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/clockchips.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, init_tss) = INIT_TSS;
44 static DEFINE_PER_CPU(unsigned char, is_idle);
45 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
47 void idle_notifier_register(struct notifier_block *n)
49 atomic_notifier_chain_register(&idle_notifier, n);
51 EXPORT_SYMBOL_GPL(idle_notifier_register);
53 void idle_notifier_unregister(struct notifier_block *n)
55 atomic_notifier_chain_unregister(&idle_notifier, n);
57 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
60 struct kmem_cache *task_xstate_cachep;
61 EXPORT_SYMBOL_GPL(task_xstate_cachep);
64 * this gets called so that we can store lazy state into memory and copy the
65 * current task into the new thread.
67 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
71 dst->thread.fpu_counter = 0;
72 dst->thread.fpu.has_fpu = 0;
73 dst->thread.fpu.last_cpu = ~0;
74 dst->thread.fpu.state = NULL;
75 if (tsk_used_math(src)) {
76 int err = fpu_alloc(&dst->thread.fpu);
84 void free_thread_xstate(struct task_struct *tsk)
86 fpu_free(&tsk->thread.fpu);
89 void arch_release_task_struct(struct task_struct *tsk)
91 free_thread_xstate(tsk);
94 void arch_task_cache_init(void)
97 kmem_cache_create("task_xstate", xstate_size,
98 __alignof__(union thread_xstate),
99 SLAB_PANIC | SLAB_NOTRACK, NULL);
104 * Free current thread data structures etc..
106 void exit_thread(void)
108 struct task_struct *me = current;
109 struct thread_struct *t = &me->thread;
110 unsigned long *bp = t->io_bitmap_ptr;
113 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
115 t->io_bitmap_ptr = NULL;
116 clear_thread_flag(TIF_IO_BITMAP);
118 * Careful, clear this in the TSS too:
120 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
121 t->io_bitmap_max = 0;
129 void flush_thread(void)
131 struct task_struct *tsk = current;
133 flush_ptrace_hw_breakpoint(tsk);
134 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
137 * Free the FPU state for non xsave platforms. They get reallocated
138 * lazily at the first use.
140 if (!use_eager_fpu())
141 free_thread_xstate(tsk);
144 static void hard_disable_TSC(void)
146 cr4_set_bits(X86_CR4_TSD);
149 void disable_TSC(void)
152 if (!test_and_set_thread_flag(TIF_NOTSC))
154 * Must flip the CPU state synchronously with
155 * TIF_NOTSC in the current running context.
161 static void hard_enable_TSC(void)
163 cr4_clear_bits(X86_CR4_TSD);
166 static void enable_TSC(void)
169 if (test_and_clear_thread_flag(TIF_NOTSC))
171 * Must flip the CPU state synchronously with
172 * TIF_NOTSC in the current running context.
178 int get_tsc_mode(unsigned long adr)
182 if (test_thread_flag(TIF_NOTSC))
183 val = PR_TSC_SIGSEGV;
187 return put_user(val, (unsigned int __user *)adr);
190 int set_tsc_mode(unsigned int val)
192 if (val == PR_TSC_SIGSEGV)
194 else if (val == PR_TSC_ENABLE)
202 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
203 struct tss_struct *tss)
205 struct thread_struct *prev, *next;
207 prev = &prev_p->thread;
208 next = &next_p->thread;
210 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
211 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
212 unsigned long debugctl = get_debugctlmsr();
214 debugctl &= ~DEBUGCTLMSR_BTF;
215 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
216 debugctl |= DEBUGCTLMSR_BTF;
218 update_debugctlmsr(debugctl);
221 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
222 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
223 /* prev and next are different */
224 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
230 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
232 * Copy the relevant range of the IO bitmap.
233 * Normally this is 128 bytes or less:
235 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
236 max(prev->io_bitmap_max, next->io_bitmap_max));
237 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
239 * Clear any possible leftover bits:
241 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
243 propagate_user_return_notify(prev_p, next_p);
247 * Idle related variables and functions
249 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
250 EXPORT_SYMBOL(boot_option_idle_override);
252 static void (*x86_idle)(void);
255 static inline void play_dead(void)
262 void enter_idle(void)
264 this_cpu_write(is_idle, 1);
265 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
268 static void __exit_idle(void)
270 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
272 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
275 /* Called from interrupts to signify idle end */
278 /* idle loop has pid 0 */
285 void arch_cpu_idle_enter(void)
291 void arch_cpu_idle_exit(void)
296 void arch_cpu_idle_dead(void)
302 * Called from the generic idle code.
304 void arch_cpu_idle(void)
310 * We use this if we don't have any better idle routine..
312 void default_idle(void)
314 trace_cpu_idle_rcuidle(1, smp_processor_id());
316 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
318 #ifdef CONFIG_APM_MODULE
319 EXPORT_SYMBOL(default_idle);
323 bool xen_set_default_idle(void)
325 bool ret = !!x86_idle;
327 x86_idle = default_idle;
332 void stop_this_cpu(void *dummy)
338 set_cpu_online(smp_processor_id(), false);
339 disable_local_APIC();
345 bool amd_e400_c1e_detected;
346 EXPORT_SYMBOL(amd_e400_c1e_detected);
348 static cpumask_var_t amd_e400_c1e_mask;
350 void amd_e400_remove_cpu(int cpu)
352 if (amd_e400_c1e_mask != NULL)
353 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
357 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
358 * pending message MSR. If we detect C1E, then we handle it the same
359 * way as C3 power states (local apic timer and TSC stop)
361 static void amd_e400_idle(void)
363 if (!amd_e400_c1e_detected) {
366 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
368 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
369 amd_e400_c1e_detected = true;
370 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
371 mark_tsc_unstable("TSC halt in AMD C1E");
372 pr_info("System has AMD C1E enabled\n");
376 if (amd_e400_c1e_detected) {
377 int cpu = smp_processor_id();
379 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
380 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
382 * Force broadcast so ACPI can not interfere.
384 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
386 pr_info("Switch to broadcast mode on CPU%d\n", cpu);
388 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
393 * The switch back from broadcast mode needs to be
394 * called with interrupts disabled.
397 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
404 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
405 * We can't rely on cpuidle installing MWAIT, because it will not load
406 * on systems that support only C1 -- so the boot default must be MWAIT.
408 * Some AMD machines are the opposite, they depend on using HALT.
410 * So for default C1, which is used during boot until cpuidle loads,
411 * use MWAIT-C1 on Intel HW that has it, else use HALT.
413 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
415 if (c->x86_vendor != X86_VENDOR_INTEL)
418 if (!cpu_has(c, X86_FEATURE_MWAIT))
425 * MONITOR/MWAIT with no hints, used for default default C1 state.
426 * This invokes MWAIT with interrutps enabled and no flags,
427 * which is backwards compatible with the original MWAIT implementation.
430 static void mwait_idle(void)
432 if (!current_set_polling_and_test()) {
433 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
434 smp_mb(); /* quirk */
435 clflush((void *)¤t_thread_info()->flags);
436 smp_mb(); /* quirk */
439 __monitor((void *)¤t_thread_info()->flags, 0, 0);
447 __current_clr_polling();
450 void select_idle_routine(const struct cpuinfo_x86 *c)
453 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
454 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
456 if (x86_idle || boot_option_idle_override == IDLE_POLL)
459 if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
460 /* E400: APIC timer interrupt does not wake up CPU from C1e */
461 pr_info("using AMD E400 aware idle routine\n");
462 x86_idle = amd_e400_idle;
463 } else if (prefer_mwait_c1_over_halt(c)) {
464 pr_info("using mwait in idle threads\n");
465 x86_idle = mwait_idle;
467 x86_idle = default_idle;
470 void __init init_amd_e400_c1e_mask(void)
472 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
473 if (x86_idle == amd_e400_idle)
474 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
477 static int __init idle_setup(char *str)
482 if (!strcmp(str, "poll")) {
483 pr_info("using polling idle threads\n");
484 boot_option_idle_override = IDLE_POLL;
485 cpu_idle_poll_ctrl(true);
486 } else if (!strcmp(str, "halt")) {
488 * When the boot option of idle=halt is added, halt is
489 * forced to be used for CPU idle. In such case CPU C2/C3
490 * won't be used again.
491 * To continue to load the CPU idle driver, don't touch
492 * the boot_option_idle_override.
494 x86_idle = default_idle;
495 boot_option_idle_override = IDLE_HALT;
496 } else if (!strcmp(str, "nomwait")) {
498 * If the boot option of "idle=nomwait" is added,
499 * it means that mwait will be disabled for CPU C2/C3
500 * states. In such case it won't touch the variable
501 * of boot_option_idle_override.
503 boot_option_idle_override = IDLE_NOMWAIT;
509 early_param("idle", idle_setup);
511 unsigned long arch_align_stack(unsigned long sp)
513 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
514 sp -= get_random_int() % 8192;
518 unsigned long arch_randomize_brk(struct mm_struct *mm)
520 unsigned long range_end = mm->brk + 0x02000000;
521 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;