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>
28 #include <asm/fpu-internal.h>
29 #include <asm/debugreg.h>
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.
39 DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss) = INIT_TSS;
42 static DEFINE_PER_CPU(unsigned char, is_idle);
43 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
45 void idle_notifier_register(struct notifier_block *n)
47 atomic_notifier_chain_register(&idle_notifier, n);
49 EXPORT_SYMBOL_GPL(idle_notifier_register);
51 void idle_notifier_unregister(struct notifier_block *n)
53 atomic_notifier_chain_unregister(&idle_notifier, n);
55 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
58 struct kmem_cache *task_xstate_cachep;
59 EXPORT_SYMBOL_GPL(task_xstate_cachep);
62 * this gets called so that we can store lazy state into memory and copy the
63 * current task into the new thread.
65 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
72 if (fpu_allocated(&src->thread.fpu)) {
73 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
74 ret = fpu_alloc(&dst->thread.fpu);
77 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
82 void free_thread_xstate(struct task_struct *tsk)
84 fpu_free(&tsk->thread.fpu);
87 void arch_release_task_struct(struct task_struct *tsk)
89 free_thread_xstate(tsk);
92 void arch_task_cache_init(void)
95 kmem_cache_create("task_xstate", xstate_size,
96 __alignof__(union thread_xstate),
97 SLAB_PANIC | SLAB_NOTRACK, NULL);
100 static inline void drop_fpu(struct task_struct *tsk)
103 * Forget coprocessor state..
105 tsk->fpu_counter = 0;
111 * Free current thread data structures etc..
113 void exit_thread(void)
115 struct task_struct *me = current;
116 struct thread_struct *t = &me->thread;
117 unsigned long *bp = t->io_bitmap_ptr;
120 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
122 t->io_bitmap_ptr = NULL;
123 clear_thread_flag(TIF_IO_BITMAP);
125 * Careful, clear this in the TSS too:
127 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
128 t->io_bitmap_max = 0;
136 void show_regs_common(void)
138 const char *vendor, *product, *board;
140 vendor = dmi_get_system_info(DMI_SYS_VENDOR);
143 product = dmi_get_system_info(DMI_PRODUCT_NAME);
147 /* Board Name is optional */
148 board = dmi_get_system_info(DMI_BOARD_NAME);
150 printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s %s %s%s%s\n",
151 current->pid, current->comm, print_tainted(),
152 init_utsname()->release,
153 (int)strcspn(init_utsname()->version, " "),
154 init_utsname()->version,
160 void flush_thread(void)
162 struct task_struct *tsk = current;
164 flush_ptrace_hw_breakpoint(tsk);
165 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
169 static void hard_disable_TSC(void)
171 write_cr4(read_cr4() | X86_CR4_TSD);
174 void disable_TSC(void)
177 if (!test_and_set_thread_flag(TIF_NOTSC))
179 * Must flip the CPU state synchronously with
180 * TIF_NOTSC in the current running context.
186 static void hard_enable_TSC(void)
188 write_cr4(read_cr4() & ~X86_CR4_TSD);
191 static void enable_TSC(void)
194 if (test_and_clear_thread_flag(TIF_NOTSC))
196 * Must flip the CPU state synchronously with
197 * TIF_NOTSC in the current running context.
203 int get_tsc_mode(unsigned long adr)
207 if (test_thread_flag(TIF_NOTSC))
208 val = PR_TSC_SIGSEGV;
212 return put_user(val, (unsigned int __user *)adr);
215 int set_tsc_mode(unsigned int val)
217 if (val == PR_TSC_SIGSEGV)
219 else if (val == PR_TSC_ENABLE)
227 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
228 struct tss_struct *tss)
230 struct thread_struct *prev, *next;
232 prev = &prev_p->thread;
233 next = &next_p->thread;
235 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
236 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
237 unsigned long debugctl = get_debugctlmsr();
239 debugctl &= ~DEBUGCTLMSR_BTF;
240 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
241 debugctl |= DEBUGCTLMSR_BTF;
243 update_debugctlmsr(debugctl);
246 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
247 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
248 /* prev and next are different */
249 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
255 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
257 * Copy the relevant range of the IO bitmap.
258 * Normally this is 128 bytes or less:
260 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
261 max(prev->io_bitmap_max, next->io_bitmap_max));
262 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
264 * Clear any possible leftover bits:
266 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
268 propagate_user_return_notify(prev_p, next_p);
271 int sys_fork(struct pt_regs *regs)
273 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
277 * This is trivial, and on the face of it looks like it
278 * could equally well be done in user mode.
280 * Not so, for quite unobvious reasons - register pressure.
281 * In user mode vfork() cannot have a stack frame, and if
282 * done by calling the "clone()" system call directly, you
283 * do not have enough call-clobbered registers to hold all
284 * the information you need.
286 int sys_vfork(struct pt_regs *regs)
288 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
293 sys_clone(unsigned long clone_flags, unsigned long newsp,
294 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
298 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
302 * This gets run with %si containing the
303 * function to call, and %di containing
306 extern void kernel_thread_helper(void);
309 * Create a kernel thread
311 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
315 memset(®s, 0, sizeof(regs));
317 regs.si = (unsigned long) fn;
318 regs.di = (unsigned long) arg;
323 regs.fs = __KERNEL_PERCPU;
324 regs.gs = __KERNEL_STACK_CANARY;
326 regs.ss = __KERNEL_DS;
330 regs.ip = (unsigned long) kernel_thread_helper;
331 regs.cs = __KERNEL_CS | get_kernel_rpl();
332 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_BIT1;
334 /* Ok, create the new process.. */
335 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
337 EXPORT_SYMBOL(kernel_thread);
340 * sys_execve() executes a new program.
342 long sys_execve(const char __user *name,
343 const char __user *const __user *argv,
344 const char __user *const __user *envp, struct pt_regs *regs)
349 filename = getname(name);
350 error = PTR_ERR(filename);
351 if (IS_ERR(filename))
353 error = do_execve(filename, argv, envp, regs);
357 /* Make sure we don't return using sysenter.. */
358 set_thread_flag(TIF_IRET);
367 * Idle related variables and functions
369 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
370 EXPORT_SYMBOL(boot_option_idle_override);
373 * Powermanagement idle function, if any..
375 void (*pm_idle)(void);
376 #ifdef CONFIG_APM_MODULE
377 EXPORT_SYMBOL(pm_idle);
380 static inline int hlt_use_halt(void)
386 static inline void play_dead(void)
393 void enter_idle(void)
395 this_cpu_write(is_idle, 1);
396 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
399 static void __exit_idle(void)
401 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
403 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
406 /* Called from interrupts to signify idle end */
409 /* idle loop has pid 0 */
417 * The idle thread. There's no useful work to be
418 * done, so just try to conserve power and have a
419 * low exit latency (ie sit in a loop waiting for
420 * somebody to say that they'd like to reschedule)
425 * If we're the non-boot CPU, nothing set the stack canary up
426 * for us. CPU0 already has it initialized but no harm in
427 * doing it again. This is a good place for updating it, as
428 * we wont ever return from this function (so the invalid
429 * canaries already on the stack wont ever trigger).
431 boot_init_stack_canary();
432 current_thread_info()->status |= TS_POLLING;
435 tick_nohz_idle_enter();
437 while (!need_resched()) {
440 if (cpu_is_offline(smp_processor_id()))
444 * Idle routines should keep interrupts disabled
445 * from here on, until they go to idle.
446 * Otherwise, idle callbacks can misfire.
453 /* Don't trace irqs off for idle */
454 stop_critical_timings();
456 /* enter_idle() needs rcu for notifiers */
459 if (cpuidle_idle_call())
463 start_critical_timings();
465 /* In many cases the interrupt that ended idle
466 has already called exit_idle. But some idle
467 loops can be woken up without interrupt. */
471 tick_nohz_idle_exit();
472 preempt_enable_no_resched();
479 * We use this if we don't have any better
482 void default_idle(void)
484 if (hlt_use_halt()) {
485 trace_power_start_rcuidle(POWER_CSTATE, 1, smp_processor_id());
486 trace_cpu_idle_rcuidle(1, smp_processor_id());
487 current_thread_info()->status &= ~TS_POLLING;
489 * TS_POLLING-cleared state must be visible before we
495 safe_halt(); /* enables interrupts racelessly */
498 current_thread_info()->status |= TS_POLLING;
499 trace_power_end_rcuidle(smp_processor_id());
500 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
503 /* loop is done by the caller */
507 #ifdef CONFIG_APM_MODULE
508 EXPORT_SYMBOL(default_idle);
511 bool set_pm_idle_to_default(void)
513 bool ret = !!pm_idle;
515 pm_idle = default_idle;
519 void stop_this_cpu(void *dummy)
525 set_cpu_online(smp_processor_id(), false);
526 disable_local_APIC();
529 if (hlt_works(smp_processor_id()))
534 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
535 static void mwait_idle(void)
537 if (!need_resched()) {
538 trace_power_start_rcuidle(POWER_CSTATE, 1, smp_processor_id());
539 trace_cpu_idle_rcuidle(1, smp_processor_id());
540 if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR))
541 clflush((void *)¤t_thread_info()->flags);
543 __monitor((void *)¤t_thread_info()->flags, 0, 0);
549 trace_power_end_rcuidle(smp_processor_id());
550 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
556 * On SMP it's slightly faster (but much more power-consuming!)
557 * to poll the ->work.need_resched flag instead of waiting for the
558 * cross-CPU IPI to arrive. Use this option with caution.
560 static void poll_idle(void)
562 trace_power_start_rcuidle(POWER_CSTATE, 0, smp_processor_id());
563 trace_cpu_idle_rcuidle(0, smp_processor_id());
565 while (!need_resched())
567 trace_power_end_rcuidle(smp_processor_id());
568 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
572 * mwait selection logic:
574 * It depends on the CPU. For AMD CPUs that support MWAIT this is
575 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
576 * then depend on a clock divisor and current Pstate of the core. If
577 * all cores of a processor are in halt state (C1) the processor can
578 * enter the C1E (C1 enhanced) state. If mwait is used this will never
581 * idle=mwait overrides this decision and forces the usage of mwait.
584 #define MWAIT_INFO 0x05
585 #define MWAIT_ECX_EXTENDED_INFO 0x01
586 #define MWAIT_EDX_C1 0xf0
588 int mwait_usable(const struct cpuinfo_x86 *c)
590 u32 eax, ebx, ecx, edx;
592 /* Use mwait if idle=mwait boot option is given */
593 if (boot_option_idle_override == IDLE_FORCE_MWAIT)
597 * Any idle= boot option other than idle=mwait means that we must not
598 * use mwait. Eg: idle=halt or idle=poll or idle=nomwait
600 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
603 if (c->cpuid_level < MWAIT_INFO)
606 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
607 /* Check, whether EDX has extended info about MWAIT */
608 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
612 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
615 return (edx & MWAIT_EDX_C1);
618 bool amd_e400_c1e_detected;
619 EXPORT_SYMBOL(amd_e400_c1e_detected);
621 static cpumask_var_t amd_e400_c1e_mask;
623 void amd_e400_remove_cpu(int cpu)
625 if (amd_e400_c1e_mask != NULL)
626 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
630 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
631 * pending message MSR. If we detect C1E, then we handle it the same
632 * way as C3 power states (local apic timer and TSC stop)
634 static void amd_e400_idle(void)
639 if (!amd_e400_c1e_detected) {
642 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
644 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
645 amd_e400_c1e_detected = true;
646 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
647 mark_tsc_unstable("TSC halt in AMD C1E");
648 pr_info("System has AMD C1E enabled\n");
652 if (amd_e400_c1e_detected) {
653 int cpu = smp_processor_id();
655 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
656 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
658 * Force broadcast so ACPI can not interfere.
660 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
662 pr_info("Switch to broadcast mode on CPU%d\n", cpu);
664 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
669 * The switch back from broadcast mode needs to be
670 * called with interrupts disabled.
673 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
679 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
682 if (pm_idle == poll_idle && smp_num_siblings > 1) {
683 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
689 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
691 * One CPU supports mwait => All CPUs supports mwait
693 pr_info("using mwait in idle threads\n");
694 pm_idle = mwait_idle;
695 } else if (cpu_has_amd_erratum(amd_erratum_400)) {
696 /* E400: APIC timer interrupt does not wake up CPU from C1e */
697 pr_info("using AMD E400 aware idle routine\n");
698 pm_idle = amd_e400_idle;
700 pm_idle = default_idle;
703 void __init init_amd_e400_c1e_mask(void)
705 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
706 if (pm_idle == amd_e400_idle)
707 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
710 static int __init idle_setup(char *str)
715 if (!strcmp(str, "poll")) {
716 pr_info("using polling idle threads\n");
718 boot_option_idle_override = IDLE_POLL;
719 } else if (!strcmp(str, "mwait")) {
720 boot_option_idle_override = IDLE_FORCE_MWAIT;
721 WARN_ONCE(1, "\"idle=mwait\" will be removed in 2012\n");
722 } else if (!strcmp(str, "halt")) {
724 * When the boot option of idle=halt is added, halt is
725 * forced to be used for CPU idle. In such case CPU C2/C3
726 * won't be used again.
727 * To continue to load the CPU idle driver, don't touch
728 * the boot_option_idle_override.
730 pm_idle = default_idle;
731 boot_option_idle_override = IDLE_HALT;
732 } else if (!strcmp(str, "nomwait")) {
734 * If the boot option of "idle=nomwait" is added,
735 * it means that mwait will be disabled for CPU C2/C3
736 * states. In such case it won't touch the variable
737 * of boot_option_idle_override.
739 boot_option_idle_override = IDLE_NOMWAIT;
745 early_param("idle", idle_setup);
747 unsigned long arch_align_stack(unsigned long sp)
749 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
750 sp -= get_random_int() % 8192;
754 unsigned long arch_randomize_brk(struct mm_struct *mm)
756 unsigned long range_end = mm->brk + 0x02000000;
757 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;