Following patch has introduced per cpu timers or works for ondemand and conservative governors.
commit 2abfa876f1117b0ab45f191fb1f82c41b1cbc8fe Author: Rickard Andersson rickard.andersson@stericsson.com Date: Thu Dec 27 14:55:38 2012 +0000
cpufreq: handle SW coordinated CPUs
This causes additional unnecessary interrupts on all cpus when the load is recently evaluated by any other cpu. i.e. When load is recently evaluated by cpu x, we don't really need any other cpu to evaluate this load again for the next sampling_rate time.
Some sort of code is present to avoid that but we are still getting timer interrupts for all cpus. A good way of avoiding this would be to modify delays for all cpus (policy->cpus) whenever any cpu has evaluated load.
This patchset tries to fix this issue.
These patches are applied here:
http://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=ref...
V1->V2: - Dropped Workqueue modifications and use system_wq directly.
Viresh Kumar (2): cpufreq: ondemand: Don't update sample_type if we don't evaluate load again cpufreq: governors: Avoid unnecessary per cpu timer interrupts
drivers/cpufreq/cpufreq_conservative.c | 8 ++++--- drivers/cpufreq/cpufreq_governor.c | 39 ++++++++++++++++++++++++---------- drivers/cpufreq/cpufreq_governor.h | 2 ++ drivers/cpufreq/cpufreq_ondemand.c | 34 ++++++++++++++--------------- 4 files changed, 52 insertions(+), 31 deletions(-)
Because we have per cpu timer now, we check if we need to evaluate load again or not (In case it is recently evaluated). Here the 2nd cpu which got timer interrupt updates core_dbs_info->sample_type irrespective of load evaluation is required or not. Which is wrong as the first cpu is dependent on this variable set to an older value.
Moreover it would be best in this case to schedule 2nd cpu's timer to sampling_rate instead of freq_lo or hi as that must be managed by the other cpu. In case the other cpu idles in between then also we wouldn't loose much power.
Signed-off-by: Viresh Kumar viresh.kumar@linaro.org --- drivers/cpufreq/cpufreq_ondemand.c | 24 +++++++++++------------- 1 file changed, 11 insertions(+), 13 deletions(-)
diff --git a/drivers/cpufreq/cpufreq_ondemand.c b/drivers/cpufreq/cpufreq_ondemand.c index c5fd794..e76f4f8 100644 --- a/drivers/cpufreq/cpufreq_ondemand.c +++ b/drivers/cpufreq/cpufreq_ondemand.c @@ -224,34 +224,32 @@ static void od_dbs_timer(struct work_struct *work) cpu); struct dbs_data *dbs_data = dbs_info->cdbs.cur_policy->governor_data; struct od_dbs_tuners *od_tuners = dbs_data->tuners; - int delay, sample_type = core_dbs_info->sample_type; - bool eval_load; + int delay = 0, sample_type = core_dbs_info->sample_type;
mutex_lock(&core_dbs_info->cdbs.timer_mutex); - eval_load = need_load_eval(&core_dbs_info->cdbs, - od_tuners->sampling_rate); + if (!need_load_eval(&core_dbs_info->cdbs, od_tuners->sampling_rate)) + goto max_delay;
/* Common NORMAL_SAMPLE setup */ core_dbs_info->sample_type = OD_NORMAL_SAMPLE; if (sample_type == OD_SUB_SAMPLE) { delay = core_dbs_info->freq_lo_jiffies; - if (eval_load) - __cpufreq_driver_target(core_dbs_info->cdbs.cur_policy, - core_dbs_info->freq_lo, - CPUFREQ_RELATION_H); + __cpufreq_driver_target(core_dbs_info->cdbs.cur_policy, + core_dbs_info->freq_lo, CPUFREQ_RELATION_H); } else { - if (eval_load) - dbs_check_cpu(dbs_data, cpu); + dbs_check_cpu(dbs_data, cpu); if (core_dbs_info->freq_lo) { /* Setup timer for SUB_SAMPLE */ core_dbs_info->sample_type = OD_SUB_SAMPLE; delay = core_dbs_info->freq_hi_jiffies; - } else { - delay = delay_for_sampling_rate(od_tuners->sampling_rate - * core_dbs_info->rate_mult); } }
+max_delay: + if (!delay) + delay = delay_for_sampling_rate(od_tuners->sampling_rate + * core_dbs_info->rate_mult); + schedule_delayed_work_on(smp_processor_id(), dw, delay); mutex_unlock(&core_dbs_info->cdbs.timer_mutex); }
Following patch has introduced per cpu timers or works for ondemand and conservative governors.
commit 2abfa876f1117b0ab45f191fb1f82c41b1cbc8fe Author: Rickard Andersson rickard.andersson@stericsson.com Date: Thu Dec 27 14:55:38 2012 +0000
cpufreq: handle SW coordinated CPUs
This causes additional unnecessary interrupts on all cpus when the load is recently evaluated by any other cpu. i.e. When load is recently evaluated by cpu x, we don't really need any other cpu to evaluate this load again for the next sampling_rate time.
Some sort of code is present to avoid that but we are still getting timer interrupts for all cpus. A good way of avoiding this would be to modify delays for all cpus (policy->cpus) whenever any cpu has evaluated load.
This patch does this change and some related code cleanup.
Signed-off-by: Viresh Kumar viresh.kumar@linaro.org --- drivers/cpufreq/cpufreq_conservative.c | 8 ++++--- drivers/cpufreq/cpufreq_governor.c | 39 ++++++++++++++++++++++++---------- drivers/cpufreq/cpufreq_governor.h | 2 ++ drivers/cpufreq/cpufreq_ondemand.c | 12 ++++++----- 4 files changed, 42 insertions(+), 19 deletions(-)
diff --git a/drivers/cpufreq/cpufreq_conservative.c b/drivers/cpufreq/cpufreq_conservative.c index d86577f..77b60a7 100644 --- a/drivers/cpufreq/cpufreq_conservative.c +++ b/drivers/cpufreq/cpufreq_conservative.c @@ -107,7 +107,6 @@ static void cs_check_cpu(int cpu, unsigned int load)
static void cs_dbs_timer(struct work_struct *work) { - struct delayed_work *dw = to_delayed_work(work); struct cs_cpu_dbs_info_s *dbs_info = container_of(work, struct cs_cpu_dbs_info_s, cdbs.work.work); unsigned int cpu = dbs_info->cdbs.cur_policy->cpu; @@ -116,12 +115,15 @@ static void cs_dbs_timer(struct work_struct *work) struct dbs_data *dbs_data = dbs_info->cdbs.cur_policy->governor_data; struct cs_dbs_tuners *cs_tuners = dbs_data->tuners; int delay = delay_for_sampling_rate(cs_tuners->sampling_rate); + bool modify_all = true;
mutex_lock(&core_dbs_info->cdbs.timer_mutex); - if (need_load_eval(&core_dbs_info->cdbs, cs_tuners->sampling_rate)) + if (!need_load_eval(&core_dbs_info->cdbs, cs_tuners->sampling_rate)) + modify_all = false; + else dbs_check_cpu(dbs_data, cpu);
- schedule_delayed_work_on(smp_processor_id(), dw, delay); + gov_queue_work(dbs_data, dbs_info->cdbs.cur_policy, delay, modify_all); mutex_unlock(&core_dbs_info->cdbs.timer_mutex); }
diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c index 7722505..f592f71 100644 --- a/drivers/cpufreq/cpufreq_governor.c +++ b/drivers/cpufreq/cpufreq_governor.c @@ -162,20 +162,38 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu) } EXPORT_SYMBOL_GPL(dbs_check_cpu);
-static inline void dbs_timer_init(struct dbs_data *dbs_data, int cpu, - unsigned int sampling_rate) +static inline void __gov_queue_work(int cpu, struct dbs_data *dbs_data, + unsigned int delay) { - int delay = delay_for_sampling_rate(sampling_rate); struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);
- schedule_delayed_work_on(cpu, &cdbs->work, delay); + mod_delayed_work_on(cpu, system_wq, &cdbs->work, delay); }
-static inline void dbs_timer_exit(struct dbs_data *dbs_data, int cpu) +void gov_queue_work(struct dbs_data *dbs_data, struct cpufreq_policy *policy, + unsigned int delay, bool all_cpus) { - struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu); + int i; + + if (!all_cpus) { + __gov_queue_work(smp_processor_id(), dbs_data, delay); + } else { + for_each_cpu(i, policy->cpus) + __gov_queue_work(i, dbs_data, delay); + } +} +EXPORT_SYMBOL_GPL(gov_queue_work); + +static inline void gov_cancel_work(struct dbs_data *dbs_data, + struct cpufreq_policy *policy) +{ + struct cpu_dbs_common_info *cdbs; + int i;
- cancel_delayed_work_sync(&cdbs->work); + for_each_cpu(i, policy->cpus) { + cdbs = dbs_data->cdata->get_cpu_cdbs(i); + cancel_delayed_work_sync(&cdbs->work); + } }
/* Will return if we need to evaluate cpu load again or not */ @@ -330,16 +348,15 @@ int cpufreq_governor_dbs(struct cpufreq_policy *policy, /* Initiate timer time stamp */ cpu_cdbs->time_stamp = ktime_get();
- for_each_cpu(j, policy->cpus) - dbs_timer_init(dbs_data, j, sampling_rate); + gov_queue_work(dbs_data, policy, + delay_for_sampling_rate(sampling_rate), true); break;
case CPUFREQ_GOV_STOP: if (dbs_data->cdata->governor == GOV_CONSERVATIVE) cs_dbs_info->enable = 0;
- for_each_cpu(j, policy->cpus) - dbs_timer_exit(dbs_data, j); + gov_cancel_work(dbs_data, policy);
mutex_lock(&dbs_data->mutex); mutex_destroy(&cpu_cdbs->timer_mutex); diff --git a/drivers/cpufreq/cpufreq_governor.h b/drivers/cpufreq/cpufreq_governor.h index 6301790..9e3d310 100644 --- a/drivers/cpufreq/cpufreq_governor.h +++ b/drivers/cpufreq/cpufreq_governor.h @@ -192,4 +192,6 @@ bool need_load_eval(struct cpu_dbs_common_info *cdbs, unsigned int sampling_rate); int cpufreq_governor_dbs(struct cpufreq_policy *policy, struct common_dbs_data *cdata, unsigned int event); +void gov_queue_work(struct dbs_data *dbs_data, struct cpufreq_policy *policy, + unsigned int delay, bool all_cpus); #endif /* _CPUFREQ_GOVERNER_H */ diff --git a/drivers/cpufreq/cpufreq_ondemand.c b/drivers/cpufreq/cpufreq_ondemand.c index e76f4f8..04f81d8 100644 --- a/drivers/cpufreq/cpufreq_ondemand.c +++ b/drivers/cpufreq/cpufreq_ondemand.c @@ -216,7 +216,6 @@ static void od_check_cpu(int cpu, unsigned int load_freq)
static void od_dbs_timer(struct work_struct *work) { - struct delayed_work *dw = to_delayed_work(work); struct od_cpu_dbs_info_s *dbs_info = container_of(work, struct od_cpu_dbs_info_s, cdbs.work.work); unsigned int cpu = dbs_info->cdbs.cur_policy->cpu; @@ -225,10 +224,13 @@ static void od_dbs_timer(struct work_struct *work) struct dbs_data *dbs_data = dbs_info->cdbs.cur_policy->governor_data; struct od_dbs_tuners *od_tuners = dbs_data->tuners; int delay = 0, sample_type = core_dbs_info->sample_type; + bool modify_all = true;
mutex_lock(&core_dbs_info->cdbs.timer_mutex); - if (!need_load_eval(&core_dbs_info->cdbs, od_tuners->sampling_rate)) + if (!need_load_eval(&core_dbs_info->cdbs, od_tuners->sampling_rate)) { + modify_all = false; goto max_delay; + }
/* Common NORMAL_SAMPLE setup */ core_dbs_info->sample_type = OD_NORMAL_SAMPLE; @@ -250,7 +252,7 @@ max_delay: delay = delay_for_sampling_rate(od_tuners->sampling_rate * core_dbs_info->rate_mult);
- schedule_delayed_work_on(smp_processor_id(), dw, delay); + gov_queue_work(dbs_data, dbs_info->cdbs.cur_policy, delay, modify_all); mutex_unlock(&core_dbs_info->cdbs.timer_mutex); }
@@ -308,8 +310,8 @@ static void update_sampling_rate(struct dbs_data *dbs_data, cancel_delayed_work_sync(&dbs_info->cdbs.work); mutex_lock(&dbs_info->cdbs.timer_mutex);
- schedule_delayed_work_on(cpu, &dbs_info->cdbs.work, - usecs_to_jiffies(new_rate)); + gov_queue_work(dbs_data, dbs_info->cdbs.cur_policy, + usecs_to_jiffies(new_rate), true);
} mutex_unlock(&dbs_info->cdbs.timer_mutex);
linaro-kernel@lists.linaro.org
-
Viresh Kumar