1 #include <linux/errno.h>
2 #include <linux/kernel.h>
5 #include <linux/prctl.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/module.h>
10 #include <linux/clockchips.h>
11 #include <linux/random.h>
12 #include <trace/events/power.h>
13 #include <asm/system.h>
15 #include <asm/syscalls.h>
17 #include <asm/uaccess.h>
21 unsigned long idle_halt;
22 EXPORT_SYMBOL(idle_halt);
23 unsigned long idle_nomwait;
24 EXPORT_SYMBOL(idle_nomwait);
26 struct kmem_cache *task_xstate_cachep;
28 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
31 if (src->thread.xstate) {
32 dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
34 if (!dst->thread.xstate)
36 WARN_ON((unsigned long)dst->thread.xstate & 15);
37 memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
42 void free_thread_xstate(struct task_struct *tsk)
44 if (tsk->thread.xstate) {
45 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
46 tsk->thread.xstate = NULL;
49 WARN(tsk->thread.ds_ctx, "leaking DS context\n");
52 void free_thread_info(struct thread_info *ti)
54 free_thread_xstate(ti->task);
55 free_pages((unsigned long)ti, get_order(THREAD_SIZE));
58 void arch_task_cache_init(void)
61 kmem_cache_create("task_xstate", xstate_size,
62 __alignof__(union thread_xstate),
63 SLAB_PANIC | SLAB_NOTRACK, NULL);
67 * Free current thread data structures etc..
69 void exit_thread(void)
71 struct task_struct *me = current;
72 struct thread_struct *t = &me->thread;
73 unsigned long *bp = t->io_bitmap_ptr;
76 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
78 t->io_bitmap_ptr = NULL;
79 clear_thread_flag(TIF_IO_BITMAP);
81 * Careful, clear this in the TSS too:
83 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
90 void flush_thread(void)
92 struct task_struct *tsk = current;
94 clear_tsk_thread_flag(tsk, TIF_DEBUG);
96 tsk->thread.debugreg0 = 0;
97 tsk->thread.debugreg1 = 0;
98 tsk->thread.debugreg2 = 0;
99 tsk->thread.debugreg3 = 0;
100 tsk->thread.debugreg6 = 0;
101 tsk->thread.debugreg7 = 0;
102 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
104 * Forget coprocessor state..
106 tsk->fpu_counter = 0;
111 static void hard_disable_TSC(void)
113 write_cr4(read_cr4() | X86_CR4_TSD);
116 void disable_TSC(void)
119 if (!test_and_set_thread_flag(TIF_NOTSC))
121 * Must flip the CPU state synchronously with
122 * TIF_NOTSC in the current running context.
128 static void hard_enable_TSC(void)
130 write_cr4(read_cr4() & ~X86_CR4_TSD);
133 static void enable_TSC(void)
136 if (test_and_clear_thread_flag(TIF_NOTSC))
138 * Must flip the CPU state synchronously with
139 * TIF_NOTSC in the current running context.
145 int get_tsc_mode(unsigned long adr)
149 if (test_thread_flag(TIF_NOTSC))
150 val = PR_TSC_SIGSEGV;
154 return put_user(val, (unsigned int __user *)adr);
157 int set_tsc_mode(unsigned int val)
159 if (val == PR_TSC_SIGSEGV)
161 else if (val == PR_TSC_ENABLE)
169 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
170 struct tss_struct *tss)
172 struct thread_struct *prev, *next;
174 prev = &prev_p->thread;
175 next = &next_p->thread;
177 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
178 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
179 ds_switch_to(prev_p, next_p);
180 else if (next->debugctlmsr != prev->debugctlmsr)
181 update_debugctlmsr(next->debugctlmsr);
183 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
184 set_debugreg(next->debugreg0, 0);
185 set_debugreg(next->debugreg1, 1);
186 set_debugreg(next->debugreg2, 2);
187 set_debugreg(next->debugreg3, 3);
189 set_debugreg(next->debugreg6, 6);
190 set_debugreg(next->debugreg7, 7);
193 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
194 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
195 /* prev and next are different */
196 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
202 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
204 * Copy the relevant range of the IO bitmap.
205 * Normally this is 128 bytes or less:
207 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
208 max(prev->io_bitmap_max, next->io_bitmap_max));
209 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
211 * Clear any possible leftover bits:
213 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
217 int sys_fork(struct pt_regs *regs)
219 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
223 * This is trivial, and on the face of it looks like it
224 * could equally well be done in user mode.
226 * Not so, for quite unobvious reasons - register pressure.
227 * In user mode vfork() cannot have a stack frame, and if
228 * done by calling the "clone()" system call directly, you
229 * do not have enough call-clobbered registers to hold all
230 * the information you need.
232 int sys_vfork(struct pt_regs *regs)
234 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
240 * Idle related variables and functions
242 unsigned long boot_option_idle_override = 0;
243 EXPORT_SYMBOL(boot_option_idle_override);
246 * Powermanagement idle function, if any..
248 void (*pm_idle)(void);
249 EXPORT_SYMBOL(pm_idle);
253 * This halt magic was a workaround for ancient floppy DMA
254 * wreckage. It should be safe to remove.
256 static int hlt_counter;
257 void disable_hlt(void)
261 EXPORT_SYMBOL(disable_hlt);
263 void enable_hlt(void)
267 EXPORT_SYMBOL(enable_hlt);
269 static inline int hlt_use_halt(void)
271 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
274 static inline int hlt_use_halt(void)
281 * We use this if we don't have any better
284 void default_idle(void)
286 if (hlt_use_halt()) {
287 trace_power_start(POWER_CSTATE, 1);
288 current_thread_info()->status &= ~TS_POLLING;
290 * TS_POLLING-cleared state must be visible before we
296 safe_halt(); /* enables interrupts racelessly */
299 current_thread_info()->status |= TS_POLLING;
302 /* loop is done by the caller */
306 #ifdef CONFIG_APM_MODULE
307 EXPORT_SYMBOL(default_idle);
310 void stop_this_cpu(void *dummy)
316 set_cpu_online(smp_processor_id(), false);
317 disable_local_APIC();
320 if (hlt_works(smp_processor_id()))
325 static void do_nothing(void *unused)
330 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
331 * pm_idle and update to new pm_idle value. Required while changing pm_idle
332 * handler on SMP systems.
334 * Caller must have changed pm_idle to the new value before the call. Old
335 * pm_idle value will not be used by any CPU after the return of this function.
337 void cpu_idle_wait(void)
340 /* kick all the CPUs so that they exit out of pm_idle */
341 smp_call_function(do_nothing, NULL, 1);
343 EXPORT_SYMBOL_GPL(cpu_idle_wait);
346 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
347 * which can obviate IPI to trigger checking of need_resched.
348 * We execute MONITOR against need_resched and enter optimized wait state
349 * through MWAIT. Whenever someone changes need_resched, we would be woken
350 * up from MWAIT (without an IPI).
352 * New with Core Duo processors, MWAIT can take some hints based on CPU
355 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
357 trace_power_start(POWER_CSTATE, (ax>>4)+1);
358 if (!need_resched()) {
359 if (cpu_has(¤t_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
360 clflush((void *)¤t_thread_info()->flags);
362 __monitor((void *)¤t_thread_info()->flags, 0, 0);
369 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
370 static void mwait_idle(void)
372 if (!need_resched()) {
373 trace_power_start(POWER_CSTATE, 1);
374 if (cpu_has(¤t_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
375 clflush((void *)¤t_thread_info()->flags);
377 __monitor((void *)¤t_thread_info()->flags, 0, 0);
388 * On SMP it's slightly faster (but much more power-consuming!)
389 * to poll the ->work.need_resched flag instead of waiting for the
390 * cross-CPU IPI to arrive. Use this option with caution.
392 static void poll_idle(void)
394 trace_power_start(POWER_CSTATE, 0);
396 while (!need_resched())
402 * mwait selection logic:
404 * It depends on the CPU. For AMD CPUs that support MWAIT this is
405 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
406 * then depend on a clock divisor and current Pstate of the core. If
407 * all cores of a processor are in halt state (C1) the processor can
408 * enter the C1E (C1 enhanced) state. If mwait is used this will never
411 * idle=mwait overrides this decision and forces the usage of mwait.
413 static int __cpuinitdata force_mwait;
415 #define MWAIT_INFO 0x05
416 #define MWAIT_ECX_EXTENDED_INFO 0x01
417 #define MWAIT_EDX_C1 0xf0
419 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
421 u32 eax, ebx, ecx, edx;
426 if (c->cpuid_level < MWAIT_INFO)
429 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
430 /* Check, whether EDX has extended info about MWAIT */
431 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
435 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
438 return (edx & MWAIT_EDX_C1);
442 * Check for AMD CPUs, where APIC timer interrupt does not wake up CPU from C1e.
443 * For more information see
444 * - Erratum #400 for NPT family 0xf and family 0x10 CPUs
445 * - Erratum #365 for family 0x11 (not affected because C1e not in use)
447 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
450 if (c->x86_vendor != X86_VENDOR_AMD)
453 /* Family 0x0f models < rev F do not have C1E */
454 if (c->x86 == 0x0F && c->x86_model >= 0x40)
457 if (c->x86 == 0x10) {
459 * check OSVW bit for CPUs that are not affected
462 if (cpu_has(c, X86_FEATURE_OSVW)) {
463 rdmsrl(MSR_AMD64_OSVW_ID_LENGTH, val);
465 rdmsrl(MSR_AMD64_OSVW_STATUS, val);
477 static cpumask_var_t c1e_mask;
478 static int c1e_detected;
480 void c1e_remove_cpu(int cpu)
482 if (c1e_mask != NULL)
483 cpumask_clear_cpu(cpu, c1e_mask);
487 * C1E aware idle routine. We check for C1E active in the interrupt
488 * pending message MSR. If we detect C1E, then we handle it the same
489 * way as C3 power states (local apic timer and TSC stop)
491 static void c1e_idle(void)
499 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
500 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
502 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
503 mark_tsc_unstable("TSC halt in AMD C1E");
504 printk(KERN_INFO "System has AMD C1E enabled\n");
505 set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
510 int cpu = smp_processor_id();
512 if (!cpumask_test_cpu(cpu, c1e_mask)) {
513 cpumask_set_cpu(cpu, c1e_mask);
515 * Force broadcast so ACPI can not interfere.
517 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
519 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
522 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
527 * The switch back from broadcast mode needs to be
528 * called with interrupts disabled.
531 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
537 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
540 if (pm_idle == poll_idle && smp_num_siblings > 1) {
541 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
542 " performance may degrade.\n");
548 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
550 * One CPU supports mwait => All CPUs supports mwait
552 printk(KERN_INFO "using mwait in idle threads.\n");
553 pm_idle = mwait_idle;
554 } else if (check_c1e_idle(c)) {
555 printk(KERN_INFO "using C1E aware idle routine\n");
558 pm_idle = default_idle;
561 void __init init_c1e_mask(void)
563 /* If we're using c1e_idle, we need to allocate c1e_mask. */
564 if (pm_idle == c1e_idle)
565 zalloc_cpumask_var(&c1e_mask, GFP_KERNEL);
568 static int __init idle_setup(char *str)
573 if (!strcmp(str, "poll")) {
574 printk("using polling idle threads.\n");
576 } else if (!strcmp(str, "mwait"))
578 else if (!strcmp(str, "halt")) {
580 * When the boot option of idle=halt is added, halt is
581 * forced to be used for CPU idle. In such case CPU C2/C3
582 * won't be used again.
583 * To continue to load the CPU idle driver, don't touch
584 * the boot_option_idle_override.
586 pm_idle = default_idle;
589 } else if (!strcmp(str, "nomwait")) {
591 * If the boot option of "idle=nomwait" is added,
592 * it means that mwait will be disabled for CPU C2/C3
593 * states. In such case it won't touch the variable
594 * of boot_option_idle_override.
601 boot_option_idle_override = 1;
604 early_param("idle", idle_setup);
606 unsigned long arch_align_stack(unsigned long sp)
608 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
609 sp -= get_random_int() % 8192;
613 unsigned long arch_randomize_brk(struct mm_struct *mm)
615 unsigned long range_end = mm->brk + 0x02000000;
616 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;