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/tick.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>
28 #include <asm/fpu/internal.h>
29 #include <asm/debugreg.h>
31 #include <asm/tlbflush.h>
36 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
37 * no more per-task TSS's. The TSS size is kept cacheline-aligned
38 * so they are allowed to end up in the .data..cacheline_aligned
39 * section. Since TSS's are completely CPU-local, we want them
40 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
42 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
44 .sp0 = TOP_OF_INIT_STACK,
48 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
53 * Note that the .io_bitmap member must be extra-big. This is because
54 * the CPU will access an additional byte beyond the end of the IO
55 * permission bitmap. The extra byte must be all 1 bits, and must
56 * be within the limit.
58 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 },
61 EXPORT_PER_CPU_SYMBOL(cpu_tss);
64 static DEFINE_PER_CPU(unsigned char, is_idle);
65 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
67 void idle_notifier_register(struct notifier_block *n)
69 atomic_notifier_chain_register(&idle_notifier, n);
71 EXPORT_SYMBOL_GPL(idle_notifier_register);
73 void idle_notifier_unregister(struct notifier_block *n)
75 atomic_notifier_chain_unregister(&idle_notifier, n);
77 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
81 * this gets called so that we can store lazy state into memory and copy the
82 * current task into the new thread.
84 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
86 memcpy(dst, src, arch_task_struct_size);
88 return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
92 * Free current thread data structures etc..
94 void exit_thread(void)
96 struct task_struct *me = current;
97 struct thread_struct *t = &me->thread;
98 unsigned long *bp = t->io_bitmap_ptr;
99 struct fpu *fpu = &t->fpu;
102 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
104 t->io_bitmap_ptr = NULL;
105 clear_thread_flag(TIF_IO_BITMAP);
107 * Careful, clear this in the TSS too:
109 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
110 t->io_bitmap_max = 0;
120 void flush_thread(void)
122 struct task_struct *tsk = current;
124 flush_ptrace_hw_breakpoint(tsk);
125 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
127 fpu__clear(&tsk->thread.fpu);
130 static void hard_disable_TSC(void)
132 cr4_set_bits(X86_CR4_TSD);
135 void disable_TSC(void)
138 if (!test_and_set_thread_flag(TIF_NOTSC))
140 * Must flip the CPU state synchronously with
141 * TIF_NOTSC in the current running context.
147 static void hard_enable_TSC(void)
149 cr4_clear_bits(X86_CR4_TSD);
152 static void enable_TSC(void)
155 if (test_and_clear_thread_flag(TIF_NOTSC))
157 * Must flip the CPU state synchronously with
158 * TIF_NOTSC in the current running context.
164 int get_tsc_mode(unsigned long adr)
168 if (test_thread_flag(TIF_NOTSC))
169 val = PR_TSC_SIGSEGV;
173 return put_user(val, (unsigned int __user *)adr);
176 int set_tsc_mode(unsigned int val)
178 if (val == PR_TSC_SIGSEGV)
180 else if (val == PR_TSC_ENABLE)
188 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
189 struct tss_struct *tss)
191 struct thread_struct *prev, *next;
193 prev = &prev_p->thread;
194 next = &next_p->thread;
196 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
197 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
198 unsigned long debugctl = get_debugctlmsr();
200 debugctl &= ~DEBUGCTLMSR_BTF;
201 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
202 debugctl |= DEBUGCTLMSR_BTF;
204 update_debugctlmsr(debugctl);
207 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
208 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
209 /* prev and next are different */
210 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
216 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
218 * Copy the relevant range of the IO bitmap.
219 * Normally this is 128 bytes or less:
221 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
222 max(prev->io_bitmap_max, next->io_bitmap_max));
223 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
225 * Clear any possible leftover bits:
227 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
229 propagate_user_return_notify(prev_p, next_p);
233 * Idle related variables and functions
235 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
236 EXPORT_SYMBOL(boot_option_idle_override);
238 static void (*x86_idle)(void);
241 static inline void play_dead(void)
248 void enter_idle(void)
250 this_cpu_write(is_idle, 1);
251 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
254 static void __exit_idle(void)
256 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
258 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
261 /* Called from interrupts to signify idle end */
264 /* idle loop has pid 0 */
271 void arch_cpu_idle_enter(void)
277 void arch_cpu_idle_exit(void)
282 void arch_cpu_idle_dead(void)
288 * Called from the generic idle code.
290 void arch_cpu_idle(void)
296 * We use this if we don't have any better idle routine..
298 void default_idle(void)
300 trace_cpu_idle_rcuidle(1, smp_processor_id());
302 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
304 #ifdef CONFIG_APM_MODULE
305 EXPORT_SYMBOL(default_idle);
309 bool xen_set_default_idle(void)
311 bool ret = !!x86_idle;
313 x86_idle = default_idle;
318 void stop_this_cpu(void *dummy)
324 set_cpu_online(smp_processor_id(), false);
325 disable_local_APIC();
326 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
332 bool amd_e400_c1e_detected;
333 EXPORT_SYMBOL(amd_e400_c1e_detected);
335 static cpumask_var_t amd_e400_c1e_mask;
337 void amd_e400_remove_cpu(int cpu)
339 if (amd_e400_c1e_mask != NULL)
340 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
344 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
345 * pending message MSR. If we detect C1E, then we handle it the same
346 * way as C3 power states (local apic timer and TSC stop)
348 static void amd_e400_idle(void)
350 if (!amd_e400_c1e_detected) {
353 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
355 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
356 amd_e400_c1e_detected = true;
357 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
358 mark_tsc_unstable("TSC halt in AMD C1E");
359 pr_info("System has AMD C1E enabled\n");
363 if (amd_e400_c1e_detected) {
364 int cpu = smp_processor_id();
366 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
367 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
368 /* Force broadcast so ACPI can not interfere. */
369 tick_broadcast_force();
370 pr_info("Switch to broadcast mode on CPU%d\n", cpu);
372 tick_broadcast_enter();
377 * The switch back from broadcast mode needs to be
378 * called with interrupts disabled.
381 tick_broadcast_exit();
388 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
389 * We can't rely on cpuidle installing MWAIT, because it will not load
390 * on systems that support only C1 -- so the boot default must be MWAIT.
392 * Some AMD machines are the opposite, they depend on using HALT.
394 * So for default C1, which is used during boot until cpuidle loads,
395 * use MWAIT-C1 on Intel HW that has it, else use HALT.
397 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
399 if (c->x86_vendor != X86_VENDOR_INTEL)
402 if (!cpu_has(c, X86_FEATURE_MWAIT))
409 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
410 * with interrupts enabled and no flags, which is backwards compatible with the
411 * original MWAIT implementation.
413 static void mwait_idle(void)
415 if (!current_set_polling_and_test()) {
416 trace_cpu_idle_rcuidle(1, smp_processor_id());
417 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
418 smp_mb(); /* quirk */
419 clflush((void *)¤t_thread_info()->flags);
420 smp_mb(); /* quirk */
423 __monitor((void *)¤t_thread_info()->flags, 0, 0);
428 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
432 __current_clr_polling();
435 void select_idle_routine(const struct cpuinfo_x86 *c)
438 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
439 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
441 if (x86_idle || boot_option_idle_override == IDLE_POLL)
444 if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
445 /* E400: APIC timer interrupt does not wake up CPU from C1e */
446 pr_info("using AMD E400 aware idle routine\n");
447 x86_idle = amd_e400_idle;
448 } else if (prefer_mwait_c1_over_halt(c)) {
449 pr_info("using mwait in idle threads\n");
450 x86_idle = mwait_idle;
452 x86_idle = default_idle;
455 void __init init_amd_e400_c1e_mask(void)
457 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
458 if (x86_idle == amd_e400_idle)
459 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
462 static int __init idle_setup(char *str)
467 if (!strcmp(str, "poll")) {
468 pr_info("using polling idle threads\n");
469 boot_option_idle_override = IDLE_POLL;
470 cpu_idle_poll_ctrl(true);
471 } else if (!strcmp(str, "halt")) {
473 * When the boot option of idle=halt is added, halt is
474 * forced to be used for CPU idle. In such case CPU C2/C3
475 * won't be used again.
476 * To continue to load the CPU idle driver, don't touch
477 * the boot_option_idle_override.
479 x86_idle = default_idle;
480 boot_option_idle_override = IDLE_HALT;
481 } else if (!strcmp(str, "nomwait")) {
483 * If the boot option of "idle=nomwait" is added,
484 * it means that mwait will be disabled for CPU C2/C3
485 * states. In such case it won't touch the variable
486 * of boot_option_idle_override.
488 boot_option_idle_override = IDLE_NOMWAIT;
494 early_param("idle", idle_setup);
496 unsigned long arch_align_stack(unsigned long sp)
498 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
499 sp -= get_random_int() % 8192;
503 unsigned long arch_randomize_brk(struct mm_struct *mm)
505 unsigned long range_end = mm->brk + 0x02000000;
506 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;