATOMIC_NOTIFIER_HEAD(idle_notifier); | |
44 | ||
45 | void idle_notifier_register(struct notifier_block *n) | |
46 | { | |
47 | atomic_notifier_chain_register(&idle_notifier, n); | |
48 | } | |
49 | EXPORT_SYMBOL_GPL(idle_notifier_register); | |
50 | ||
51 | void idle_notifier_unregister(struct notifier_block *n) | |
52 | { | |
53 | atomic_notifier_chain_unregister(&idle_notifier, n); | |
54 | } | |
55 | EXPORT_SYMBOL_GPL(idle_notifier_unregister); | |
56 | #endif | |
c1e3b377 | 57 | |
aa283f49 | 58 | struct kmem_cache *task_xstate_cachep; |
5ee481da | 59 | EXPORT_SYMBOL_GPL(task_xstate_cachep); |
61c4628b | 60 | |
55ccf3fe SS |
61 | /* |
62 | * this gets called so that we can store lazy state into memory and copy the | |
63 | * current task into the new thread. | |
64 | */ | |
61c4628b SS |
65 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) |
66 | { | |
86603283 AK |
67 | int ret; |
68 | ||
61c4628b | 69 | *dst = *src; |
86603283 AK |
70 | if (fpu_allocated(&src->thread.fpu)) { |
71 | memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu)); | |
72 | ret = fpu_alloc(&dst->thread.fpu); | |
73 | if (ret) | |
74 | return ret; | |
304bceda | 75 | fpu_copy(dst, src); |
aa283f49 | 76 | } |
61c4628b SS |
77 | return 0; |
78 | } | |
79 | ||
aa283f49 | 80 | void free_thread_xstate(struct task_struct *tsk) |
61c4628b | 81 | { |
86603283 | 82 | fpu_free(&tsk->thread.fpu); |
aa283f49 SS |
83 | } |
84 | ||
38e7c572 | 85 | void arch_release_task_struct(struct task_struct *tsk) |
aa283f49 | 86 | { |
38e7c572 | 87 | free_thread_xstate(tsk); |
61c4628b SS |
88 | } |
89 | ||
90 | void arch_task_cache_init(void) | |
91 | { | |
92 | task_xstate_cachep = | |
93 | kmem_cache_create("task_xstate", xstate_size, | |
94 | __alignof__(union thread_xstate), | |
2dff4405 | 95 | SLAB_PANIC | SLAB_NOTRACK, NULL); |
7496d645 | 96 | setup_xstate_comp(); |
61c4628b | 97 | } |
7f424a8b | 98 | |
389d1fb1 JF |
99 | /* |
100 | * Free current thread data structures etc.. | |
101 | */ | |
102 | void exit_thread(void) | |
103 | { | |
104 | struct task_struct *me = current; | |
105 | struct thread_struct *t = &me->thread; | |
250981e6 | 106 | unsigned long *bp = t->io_bitmap_ptr; |
389d1fb1 | 107 | |
250981e6 | 108 | if (bp) { |
389d1fb1 JF |
109 | struct tss_struct *tss = &per_cpu(init_tss, get_cpu()); |
110 | ||
389d1fb1 JF |
111 | t->io_bitmap_ptr = NULL; |
112 | clear_thread_flag(TIF_IO_BITMAP); | |
113 | /* | |
114 | * Careful, clear this in the TSS too: | |
115 | */ | |
116 | memset(tss->io_bitmap, 0xff, t->io_bitmap_max); | |
117 | t->io_bitmap_max = 0; | |
118 | put_cpu(); | |
250981e6 | 119 | kfree(bp); |
389d1fb1 | 120 | } |
1dcc8d7b SS |
121 | |
122 | drop_fpu(me); | |
389d1fb1 JF |
123 | } |
124 | ||
125 | void flush_thread(void) | |
126 | { | |
127 | struct task_struct *tsk = current; | |
128 | ||
24f1e32c | 129 | flush_ptrace_hw_breakpoint(tsk); |
389d1fb1 | 130 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); |
304bceda SS |
131 | drop_init_fpu(tsk); |
132 | /* | |
133 | * Free the FPU state for non xsave platforms. They get reallocated | |
134 | * lazily at the first use. | |
135 | */ | |
5d2bd700 | 136 | if (!use_eager_fpu()) |
304bceda | 137 | free_thread_xstate(tsk); |
389d1fb1 JF |
138 | } |
139 | ||
140 | static void hard_disable_TSC(void) | |
141 | { | |
142 | write_cr4(read_cr4() | X86_CR4_TSD); | |
143 | } | |
144 | ||
145 | void disable_TSC(void) | |
146 | { | |
147 | preempt_disable(); | |
148 | if (!test_and_set_thread_flag(TIF_NOTSC)) | |
149 | /* | |
150 | * Must flip the CPU state synchronously with | |
151 | * TIF_NOTSC in the current running context. | |
152 | */ | |
153 | hard_disable_TSC(); | |
154 | preempt_enable(); | |
155 | } | |
156 | ||
157 | static void hard_enable_TSC(void) | |
158 | { | |
159 | write_cr4(read_cr4() & ~X86_CR4_TSD); | |
160 | } | |
161 | ||
162 | static void enable_TSC(void) | |
163 | { | |
164 | preempt_disable(); | |
165 | if (test_and_clear_thread_flag(TIF_NOTSC)) | |
166 | /* | |
167 | * Must flip the CPU state synchronously with | |
168 | * TIF_NOTSC in the current running context. | |
169 | */ | |
170 | hard_enable_TSC(); | |
171 | preempt_enable(); | |
172 | } | |
173 | ||
174 | int get_tsc_mode(unsigned long adr) | |
175 | { | |
176 | unsigned int val; | |
177 | ||
178 | if (test_thread_flag(TIF_NOTSC)) | |
179 | val = PR_TSC_SIGSEGV; | |
180 | else | |
181 | val = PR_TSC_ENABLE; | |
182 | ||
183 | return put_user(val, (unsigned int __user *)adr); | |
184 | } | |
185 | ||
186 | int set_tsc_mode(unsigned int val) | |
187 | { | |
188 | if (val == PR_TSC_SIGSEGV) | |
189 | disable_TSC(); | |
190 | else if (val == PR_TSC_ENABLE) | |
191 | enable_TSC(); | |
192 | else | |
193 | return -EINVAL; | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, | |
199 | struct tss_struct *tss) | |
200 | { | |
201 | struct thread_struct *prev, *next; | |
202 | ||
203 | prev = &prev_p->thread; | |
204 | next = &next_p->thread; | |
205 | ||
ea8e61b7 PZ |
206 | if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^ |
207 | test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) { | |
208 | unsigned long debugctl = get_debugctlmsr(); | |
209 | ||
210 | debugctl &= ~DEBUGCTLMSR_BTF; | |
211 | if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) | |
212 | debugctl |= DEBUGCTLMSR_BTF; | |
213 | ||
214 | update_debugctlmsr(debugctl); | |
215 | } | |
389d1fb1 | 216 | |
389d1fb1 JF |
217 | if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^ |
218 | test_tsk_thread_flag(next_p, TIF_NOTSC)) { | |
219 | /* prev and next are different */ | |
220 | if (test_tsk_thread_flag(next_p, TIF_NOTSC)) | |
221 | hard_disable_TSC(); | |
222 | else | |
223 | hard_enable_TSC(); | |
224 | } | |
225 | ||
226 | if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { | |
227 | /* | |
228 | * Copy the relevant range of the IO bitmap. | |
229 | * Normally this is 128 bytes or less: | |
230 | */ | |
231 | memcpy(tss->io_bitmap, next->io_bitmap_ptr, | |
232 | max(prev->io_bitmap_max, next->io_bitmap_max)); | |
233 | } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) { | |
234 | /* | |
235 | * Clear any possible leftover bits: | |
236 | */ | |
237 | memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); | |
238 | } | |
7c68af6e | 239 | propagate_user_return_notify(prev_p, next_p); |
389d1fb1 JF |
240 | } |
241 | ||
00dba564 TG |
242 | /* |
243 | * Idle related variables and functions | |
244 | */ | |
d1896049 | 245 | unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; |
00dba564 TG |
246 | EXPORT_SYMBOL(boot_option_idle_override); |
247 | ||
a476bda3 | 248 | static void (*x86_idle)(void); |
00dba564 | 249 | |
90e24014 RW |
250 | #ifndef CONFIG_SMP |
251 | static inline void play_dead(void) | |
252 | { | |
253 | BUG(); | |
254 | } | |
255 | #endif | |
256 | ||
257 | #ifdef CONFIG_X86_64 | |
258 | void enter_idle(void) | |
259 | { | |
c6ae41e7 | 260 | this_cpu_write(is_idle, 1); |
90e24014 RW |
261 | atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL); |
262 | } | |
263 | ||
264 | static void __exit_idle(void) | |
265 | { | |
266 | if (x86_test_and_clear_bit_percpu(0, is_idle) == 0) | |
267 | return; | |
268 | atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL); | |
269 | } | |
270 | ||
271 | /* Called from interrupts to signify idle end */ | |
272 | void exit_idle(void) | |
273 | { | |
274 | /* idle loop has pid 0 */ | |
275 | if (current->pid) | |
276 | return; | |
277 | __exit_idle(); | |
278 | } | |
279 | #endif | |
280 | ||
7d1a9417 TG |
281 | void arch_cpu_idle_enter(void) |
282 | { | |
283 | local_touch_nmi(); | |
284 | enter_idle(); | |
285 | } | |
90e24014 | 286 | |
7d1a9417 TG |
287 | void arch_cpu_idle_exit(void) |
288 | { | |
289 | __exit_idle(); | |
290 | } | |
90e24014 | 291 | |
7d1a9417 TG |
292 | void arch_cpu_idle_dead(void) |
293 | { | |
294 | play_dead(); | |
295 | } | |
90e24014 | 296 | |
7d1a9417 TG |
297 | /* |
298 | * Called from the generic idle code. | |
299 | */ | |
300 | void arch_cpu_idle(void) | |
301 | { | |
16f8b05a | 302 | x86_idle(); |
90e24014 RW |
303 | } |
304 | ||
00dba564 | 305 | /* |
7d1a9417 | 306 | * We use this if we don't have any better idle routine.. |
00dba564 TG |
307 | */ |
308 | void default_idle(void) | |
309 | { | |
4d0e42cc | 310 | trace_cpu_idle_rcuidle(1, smp_processor_id()); |
7d1a9417 | 311 | safe_halt(); |
4d0e42cc | 312 | trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); |
00dba564 | 313 | } |
60b8b1de | 314 | #ifdef CONFIG_APM_MODULE |
00dba564 TG |
315 | EXPORT_SYMBOL(default_idle); |
316 | #endif | |
317 | ||
6a377ddc LB |
318 | #ifdef CONFIG_XEN |
319 | bool xen_set_default_idle(void) | |
e5fd47bf | 320 | { |
a476bda3 | 321 | bool ret = !!x86_idle; |
e5fd47bf | 322 | |
a476bda3 | 323 | x86_idle = default_idle; |
e5fd47bf KRW |
324 | |
325 | return ret; | |
326 | } | |
6a377ddc | 327 | #endif |
d3ec5cae IV |
328 | void stop_this_cpu(void *dummy) |
329 | { | |
330 | local_irq_disable(); | |
331 | /* | |
332 | * Remove this CPU: | |
333 | */ | |
4f062896 | 334 | set_cpu_online(smp_processor_id(), false); |
d3ec5cae IV |
335 | disable_local_APIC(); |
336 | ||
27be4570 LB |
337 | for (;;) |
338 | halt(); | |
7f424a8b PZ |
339 | } |
340 | ||
02c68a02 LB |
341 | bool amd_e400_c1e_detected; |
342 | EXPORT_SYMBOL(amd_e400_c1e_detected); | |
aa276e1c | 343 | |
02c68a02 | 344 | static cpumask_var_t amd_e400_c1e_mask; |
4faac97d | 345 | |
02c68a02 | 346 | void amd_e400_remove_cpu(int cpu) |
4faac97d | 347 | { |
02c68a02 LB |
348 | if (amd_e400_c1e_mask != NULL) |
349 | cpumask_clear_cpu(cpu, amd_e400_c1e_mask); | |
4faac97d TG |
350 | } |
351 | ||
aa276e1c | 352 | /* |
02c68a02 | 353 | * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt |
aa276e1c TG |
354 | * pending message MSR. If we detect C1E, then we handle it the same |
355 | * way as C3 power states (local apic timer and TSC stop) | |
356 | */ | |
02c68a02 | 357 | static void amd_e400_idle(void) |
aa276e1c | 358 | { |
02c68a02 | 359 | if (!amd_e400_c1e_detected) { |
aa276e1c TG |
360 | u32 lo, hi; |
361 | ||
362 | rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); | |
e8c534ec | 363 | |
aa276e1c | 364 | if (lo & K8_INTP_C1E_ACTIVE_MASK) { |
02c68a02 | 365 | amd_e400_c1e_detected = true; |
40fb1715 | 366 | if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) |
09bfeea1 | 367 | mark_tsc_unstable("TSC halt in AMD C1E"); |
c767a54b | 368 | pr_info("System has AMD C1E enabled\n"); |
aa276e1c TG |
369 | } |
370 | } | |
371 | ||
02c68a02 | 372 | if (amd_e400_c1e_detected) { |
aa276e1c TG |
373 | int cpu = smp_processor_id(); |
374 | ||
02c68a02 LB |
375 | if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) { |
376 | cpumask_set_cpu(cpu, amd_e400_c1e_mask); | |
0beefa20 | 377 | /* |
f833bab8 | 378 | * Force broadcast so ACPI can not interfere. |
0beefa20 | 379 | */ |
aa276e1c TG |
380 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, |
381 | &cpu); | |
c767a54b | 382 | pr_info("Switch to broadcast mode on CPU%d\n", cpu); |
aa276e1c TG |
383 | } |
384 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); | |
0beefa20 | 385 | |
aa276e1c | 386 | default_idle(); |
0beefa20 TG |
387 | |
388 | /* | |
389 | * The switch back from broadcast mode needs to be | |
390 | * called with interrupts disabled. | |
391 | */ | |
ea811747 PZ |
392 | local_irq_disable(); |
393 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu); | |
394 | local_irq_enable(); | |
aa276e1c TG |
395 | } else |
396 | default_idle(); | |
397 | } | |
398 | ||
148f9bb8 | 399 | void select_idle_routine(const struct cpuinfo_x86 *c) |
7f424a8b | 400 | { |
3e5095d1 | 401 | #ifdef CONFIG_SMP |
7d1a9417 | 402 | if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) |
c767a54b | 403 | pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n"); |
7f424a8b | 404 | #endif |
7d1a9417 | 405 | if (x86_idle || boot_option_idle_override == IDLE_POLL) |
6ddd2a27 TG |
406 | return; |
407 | ||
7d7dc116 | 408 | if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) { |
9d8888c2 | 409 | /* E400: APIC timer interrupt does not wake up CPU from C1e */ |
c767a54b | 410 | pr_info("using AMD E400 aware idle routine\n"); |
a476bda3 | 411 | x86_idle = amd_e400_idle; |
6ddd2a27 | 412 | } else |
a476bda3 | 413 | x86_idle = default_idle; |
7f424a8b PZ |
414 | } |
415 | ||
02c68a02 | 416 | void __init init_amd_e400_c1e_mask(void) |
30e1e6d1 | 417 | { |
02c68a02 | 418 | /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */ |
a476bda3 | 419 | if (x86_idle == amd_e400_idle) |
02c68a02 | 420 | zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL); |
30e1e6d1 RR |
421 | } |
422 | ||
7f424a8b PZ |
423 | static int __init idle_setup(char *str) |
424 | { | |
ab6bc3e3 CG |
425 | if (!str) |
426 | return -EINVAL; | |
427 | ||
7f424a8b | 428 | if (!strcmp(str, "poll")) { |
c767a54b | 429 | pr_info("using polling idle threads\n"); |
d1896049 | 430 | boot_option_idle_override = IDLE_POLL; |
7d1a9417 | 431 | cpu_idle_poll_ctrl(true); |
d1896049 | 432 | } else if (!strcmp(str, "halt")) { |
c1e3b377 ZY |
433 | /* |
434 | * When the boot option of idle=halt is added, halt is | |
435 | * forced to be used for CPU idle. In such case CPU C2/C3 | |
436 | * won't be used again. | |
437 | * To continue to load the CPU idle driver, don't touch | |
438 | * the boot_option_idle_override. | |
439 | */ | |
a476bda3 | 440 | x86_idle = default_idle; |
d1896049 | 441 | boot_option_idle_override = IDLE_HALT; |
da5e09a1 ZY |
442 | } else if (!strcmp(str, "nomwait")) { |
443 | /* | |
444 | * If the boot option of "idle=nomwait" is added, | |
445 | * it means that mwait will be disabled for CPU C2/C3 | |
446 | * states. In such case it won't touch the variable | |
447 | * of boot_option_idle_override. | |
448 | */ | |
d1896049 | 449 | boot_option_idle_override = IDLE_NOMWAIT; |
c1e3b377 | 450 | } else |
7f424a8b PZ |
451 | return -1; |
452 | ||
7f424a8b PZ |
453 | return 0; |
454 | } | |
455 | early_param("idle", idle_setup); | |
456 | ||
9d62dcdf AW |
457 | unsigned long arch_align_stack(unsigned long sp) |
458 | { | |
459 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) | |
460 | sp -= get_random_int() % 8192; | |
461 | return sp & ~0xf; | |
462 | } | |
463 | ||
464 | unsigned long arch_randomize_brk(struct mm_struct *mm) | |
465 | { | |
466 | unsigned long range_end = mm->brk + 0x02000000; | |
467 | return randomize_range(mm->brk, range_end, 0) ? : mm->brk; | |
468 | } | |
469 |