aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--drivers/acpi/battery.c2
-rw-r--r--drivers/cpufreq/cpufreq.c19
-rw-r--r--drivers/cpuidle/governors/menu.c106
-rw-r--r--include/linux/sched.h1
-rw-r--r--include/linux/tick.h6
-rw-r--r--kernel/freezer.c2
-rw-r--r--kernel/power/process.c11
-rw-r--r--kernel/time/tick-sched.c9
8 files changed, 31 insertions, 125 deletions
diff --git a/drivers/acpi/battery.c b/drivers/acpi/battery.c
index 082b4dd252a8..d405fbad406a 100644
--- a/drivers/acpi/battery.c
+++ b/drivers/acpi/battery.c
@@ -117,6 +117,7 @@ struct acpi_battery {
117 struct acpi_device *device; 117 struct acpi_device *device;
118 struct notifier_block pm_nb; 118 struct notifier_block pm_nb;
119 unsigned long update_time; 119 unsigned long update_time;
120 int revision;
120 int rate_now; 121 int rate_now;
121 int capacity_now; 122 int capacity_now;
122 int voltage_now; 123 int voltage_now;
@@ -359,6 +360,7 @@ static struct acpi_offsets info_offsets[] = {
359}; 360};
360 361
361static struct acpi_offsets extended_info_offsets[] = { 362static struct acpi_offsets extended_info_offsets[] = {
363 {offsetof(struct acpi_battery, revision), 0},
362 {offsetof(struct acpi_battery, power_unit), 0}, 364 {offsetof(struct acpi_battery, power_unit), 0},
363 {offsetof(struct acpi_battery, design_capacity), 0}, 365 {offsetof(struct acpi_battery, design_capacity), 0},
364 {offsetof(struct acpi_battery, full_charge_capacity), 0}, 366 {offsetof(struct acpi_battery, full_charge_capacity), 0},
diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index a4ad7339588d..f0a5e2b0eb8a 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1177,14 +1177,11 @@ static int __cpufreq_remove_dev(struct device *dev,
1177 __func__, cpu_dev->id, cpu); 1177 __func__, cpu_dev->id, cpu);
1178 } 1178 }
1179 1179
1180 if ((cpus == 1) && (cpufreq_driver->target))
1181 __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1182
1183 pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1184 cpufreq_cpu_put(data);
1185
1186 /* If cpu is last user of policy, free policy */ 1180 /* If cpu is last user of policy, free policy */
1187 if (cpus == 1) { 1181 if (cpus == 1) {
1182 if (cpufreq_driver->target)
1183 __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1184
1188 lock_policy_rwsem_read(cpu); 1185 lock_policy_rwsem_read(cpu);
1189 kobj = &data->kobj; 1186 kobj = &data->kobj;
1190 cmp = &data->kobj_unregister; 1187 cmp = &data->kobj_unregister;
@@ -1205,9 +1202,13 @@ static int __cpufreq_remove_dev(struct device *dev,
1205 free_cpumask_var(data->related_cpus); 1202 free_cpumask_var(data->related_cpus);
1206 free_cpumask_var(data->cpus); 1203 free_cpumask_var(data->cpus);
1207 kfree(data); 1204 kfree(data);
1208 } else if (cpufreq_driver->target) { 1205 } else {
1209 __cpufreq_governor(data, CPUFREQ_GOV_START); 1206 pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1210 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS); 1207 cpufreq_cpu_put(data);
1208 if (cpufreq_driver->target) {
1209 __cpufreq_governor(data, CPUFREQ_GOV_START);
1210 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1211 }
1211 } 1212 }
1212 1213
1213 per_cpu(cpufreq_policy_cpu, cpu) = -1; 1214 per_cpu(cpufreq_policy_cpu, cpu) = -1;
diff --git a/drivers/cpuidle/governors/menu.c b/drivers/cpuidle/governors/menu.c
index fe343a06b7da..bc580b67a652 100644
--- a/drivers/cpuidle/governors/menu.c
+++ b/drivers/cpuidle/governors/menu.c
@@ -28,13 +28,6 @@
28#define MAX_INTERESTING 50000 28#define MAX_INTERESTING 50000
29#define STDDEV_THRESH 400 29#define STDDEV_THRESH 400
30 30
31/* 60 * 60 > STDDEV_THRESH * INTERVALS = 400 * 8 */
32#define MAX_DEVIATION 60
33
34static DEFINE_PER_CPU(struct hrtimer, menu_hrtimer);
35static DEFINE_PER_CPU(int, hrtimer_status);
36/* menu hrtimer mode */
37enum {MENU_HRTIMER_STOP, MENU_HRTIMER_REPEAT, MENU_HRTIMER_GENERAL};
38 31
39/* 32/*
40 * Concepts and ideas behind the menu governor 33 * Concepts and ideas behind the menu governor
@@ -116,13 +109,6 @@ enum {MENU_HRTIMER_STOP, MENU_HRTIMER_REPEAT, MENU_HRTIMER_GENERAL};
116 * 109 *
117 */ 110 */
118 111
119/*
120 * The C-state residency is so long that is is worthwhile to exit
121 * from the shallow C-state and re-enter into a deeper C-state.
122 */
123static unsigned int perfect_cstate_ms __read_mostly = 30;
124module_param(perfect_cstate_ms, uint, 0000);
125
126struct menu_device { 112struct menu_device {
127 int last_state_idx; 113 int last_state_idx;
128 int needs_update; 114 int needs_update;
@@ -205,52 +191,17 @@ static u64 div_round64(u64 dividend, u32 divisor)
205 return div_u64(dividend + (divisor / 2), divisor); 191 return div_u64(dividend + (divisor / 2), divisor);
206} 192}
207 193
208/* Cancel the hrtimer if it is not triggered yet */
209void menu_hrtimer_cancel(void)
210{
211 int cpu = smp_processor_id();
212 struct hrtimer *hrtmr = &per_cpu(menu_hrtimer, cpu);
213
214 /* The timer is still not time out*/
215 if (per_cpu(hrtimer_status, cpu)) {
216 hrtimer_cancel(hrtmr);
217 per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_STOP;
218 }
219}
220EXPORT_SYMBOL_GPL(menu_hrtimer_cancel);
221
222/* Call back for hrtimer is triggered */
223static enum hrtimer_restart menu_hrtimer_notify(struct hrtimer *hrtimer)
224{
225 int cpu = smp_processor_id();
226 struct menu_device *data = &per_cpu(menu_devices, cpu);
227
228 /* In general case, the expected residency is much larger than
229 * deepest C-state target residency, but prediction logic still
230 * predicts a small predicted residency, so the prediction
231 * history is totally broken if the timer is triggered.
232 * So reset the correction factor.
233 */
234 if (per_cpu(hrtimer_status, cpu) == MENU_HRTIMER_GENERAL)
235 data->correction_factor[data->bucket] = RESOLUTION * DECAY;
236
237 per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_STOP;
238
239 return HRTIMER_NORESTART;
240}
241
242/* 194/*
243 * Try detecting repeating patterns by keeping track of the last 8 195 * Try detecting repeating patterns by keeping track of the last 8
244 * intervals, and checking if the standard deviation of that set 196 * intervals, and checking if the standard deviation of that set
245 * of points is below a threshold. If it is... then use the 197 * of points is below a threshold. If it is... then use the
246 * average of these 8 points as the estimated value. 198 * average of these 8 points as the estimated value.
247 */ 199 */
248static u32 get_typical_interval(struct menu_device *data) 200static void get_typical_interval(struct menu_device *data)
249{ 201{
250 int i = 0, divisor = 0; 202 int i = 0, divisor = 0;
251 uint64_t max = 0, avg = 0, stddev = 0; 203 uint64_t max = 0, avg = 0, stddev = 0;
252 int64_t thresh = LLONG_MAX; /* Discard outliers above this value. */ 204 int64_t thresh = LLONG_MAX; /* Discard outliers above this value. */
253 unsigned int ret = 0;
254 205
255again: 206again:
256 207
@@ -291,16 +242,13 @@ again:
291 if (((avg > stddev * 6) && (divisor * 4 >= INTERVALS * 3)) 242 if (((avg > stddev * 6) && (divisor * 4 >= INTERVALS * 3))
292 || stddev <= 20) { 243 || stddev <= 20) {
293 data->predicted_us = avg; 244 data->predicted_us = avg;
294 ret = 1; 245 return;
295 return ret;
296 246
297 } else if ((divisor * 4) > INTERVALS * 3) { 247 } else if ((divisor * 4) > INTERVALS * 3) {
298 /* Exclude the max interval */ 248 /* Exclude the max interval */
299 thresh = max - 1; 249 thresh = max - 1;
300 goto again; 250 goto again;
301 } 251 }
302
303 return ret;
304} 252}
305 253
306/** 254/**
@@ -315,9 +263,6 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
315 int i; 263 int i;
316 int multiplier; 264 int multiplier;
317 struct timespec t; 265 struct timespec t;
318 int repeat = 0, low_predicted = 0;
319 int cpu = smp_processor_id();
320 struct hrtimer *hrtmr = &per_cpu(menu_hrtimer, cpu);
321 266
322 if (data->needs_update) { 267 if (data->needs_update) {
323 menu_update(drv, dev); 268 menu_update(drv, dev);
@@ -352,7 +297,7 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
352 data->predicted_us = div_round64(data->expected_us * data->correction_factor[data->bucket], 297 data->predicted_us = div_round64(data->expected_us * data->correction_factor[data->bucket],
353 RESOLUTION * DECAY); 298 RESOLUTION * DECAY);
354 299
355 repeat = get_typical_interval(data); 300 get_typical_interval(data);
356 301
357 /* 302 /*
358 * We want to default to C1 (hlt), not to busy polling 303 * We want to default to C1 (hlt), not to busy polling
@@ -373,10 +318,8 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
373 318
374 if (s->disabled || su->disable) 319 if (s->disabled || su->disable)
375 continue; 320 continue;
376 if (s->target_residency > data->predicted_us) { 321 if (s->target_residency > data->predicted_us)
377 low_predicted = 1;
378 continue; 322 continue;
379 }
380 if (s->exit_latency > latency_req) 323 if (s->exit_latency > latency_req)
381 continue; 324 continue;
382 if (s->exit_latency * multiplier > data->predicted_us) 325 if (s->exit_latency * multiplier > data->predicted_us)
@@ -386,44 +329,6 @@ static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
386 data->exit_us = s->exit_latency; 329 data->exit_us = s->exit_latency;
387 } 330 }
388 331
389 /* not deepest C-state chosen for low predicted residency */
390 if (low_predicted) {
391 unsigned int timer_us = 0;
392 unsigned int perfect_us = 0;
393
394 /*
395 * Set a timer to detect whether this sleep is much
396 * longer than repeat mode predicted. If the timer
397 * triggers, the code will evaluate whether to put
398 * the CPU into a deeper C-state.
399 * The timer is cancelled on CPU wakeup.
400 */
401 timer_us = 2 * (data->predicted_us + MAX_DEVIATION);
402
403 perfect_us = perfect_cstate_ms * 1000;
404
405 if (repeat && (4 * timer_us < data->expected_us)) {
406 RCU_NONIDLE(hrtimer_start(hrtmr,
407 ns_to_ktime(1000 * timer_us),
408 HRTIMER_MODE_REL_PINNED));
409 /* In repeat case, menu hrtimer is started */
410 per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_REPEAT;
411 } else if (perfect_us < data->expected_us) {
412 /*
413 * The next timer is long. This could be because
414 * we did not make a useful prediction.
415 * In that case, it makes sense to re-enter
416 * into a deeper C-state after some time.
417 */
418 RCU_NONIDLE(hrtimer_start(hrtmr,
419 ns_to_ktime(1000 * timer_us),
420 HRTIMER_MODE_REL_PINNED));
421 /* In general case, menu hrtimer is started */
422 per_cpu(hrtimer_status, cpu) = MENU_HRTIMER_GENERAL;
423 }
424
425 }
426
427 return data->last_state_idx; 332 return data->last_state_idx;
428} 333}
429 334
@@ -514,9 +419,6 @@ static int menu_enable_device(struct cpuidle_driver *drv,
514 struct cpuidle_device *dev) 419 struct cpuidle_device *dev)
515{ 420{
516 struct menu_device *data = &per_cpu(menu_devices, dev->cpu); 421 struct menu_device *data = &per_cpu(menu_devices, dev->cpu);
517 struct hrtimer *t = &per_cpu(menu_hrtimer, dev->cpu);
518 hrtimer_init(t, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
519 t->function = menu_hrtimer_notify;
520 422
521 memset(data, 0, sizeof(struct menu_device)); 423 memset(data, 0, sizeof(struct menu_device));
522 424
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 50d04b92ceda..d722490da030 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1628,6 +1628,7 @@ extern void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut,
1628#define PF_MEMPOLICY 0x10000000 /* Non-default NUMA mempolicy */ 1628#define PF_MEMPOLICY 0x10000000 /* Non-default NUMA mempolicy */
1629#define PF_MUTEX_TESTER 0x20000000 /* Thread belongs to the rt mutex tester */ 1629#define PF_MUTEX_TESTER 0x20000000 /* Thread belongs to the rt mutex tester */
1630#define PF_FREEZER_SKIP 0x40000000 /* Freezer should not count it as freezable */ 1630#define PF_FREEZER_SKIP 0x40000000 /* Freezer should not count it as freezable */
1631#define PF_SUSPEND_TASK 0x80000000 /* this thread called freeze_processes and should not be frozen */
1631 1632
1632/* 1633/*
1633 * Only the _current_ task can read/write to tsk->flags, but other 1634 * Only the _current_ task can read/write to tsk->flags, but other
diff --git a/include/linux/tick.h b/include/linux/tick.h
index 9180f4b85e6d..62bd8b72873c 100644
--- a/include/linux/tick.h
+++ b/include/linux/tick.h
@@ -174,10 +174,4 @@ static inline void tick_nohz_task_switch(struct task_struct *tsk) { }
174#endif 174#endif
175 175
176 176
177# ifdef CONFIG_CPU_IDLE_GOV_MENU
178extern void menu_hrtimer_cancel(void);
179# else
180static inline void menu_hrtimer_cancel(void) {}
181# endif /* CONFIG_CPU_IDLE_GOV_MENU */
182
183#endif 177#endif
diff --git a/kernel/freezer.c b/kernel/freezer.c
index 8b2afc1c9df0..b462fa197517 100644
--- a/kernel/freezer.c
+++ b/kernel/freezer.c
@@ -33,7 +33,7 @@ static DEFINE_SPINLOCK(freezer_lock);
33 */ 33 */
34bool freezing_slow_path(struct task_struct *p) 34bool freezing_slow_path(struct task_struct *p)
35{ 35{
36 if (p->flags & PF_NOFREEZE) 36 if (p->flags & (PF_NOFREEZE | PF_SUSPEND_TASK))
37 return false; 37 return false;
38 38
39 if (pm_nosig_freezing || cgroup_freezing(p)) 39 if (pm_nosig_freezing || cgroup_freezing(p))
diff --git a/kernel/power/process.c b/kernel/power/process.c
index fc0df8486449..06ec8869dbf1 100644
--- a/kernel/power/process.c
+++ b/kernel/power/process.c
@@ -109,6 +109,8 @@ static int try_to_freeze_tasks(bool user_only)
109 109
110/** 110/**
111 * freeze_processes - Signal user space processes to enter the refrigerator. 111 * freeze_processes - Signal user space processes to enter the refrigerator.
112 * The current thread will not be frozen. The same process that calls
113 * freeze_processes must later call thaw_processes.
112 * 114 *
113 * On success, returns 0. On failure, -errno and system is fully thawed. 115 * On success, returns 0. On failure, -errno and system is fully thawed.
114 */ 116 */
@@ -120,6 +122,9 @@ int freeze_processes(void)
120 if (error) 122 if (error)
121 return error; 123 return error;
122 124
125 /* Make sure this task doesn't get frozen */
126 current->flags |= PF_SUSPEND_TASK;
127
123 if (!pm_freezing) 128 if (!pm_freezing)
124 atomic_inc(&system_freezing_cnt); 129 atomic_inc(&system_freezing_cnt);
125 130
@@ -168,6 +173,7 @@ int freeze_kernel_threads(void)
168void thaw_processes(void) 173void thaw_processes(void)
169{ 174{
170 struct task_struct *g, *p; 175 struct task_struct *g, *p;
176 struct task_struct *curr = current;
171 177
172 if (pm_freezing) 178 if (pm_freezing)
173 atomic_dec(&system_freezing_cnt); 179 atomic_dec(&system_freezing_cnt);
@@ -182,10 +188,15 @@ void thaw_processes(void)
182 188
183 read_lock(&tasklist_lock); 189 read_lock(&tasklist_lock);
184 do_each_thread(g, p) { 190 do_each_thread(g, p) {
191 /* No other threads should have PF_SUSPEND_TASK set */
192 WARN_ON((p != curr) && (p->flags & PF_SUSPEND_TASK));
185 __thaw_task(p); 193 __thaw_task(p);
186 } while_each_thread(g, p); 194 } while_each_thread(g, p);
187 read_unlock(&tasklist_lock); 195 read_unlock(&tasklist_lock);
188 196
197 WARN_ON(!(curr->flags & PF_SUSPEND_TASK));
198 curr->flags &= ~PF_SUSPEND_TASK;
199
189 usermodehelper_enable(); 200 usermodehelper_enable();
190 201
191 schedule(); 202 schedule();
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index e80183f4a6c4..e77edc97e036 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -827,13 +827,10 @@ void tick_nohz_irq_exit(void)
827{ 827{
828 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); 828 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
829 829
830 if (ts->inidle) { 830 if (ts->inidle)
831 /* Cancel the timer because CPU already waken up from the C-states*/
832 menu_hrtimer_cancel();
833 __tick_nohz_idle_enter(ts); 831 __tick_nohz_idle_enter(ts);
834 } else { 832 else
835 tick_nohz_full_stop_tick(ts); 833 tick_nohz_full_stop_tick(ts);
836 }
837} 834}
838 835
839/** 836/**
@@ -931,8 +928,6 @@ void tick_nohz_idle_exit(void)
931 928
932 ts->inidle = 0; 929 ts->inidle = 0;
933 930
934 /* Cancel the timer because CPU already waken up from the C-states*/
935 menu_hrtimer_cancel();
936 if (ts->idle_active || ts->tick_stopped) 931 if (ts->idle_active || ts->tick_stopped)
937 now = ktime_get(); 932 now = ktime_get();
938 933