From nobody Mon Feb 9 06:01:42 2026 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4C4EEC77B7F for ; Fri, 12 May 2023 10:11:56 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S240495AbjELKLy (ORCPT ); Fri, 12 May 2023 06:11:54 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39316 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S240240AbjELKLn (ORCPT ); Fri, 12 May 2023 06:11:43 -0400 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by lindbergh.monkeyblade.net (Postfix) with ESMTP id E5BE211D8B for ; Fri, 12 May 2023 03:11:05 -0700 (PDT) Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 7F62F1063; Fri, 12 May 2023 03:11:28 -0700 (PDT) Received: from e125579.fritz.box (unknown [172.31.20.19]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 515533F5A1; Fri, 12 May 2023 03:10:42 -0700 (PDT) From: Dietmar Eggemann To: Ingo Molnar , Peter Zijlstra , Vincent Guittot Cc: Qais Yousef , Kajetan Puchalski , Morten Rasmussen , Vincent Donnefort , Quentin Perret , Abhijeet Dharmapurikar , linux-kernel@vger.kernel.org Subject: [PATCH v2 1/2] sched/fair: Refactor CPU utilization functions Date: Fri, 12 May 2023 12:10:28 +0200 Message-Id: <20230512101029.342823-2-dietmar.eggemann@arm.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20230512101029.342823-1-dietmar.eggemann@arm.com> References: <20230512101029.342823-1-dietmar.eggemann@arm.com> MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Type: text/plain; charset="utf-8" There is a lot of code duplication in cpu_util_next() & cpu_util_cfs(). Remove this by allowing cpu_util_next() to be called with p =3D NULL. Rename cpu_util_next() to cpu_util() since the '_next' suffix is no longer necessary to distinct cpu utilization related functions. Implement cpu_util_cfs(cpu) as cpu_util(cpu, p =3D NULL, -1). This will allow to code future related cpu util changes only in one place, namely in cpu_util(). Signed-off-by: Dietmar Eggemann --- kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++-------- kernel/sched/sched.h | 47 +----------------------------------- 2 files changed, 48 insertions(+), 56 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 3f8135d7c89d..1220cfbee258 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -7149,7 +7149,7 @@ static int select_idle_sibling(struct task_struct *p,= int prev, int target) * Predicts what cpu_util(@cpu) would return if @p was removed from @cpu * (@dst_cpu =3D -1) or migrated to @dst_cpu. */ -static unsigned long cpu_util_next(int cpu, struct task_struct *p, int dst= _cpu) +static unsigned long cpu_util(int cpu, struct task_struct *p, int dst_cpu) { struct cfs_rq *cfs_rq =3D &cpu_rq(cpu)->cfs; unsigned long util =3D READ_ONCE(cfs_rq->avg.util_avg); @@ -7160,9 +7160,9 @@ static unsigned long cpu_util_next(int cpu, struct ta= sk_struct *p, int dst_cpu) * contribution. In all the other cases @cpu is not impacted by the * migration so its util_avg is already correct. */ - if (task_cpu(p) =3D=3D cpu && dst_cpu !=3D cpu) + if (p && task_cpu(p) =3D=3D cpu && dst_cpu !=3D cpu) lsub_positive(&util, task_util(p)); - else if (task_cpu(p) !=3D cpu && dst_cpu =3D=3D cpu) + else if (p && task_cpu(p) !=3D cpu && dst_cpu =3D=3D cpu) util +=3D task_util(p); =20 if (sched_feat(UTIL_EST)) { @@ -7198,7 +7198,7 @@ static unsigned long cpu_util_next(int cpu, struct ta= sk_struct *p, int dst_cpu) */ if (dst_cpu =3D=3D cpu) util_est +=3D _task_util_est(p); - else if (unlikely(task_on_rq_queued(p) || current =3D=3D p)) + else if (p && unlikely(task_on_rq_queued(p) || current =3D=3D p)) lsub_positive(&util_est, _task_util_est(p)); =20 util =3D max(util, util_est); @@ -7207,6 +7207,43 @@ static unsigned long cpu_util_next(int cpu, struct t= ask_struct *p, int dst_cpu) return min(util, capacity_orig_of(cpu)); } =20 +/** + * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks. + * @cpu: the CPU to get the utilization for. + * + * The unit of the return value must be the same as the one of CPU capacity + * so that CPU utilization can be compared with CPU capacity. + * + * CPU utilization is the sum of running time of runnable tasks plus the + * recent utilization of currently non-runnable tasks on that CPU. + * It represents the amount of CPU capacity currently used by CFS tasks in + * the range [0..max CPU capacity] with max CPU capacity being the CPU + * capacity at f_max. + * + * The estimated CPU utilization is defined as the maximum between CPU + * utilization and sum of the estimated utilization of the currently + * runnable tasks on that CPU. It preserves a utilization "snapshot" of + * previously-executed tasks, which helps better deduce how busy a CPU will + * be when a long-sleeping task wakes up. The contribution to CPU utilizat= ion + * of such a task would be significantly decayed at this point of time. + * + * CPU utilization can be higher than the current CPU capacity + * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because + * of rounding errors as well as task migrations or wakeups of new tasks. + * CPU utilization has to be capped to fit into the [0..max CPU capacity] + * range. Otherwise a group of CPUs (CPU0 util =3D 121% + CPU1 util =3D 80= %) + * could be seen as over-utilized even though CPU1 has 20% of spare CPU + * capacity. CPU utilization is allowed to overshoot current CPU capacity + * though since this is useful for predicting the CPU capacity required + * after task migrations (scheduler-driven DVFS). + * + * Return: (Estimated) utilization for the specified CPU. + */ +unsigned long cpu_util_cfs(int cpu) +{ + return cpu_util(cpu, NULL, -1); +} + /* * cpu_util_without: compute cpu utilization without any contributions fro= m *p * @cpu: the CPU which utilization is requested @@ -7224,9 +7261,9 @@ static unsigned long cpu_util_without(int cpu, struct= task_struct *p) { /* Task has no contribution or is new */ if (cpu !=3D task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time)) - return cpu_util_cfs(cpu); + p =3D NULL; =20 - return cpu_util_next(cpu, p, -1); + return cpu_util(cpu, p, -1); } =20 /* @@ -7273,7 +7310,7 @@ static inline void eenv_task_busy_time(struct energy_= env *eenv, * cpu_capacity. * * The contribution of the task @p for which we want to estimate the - * energy cost is removed (by cpu_util_next()) and must be calculated + * energy cost is removed (by cpu_util()) and must be calculated * separately (see eenv_task_busy_time). This ensures: * * - A stable PD utilization, no matter which CPU of that PD we want to = place @@ -7294,7 +7331,7 @@ static inline void eenv_pd_busy_time(struct energy_en= v *eenv, int cpu; =20 for_each_cpu(cpu, pd_cpus) { - unsigned long util =3D cpu_util_next(cpu, p, -1); + unsigned long util =3D cpu_util(cpu, p, -1); =20 busy_time +=3D effective_cpu_util(cpu, util, ENERGY_UTIL, NULL); } @@ -7318,7 +7355,7 @@ eenv_pd_max_util(struct energy_env *eenv, struct cpum= ask *pd_cpus, =20 for_each_cpu(cpu, pd_cpus) { struct task_struct *tsk =3D (cpu =3D=3D dst_cpu) ? p : NULL; - unsigned long util =3D cpu_util_next(cpu, p, dst_cpu); + unsigned long util =3D cpu_util(cpu, p, dst_cpu); unsigned long cpu_util; =20 /* @@ -7464,7 +7501,7 @@ static int find_energy_efficient_cpu(struct task_stru= ct *p, int prev_cpu) if (!cpumask_test_cpu(cpu, p->cpus_ptr)) continue; =20 - util =3D cpu_util_next(cpu, p, cpu); + util =3D cpu_util(cpu, p, cpu); cpu_cap =3D capacity_of(cpu); =20 /* diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index ec7b3e0a2b20..f78c0f85cc76 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2946,53 +2946,8 @@ static inline unsigned long cpu_util_dl(struct rq *r= q) return READ_ONCE(rq->avg_dl.util_avg); } =20 -/** - * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks. - * @cpu: the CPU to get the utilization for. - * - * The unit of the return value must be the same as the one of CPU capacity - * so that CPU utilization can be compared with CPU capacity. - * - * CPU utilization is the sum of running time of runnable tasks plus the - * recent utilization of currently non-runnable tasks on that CPU. - * It represents the amount of CPU capacity currently used by CFS tasks in - * the range [0..max CPU capacity] with max CPU capacity being the CPU - * capacity at f_max. - * - * The estimated CPU utilization is defined as the maximum between CPU - * utilization and sum of the estimated utilization of the currently - * runnable tasks on that CPU. It preserves a utilization "snapshot" of - * previously-executed tasks, which helps better deduce how busy a CPU will - * be when a long-sleeping task wakes up. The contribution to CPU utilizat= ion - * of such a task would be significantly decayed at this point of time. - * - * CPU utilization can be higher than the current CPU capacity - * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because - * of rounding errors as well as task migrations or wakeups of new tasks. - * CPU utilization has to be capped to fit into the [0..max CPU capacity] - * range. Otherwise a group of CPUs (CPU0 util =3D 121% + CPU1 util =3D 80= %) - * could be seen as over-utilized even though CPU1 has 20% of spare CPU - * capacity. CPU utilization is allowed to overshoot current CPU capacity - * though since this is useful for predicting the CPU capacity required - * after task migrations (scheduler-driven DVFS). - * - * Return: (Estimated) utilization for the specified CPU. - */ -static inline unsigned long cpu_util_cfs(int cpu) -{ - struct cfs_rq *cfs_rq; - unsigned long util; - - cfs_rq =3D &cpu_rq(cpu)->cfs; - util =3D READ_ONCE(cfs_rq->avg.util_avg); =20 - if (sched_feat(UTIL_EST)) { - util =3D max_t(unsigned long, util, - READ_ONCE(cfs_rq->avg.util_est.enqueued)); - } - - return min(util, capacity_orig_of(cpu)); -} +extern unsigned long cpu_util_cfs(int cpu); =20 static inline unsigned long cpu_util_rt(struct rq *rq) { --=20 2.25.1 From nobody Mon Feb 9 06:01:42 2026 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 93907C77B75 for ; Fri, 12 May 2023 10:11:51 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S240425AbjELKLt (ORCPT ); Fri, 12 May 2023 06:11:49 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39336 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S240134AbjELKLn (ORCPT ); Fri, 12 May 2023 06:11:43 -0400 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by lindbergh.monkeyblade.net (Postfix) with ESMTP id 80C3F11B66 for ; Fri, 12 May 2023 03:11:09 -0700 (PDT) Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 61EC9139F; Fri, 12 May 2023 03:11:30 -0700 (PDT) Received: from e125579.fritz.box (unknown [172.31.20.19]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 3638D3F5A1; Fri, 12 May 2023 03:10:44 -0700 (PDT) From: Dietmar Eggemann To: Ingo Molnar , Peter Zijlstra , Vincent Guittot Cc: Qais Yousef , Kajetan Puchalski , Morten Rasmussen , Vincent Donnefort , Quentin Perret , Abhijeet Dharmapurikar , linux-kernel@vger.kernel.org Subject: [PATCH v2 2/2] sched/fair, cpufreq: Introduce 'runnable boosting' Date: Fri, 12 May 2023 12:10:29 +0200 Message-Id: <20230512101029.342823-3-dietmar.eggemann@arm.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20230512101029.342823-1-dietmar.eggemann@arm.com> References: <20230512101029.342823-1-dietmar.eggemann@arm.com> MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Type: text/plain; charset="utf-8" The responsiveness of the Per Entity Load Tracking (PELT) util_avg in mobile devices is still considered too low for utilization changes during task ramp-up. In Android this manifests in the fact that the first frames of a UI activity are very prone to be jankframes (a frame which doesn't meet the required frame rendering time, e.g. 16ms@60Hz) since the CPU frequency is normally low at this point and has to ramp up quickly. The beginning of an UI activity is also characterized by the occurrence of CPU contention, especially on little CPUs. Current little CPUs can have an original CPU capacity of only ~ 150 which means that the actual CPU capacity at lower frequency can even be much smaller. Schedutil maps CPU util_avg into CPU frequency request via: util =3D effective_cpu_util(..., cpu_util_cfs(cpu), ...) -> util =3D map_util_perf(util) -> freq =3D map_util_freq(util, ...) CPU contention for CFS tasks can be detected by the 'CPU runnable_avg > util_avg' condition in cpu_util_cfs(..., boost) -> cpu_util(..., boost). Schedutil activates 'runnable boosting' by setting the new parameter 'boost =3D 1'. To be in sync with schedutil's CPU frequency selection, Energy Aware Scheduling (EAS) also calls cpu_util(..., boost =3D 1) during max util detection. Moreover, 'runnable boosting' is also used in load-balance for busiest CPU selection when the migration type is 'migrate_util', i.e. only at sched domains which don't have the SD_SHARE_PKG_RESOURCES flag set. Suggested-by: Vincent Guittot Signed-off-by: Dietmar Eggemann --- kernel/sched/core.c | 2 +- kernel/sched/cpufreq_schedutil.c | 3 ++- kernel/sched/fair.c | 29 ++++++++++++++++++----------- kernel/sched/sched.h | 4 ++-- 4 files changed, 23 insertions(+), 15 deletions(-) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 898fa3bc2765..8b776db1d24c 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -7441,7 +7441,7 @@ unsigned long effective_cpu_util(int cpu, unsigned lo= ng util_cfs, =20 unsigned long sched_cpu_util(int cpu) { - return effective_cpu_util(cpu, cpu_util_cfs(cpu), ENERGY_UTIL, NULL); + return effective_cpu_util(cpu, cpu_util_cfs(cpu, 0), ENERGY_UTIL, NULL); } #endif /* CONFIG_SMP */ =20 diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedu= til.c index e3211455b203..3b902f533214 100644 --- a/kernel/sched/cpufreq_schedutil.c +++ b/kernel/sched/cpufreq_schedutil.c @@ -155,10 +155,11 @@ static unsigned int get_next_freq(struct sugov_policy= *sg_policy, =20 static void sugov_get_util(struct sugov_cpu *sg_cpu) { + unsigned long util =3D cpu_util_cfs(sg_cpu->cpu, 1); struct rq *rq =3D cpu_rq(sg_cpu->cpu); =20 sg_cpu->bw_dl =3D cpu_bw_dl(rq); - sg_cpu->util =3D effective_cpu_util(sg_cpu->cpu, cpu_util_cfs(sg_cpu->cpu= ), + sg_cpu->util =3D effective_cpu_util(sg_cpu->cpu, util, FREQUENCY_UTIL, NULL); } =20 diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 1220cfbee258..3a10fe5988d6 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1800,7 +1800,7 @@ static void update_numa_stats(struct task_numa_env *e= nv, =20 ns->load +=3D cpu_load(rq); ns->runnable +=3D cpu_runnable(rq); - ns->util +=3D cpu_util_cfs(cpu); + ns->util +=3D cpu_util_cfs(cpu, 0); ns->nr_running +=3D rq->cfs.h_nr_running; ns->compute_capacity +=3D capacity_of(cpu); =20 @@ -6184,9 +6184,10 @@ static inline bool cpu_overutilized(int cpu) { unsigned long rq_util_min =3D uclamp_rq_get(cpu_rq(cpu), UCLAMP_MIN); unsigned long rq_util_max =3D uclamp_rq_get(cpu_rq(cpu), UCLAMP_MAX); + unsigned long util =3D cpu_util_cfs(cpu, 0); =20 /* Return true only if the utilization doesn't fit CPU's capacity */ - return !util_fits_cpu(cpu_util_cfs(cpu), rq_util_min, rq_util_max, cpu); + return !util_fits_cpu(util, rq_util_min, rq_util_max, cpu); } =20 static inline void update_overutilized_status(struct rq *rq) @@ -7149,10 +7150,15 @@ static int select_idle_sibling(struct task_struct *= p, int prev, int target) * Predicts what cpu_util(@cpu) would return if @p was removed from @cpu * (@dst_cpu =3D -1) or migrated to @dst_cpu. */ -static unsigned long cpu_util(int cpu, struct task_struct *p, int dst_cpu) +static unsigned long +cpu_util(int cpu, struct task_struct *p, int dst_cpu, int boost) { struct cfs_rq *cfs_rq =3D &cpu_rq(cpu)->cfs; unsigned long util =3D READ_ONCE(cfs_rq->avg.util_avg); + unsigned long runnable; + + runnable =3D boost ? READ_ONCE(cfs_rq->avg.runnable_avg) : 0; + util =3D max(util, runnable); =20 /* * If @dst_cpu is -1 or @p migrates from @cpu to @dst_cpu remove its @@ -7169,6 +7175,7 @@ static unsigned long cpu_util(int cpu, struct task_st= ruct *p, int dst_cpu) unsigned long util_est; =20 util_est =3D READ_ONCE(cfs_rq->avg.util_est.enqueued); + util_est =3D max(util_est, runnable); =20 /* * During wake-up @p isn't enqueued yet and doesn't contribute @@ -7239,9 +7246,9 @@ static unsigned long cpu_util(int cpu, struct task_st= ruct *p, int dst_cpu) * * Return: (Estimated) utilization for the specified CPU. */ -unsigned long cpu_util_cfs(int cpu) +unsigned long cpu_util_cfs(int cpu, int boost) { - return cpu_util(cpu, NULL, -1); + return cpu_util(cpu, NULL, -1, boost); } =20 /* @@ -7263,7 +7270,7 @@ static unsigned long cpu_util_without(int cpu, struct= task_struct *p) if (cpu !=3D task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time)) p =3D NULL; =20 - return cpu_util(cpu, p, -1); + return cpu_util(cpu, p, -1, 0); } =20 /* @@ -7331,7 +7338,7 @@ static inline void eenv_pd_busy_time(struct energy_en= v *eenv, int cpu; =20 for_each_cpu(cpu, pd_cpus) { - unsigned long util =3D cpu_util(cpu, p, -1); + unsigned long util =3D cpu_util(cpu, p, -1, 0); =20 busy_time +=3D effective_cpu_util(cpu, util, ENERGY_UTIL, NULL); } @@ -7355,7 +7362,7 @@ eenv_pd_max_util(struct energy_env *eenv, struct cpum= ask *pd_cpus, =20 for_each_cpu(cpu, pd_cpus) { struct task_struct *tsk =3D (cpu =3D=3D dst_cpu) ? p : NULL; - unsigned long util =3D cpu_util(cpu, p, dst_cpu); + unsigned long util =3D cpu_util(cpu, p, dst_cpu, 1); unsigned long cpu_util; =20 /* @@ -7501,7 +7508,7 @@ static int find_energy_efficient_cpu(struct task_stru= ct *p, int prev_cpu) if (!cpumask_test_cpu(cpu, p->cpus_ptr)) continue; =20 - util =3D cpu_util(cpu, p, cpu); + util =3D cpu_util(cpu, p, cpu, 0); cpu_cap =3D capacity_of(cpu); =20 /* @@ -9443,7 +9450,7 @@ static inline void update_sg_lb_stats(struct lb_env *= env, unsigned long load =3D cpu_load(rq); =20 sgs->group_load +=3D load; - sgs->group_util +=3D cpu_util_cfs(i); + sgs->group_util +=3D cpu_util_cfs(i, 0); sgs->group_runnable +=3D cpu_runnable(rq); sgs->sum_h_nr_running +=3D rq->cfs.h_nr_running; =20 @@ -10561,7 +10568,7 @@ static struct rq *find_busiest_queue(struct lb_env = *env, break; =20 case migrate_util: - util =3D cpu_util_cfs(i); + util =3D cpu_util_cfs(i, 1); =20 /* * Don't try to pull utilization from a CPU with one diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index f78c0f85cc76..b4706874eec1 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2947,7 +2947,7 @@ static inline unsigned long cpu_util_dl(struct rq *rq) } =20 =20 -extern unsigned long cpu_util_cfs(int cpu); +extern unsigned long cpu_util_cfs(int cpu, int boost); =20 static inline unsigned long cpu_util_rt(struct rq *rq) { @@ -3037,7 +3037,7 @@ static inline bool uclamp_rq_is_capped(struct rq *rq) if (!static_branch_likely(&sched_uclamp_used)) return false; =20 - rq_util =3D cpu_util_cfs(cpu_of(rq)) + cpu_util_rt(rq); + rq_util =3D cpu_util_cfs(cpu_of(rq), 0) + cpu_util_rt(rq); max_util =3D READ_ONCE(rq->uclamp[UCLAMP_MAX].value); =20 return max_util !=3D SCHED_CAPACITY_SCALE && rq_util >=3D max_util; --=20 2.25.1