1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 #include <linux/errno.h>
4 #include <linux/kernel.h>
7 #include <linux/prctl.h>
8 #include <linux/slab.h>
9 #include <linux/sched.h>
10 #include <linux/module.h>
12 #include <linux/tick.h>
13 #include <linux/random.h>
14 #include <linux/user-return-notifier.h>
15 #include <linux/dmi.h>
16 #include <linux/utsname.h>
17 #include <linux/stackprotector.h>
18 #include <linux/tick.h>
19 #include <linux/cpuidle.h>
20 #include <trace/events/power.h>
21 #include <linux/hw_breakpoint.h>
24 #include <asm/syscalls.h>
26 #include <asm/uaccess.h>
27 #include <asm/mwait.h>
29 #include <asm/fpu-internal.h>
30 #include <asm/debugreg.h>
32 #include <asm/tlbflush.h>
35 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
36 * no more per-task TSS's. The TSS size is kept cacheline-aligned
37 * so they are allowed to end up in the .data..cacheline_aligned
38 * section. Since TSS's are completely CPU-local, we want them
39 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
41 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
43 .sp0 = TOP_OF_INIT_STACK,
47 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
52 * Note that the .io_bitmap member must be extra-big. This is because
53 * the CPU will access an additional byte beyond the end of the IO
54 * permission bitmap. The extra byte must be all 1 bits, and must
55 * be within the limit.
57 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 },
60 EXPORT_PER_CPU_SYMBOL(cpu_tss);
63 static DEFINE_PER_CPU(unsigned char, is_idle);
64 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
66 void idle_notifier_register(struct notifier_block *n)
68 atomic_notifier_chain_register(&idle_notifier, n);
70 EXPORT_SYMBOL_GPL(idle_notifier_register);
72 void idle_notifier_unregister(struct notifier_block *n)
74 atomic_notifier_chain_unregister(&idle_notifier, n);
76 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
79 struct kmem_cache *task_xstate_cachep;
80 EXPORT_SYMBOL_GPL(task_xstate_cachep);
83 * this gets called so that we can store lazy state into memory and copy the
84 * current task into the new thread.
86 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
90 dst->thread.fpu_counter = 0;
91 dst->thread.fpu.has_fpu = 0;
92 dst->thread.fpu.state = NULL;
93 task_disable_lazy_fpu_restore(dst);
94 if (tsk_used_math(src)) {
95 int err = fpu_alloc(&dst->thread.fpu);
103 void free_thread_xstate(struct task_struct *tsk)
105 fpu_free(&tsk->thread.fpu);
108 void arch_release_task_struct(struct task_struct *tsk)
110 free_thread_xstate(tsk);
113 void arch_task_cache_init(void)
116 kmem_cache_create("task_xstate", xstate_size,
117 __alignof__(union thread_xstate),
118 SLAB_PANIC | SLAB_NOTRACK, NULL);
123 * Free current thread data structures etc..
125 void exit_thread(void)
127 struct task_struct *me = current;
128 struct thread_struct *t = &me->thread;
129 unsigned long *bp = t->io_bitmap_ptr;
132 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
134 t->io_bitmap_ptr = NULL;
135 clear_thread_flag(TIF_IO_BITMAP);
137 * Careful, clear this in the TSS too:
139 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
140 t->io_bitmap_max = 0;
148 void flush_thread(void)
150 struct task_struct *tsk = current;
152 flush_ptrace_hw_breakpoint(tsk);
153 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
155 if (!use_eager_fpu()) {
156 /* FPU state will be reallocated lazily at the first use. */
158 free_thread_xstate(tsk);
160 if (!tsk_used_math(tsk)) {
161 /* kthread execs. TODO: cleanup this horror. */
162 if (WARN_ON(init_fpu(tsk)))
163 force_sig(SIGKILL, tsk);
166 restore_init_xstate();
170 static void hard_disable_TSC(void)
172 cr4_set_bits(X86_CR4_TSD);
175 void disable_TSC(void)
178 if (!test_and_set_thread_flag(TIF_NOTSC))
180 * Must flip the CPU state synchronously with
181 * TIF_NOTSC in the current running context.
187 static void hard_enable_TSC(void)
189 cr4_clear_bits(X86_CR4_TSD);
192 static void enable_TSC(void)
195 if (test_and_clear_thread_flag(TIF_NOTSC))
197 * Must flip the CPU state synchronously with
198 * TIF_NOTSC in the current running context.
204 int get_tsc_mode(unsigned long adr)
208 if (test_thread_flag(TIF_NOTSC))
209 val = PR_TSC_SIGSEGV;
213 return put_user(val, (unsigned int __user *)adr);
216 int set_tsc_mode(unsigned int val)
218 if (val == PR_TSC_SIGSEGV)
220 else if (val == PR_TSC_ENABLE)
228 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
229 struct tss_struct *tss)
231 struct thread_struct *prev, *next;
233 prev = &prev_p->thread;
234 next = &next_p->thread;
236 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
237 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
238 unsigned long debugctl = get_debugctlmsr();
240 debugctl &= ~DEBUGCTLMSR_BTF;
241 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
242 debugctl |= DEBUGCTLMSR_BTF;
244 update_debugctlmsr(debugctl);
247 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
248 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
249 /* prev and next are different */
250 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
256 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
258 * Copy the relevant range of the IO bitmap.
259 * Normally this is 128 bytes or less:
261 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
262 max(prev->io_bitmap_max, next->io_bitmap_max));
263 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
265 * Clear any possible leftover bits:
267 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
269 propagate_user_return_notify(prev_p, next_p);
273 * Idle related variables and functions
275 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
276 EXPORT_SYMBOL(boot_option_idle_override);
278 static void (*x86_idle)(void);
281 static inline void play_dead(void)
288 void enter_idle(void)
290 this_cpu_write(is_idle, 1);
291 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
294 static void __exit_idle(void)
296 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
298 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
301 /* Called from interrupts to signify idle end */
304 /* idle loop has pid 0 */
311 void arch_cpu_idle_enter(void)
317 void arch_cpu_idle_exit(void)
322 void arch_cpu_idle_dead(void)
328 * Called from the generic idle code.
330 void arch_cpu_idle(void)
336 * We use this if we don't have any better idle routine..
338 void default_idle(void)
340 trace_cpu_idle_rcuidle(1, smp_processor_id());
342 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
344 #ifdef CONFIG_APM_MODULE
345 EXPORT_SYMBOL(default_idle);
349 bool xen_set_default_idle(void)
351 bool ret = !!x86_idle;
353 x86_idle = default_idle;
358 void stop_this_cpu(void *dummy)
364 set_cpu_online(smp_processor_id(), false);
365 disable_local_APIC();
371 bool amd_e400_c1e_detected;
372 EXPORT_SYMBOL(amd_e400_c1e_detected);
374 static cpumask_var_t amd_e400_c1e_mask;
376 void amd_e400_remove_cpu(int cpu)
378 if (amd_e400_c1e_mask != NULL)
379 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
383 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
384 * pending message MSR. If we detect C1E, then we handle it the same
385 * way as C3 power states (local apic timer and TSC stop)
387 static void amd_e400_idle(void)
389 if (!amd_e400_c1e_detected) {
392 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
394 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
395 amd_e400_c1e_detected = true;
396 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
397 mark_tsc_unstable("TSC halt in AMD C1E");
398 pr_info("System has AMD C1E enabled\n");
402 if (amd_e400_c1e_detected) {
403 int cpu = smp_processor_id();
405 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
406 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
407 /* Force broadcast so ACPI can not interfere. */
408 tick_broadcast_force();
409 pr_info("Switch to broadcast mode on CPU%d\n", cpu);
411 tick_broadcast_enter();
416 * The switch back from broadcast mode needs to be
417 * called with interrupts disabled.
420 tick_broadcast_exit();
427 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
428 * We can't rely on cpuidle installing MWAIT, because it will not load
429 * on systems that support only C1 -- so the boot default must be MWAIT.
431 * Some AMD machines are the opposite, they depend on using HALT.
433 * So for default C1, which is used during boot until cpuidle loads,
434 * use MWAIT-C1 on Intel HW that has it, else use HALT.
436 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
438 if (c->x86_vendor != X86_VENDOR_INTEL)
441 if (!cpu_has(c, X86_FEATURE_MWAIT))
448 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
449 * with interrupts enabled and no flags, which is backwards compatible with the
450 * original MWAIT implementation.
452 static void mwait_idle(void)
454 if (!current_set_polling_and_test()) {
455 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
456 smp_mb(); /* quirk */
457 clflush((void *)¤t_thread_info()->flags);
458 smp_mb(); /* quirk */
461 __monitor((void *)¤t_thread_info()->flags, 0, 0);
469 __current_clr_polling();
472 void select_idle_routine(const struct cpuinfo_x86 *c)
475 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
476 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
478 if (x86_idle || boot_option_idle_override == IDLE_POLL)
481 if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
482 /* E400: APIC timer interrupt does not wake up CPU from C1e */
483 pr_info("using AMD E400 aware idle routine\n");
484 x86_idle = amd_e400_idle;
485 } else if (prefer_mwait_c1_over_halt(c)) {
486 pr_info("using mwait in idle threads\n");
487 x86_idle = mwait_idle;
489 x86_idle = default_idle;
492 void __init init_amd_e400_c1e_mask(void)
494 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
495 if (x86_idle == amd_e400_idle)
496 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
499 static int __init idle_setup(char *str)
504 if (!strcmp(str, "poll")) {
505 pr_info("using polling idle threads\n");
506 boot_option_idle_override = IDLE_POLL;
507 cpu_idle_poll_ctrl(true);
508 } else if (!strcmp(str, "halt")) {
510 * When the boot option of idle=halt is added, halt is
511 * forced to be used for CPU idle. In such case CPU C2/C3
512 * won't be used again.
513 * To continue to load the CPU idle driver, don't touch
514 * the boot_option_idle_override.
516 x86_idle = default_idle;
517 boot_option_idle_override = IDLE_HALT;
518 } else if (!strcmp(str, "nomwait")) {
520 * If the boot option of "idle=nomwait" is added,
521 * it means that mwait will be disabled for CPU C2/C3
522 * states. In such case it won't touch the variable
523 * of boot_option_idle_override.
525 boot_option_idle_override = IDLE_NOMWAIT;
531 early_param("idle", idle_setup);
533 unsigned long arch_align_stack(unsigned long sp)
535 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
536 sp -= get_random_int() % 8192;
540 unsigned long arch_randomize_brk(struct mm_struct *mm)
542 unsigned long range_end = mm->brk + 0x02000000;
543 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;