| Message ID | 1431647748-13221-3-git-send-email-kandoiruchi@google.com (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
On Thu, May 14, 2015 at 04:55:48PM -0700, Ruchi Kandoi wrote: > cpu_power has been added to keep track of amount of power each task is > consuming. cpu_power is updated whenever stime and utime are updated for > a task. power is computed by taking into account the frequency at which > the current core was running and the current for cpu actively > running at hat frequency. > Both you patches completely lack any reason for me to even start considering this. _WHY_ and _what_ are you doing? -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, May 15, 2015 at 9:07 AM, Peter Zijlstra <peterz@infradead.org> wrote: > > On Thu, May 14, 2015 at 04:55:48PM -0700, Ruchi Kandoi wrote: > > cpu_power has been added to keep track of amount of power each task is > > consuming. cpu_power is updated whenever stime and utime are updated for > > a task. power is computed by taking into account the frequency at which > > the current core was running and the current for cpu actively > > running at hat frequency. > > > > Both you patches completely lack any reason for me to even start > considering this. > > _WHY_ and _what_ are you doing? We need a mechanism in which we can get information about how much cpu power each of the process(which is then aggregated fro each uid/application) is consuming. In the current architecture, it is based on the amount of the time the process ran. This brings in inaccuracy because running x seconds at low frequency will have different power consumption as compared to running at the higher frequency. With these changes we have the information about the power which is not only dependent on the time it was running but also takes into account the frequency it was running at as well as the CPU # it was running at. Because the cost of running at different CPUs at the same frequency is different. This gives a better overview of the current power state of the system wrt cpu power. Please let me know if more information is required for the same. Thanks, Ruchi Kandoi -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/cpufreq/cpufreq_stats.c b/drivers/cpufreq/cpufreq_stats.c index 6f0b562..4a0bd9a 100644 --- a/drivers/cpufreq/cpufreq_stats.c +++ b/drivers/cpufreq/cpufreq_stats.c @@ -14,6 +14,7 @@ #include <linux/module.h> #include <linux/slab.h> #include <linux/cputime.h> +#include <linux/sched.h> static spinlock_t cpufreq_stats_lock; @@ -83,6 +84,28 @@ static void store_current_value(struct cpufreq_power_stats *powerstats, } } +void acct_update_power(struct task_struct *task, cputime_t cputime) +{ + struct cpufreq_power_stats *powerstats; + struct cpufreq_stats *stats; + struct cpufreq_policy *policy; + unsigned int cpu_num, curr; + + if (!task) + return; + cpu_num = task_cpu(task); + powerstats = per_cpu(cpufreq_power_stats, cpu_num); + policy = cpufreq_cpu_get(cpu_num); + if (!powerstats || !policy || !(policy->stats)) + return; + + stats = policy->stats; + curr = powerstats->curr[stats->last_index]; + task->cpu_power += curr * cputime_to_usecs(cputime); + cpufreq_cpu_put(cpu_num); +} +EXPORT_SYMBOL_GPL(acct_update_power); + static ssize_t store_current_in_state(struct cpufreq_policy *policy, const char *buf, size_t len) { diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h index 2ee4888..86826c8 100644 --- a/include/linux/cpufreq.h +++ b/include/linux/cpufreq.h @@ -18,6 +18,7 @@ #include <linux/notifier.h> #include <linux/spinlock.h> #include <linux/sysfs.h> +#include <asm/cputime.h> /********************************************************************* * CPUFREQ INTERFACE * @@ -601,4 +602,11 @@ unsigned int cpufreq_generic_get(unsigned int cpu); int cpufreq_generic_init(struct cpufreq_policy *policy, struct cpufreq_frequency_table *table, unsigned int transition_latency); + +/********************************************************************* + * CPUFREQ STATS * + *********************************************************************/ + +void acct_update_power(struct task_struct *p, cputime_t cputime); + #endif /* _LINUX_CPUFREQ_H */ diff --git a/include/linux/sched.h b/include/linux/sched.h index 26a2e61..1f2400a 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1429,6 +1429,7 @@ struct task_struct { int __user *clear_child_tid; /* CLONE_CHILD_CLEARTID */ cputime_t utime, stime, utimescaled, stimescaled; + unsigned long long cpu_power; cputime_t gtime; #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE struct cputime prev_cputime; @@ -1441,6 +1442,7 @@ struct task_struct { VTIME_USER, VTIME_SYS, } vtime_snap_whence; + #endif unsigned long nvcsw, nivcsw; /* context switch counts */ u64 start_time; /* monotonic time in nsec */ diff --git a/kernel/fork.c b/kernel/fork.c index 03c1eaa..2ca0e9e 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1341,6 +1341,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, p->utime = p->stime = p->gtime = 0; p->utimescaled = p->stimescaled = 0; + p->cpu_power = 0; #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE p->prev_cputime.utime = p->prev_cputime.stime = 0; #endif diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c index 8394b1e..53a79d5 100644 --- a/kernel/sched/cputime.c +++ b/kernel/sched/cputime.c @@ -4,6 +4,7 @@ #include <linux/kernel_stat.h> #include <linux/static_key.h> #include <linux/context_tracking.h> +#include <linux/cpufreq.h> #include "sched.h" @@ -149,6 +150,9 @@ void account_user_time(struct task_struct *p, cputime_t cputime, /* Account for user time used */ acct_account_cputime(p); + + /* Account power usage for user time */ + acct_update_power(p, cputime); } /* @@ -199,6 +203,9 @@ void __account_system_time(struct task_struct *p, cputime_t cputime, /* Account for system time used */ acct_account_cputime(p); + + /* Account power usage for system time */ + acct_update_power(p, cputime); } /*
cpu_power has been added to keep track of amount of power each task is consuming. cpu_power is updated whenever stime and utime are updated for a task. power is computed by taking into account the frequency at which the current core was running and the current for cpu actively running at hat frequency. Signed-off-by: Ruchi Kandoi <kandoiruchi@google.com> --- drivers/cpufreq/cpufreq_stats.c | 23 +++++++++++++++++++++++ include/linux/cpufreq.h | 8 ++++++++ include/linux/sched.h | 2 ++ kernel/fork.c | 1 + kernel/sched/cputime.c | 7 +++++++ 5 files changed, 41 insertions(+)