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/init.h>
11 #include <linux/export.h>
13 #include <linux/tick.h>
14 #include <linux/random.h>
15 #include <linux/user-return-notifier.h>
16 #include <linux/dmi.h>
17 #include <linux/utsname.h>
18 #include <linux/stackprotector.h>
19 #include <linux/tick.h>
20 #include <linux/cpuidle.h>
21 #include <trace/events/power.h>
22 #include <linux/hw_breakpoint.h>
25 #include <asm/syscalls.h>
27 #include <asm/uaccess.h>
28 #include <asm/mwait.h>
29 #include <asm/fpu/internal.h>
30 #include <asm/debugreg.h>
32 #include <asm/tlbflush.h>
37 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
38 * no more per-task TSS's. The TSS size is kept cacheline-aligned
39 * so they are allowed to end up in the .data..cacheline_aligned
40 * section. Since TSS's are completely CPU-local, we want them
41 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
43 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
45 .sp0 = TOP_OF_INIT_STACK,
49 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
54 * Note that the .io_bitmap member must be extra-big. This is because
55 * the CPU will access an additional byte beyond the end of the IO
56 * permission bitmap. The extra byte must be all 1 bits, and must
57 * be within the limit.
59 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 },
62 .SYSENTER_stack_canary = STACK_END_MAGIC,
65 EXPORT_PER_CPU_SYMBOL(cpu_tss);
68 static DEFINE_PER_CPU(unsigned char, is_idle);
69 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
71 void idle_notifier_register(struct notifier_block *n)
73 atomic_notifier_chain_register(&idle_notifier, n);
75 EXPORT_SYMBOL_GPL(idle_notifier_register);
77 void idle_notifier_unregister(struct notifier_block *n)
79 atomic_notifier_chain_unregister(&idle_notifier, n);
81 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
85 * this gets called so that we can store lazy state into memory and copy the
86 * current task into the new thread.
88 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
90 memcpy(dst, src, arch_task_struct_size);
92 dst->thread.vm86 = NULL;
95 return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
99 * Free current thread data structures etc..
101 void exit_thread(struct task_struct *tsk)
103 struct thread_struct *t = &tsk->thread;
104 unsigned long *bp = t->io_bitmap_ptr;
105 struct fpu *fpu = &t->fpu;
108 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
110 t->io_bitmap_ptr = NULL;
111 clear_thread_flag(TIF_IO_BITMAP);
113 * Careful, clear this in the TSS too:
115 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
116 t->io_bitmap_max = 0;
126 void flush_thread(void)
128 struct task_struct *tsk = current;
130 flush_ptrace_hw_breakpoint(tsk);
131 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
133 fpu__clear(&tsk->thread.fpu);
136 static void hard_disable_TSC(void)
138 cr4_set_bits(X86_CR4_TSD);
141 void disable_TSC(void)
144 if (!test_and_set_thread_flag(TIF_NOTSC))
146 * Must flip the CPU state synchronously with
147 * TIF_NOTSC in the current running context.
153 static void hard_enable_TSC(void)
155 cr4_clear_bits(X86_CR4_TSD);
158 static void enable_TSC(void)
161 if (test_and_clear_thread_flag(TIF_NOTSC))
163 * Must flip the CPU state synchronously with
164 * TIF_NOTSC in the current running context.
170 int get_tsc_mode(unsigned long adr)
174 if (test_thread_flag(TIF_NOTSC))
175 val = PR_TSC_SIGSEGV;
179 return put_user(val, (unsigned int __user *)adr);
182 int set_tsc_mode(unsigned int val)
184 if (val == PR_TSC_SIGSEGV)
186 else if (val == PR_TSC_ENABLE)
194 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
195 struct tss_struct *tss)
197 struct thread_struct *prev, *next;
199 prev = &prev_p->thread;
200 next = &next_p->thread;
202 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
203 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
204 unsigned long debugctl = get_debugctlmsr();
206 debugctl &= ~DEBUGCTLMSR_BTF;
207 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
208 debugctl |= DEBUGCTLMSR_BTF;
210 update_debugctlmsr(debugctl);
213 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
214 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
215 /* prev and next are different */
216 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
222 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
224 * Copy the relevant range of the IO bitmap.
225 * Normally this is 128 bytes or less:
227 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
228 max(prev->io_bitmap_max, next->io_bitmap_max));
229 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
231 * Clear any possible leftover bits:
233 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
235 propagate_user_return_notify(prev_p, next_p);
239 * Idle related variables and functions
241 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
242 EXPORT_SYMBOL(boot_option_idle_override);
244 static void (*x86_idle)(void);
247 static inline void play_dead(void)
254 void enter_idle(void)
256 this_cpu_write(is_idle, 1);
257 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
260 static void __exit_idle(void)
262 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
264 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
267 /* Called from interrupts to signify idle end */
270 /* idle loop has pid 0 */
277 void arch_cpu_idle_enter(void)
283 void arch_cpu_idle_exit(void)
288 void arch_cpu_idle_dead(void)
294 * Called from the generic idle code.
296 void arch_cpu_idle(void)
302 * We use this if we don't have any better idle routine..
304 void default_idle(void)
306 trace_cpu_idle_rcuidle(1, smp_processor_id());
308 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
310 #ifdef CONFIG_APM_MODULE
311 EXPORT_SYMBOL(default_idle);
315 bool xen_set_default_idle(void)
317 bool ret = !!x86_idle;
319 x86_idle = default_idle;
324 void stop_this_cpu(void *dummy)
330 set_cpu_online(smp_processor_id(), false);
331 disable_local_APIC();
332 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
338 bool amd_e400_c1e_detected;
339 EXPORT_SYMBOL(amd_e400_c1e_detected);
341 static cpumask_var_t amd_e400_c1e_mask;
343 void amd_e400_remove_cpu(int cpu)
345 if (amd_e400_c1e_mask != NULL)
346 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
350 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
351 * pending message MSR. If we detect C1E, then we handle it the same
352 * way as C3 power states (local apic timer and TSC stop)
354 static void amd_e400_idle(void)
356 if (!amd_e400_c1e_detected) {
359 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
361 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
362 amd_e400_c1e_detected = true;
363 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
364 mark_tsc_unstable("TSC halt in AMD C1E");
365 pr_info("System has AMD C1E enabled\n");
369 if (amd_e400_c1e_detected) {
370 int cpu = smp_processor_id();
372 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
373 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
374 /* Force broadcast so ACPI can not interfere. */
375 tick_broadcast_force();
376 pr_info("Switch to broadcast mode on CPU%d\n", cpu);
378 tick_broadcast_enter();
383 * The switch back from broadcast mode needs to be
384 * called with interrupts disabled.
387 tick_broadcast_exit();
394 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
395 * We can't rely on cpuidle installing MWAIT, because it will not load
396 * on systems that support only C1 -- so the boot default must be MWAIT.
398 * Some AMD machines are the opposite, they depend on using HALT.
400 * So for default C1, which is used during boot until cpuidle loads,
401 * use MWAIT-C1 on Intel HW that has it, else use HALT.
403 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
405 if (c->x86_vendor != X86_VENDOR_INTEL)
408 if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR))
415 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
416 * with interrupts enabled and no flags, which is backwards compatible with the
417 * original MWAIT implementation.
419 static void mwait_idle(void)
421 if (!current_set_polling_and_test()) {
422 trace_cpu_idle_rcuidle(1, smp_processor_id());
423 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
425 clflush((void *)¤t_thread_info()->flags);
429 __monitor((void *)¤t_thread_info()->flags, 0, 0);
434 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
438 __current_clr_polling();
441 void select_idle_routine(const struct cpuinfo_x86 *c)
444 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
445 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
447 if (x86_idle || boot_option_idle_override == IDLE_POLL)
450 if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
451 /* E400: APIC timer interrupt does not wake up CPU from C1e */
452 pr_info("using AMD E400 aware idle routine\n");
453 x86_idle = amd_e400_idle;
454 } else if (prefer_mwait_c1_over_halt(c)) {
455 pr_info("using mwait in idle threads\n");
456 x86_idle = mwait_idle;
458 x86_idle = default_idle;
461 void __init init_amd_e400_c1e_mask(void)
463 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
464 if (x86_idle == amd_e400_idle)
465 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
468 static int __init idle_setup(char *str)
473 if (!strcmp(str, "poll")) {
474 pr_info("using polling idle threads\n");
475 boot_option_idle_override = IDLE_POLL;
476 cpu_idle_poll_ctrl(true);
477 } else if (!strcmp(str, "halt")) {
479 * When the boot option of idle=halt is added, halt is
480 * forced to be used for CPU idle. In such case CPU C2/C3
481 * won't be used again.
482 * To continue to load the CPU idle driver, don't touch
483 * the boot_option_idle_override.
485 x86_idle = default_idle;
486 boot_option_idle_override = IDLE_HALT;
487 } else if (!strcmp(str, "nomwait")) {
489 * If the boot option of "idle=nomwait" is added,
490 * it means that mwait will be disabled for CPU C2/C3
491 * states. In such case it won't touch the variable
492 * of boot_option_idle_override.
494 boot_option_idle_override = IDLE_NOMWAIT;
500 early_param("idle", idle_setup);
502 unsigned long arch_align_stack(unsigned long sp)
504 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
505 sp -= get_random_int() % 8192;
509 unsigned long arch_randomize_brk(struct mm_struct *mm)
511 unsigned long range_end = mm->brk + 0x02000000;
512 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
516 * Called from fs/proc with a reference on @p to find the function
517 * which called into schedule(). This needs to be done carefully
518 * because the task might wake up and we might look at a stack
521 unsigned long get_wchan(struct task_struct *p)
523 unsigned long start, bottom, top, sp, fp, ip;
526 if (!p || p == current || p->state == TASK_RUNNING)
529 start = (unsigned long)task_stack_page(p);
534 * Layout of the stack page:
536 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
538 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
540 * ----------- bottom = start + sizeof(thread_info)
544 * The tasks stack pointer points at the location where the
545 * framepointer is stored. The data on the stack is:
546 * ... IP FP ... IP FP
548 * We need to read FP and IP, so we need to adjust the upper
549 * bound by another unsigned long.
551 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
552 top -= 2 * sizeof(unsigned long);
553 bottom = start + sizeof(struct thread_info);
555 sp = READ_ONCE(p->thread.sp);
556 if (sp < bottom || sp > top)
559 fp = READ_ONCE_NOCHECK(*(unsigned long *)sp);
561 if (fp < bottom || fp > top)
563 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
564 if (!in_sched_functions(ip))
566 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
567 } while (count++ < 16 && p->state != TASK_RUNNING);