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 | 30 | #include <asm/nmi.h> |
375074cc | 31 | #include <asm/tlbflush.h> |
90e24014 | 32 | |
45046892 TG |
33 | /* |
34 | * per-CPU TSS segments. Threads are completely 'soft' on Linux, | |
35 | * no more per-task TSS's. The TSS size is kept cacheline-aligned | |
36 | * so they are allowed to end up in the .data..cacheline_aligned | |
37 | * section. Since TSS's are completely CPU-local, we want them | |
38 | * on exact cacheline boundaries, to eliminate cacheline ping-pong. | |
39 | */ | |
277d5b40 | 40 | __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss) = INIT_TSS; |
45046892 | 41 | |
90e24014 RW |
42 | #ifdef CONFIG_X86_64 |
43 | static DEFINE_PER_CPU(unsigned char, is_idle); | |
44 | static ATOMIC_NOTIFIER_HEAD(idle_notifier); | |
45 | ||
46 | void idle_notifier_register(struct notifier_block *n) | |
47 | { | |
48 | atomic_notifier_chain_register(&idle_notifier, n); | |
49 | } | |
50 | EXPORT_SYMBOL_GPL(idle_notifier_register); | |
51 | ||
52 | void idle_notifier_unregister(struct notifier_block *n) | |
53 | { | |
54 | atomic_notifier_chain_unregister(&idle_notifier, n); | |
55 | } | |
56 | EXPORT_SYMBOL_GPL(idle_notifier_unregister); | |
57 | #endif | |
c1e3b377 | 58 | |
aa283f49 | 59 | struct kmem_cache *task_xstate_cachep; |
5ee481da | 60 | EXPORT_SYMBOL_GPL(task_xstate_cachep); |
61c4628b | 61 | |
55ccf3fe SS |
62 | /* |
63 | * this gets called so that we can store lazy state into memory and copy the | |
64 | * current task into the new thread. | |
65 | */ | |
61c4628b SS |
66 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) |
67 | { | |
68 | *dst = *src; | |
f1853505 | 69 | |
dc56c0f9 | 70 | dst->thread.fpu_counter = 0; |
5e23fee2 | 71 | dst->thread.fpu.has_fpu = 0; |
5e23fee2 | 72 | dst->thread.fpu.state = NULL; |
6a5fe895 | 73 | task_disable_lazy_fpu_restore(dst); |
f1853505 ON |
74 | if (tsk_used_math(src)) { |
75 | int err = fpu_alloc(&dst->thread.fpu); | |
76 | if (err) | |
77 | return err; | |
304bceda | 78 | fpu_copy(dst, src); |
aa283f49 | 79 | } |
61c4628b SS |
80 | return 0; |
81 | } | |
82 | ||
aa283f49 | 83 | void free_thread_xstate(struct task_struct *tsk) |
61c4628b | 84 | { |
86603283 | 85 | fpu_free(&tsk->thread.fpu); |
aa283f49 SS |
86 | } |
87 | ||
38e7c572 | 88 | void arch_release_task_struct(struct task_struct *tsk) |
aa283f49 | 89 | { |
38e7c572 | 90 | free_thread_xstate(tsk); |
61c4628b SS |
91 | } |
92 | ||
93 | void arch_task_cache_init(void) | |
94 | { | |
95 | task_xstate_cachep = | |
96 | kmem_cache_create("task_xstate", xstate_size, | |
97 | __alignof__(union thread_xstate), | |
2dff4405 | 98 | SLAB_PANIC | SLAB_NOTRACK, NULL); |
7496d645 | 99 | setup_xstate_comp(); |
61c4628b | 100 | } |
7f424a8b | 101 | |
389d1fb1 JF |
102 | /* |
103 | * Free current thread data structures etc.. | |
104 | */ | |
105 | void exit_thread(void) | |
106 | { | |
107 | struct task_struct *me = current; | |
108 | struct thread_struct *t = &me->thread; | |
250981e6 | 109 | unsigned long *bp = t->io_bitmap_ptr; |
389d1fb1 | 110 | |
250981e6 | 111 | if (bp) { |
389d1fb1 JF |
112 | struct tss_struct *tss = &per_cpu(init_tss, get_cpu()); |
113 | ||
389d1fb1 JF |
114 | t->io_bitmap_ptr = NULL; |
115 | clear_thread_flag(TIF_IO_BITMAP); | |
116 | /* | |
117 | * Careful, clear this in the TSS too: | |
118 | */ | |
119 | memset(tss->io_bitmap, 0xff, t->io_bitmap_max); | |
120 | t->io_bitmap_max = 0; | |
121 | put_cpu(); | |
250981e6 | 122 | kfree(bp); |
389d1fb1 | 123 | } |
1dcc8d7b SS |
124 | |
125 | drop_fpu(me); | |
389d1fb1 JF |
126 | } |
127 | ||
128 | void flush_thread(void) | |
129 | { | |
130 | struct task_struct *tsk = current; | |
131 | ||
24f1e32c | 132 | flush_ptrace_hw_breakpoint(tsk); |
389d1fb1 | 133 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); |
110d7f75 | 134 | |
304bceda SS |
135 | drop_init_fpu(tsk); |
136 | /* | |
137 | * Free the FPU state for non xsave platforms. They get reallocated | |
138 | * lazily at the first use. | |
139 | */ | |
5d2bd700 | 140 | if (!use_eager_fpu()) |
304bceda | 141 | free_thread_xstate(tsk); |
110d7f75 ON |
142 | else if (!used_math()) { |
143 | /* kthread execs. TODO: cleanup this horror. */ | |
144 | if (WARN_ON(init_fpu(current))) | |
145 | force_sig(SIGKILL, current); | |
146 | math_state_restore(); | |
147 | } | |
389d1fb1 JF |
148 | } |
149 | ||
150 | static void hard_disable_TSC(void) | |
151 | { | |
375074cc | 152 | cr4_set_bits(X86_CR4_TSD); |
389d1fb1 JF |
153 | } |
154 | ||
155 | void disable_TSC(void) | |
156 | { | |
157 | preempt_disable(); | |
158 | if (!test_and_set_thread_flag(TIF_NOTSC)) | |
159 | /* | |
160 | * Must flip the CPU state synchronously with | |
161 | * TIF_NOTSC in the current running context. | |
162 | */ | |
163 | hard_disable_TSC(); | |
164 | preempt_enable(); | |
165 | } | |
166 | ||
167 | static void hard_enable_TSC(void) | |
168 | { | |
375074cc | 169 | cr4_clear_bits(X86_CR4_TSD); |
389d1fb1 JF |
170 | } |
171 | ||
172 | static void enable_TSC(void) | |
173 | { | |
174 | preempt_disable(); | |
175 | if (test_and_clear_thread_flag(TIF_NOTSC)) | |
176 | /* | |
177 | * Must flip the CPU state synchronously with | |
178 | * TIF_NOTSC in the current running context. | |
179 | */ | |
180 | hard_enable_TSC(); | |
181 | preempt_enable(); | |
182 | } | |
183 | ||
184 | int get_tsc_mode(unsigned long adr) | |
185 | { | |
186 | unsigned int val; | |
187 | ||
188 | if (test_thread_flag(TIF_NOTSC)) | |
189 | val = PR_TSC_SIGSEGV; | |
190 | else | |
191 | val = PR_TSC_ENABLE; | |
192 | ||
193 | return put_user(val, (unsigned int __user *)adr); | |
194 | } | |
195 | ||
196 | int set_tsc_mode(unsigned int val) | |
197 | { | |
198 | if (val == PR_TSC_SIGSEGV) | |
199 | disable_TSC(); | |
200 | else if (val == PR_TSC_ENABLE) | |
201 | enable_TSC(); | |
202 | else | |
203 | return -EINVAL; | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
208 | void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, | |
209 | struct tss_struct *tss) | |
210 | { | |
211 | struct thread_struct *prev, *next; | |
212 | ||
213 | prev = &prev_p->thread; | |
214 | next = &next_p->thread; | |
215 | ||
ea8e61b7 PZ |
216 | if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^ |
217 | test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) { | |
218 | unsigned long debugctl = get_debugctlmsr(); | |
219 | ||
220 | debugctl &= ~DEBUGCTLMSR_BTF; | |
221 | if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) | |
222 | debugctl |= DEBUGCTLMSR_BTF; | |
223 | ||
224 | update_debugctlmsr(debugctl); | |
225 | } | |
389d1fb1 | 226 | |
389d1fb1 JF |
227 | if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^ |
228 | test_tsk_thread_flag(next_p, TIF_NOTSC)) { | |
229 | /* prev and next are different */ | |
230 | if (test_tsk_thread_flag(next_p, TIF_NOTSC)) | |
231 | hard_disable_TSC(); | |
232 | else | |
233 | hard_enable_TSC(); | |
234 | } | |
235 | ||
236 | if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { | |
237 | /* | |
238 | * Copy the relevant range of the IO bitmap. | |
239 | * Normally this is 128 bytes or less: | |
240 | */ | |
241 | memcpy(tss->io_bitmap, next->io_bitmap_ptr, | |
242 | max(prev->io_bitmap_max, next->io_bitmap_max)); | |
243 | } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) { | |
244 | /* | |
245 | * Clear any possible leftover bits: | |
246 | */ | |
247 | memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); | |
248 | } | |
7c68af6e | 249 | propagate_user_return_notify(prev_p, next_p); |
389d1fb1 JF |
250 | } |
251 | ||
00dba564 TG |
252 | /* |
253 | * Idle related variables and functions | |
254 | */ | |
d1896049 | 255 | unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; |
00dba564 TG |
256 | EXPORT_SYMBOL(boot_option_idle_override); |
257 | ||
a476bda3 | 258 | static void (*x86_idle)(void); |
00dba564 | 259 | |
90e24014 RW |
260 | #ifndef CONFIG_SMP |
261 | static inline void play_dead(void) | |
262 | { | |
263 | BUG(); | |
264 | } | |
265 | #endif | |
266 | ||
267 | #ifdef CONFIG_X86_64 | |
268 | void enter_idle(void) | |
269 | { | |
c6ae41e7 | 270 | this_cpu_write(is_idle, 1); |
90e24014 RW |
271 | atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL); |
272 | } | |
273 | ||
274 | static void __exit_idle(void) | |
275 | { | |
276 | if (x86_test_and_clear_bit_percpu(0, is_idle) == 0) | |
277 | return; | |
278 | atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL); | |
279 | } | |
280 | ||
281 | /* Called from interrupts to signify idle end */ | |
282 | void exit_idle(void) | |
283 | { | |
284 | /* idle loop has pid 0 */ | |
285 | if (current->pid) | |
286 | return; | |
287 | __exit_idle(); | |
288 | } | |
289 | #endif | |
290 | ||
7d1a9417 TG |
291 | void arch_cpu_idle_enter(void) |
292 | { | |
293 | local_touch_nmi(); | |
294 | enter_idle(); | |
295 | } | |
90e24014 | 296 | |
7d1a9417 TG |
297 | void arch_cpu_idle_exit(void) |
298 | { | |
299 | __exit_idle(); | |
300 | } | |
90e24014 | 301 | |
7d1a9417 TG |
302 | void arch_cpu_idle_dead(void) |
303 | { | |
304 | play_dead(); | |
305 | } | |
90e24014 | 306 | |
7d1a9417 TG |
307 | /* |
308 | * Called from the generic idle code. | |
309 | */ | |
310 | void arch_cpu_idle(void) | |
311 | { | |
16f8b05a | 312 | x86_idle(); |
90e24014 RW |
313 | } |
314 | ||
00dba564 | 315 | /* |
7d1a9417 | 316 | * We use this if we don't have any better idle routine.. |
00dba564 TG |
317 | */ |
318 | void default_idle(void) | |
319 | { | |
4d0e42cc | 320 | trace_cpu_idle_rcuidle(1, smp_processor_id()); |
7d1a9417 | 321 | safe_halt(); |
4d0e42cc | 322 | trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); |
00dba564 | 323 | } |
60b8b1de | 324 | #ifdef CONFIG_APM_MODULE |
00dba564 TG |
325 | EXPORT_SYMBOL(default_idle); |
326 | #endif | |
327 | ||
6a377ddc LB |
328 | #ifdef CONFIG_XEN |
329 | bool xen_set_default_idle(void) | |
e5fd47bf | 330 | { |
a476bda3 | 331 | bool ret = !!x86_idle; |
e5fd47bf | 332 | |
a476bda3 | 333 | x86_idle = default_idle; |
e5fd47bf KRW |
334 | |
335 | return ret; | |
336 | } | |
6a377ddc | 337 | #endif |
d3ec5cae IV |
338 | void stop_this_cpu(void *dummy) |
339 | { | |
340 | local_irq_disable(); | |
341 | /* | |
342 | * Remove this CPU: | |
343 | */ | |
4f062896 | 344 | set_cpu_online(smp_processor_id(), false); |
d3ec5cae IV |
345 | disable_local_APIC(); |
346 | ||
27be4570 LB |
347 | for (;;) |
348 | halt(); | |
7f424a8b PZ |
349 | } |
350 | ||
02c68a02 LB |
351 | bool amd_e400_c1e_detected; |
352 | EXPORT_SYMBOL(amd_e400_c1e_detected); | |
aa276e1c | 353 | |
02c68a02 | 354 | static cpumask_var_t amd_e400_c1e_mask; |
4faac97d | 355 | |
02c68a02 | 356 | void amd_e400_remove_cpu(int cpu) |
4faac97d | 357 | { |
02c68a02 LB |
358 | if (amd_e400_c1e_mask != NULL) |
359 | cpumask_clear_cpu(cpu, amd_e400_c1e_mask); | |
4faac97d TG |
360 | } |
361 | ||
aa276e1c | 362 | /* |
02c68a02 | 363 | * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt |
aa276e1c TG |
364 | * pending message MSR. If we detect C1E, then we handle it the same |
365 | * way as C3 power states (local apic timer and TSC stop) | |
366 | */ | |
02c68a02 | 367 | static void amd_e400_idle(void) |
aa276e1c | 368 | { |
02c68a02 | 369 | if (!amd_e400_c1e_detected) { |
aa276e1c TG |
370 | u32 lo, hi; |
371 | ||
372 | rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); | |
e8c534ec | 373 | |
aa276e1c | 374 | if (lo & K8_INTP_C1E_ACTIVE_MASK) { |
02c68a02 | 375 | amd_e400_c1e_detected = true; |
40fb1715 | 376 | if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) |
09bfeea1 | 377 | mark_tsc_unstable("TSC halt in AMD C1E"); |
c767a54b | 378 | pr_info("System has AMD C1E enabled\n"); |
aa276e1c TG |
379 | } |
380 | } | |
381 | ||
02c68a02 | 382 | if (amd_e400_c1e_detected) { |
aa276e1c TG |
383 | int cpu = smp_processor_id(); |
384 | ||
02c68a02 LB |
385 | if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) { |
386 | cpumask_set_cpu(cpu, amd_e400_c1e_mask); | |
0beefa20 | 387 | /* |
f833bab8 | 388 | * Force broadcast so ACPI can not interfere. |
0beefa20 | 389 | */ |
aa276e1c TG |
390 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, |
391 | &cpu); | |
c767a54b | 392 | pr_info("Switch to broadcast mode on CPU%d\n", cpu); |
aa276e1c TG |
393 | } |
394 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); | |
0beefa20 | 395 | |
aa276e1c | 396 | default_idle(); |
0beefa20 TG |
397 | |
398 | /* | |
399 | * The switch back from broadcast mode needs to be | |
400 | * called with interrupts disabled. | |
401 | */ | |
ea811747 PZ |
402 | local_irq_disable(); |
403 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu); | |
404 | local_irq_enable(); | |
aa276e1c TG |
405 | } else |
406 | default_idle(); | |
407 | } | |
408 | ||
148f9bb8 | 409 | void select_idle_routine(const struct cpuinfo_x86 *c) |
7f424a8b | 410 | { |
3e5095d1 | 411 | #ifdef CONFIG_SMP |
7d1a9417 | 412 | if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) |
c767a54b | 413 | pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n"); |
7f424a8b | 414 | #endif |
7d1a9417 | 415 | if (x86_idle || boot_option_idle_override == IDLE_POLL) |
6ddd2a27 TG |
416 | return; |
417 | ||
7d7dc116 | 418 | if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) { |
9d8888c2 | 419 | /* E400: APIC timer interrupt does not wake up CPU from C1e */ |
c767a54b | 420 | pr_info("using AMD E400 aware idle routine\n"); |
a476bda3 | 421 | x86_idle = amd_e400_idle; |
6ddd2a27 | 422 | } else |
a476bda3 | 423 | x86_idle = default_idle; |
7f424a8b PZ |
424 | } |
425 | ||
02c68a02 | 426 | void __init init_amd_e400_c1e_mask(void) |
30e1e6d1 | 427 | { |
02c68a02 | 428 | /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */ |
a476bda3 | 429 | if (x86_idle == amd_e400_idle) |
02c68a02 | 430 | zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL); |
30e1e6d1 RR |
431 | } |
432 | ||
7f424a8b PZ |
433 | static int __init idle_setup(char *str) |
434 | { | |
ab6bc3e3 CG |
435 | if (!str) |
436 | return -EINVAL; | |
437 | ||
7f424a8b | 438 | if (!strcmp(str, "poll")) { |
c767a54b | 439 | pr_info("using polling idle threads\n"); |
d1896049 | 440 | boot_option_idle_override = IDLE_POLL; |
7d1a9417 | 441 | cpu_idle_poll_ctrl(true); |
d1896049 | 442 | } else if (!strcmp(str, "halt")) { |
c1e3b377 ZY |
443 | /* |
444 | * When the boot option of idle=halt is added, halt is | |
445 | * forced to be used for CPU idle. In such case CPU C2/C3 | |
446 | * won't be used again. | |
447 | * To continue to load the CPU idle driver, don't touch | |
448 | * the boot_option_idle_override. | |
449 | */ | |
a476bda3 | 450 | x86_idle = default_idle; |
d1896049 | 451 | boot_option_idle_override = IDLE_HALT; |
da5e09a1 ZY |
452 | } else if (!strcmp(str, "nomwait")) { |
453 | /* | |
454 | * If the boot option of "idle=nomwait" is added, | |
455 | * it means that mwait will be disabled for CPU C2/C3 | |
456 | * states. In such case it won't touch the variable | |
457 | * of boot_option_idle_override. | |
458 | */ | |
d1896049 | 459 | boot_option_idle_override = IDLE_NOMWAIT; |
c1e3b377 | 460 | } else |
7f424a8b PZ |
461 | return -1; |
462 | ||
7f424a8b PZ |
463 | return 0; |
464 | } | |
465 | early_param("idle", idle_setup); | |
466 | ||
9d62dcdf AW |
467 | unsigned long arch_align_stack(unsigned long sp) |
468 | { | |
469 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) | |
470 | sp -= get_random_int() % 8192; | |
471 | return sp & ~0xf; | |
472 | } | |
473 | ||
474 | unsigned long arch_randomize_brk(struct mm_struct *mm) | |
475 | { | |
476 | unsigned long range_end = mm->brk + 0x02000000; | |
477 | return randomize_range(mm->brk, range_end, 0) ? : mm->brk; | |
478 | } | |
479 |