x86 idle: APM requires pm_idle/default_idle unconditionally when a module
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / process.c
blobe1ba8cb24e4edb6629a9107312c071739c32de75
1 #include <linux/errno.h>
2 #include <linux/kernel.h>
3 #include <linux/mm.h>
4 #include <linux/smp.h>
5 #include <linux/prctl.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/module.h>
9 #include <linux/pm.h>
10 #include <linux/clockchips.h>
11 #include <linux/random.h>
12 #include <linux/user-return-notifier.h>
13 #include <linux/dmi.h>
14 #include <linux/utsname.h>
15 #include <trace/events/power.h>
16 #include <linux/hw_breakpoint.h>
17 #include <asm/cpu.h>
18 #include <asm/system.h>
19 #include <asm/apic.h>
20 #include <asm/syscalls.h>
21 #include <asm/idle.h>
22 #include <asm/uaccess.h>
23 #include <asm/i387.h>
24 #include <asm/debugreg.h>
26 struct kmem_cache *task_xstate_cachep;
27 EXPORT_SYMBOL_GPL(task_xstate_cachep);
29 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
31 int ret;
33 *dst = *src;
34 if (fpu_allocated(&src->thread.fpu)) {
35 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
36 ret = fpu_alloc(&dst->thread.fpu);
37 if (ret)
38 return ret;
39 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
41 return 0;
44 void free_thread_xstate(struct task_struct *tsk)
46 fpu_free(&tsk->thread.fpu);
49 void free_thread_info(struct thread_info *ti)
51 free_thread_xstate(ti->task);
52 free_pages((unsigned long)ti, get_order(THREAD_SIZE));
55 void arch_task_cache_init(void)
57 task_xstate_cachep =
58 kmem_cache_create("task_xstate", xstate_size,
59 __alignof__(union thread_xstate),
60 SLAB_PANIC | SLAB_NOTRACK, NULL);
64 * Free current thread data structures etc..
66 void exit_thread(void)
68 struct task_struct *me = current;
69 struct thread_struct *t = &me->thread;
70 unsigned long *bp = t->io_bitmap_ptr;
72 if (bp) {
73 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
75 t->io_bitmap_ptr = NULL;
76 clear_thread_flag(TIF_IO_BITMAP);
78 * Careful, clear this in the TSS too:
80 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
81 t->io_bitmap_max = 0;
82 put_cpu();
83 kfree(bp);
87 void show_regs(struct pt_regs *regs)
89 show_registers(regs);
90 show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs), 0);
93 void show_regs_common(void)
95 const char *vendor, *product, *board;
97 vendor = dmi_get_system_info(DMI_SYS_VENDOR);
98 if (!vendor)
99 vendor = "";
100 product = dmi_get_system_info(DMI_PRODUCT_NAME);
101 if (!product)
102 product = "";
104 /* Board Name is optional */
105 board = dmi_get_system_info(DMI_BOARD_NAME);
107 printk(KERN_CONT "\n");
108 printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s",
109 current->pid, current->comm, print_tainted(),
110 init_utsname()->release,
111 (int)strcspn(init_utsname()->version, " "),
112 init_utsname()->version);
113 printk(KERN_CONT " %s %s", vendor, product);
114 if (board)
115 printk(KERN_CONT "/%s", board);
116 printk(KERN_CONT "\n");
119 void flush_thread(void)
121 struct task_struct *tsk = current;
123 flush_ptrace_hw_breakpoint(tsk);
124 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
126 * Forget coprocessor state..
128 tsk->fpu_counter = 0;
129 clear_fpu(tsk);
130 clear_used_math();
133 static void hard_disable_TSC(void)
135 write_cr4(read_cr4() | X86_CR4_TSD);
138 void disable_TSC(void)
140 preempt_disable();
141 if (!test_and_set_thread_flag(TIF_NOTSC))
143 * Must flip the CPU state synchronously with
144 * TIF_NOTSC in the current running context.
146 hard_disable_TSC();
147 preempt_enable();
150 static void hard_enable_TSC(void)
152 write_cr4(read_cr4() & ~X86_CR4_TSD);
155 static void enable_TSC(void)
157 preempt_disable();
158 if (test_and_clear_thread_flag(TIF_NOTSC))
160 * Must flip the CPU state synchronously with
161 * TIF_NOTSC in the current running context.
163 hard_enable_TSC();
164 preempt_enable();
167 int get_tsc_mode(unsigned long adr)
169 unsigned int val;
171 if (test_thread_flag(TIF_NOTSC))
172 val = PR_TSC_SIGSEGV;
173 else
174 val = PR_TSC_ENABLE;
176 return put_user(val, (unsigned int __user *)adr);
179 int set_tsc_mode(unsigned int val)
181 if (val == PR_TSC_SIGSEGV)
182 disable_TSC();
183 else if (val == PR_TSC_ENABLE)
184 enable_TSC();
185 else
186 return -EINVAL;
188 return 0;
191 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
192 struct tss_struct *tss)
194 struct thread_struct *prev, *next;
196 prev = &prev_p->thread;
197 next = &next_p->thread;
199 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
200 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
201 unsigned long debugctl = get_debugctlmsr();
203 debugctl &= ~DEBUGCTLMSR_BTF;
204 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
205 debugctl |= DEBUGCTLMSR_BTF;
207 update_debugctlmsr(debugctl);
210 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
211 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
212 /* prev and next are different */
213 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
214 hard_disable_TSC();
215 else
216 hard_enable_TSC();
219 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
221 * Copy the relevant range of the IO bitmap.
222 * Normally this is 128 bytes or less:
224 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
225 max(prev->io_bitmap_max, next->io_bitmap_max));
226 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
228 * Clear any possible leftover bits:
230 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
232 propagate_user_return_notify(prev_p, next_p);
235 int sys_fork(struct pt_regs *regs)
237 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
241 * This is trivial, and on the face of it looks like it
242 * could equally well be done in user mode.
244 * Not so, for quite unobvious reasons - register pressure.
245 * In user mode vfork() cannot have a stack frame, and if
246 * done by calling the "clone()" system call directly, you
247 * do not have enough call-clobbered registers to hold all
248 * the information you need.
250 int sys_vfork(struct pt_regs *regs)
252 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
253 NULL, NULL);
256 long
257 sys_clone(unsigned long clone_flags, unsigned long newsp,
258 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
260 if (!newsp)
261 newsp = regs->sp;
262 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
266 * This gets run with %si containing the
267 * function to call, and %di containing
268 * the "args".
270 extern void kernel_thread_helper(void);
273 * Create a kernel thread
275 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
277 struct pt_regs regs;
279 memset(&regs, 0, sizeof(regs));
281 regs.si = (unsigned long) fn;
282 regs.di = (unsigned long) arg;
284 #ifdef CONFIG_X86_32
285 regs.ds = __USER_DS;
286 regs.es = __USER_DS;
287 regs.fs = __KERNEL_PERCPU;
288 regs.gs = __KERNEL_STACK_CANARY;
289 #else
290 regs.ss = __KERNEL_DS;
291 #endif
293 regs.orig_ax = -1;
294 regs.ip = (unsigned long) kernel_thread_helper;
295 regs.cs = __KERNEL_CS | get_kernel_rpl();
296 regs.flags = X86_EFLAGS_IF | 0x2;
298 /* Ok, create the new process.. */
299 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
301 EXPORT_SYMBOL(kernel_thread);
304 * sys_execve() executes a new program.
306 long sys_execve(const char __user *name,
307 const char __user *const __user *argv,
308 const char __user *const __user *envp, struct pt_regs *regs)
310 long error;
311 char *filename;
313 filename = getname(name);
314 error = PTR_ERR(filename);
315 if (IS_ERR(filename))
316 return error;
317 error = do_execve(filename, argv, envp, regs);
319 #ifdef CONFIG_X86_32
320 if (error == 0) {
321 /* Make sure we don't return using sysenter.. */
322 set_thread_flag(TIF_IRET);
324 #endif
326 putname(filename);
327 return error;
331 * Idle related variables and functions
333 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
334 EXPORT_SYMBOL(boot_option_idle_override);
337 * Powermanagement idle function, if any..
339 void (*pm_idle)(void);
340 #ifdef CONFIG_APM_MODULE
341 EXPORT_SYMBOL(pm_idle);
342 #endif
344 #ifdef CONFIG_X86_32
346 * This halt magic was a workaround for ancient floppy DMA
347 * wreckage. It should be safe to remove.
349 static int hlt_counter;
350 void disable_hlt(void)
352 hlt_counter++;
354 EXPORT_SYMBOL(disable_hlt);
356 void enable_hlt(void)
358 hlt_counter--;
360 EXPORT_SYMBOL(enable_hlt);
362 static inline int hlt_use_halt(void)
364 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
366 #else
367 static inline int hlt_use_halt(void)
369 return 1;
371 #endif
374 * We use this if we don't have any better
375 * idle routine..
377 void default_idle(void)
379 if (hlt_use_halt()) {
380 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
381 trace_cpu_idle(1, smp_processor_id());
382 current_thread_info()->status &= ~TS_POLLING;
384 * TS_POLLING-cleared state must be visible before we
385 * test NEED_RESCHED:
387 smp_mb();
389 if (!need_resched())
390 safe_halt(); /* enables interrupts racelessly */
391 else
392 local_irq_enable();
393 current_thread_info()->status |= TS_POLLING;
394 trace_power_end(smp_processor_id());
395 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
396 } else {
397 local_irq_enable();
398 /* loop is done by the caller */
399 cpu_relax();
402 #ifdef CONFIG_APM_MODULE
403 EXPORT_SYMBOL(default_idle);
404 #endif
406 void stop_this_cpu(void *dummy)
408 local_irq_disable();
410 * Remove this CPU:
412 set_cpu_online(smp_processor_id(), false);
413 disable_local_APIC();
415 for (;;) {
416 if (hlt_works(smp_processor_id()))
417 halt();
421 static void do_nothing(void *unused)
426 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
427 * pm_idle and update to new pm_idle value. Required while changing pm_idle
428 * handler on SMP systems.
430 * Caller must have changed pm_idle to the new value before the call. Old
431 * pm_idle value will not be used by any CPU after the return of this function.
433 void cpu_idle_wait(void)
435 smp_mb();
436 /* kick all the CPUs so that they exit out of pm_idle */
437 smp_call_function(do_nothing, NULL, 1);
439 EXPORT_SYMBOL_GPL(cpu_idle_wait);
442 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
443 * which can obviate IPI to trigger checking of need_resched.
444 * We execute MONITOR against need_resched and enter optimized wait state
445 * through MWAIT. Whenever someone changes need_resched, we would be woken
446 * up from MWAIT (without an IPI).
448 * New with Core Duo processors, MWAIT can take some hints based on CPU
449 * capability.
451 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
453 if (!need_resched()) {
454 if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR))
455 clflush((void *)&current_thread_info()->flags);
457 __monitor((void *)&current_thread_info()->flags, 0, 0);
458 smp_mb();
459 if (!need_resched())
460 __mwait(ax, cx);
464 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
465 static void mwait_idle(void)
467 if (!need_resched()) {
468 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
469 trace_cpu_idle(1, smp_processor_id());
470 if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR))
471 clflush((void *)&current_thread_info()->flags);
473 __monitor((void *)&current_thread_info()->flags, 0, 0);
474 smp_mb();
475 if (!need_resched())
476 __sti_mwait(0, 0);
477 else
478 local_irq_enable();
479 trace_power_end(smp_processor_id());
480 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
481 } else
482 local_irq_enable();
486 * On SMP it's slightly faster (but much more power-consuming!)
487 * to poll the ->work.need_resched flag instead of waiting for the
488 * cross-CPU IPI to arrive. Use this option with caution.
490 static void poll_idle(void)
492 trace_power_start(POWER_CSTATE, 0, smp_processor_id());
493 trace_cpu_idle(0, smp_processor_id());
494 local_irq_enable();
495 while (!need_resched())
496 cpu_relax();
497 trace_power_end(smp_processor_id());
498 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
502 * mwait selection logic:
504 * It depends on the CPU. For AMD CPUs that support MWAIT this is
505 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
506 * then depend on a clock divisor and current Pstate of the core. If
507 * all cores of a processor are in halt state (C1) the processor can
508 * enter the C1E (C1 enhanced) state. If mwait is used this will never
509 * happen.
511 * idle=mwait overrides this decision and forces the usage of mwait.
514 #define MWAIT_INFO 0x05
515 #define MWAIT_ECX_EXTENDED_INFO 0x01
516 #define MWAIT_EDX_C1 0xf0
518 int mwait_usable(const struct cpuinfo_x86 *c)
520 u32 eax, ebx, ecx, edx;
522 if (boot_option_idle_override == IDLE_FORCE_MWAIT)
523 return 1;
525 if (c->cpuid_level < MWAIT_INFO)
526 return 0;
528 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
529 /* Check, whether EDX has extended info about MWAIT */
530 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
531 return 1;
534 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
535 * C1 supports MWAIT
537 return (edx & MWAIT_EDX_C1);
540 bool amd_e400_c1e_detected;
541 EXPORT_SYMBOL(amd_e400_c1e_detected);
543 static cpumask_var_t amd_e400_c1e_mask;
545 void amd_e400_remove_cpu(int cpu)
547 if (amd_e400_c1e_mask != NULL)
548 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
552 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
553 * pending message MSR. If we detect C1E, then we handle it the same
554 * way as C3 power states (local apic timer and TSC stop)
556 static void amd_e400_idle(void)
558 if (need_resched())
559 return;
561 if (!amd_e400_c1e_detected) {
562 u32 lo, hi;
564 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
566 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
567 amd_e400_c1e_detected = true;
568 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
569 mark_tsc_unstable("TSC halt in AMD C1E");
570 printk(KERN_INFO "System has AMD C1E enabled\n");
574 if (amd_e400_c1e_detected) {
575 int cpu = smp_processor_id();
577 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
578 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
580 * Force broadcast so ACPI can not interfere.
582 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
583 &cpu);
584 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
585 cpu);
587 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
589 default_idle();
592 * The switch back from broadcast mode needs to be
593 * called with interrupts disabled.
595 local_irq_disable();
596 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
597 local_irq_enable();
598 } else
599 default_idle();
602 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
604 #ifdef CONFIG_SMP
605 if (pm_idle == poll_idle && smp_num_siblings > 1) {
606 printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
607 " performance may degrade.\n");
609 #endif
610 if (pm_idle)
611 return;
613 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
615 * One CPU supports mwait => All CPUs supports mwait
617 printk(KERN_INFO "using mwait in idle threads.\n");
618 pm_idle = mwait_idle;
619 } else if (cpu_has_amd_erratum(amd_erratum_400)) {
620 /* E400: APIC timer interrupt does not wake up CPU from C1e */
621 printk(KERN_INFO "using AMD E400 aware idle routine\n");
622 pm_idle = amd_e400_idle;
623 } else
624 pm_idle = default_idle;
627 void __init init_amd_e400_c1e_mask(void)
629 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
630 if (pm_idle == amd_e400_idle)
631 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
634 static int __init idle_setup(char *str)
636 if (!str)
637 return -EINVAL;
639 if (!strcmp(str, "poll")) {
640 printk("using polling idle threads.\n");
641 pm_idle = poll_idle;
642 boot_option_idle_override = IDLE_POLL;
643 } else if (!strcmp(str, "mwait")) {
644 boot_option_idle_override = IDLE_FORCE_MWAIT;
645 WARN_ONCE(1, "\"idle=mwait\" will be removed in 2012\n");
646 } else if (!strcmp(str, "halt")) {
648 * When the boot option of idle=halt is added, halt is
649 * forced to be used for CPU idle. In such case CPU C2/C3
650 * won't be used again.
651 * To continue to load the CPU idle driver, don't touch
652 * the boot_option_idle_override.
654 pm_idle = default_idle;
655 boot_option_idle_override = IDLE_HALT;
656 } else if (!strcmp(str, "nomwait")) {
658 * If the boot option of "idle=nomwait" is added,
659 * it means that mwait will be disabled for CPU C2/C3
660 * states. In such case it won't touch the variable
661 * of boot_option_idle_override.
663 boot_option_idle_override = IDLE_NOMWAIT;
664 } else
665 return -1;
667 return 0;
669 early_param("idle", idle_setup);
671 unsigned long arch_align_stack(unsigned long sp)
673 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
674 sp -= get_random_int() % 8192;
675 return sp & ~0xf;
678 unsigned long arch_randomize_brk(struct mm_struct *mm)
680 unsigned long range_end = mm->brk + 0x02000000;
681 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;