OpenCloudOS-Kernel/kernel/sched/pelt.h

236 lines
6.0 KiB
C
Raw Normal View History

#ifdef CONFIG_SMP
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
#include "sched-pelt.h"
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
int __update_load_avg_blocked_se(u64 now, struct sched_entity *se);
int __update_load_avg_se(u64 now, struct cfs_rq *cfs_rq, struct sched_entity *se);
int __update_load_avg_cfs_rq(u64 now, struct cfs_rq *cfs_rq);
2018-06-28 23:45:05 +08:00
int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);
int update_dl_rq_load_avg(u64 now, struct rq *rq, int running);
#ifdef CONFIG_SCHED_THERMAL_PRESSURE
int update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity);
static inline u64 thermal_load_avg(struct rq *rq)
{
return READ_ONCE(rq->avg_thermal.load_avg);
}
#else
static inline int
update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity)
{
return 0;
}
static inline u64 thermal_load_avg(struct rq *rq)
{
return 0;
}
#endif
#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
sched/irq: Add IRQ utilization tracking interrupt and steal time are the only remaining activities tracked by rt_avg. Like for sched classes, we can use PELT to track their average utilization of the CPU. But unlike sched class, we don't track when entering/leaving interrupt; Instead, we take into account the time spent under interrupt context when we update rqs' clock (rq_clock_task). This also means that we have to decay the normal context time and account for interrupt time during the update. That's also important to note that because: rq_clock == rq_clock_task + interrupt time and rq_clock_task is used by a sched class to compute its utilization, the util_avg of a sched class only reflects the utilization of the time spent in normal context and not of the whole time of the CPU. The utilization of interrupt gives an more accurate level of utilization of CPU. The CPU utilization is: avg_irq + (1 - avg_irq / max capacity) * /Sum avg_rq Most of the time, avg_irq is small and neglictible so the use of the approximation CPU utilization = /Sum avg_rq was enough. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: claudio@evidence.eu.com Cc: daniel.lezcano@linaro.org Cc: dietmar.eggemann@arm.com Cc: joel@joelfernandes.org Cc: juri.lelli@redhat.com Cc: luca.abeni@santannapisa.it Cc: patrick.bellasi@arm.com Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: valentin.schneider@arm.com Cc: viresh.kumar@linaro.org Link: http://lkml.kernel.org/r/1530200714-4504-7-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2018-06-28 23:45:09 +08:00
int update_irq_load_avg(struct rq *rq, u64 running);
#else
static inline int
update_irq_load_avg(struct rq *rq, u64 running)
{
return 0;
}
#endif
#define PELT_MIN_DIVIDER (LOAD_AVG_MAX - 1024)
static inline u32 get_pelt_divider(struct sched_avg *avg)
{
return PELT_MIN_DIVIDER + avg->period_contrib;
}
static inline void cfs_se_util_change(struct sched_avg *avg)
{
unsigned int enqueued;
if (!sched_feat(UTIL_EST))
return;
sched/fair: Fix util_est UTIL_AVG_UNCHANGED handling The util_est internal UTIL_AVG_UNCHANGED flag which is used to prevent unnecessary util_est updates uses the LSB of util_est.enqueued. It is exposed via _task_util_est() (and task_util_est()). Commit 92a801e5d5b7 ("sched/fair: Mask UTIL_AVG_UNCHANGED usages") mentions that the LSB is lost for util_est resolution but find_energy_efficient_cpu() checks if task_util_est() returns 0 to return prev_cpu early. _task_util_est() returns the max value of util_est.ewma and util_est.enqueued or'ed w/ UTIL_AVG_UNCHANGED. So task_util_est() returning the max of task_util() and _task_util_est() will never return 0 under the default SCHED_FEAT(UTIL_EST, true). To fix this use the MSB of util_est.enqueued instead and keep the flag util_est internal, i.e. don't export it via _task_util_est(). The maximal possible util_avg value for a task is 1024 so the MSB of 'unsigned int util_est.enqueued' isn't used to store a util value. As a caveat the code behind the util_est_se trace point has to filter UTIL_AVG_UNCHANGED to see the real util_est.enqueued value which should be easy to do. This also fixes an issue report by Xuewen Yan that util_est_update() only used UTIL_AVG_UNCHANGED for the subtrahend of the equation: last_enqueued_diff = ue.enqueued - (task_util() | UTIL_AVG_UNCHANGED) Fixes: b89997aa88f0b sched/pelt: Fix task util_est update filtering Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Xuewen Yan <xuewen.yan@unisoc.com> Reviewed-by: Vincent Donnefort <vincent.donnefort@arm.com> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lore.kernel.org/r/20210602145808.1562603-1-dietmar.eggemann@arm.com
2021-06-02 22:58:08 +08:00
/* Avoid store if the flag has been already reset */
enqueued = avg->util_est.enqueued;
if (!(enqueued & UTIL_AVG_UNCHANGED))
return;
/* Reset flag to report util_avg has been updated */
enqueued &= ~UTIL_AVG_UNCHANGED;
WRITE_ONCE(avg->util_est.enqueued, enqueued);
}
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
static inline u64 rq_clock_pelt(struct rq *rq)
{
lockdep_assert_rq_held(rq);
assert_clock_updated(rq);
return rq->clock_pelt - rq->lost_idle_time;
}
/* The rq is idle, we can sync to clock_task */
static inline void _update_idle_rq_clock_pelt(struct rq *rq)
{
rq->clock_pelt = rq_clock_task(rq);
u64_u32_store(rq->clock_idle, rq_clock(rq));
/* Paired with smp_rmb in migrate_se_pelt_lag() */
smp_wmb();
u64_u32_store(rq->clock_pelt_idle, rq_clock_pelt(rq));
}
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
/*
* The clock_pelt scales the time to reflect the effective amount of
* computation done during the running delta time but then sync back to
* clock_task when rq is idle.
*
*
* absolute time | 1| 2| 3| 4| 5| 6| 7| 8| 9|10|11|12|13|14|15|16
* @ max capacity ------******---------------******---------------
* @ half capacity ------************---------************---------
* clock pelt | 1| 2| 3| 4| 7| 8| 9| 10| 11|14|15|16
*
*/
static inline void update_rq_clock_pelt(struct rq *rq, s64 delta)
{
if (unlikely(is_idle_task(rq->curr))) {
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
_update_idle_rq_clock_pelt(rq);
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
return;
}
/*
* When a rq runs at a lower compute capacity, it will need
* more time to do the same amount of work than at max
* capacity. In order to be invariant, we scale the delta to
* reflect how much work has been really done.
* Running longer results in stealing idle time that will
* disturb the load signal compared to max capacity. This
* stolen idle time will be automatically reflected when the
* rq will be idle and the clock will be synced with
* rq_clock_task.
*/
/*
* Scale the elapsed time to reflect the real amount of
* computation
*/
delta = cap_scale(delta, arch_scale_cpu_capacity(cpu_of(rq)));
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
delta = cap_scale(delta, arch_scale_freq_capacity(cpu_of(rq)));
rq->clock_pelt += delta;
}
/*
* When rq becomes idle, we have to check if it has lost idle time
* because it was fully busy. A rq is fully used when the /Sum util_sum
* is greater or equal to:
* (LOAD_AVG_MAX - 1024 + rq->cfs.avg.period_contrib) << SCHED_CAPACITY_SHIFT;
* For optimization and computing rounding purpose, we don't take into account
* the position in the current window (period_contrib) and we use the higher
* bound of util_sum to decide.
*/
static inline void update_idle_rq_clock_pelt(struct rq *rq)
{
u32 divider = ((LOAD_AVG_MAX - 1024) << SCHED_CAPACITY_SHIFT) - LOAD_AVG_MAX;
u32 util_sum = rq->cfs.avg.util_sum;
util_sum += rq->avg_rt.util_sum;
util_sum += rq->avg_dl.util_sum;
/*
* Reflecting stolen time makes sense only if the idle
* phase would be present at max capacity. As soon as the
* utilization of a rq has reached the maximum value, it is
* considered as an always running rq without idle time to
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
* steal. This potential idle time is considered as lost in
* this case. We keep track of this lost idle time compare to
* rq's clock_task.
*/
if (util_sum >= divider)
rq->lost_idle_time += rq_clock_task(rq) - rq->clock_pelt;
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
_update_idle_rq_clock_pelt(rq);
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
}
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
#ifdef CONFIG_CFS_BANDWIDTH
static inline void update_idle_cfs_rq_clock_pelt(struct cfs_rq *cfs_rq)
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
{
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
u64 throttled;
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
if (unlikely(cfs_rq->throttle_count))
throttled = U64_MAX;
else
throttled = cfs_rq->throttled_clock_pelt_time;
u64_u32_store(cfs_rq->throttled_pelt_idle, throttled);
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
}
/* rq->task_clock normalized against any time this cfs_rq has spent throttled */
static inline u64 cfs_rq_clock_pelt(struct cfs_rq *cfs_rq)
{
if (unlikely(cfs_rq->throttle_count))
return cfs_rq->throttled_clock_pelt - cfs_rq->throttled_clock_pelt_time;
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
return rq_clock_pelt(rq_of(cfs_rq)) - cfs_rq->throttled_clock_pelt_time;
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
}
#else
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
static inline void update_idle_cfs_rq_clock_pelt(struct cfs_rq *cfs_rq) { }
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
static inline u64 cfs_rq_clock_pelt(struct cfs_rq *cfs_rq)
{
return rq_clock_pelt(rq_of(cfs_rq));
}
#endif
#else
static inline int
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
{
return 0;
}
2018-06-28 23:45:05 +08:00
static inline int
update_rt_rq_load_avg(u64 now, struct rq *rq, int running)
{
return 0;
}
static inline int
update_dl_rq_load_avg(u64 now, struct rq *rq, int running)
{
return 0;
}
sched/irq: Add IRQ utilization tracking interrupt and steal time are the only remaining activities tracked by rt_avg. Like for sched classes, we can use PELT to track their average utilization of the CPU. But unlike sched class, we don't track when entering/leaving interrupt; Instead, we take into account the time spent under interrupt context when we update rqs' clock (rq_clock_task). This also means that we have to decay the normal context time and account for interrupt time during the update. That's also important to note that because: rq_clock == rq_clock_task + interrupt time and rq_clock_task is used by a sched class to compute its utilization, the util_avg of a sched class only reflects the utilization of the time spent in normal context and not of the whole time of the CPU. The utilization of interrupt gives an more accurate level of utilization of CPU. The CPU utilization is: avg_irq + (1 - avg_irq / max capacity) * /Sum avg_rq Most of the time, avg_irq is small and neglictible so the use of the approximation CPU utilization = /Sum avg_rq was enough. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: claudio@evidence.eu.com Cc: daniel.lezcano@linaro.org Cc: dietmar.eggemann@arm.com Cc: joel@joelfernandes.org Cc: juri.lelli@redhat.com Cc: luca.abeni@santannapisa.it Cc: patrick.bellasi@arm.com Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: valentin.schneider@arm.com Cc: viresh.kumar@linaro.org Link: http://lkml.kernel.org/r/1530200714-4504-7-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2018-06-28 23:45:09 +08:00
static inline int
update_thermal_load_avg(u64 now, struct rq *rq, u64 capacity)
{
return 0;
}
static inline u64 thermal_load_avg(struct rq *rq)
{
return 0;
}
sched/irq: Add IRQ utilization tracking interrupt and steal time are the only remaining activities tracked by rt_avg. Like for sched classes, we can use PELT to track their average utilization of the CPU. But unlike sched class, we don't track when entering/leaving interrupt; Instead, we take into account the time spent under interrupt context when we update rqs' clock (rq_clock_task). This also means that we have to decay the normal context time and account for interrupt time during the update. That's also important to note that because: rq_clock == rq_clock_task + interrupt time and rq_clock_task is used by a sched class to compute its utilization, the util_avg of a sched class only reflects the utilization of the time spent in normal context and not of the whole time of the CPU. The utilization of interrupt gives an more accurate level of utilization of CPU. The CPU utilization is: avg_irq + (1 - avg_irq / max capacity) * /Sum avg_rq Most of the time, avg_irq is small and neglictible so the use of the approximation CPU utilization = /Sum avg_rq was enough. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: claudio@evidence.eu.com Cc: daniel.lezcano@linaro.org Cc: dietmar.eggemann@arm.com Cc: joel@joelfernandes.org Cc: juri.lelli@redhat.com Cc: luca.abeni@santannapisa.it Cc: patrick.bellasi@arm.com Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: valentin.schneider@arm.com Cc: viresh.kumar@linaro.org Link: http://lkml.kernel.org/r/1530200714-4504-7-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2018-06-28 23:45:09 +08:00
static inline int
update_irq_load_avg(struct rq *rq, u64 running)
{
return 0;
}
sched/fair: Update scale invariance of PELT The current implementation of load tracking invariance scales the contribution with current frequency and uarch performance (only for utilization) of the CPU. One main result of this formula is that the figures are capped by current capacity of CPU. Another one is that the load_avg is not invariant because not scaled with uarch. The util_avg of a periodic task that runs r time slots every p time slots varies in the range : U * (1-y^r)/(1-y^p) * y^i < Utilization < U * (1-y^r)/(1-y^p) with U is the max util_avg value = SCHED_CAPACITY_SCALE At a lower capacity, the range becomes: U * C * (1-y^r')/(1-y^p) * y^i' < Utilization < U * C * (1-y^r')/(1-y^p) with C reflecting the compute capacity ratio between current capacity and max capacity. so C tries to compensate changes in (1-y^r') but it can't be accurate. Instead of scaling the contribution value of PELT algo, we should scale the running time. The PELT signal aims to track the amount of computation of tasks and/or rq so it seems more correct to scale the running time to reflect the effective amount of computation done since the last update. In order to be fully invariant, we need to apply the same amount of running time and idle time whatever the current capacity. Because running at lower capacity implies that the task will run longer, we have to ensure that the same amount of idle time will be applied when system becomes idle and no idle time has been "stolen". But reaching the maximum utilization value (SCHED_CAPACITY_SCALE) means that the task is seen as an always-running task whatever the capacity of the CPU (even at max compute capacity). In this case, we can discard this "stolen" idle times which becomes meaningless. In order to achieve this time scaling, a new clock_pelt is created per rq. The increase of this clock scales with current capacity when something is running on rq and synchronizes with clock_task when rq is idle. With this mechanism, we ensure the same running and idle time whatever the current capacity. This also enables to simplify the pelt algorithm by removing all references of uarch and frequency and applying the same contribution to utilization and loads. Furthermore, the scaling is done only once per update of clock (update_rq_clock_task()) instead of during each update of sched_entities and cfs/rt/dl_rq of the rq like the current implementation. This is interesting when cgroup are involved as shown in the results below: On a hikey (octo Arm64 platform). Performance cpufreq governor and only shallowest c-state to remove variance generated by those power features so we only track the impact of pelt algo. each test runs 16 times: ./perf bench sched pipe (higher is better) kernel tip/sched/core + patch ops/seconds ops/seconds diff cgroup root 59652(+/- 0.18%) 59876(+/- 0.24%) +0.38% level1 55608(+/- 0.27%) 55923(+/- 0.24%) +0.57% level2 52115(+/- 0.29%) 52564(+/- 0.22%) +0.86% hackbench -l 1000 (lower is better) kernel tip/sched/core + patch duration(sec) duration(sec) diff cgroup root 4.453(+/- 2.37%) 4.383(+/- 2.88%) -1.57% level1 4.859(+/- 8.50%) 4.830(+/- 7.07%) -0.60% level2 5.063(+/- 9.83%) 4.928(+/- 9.66%) -2.66% Then, the responsiveness of PELT is improved when CPU is not running at max capacity with this new algorithm. I have put below some examples of duration to reach some typical load values according to the capacity of the CPU with current implementation and with this patch. These values has been computed based on the geometric series and the half period value: Util (%) max capacity half capacity(mainline) half capacity(w/ patch) 972 (95%) 138ms not reachable 276ms 486 (47.5%) 30ms 138ms 60ms 256 (25%) 13ms 32ms 26ms On my hikey (octo Arm64 platform) with schedutil governor, the time to reach max OPP when starting from a null utilization, decreases from 223ms with current scale invariance down to 121ms with the new algorithm. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Morten.Rasmussen@arm.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bsegall@google.com Cc: dietmar.eggemann@arm.com Cc: patrick.bellasi@arm.com Cc: pjt@google.com Cc: pkondeti@codeaurora.org Cc: quentin.perret@arm.com Cc: rjw@rjwysocki.net Cc: srinivas.pandruvada@linux.intel.com Cc: thara.gopinath@linaro.org Link: https://lkml.kernel.org/r/1548257214-13745-3-git-send-email-vincent.guittot@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-01-23 23:26:53 +08:00
static inline u64 rq_clock_pelt(struct rq *rq)
{
return rq_clock_task(rq);
}
static inline void
update_rq_clock_pelt(struct rq *rq, s64 delta) { }
static inline void
update_idle_rq_clock_pelt(struct rq *rq) { }
sched/fair: Decay task PELT values during wakeup migration Before being migrated to a new CPU, a task sees its PELT values synchronized with rq last_update_time. Once done, that same task will also have its sched_avg last_update_time reset. This means the time between the migration and the last clock update will not be accounted for in util_avg and a discontinuity will appear. This issue is amplified by the PELT clock scaling. It takes currently one tick after the CPU being idle to let clock_pelt catching up clock_task. This is especially problematic for asymmetric CPU capacity systems which need stable util_avg signals for task placement and energy estimation. Ideally, this problem would be solved by updating the runqueue clocks before the migration. But that would require taking the runqueue lock which is quite expensive [1]. Instead estimate the missing time and update the task util_avg with that value. To that end, we need sched_clock_cpu() but it is a costly function. Limit the usage to the case where the source CPU is idle as we know this is when the clock is having the biggest risk of being outdated. See comment in migrate_se_pelt_lag() for more details about how the PELT value is estimated. Notice though this estimation doesn't take into account IRQ and Paravirt time. [1] https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Vincent Donnefort <vdonnefort@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Tested-by: Lukasz Luba <lukasz.luba@arm.com> Link: https://lkml.kernel.org/r/20220621090414.433602-3-vdonnefort@google.com
2022-06-21 17:04:09 +08:00
static inline void update_idle_cfs_rq_clock_pelt(struct cfs_rq *cfs_rq) { }
#endif