1 #include <linux/errno.h>
2 #include <linux/kernel.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/module.h>
10 #include <linux/clockchips.h>
11 #include <linux/ftrace.h>
12 #include <asm/system.h>
15 unsigned long idle_halt;
16 EXPORT_SYMBOL(idle_halt);
17 unsigned long idle_nomwait;
18 EXPORT_SYMBOL(idle_nomwait);
20 struct kmem_cache *task_xstate_cachep;
22 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
25 if (src->thread.xstate) {
26 dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
28 if (!dst->thread.xstate)
30 WARN_ON((unsigned long)dst->thread.xstate & 15);
31 memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
36 void free_thread_xstate(struct task_struct *tsk)
38 if (tsk->thread.xstate) {
39 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
40 tsk->thread.xstate = NULL;
44 void free_thread_info(struct thread_info *ti)
46 free_thread_xstate(ti->task);
47 free_pages((unsigned long)ti, get_order(THREAD_SIZE));
50 void arch_task_cache_init(void)
53 kmem_cache_create("task_xstate", xstate_size,
54 __alignof__(union thread_xstate),
59 * Idle related variables and functions
61 unsigned long boot_option_idle_override = 0;
62 EXPORT_SYMBOL(boot_option_idle_override);
65 * Powermanagement idle function, if any..
67 void (*pm_idle)(void);
68 EXPORT_SYMBOL(pm_idle);
72 * This halt magic was a workaround for ancient floppy DMA
73 * wreckage. It should be safe to remove.
75 static int hlt_counter;
76 void disable_hlt(void)
80 EXPORT_SYMBOL(disable_hlt);
86 EXPORT_SYMBOL(enable_hlt);
88 static inline int hlt_use_halt(void)
90 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
93 static inline int hlt_use_halt(void)
100 * We use this if we don't have any better
103 void default_idle(void)
105 if (hlt_use_halt()) {
106 struct power_trace it;
108 trace_power_start(&it, POWER_CSTATE, 1);
109 current_thread_info()->status &= ~TS_POLLING;
111 * TS_POLLING-cleared state must be visible before we
117 safe_halt(); /* enables interrupts racelessly */
120 current_thread_info()->status |= TS_POLLING;
121 trace_power_end(&it);
124 /* loop is done by the caller */
128 #ifdef CONFIG_APM_MODULE
129 EXPORT_SYMBOL(default_idle);
132 void stop_this_cpu(void *dummy)
138 cpu_clear(smp_processor_id(), cpu_online_map);
139 disable_local_APIC();
142 if (hlt_works(smp_processor_id()))
147 static void do_nothing(void *unused)
152 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
153 * pm_idle and update to new pm_idle value. Required while changing pm_idle
154 * handler on SMP systems.
156 * Caller must have changed pm_idle to the new value before the call. Old
157 * pm_idle value will not be used by any CPU after the return of this function.
159 void cpu_idle_wait(void)
162 /* kick all the CPUs so that they exit out of pm_idle */
163 smp_call_function(do_nothing, NULL, 1);
165 EXPORT_SYMBOL_GPL(cpu_idle_wait);
168 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
169 * which can obviate IPI to trigger checking of need_resched.
170 * We execute MONITOR against need_resched and enter optimized wait state
171 * through MWAIT. Whenever someone changes need_resched, we would be woken
172 * up from MWAIT (without an IPI).
174 * New with Core Duo processors, MWAIT can take some hints based on CPU
177 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
179 struct power_trace it;
181 trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
182 if (!need_resched()) {
183 __monitor((void *)¤t_thread_info()->flags, 0, 0);
188 trace_power_end(&it);
191 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
192 static void mwait_idle(void)
194 struct power_trace it;
195 if (!need_resched()) {
196 trace_power_start(&it, POWER_CSTATE, 1);
197 __monitor((void *)¤t_thread_info()->flags, 0, 0);
203 trace_power_end(&it);
209 * On SMP it's slightly faster (but much more power-consuming!)
210 * to poll the ->work.need_resched flag instead of waiting for the
211 * cross-CPU IPI to arrive. Use this option with caution.
213 static void poll_idle(void)
215 struct power_trace it;
217 trace_power_start(&it, POWER_CSTATE, 0);
219 while (!need_resched())
221 trace_power_end(&it);
225 * mwait selection logic:
227 * It depends on the CPU. For AMD CPUs that support MWAIT this is
228 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
229 * then depend on a clock divisor and current Pstate of the core. If
230 * all cores of a processor are in halt state (C1) the processor can
231 * enter the C1E (C1 enhanced) state. If mwait is used this will never
234 * idle=mwait overrides this decision and forces the usage of mwait.
236 static int __cpuinitdata force_mwait;
238 #define MWAIT_INFO 0x05
239 #define MWAIT_ECX_EXTENDED_INFO 0x01
240 #define MWAIT_EDX_C1 0xf0
242 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
244 u32 eax, ebx, ecx, edx;
249 if (c->cpuid_level < MWAIT_INFO)
252 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
253 /* Check, whether EDX has extended info about MWAIT */
254 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
258 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
261 return (edx & MWAIT_EDX_C1);
265 * Check for AMD CPUs, which have potentially C1E support
267 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
269 if (c->x86_vendor != X86_VENDOR_AMD)
275 /* Family 0x0f models < rev F do not have C1E */
276 if (c->x86 == 0x0f && c->x86_model < 0x40)
282 static cpumask_t c1e_mask = CPU_MASK_NONE;
283 static int c1e_detected;
285 void c1e_remove_cpu(int cpu)
287 cpu_clear(cpu, c1e_mask);
291 * C1E aware idle routine. We check for C1E active in the interrupt
292 * pending message MSR. If we detect C1E, then we handle it the same
293 * way as C3 power states (local apic timer and TSC stop)
295 static void c1e_idle(void)
303 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
304 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
306 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
307 mark_tsc_unstable("TSC halt in AMD C1E");
308 printk(KERN_INFO "System has AMD C1E enabled\n");
309 set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
314 int cpu = smp_processor_id();
316 if (!cpu_isset(cpu, c1e_mask)) {
317 cpu_set(cpu, c1e_mask);
319 * Force broadcast so ACPI can not interfere. Needs
320 * to run with interrupts enabled as it uses
324 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
326 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
330 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
335 * The switch back from broadcast mode needs to be
336 * called with interrupts disabled.
339 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
345 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
347 #ifdef CONFIG_X86_SMP
348 if (pm_idle == poll_idle && smp_num_siblings > 1) {
349 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
350 " performance may degrade.\n");
356 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
358 * One CPU supports mwait => All CPUs supports mwait
360 printk(KERN_INFO "using mwait in idle threads.\n");
361 pm_idle = mwait_idle;
362 } else if (check_c1e_idle(c)) {
363 printk(KERN_INFO "using C1E aware idle routine\n");
366 pm_idle = default_idle;
369 static int __init idle_setup(char *str)
374 if (!strcmp(str, "poll")) {
375 printk("using polling idle threads.\n");
377 } else if (!strcmp(str, "mwait"))
379 else if (!strcmp(str, "halt")) {
381 * When the boot option of idle=halt is added, halt is
382 * forced to be used for CPU idle. In such case CPU C2/C3
383 * won't be used again.
384 * To continue to load the CPU idle driver, don't touch
385 * the boot_option_idle_override.
387 pm_idle = default_idle;
390 } else if (!strcmp(str, "nomwait")) {
392 * If the boot option of "idle=nomwait" is added,
393 * it means that mwait will be disabled for CPU C2/C3
394 * states. In such case it won't touch the variable
395 * of boot_option_idle_override.
402 boot_option_idle_override = 1;
405 early_param("idle", idle_setup);