Commit | Line | Data |
---|---|---|
61c4628b SS |
1 | #include <linux/errno.h> |
2 | #include <linux/kernel.h> | |
3 | #include <linux/mm.h> | |
4 | #include <linux/smp.h> | |
5 | #include <linux/slab.h> | |
6 | #include <linux/sched.h> | |
7f424a8b PZ |
7 | #include <linux/module.h> |
8 | #include <linux/pm.h> | |
aa276e1c | 9 | #include <linux/clockchips.h> |
c1e3b377 | 10 | #include <asm/system.h> |
d3ec5cae | 11 | #include <asm/apic.h> |
c1e3b377 ZY |
12 | |
13 | unsigned long idle_halt; | |
14 | EXPORT_SYMBOL(idle_halt); | |
da5e09a1 ZY |
15 | unsigned long idle_nomwait; |
16 | EXPORT_SYMBOL(idle_nomwait); | |
61c4628b | 17 | |
aa283f49 | 18 | struct kmem_cache *task_xstate_cachep; |
61c4628b SS |
19 | |
20 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) | |
21 | { | |
22 | *dst = *src; | |
aa283f49 SS |
23 | if (src->thread.xstate) { |
24 | dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep, | |
25 | GFP_KERNEL); | |
26 | if (!dst->thread.xstate) | |
27 | return -ENOMEM; | |
28 | WARN_ON((unsigned long)dst->thread.xstate & 15); | |
29 | memcpy(dst->thread.xstate, src->thread.xstate, xstate_size); | |
30 | } | |
61c4628b SS |
31 | return 0; |
32 | } | |
33 | ||
aa283f49 | 34 | void free_thread_xstate(struct task_struct *tsk) |
61c4628b | 35 | { |
aa283f49 SS |
36 | if (tsk->thread.xstate) { |
37 | kmem_cache_free(task_xstate_cachep, tsk->thread.xstate); | |
38 | tsk->thread.xstate = NULL; | |
39 | } | |
40 | } | |
41 | ||
aa283f49 SS |
42 | void free_thread_info(struct thread_info *ti) |
43 | { | |
44 | free_thread_xstate(ti->task); | |
1679f271 | 45 | free_pages((unsigned long)ti, get_order(THREAD_SIZE)); |
61c4628b SS |
46 | } |
47 | ||
48 | void arch_task_cache_init(void) | |
49 | { | |
50 | task_xstate_cachep = | |
51 | kmem_cache_create("task_xstate", xstate_size, | |
52 | __alignof__(union thread_xstate), | |
53 | SLAB_PANIC, NULL); | |
54 | } | |
7f424a8b | 55 | |
00dba564 TG |
56 | /* |
57 | * Idle related variables and functions | |
58 | */ | |
59 | unsigned long boot_option_idle_override = 0; | |
60 | EXPORT_SYMBOL(boot_option_idle_override); | |
61 | ||
62 | /* | |
63 | * Powermanagement idle function, if any.. | |
64 | */ | |
65 | void (*pm_idle)(void); | |
66 | EXPORT_SYMBOL(pm_idle); | |
67 | ||
68 | #ifdef CONFIG_X86_32 | |
69 | /* | |
70 | * This halt magic was a workaround for ancient floppy DMA | |
71 | * wreckage. It should be safe to remove. | |
72 | */ | |
73 | static int hlt_counter; | |
74 | void disable_hlt(void) | |
75 | { | |
76 | hlt_counter++; | |
77 | } | |
78 | EXPORT_SYMBOL(disable_hlt); | |
79 | ||
80 | void enable_hlt(void) | |
81 | { | |
82 | hlt_counter--; | |
83 | } | |
84 | EXPORT_SYMBOL(enable_hlt); | |
85 | ||
86 | static inline int hlt_use_halt(void) | |
87 | { | |
88 | return (!hlt_counter && boot_cpu_data.hlt_works_ok); | |
89 | } | |
90 | #else | |
91 | static inline int hlt_use_halt(void) | |
92 | { | |
93 | return 1; | |
94 | } | |
95 | #endif | |
96 | ||
97 | /* | |
98 | * We use this if we don't have any better | |
99 | * idle routine.. | |
100 | */ | |
101 | void default_idle(void) | |
102 | { | |
103 | if (hlt_use_halt()) { | |
104 | current_thread_info()->status &= ~TS_POLLING; | |
105 | /* | |
106 | * TS_POLLING-cleared state must be visible before we | |
107 | * test NEED_RESCHED: | |
108 | */ | |
109 | smp_mb(); | |
110 | ||
111 | if (!need_resched()) | |
112 | safe_halt(); /* enables interrupts racelessly */ | |
113 | else | |
114 | local_irq_enable(); | |
115 | current_thread_info()->status |= TS_POLLING; | |
116 | } else { | |
117 | local_irq_enable(); | |
118 | /* loop is done by the caller */ | |
119 | cpu_relax(); | |
120 | } | |
121 | } | |
122 | #ifdef CONFIG_APM_MODULE | |
123 | EXPORT_SYMBOL(default_idle); | |
124 | #endif | |
125 | ||
d3ec5cae IV |
126 | void stop_this_cpu(void *dummy) |
127 | { | |
128 | local_irq_disable(); | |
129 | /* | |
130 | * Remove this CPU: | |
131 | */ | |
132 | cpu_clear(smp_processor_id(), cpu_online_map); | |
133 | disable_local_APIC(); | |
134 | ||
135 | for (;;) { | |
136 | if (hlt_works(smp_processor_id())) | |
137 | halt(); | |
138 | } | |
139 | } | |
140 | ||
7f424a8b PZ |
141 | static void do_nothing(void *unused) |
142 | { | |
143 | } | |
144 | ||
145 | /* | |
146 | * cpu_idle_wait - Used to ensure that all the CPUs discard old value of | |
147 | * pm_idle and update to new pm_idle value. Required while changing pm_idle | |
148 | * handler on SMP systems. | |
149 | * | |
150 | * Caller must have changed pm_idle to the new value before the call. Old | |
151 | * pm_idle value will not be used by any CPU after the return of this function. | |
152 | */ | |
153 | void cpu_idle_wait(void) | |
154 | { | |
155 | smp_mb(); | |
156 | /* kick all the CPUs so that they exit out of pm_idle */ | |
127a237a | 157 | smp_call_function(do_nothing, NULL, 1); |
7f424a8b PZ |
158 | } |
159 | EXPORT_SYMBOL_GPL(cpu_idle_wait); | |
160 | ||
161 | /* | |
162 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, | |
163 | * which can obviate IPI to trigger checking of need_resched. | |
164 | * We execute MONITOR against need_resched and enter optimized wait state | |
165 | * through MWAIT. Whenever someone changes need_resched, we would be woken | |
166 | * up from MWAIT (without an IPI). | |
167 | * | |
168 | * New with Core Duo processors, MWAIT can take some hints based on CPU | |
169 | * capability. | |
170 | */ | |
171 | void mwait_idle_with_hints(unsigned long ax, unsigned long cx) | |
172 | { | |
173 | if (!need_resched()) { | |
174 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
175 | smp_mb(); | |
176 | if (!need_resched()) | |
177 | __mwait(ax, cx); | |
178 | } | |
179 | } | |
180 | ||
181 | /* Default MONITOR/MWAIT with no hints, used for default C1 state */ | |
182 | static void mwait_idle(void) | |
183 | { | |
184 | if (!need_resched()) { | |
185 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
186 | smp_mb(); | |
187 | if (!need_resched()) | |
188 | __sti_mwait(0, 0); | |
189 | else | |
190 | local_irq_enable(); | |
191 | } else | |
192 | local_irq_enable(); | |
193 | } | |
194 | ||
7f424a8b PZ |
195 | /* |
196 | * On SMP it's slightly faster (but much more power-consuming!) | |
197 | * to poll the ->work.need_resched flag instead of waiting for the | |
198 | * cross-CPU IPI to arrive. Use this option with caution. | |
199 | */ | |
200 | static void poll_idle(void) | |
201 | { | |
202 | local_irq_enable(); | |
2c7e9fd4 JK |
203 | while (!need_resched()) |
204 | cpu_relax(); | |
7f424a8b PZ |
205 | } |
206 | ||
e9623b35 TG |
207 | /* |
208 | * mwait selection logic: | |
209 | * | |
210 | * It depends on the CPU. For AMD CPUs that support MWAIT this is | |
211 | * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings | |
212 | * then depend on a clock divisor and current Pstate of the core. If | |
213 | * all cores of a processor are in halt state (C1) the processor can | |
214 | * enter the C1E (C1 enhanced) state. If mwait is used this will never | |
215 | * happen. | |
216 | * | |
217 | * idle=mwait overrides this decision and forces the usage of mwait. | |
218 | */ | |
08ad8afa | 219 | static int __cpuinitdata force_mwait; |
09fd4b4e TG |
220 | |
221 | #define MWAIT_INFO 0x05 | |
222 | #define MWAIT_ECX_EXTENDED_INFO 0x01 | |
223 | #define MWAIT_EDX_C1 0xf0 | |
224 | ||
e9623b35 TG |
225 | static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c) |
226 | { | |
09fd4b4e TG |
227 | u32 eax, ebx, ecx, edx; |
228 | ||
e9623b35 TG |
229 | if (force_mwait) |
230 | return 1; | |
231 | ||
09fd4b4e TG |
232 | if (c->cpuid_level < MWAIT_INFO) |
233 | return 0; | |
234 | ||
235 | cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx); | |
236 | /* Check, whether EDX has extended info about MWAIT */ | |
237 | if (!(ecx & MWAIT_ECX_EXTENDED_INFO)) | |
238 | return 1; | |
239 | ||
240 | /* | |
241 | * edx enumeratios MONITOR/MWAIT extensions. Check, whether | |
242 | * C1 supports MWAIT | |
243 | */ | |
244 | return (edx & MWAIT_EDX_C1); | |
e9623b35 TG |
245 | } |
246 | ||
aa276e1c TG |
247 | /* |
248 | * Check for AMD CPUs, which have potentially C1E support | |
249 | */ | |
250 | static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c) | |
251 | { | |
252 | if (c->x86_vendor != X86_VENDOR_AMD) | |
253 | return 0; | |
254 | ||
255 | if (c->x86 < 0x0F) | |
256 | return 0; | |
257 | ||
258 | /* Family 0x0f models < rev F do not have C1E */ | |
259 | if (c->x86 == 0x0f && c->x86_model < 0x40) | |
260 | return 0; | |
261 | ||
262 | return 1; | |
263 | } | |
264 | ||
4faac97d TG |
265 | static cpumask_t c1e_mask = CPU_MASK_NONE; |
266 | static int c1e_detected; | |
267 | ||
268 | void c1e_remove_cpu(int cpu) | |
269 | { | |
270 | cpu_clear(cpu, c1e_mask); | |
271 | } | |
272 | ||
aa276e1c TG |
273 | /* |
274 | * C1E aware idle routine. We check for C1E active in the interrupt | |
275 | * pending message MSR. If we detect C1E, then we handle it the same | |
276 | * way as C3 power states (local apic timer and TSC stop) | |
277 | */ | |
278 | static void c1e_idle(void) | |
279 | { | |
aa276e1c TG |
280 | if (need_resched()) |
281 | return; | |
282 | ||
283 | if (!c1e_detected) { | |
284 | u32 lo, hi; | |
285 | ||
286 | rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); | |
287 | if (lo & K8_INTP_C1E_ACTIVE_MASK) { | |
288 | c1e_detected = 1; | |
09bfeea1 AH |
289 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
290 | mark_tsc_unstable("TSC halt in AMD C1E"); | |
291 | printk(KERN_INFO "System has AMD C1E enabled\n"); | |
a8d68290 | 292 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E); |
aa276e1c TG |
293 | } |
294 | } | |
295 | ||
296 | if (c1e_detected) { | |
297 | int cpu = smp_processor_id(); | |
298 | ||
299 | if (!cpu_isset(cpu, c1e_mask)) { | |
300 | cpu_set(cpu, c1e_mask); | |
0beefa20 TG |
301 | /* |
302 | * Force broadcast so ACPI can not interfere. Needs | |
303 | * to run with interrupts enabled as it uses | |
304 | * smp_function_call. | |
305 | */ | |
306 | local_irq_enable(); | |
aa276e1c TG |
307 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, |
308 | &cpu); | |
309 | printk(KERN_INFO "Switch to broadcast mode on CPU%d\n", | |
310 | cpu); | |
0beefa20 | 311 | local_irq_disable(); |
aa276e1c TG |
312 | } |
313 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); | |
0beefa20 | 314 | |
aa276e1c | 315 | default_idle(); |
0beefa20 TG |
316 | |
317 | /* | |
318 | * The switch back from broadcast mode needs to be | |
319 | * called with interrupts disabled. | |
320 | */ | |
321 | local_irq_disable(); | |
322 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu); | |
323 | local_irq_enable(); | |
aa276e1c TG |
324 | } else |
325 | default_idle(); | |
326 | } | |
327 | ||
7f424a8b PZ |
328 | void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) |
329 | { | |
7f424a8b PZ |
330 | #ifdef CONFIG_X86_SMP |
331 | if (pm_idle == poll_idle && smp_num_siblings > 1) { | |
332 | printk(KERN_WARNING "WARNING: polling idle and HT enabled," | |
333 | " performance may degrade.\n"); | |
334 | } | |
335 | #endif | |
6ddd2a27 TG |
336 | if (pm_idle) |
337 | return; | |
338 | ||
e9623b35 | 339 | if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) { |
7f424a8b | 340 | /* |
7f424a8b PZ |
341 | * One CPU supports mwait => All CPUs supports mwait |
342 | */ | |
6ddd2a27 TG |
343 | printk(KERN_INFO "using mwait in idle threads.\n"); |
344 | pm_idle = mwait_idle; | |
aa276e1c TG |
345 | } else if (check_c1e_idle(c)) { |
346 | printk(KERN_INFO "using C1E aware idle routine\n"); | |
347 | pm_idle = c1e_idle; | |
6ddd2a27 TG |
348 | } else |
349 | pm_idle = default_idle; | |
7f424a8b PZ |
350 | } |
351 | ||
352 | static int __init idle_setup(char *str) | |
353 | { | |
ab6bc3e3 CG |
354 | if (!str) |
355 | return -EINVAL; | |
356 | ||
7f424a8b PZ |
357 | if (!strcmp(str, "poll")) { |
358 | printk("using polling idle threads.\n"); | |
359 | pm_idle = poll_idle; | |
360 | } else if (!strcmp(str, "mwait")) | |
361 | force_mwait = 1; | |
c1e3b377 ZY |
362 | else if (!strcmp(str, "halt")) { |
363 | /* | |
364 | * When the boot option of idle=halt is added, halt is | |
365 | * forced to be used for CPU idle. In such case CPU C2/C3 | |
366 | * won't be used again. | |
367 | * To continue to load the CPU idle driver, don't touch | |
368 | * the boot_option_idle_override. | |
369 | */ | |
370 | pm_idle = default_idle; | |
371 | idle_halt = 1; | |
372 | return 0; | |
da5e09a1 ZY |
373 | } else if (!strcmp(str, "nomwait")) { |
374 | /* | |
375 | * If the boot option of "idle=nomwait" is added, | |
376 | * it means that mwait will be disabled for CPU C2/C3 | |
377 | * states. In such case it won't touch the variable | |
378 | * of boot_option_idle_override. | |
379 | */ | |
380 | idle_nomwait = 1; | |
381 | return 0; | |
c1e3b377 | 382 | } else |
7f424a8b PZ |
383 | return -1; |
384 | ||
385 | boot_option_idle_override = 1; | |
386 | return 0; | |
387 | } | |
388 | early_param("idle", idle_setup); | |
389 |