1 #include <linux/errno.h>
2 #include <linux/kernel.h>
5 #include <linux/slab.h>
6 #include <linux/sched.h>
7 #include <linux/module.h>
9 #include <linux/clockchips.h>
10 #include <asm/system.h>
12 unsigned long idle_halt
;
13 EXPORT_SYMBOL(idle_halt
);
15 struct kmem_cache
*task_xstate_cachep
;
17 int arch_dup_task_struct(struct task_struct
*dst
, struct task_struct
*src
)
20 if (src
->thread
.xstate
) {
21 dst
->thread
.xstate
= kmem_cache_alloc(task_xstate_cachep
,
23 if (!dst
->thread
.xstate
)
25 WARN_ON((unsigned long)dst
->thread
.xstate
& 15);
26 memcpy(dst
->thread
.xstate
, src
->thread
.xstate
, xstate_size
);
31 void free_thread_xstate(struct task_struct
*tsk
)
33 if (tsk
->thread
.xstate
) {
34 kmem_cache_free(task_xstate_cachep
, tsk
->thread
.xstate
);
35 tsk
->thread
.xstate
= NULL
;
39 void free_thread_info(struct thread_info
*ti
)
41 free_thread_xstate(ti
->task
);
42 free_pages((unsigned long)ti
, get_order(THREAD_SIZE
));
45 void arch_task_cache_init(void)
48 kmem_cache_create("task_xstate", xstate_size
,
49 __alignof__(union thread_xstate
),
54 * Idle related variables and functions
56 unsigned long boot_option_idle_override
= 0;
57 EXPORT_SYMBOL(boot_option_idle_override
);
60 * Powermanagement idle function, if any..
62 void (*pm_idle
)(void);
63 EXPORT_SYMBOL(pm_idle
);
67 * This halt magic was a workaround for ancient floppy DMA
68 * wreckage. It should be safe to remove.
70 static int hlt_counter
;
71 void disable_hlt(void)
75 EXPORT_SYMBOL(disable_hlt
);
81 EXPORT_SYMBOL(enable_hlt
);
83 static inline int hlt_use_halt(void)
85 return (!hlt_counter
&& boot_cpu_data
.hlt_works_ok
);
88 static inline int hlt_use_halt(void)
95 * We use this if we don't have any better
98 void default_idle(void)
100 if (hlt_use_halt()) {
101 current_thread_info()->status
&= ~TS_POLLING
;
103 * TS_POLLING-cleared state must be visible before we
109 safe_halt(); /* enables interrupts racelessly */
112 current_thread_info()->status
|= TS_POLLING
;
115 /* loop is done by the caller */
119 #ifdef CONFIG_APM_MODULE
120 EXPORT_SYMBOL(default_idle
);
123 static void do_nothing(void *unused
)
128 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
129 * pm_idle and update to new pm_idle value. Required while changing pm_idle
130 * handler on SMP systems.
132 * Caller must have changed pm_idle to the new value before the call. Old
133 * pm_idle value will not be used by any CPU after the return of this function.
135 void cpu_idle_wait(void)
138 /* kick all the CPUs so that they exit out of pm_idle */
139 smp_call_function(do_nothing
, NULL
, 1);
141 EXPORT_SYMBOL_GPL(cpu_idle_wait
);
144 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
145 * which can obviate IPI to trigger checking of need_resched.
146 * We execute MONITOR against need_resched and enter optimized wait state
147 * through MWAIT. Whenever someone changes need_resched, we would be woken
148 * up from MWAIT (without an IPI).
150 * New with Core Duo processors, MWAIT can take some hints based on CPU
153 void mwait_idle_with_hints(unsigned long ax
, unsigned long cx
)
155 if (!need_resched()) {
156 __monitor((void *)¤t_thread_info()->flags
, 0, 0);
163 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
164 static void mwait_idle(void)
166 if (!need_resched()) {
167 __monitor((void *)¤t_thread_info()->flags
, 0, 0);
178 * On SMP it's slightly faster (but much more power-consuming!)
179 * to poll the ->work.need_resched flag instead of waiting for the
180 * cross-CPU IPI to arrive. Use this option with caution.
182 static void poll_idle(void)
189 * mwait selection logic:
191 * It depends on the CPU. For AMD CPUs that support MWAIT this is
192 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
193 * then depend on a clock divisor and current Pstate of the core. If
194 * all cores of a processor are in halt state (C1) the processor can
195 * enter the C1E (C1 enhanced) state. If mwait is used this will never
198 * idle=mwait overrides this decision and forces the usage of mwait.
201 #define MWAIT_INFO 0x05
202 #define MWAIT_ECX_EXTENDED_INFO 0x01
203 #define MWAIT_EDX_C1 0xf0
205 static int __cpuinit
mwait_usable(const struct cpuinfo_x86
*c
)
207 u32 eax
, ebx
, ecx
, edx
;
212 if (c
->cpuid_level
< MWAIT_INFO
)
215 cpuid(MWAIT_INFO
, &eax
, &ebx
, &ecx
, &edx
);
216 /* Check, whether EDX has extended info about MWAIT */
217 if (!(ecx
& MWAIT_ECX_EXTENDED_INFO
))
221 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
224 return (edx
& MWAIT_EDX_C1
);
228 * Check for AMD CPUs, which have potentially C1E support
230 static int __cpuinit
check_c1e_idle(const struct cpuinfo_x86
*c
)
232 if (c
->x86_vendor
!= X86_VENDOR_AMD
)
238 /* Family 0x0f models < rev F do not have C1E */
239 if (c
->x86
== 0x0f && c
->x86_model
< 0x40)
246 * C1E aware idle routine. We check for C1E active in the interrupt
247 * pending message MSR. If we detect C1E, then we handle it the same
248 * way as C3 power states (local apic timer and TSC stop)
250 static void c1e_idle(void)
252 static cpumask_t c1e_mask
= CPU_MASK_NONE
;
253 static int c1e_detected
;
261 rdmsr(MSR_K8_INT_PENDING_MSG
, lo
, hi
);
262 if (lo
& K8_INTP_C1E_ACTIVE_MASK
) {
264 mark_tsc_unstable("TSC halt in C1E");
265 printk(KERN_INFO
"System has C1E enabled\n");
270 int cpu
= smp_processor_id();
272 if (!cpu_isset(cpu
, c1e_mask
)) {
273 cpu_set(cpu
, c1e_mask
);
275 * Force broadcast so ACPI can not interfere. Needs
276 * to run with interrupts enabled as it uses
280 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE
,
282 printk(KERN_INFO
"Switch to broadcast mode on CPU%d\n",
286 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER
, &cpu
);
291 * The switch back from broadcast mode needs to be
292 * called with interrupts disabled.
295 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT
, &cpu
);
301 void __cpuinit
select_idle_routine(const struct cpuinfo_x86
*c
)
303 #ifdef CONFIG_X86_SMP
304 if (pm_idle
== poll_idle
&& smp_num_siblings
> 1) {
305 printk(KERN_WARNING
"WARNING: polling idle and HT enabled,"
306 " performance may degrade.\n");
312 if (cpu_has(c
, X86_FEATURE_MWAIT
) && mwait_usable(c
)) {
314 * One CPU supports mwait => All CPUs supports mwait
316 printk(KERN_INFO
"using mwait in idle threads.\n");
317 pm_idle
= mwait_idle
;
318 } else if (check_c1e_idle(c
)) {
319 printk(KERN_INFO
"using C1E aware idle routine\n");
322 pm_idle
= default_idle
;
325 static int __init
idle_setup(char *str
)
327 if (!strcmp(str
, "poll")) {
328 printk("using polling idle threads.\n");
330 } else if (!strcmp(str
, "mwait"))
332 else if (!strcmp(str
, "halt")) {
334 * When the boot option of idle=halt is added, halt is
335 * forced to be used for CPU idle. In such case CPU C2/C3
336 * won't be used again.
337 * To continue to load the CPU idle driver, don't touch
338 * the boot_option_idle_override.
340 pm_idle
= default_idle
;
346 boot_option_idle_override
= 1;
349 early_param("idle", idle_setup
);