From d6111cf45c5787282b2e20d77bdb6b28881d516a Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Tue, 31 Oct 2023 11:12:01 -0700 Subject: [PATCH 01/30] sched: Use WRITE_ONCE() for p->on_rq Since RCU-tasks uses READ_ONCE(p->on_rq), ensure the write-side matches with WRITE_ONCE(). Signed-off-by: "Paul E. McKenney" Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/e4896e0b-eacc-45a2-a7a8-de2280a51ecc@paulmck-laptop --- kernel/sched/core.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index a708d225c28e..9d5099d02dbc 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -2124,12 +2124,14 @@ void activate_task(struct rq *rq, struct task_struct *p, int flags) enqueue_task(rq, p, flags); - p->on_rq = TASK_ON_RQ_QUEUED; + WRITE_ONCE(p->on_rq, TASK_ON_RQ_QUEUED); + ASSERT_EXCLUSIVE_WRITER(p->on_rq); } void deactivate_task(struct rq *rq, struct task_struct *p, int flags) { - p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; + WRITE_ONCE(p->on_rq, (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING); + ASSERT_EXCLUSIVE_WRITER(p->on_rq); dequeue_task(rq, p, flags); } From 84db47ca7146d7bd00eb5cf2b93989a971c84650 Mon Sep 17 00:00:00 2001 From: Raghavendra K T Date: Fri, 20 Oct 2023 21:27:46 +0530 Subject: [PATCH 02/30] sched/numa: Fix mm numa_scan_seq based unconditional scan Since commit fc137c0ddab2 ("sched/numa: enhance vma scanning logic") NUMA Balancing allows updating PTEs to trap NUMA hinting faults if the task had previously accessed VMA. However unconditional scan of VMAs are allowed during initial phase of VMA creation until process's mm numa_scan_seq reaches 2 even though current task had not accessed VMA. Rationale: - Without initial scan subsequent PTE update may never happen. - Give fair opportunity to all the VMAs to be scanned and subsequently understand the access pattern of all the VMAs. But it has a corner case where, if a VMA is created after some time, process's mm numa_scan_seq could be already greater than 2. For e.g., values of mm numa_scan_seq when VMAs are created by running mmtest autonuma benchmark briefly looks like: start_seq=0 : 459 start_seq=2 : 138 start_seq=3 : 144 start_seq=4 : 8 start_seq=8 : 1 start_seq=9 : 1 This results in no unconditional PTE updates for those VMAs created after some time. Fix: - Note down the initial value of mm numa_scan_seq in per VMA start_seq. - Allow unconditional scan till start_seq + 2. Result: SUT: AMD EPYC Milan with 2 NUMA nodes 256 cpus. base kernel: upstream 6.6-rc6 with Mels patches [1] applied. kernbench ========== base patched %gain Amean elsp-128 165.09 ( 0.00%) 164.78 * 0.19%* Duration User 41404.28 41375.08 Duration System 9862.22 9768.48 Duration Elapsed 519.87 518.72 Ops NUMA PTE updates 1041416.00 831536.00 Ops NUMA hint faults 263296.00 220966.00 Ops NUMA pages migrated 258021.00 212769.00 Ops AutoNUMA cost 1328.67 1114.69 autonumabench NUMA01_THREADLOCAL ================== Amean elsp-NUMA01_THREADLOCAL 81.79 (0.00%) 67.74 * 17.18%* Duration User 54832.73 47379.67 Duration System 75.00 185.75 Duration Elapsed 576.72 476.09 Ops NUMA PTE updates 394429.00 11121044.00 Ops NUMA hint faults 1001.00 8906404.00 Ops NUMA pages migrated 288.00 2998694.00 Ops AutoNUMA cost 7.77 44666.84 Signed-off-by: Raghavendra K T Signed-off-by: Peter Zijlstra (Intel) Acked-by: Mel Gorman Link: https://lore.kernel.org/r/2ea7cbce80ac7c62e90cbfb9653a7972f902439f.1697816692.git.raghavendra.kt@amd.com --- include/linux/mm_types.h | 3 +++ kernel/sched/fair.c | 4 +++- 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 957ce38768b2..950df415d7de 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -600,6 +600,9 @@ struct vma_numab_state { */ unsigned long pids_active[2]; + /* MM scan sequence ID when scan first started after VMA creation */ + int start_scan_seq; + /* * MM scan sequence ID when the VMA was last completely scanned. * A VMA is not eligible for scanning if prev_scan_seq == numa_scan_seq diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index d7a3c63a2171..44b5262b6657 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -3164,7 +3164,7 @@ static bool vma_is_accessed(struct mm_struct *mm, struct vm_area_struct *vma) * This is also done to avoid any side effect of task scanning * amplifying the unfairness of disjoint set of VMAs' access. */ - if (READ_ONCE(current->mm->numa_scan_seq) < 2) + if ((READ_ONCE(current->mm->numa_scan_seq) - vma->numab_state->start_scan_seq) < 2) return true; pids = vma->numab_state->pids_active[0] | vma->numab_state->pids_active[1]; @@ -3307,6 +3307,8 @@ retry_pids: if (!vma->numab_state) continue; + vma->numab_state->start_scan_seq = mm->numa_scan_seq; + vma->numab_state->next_scan = now + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); From 2227a957e1d5b1941be4e4207879ec74f4bb37f8 Mon Sep 17 00:00:00 2001 From: Abel Wu Date: Wed, 15 Nov 2023 11:36:45 +0800 Subject: [PATCH 03/30] sched/eevdf: Sort the rbtree by virtual deadline Sort the task timeline by virtual deadline and keep the min_vruntime in the augmented tree, so we can avoid doubling the worst case cost and make full use of the cached leftmost node to enable O(1) fastpath picking in next patch. Signed-off-by: Abel Wu Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20231115033647.80785-3-wuyun.abel@bytedance.com --- include/linux/sched.h | 2 +- kernel/sched/debug.c | 11 ++- kernel/sched/fair.c | 170 +++++++++++++++++------------------------- kernel/sched/sched.h | 1 + 4 files changed, 78 insertions(+), 106 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index 292c31697248..cd56d4018527 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -553,7 +553,7 @@ struct sched_entity { struct load_weight load; struct rb_node run_node; u64 deadline; - u64 min_deadline; + u64 min_vruntime; struct list_head group_node; unsigned int on_rq; diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c index 4580a450700e..168eecc209b4 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -628,8 +628,8 @@ static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) { - s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, spread; - struct sched_entity *last, *first; + s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread; + struct sched_entity *last, *first, *root; struct rq *rq = cpu_rq(cpu); unsigned long flags; @@ -644,15 +644,20 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) SPLIT_NS(cfs_rq->exec_clock)); raw_spin_rq_lock_irqsave(rq, flags); + root = __pick_root_entity(cfs_rq); + if (root) + left_vruntime = root->min_vruntime; first = __pick_first_entity(cfs_rq); if (first) - left_vruntime = first->vruntime; + left_deadline = first->deadline; last = __pick_last_entity(cfs_rq); if (last) right_vruntime = last->vruntime; min_vruntime = cfs_rq->min_vruntime; raw_spin_rq_unlock_irqrestore(rq, flags); + SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline", + SPLIT_NS(left_deadline)); SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime", SPLIT_NS(left_vruntime)); SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 44b5262b6657..31bca05c3612 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -551,7 +551,11 @@ static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) static inline bool entity_before(const struct sched_entity *a, const struct sched_entity *b) { - return (s64)(a->vruntime - b->vruntime) < 0; + /* + * Tiebreak on vruntime seems unnecessary since it can + * hardly happen. + */ + return (s64)(a->deadline - b->deadline) < 0; } static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) @@ -720,7 +724,7 @@ static void update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se) * Note: using 'avg_vruntime() > se->vruntime' is inacurate due * to the loss in precision caused by the division. */ -int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se) +static int vruntime_eligible(struct cfs_rq *cfs_rq, u64 vruntime) { struct sched_entity *curr = cfs_rq->curr; s64 avg = cfs_rq->avg_vruntime; @@ -733,7 +737,12 @@ int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se) load += weight; } - return avg >= entity_key(cfs_rq, se) * load; + return avg >= (s64)(vruntime - cfs_rq->min_vruntime) * load; +} + +int entity_eligible(struct cfs_rq *cfs_rq, struct sched_entity *se) +{ + return vruntime_eligible(cfs_rq, se->vruntime); } static u64 __update_min_vruntime(struct cfs_rq *cfs_rq, u64 vruntime) @@ -752,9 +761,8 @@ static u64 __update_min_vruntime(struct cfs_rq *cfs_rq, u64 vruntime) static void update_min_vruntime(struct cfs_rq *cfs_rq) { - struct sched_entity *se = __pick_first_entity(cfs_rq); + struct sched_entity *se = __pick_root_entity(cfs_rq); struct sched_entity *curr = cfs_rq->curr; - u64 vruntime = cfs_rq->min_vruntime; if (curr) { @@ -766,9 +774,9 @@ static void update_min_vruntime(struct cfs_rq *cfs_rq) if (se) { if (!curr) - vruntime = se->vruntime; + vruntime = se->min_vruntime; else - vruntime = min_vruntime(vruntime, se->vruntime); + vruntime = min_vruntime(vruntime, se->min_vruntime); } /* ensure we never gain time by being placed backwards. */ @@ -781,34 +789,34 @@ static inline bool __entity_less(struct rb_node *a, const struct rb_node *b) return entity_before(__node_2_se(a), __node_2_se(b)); } -#define deadline_gt(field, lse, rse) ({ (s64)((lse)->field - (rse)->field) > 0; }) +#define vruntime_gt(field, lse, rse) ({ (s64)((lse)->field - (rse)->field) > 0; }) -static inline void __update_min_deadline(struct sched_entity *se, struct rb_node *node) +static inline void __min_vruntime_update(struct sched_entity *se, struct rb_node *node) { if (node) { struct sched_entity *rse = __node_2_se(node); - if (deadline_gt(min_deadline, se, rse)) - se->min_deadline = rse->min_deadline; + if (vruntime_gt(min_vruntime, se, rse)) + se->min_vruntime = rse->min_vruntime; } } /* - * se->min_deadline = min(se->deadline, left->min_deadline, right->min_deadline) + * se->min_vruntime = min(se->vruntime, {left,right}->min_vruntime) */ -static inline bool min_deadline_update(struct sched_entity *se, bool exit) +static inline bool min_vruntime_update(struct sched_entity *se, bool exit) { - u64 old_min_deadline = se->min_deadline; + u64 old_min_vruntime = se->min_vruntime; struct rb_node *node = &se->run_node; - se->min_deadline = se->deadline; - __update_min_deadline(se, node->rb_right); - __update_min_deadline(se, node->rb_left); + se->min_vruntime = se->vruntime; + __min_vruntime_update(se, node->rb_right); + __min_vruntime_update(se, node->rb_left); - return se->min_deadline == old_min_deadline; + return se->min_vruntime == old_min_vruntime; } -RB_DECLARE_CALLBACKS(static, min_deadline_cb, struct sched_entity, - run_node, min_deadline, min_deadline_update); +RB_DECLARE_CALLBACKS(static, min_vruntime_cb, struct sched_entity, + run_node, min_vruntime, min_vruntime_update); /* * Enqueue an entity into the rb-tree: @@ -816,18 +824,28 @@ RB_DECLARE_CALLBACKS(static, min_deadline_cb, struct sched_entity, static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) { avg_vruntime_add(cfs_rq, se); - se->min_deadline = se->deadline; + se->min_vruntime = se->vruntime; rb_add_augmented_cached(&se->run_node, &cfs_rq->tasks_timeline, - __entity_less, &min_deadline_cb); + __entity_less, &min_vruntime_cb); } static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) { rb_erase_augmented_cached(&se->run_node, &cfs_rq->tasks_timeline, - &min_deadline_cb); + &min_vruntime_cb); avg_vruntime_sub(cfs_rq, se); } +struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq) +{ + struct rb_node *root = cfs_rq->tasks_timeline.rb_root.rb_node; + + if (!root) + return NULL; + + return __node_2_se(root); +} + struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq) { struct rb_node *left = rb_first_cached(&cfs_rq->tasks_timeline); @@ -850,23 +868,28 @@ struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq) * with the earliest virtual deadline. * * We can do this in O(log n) time due to an augmented RB-tree. The - * tree keeps the entries sorted on service, but also functions as a - * heap based on the deadline by keeping: + * tree keeps the entries sorted on deadline, but also functions as a + * heap based on the vruntime by keeping: * - * se->min_deadline = min(se->deadline, se->{left,right}->min_deadline) + * se->min_vruntime = min(se->vruntime, se->{left,right}->min_vruntime) * - * Which allows an EDF like search on (sub)trees. + * Which allows tree pruning through eligibility. */ -static struct sched_entity *__pick_eevdf(struct cfs_rq *cfs_rq) +static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) { struct rb_node *node = cfs_rq->tasks_timeline.rb_root.rb_node; struct sched_entity *curr = cfs_rq->curr; struct sched_entity *best = NULL; - struct sched_entity *best_left = NULL; + + /* + * We can safely skip eligibility check if there is only one entity + * in this cfs_rq, saving some cycles. + */ + if (cfs_rq->nr_running == 1) + return curr && curr->on_rq ? curr : __node_2_se(node); if (curr && (!curr->on_rq || !entity_eligible(cfs_rq, curr))) curr = NULL; - best = curr; /* * Once selected, run a task until it either becomes non-eligible or @@ -875,95 +898,38 @@ static struct sched_entity *__pick_eevdf(struct cfs_rq *cfs_rq) if (sched_feat(RUN_TO_PARITY) && curr && curr->vlag == curr->deadline) return curr; + /* Heap search for the EEVD entity */ while (node) { struct sched_entity *se = __node_2_se(node); + struct rb_node *left = node->rb_left; /* - * If this entity is not eligible, try the left subtree. + * Eligible entities in left subtree are always better + * choices, since they have earlier deadlines. */ - if (!entity_eligible(cfs_rq, se)) { - node = node->rb_left; + if (left && vruntime_eligible(cfs_rq, + __node_2_se(left)->min_vruntime)) { + node = left; continue; } /* - * Now we heap search eligible trees for the best (min_)deadline + * The left subtree either is empty or has no eligible + * entity, so check the current node since it is the one + * with earliest deadline that might be eligible. */ - if (!best || deadline_gt(deadline, best, se)) + if (entity_eligible(cfs_rq, se)) { best = se; - - /* - * Every se in a left branch is eligible, keep track of the - * branch with the best min_deadline - */ - if (node->rb_left) { - struct sched_entity *left = __node_2_se(node->rb_left); - - if (!best_left || deadline_gt(min_deadline, best_left, left)) - best_left = left; - - /* - * min_deadline is in the left branch. rb_left and all - * descendants are eligible, so immediately switch to the second - * loop. - */ - if (left->min_deadline == se->min_deadline) - break; - } - - /* min_deadline is at this node, no need to look right */ - if (se->deadline == se->min_deadline) break; + } - /* else min_deadline is in the right branch. */ node = node->rb_right; } - /* - * We ran into an eligible node which is itself the best. - * (Or nr_running == 0 and both are NULL) - */ - if (!best_left || (s64)(best_left->min_deadline - best->deadline) > 0) - return best; + if (!best || (curr && entity_before(curr, best))) + best = curr; - /* - * Now best_left and all of its children are eligible, and we are just - * looking for deadline == min_deadline - */ - node = &best_left->run_node; - while (node) { - struct sched_entity *se = __node_2_se(node); - - /* min_deadline is the current node */ - if (se->deadline == se->min_deadline) - return se; - - /* min_deadline is in the left branch */ - if (node->rb_left && - __node_2_se(node->rb_left)->min_deadline == se->min_deadline) { - node = node->rb_left; - continue; - } - - /* else min_deadline is in the right branch */ - node = node->rb_right; - } - return NULL; -} - -static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) -{ - struct sched_entity *se = __pick_eevdf(cfs_rq); - - if (!se) { - struct sched_entity *left = __pick_first_entity(cfs_rq); - if (left) { - pr_err("EEVDF scheduling fail, picking leftmost\n"); - return left; - } - } - - return se; + return best; } #ifdef CONFIG_SCHED_DEBUG diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 2e5a95486a42..539c7e763f15 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2822,6 +2822,7 @@ DEFINE_LOCK_GUARD_2(double_rq_lock, struct rq, double_rq_lock(_T->lock, _T->lock2), double_rq_unlock(_T->lock, _T->lock2)) +extern struct sched_entity *__pick_root_entity(struct cfs_rq *cfs_rq); extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); From ee4373dc902c0a403dd084b254ce70a78f95466f Mon Sep 17 00:00:00 2001 From: Abel Wu Date: Wed, 15 Nov 2023 11:36:46 +0800 Subject: [PATCH 04/30] sched/eevdf: O(1) fastpath for task selection Since the RB-tree is now sorted by deadline, let's first try the leftmost entity which has the earliest virtual deadline. I've done some benchmarks to see its effectiveness. All the benchmarks are done inside a normal cpu cgroup in a clean environment with cpu turbo disabled, on a dual-CPU Intel Xeon(R) Platinum 8260 with 2 NUMA nodes each of which has 24C/48T. hackbench: process/thread + pipe/socket + 1/2/4/8 groups netperf: TCP/UDP + STREAM/RR + 24/48/72/96/192 threads tbench: loopback 24/48/72/96/192 threads schbench: 1/2/4/8 mthreads direct: cfs_rq has only one entity parity: RUN_TO_PARITY fast: O(1) fastpath slow: heap search (%) direct parity fast slow hackbench 92.95 2.02 4.91 0.12 netperf 68.08 6.60 24.18 1.14 tbench 67.55 11.22 20.61 0.62 schbench 69.91 2.65 25.73 1.71 The above results indicate that this fastpath really makes task selection more efficient. Signed-off-by: Abel Wu Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20231115033647.80785-4-wuyun.abel@bytedance.com --- kernel/sched/fair.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 31bca05c3612..d3e045d80cab 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -878,6 +878,7 @@ struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq) static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) { struct rb_node *node = cfs_rq->tasks_timeline.rb_root.rb_node; + struct sched_entity *se = __pick_first_entity(cfs_rq); struct sched_entity *curr = cfs_rq->curr; struct sched_entity *best = NULL; @@ -886,7 +887,7 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) * in this cfs_rq, saving some cycles. */ if (cfs_rq->nr_running == 1) - return curr && curr->on_rq ? curr : __node_2_se(node); + return curr && curr->on_rq ? curr : se; if (curr && (!curr->on_rq || !entity_eligible(cfs_rq, curr))) curr = NULL; @@ -898,9 +899,14 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) if (sched_feat(RUN_TO_PARITY) && curr && curr->vlag == curr->deadline) return curr; + /* Pick the leftmost entity if it's eligible */ + if (se && entity_eligible(cfs_rq, se)) { + best = se; + goto found; + } + /* Heap search for the EEVD entity */ while (node) { - struct sched_entity *se = __node_2_se(node); struct rb_node *left = node->rb_left; /* @@ -913,6 +919,8 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) continue; } + se = __node_2_se(node); + /* * The left subtree either is empty or has no eligible * entity, so check the current node since it is the one @@ -925,7 +933,7 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) node = node->rb_right; } - +found: if (!best || (curr && entity_before(curr, best))) best = curr; From 5d69eca542ee17c618f9a55da52191d5e28b435f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 4 Nov 2023 11:59:18 +0100 Subject: [PATCH 05/30] sched: Unify runtime accounting across classes All classes use sched_entity::exec_start to track runtime and have copies of the exact same code around to compute runtime. Collapse all that. Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Daniel Bristot de Oliveira Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Phil Auld Reviewed-by: Valentin Schneider Reviewed-by: Steven Rostedt (Google) Link: https://lkml.kernel.org/r/54d148a144f26d9559698c4dd82d8859038a7380.1699095159.git.bristot@kernel.org --- include/linux/sched.h | 2 +- kernel/sched/deadline.c | 15 +++-------- kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++---------- kernel/sched/rt.c | 15 +++-------- kernel/sched/sched.h | 12 ++------- kernel/sched/stop_task.c | 13 +-------- 6 files changed, 53 insertions(+), 61 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index cd56d4018527..44b46d9743bf 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -523,7 +523,7 @@ struct sched_statistics { u64 block_max; s64 sum_block_runtime; - u64 exec_max; + s64 exec_max; u64 slice_max; u64 nr_migrations_cold; diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index b28114478b82..de79719c63c0 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -1275,9 +1275,8 @@ static void update_curr_dl(struct rq *rq) { struct task_struct *curr = rq->curr; struct sched_dl_entity *dl_se = &curr->dl; - u64 delta_exec, scaled_delta_exec; + s64 delta_exec, scaled_delta_exec; int cpu = cpu_of(rq); - u64 now; if (!dl_task(curr) || !on_dl_rq(dl_se)) return; @@ -1290,21 +1289,13 @@ static void update_curr_dl(struct rq *rq) * natural solution, but the full ramifications of this * approach need further study. */ - now = rq_clock_task(rq); - delta_exec = now - curr->se.exec_start; - if (unlikely((s64)delta_exec <= 0)) { + delta_exec = update_curr_common(rq); + if (unlikely(delta_exec <= 0)) { if (unlikely(dl_se->dl_yielded)) goto throttle; return; } - schedstat_set(curr->stats.exec_max, - max(curr->stats.exec_max, delta_exec)); - - trace_sched_stat_runtime(curr, delta_exec, 0); - - update_current_exec_runtime(curr, now, delta_exec); - if (dl_entity_is_special(dl_se)) return; diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index d3e045d80cab..11073cf00134 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1103,23 +1103,17 @@ static void update_tg_load_avg(struct cfs_rq *cfs_rq) } #endif /* CONFIG_SMP */ -/* - * Update the current task's runtime statistics. - */ -static void update_curr(struct cfs_rq *cfs_rq) +static s64 update_curr_se(struct rq *rq, struct sched_entity *curr) { - struct sched_entity *curr = cfs_rq->curr; - u64 now = rq_clock_task(rq_of(cfs_rq)); - u64 delta_exec; - - if (unlikely(!curr)) - return; + u64 now = rq_clock_task(rq); + s64 delta_exec; delta_exec = now - curr->exec_start; - if (unlikely((s64)delta_exec <= 0)) - return; + if (unlikely(delta_exec <= 0)) + return delta_exec; curr->exec_start = now; + curr->sum_exec_runtime += delta_exec; if (schedstat_enabled()) { struct sched_statistics *stats; @@ -1129,8 +1123,43 @@ static void update_curr(struct cfs_rq *cfs_rq) max(delta_exec, stats->exec_max)); } - curr->sum_exec_runtime += delta_exec; - schedstat_add(cfs_rq->exec_clock, delta_exec); + return delta_exec; +} + +/* + * Used by other classes to account runtime. + */ +s64 update_curr_common(struct rq *rq) +{ + struct task_struct *curr = rq->curr; + s64 delta_exec; + + delta_exec = update_curr_se(rq, &curr->se); + if (unlikely(delta_exec <= 0)) + return delta_exec; + + trace_sched_stat_runtime(curr, delta_exec, 0); + + account_group_exec_runtime(curr, delta_exec); + cgroup_account_cputime(curr, delta_exec); + + return delta_exec; +} + +/* + * Update the current task's runtime statistics. + */ +static void update_curr(struct cfs_rq *cfs_rq) +{ + struct sched_entity *curr = cfs_rq->curr; + s64 delta_exec; + + if (unlikely(!curr)) + return; + + delta_exec = update_curr_se(rq_of(cfs_rq), curr); + if (unlikely(delta_exec <= 0)) + return; curr->vruntime += calc_delta_fair(delta_exec, curr); update_deadline(cfs_rq, curr); diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 6aaf0a3d6081..3261b067b67e 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -1002,24 +1002,15 @@ static void update_curr_rt(struct rq *rq) { struct task_struct *curr = rq->curr; struct sched_rt_entity *rt_se = &curr->rt; - u64 delta_exec; - u64 now; + s64 delta_exec; if (curr->sched_class != &rt_sched_class) return; - now = rq_clock_task(rq); - delta_exec = now - curr->se.exec_start; - if (unlikely((s64)delta_exec <= 0)) + delta_exec = update_curr_common(rq); + if (unlikely(delta_exec <= 0)) return; - schedstat_set(curr->stats.exec_max, - max(curr->stats.exec_max, delta_exec)); - - trace_sched_stat_runtime(curr, delta_exec, 0); - - update_current_exec_runtime(curr, now, delta_exec); - if (!rt_bandwidth_enabled()) return; diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 539c7e763f15..6703e9e81b1d 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2212,6 +2212,8 @@ struct affinity_context { unsigned int flags; }; +extern s64 update_curr_common(struct rq *rq); + struct sched_class { #ifdef CONFIG_UCLAMP_TASK @@ -3262,16 +3264,6 @@ extern int sched_dynamic_mode(const char *str); extern void sched_dynamic_update(int mode); #endif -static inline void update_current_exec_runtime(struct task_struct *curr, - u64 now, u64 delta_exec) -{ - curr->se.sum_exec_runtime += delta_exec; - account_group_exec_runtime(curr, delta_exec); - - curr->se.exec_start = now; - cgroup_account_cputime(curr, delta_exec); -} - #ifdef CONFIG_SCHED_MM_CID #define SCHED_MM_CID_PERIOD_NS (100ULL * 1000000) /* 100ms */ diff --git a/kernel/sched/stop_task.c b/kernel/sched/stop_task.c index 6cf7304e6449..b1b8fe61c532 100644 --- a/kernel/sched/stop_task.c +++ b/kernel/sched/stop_task.c @@ -70,18 +70,7 @@ static void yield_task_stop(struct rq *rq) static void put_prev_task_stop(struct rq *rq, struct task_struct *prev) { - struct task_struct *curr = rq->curr; - u64 now, delta_exec; - - now = rq_clock_task(rq); - delta_exec = now - curr->se.exec_start; - if (unlikely((s64)delta_exec < 0)) - delta_exec = 0; - - schedstat_set(curr->stats.exec_max, - max(curr->stats.exec_max, delta_exec)); - - update_current_exec_runtime(curr, now, delta_exec); + update_curr_common(rq); } /* From 5fe6ec8f6ab549b6422e41551abb51802bd48bc7 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Nov 2023 13:41:43 +0100 Subject: [PATCH 06/30] sched: Remove vruntime from trace_sched_stat_runtime() Tracing the runtime delta makes sense, observer can sum over time. Tracing the absolute vruntime makes less sense, inconsistent: absolute-vs-delta, but also vruntime delta can be computed from runtime delta. Removing the vruntime thing also makes the two tracepoint sites identical, allowing to unify the code in a later patch. Signed-off-by: Peter Zijlstra (Intel) --- include/trace/events/sched.h | 15 ++++++--------- kernel/sched/fair.c | 5 ++--- 2 files changed, 8 insertions(+), 12 deletions(-) diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h index 6188ad0d9e0d..dbb01b4b7451 100644 --- a/include/trace/events/sched.h +++ b/include/trace/events/sched.h @@ -493,33 +493,30 @@ DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_blocked, */ DECLARE_EVENT_CLASS(sched_stat_runtime, - TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime), + TP_PROTO(struct task_struct *tsk, u64 runtime), - TP_ARGS(tsk, __perf_count(runtime), vruntime), + TP_ARGS(tsk, __perf_count(runtime)), TP_STRUCT__entry( __array( char, comm, TASK_COMM_LEN ) __field( pid_t, pid ) __field( u64, runtime ) - __field( u64, vruntime ) ), TP_fast_assign( memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN); __entry->pid = tsk->pid; __entry->runtime = runtime; - __entry->vruntime = vruntime; ), - TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]", + TP_printk("comm=%s pid=%d runtime=%Lu [ns]", __entry->comm, __entry->pid, - (unsigned long long)__entry->runtime, - (unsigned long long)__entry->vruntime) + (unsigned long long)__entry->runtime) ); DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime, - TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime), - TP_ARGS(tsk, runtime, vruntime)); + TP_PROTO(struct task_struct *tsk, u64 runtime), + TP_ARGS(tsk, runtime)); /* * Tracepoint for showing priority inheritance modifying a tasks diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 11073cf00134..33db70c6b582 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1138,8 +1138,7 @@ s64 update_curr_common(struct rq *rq) if (unlikely(delta_exec <= 0)) return delta_exec; - trace_sched_stat_runtime(curr, delta_exec, 0); - + trace_sched_stat_runtime(curr, delta_exec); account_group_exec_runtime(curr, delta_exec); cgroup_account_cputime(curr, delta_exec); @@ -1168,7 +1167,7 @@ static void update_curr(struct cfs_rq *cfs_rq) if (entity_is_task(curr)) { struct task_struct *curtask = task_of(curr); - trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime); + trace_sched_stat_runtime(curtask, delta_exec); cgroup_account_cputime(curtask, delta_exec); account_group_exec_runtime(curtask, delta_exec); } From c708a4dc5ab547edc3d6537233ca9e79ea30ce47 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Nov 2023 14:04:01 +0100 Subject: [PATCH 07/30] sched: Unify more update_curr*() Now that trace_sched_stat_runtime() no longer takes a vruntime argument, the task specific bits are identical between update_curr_common() and update_curr(). Signed-off-by: Peter Zijlstra (Intel) --- kernel/sched/fair.c | 24 +++++++++++------------- 1 file changed, 11 insertions(+), 13 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 33db70c6b582..1cd92b11b289 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1126,6 +1126,13 @@ static s64 update_curr_se(struct rq *rq, struct sched_entity *curr) return delta_exec; } +static inline void update_curr_task(struct task_struct *p, s64 delta_exec) +{ + trace_sched_stat_runtime(p, delta_exec); + account_group_exec_runtime(p, delta_exec); + cgroup_account_cputime(p, delta_exec); +} + /* * Used by other classes to account runtime. */ @@ -1135,12 +1142,8 @@ s64 update_curr_common(struct rq *rq) s64 delta_exec; delta_exec = update_curr_se(rq, &curr->se); - if (unlikely(delta_exec <= 0)) - return delta_exec; - - trace_sched_stat_runtime(curr, delta_exec); - account_group_exec_runtime(curr, delta_exec); - cgroup_account_cputime(curr, delta_exec); + if (likely(delta_exec > 0)) + update_curr_task(curr, delta_exec); return delta_exec; } @@ -1164,13 +1167,8 @@ static void update_curr(struct cfs_rq *cfs_rq) update_deadline(cfs_rq, curr); update_min_vruntime(cfs_rq); - if (entity_is_task(curr)) { - struct task_struct *curtask = task_of(curr); - - trace_sched_stat_runtime(curtask, delta_exec); - cgroup_account_cputime(curtask, delta_exec); - account_group_exec_runtime(curtask, delta_exec); - } + if (entity_is_task(curr)) + update_curr_task(task_of(curr), delta_exec); account_cfs_rq_runtime(cfs_rq, delta_exec); } From 9e07d45c5210f5dd6701c00d55791983db7320fa Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 4 Nov 2023 11:59:19 +0100 Subject: [PATCH 08/30] sched/deadline: Collect sched_dl_entity initialization Create a single function that initializes a sched_dl_entity. Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Daniel Bristot de Oliveira Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Phil Auld Reviewed-by: Valentin Schneider Link: https://lkml.kernel.org/r/51acc695eecf0a1a2f78f9a044e11ffd9b316bcf.1699095159.git.bristot@kernel.org --- kernel/sched/core.c | 5 +---- kernel/sched/deadline.c | 22 +++++++++++++++------- kernel/sched/sched.h | 5 +---- 3 files changed, 17 insertions(+), 15 deletions(-) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 9d5099d02dbc..966631f05d71 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4511,10 +4511,7 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p) memset(&p->stats, 0, sizeof(p->stats)); #endif - RB_CLEAR_NODE(&p->dl.rb_node); - init_dl_task_timer(&p->dl); - init_dl_inactive_task_timer(&p->dl); - __dl_clear_params(p); + init_dl_entity(&p->dl); INIT_LIST_HEAD(&p->rt.run_list); p->rt.timeout = 0; diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index de79719c63c0..e80bb884262d 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -335,6 +335,8 @@ static void dl_change_utilization(struct task_struct *p, u64 new_bw) __add_rq_bw(new_bw, &rq->dl); } +static void __dl_clear_params(struct sched_dl_entity *dl_se); + /* * The utilization of a task cannot be immediately removed from * the rq active utilization (running_bw) when the task blocks. @@ -434,7 +436,7 @@ static void task_non_contending(struct task_struct *p) raw_spin_lock(&dl_b->lock); __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p))); raw_spin_unlock(&dl_b->lock); - __dl_clear_params(p); + __dl_clear_params(dl_se); } return; @@ -1183,7 +1185,7 @@ unlock: return HRTIMER_NORESTART; } -void init_dl_task_timer(struct sched_dl_entity *dl_se) +static void init_dl_task_timer(struct sched_dl_entity *dl_se) { struct hrtimer *timer = &dl_se->dl_timer; @@ -1389,7 +1391,7 @@ static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer) raw_spin_lock(&dl_b->lock); __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p))); raw_spin_unlock(&dl_b->lock); - __dl_clear_params(p); + __dl_clear_params(dl_se); goto unlock; } @@ -1405,7 +1407,7 @@ unlock: return HRTIMER_NORESTART; } -void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se) +static void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se) { struct hrtimer *timer = &dl_se->inactive_timer; @@ -2957,10 +2959,8 @@ bool __checkparam_dl(const struct sched_attr *attr) /* * This function clears the sched_dl_entity static params. */ -void __dl_clear_params(struct task_struct *p) +static void __dl_clear_params(struct sched_dl_entity *dl_se) { - struct sched_dl_entity *dl_se = &p->dl; - dl_se->dl_runtime = 0; dl_se->dl_deadline = 0; dl_se->dl_period = 0; @@ -2978,6 +2978,14 @@ void __dl_clear_params(struct task_struct *p) #endif } +void init_dl_entity(struct sched_dl_entity *dl_se) +{ + RB_CLEAR_NODE(&dl_se->rb_node); + init_dl_task_timer(dl_se); + init_dl_inactive_task_timer(dl_se); + __dl_clear_params(dl_se); +} + bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr) { struct sched_dl_entity *dl_se = &p->dl; diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 6703e9e81b1d..3c62df1511e7 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -273,8 +273,6 @@ struct rt_bandwidth { unsigned int rt_period_active; }; -void __dl_clear_params(struct task_struct *p); - static inline int dl_bandwidth_enabled(void) { return sysctl_sched_rt_runtime >= 0; @@ -2427,8 +2425,7 @@ extern struct rt_bandwidth def_rt_bandwidth; extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq); -extern void init_dl_task_timer(struct sched_dl_entity *dl_se); -extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); +extern void init_dl_entity(struct sched_dl_entity *dl_se); #define BW_SHIFT 20 #define BW_UNIT (1 << BW_SHIFT) From 2f7a0f58948d8231236e2facecc500f1930fb996 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 4 Nov 2023 11:59:20 +0100 Subject: [PATCH 09/30] sched/deadline: Move bandwidth accounting into {en,de}queue_dl_entity In preparation of introducing !task sched_dl_entity; move the bandwidth accounting into {en.de}queue_dl_entity(). Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Daniel Bristot de Oliveira Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Phil Auld Reviewed-by: Valentin Schneider Link: https://lkml.kernel.org/r/a86dccbbe44e021b8771627e1dae01a69b73466d.1699095159.git.bristot@kernel.org --- kernel/sched/deadline.c | 132 ++++++++++++++++++++++------------------ kernel/sched/sched.h | 6 ++ 2 files changed, 79 insertions(+), 59 deletions(-) diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index e80bb884262d..81810f67df7a 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -391,12 +391,12 @@ static void __dl_clear_params(struct sched_dl_entity *dl_se); * up, and checks if the task is still in the "ACTIVE non contending" * state or not (in the second case, it updates running_bw). */ -static void task_non_contending(struct task_struct *p) +static void task_non_contending(struct sched_dl_entity *dl_se) { - struct sched_dl_entity *dl_se = &p->dl; struct hrtimer *timer = &dl_se->inactive_timer; struct dl_rq *dl_rq = dl_rq_of_se(dl_se); struct rq *rq = rq_of_dl_rq(dl_rq); + struct task_struct *p = dl_task_of(dl_se); s64 zerolag_time; /* @@ -428,13 +428,14 @@ static void task_non_contending(struct task_struct *p) if ((zerolag_time < 0) || hrtimer_active(&dl_se->inactive_timer)) { if (dl_task(p)) sub_running_bw(dl_se, dl_rq); + if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) { struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); if (READ_ONCE(p->__state) == TASK_DEAD) - sub_rq_bw(&p->dl, &rq->dl); + sub_rq_bw(dl_se, &rq->dl); raw_spin_lock(&dl_b->lock); - __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p))); + __dl_sub(dl_b, dl_se->dl_bw, dl_bw_cpus(task_cpu(p))); raw_spin_unlock(&dl_b->lock); __dl_clear_params(dl_se); } @@ -1601,6 +1602,41 @@ enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags) update_stats_enqueue_dl(dl_rq_of_se(dl_se), dl_se, flags); + /* + * Check if a constrained deadline task was activated + * after the deadline but before the next period. + * If that is the case, the task will be throttled and + * the replenishment timer will be set to the next period. + */ + if (!dl_se->dl_throttled && !dl_is_implicit(dl_se)) + dl_check_constrained_dl(dl_se); + + if (flags & (ENQUEUE_RESTORE|ENQUEUE_MIGRATING)) { + struct dl_rq *dl_rq = dl_rq_of_se(dl_se); + + add_rq_bw(dl_se, dl_rq); + add_running_bw(dl_se, dl_rq); + } + + /* + * If p is throttled, we do not enqueue it. In fact, if it exhausted + * its budget it needs a replenishment and, since it now is on + * its rq, the bandwidth timer callback (which clearly has not + * run yet) will take care of this. + * However, the active utilization does not depend on the fact + * that the task is on the runqueue or not (but depends on the + * task's state - in GRUB parlance, "inactive" vs "active contending"). + * In other words, even if a task is throttled its utilization must + * be counted in the active utilization; hence, we need to call + * add_running_bw(). + */ + if (dl_se->dl_throttled && !(flags & ENQUEUE_REPLENISH)) { + if (flags & ENQUEUE_WAKEUP) + task_contending(dl_se, flags); + + return; + } + /* * If this is a wakeup or a new instance, the scheduling * parameters of the task might need updating. Otherwise, @@ -1620,9 +1656,28 @@ enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags) __enqueue_dl_entity(dl_se); } -static void dequeue_dl_entity(struct sched_dl_entity *dl_se) +static void dequeue_dl_entity(struct sched_dl_entity *dl_se, int flags) { __dequeue_dl_entity(dl_se); + + if (flags & (DEQUEUE_SAVE|DEQUEUE_MIGRATING)) { + struct dl_rq *dl_rq = dl_rq_of_se(dl_se); + + sub_running_bw(dl_se, dl_rq); + sub_rq_bw(dl_se, dl_rq); + } + + /* + * This check allows to start the inactive timer (or to immediately + * decrease the active utilization, if needed) in two cases: + * when the task blocks and when it is terminating + * (p->state == TASK_DEAD). We can handle the two cases in the same + * way, because from GRUB's point of view the same thing is happening + * (the task moves from "active contending" to "active non contending" + * or "inactive") + */ + if (flags & DEQUEUE_SLEEP) + task_non_contending(dl_se); } static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags) @@ -1667,76 +1722,35 @@ static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags) return; } - /* - * Check if a constrained deadline task was activated - * after the deadline but before the next period. - * If that is the case, the task will be throttled and - * the replenishment timer will be set to the next period. - */ - if (!p->dl.dl_throttled && !dl_is_implicit(&p->dl)) - dl_check_constrained_dl(&p->dl); - - if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & ENQUEUE_RESTORE) { - add_rq_bw(&p->dl, &rq->dl); - add_running_bw(&p->dl, &rq->dl); - } - - /* - * If p is throttled, we do not enqueue it. In fact, if it exhausted - * its budget it needs a replenishment and, since it now is on - * its rq, the bandwidth timer callback (which clearly has not - * run yet) will take care of this. - * However, the active utilization does not depend on the fact - * that the task is on the runqueue or not (but depends on the - * task's state - in GRUB parlance, "inactive" vs "active contending"). - * In other words, even if a task is throttled its utilization must - * be counted in the active utilization; hence, we need to call - * add_running_bw(). - */ - if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH)) { - if (flags & ENQUEUE_WAKEUP) - task_contending(&p->dl, flags); - - return; - } - check_schedstat_required(); update_stats_wait_start_dl(dl_rq_of_se(&p->dl), &p->dl); + if (p->on_rq == TASK_ON_RQ_MIGRATING) + flags |= ENQUEUE_MIGRATING; + enqueue_dl_entity(&p->dl, flags); - if (!task_current(rq, p) && p->nr_cpus_allowed > 1) + if (!task_current(rq, p) && !p->dl.dl_throttled && p->nr_cpus_allowed > 1) enqueue_pushable_dl_task(rq, p); } static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags) { update_stats_dequeue_dl(&rq->dl, &p->dl, flags); - dequeue_dl_entity(&p->dl); - dequeue_pushable_dl_task(rq, p); + dequeue_dl_entity(&p->dl, flags); + + if (!p->dl.dl_throttled) + dequeue_pushable_dl_task(rq, p); } static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags) { update_curr_dl(rq); + + if (p->on_rq == TASK_ON_RQ_MIGRATING) + flags |= DEQUEUE_MIGRATING; + __dequeue_task_dl(rq, p, flags); - - if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & DEQUEUE_SAVE) { - sub_running_bw(&p->dl, &rq->dl); - sub_rq_bw(&p->dl, &rq->dl); - } - - /* - * This check allows to start the inactive timer (or to immediately - * decrease the active utilization, if needed) in two cases: - * when the task blocks and when it is terminating - * (p->state == TASK_DEAD). We can handle the two cases in the same - * way, because from GRUB's point of view the same thing is happening - * (the task moves from "active contending" to "active non contending" - * or "inactive") - */ - if (flags & DEQUEUE_SLEEP) - task_non_contending(p); } /* @@ -2551,7 +2565,7 @@ static void switched_from_dl(struct rq *rq, struct task_struct *p) * will reset the task parameters. */ if (task_on_rq_queued(p) && p->dl.dl_runtime) - task_non_contending(p); + task_non_contending(&p->dl); /* * In case a task is setscheduled out from SCHED_DEADLINE we need to diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 3c62df1511e7..1cda787172f0 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2177,6 +2177,10 @@ extern const u32 sched_prio_to_wmult[40]; * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location * in the runqueue. * + * NOCLOCK - skip the update_rq_clock() (avoids double updates) + * + * MIGRATION - p->on_rq == TASK_ON_RQ_MIGRATING (used for DEADLINE) + * * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) * ENQUEUE_MIGRATED - the task was migrated during wakeup @@ -2187,6 +2191,7 @@ extern const u32 sched_prio_to_wmult[40]; #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ +#define DEQUEUE_MIGRATING 0x100 /* Matches ENQUEUE_MIGRATING */ #define ENQUEUE_WAKEUP 0x01 #define ENQUEUE_RESTORE 0x02 @@ -2201,6 +2206,7 @@ extern const u32 sched_prio_to_wmult[40]; #define ENQUEUE_MIGRATED 0x00 #endif #define ENQUEUE_INITIAL 0x80 +#define ENQUEUE_MIGRATING 0x100 #define RETRY_TASK ((void *)-1UL) From 63ba8422f876e32ee564ea95da9a7313b13ff0a1 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 4 Nov 2023 11:59:21 +0100 Subject: [PATCH 10/30] sched/deadline: Introduce deadline servers Low priority tasks (e.g., SCHED_OTHER) can suffer starvation if tasks with higher priority (e.g., SCHED_FIFO) monopolize CPU(s). RT Throttling has been introduced a while ago as a (mostly debug) countermeasure one can utilize to reserve some CPU time for low priority tasks (usually background type of work, e.g. workqueues, timers, etc.). It however has its own problems (see documentation) and the undesired effect of unconditionally throttling FIFO tasks even when no lower priority activity needs to run (there are mechanisms to fix this issue as well, but, again, with their own problems). Introduce deadline servers to service low priority tasks needs under starvation conditions. Deadline servers are built extending SCHED_DEADLINE implementation to allow 2-level scheduling (a sched_deadline entity becomes a container for lower priority scheduling entities). Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Daniel Bristot de Oliveira Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/4968601859d920335cf85822eb573a5f179f04b8.1699095159.git.bristot@kernel.org --- include/linux/sched.h | 22 ++- kernel/sched/core.c | 17 ++ kernel/sched/deadline.c | 332 +++++++++++++++++++++++++++------------- kernel/sched/fair.c | 2 + kernel/sched/sched.h | 27 ++++ 5 files changed, 292 insertions(+), 108 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index 44b46d9743bf..8d258162deb0 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -63,11 +63,13 @@ struct robust_list_head; struct root_domain; struct rq; struct sched_attr; +struct sched_dl_entity; struct seq_file; struct sighand_struct; struct signal_struct; struct task_delay_info; struct task_group; +struct task_struct; struct user_event_mm; /* @@ -607,6 +609,9 @@ struct sched_rt_entity { #endif } __randomize_layout; +typedef bool (*dl_server_has_tasks_f)(struct sched_dl_entity *); +typedef struct task_struct *(*dl_server_pick_f)(struct sched_dl_entity *); + struct sched_dl_entity { struct rb_node rb_node; @@ -654,6 +659,7 @@ struct sched_dl_entity { unsigned int dl_yielded : 1; unsigned int dl_non_contending : 1; unsigned int dl_overrun : 1; + unsigned int dl_server : 1; /* * Bandwidth enforcement timer. Each -deadline task has its @@ -668,7 +674,20 @@ struct sched_dl_entity { * timer is needed to decrease the active utilization at the correct * time. */ - struct hrtimer inactive_timer; + struct hrtimer inactive_timer; + + /* + * Bits for DL-server functionality. Also see the comment near + * dl_server_update(). + * + * @rq the runqueue this server is for + * + * @server_has_tasks() returns true if @server_pick return a + * runnable task. + */ + struct rq *rq; + dl_server_has_tasks_f server_has_tasks; + dl_server_pick_f server_pick; #ifdef CONFIG_RT_MUTEXES /* @@ -795,6 +814,7 @@ struct task_struct { struct sched_entity se; struct sched_rt_entity rt; struct sched_dl_entity dl; + struct sched_dl_entity *dl_server; const struct sched_class *sched_class; #ifdef CONFIG_SCHED_CORE diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 966631f05d71..f5f4495d1768 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -3797,6 +3797,8 @@ ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, rq->idle_stamp = 0; } #endif + + p->dl_server = NULL; } /* @@ -6003,12 +6005,27 @@ __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) p = pick_next_task_idle(rq); } + /* + * This is the fast path; it cannot be a DL server pick; + * therefore even if @p == @prev, ->dl_server must be NULL. + */ + if (p->dl_server) + p->dl_server = NULL; + return p; } restart: put_prev_task_balance(rq, prev, rf); + /* + * We've updated @prev and no longer need the server link, clear it. + * Must be done before ->pick_next_task() because that can (re)set + * ->dl_server. + */ + if (prev->dl_server) + prev->dl_server = NULL; + for_each_class(class) { p = class->pick_next_task(rq); if (p) diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index 81810f67df7a..a04a436af8cc 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -54,8 +54,14 @@ static int __init sched_dl_sysctl_init(void) late_initcall(sched_dl_sysctl_init); #endif +static bool dl_server(struct sched_dl_entity *dl_se) +{ + return dl_se->dl_server; +} + static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se) { + BUG_ON(dl_server(dl_se)); return container_of(dl_se, struct task_struct, dl); } @@ -64,12 +70,19 @@ static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq) return container_of(dl_rq, struct rq, dl); } +static inline struct rq *rq_of_dl_se(struct sched_dl_entity *dl_se) +{ + struct rq *rq = dl_se->rq; + + if (!dl_server(dl_se)) + rq = task_rq(dl_task_of(dl_se)); + + return rq; +} + static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se) { - struct task_struct *p = dl_task_of(dl_se); - struct rq *rq = task_rq(p); - - return &rq->dl; + return &rq_of_dl_se(dl_se)->dl; } static inline int on_dl_rq(struct sched_dl_entity *dl_se) @@ -394,9 +407,8 @@ static void __dl_clear_params(struct sched_dl_entity *dl_se); static void task_non_contending(struct sched_dl_entity *dl_se) { struct hrtimer *timer = &dl_se->inactive_timer; - struct dl_rq *dl_rq = dl_rq_of_se(dl_se); - struct rq *rq = rq_of_dl_rq(dl_rq); - struct task_struct *p = dl_task_of(dl_se); + struct rq *rq = rq_of_dl_se(dl_se); + struct dl_rq *dl_rq = &rq->dl; s64 zerolag_time; /* @@ -426,25 +438,33 @@ static void task_non_contending(struct sched_dl_entity *dl_se) * utilization now, instead of starting a timer */ if ((zerolag_time < 0) || hrtimer_active(&dl_se->inactive_timer)) { - if (dl_task(p)) + if (dl_server(dl_se)) { sub_running_bw(dl_se, dl_rq); + } else { + struct task_struct *p = dl_task_of(dl_se); - if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) { - struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); + if (dl_task(p)) + sub_running_bw(dl_se, dl_rq); - if (READ_ONCE(p->__state) == TASK_DEAD) - sub_rq_bw(dl_se, &rq->dl); - raw_spin_lock(&dl_b->lock); - __dl_sub(dl_b, dl_se->dl_bw, dl_bw_cpus(task_cpu(p))); - raw_spin_unlock(&dl_b->lock); - __dl_clear_params(dl_se); + if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) { + struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); + + if (READ_ONCE(p->__state) == TASK_DEAD) + sub_rq_bw(dl_se, &rq->dl); + raw_spin_lock(&dl_b->lock); + __dl_sub(dl_b, dl_se->dl_bw, dl_bw_cpus(task_cpu(p))); + raw_spin_unlock(&dl_b->lock); + __dl_clear_params(dl_se); + } } return; } dl_se->dl_non_contending = 1; - get_task_struct(p); + if (!dl_server(dl_se)) + get_task_struct(dl_task_of(dl_se)); + hrtimer_start(timer, ns_to_ktime(zerolag_time), HRTIMER_MODE_REL_HARD); } @@ -471,8 +491,10 @@ static void task_contending(struct sched_dl_entity *dl_se, int flags) * will not touch the rq's active utilization, * so we are still safe. */ - if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1) - put_task_struct(dl_task_of(dl_se)); + if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1) { + if (!dl_server(dl_se)) + put_task_struct(dl_task_of(dl_se)); + } } else { /* * Since "dl_non_contending" is not set, the @@ -485,10 +507,8 @@ static void task_contending(struct sched_dl_entity *dl_se, int flags) } } -static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq) +static inline int is_leftmost(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) { - struct sched_dl_entity *dl_se = &p->dl; - return rb_first_cached(&dl_rq->root) == &dl_se->rb_node; } @@ -740,8 +760,10 @@ static inline void deadline_queue_pull_task(struct rq *rq) } #endif /* CONFIG_SMP */ +static void +enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags); static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags); -static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags); +static void dequeue_dl_entity(struct sched_dl_entity *dl_se, int flags); static void wakeup_preempt_dl(struct rq *rq, struct task_struct *p, int flags); static inline void replenish_dl_new_period(struct sched_dl_entity *dl_se, @@ -989,8 +1011,7 @@ static inline bool dl_is_implicit(struct sched_dl_entity *dl_se) */ static void update_dl_entity(struct sched_dl_entity *dl_se) { - struct dl_rq *dl_rq = dl_rq_of_se(dl_se); - struct rq *rq = rq_of_dl_rq(dl_rq); + struct rq *rq = rq_of_dl_se(dl_se); if (dl_time_before(dl_se->deadline, rq_clock(rq)) || dl_entity_overflow(dl_se, rq_clock(rq))) { @@ -1021,11 +1042,11 @@ static inline u64 dl_next_period(struct sched_dl_entity *dl_se) * actually started or not (i.e., the replenishment instant is in * the future or in the past). */ -static int start_dl_timer(struct task_struct *p) +static int start_dl_timer(struct sched_dl_entity *dl_se) { - struct sched_dl_entity *dl_se = &p->dl; struct hrtimer *timer = &dl_se->dl_timer; - struct rq *rq = task_rq(p); + struct dl_rq *dl_rq = dl_rq_of_se(dl_se); + struct rq *rq = rq_of_dl_rq(dl_rq); ktime_t now, act; s64 delta; @@ -1059,13 +1080,33 @@ static int start_dl_timer(struct task_struct *p) * and observe our state. */ if (!hrtimer_is_queued(timer)) { - get_task_struct(p); + if (!dl_server(dl_se)) + get_task_struct(dl_task_of(dl_se)); hrtimer_start(timer, act, HRTIMER_MODE_ABS_HARD); } return 1; } +static void __push_dl_task(struct rq *rq, struct rq_flags *rf) +{ +#ifdef CONFIG_SMP + /* + * Queueing this task back might have overloaded rq, check if we need + * to kick someone away. + */ + if (has_pushable_dl_tasks(rq)) { + /* + * Nothing relies on rq->lock after this, so its safe to drop + * rq->lock. + */ + rq_unpin_lock(rq, rf); + push_dl_task(rq); + rq_repin_lock(rq, rf); + } +#endif +} + /* * This is the bandwidth enforcement timer callback. If here, we know * a task is not on its dl_rq, since the fact that the timer was running @@ -1084,10 +1125,34 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer) struct sched_dl_entity *dl_se = container_of(timer, struct sched_dl_entity, dl_timer); - struct task_struct *p = dl_task_of(dl_se); + struct task_struct *p; struct rq_flags rf; struct rq *rq; + if (dl_server(dl_se)) { + struct rq *rq = rq_of_dl_se(dl_se); + struct rq_flags rf; + + rq_lock(rq, &rf); + if (dl_se->dl_throttled) { + sched_clock_tick(); + update_rq_clock(rq); + + if (dl_se->server_has_tasks(dl_se)) { + enqueue_dl_entity(dl_se, ENQUEUE_REPLENISH); + resched_curr(rq); + __push_dl_task(rq, &rf); + } else { + replenish_dl_entity(dl_se); + } + + } + rq_unlock(rq, &rf); + + return HRTIMER_NORESTART; + } + + p = dl_task_of(dl_se); rq = task_rq_lock(p, &rf); /* @@ -1158,21 +1223,7 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer) else resched_curr(rq); -#ifdef CONFIG_SMP - /* - * Queueing this task back might have overloaded rq, check if we need - * to kick someone away. - */ - if (has_pushable_dl_tasks(rq)) { - /* - * Nothing relies on rq->lock after this, so its safe to drop - * rq->lock. - */ - rq_unpin_lock(rq, &rf); - push_dl_task(rq); - rq_repin_lock(rq, &rf); - } -#endif + __push_dl_task(rq, &rf); unlock: task_rq_unlock(rq, p, &rf); @@ -1214,12 +1265,11 @@ static void init_dl_task_timer(struct sched_dl_entity *dl_se) */ static inline void dl_check_constrained_dl(struct sched_dl_entity *dl_se) { - struct task_struct *p = dl_task_of(dl_se); - struct rq *rq = rq_of_dl_rq(dl_rq_of_se(dl_se)); + struct rq *rq = rq_of_dl_se(dl_se); if (dl_time_before(dl_se->deadline, rq_clock(rq)) && dl_time_before(rq_clock(rq), dl_next_period(dl_se))) { - if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(p))) + if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(dl_se))) return; dl_se->dl_throttled = 1; if (dl_se->runtime > 0) @@ -1270,29 +1320,13 @@ static u64 grub_reclaim(u64 delta, struct rq *rq, struct sched_dl_entity *dl_se) return (delta * u_act) >> BW_SHIFT; } -/* - * Update the current task's runtime statistics (provided it is still - * a -deadline task and has not been removed from the dl_rq). - */ -static void update_curr_dl(struct rq *rq) +static inline void +update_stats_dequeue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se, + int flags); +static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64 delta_exec) { - struct task_struct *curr = rq->curr; - struct sched_dl_entity *dl_se = &curr->dl; - s64 delta_exec, scaled_delta_exec; - int cpu = cpu_of(rq); + s64 scaled_delta_exec; - if (!dl_task(curr) || !on_dl_rq(dl_se)) - return; - - /* - * Consumed budget is computed considering the time as - * observed by schedulable tasks (excluding time spent - * in hardirq context, etc.). Deadlines are instead - * computed using hard walltime. This seems to be the more - * natural solution, but the full ramifications of this - * approach need further study. - */ - delta_exec = update_curr_common(rq); if (unlikely(delta_exec <= 0)) { if (unlikely(dl_se->dl_yielded)) goto throttle; @@ -1310,10 +1344,9 @@ static void update_curr_dl(struct rq *rq) * according to current frequency and CPU maximum capacity. */ if (unlikely(dl_se->flags & SCHED_FLAG_RECLAIM)) { - scaled_delta_exec = grub_reclaim(delta_exec, - rq, - &curr->dl); + scaled_delta_exec = grub_reclaim(delta_exec, rq, dl_se); } else { + int cpu = cpu_of(rq); unsigned long scale_freq = arch_scale_freq_capacity(cpu); unsigned long scale_cpu = arch_scale_cpu_capacity(cpu); @@ -1332,11 +1365,20 @@ throttle: (dl_se->flags & SCHED_FLAG_DL_OVERRUN)) dl_se->dl_overrun = 1; - __dequeue_task_dl(rq, curr, 0); - if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(curr))) - enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH); + dequeue_dl_entity(dl_se, 0); + if (!dl_server(dl_se)) { + update_stats_dequeue_dl(&rq->dl, dl_se, 0); + dequeue_pushable_dl_task(rq, dl_task_of(dl_se)); + } - if (!is_leftmost(curr, &rq->dl)) + if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(dl_se))) { + if (dl_server(dl_se)) + enqueue_dl_entity(dl_se, ENQUEUE_REPLENISH); + else + enqueue_task_dl(rq, dl_task_of(dl_se), ENQUEUE_REPLENISH); + } + + if (!is_leftmost(dl_se, &rq->dl)) resched_curr(rq); } @@ -1366,20 +1408,82 @@ throttle: } } +void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec) +{ + update_curr_dl_se(dl_se->rq, dl_se, delta_exec); +} + +void dl_server_start(struct sched_dl_entity *dl_se) +{ + if (!dl_server(dl_se)) { + dl_se->dl_server = 1; + setup_new_dl_entity(dl_se); + } + enqueue_dl_entity(dl_se, ENQUEUE_WAKEUP); +} + +void dl_server_stop(struct sched_dl_entity *dl_se) +{ + dequeue_dl_entity(dl_se, DEQUEUE_SLEEP); +} + +void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq, + dl_server_has_tasks_f has_tasks, + dl_server_pick_f pick) +{ + dl_se->rq = rq; + dl_se->server_has_tasks = has_tasks; + dl_se->server_pick = pick; +} + +/* + * Update the current task's runtime statistics (provided it is still + * a -deadline task and has not been removed from the dl_rq). + */ +static void update_curr_dl(struct rq *rq) +{ + struct task_struct *curr = rq->curr; + struct sched_dl_entity *dl_se = &curr->dl; + s64 delta_exec; + + if (!dl_task(curr) || !on_dl_rq(dl_se)) + return; + + /* + * Consumed budget is computed considering the time as + * observed by schedulable tasks (excluding time spent + * in hardirq context, etc.). Deadlines are instead + * computed using hard walltime. This seems to be the more + * natural solution, but the full ramifications of this + * approach need further study. + */ + delta_exec = update_curr_common(rq); + update_curr_dl_se(rq, dl_se, delta_exec); +} + static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer) { struct sched_dl_entity *dl_se = container_of(timer, struct sched_dl_entity, inactive_timer); - struct task_struct *p = dl_task_of(dl_se); + struct task_struct *p = NULL; struct rq_flags rf; struct rq *rq; - rq = task_rq_lock(p, &rf); + if (!dl_server(dl_se)) { + p = dl_task_of(dl_se); + rq = task_rq_lock(p, &rf); + } else { + rq = dl_se->rq; + rq_lock(rq, &rf); + } sched_clock_tick(); update_rq_clock(rq); + if (dl_server(dl_se)) + goto no_task; + if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) { struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); @@ -1396,14 +1500,21 @@ static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer) goto unlock; } + +no_task: if (dl_se->dl_non_contending == 0) goto unlock; sub_running_bw(dl_se, &rq->dl); dl_se->dl_non_contending = 0; unlock: - task_rq_unlock(rq, p, &rf); - put_task_struct(p); + + if (!dl_server(dl_se)) { + task_rq_unlock(rq, p, &rf); + put_task_struct(p); + } else { + rq_unlock(rq, &rf); + } return HRTIMER_NORESTART; } @@ -1466,10 +1577,8 @@ static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {} static inline void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) { - int prio = dl_task_of(dl_se)->prio; u64 deadline = dl_se->deadline; - WARN_ON(!dl_prio(prio)); dl_rq->dl_nr_running++; add_nr_running(rq_of_dl_rq(dl_rq), 1); @@ -1479,9 +1588,6 @@ void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) static inline void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) { - int prio = dl_task_of(dl_se)->prio; - - WARN_ON(!dl_prio(prio)); WARN_ON(!dl_rq->dl_nr_running); dl_rq->dl_nr_running--; sub_nr_running(rq_of_dl_rq(dl_rq), 1); @@ -1648,8 +1754,7 @@ enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags) } else if (flags & ENQUEUE_REPLENISH) { replenish_dl_entity(dl_se); } else if ((flags & ENQUEUE_RESTORE) && - dl_time_before(dl_se->deadline, - rq_clock(rq_of_dl_rq(dl_rq_of_se(dl_se))))) { + dl_time_before(dl_se->deadline, rq_clock(rq_of_dl_se(dl_se)))) { setup_new_dl_entity(dl_se); } @@ -1730,19 +1835,13 @@ static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags) enqueue_dl_entity(&p->dl, flags); + if (dl_server(&p->dl)) + return; + if (!task_current(rq, p) && !p->dl.dl_throttled && p->nr_cpus_allowed > 1) enqueue_pushable_dl_task(rq, p); } -static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags) -{ - update_stats_dequeue_dl(&rq->dl, &p->dl, flags); - dequeue_dl_entity(&p->dl, flags); - - if (!p->dl.dl_throttled) - dequeue_pushable_dl_task(rq, p); -} - static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags) { update_curr_dl(rq); @@ -1750,7 +1849,9 @@ static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags) if (p->on_rq == TASK_ON_RQ_MIGRATING) flags |= DEQUEUE_MIGRATING; - __dequeue_task_dl(rq, p, flags); + dequeue_dl_entity(&p->dl, flags); + if (!p->dl.dl_throttled && !dl_server(&p->dl)) + dequeue_pushable_dl_task(rq, p); } /* @@ -1940,12 +2041,12 @@ static void wakeup_preempt_dl(struct rq *rq, struct task_struct *p, } #ifdef CONFIG_SCHED_HRTICK -static void start_hrtick_dl(struct rq *rq, struct task_struct *p) +static void start_hrtick_dl(struct rq *rq, struct sched_dl_entity *dl_se) { - hrtick_start(rq, p->dl.runtime); + hrtick_start(rq, dl_se->runtime); } #else /* !CONFIG_SCHED_HRTICK */ -static void start_hrtick_dl(struct rq *rq, struct task_struct *p) +static void start_hrtick_dl(struct rq *rq, struct sched_dl_entity *dl_se) { } #endif @@ -1965,9 +2066,6 @@ static void set_next_task_dl(struct rq *rq, struct task_struct *p, bool first) if (!first) return; - if (hrtick_enabled_dl(rq)) - start_hrtick_dl(rq, p); - if (rq->curr->sched_class != &dl_sched_class) update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0); @@ -1990,12 +2088,25 @@ static struct task_struct *pick_task_dl(struct rq *rq) struct dl_rq *dl_rq = &rq->dl; struct task_struct *p; +again: if (!sched_dl_runnable(rq)) return NULL; dl_se = pick_next_dl_entity(dl_rq); WARN_ON_ONCE(!dl_se); - p = dl_task_of(dl_se); + + if (dl_server(dl_se)) { + p = dl_se->server_pick(dl_se); + if (!p) { + WARN_ON_ONCE(1); + dl_se->dl_yielded = 1; + update_curr_dl_se(rq, dl_se, 0); + goto again; + } + p->dl_server = dl_se; + } else { + p = dl_task_of(dl_se); + } return p; } @@ -2005,9 +2116,15 @@ static struct task_struct *pick_next_task_dl(struct rq *rq) struct task_struct *p; p = pick_task_dl(rq); - if (p) + if (!p) + return p; + + if (!p->dl_server) set_next_task_dl(rq, p, true); + if (hrtick_enabled(rq)) + start_hrtick_dl(rq, &p->dl); + return p; } @@ -2045,8 +2162,8 @@ static void task_tick_dl(struct rq *rq, struct task_struct *p, int queued) * be set and schedule() will start a new hrtick for the next task. */ if (hrtick_enabled_dl(rq) && queued && p->dl.runtime > 0 && - is_leftmost(p, &rq->dl)) - start_hrtick_dl(rq, p); + is_leftmost(&p->dl, &rq->dl)) + start_hrtick_dl(rq, &p->dl); } static void task_fork_dl(struct task_struct *p) @@ -2986,6 +3103,7 @@ static void __dl_clear_params(struct sched_dl_entity *dl_se) dl_se->dl_yielded = 0; dl_se->dl_non_contending = 0; dl_se->dl_overrun = 0; + dl_se->dl_server = 0; #ifdef CONFIG_RT_MUTEXES dl_se->pi_se = dl_se; diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 1cd92b11b289..07f555857698 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1131,6 +1131,8 @@ static inline void update_curr_task(struct task_struct *p, s64 delta_exec) trace_sched_stat_runtime(p, delta_exec); account_group_exec_runtime(p, delta_exec); cgroup_account_cputime(p, delta_exec); + if (p->dl_server) + dl_server_update(p->dl_server, delta_exec); } /* diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 1cda787172f0..8a70d51ffa33 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -313,6 +313,33 @@ extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *att extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); extern int dl_bw_check_overflow(int cpu); +/* + * SCHED_DEADLINE supports servers (nested scheduling) with the following + * interface: + * + * dl_se::rq -- runqueue we belong to. + * + * dl_se::server_has_tasks() -- used on bandwidth enforcement; we 'stop' the + * server when it runs out of tasks to run. + * + * dl_se::server_pick() -- nested pick_next_task(); we yield the period if this + * returns NULL. + * + * dl_server_update() -- called from update_curr_common(), propagates runtime + * to the server. + * + * dl_server_start() + * dl_server_stop() -- start/stop the server when it has (no) tasks. + * + * dl_server_init() -- initializes the server. + */ +extern void dl_server_update(struct sched_dl_entity *dl_se, s64 delta_exec); +extern void dl_server_start(struct sched_dl_entity *dl_se); +extern void dl_server_stop(struct sched_dl_entity *dl_se); +extern void dl_server_init(struct sched_dl_entity *dl_se, struct rq *rq, + dl_server_has_tasks_f has_tasks, + dl_server_pick_f pick); + #ifdef CONFIG_CGROUP_SCHED struct cfs_rq; From dd5403869a40595eb953f12e8cd2bb57bb88bb67 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 14 Nov 2023 14:38:39 -0500 Subject: [PATCH 11/30] sched/cpuidle: Comment about timers requirements VS idle handler Add missing explanation concerning IRQs re-enablement constraints in the cpuidle path against timers. Signed-off-by: Frederic Weisbecker Signed-off-by: Peter Zijlstra (Intel) Acked-by: Rafael J. Wysocki Link: https://lkml.kernel.org/r/20231114193840.4041-2-frederic@kernel.org --- kernel/sched/idle.c | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+) diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c index 565f8374ddbb..31231925f1ec 100644 --- a/kernel/sched/idle.c +++ b/kernel/sched/idle.c @@ -258,6 +258,36 @@ static void do_idle(void) while (!need_resched()) { rmb(); + /* + * Interrupts shouldn't be re-enabled from that point on until + * the CPU sleeping instruction is reached. Otherwise an interrupt + * may fire and queue a timer that would be ignored until the CPU + * wakes from the sleeping instruction. And testing need_resched() + * doesn't tell about pending needed timer reprogram. + * + * Several cases to consider: + * + * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as + * "wfi" or "mwait" are fine because they can be entered with + * interrupt disabled. + * + * - sti;mwait() couple is fine because the interrupts are + * re-enabled only upon the execution of mwait, leaving no gap + * in-between. + * + * - ROLLBACK based idle handlers with the sleeping instruction + * called with interrupts enabled are NOT fine. In this scheme + * when the interrupt detects it has interrupted an idle handler, + * it rolls back to its beginning which performs the + * need_resched() check before re-executing the sleeping + * instruction. This can leak a pending needed timer reprogram. + * If such a scheme is really mandatory due to the lack of an + * appropriate CPU sleeping instruction, then a FAST-FORWARD + * must instead be applied: when the interrupt detects it has + * interrupted an idle handler, it must resume to the end of + * this idle handler so that the generic idle loop is iterated + * again to reprogram the tick. + */ local_irq_disable(); if (cpu_is_offline(cpu)) { From 194600008d5c43b5a4ba98c4b81633397e34ffad Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 14 Nov 2023 14:38:40 -0500 Subject: [PATCH 12/30] sched/timers: Explain why idle task schedules out on remote timer enqueue Trying to avoid that didn't bring much value after testing, add comment about this. Signed-off-by: Frederic Weisbecker Signed-off-by: Peter Zijlstra (Intel) Acked-by: Rafael J. Wysocki Link: https://lkml.kernel.org/r/20231114193840.4041-3-frederic@kernel.org --- kernel/sched/core.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index f5f4495d1768..2de77a6d5ef8 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -1131,6 +1131,28 @@ static void wake_up_idle_cpu(int cpu) if (cpu == smp_processor_id()) return; + /* + * Set TIF_NEED_RESCHED and send an IPI if in the non-polling + * part of the idle loop. This forces an exit from the idle loop + * and a round trip to schedule(). Now this could be optimized + * because a simple new idle loop iteration is enough to + * re-evaluate the next tick. Provided some re-ordering of tick + * nohz functions that would need to follow TIF_NR_POLLING + * clearing: + * + * - On most archs, a simple fetch_or on ti::flags with a + * "0" value would be enough to know if an IPI needs to be sent. + * + * - x86 needs to perform a last need_resched() check between + * monitor and mwait which doesn't take timers into account. + * There a dedicated TIF_TIMER flag would be required to + * fetch_or here and be checked along with TIF_NEED_RESCHED + * before mwait(). + * + * However, remote timer enqueue is not such a frequent event + * and testing of the above solutions didn't appear to report + * much benefits. + */ if (set_nr_and_not_polling(rq->idle)) smp_send_reschedule(cpu); else From 50181c0cff31281b9f1071575ffba8a102375ece Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Wed, 22 Nov 2023 15:01:19 +0100 Subject: [PATCH 13/30] sched/pelt: Avoid underestimation of task utilization Lukasz Luba reported that a thread's util_est can significantly decrease as a result of sharing the CPU with other threads. The use case can be easily reproduced with a periodic task TA that runs 1ms and sleeps 100us. When the task is alone on the CPU, its max utilization and its util_est is around 888. If another similar task starts to run on the same CPU, TA will have to share the CPU runtime and its maximum utilization will decrease around half the CPU capacity (512) then TA's util_est will follow this new maximum trend which is only the result of sharing the CPU with others tasks. Such situation can be detected with runnable_avg wich is close or equal to util_avg when TA is alone, but increases above util_avg when TA shares the CPU with other threads and wait on the runqueue. [ We prefer an util_est that overestimate rather than under estimate because in 1st case we will not provide enough performance to the task which will remain under-provisioned, whereas in the other case we will create some idle time which will enable to reduce contention and as a result reduces the util_est so the overestimate will be transient whereas the underestimate will remain. ] [ mingo: Refined the changelog, added comments from the LKML discussion. ] Reported-by: Lukasz Luba Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Link: https://lore.kernel.org/lkml/CAKfTPtDd-HhF-YiNTtL9i5k0PfJbF819Yxu4YquzfXgwi7voyw@mail.gmail.com/#t Link: https://lore.kernel.org/r/20231122140119.472110-1-vincent.guittot@linaro.org Cc: Hongyan Xia --- kernel/sched/fair.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 07f555857698..53dea95ad8c9 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4774,6 +4774,11 @@ static inline unsigned long task_util(struct task_struct *p) return READ_ONCE(p->se.avg.util_avg); } +static inline unsigned long task_runnable(struct task_struct *p) +{ + return READ_ONCE(p->se.avg.runnable_avg); +} + static inline unsigned long _task_util_est(struct task_struct *p) { struct util_est ue = READ_ONCE(p->se.avg.util_est); @@ -4892,6 +4897,14 @@ static inline void util_est_update(struct cfs_rq *cfs_rq, if (task_util(p) > arch_scale_cpu_capacity(cpu_of(rq_of(cfs_rq)))) return; + /* + * To avoid underestimate of task utilization, skip updates of EWMA if + * we cannot grant that thread got all CPU time it wanted. + */ + if ((ue.enqueued + UTIL_EST_MARGIN) < task_runnable(p)) + goto done; + + /* * Update Task's estimated utilization * From 9c0b4bb7f6303c9c4e2e34984c46f5a86478f84d Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Wed, 22 Nov 2023 14:39:03 +0100 Subject: [PATCH 14/30] sched/cpufreq: Rework schedutil governor performance estimation The current method to take into account uclamp hints when estimating the target frequency can end in a situation where the selected target frequency is finally higher than uclamp hints, whereas there are no real needs. Such cases mainly happen because we are currently mixing the traditional scheduler utilization signal with the uclamp performance hints. By adding these 2 metrics, we loose an important information when it comes to select the target frequency, and we have to make some assumptions which can't fit all cases. Rework the interface between the scheduler and schedutil governor in order to propagate all information down to the cpufreq governor. effective_cpu_util() interface changes and now returns the actual utilization of the CPU with 2 optional inputs: - The minimum performance for this CPU; typically the capacity to handle the deadline task and the interrupt pressure. But also uclamp_min request when available. - The maximum targeting performance for this CPU which reflects the maximum level that we would like to not exceed. By default it will be the CPU capacity but can be reduced because of some performance hints set with uclamp. The value can be lower than actual utilization and/or min performance level. A new sugov_effective_cpu_perf() interface is also available to compute the final performance level that is targeted for the CPU, after applying some cpufreq headroom and taking into account all inputs. With these 2 functions, schedutil is now able to decide when it must go above uclamp hints. It now also has a generic way to get the min performance level. The dependency between energy model and cpufreq governor and its headroom policy doesn't exist anymore. eenv_pd_max_util() asks schedutil for the targeted performance after applying the impact of the waking task. [ mingo: Refined the changelog & C comments. ] Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Acked-by: Rafael J. Wysocki Link: https://lore.kernel.org/r/20231122133904.446032-2-vincent.guittot@linaro.org --- include/linux/energy_model.h | 1 - kernel/sched/core.c | 90 ++++++++++++++------------------ kernel/sched/cpufreq_schedutil.c | 35 +++++++++---- kernel/sched/fair.c | 22 ++++++-- kernel/sched/sched.h | 24 +++------ 5 files changed, 89 insertions(+), 83 deletions(-) diff --git a/include/linux/energy_model.h b/include/linux/energy_model.h index b9caa01dfac4..adec808b371a 100644 --- a/include/linux/energy_model.h +++ b/include/linux/energy_model.h @@ -243,7 +243,6 @@ static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, scale_cpu = arch_scale_cpu_capacity(cpu); ps = &pd->table[pd->nr_perf_states - 1]; - max_util = map_util_perf(max_util); max_util = min(max_util, allowed_cpu_cap); freq = map_util_freq(max_util, ps->frequency, scale_cpu); diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 2de77a6d5ef8..db4be4921e7f 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -7467,18 +7467,13 @@ int sched_core_idle_cpu(int cpu) * required to meet deadlines. */ unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, - enum cpu_util_type type, - struct task_struct *p) + unsigned long *min, + unsigned long *max) { - unsigned long dl_util, util, irq, max; + unsigned long util, irq, scale; struct rq *rq = cpu_rq(cpu); - max = arch_scale_cpu_capacity(cpu); - - if (!uclamp_is_used() && - type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { - return max; - } + scale = arch_scale_cpu_capacity(cpu); /* * Early check to see if IRQ/steal time saturates the CPU, can be @@ -7486,45 +7481,49 @@ unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, * update_irq_load_avg(). */ irq = cpu_util_irq(rq); - if (unlikely(irq >= max)) - return max; + if (unlikely(irq >= scale)) { + if (min) + *min = scale; + if (max) + *max = scale; + return scale; + } + + if (min) { + /* + * The minimum utilization returns the highest level between: + * - the computed DL bandwidth needed with the IRQ pressure which + * steals time to the deadline task. + * - The minimum performance requirement for CFS and/or RT. + */ + *min = max(irq + cpu_bw_dl(rq), uclamp_rq_get(rq, UCLAMP_MIN)); + + /* + * When an RT task is runnable and uclamp is not used, we must + * ensure that the task will run at maximum compute capacity. + */ + if (!uclamp_is_used() && rt_rq_is_runnable(&rq->rt)) + *min = max(*min, scale); + } /* * Because the time spend on RT/DL tasks is visible as 'lost' time to * CFS tasks and we use the same metric to track the effective * utilization (PELT windows are synchronized) we can directly add them * to obtain the CPU's actual utilization. - * - * CFS and RT utilization can be boosted or capped, depending on - * utilization clamp constraints requested by currently RUNNABLE - * tasks. - * When there are no CFS RUNNABLE tasks, clamps are released and - * frequency will be gracefully reduced with the utilization decay. */ util = util_cfs + cpu_util_rt(rq); - if (type == FREQUENCY_UTIL) - util = uclamp_rq_util_with(rq, util, p); - - dl_util = cpu_util_dl(rq); + util += cpu_util_dl(rq); /* - * For frequency selection we do not make cpu_util_dl() a permanent part - * of this sum because we want to use cpu_bw_dl() later on, but we need - * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such - * that we select f_max when there is no idle time. - * - * NOTE: numerical errors or stop class might cause us to not quite hit - * saturation when we should -- something for later. + * The maximum hint is a soft bandwidth requirement, which can be lower + * than the actual utilization because of uclamp_max requirements. */ - if (util + dl_util >= max) - return max; + if (max) + *max = min(scale, uclamp_rq_get(rq, UCLAMP_MAX)); - /* - * OTOH, for energy computation we need the estimated running time, so - * include util_dl and ignore dl_bw. - */ - if (type == ENERGY_UTIL) - util += dl_util; + if (util >= scale) + return scale; /* * There is still idle time; further improve the number by using the @@ -7535,28 +7534,15 @@ unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, * U' = irq + --------- * U * max */ - util = scale_irq_capacity(util, irq, max); + util = scale_irq_capacity(util, irq, scale); util += irq; - /* - * Bandwidth required by DEADLINE must always be granted while, for - * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism - * to gracefully reduce the frequency when no tasks show up for longer - * periods of time. - * - * Ideally we would like to set bw_dl as min/guaranteed freq and util + - * bw_dl as requested freq. However, cpufreq is not yet ready for such - * an interface. So, we only do the latter for now. - */ - if (type == FREQUENCY_UTIL) - util += cpu_bw_dl(rq); - - return min(max, util); + return min(scale, util); } 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), NULL, NULL); } #endif /* CONFIG_SMP */ diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c index 5888176354e2..f3acf2cf26ed 100644 --- a/kernel/sched/cpufreq_schedutil.c +++ b/kernel/sched/cpufreq_schedutil.c @@ -47,7 +47,7 @@ struct sugov_cpu { u64 last_update; unsigned long util; - unsigned long bw_dl; + unsigned long bw_min; /* The field below is for single-CPU policies only: */ #ifdef CONFIG_NO_HZ_COMMON @@ -143,7 +143,6 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy, unsigned int freq = arch_scale_freq_invariant() ? policy->cpuinfo.max_freq : policy->cur; - util = map_util_perf(util); freq = map_util_freq(util, freq, max); if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update) @@ -153,14 +152,30 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy, return cpufreq_driver_resolve_freq(policy, freq); } +unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual, + unsigned long min, + unsigned long max) +{ + /* Add dvfs headroom to actual utilization */ + actual = map_util_perf(actual); + /* Actually we don't need to target the max performance */ + if (actual < max) + max = actual; + + /* + * Ensure at least minimum performance while providing more compute + * capacity when possible. + */ + return max(min, max); +} + static void sugov_get_util(struct sugov_cpu *sg_cpu) { - unsigned long util = cpu_util_cfs_boost(sg_cpu->cpu); - struct rq *rq = cpu_rq(sg_cpu->cpu); + unsigned long min, max, util = cpu_util_cfs_boost(sg_cpu->cpu); - sg_cpu->bw_dl = cpu_bw_dl(rq); - sg_cpu->util = effective_cpu_util(sg_cpu->cpu, util, - FREQUENCY_UTIL, NULL); + util = effective_cpu_util(sg_cpu->cpu, util, &min, &max); + sg_cpu->bw_min = min; + sg_cpu->util = sugov_effective_cpu_perf(sg_cpu->cpu, util, min, max); } /** @@ -306,7 +321,7 @@ static inline bool sugov_cpu_is_busy(struct sugov_cpu *sg_cpu) { return false; } */ static inline void ignore_dl_rate_limit(struct sugov_cpu *sg_cpu) { - if (cpu_bw_dl(cpu_rq(sg_cpu->cpu)) > sg_cpu->bw_dl) + if (cpu_bw_dl(cpu_rq(sg_cpu->cpu)) > sg_cpu->bw_min) sg_cpu->sg_policy->limits_changed = true; } @@ -407,8 +422,8 @@ static void sugov_update_single_perf(struct update_util_data *hook, u64 time, sugov_cpu_is_busy(sg_cpu) && sg_cpu->util < prev_util) sg_cpu->util = prev_util; - cpufreq_driver_adjust_perf(sg_cpu->cpu, map_util_perf(sg_cpu->bw_dl), - map_util_perf(sg_cpu->util), max_cap); + cpufreq_driver_adjust_perf(sg_cpu->cpu, sg_cpu->bw_min, + sg_cpu->util, max_cap); sg_cpu->sg_policy->last_freq_update_time = time; } diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 53dea95ad8c9..34fe6e9490c2 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -7793,7 +7793,7 @@ static inline void eenv_pd_busy_time(struct energy_env *eenv, for_each_cpu(cpu, pd_cpus) { unsigned long util = cpu_util(cpu, p, -1, 0); - busy_time += effective_cpu_util(cpu, util, ENERGY_UTIL, NULL); + busy_time += effective_cpu_util(cpu, util, NULL, NULL); } eenv->pd_busy_time = min(eenv->pd_cap, busy_time); @@ -7816,7 +7816,7 @@ eenv_pd_max_util(struct energy_env *eenv, struct cpumask *pd_cpus, for_each_cpu(cpu, pd_cpus) { struct task_struct *tsk = (cpu == dst_cpu) ? p : NULL; unsigned long util = cpu_util(cpu, p, dst_cpu, 1); - unsigned long eff_util; + unsigned long eff_util, min, max; /* * Performance domain frequency: utilization clamping @@ -7825,7 +7825,23 @@ eenv_pd_max_util(struct energy_env *eenv, struct cpumask *pd_cpus, * NOTE: in case RT tasks are running, by default the * FREQUENCY_UTIL's utilization can be max OPP. */ - eff_util = effective_cpu_util(cpu, util, FREQUENCY_UTIL, tsk); + eff_util = effective_cpu_util(cpu, util, &min, &max); + + /* Task's uclamp can modify min and max value */ + if (tsk && uclamp_is_used()) { + min = max(min, uclamp_eff_value(p, UCLAMP_MIN)); + + /* + * If there is no active max uclamp constraint, + * directly use task's one, otherwise keep max. + */ + if (uclamp_rq_is_idle(cpu_rq(cpu))) + max = uclamp_eff_value(p, UCLAMP_MAX); + else + max = max(max, uclamp_eff_value(p, UCLAMP_MAX)); + } + + eff_util = sugov_effective_cpu_perf(cpu, eff_util, min, max); max_util = max(max_util, eff_util); } diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 8a70d51ffa33..c1574cd388e7 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2994,24 +2994,14 @@ static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} #endif #ifdef CONFIG_SMP -/** - * enum cpu_util_type - CPU utilization type - * @FREQUENCY_UTIL: Utilization used to select frequency - * @ENERGY_UTIL: Utilization used during energy calculation - * - * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time - * need to be aggregated differently depending on the usage made of them. This - * enum is used within effective_cpu_util() to differentiate the types of - * utilization expected by the callers, and adjust the aggregation accordingly. - */ -enum cpu_util_type { - FREQUENCY_UTIL, - ENERGY_UTIL, -}; - unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, - enum cpu_util_type type, - struct task_struct *p); + unsigned long *min, + unsigned long *max); + +unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual, + unsigned long min, + unsigned long max); + /* * Verify the fitness of task @p to run on @cpu taking into account the From f12560779f9d734446508f3df17f5632e9aaa2c8 Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Wed, 22 Nov 2023 14:39:04 +0100 Subject: [PATCH 15/30] sched/cpufreq: Rework iowait boost Use the max value that has already been computed inside sugov_get_util() to cap the iowait boost and remove dependency with uclamp_rq_util_with() which is not used anymore. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Acked-by: Rafael J. Wysocki Link: https://lore.kernel.org/r/20231122133904.446032-3-vincent.guittot@linaro.org --- kernel/sched/cpufreq_schedutil.c | 29 ++++++++------- kernel/sched/sched.h | 60 -------------------------------- 2 files changed, 14 insertions(+), 75 deletions(-) diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c index f3acf2cf26ed..4ee8ad70be99 100644 --- a/kernel/sched/cpufreq_schedutil.c +++ b/kernel/sched/cpufreq_schedutil.c @@ -169,11 +169,12 @@ unsigned long sugov_effective_cpu_perf(int cpu, unsigned long actual, return max(min, max); } -static void sugov_get_util(struct sugov_cpu *sg_cpu) +static void sugov_get_util(struct sugov_cpu *sg_cpu, unsigned long boost) { unsigned long min, max, util = cpu_util_cfs_boost(sg_cpu->cpu); util = effective_cpu_util(sg_cpu->cpu, util, &min, &max); + util = max(util, boost); sg_cpu->bw_min = min; sg_cpu->util = sugov_effective_cpu_perf(sg_cpu->cpu, util, min, max); } @@ -266,18 +267,16 @@ static void sugov_iowait_boost(struct sugov_cpu *sg_cpu, u64 time, * This mechanism is designed to boost high frequently IO waiting tasks, while * being more conservative on tasks which does sporadic IO operations. */ -static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time, +static unsigned long sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time, unsigned long max_cap) { - unsigned long boost; - /* No boost currently required */ if (!sg_cpu->iowait_boost) - return; + return 0; /* Reset boost if the CPU appears to have been idle enough */ if (sugov_iowait_reset(sg_cpu, time, false)) - return; + return 0; if (!sg_cpu->iowait_boost_pending) { /* @@ -286,7 +285,7 @@ static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time, sg_cpu->iowait_boost >>= 1; if (sg_cpu->iowait_boost < IOWAIT_BOOST_MIN) { sg_cpu->iowait_boost = 0; - return; + return 0; } } @@ -296,10 +295,7 @@ static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time, * sg_cpu->util is already in capacity scale; convert iowait_boost * into the same scale so we can compare. */ - boost = (sg_cpu->iowait_boost * max_cap) >> SCHED_CAPACITY_SHIFT; - boost = uclamp_rq_util_with(cpu_rq(sg_cpu->cpu), boost, NULL); - if (sg_cpu->util < boost) - sg_cpu->util = boost; + return (sg_cpu->iowait_boost * max_cap) >> SCHED_CAPACITY_SHIFT; } #ifdef CONFIG_NO_HZ_COMMON @@ -329,6 +325,8 @@ static inline bool sugov_update_single_common(struct sugov_cpu *sg_cpu, u64 time, unsigned long max_cap, unsigned int flags) { + unsigned long boost; + sugov_iowait_boost(sg_cpu, time, flags); sg_cpu->last_update = time; @@ -337,8 +335,8 @@ static inline bool sugov_update_single_common(struct sugov_cpu *sg_cpu, if (!sugov_should_update_freq(sg_cpu->sg_policy, time)) return false; - sugov_get_util(sg_cpu); - sugov_iowait_apply(sg_cpu, time, max_cap); + boost = sugov_iowait_apply(sg_cpu, time, max_cap); + sugov_get_util(sg_cpu, boost); return true; } @@ -439,9 +437,10 @@ static unsigned int sugov_next_freq_shared(struct sugov_cpu *sg_cpu, u64 time) for_each_cpu(j, policy->cpus) { struct sugov_cpu *j_sg_cpu = &per_cpu(sugov_cpu, j); + unsigned long boost; - sugov_get_util(j_sg_cpu); - sugov_iowait_apply(j_sg_cpu, time, max_cap); + boost = sugov_iowait_apply(j_sg_cpu, time, max_cap); + sugov_get_util(j_sg_cpu, boost); util = max(j_sg_cpu->util, util); } diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index c1574cd388e7..e58a54bda77d 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -3058,59 +3058,6 @@ static inline bool uclamp_rq_is_idle(struct rq *rq) return rq->uclamp_flags & UCLAMP_FLAG_IDLE; } -/** - * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values. - * @rq: The rq to clamp against. Must not be NULL. - * @util: The util value to clamp. - * @p: The task to clamp against. Can be NULL if you want to clamp - * against @rq only. - * - * Clamps the passed @util to the max(@rq, @p) effective uclamp values. - * - * If sched_uclamp_used static key is disabled, then just return the util - * without any clamping since uclamp aggregation at the rq level in the fast - * path is disabled, rendering this operation a NOP. - * - * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It - * will return the correct effective uclamp value of the task even if the - * static key is disabled. - */ -static __always_inline -unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, - struct task_struct *p) -{ - unsigned long min_util = 0; - unsigned long max_util = 0; - - if (!static_branch_likely(&sched_uclamp_used)) - return util; - - if (p) { - min_util = uclamp_eff_value(p, UCLAMP_MIN); - max_util = uclamp_eff_value(p, UCLAMP_MAX); - - /* - * Ignore last runnable task's max clamp, as this task will - * reset it. Similarly, no need to read the rq's min clamp. - */ - if (uclamp_rq_is_idle(rq)) - goto out; - } - - min_util = max_t(unsigned long, min_util, uclamp_rq_get(rq, UCLAMP_MIN)); - max_util = max_t(unsigned long, max_util, uclamp_rq_get(rq, UCLAMP_MAX)); -out: - /* - * Since CPU's {min,max}_util clamps are MAX aggregated considering - * RUNNABLE tasks with _different_ clamps, we can end up with an - * inversion. Fix it now when the clamps are applied. - */ - if (unlikely(min_util >= max_util)) - return min_util; - - return clamp(util, min_util, max_util); -} - /* Is the rq being capped/throttled by uclamp_max? */ static inline bool uclamp_rq_is_capped(struct rq *rq) { @@ -3148,13 +3095,6 @@ static inline unsigned long uclamp_eff_value(struct task_struct *p, return SCHED_CAPACITY_SCALE; } -static inline -unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, - struct task_struct *p) -{ - return util; -} - static inline bool uclamp_rq_is_capped(struct rq *rq) { return false; } static inline bool uclamp_is_used(void) From 0257e5a3c26b3810831359d39c0821397af8bf29 Mon Sep 17 00:00:00 2001 From: Wenyu Huang Date: Sat, 25 Nov 2023 02:05:27 +0000 Subject: [PATCH 16/30] sched/doc: Update documentation after renames and synchronize Chinese version Update the documentation after these changes, which didn't entirely propagate the changes: e23edc86b09d ("sched/fair: Rename check_preempt_curr() to wakeup_preempt()") 03b7fad167ef ("sched: Add task_struct pointer to sched_class::set_curr_task") 2f88c8e802c8 ("sched/eevdf/doc: Modify the documented knob to base_slice_ns as well") [ mingo: Reworked the changelog. ] Signed-off-by: Wenyu Huang Signed-off-by: Ingo Molnar Cc: linux-kernel@vger.kernel.org --- Documentation/scheduler/sched-design-CFS.rst | 8 ++++---- .../translations/zh_CN/scheduler/sched-design-CFS.rst | 8 ++++---- 2 files changed, 8 insertions(+), 8 deletions(-) diff --git a/Documentation/scheduler/sched-design-CFS.rst b/Documentation/scheduler/sched-design-CFS.rst index f68919800f05..6cffffe26500 100644 --- a/Documentation/scheduler/sched-design-CFS.rst +++ b/Documentation/scheduler/sched-design-CFS.rst @@ -180,7 +180,7 @@ This is the (partial) list of the hooks: compat_yield sysctl is turned on; in that case, it places the scheduling entity at the right-most end of the red-black tree. - - check_preempt_curr(...) + - wakeup_preempt(...) This function checks if a task that entered the runnable state should preempt the currently running task. @@ -189,10 +189,10 @@ This is the (partial) list of the hooks: This function chooses the most appropriate task eligible to run next. - - set_curr_task(...) + - set_next_task(...) - This function is called when a task changes its scheduling class or changes - its task group. + This function is called when a task changes its scheduling class, changes + its task group or is scheduled. - task_tick(...) diff --git a/Documentation/translations/zh_CN/scheduler/sched-design-CFS.rst b/Documentation/translations/zh_CN/scheduler/sched-design-CFS.rst index 3076402406c4..abc6709ec3b2 100644 --- a/Documentation/translations/zh_CN/scheduler/sched-design-CFS.rst +++ b/Documentation/translations/zh_CN/scheduler/sched-design-CFS.rst @@ -80,7 +80,7 @@ p->se.vruntime。一旦p->se.vruntime变得足够大，其它的任务将成为 CFS使用纳秒粒度的计时，不依赖于任何jiffies或HZ的细节。因此CFS并不像之前的调度器那样 有“时间片”的概念，也没有任何启发式的设计。唯一可调的参数（你需要打开CONFIG_SCHED_DEBUG）是： - /sys/kernel/debug/sched/min_granularity_ns + /sys/kernel/debug/sched/base_slice_ns 它可以用来将调度器从“桌面”模式（也就是低时延）调节为“服务器”（也就是高批处理）模式。 它的默认设置是适合桌面的工作负载。SCHED_BATCH也被CFS调度器模块处理。 @@ -147,7 +147,7 @@ array）。 这个函数的行为基本上是出队，紧接着入队，除非compat_yield sysctl被开启。在那种情况下， 它将调度实体放在红黑树的最右端。 - - check_preempt_curr(...) + - wakeup_preempt(...) 这个函数检查进入可运行状态的任务能否抢占当前正在运行的任务。 @@ -155,9 +155,9 @@ array）。 这个函数选择接下来最适合运行的任务。 - - set_curr_task(...) + - set_next_task(...) - 这个函数在任务改变调度类或改变任务组时被调用。 + 这个函数在任务改变调度类，改变任务组时，或者任务被调度时被调用。 - task_tick(...) From 5068d84054b766efe7c6202fc71b2350d1c326f1 Mon Sep 17 00:00:00 2001 From: Yiwei Lin Date: Fri, 17 Nov 2023 16:01:06 +0800 Subject: [PATCH 17/30] sched/fair: Update min_vruntime for reweight_entity() correctly Since reweight_entity() may have chance to change the weight of cfs_rq->curr entity, we should also update_min_vruntime() if this is the case Fixes: eab03c23c2a1 ("sched/eevdf: Fix vruntime adjustment on reweight") Signed-off-by: Yiwei Lin Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Abel Wu Link: https://lore.kernel.org/r/20231117080106.12890-1-s921975628@gmail.com --- kernel/sched/fair.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 34fe6e9490c2..bcea3d55d95d 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -3815,17 +3815,17 @@ static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, enqueue_load_avg(cfs_rq, se); if (se->on_rq) { update_load_add(&cfs_rq->load, se->load.weight); - if (!curr) { - /* - * The entity's vruntime has been adjusted, so let's check - * whether the rq-wide min_vruntime needs updated too. Since - * the calculations above require stable min_vruntime rather - * than up-to-date one, we do the update at the end of the - * reweight process. - */ + if (!curr) __enqueue_entity(cfs_rq, se); - update_min_vruntime(cfs_rq); - } + + /* + * The entity's vruntime has been adjusted, so let's check + * whether the rq-wide min_vruntime needs updated too. Since + * the calculations above require stable min_vruntime rather + * than up-to-date one, we do the update at the end of the + * reweight process. + */ + update_min_vruntime(cfs_rq); } } From 418146e39891ef1fb2284dee4cabbfe616cd21cf Mon Sep 17 00:00:00 2001 From: Elliot Berman Date: Mon, 20 Nov 2023 09:36:32 -0800 Subject: [PATCH 18/30] freezer,sched: Clean saved_state when restoring it during thaw Clean saved_state after using it during thaw. Cleaning the saved_state allows us to avoid some unnecessary branches in ttwu_state_match. Signed-off-by: Elliot Berman Signed-off-by: Peter Zijlstra (Intel) Link: https://lore.kernel.org/r/20231120-freezer-state-multiple-thaws-v1-2-f2e1dd7ce5a2@quicinc.com --- kernel/freezer.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/freezer.c b/kernel/freezer.c index c450fa8b8b5e..43b1d1b94d9e 100644 --- a/kernel/freezer.c +++ b/kernel/freezer.c @@ -187,6 +187,7 @@ static int __restore_freezer_state(struct task_struct *p, void *arg) if (state != TASK_RUNNING) { WRITE_ONCE(p->__state, state); + p->saved_state = TASK_RUNNING; return 1; } From 9942cb22ea458c34fa17b73d143ea32d4df1caca Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:49 +0100 Subject: [PATCH 19/30] sched/topology: Add a new arch_scale_freq_ref() method Create a new method to get a unique and fixed max frequency. Currently cpuinfo.max_freq or the highest (or last) state of performance domain are used as the max frequency when computing the frequency for a level of utilization, but: - cpuinfo_max_freq can change at runtime. boost is one example of such change. - cpuinfo.max_freq and last item of the PD can be different leading to different results between cpufreq and energy model. We need to save the reference frequency that has been used when computing the CPUs capacity and use this fixed and coherent value to convert between frequency and CPU's capacity. In fact, we already save the frequency that has been used when computing the capacity of each CPU. We extend the precision to save kHz instead of MHz currently and we modify the type to be aligned with other variables used when converting frequency to capacity and the other way. [ mingo: Minor edits. ] Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Acked-by: Sudeep Holla Link: https://lore.kernel.org/r/20231211104855.558096-2-vincent.guittot@linaro.org --- arch/arm/include/asm/topology.h | 1 + arch/arm64/include/asm/topology.h | 1 + arch/riscv/include/asm/topology.h | 1 + drivers/base/arch_topology.c | 29 ++++++++++++++--------------- include/linux/arch_topology.h | 7 +++++++ include/linux/sched/topology.h | 8 ++++++++ 6 files changed, 32 insertions(+), 15 deletions(-) diff --git a/arch/arm/include/asm/topology.h b/arch/arm/include/asm/topology.h index c7d2510e5a78..853c4f81ba4a 100644 --- a/arch/arm/include/asm/topology.h +++ b/arch/arm/include/asm/topology.h @@ -13,6 +13,7 @@ #define arch_set_freq_scale topology_set_freq_scale #define arch_scale_freq_capacity topology_get_freq_scale #define arch_scale_freq_invariant topology_scale_freq_invariant +#define arch_scale_freq_ref topology_get_freq_ref #endif /* Replace task scheduler's default cpu-invariant accounting */ diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h index 9fab663dd2de..a323b109b9c4 100644 --- a/arch/arm64/include/asm/topology.h +++ b/arch/arm64/include/asm/topology.h @@ -23,6 +23,7 @@ void update_freq_counters_refs(void); #define arch_set_freq_scale topology_set_freq_scale #define arch_scale_freq_capacity topology_get_freq_scale #define arch_scale_freq_invariant topology_scale_freq_invariant +#define arch_scale_freq_ref topology_get_freq_ref #ifdef CONFIG_ACPI_CPPC_LIB #define arch_init_invariance_cppc topology_init_cpu_capacity_cppc diff --git a/arch/riscv/include/asm/topology.h b/arch/riscv/include/asm/topology.h index e316ab3b77f3..61183688bdd5 100644 --- a/arch/riscv/include/asm/topology.h +++ b/arch/riscv/include/asm/topology.h @@ -9,6 +9,7 @@ #define arch_set_freq_scale topology_set_freq_scale #define arch_scale_freq_capacity topology_get_freq_scale #define arch_scale_freq_invariant topology_scale_freq_invariant +#define arch_scale_freq_ref topology_get_freq_ref /* Replace task scheduler's default cpu-invariant accounting */ #define arch_scale_cpu_capacity topology_get_cpu_scale diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c index b741b5ba82bd..0c9ae5b157b1 100644 --- a/drivers/base/arch_topology.c +++ b/drivers/base/arch_topology.c @@ -19,6 +19,7 @@ #include #include #include +#include #define CREATE_TRACE_POINTS #include @@ -26,7 +27,8 @@ static DEFINE_PER_CPU(struct scale_freq_data __rcu *, sft_data); static struct cpumask scale_freq_counters_mask; static bool scale_freq_invariant; -static DEFINE_PER_CPU(u32, freq_factor) = 1; +DEFINE_PER_CPU(unsigned long, capacity_freq_ref) = 1; +EXPORT_PER_CPU_SYMBOL_GPL(capacity_freq_ref); static bool supports_scale_freq_counters(const struct cpumask *cpus) { @@ -170,9 +172,9 @@ DEFINE_PER_CPU(unsigned long, thermal_pressure); * operating on stale data when hot-plug is used for some CPUs. The * @capped_freq reflects the currently allowed max CPUs frequency due to * thermal capping. It might be also a boost frequency value, which is bigger - * than the internal 'freq_factor' max frequency. In such case the pressure - * value should simply be removed, since this is an indication that there is - * no thermal throttling. The @capped_freq must be provided in kHz. + * than the internal 'capacity_freq_ref' max frequency. In such case the + * pressure value should simply be removed, since this is an indication that + * there is no thermal throttling. The @capped_freq must be provided in kHz. */ void topology_update_thermal_pressure(const struct cpumask *cpus, unsigned long capped_freq) @@ -183,10 +185,7 @@ void topology_update_thermal_pressure(const struct cpumask *cpus, cpu = cpumask_first(cpus); max_capacity = arch_scale_cpu_capacity(cpu); - max_freq = per_cpu(freq_factor, cpu); - - /* Convert to MHz scale which is used in 'freq_factor' */ - capped_freq /= 1000; + max_freq = arch_scale_freq_ref(cpu); /* * Handle properly the boost frequencies, which should simply clean @@ -279,13 +278,13 @@ void topology_normalize_cpu_scale(void) capacity_scale = 1; for_each_possible_cpu(cpu) { - capacity = raw_capacity[cpu] * per_cpu(freq_factor, cpu); + capacity = raw_capacity[cpu] * per_cpu(capacity_freq_ref, cpu); capacity_scale = max(capacity, capacity_scale); } pr_debug("cpu_capacity: capacity_scale=%llu\n", capacity_scale); for_each_possible_cpu(cpu) { - capacity = raw_capacity[cpu] * per_cpu(freq_factor, cpu); + capacity = raw_capacity[cpu] * per_cpu(capacity_freq_ref, cpu); capacity = div64_u64(capacity << SCHED_CAPACITY_SHIFT, capacity_scale); topology_set_cpu_scale(cpu, capacity); @@ -321,15 +320,15 @@ bool __init topology_parse_cpu_capacity(struct device_node *cpu_node, int cpu) cpu_node, raw_capacity[cpu]); /* - * Update freq_factor for calculating early boot cpu capacities. + * Update capacity_freq_ref for calculating early boot CPU capacities. * For non-clk CPU DVFS mechanism, there's no way to get the * frequency value now, assuming they are running at the same - * frequency (by keeping the initial freq_factor value). + * frequency (by keeping the initial capacity_freq_ref value). */ cpu_clk = of_clk_get(cpu_node, 0); if (!PTR_ERR_OR_ZERO(cpu_clk)) { - per_cpu(freq_factor, cpu) = - clk_get_rate(cpu_clk) / 1000; + per_cpu(capacity_freq_ref, cpu) = + clk_get_rate(cpu_clk) / HZ_PER_KHZ; clk_put(cpu_clk); } } else { @@ -411,7 +410,7 @@ init_cpu_capacity_callback(struct notifier_block *nb, cpumask_andnot(cpus_to_visit, cpus_to_visit, policy->related_cpus); for_each_cpu(cpu, policy->related_cpus) - per_cpu(freq_factor, cpu) = policy->cpuinfo.max_freq / 1000; + per_cpu(capacity_freq_ref, cpu) = policy->cpuinfo.max_freq; if (cpumask_empty(cpus_to_visit)) { topology_normalize_cpu_scale(); diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h index a07b510e7dc5..32c24ff4f2a8 100644 --- a/include/linux/arch_topology.h +++ b/include/linux/arch_topology.h @@ -27,6 +27,13 @@ static inline unsigned long topology_get_cpu_scale(int cpu) void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity); +DECLARE_PER_CPU(unsigned long, capacity_freq_ref); + +static inline unsigned long topology_get_freq_ref(int cpu) +{ + return per_cpu(capacity_freq_ref, cpu); +} + DECLARE_PER_CPU(unsigned long, arch_freq_scale); static inline unsigned long topology_get_freq_scale(int cpu) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index de545ba85218..a6e04b4a21d7 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -279,6 +279,14 @@ void arch_update_thermal_pressure(const struct cpumask *cpus, { } #endif +#ifndef arch_scale_freq_ref +static __always_inline +unsigned int arch_scale_freq_ref(int cpu) +{ + return 0; +} +#endif + static inline int task_node(const struct task_struct *p) { return cpu_to_node(task_cpu(p)); From 599457ba15403037b489fe536266a3d5f9efaed7 Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:50 +0100 Subject: [PATCH 20/30] cpufreq: Use the fixed and coherent frequency for scaling capacity cpuinfo.max_freq can change at runtime because of boost as an example. This implies that the value could be different from the frequency that has been used to compute the capacity of a CPU. The new arch_scale_freq_ref() returns a fixed and coherent frequency that can be used to compute the capacity for a given frequency. [ Also fix a arch_set_freq_scale() newline style wart in . ] Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Acked-by: Viresh Kumar Acked-by: Rafael J. Wysocki Link: https://lore.kernel.org/r/20231211104855.558096-3-vincent.guittot@linaro.org --- drivers/cpufreq/cpufreq.c | 4 ++-- include/linux/cpufreq.h | 1 + 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index 934d35f570b7..44db4f59c4cc 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -454,7 +454,7 @@ void cpufreq_freq_transition_end(struct cpufreq_policy *policy, arch_set_freq_scale(policy->related_cpus, policy->cur, - policy->cpuinfo.max_freq); + arch_scale_freq_ref(policy->cpu)); spin_lock(&policy->transition_lock); policy->transition_ongoing = false; @@ -2174,7 +2174,7 @@ unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy, policy->cur = freq; arch_set_freq_scale(policy->related_cpus, freq, - policy->cpuinfo.max_freq); + arch_scale_freq_ref(policy->cpu)); cpufreq_stats_record_transition(policy, freq); if (trace_cpu_frequency_enabled()) { diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h index 1c5ca92a0555..afda5f24d3dd 100644 --- a/include/linux/cpufreq.h +++ b/include/linux/cpufreq.h @@ -1203,6 +1203,7 @@ void arch_set_freq_scale(const struct cpumask *cpus, { } #endif + /* the following are really really optional */ extern struct freq_attr cpufreq_freq_attr_scaling_available_freqs; extern struct freq_attr cpufreq_freq_attr_scaling_boost_freqs; From b3edde44e5d4504c23a176819865cd603fd16d6c Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:51 +0100 Subject: [PATCH 21/30] cpufreq/schedutil: Use a fixed reference frequency cpuinfo.max_freq can change at runtime because of boost as an example. This implies that the value could be different than the one that has been used when computing the capacity of a CPU. The new arch_scale_freq_ref() returns a fixed and coherent reference frequency that can be used when computing a frequency based on utilization. Use this arch_scale_freq_ref() when available and fallback to policy otherwise. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Reviewed-by: Dietmar Eggemann Acked-by: Rafael J. Wysocki Acked-by: Viresh Kumar Link: https://lore.kernel.org/r/20231211104855.558096-4-vincent.guittot@linaro.org --- kernel/sched/cpufreq_schedutil.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c index 4ee8ad70be99..95c3c097083e 100644 --- a/kernel/sched/cpufreq_schedutil.c +++ b/kernel/sched/cpufreq_schedutil.c @@ -114,6 +114,28 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy) } } +/** + * get_capacity_ref_freq - get the reference frequency that has been used to + * correlate frequency and compute capacity for a given cpufreq policy. We use + * the CPU managing it for the arch_scale_freq_ref() call in the function. + * @policy: the cpufreq policy of the CPU in question. + * + * Return: the reference CPU frequency to compute a capacity. + */ +static __always_inline +unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy) +{ + unsigned int freq = arch_scale_freq_ref(policy->cpu); + + if (freq) + return freq; + + if (arch_scale_freq_invariant()) + return policy->cpuinfo.max_freq; + + return policy->cur; +} + /** * get_next_freq - Compute a new frequency for a given cpufreq policy. * @sg_policy: schedutil policy object to compute the new frequency for. @@ -140,9 +162,9 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy, unsigned long util, unsigned long max) { struct cpufreq_policy *policy = sg_policy->policy; - unsigned int freq = arch_scale_freq_invariant() ? - policy->cpuinfo.max_freq : policy->cur; + unsigned int freq; + freq = get_capacity_ref_freq(policy); freq = map_util_freq(util, freq, max); if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update) From 15cbbd1d317e07b4e5c6aca5d4c5579539a82784 Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:52 +0100 Subject: [PATCH 22/30] energy_model: Use a fixed reference frequency The last item of a performance domain is not always the performance point that has been used to compute CPU's capacity. This can lead to different target frequency compared with other part of the system like schedutil and would result in wrong energy estimation. A new arch_scale_freq_ref() is available to return a fixed and coherent frequency reference that can be used when computing the CPU's frequency for an level of utilization. Use this function to get this reference frequency. Energy model is never used without defining arch_scale_freq_ref() but can be compiled. Define a default arch_scale_freq_ref() returning 0 in such case. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Link: https://lore.kernel.org/r/20231211104855.558096-5-vincent.guittot@linaro.org --- include/linux/energy_model.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/include/linux/energy_model.h b/include/linux/energy_model.h index adec808b371a..88d91e087471 100644 --- a/include/linux/energy_model.h +++ b/include/linux/energy_model.h @@ -224,7 +224,7 @@ static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, unsigned long max_util, unsigned long sum_util, unsigned long allowed_cpu_cap) { - unsigned long freq, scale_cpu; + unsigned long freq, ref_freq, scale_cpu; struct em_perf_state *ps; int cpu; @@ -241,10 +241,10 @@ static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, */ cpu = cpumask_first(to_cpumask(pd->cpus)); scale_cpu = arch_scale_cpu_capacity(cpu); - ps = &pd->table[pd->nr_perf_states - 1]; + ref_freq = arch_scale_freq_ref(cpu); max_util = min(max_util, allowed_cpu_cap); - freq = map_util_freq(max_util, ps->frequency, scale_cpu); + freq = map_util_freq(max_util, ref_freq, scale_cpu); /* * Find the lowest performance state of the Energy Model above the From 50b813b147e9eb6546a1fc49d4e703e6d23691f2 Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:53 +0100 Subject: [PATCH 23/30] cpufreq/cppc: Move and rename cppc_cpufreq_{perf_to_khz|khz_to_perf}() Move and rename cppc_cpufreq_perf_to_khz() and cppc_cpufreq_khz_to_perf() to use them outside cppc_cpufreq in topology_init_cpu_capacity_cppc(). Modify the interface to use struct cppc_perf_caps *caps instead of struct cppc_cpudata *cpu_data as we only use the fields of cppc_perf_caps. cppc_cpufreq was converting the lowest and nominal freq from MHz to kHz before using them. We move this conversion inside cppc_perf_to_khz and cppc_khz_to_perf to make them generic and usable outside cppc_cpufreq. No functional change Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Pierre Gondois Acked-by: Rafael J. Wysocki Acked-by: Viresh Kumar Link: https://lore.kernel.org/r/20231211104855.558096-6-vincent.guittot@linaro.org --- drivers/acpi/cppc_acpi.c | 104 ++++++++++++++++++++++++ drivers/cpufreq/cppc_cpufreq.c | 139 ++++----------------------------- include/acpi/cppc_acpi.h | 2 + 3 files changed, 123 insertions(+), 122 deletions(-) diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c index 7ff269a78c20..d155a86a8614 100644 --- a/drivers/acpi/cppc_acpi.c +++ b/drivers/acpi/cppc_acpi.c @@ -39,6 +39,9 @@ #include #include #include +#include +#include +#include #include @@ -1760,3 +1763,104 @@ unsigned int cppc_get_transition_latency(int cpu_num) return latency_ns; } EXPORT_SYMBOL_GPL(cppc_get_transition_latency); + +/* Minimum struct length needed for the DMI processor entry we want */ +#define DMI_ENTRY_PROCESSOR_MIN_LENGTH 48 + +/* Offset in the DMI processor structure for the max frequency */ +#define DMI_PROCESSOR_MAX_SPEED 0x14 + +/* Callback function used to retrieve the max frequency from DMI */ +static void cppc_find_dmi_mhz(const struct dmi_header *dm, void *private) +{ + const u8 *dmi_data = (const u8 *)dm; + u16 *mhz = (u16 *)private; + + if (dm->type == DMI_ENTRY_PROCESSOR && + dm->length >= DMI_ENTRY_PROCESSOR_MIN_LENGTH) { + u16 val = (u16)get_unaligned((const u16 *) + (dmi_data + DMI_PROCESSOR_MAX_SPEED)); + *mhz = val > *mhz ? val : *mhz; + } +} + +/* Look up the max frequency in DMI */ +static u64 cppc_get_dmi_max_khz(void) +{ + u16 mhz = 0; + + dmi_walk(cppc_find_dmi_mhz, &mhz); + + /* + * Real stupid fallback value, just in case there is no + * actual value set. + */ + mhz = mhz ? mhz : 1; + + return KHZ_PER_MHZ * mhz; +} + +/* + * If CPPC lowest_freq and nominal_freq registers are exposed then we can + * use them to convert perf to freq and vice versa. The conversion is + * extrapolated as an affine function passing by the 2 points: + * - (Low perf, Low freq) + * - (Nominal perf, Nominal freq) + */ +unsigned int cppc_perf_to_khz(struct cppc_perf_caps *caps, unsigned int perf) +{ + s64 retval, offset = 0; + static u64 max_khz; + u64 mul, div; + + if (caps->lowest_freq && caps->nominal_freq) { + mul = caps->nominal_freq - caps->lowest_freq; + mul *= KHZ_PER_MHZ; + div = caps->nominal_perf - caps->lowest_perf; + offset = caps->nominal_freq * KHZ_PER_MHZ - + div64_u64(caps->nominal_perf * mul, div); + } else { + if (!max_khz) + max_khz = cppc_get_dmi_max_khz(); + mul = max_khz; + div = caps->highest_perf; + } + + retval = offset + div64_u64(perf * mul, div); + if (retval >= 0) + return retval; + return 0; +} +EXPORT_SYMBOL_GPL(cppc_perf_to_khz); + +unsigned int cppc_khz_to_perf(struct cppc_perf_caps *caps, unsigned int freq) +{ + s64 retval, offset = 0; + static u64 max_khz; + u64 mul, div; + + if (caps->lowest_freq && caps->nominal_freq) { + mul = caps->nominal_perf - caps->lowest_perf; + div = caps->nominal_freq - caps->lowest_freq; + /* + * We don't need to convert to kHz for computing offset and can + * directly use nominal_freq and lowest_freq as the div64_u64 + * will remove the frequency unit. + */ + offset = caps->nominal_perf - + div64_u64(caps->nominal_freq * mul, div); + /* But we need it for computing the perf level. */ + div *= KHZ_PER_MHZ; + } else { + if (!max_khz) + max_khz = cppc_get_dmi_max_khz(); + mul = caps->highest_perf; + div = max_khz; + } + + retval = offset + div64_u64(freq * mul, div); + if (retval >= 0) + return retval; + return 0; +} +EXPORT_SYMBOL_GPL(cppc_khz_to_perf); diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c index fe08ca419b3d..64420d9cfd1e 100644 --- a/drivers/cpufreq/cppc_cpufreq.c +++ b/drivers/cpufreq/cppc_cpufreq.c @@ -16,7 +16,6 @@ #include #include #include -#include #include #include #include @@ -27,12 +26,6 @@ #include -/* Minimum struct length needed for the DMI processor entry we want */ -#define DMI_ENTRY_PROCESSOR_MIN_LENGTH 48 - -/* Offset in the DMI processor structure for the max frequency */ -#define DMI_PROCESSOR_MAX_SPEED 0x14 - /* * This list contains information parsed from per CPU ACPI _CPC and _PSD * structures: e.g. the highest and lowest supported performance, capabilities, @@ -291,97 +284,9 @@ static inline void cppc_freq_invariance_exit(void) } #endif /* CONFIG_ACPI_CPPC_CPUFREQ_FIE */ -/* Callback function used to retrieve the max frequency from DMI */ -static void cppc_find_dmi_mhz(const struct dmi_header *dm, void *private) -{ - const u8 *dmi_data = (const u8 *)dm; - u16 *mhz = (u16 *)private; - - if (dm->type == DMI_ENTRY_PROCESSOR && - dm->length >= DMI_ENTRY_PROCESSOR_MIN_LENGTH) { - u16 val = (u16)get_unaligned((const u16 *) - (dmi_data + DMI_PROCESSOR_MAX_SPEED)); - *mhz = val > *mhz ? val : *mhz; - } -} - -/* Look up the max frequency in DMI */ -static u64 cppc_get_dmi_max_khz(void) -{ - u16 mhz = 0; - - dmi_walk(cppc_find_dmi_mhz, &mhz); - - /* - * Real stupid fallback value, just in case there is no - * actual value set. - */ - mhz = mhz ? mhz : 1; - - return (1000 * mhz); -} - -/* - * If CPPC lowest_freq and nominal_freq registers are exposed then we can - * use them to convert perf to freq and vice versa. The conversion is - * extrapolated as an affine function passing by the 2 points: - * - (Low perf, Low freq) - * - (Nominal perf, Nominal perf) - */ -static unsigned int cppc_cpufreq_perf_to_khz(struct cppc_cpudata *cpu_data, - unsigned int perf) -{ - struct cppc_perf_caps *caps = &cpu_data->perf_caps; - s64 retval, offset = 0; - static u64 max_khz; - u64 mul, div; - - if (caps->lowest_freq && caps->nominal_freq) { - mul = caps->nominal_freq - caps->lowest_freq; - div = caps->nominal_perf - caps->lowest_perf; - offset = caps->nominal_freq - div64_u64(caps->nominal_perf * mul, div); - } else { - if (!max_khz) - max_khz = cppc_get_dmi_max_khz(); - mul = max_khz; - div = caps->highest_perf; - } - - retval = offset + div64_u64(perf * mul, div); - if (retval >= 0) - return retval; - return 0; -} - -static unsigned int cppc_cpufreq_khz_to_perf(struct cppc_cpudata *cpu_data, - unsigned int freq) -{ - struct cppc_perf_caps *caps = &cpu_data->perf_caps; - s64 retval, offset = 0; - static u64 max_khz; - u64 mul, div; - - if (caps->lowest_freq && caps->nominal_freq) { - mul = caps->nominal_perf - caps->lowest_perf; - div = caps->nominal_freq - caps->lowest_freq; - offset = caps->nominal_perf - div64_u64(caps->nominal_freq * mul, div); - } else { - if (!max_khz) - max_khz = cppc_get_dmi_max_khz(); - mul = caps->highest_perf; - div = max_khz; - } - - retval = offset + div64_u64(freq * mul, div); - if (retval >= 0) - return retval; - return 0; -} - static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int target_freq, unsigned int relation) - { struct cppc_cpudata *cpu_data = policy->driver_data; unsigned int cpu = policy->cpu; @@ -389,7 +294,7 @@ static int cppc_cpufreq_set_target(struct cpufreq_policy *policy, u32 desired_perf; int ret = 0; - desired_perf = cppc_cpufreq_khz_to_perf(cpu_data, target_freq); + desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); /* Return if it is exactly the same perf */ if (desired_perf == cpu_data->perf_ctrls.desired_perf) return ret; @@ -417,7 +322,7 @@ static unsigned int cppc_cpufreq_fast_switch(struct cpufreq_policy *policy, u32 desired_perf; int ret; - desired_perf = cppc_cpufreq_khz_to_perf(cpu_data, target_freq); + desired_perf = cppc_khz_to_perf(&cpu_data->perf_caps, target_freq); cpu_data->perf_ctrls.desired_perf = desired_perf; ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls); @@ -530,7 +435,7 @@ static int cppc_get_cpu_power(struct device *cpu_dev, min_step = min_cap / CPPC_EM_CAP_STEP; max_step = max_cap / CPPC_EM_CAP_STEP; - perf_prev = cppc_cpufreq_khz_to_perf(cpu_data, *KHz); + perf_prev = cppc_khz_to_perf(perf_caps, *KHz); step = perf_prev / perf_step; if (step > max_step) @@ -550,8 +455,8 @@ static int cppc_get_cpu_power(struct device *cpu_dev, perf = step * perf_step; } - *KHz = cppc_cpufreq_perf_to_khz(cpu_data, perf); - perf_check = cppc_cpufreq_khz_to_perf(cpu_data, *KHz); + *KHz = cppc_perf_to_khz(perf_caps, perf); + perf_check = cppc_khz_to_perf(perf_caps, *KHz); step_check = perf_check / perf_step; /* @@ -561,8 +466,8 @@ static int cppc_get_cpu_power(struct device *cpu_dev, */ while ((*KHz == prev_freq) || (step_check != step)) { perf++; - *KHz = cppc_cpufreq_perf_to_khz(cpu_data, perf); - perf_check = cppc_cpufreq_khz_to_perf(cpu_data, *KHz); + *KHz = cppc_perf_to_khz(perf_caps, perf); + perf_check = cppc_khz_to_perf(perf_caps, *KHz); step_check = perf_check / perf_step; } @@ -591,7 +496,7 @@ static int cppc_get_cpu_cost(struct device *cpu_dev, unsigned long KHz, perf_caps = &cpu_data->perf_caps; max_cap = arch_scale_cpu_capacity(cpu_dev->id); - perf_prev = cppc_cpufreq_khz_to_perf(cpu_data, KHz); + perf_prev = cppc_khz_to_perf(perf_caps, KHz); perf_step = CPPC_EM_CAP_STEP * perf_caps->highest_perf / max_cap; step = perf_prev / perf_step; @@ -679,10 +584,6 @@ static struct cppc_cpudata *cppc_cpufreq_get_cpu_data(unsigned int cpu) goto free_mask; } - /* Convert the lowest and nominal freq from MHz to KHz */ - cpu_data->perf_caps.lowest_freq *= 1000; - cpu_data->perf_caps.nominal_freq *= 1000; - list_add(&cpu_data->node, &cpu_data_list); return cpu_data; @@ -724,20 +625,16 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy) * Set min to lowest nonlinear perf to avoid any efficiency penalty (see * Section 8.4.7.1.1.5 of ACPI 6.1 spec) */ - policy->min = cppc_cpufreq_perf_to_khz(cpu_data, - caps->lowest_nonlinear_perf); - policy->max = cppc_cpufreq_perf_to_khz(cpu_data, - caps->nominal_perf); + policy->min = cppc_perf_to_khz(caps, caps->lowest_nonlinear_perf); + policy->max = cppc_perf_to_khz(caps, caps->nominal_perf); /* * Set cpuinfo.min_freq to Lowest to make the full range of performance * available if userspace wants to use any perf between lowest & lowest * nonlinear perf */ - policy->cpuinfo.min_freq = cppc_cpufreq_perf_to_khz(cpu_data, - caps->lowest_perf); - policy->cpuinfo.max_freq = cppc_cpufreq_perf_to_khz(cpu_data, - caps->nominal_perf); + policy->cpuinfo.min_freq = cppc_perf_to_khz(caps, caps->lowest_perf); + policy->cpuinfo.max_freq = cppc_perf_to_khz(caps, caps->nominal_perf); policy->transition_delay_us = cppc_cpufreq_get_transition_delay_us(cpu); policy->shared_type = cpu_data->shared_type; @@ -773,7 +670,7 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy) boost_supported = true; /* Set policy->cur to max now. The governors will adjust later. */ - policy->cur = cppc_cpufreq_perf_to_khz(cpu_data, caps->highest_perf); + policy->cur = cppc_perf_to_khz(caps, caps->highest_perf); cpu_data->perf_ctrls.desired_perf = caps->highest_perf; ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls); @@ -863,7 +760,7 @@ static unsigned int cppc_cpufreq_get_rate(unsigned int cpu) delivered_perf = cppc_perf_from_fbctrs(cpu_data, &fb_ctrs_t0, &fb_ctrs_t1); - return cppc_cpufreq_perf_to_khz(cpu_data, delivered_perf); + return cppc_perf_to_khz(&cpu_data->perf_caps, delivered_perf); } static int cppc_cpufreq_set_boost(struct cpufreq_policy *policy, int state) @@ -878,11 +775,9 @@ static int cppc_cpufreq_set_boost(struct cpufreq_policy *policy, int state) } if (state) - policy->max = cppc_cpufreq_perf_to_khz(cpu_data, - caps->highest_perf); + policy->max = cppc_perf_to_khz(caps, caps->highest_perf); else - policy->max = cppc_cpufreq_perf_to_khz(cpu_data, - caps->nominal_perf); + policy->max = cppc_perf_to_khz(caps, caps->nominal_perf); policy->cpuinfo.max_freq = policy->max; ret = freq_qos_update_request(policy->max_freq_req, policy->max); @@ -937,7 +832,7 @@ static unsigned int hisi_cppc_cpufreq_get_rate(unsigned int cpu) if (ret < 0) return -EIO; - return cppc_cpufreq_perf_to_khz(cpu_data, desired_perf); + return cppc_perf_to_khz(&cpu_data->perf_caps, desired_perf); } static void cppc_check_hisi_workaround(void) diff --git a/include/acpi/cppc_acpi.h b/include/acpi/cppc_acpi.h index 6126c977ece0..3a0995f8bce8 100644 --- a/include/acpi/cppc_acpi.h +++ b/include/acpi/cppc_acpi.h @@ -144,6 +144,8 @@ extern int cppc_set_perf(int cpu, struct cppc_perf_ctrls *perf_ctrls); extern int cppc_set_enable(int cpu, bool enable); extern int cppc_get_perf_caps(int cpu, struct cppc_perf_caps *caps); extern bool cppc_perf_ctrs_in_pcc(void); +extern unsigned int cppc_perf_to_khz(struct cppc_perf_caps *caps, unsigned int perf); +extern unsigned int cppc_khz_to_perf(struct cppc_perf_caps *caps, unsigned int freq); extern bool acpi_cpc_valid(void); extern bool cppc_allow_fast_switch(void); extern int acpi_get_psd_map(unsigned int cpu, struct cppc_cpudata *cpu_data); From 5477fa249b56c59c3baa1b237bf083cffa64c84a Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:54 +0100 Subject: [PATCH 24/30] cpufreq/cppc: Set the frequency used for computing the capacity Save the frequency associated to the performance that has been used when initializing the capacity of CPUs. Also, cppc cpufreq driver can register an artificial energy model. In such case, it needs the frequency for this compute capacity. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Pierre Gondois Acked-by: Sudeep Holla Acked-by: Viresh Kumar Link: https://lore.kernel.org/r/20231211104855.558096-7-vincent.guittot@linaro.org --- drivers/base/arch_topology.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-) diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c index 0c9ae5b157b1..1aa76b5c96c2 100644 --- a/drivers/base/arch_topology.c +++ b/drivers/base/arch_topology.c @@ -349,6 +349,7 @@ bool __init topology_parse_cpu_capacity(struct device_node *cpu_node, int cpu) void topology_init_cpu_capacity_cppc(void) { + u64 capacity, capacity_scale = 0; struct cppc_perf_caps perf_caps; int cpu; @@ -365,6 +366,10 @@ void topology_init_cpu_capacity_cppc(void) (perf_caps.highest_perf >= perf_caps.nominal_perf) && (perf_caps.highest_perf >= perf_caps.lowest_perf)) { raw_capacity[cpu] = perf_caps.highest_perf; + capacity_scale = max_t(u64, capacity_scale, raw_capacity[cpu]); + + per_cpu(capacity_freq_ref, cpu) = cppc_perf_to_khz(&perf_caps, raw_capacity[cpu]); + pr_debug("cpu_capacity: CPU%d cpu_capacity=%u (raw).\n", cpu, raw_capacity[cpu]); continue; @@ -375,7 +380,15 @@ void topology_init_cpu_capacity_cppc(void) goto exit; } - topology_normalize_cpu_scale(); + for_each_possible_cpu(cpu) { + capacity = raw_capacity[cpu]; + capacity = div64_u64(capacity << SCHED_CAPACITY_SHIFT, + capacity_scale); + topology_set_cpu_scale(cpu, capacity); + pr_debug("cpu_capacity: CPU%d cpu_capacity=%lu\n", + cpu, topology_get_cpu_scale(cpu)); + } + schedule_work(&update_topology_flags_work); pr_debug("cpu_capacity: cpu_capacity initialization done\n"); From 1f023007f5e782bda19ad9104830c404fd622c5d Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Mon, 11 Dec 2023 11:48:55 +0100 Subject: [PATCH 25/30] arm64/amu: Use capacity_ref_freq() to set AMU ratio Use the new capacity_ref_freq() method to set the ratio that is used by AMU for computing the arch_scale_freq_capacity(). This helps to keep everything aligned using the same reference for computing CPUs capacity. The default value of the ratio (stored in per_cpu(arch_max_freq_scale)) ensures that arch_scale_freq_capacity() returns max capacity until it is set to its correct value with the cpu capacity and capacity_ref_freq(). Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Acked-by: Sudeep Holla Acked-by: Will Deacon Link: https://lore.kernel.org/r/20231211104855.558096-8-vincent.guittot@linaro.org --- arch/arm64/kernel/topology.c | 26 +++++++++++++------------- drivers/base/arch_topology.c | 12 +++++++++++- include/linux/arch_topology.h | 1 + 3 files changed, 25 insertions(+), 14 deletions(-) diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c index 817d788cd866..1a2c72f3e7f8 100644 --- a/arch/arm64/kernel/topology.c +++ b/arch/arm64/kernel/topology.c @@ -82,7 +82,12 @@ int __init parse_acpi_topology(void) #undef pr_fmt #define pr_fmt(fmt) "AMU: " fmt -static DEFINE_PER_CPU_READ_MOSTLY(unsigned long, arch_max_freq_scale); +/* + * Ensure that amu_scale_freq_tick() will return SCHED_CAPACITY_SCALE until + * the CPU capacity and its associated frequency have been correctly + * initialized. + */ +static DEFINE_PER_CPU_READ_MOSTLY(unsigned long, arch_max_freq_scale) = 1UL << (2 * SCHED_CAPACITY_SHIFT); static DEFINE_PER_CPU(u64, arch_const_cycles_prev); static DEFINE_PER_CPU(u64, arch_core_cycles_prev); static cpumask_var_t amu_fie_cpus; @@ -112,14 +117,14 @@ static inline bool freq_counters_valid(int cpu) return true; } -static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate) +void freq_inv_set_max_ratio(int cpu, u64 max_rate) { - u64 ratio; + u64 ratio, ref_rate = arch_timer_get_rate(); if (unlikely(!max_rate || !ref_rate)) { - pr_debug("CPU%d: invalid maximum or reference frequency.\n", + WARN_ONCE(1, "CPU%d: invalid maximum or reference frequency.\n", cpu); - return -EINVAL; + return; } /* @@ -139,12 +144,10 @@ static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate) ratio = div64_u64(ratio, max_rate); if (!ratio) { WARN_ONCE(1, "Reference frequency too low.\n"); - return -EINVAL; + return; } - per_cpu(arch_max_freq_scale, cpu) = (unsigned long)ratio; - - return 0; + WRITE_ONCE(per_cpu(arch_max_freq_scale, cpu), (unsigned long)ratio); } static void amu_scale_freq_tick(void) @@ -195,10 +198,7 @@ static void amu_fie_setup(const struct cpumask *cpus) return; for_each_cpu(cpu, cpus) { - if (!freq_counters_valid(cpu) || - freq_inv_set_max_ratio(cpu, - cpufreq_get_hw_max_freq(cpu) * 1000ULL, - arch_timer_get_rate())) + if (!freq_counters_valid(cpu)) return; } diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c index 1aa76b5c96c2..5aaa0865625d 100644 --- a/drivers/base/arch_topology.c +++ b/drivers/base/arch_topology.c @@ -344,6 +344,10 @@ bool __init topology_parse_cpu_capacity(struct device_node *cpu_node, int cpu) return !ret; } +void __weak freq_inv_set_max_ratio(int cpu, u64 max_rate) +{ +} + #ifdef CONFIG_ACPI_CPPC_LIB #include @@ -381,6 +385,9 @@ void topology_init_cpu_capacity_cppc(void) } for_each_possible_cpu(cpu) { + freq_inv_set_max_ratio(cpu, + per_cpu(capacity_freq_ref, cpu) * HZ_PER_KHZ); + capacity = raw_capacity[cpu]; capacity = div64_u64(capacity << SCHED_CAPACITY_SHIFT, capacity_scale); @@ -422,8 +429,11 @@ init_cpu_capacity_callback(struct notifier_block *nb, cpumask_andnot(cpus_to_visit, cpus_to_visit, policy->related_cpus); - for_each_cpu(cpu, policy->related_cpus) + for_each_cpu(cpu, policy->related_cpus) { per_cpu(capacity_freq_ref, cpu) = policy->cpuinfo.max_freq; + freq_inv_set_max_ratio(cpu, + per_cpu(capacity_freq_ref, cpu) * HZ_PER_KHZ); + } if (cpumask_empty(cpus_to_visit)) { topology_normalize_cpu_scale(); diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h index 32c24ff4f2a8..a63d61ca55af 100644 --- a/include/linux/arch_topology.h +++ b/include/linux/arch_topology.h @@ -99,6 +99,7 @@ void update_siblings_masks(unsigned int cpu); void remove_cpu_topology(unsigned int cpuid); void reset_cpu_topology(void); int parse_acpi_topology(void); +void freq_inv_set_max_ratio(int cpu, u64 max_rate); #endif #endif /* _LINUX_ARCH_TOPOLOGY_H_ */ From 7736ae5572eb344c090fbef9621a228e7e3d6276 Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Fri, 1 Dec 2023 17:16:51 +0100 Subject: [PATCH 26/30] sched/fair: Remove SCHED_FEAT(UTIL_EST_FASTUP, true) sched_feat(UTIL_EST_FASTUP) has been added to easily disable the feature in order to check for possibly related regressions. After 3 years, it has never been used and no regression has been reported. Let's remove it and make fast increase a permanent behavior. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Reviewed-by: Dietmar Eggemann Reviewed-by: Hongyan Xia Reviewed-by: Tang Yizhou Reviewed-by: Yanteng Si [for the Chinese translation] Reviewed-by: Alex Shi Link: https://lore.kernel.org/r/20231201161652.1241695-2-vincent.guittot@linaro.org --- Documentation/scheduler/schedutil.rst | 7 +++---- Documentation/translations/zh_CN/scheduler/schedutil.rst | 7 +++---- kernel/sched/fair.c | 8 +++----- kernel/sched/features.h | 1 - 4 files changed, 9 insertions(+), 14 deletions(-) diff --git a/Documentation/scheduler/schedutil.rst b/Documentation/scheduler/schedutil.rst index 32c7d69fc86c..803fba8fc714 100644 --- a/Documentation/scheduler/schedutil.rst +++ b/Documentation/scheduler/schedutil.rst @@ -90,8 +90,8 @@ For more detail see: - Documentation/scheduler/sched-capacity.rst:"1. CPU Capacity + 2. Task utilization" -UTIL_EST / UTIL_EST_FASTUP -========================== +UTIL_EST +======== Because periodic tasks have their averages decayed while they sleep, even though when running their expected utilization will be the same, they suffer a @@ -99,8 +99,7 @@ though when running their expected utilization will be the same, they suffer a To alleviate this (a default enabled option) UTIL_EST drives an Infinite Impulse Response (IIR) EWMA with the 'running' value on dequeue -- when it is -highest. A further default enabled option UTIL_EST_FASTUP modifies the IIR -filter to instantly increase and only decay on decrease. +highest. UTIL_EST filters to instantly increase and only decay on decrease. A further runqueue wide sum (of runnable tasks) is maintained of: diff --git a/Documentation/translations/zh_CN/scheduler/schedutil.rst b/Documentation/translations/zh_CN/scheduler/schedutil.rst index d1ea68007520..7c8d87f21c42 100644 --- a/Documentation/translations/zh_CN/scheduler/schedutil.rst +++ b/Documentation/translations/zh_CN/scheduler/schedutil.rst @@ -89,16 +89,15 @@ r_cpu被定义为当前CPU的最高性能水平与系统中任何其它CPU的最 - Documentation/translations/zh_CN/scheduler/sched-capacity.rst:"1. CPU Capacity + 2. Task utilization" -UTIL_EST / UTIL_EST_FASTUP -========================== +UTIL_EST +======== 由于周期性任务的平均数在睡眠时会衰减，而在运行时其预期利用率会和睡眠前相同， 因此它们在再次运行后会面临（DVFS）的上涨。 为了缓解这个问题，（一个默认使能的编译选项）UTIL_EST驱动一个无限脉冲响应 （Infinite Impulse Response，IIR）的EWMA，“运行”值在出队时是最高的。 -另一个默认使能的编译选项UTIL_EST_FASTUP修改了IIR滤波器，使其允许立即增加， -仅在利用率下降时衰减。 +UTIL_EST滤波使其在遇到更高值时立刻增加，而遇到低值时会缓慢衰减。 进一步，运行队列的（可运行任务的）利用率之和由下式计算： diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index bcea3d55d95d..e94d65da8d66 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4870,11 +4870,9 @@ static inline void util_est_update(struct cfs_rq *cfs_rq, * to smooth utilization decreases. */ ue.enqueued = task_util(p); - if (sched_feat(UTIL_EST_FASTUP)) { - if (ue.ewma < ue.enqueued) { - ue.ewma = ue.enqueued; - goto done; - } + if (ue.ewma < ue.enqueued) { + ue.ewma = ue.enqueued; + goto done; } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index a3ddf84de430..143f55df890b 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -83,7 +83,6 @@ SCHED_FEAT(WA_BIAS, true) * UtilEstimation. Use estimated CPU utilization. */ SCHED_FEAT(UTIL_EST, true) -SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) From 11137d384996bb05cf33c8163db271e1bac3f4bf Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Fri, 1 Dec 2023 17:16:52 +0100 Subject: [PATCH 27/30] sched/fair: Simplify util_est With UTIL_EST_FASTUP now being permanent, we can take advantage of the fact that the ewma jumps directly to a higher utilization at dequeue to simplify util_est and remove the enqueued field. Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Tested-by: Lukasz Luba Reviewed-by: Lukasz Luba Reviewed-by: Dietmar Eggemann Reviewed-by: Hongyan Xia Reviewed-by: Alex Shi Link: https://lore.kernel.org/r/20231201161652.1241695-3-vincent.guittot@linaro.org --- include/linux/sched.h | 49 +++++++------------------- kernel/sched/debug.c | 7 ++-- kernel/sched/fair.c | 82 ++++++++++++++++--------------------------- kernel/sched/pelt.h | 4 +-- 4 files changed, 48 insertions(+), 94 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index 8d258162deb0..03bfe9ab2951 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -415,42 +415,6 @@ struct load_weight { u32 inv_weight; }; -/** - * struct util_est - Estimation utilization of FAIR tasks - * @enqueued: instantaneous estimated utilization of a task/cpu - * @ewma: the Exponential Weighted Moving Average (EWMA) - * utilization of a task - * - * Support data structure to track an Exponential Weighted Moving Average - * (EWMA) of a FAIR task's utilization. New samples are added to the moving - * average each time a task completes an activation. Sample's weight is chosen - * so that the EWMA will be relatively insensitive to transient changes to the - * task's workload. - * - * The enqueued attribute has a slightly different meaning for tasks and cpus: - * - task: the task's util_avg at last task dequeue time - * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU - * Thus, the util_est.enqueued of a task represents the contribution on the - * estimated utilization of the CPU where that task is currently enqueued. - * - * Only for tasks we track a moving average of the past instantaneous - * estimated utilization. This allows to absorb sporadic drops in utilization - * of an otherwise almost periodic task. - * - * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg - * updates. When a task is dequeued, its util_est should not be updated if its - * util_avg has not been updated in the meantime. - * This information is mapped into the MSB bit of util_est.enqueued at dequeue - * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg - * for a task) it is safe to use MSB. - */ -struct util_est { - unsigned int enqueued; - unsigned int ewma; -#define UTIL_EST_WEIGHT_SHIFT 2 -#define UTIL_AVG_UNCHANGED 0x80000000 -} __attribute__((__aligned__(sizeof(u64)))); - /* * The load/runnable/util_avg accumulates an infinite geometric series * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c). @@ -505,9 +469,20 @@ struct sched_avg { unsigned long load_avg; unsigned long runnable_avg; unsigned long util_avg; - struct util_est util_est; + unsigned int util_est; } ____cacheline_aligned; +/* + * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg + * updates. When a task is dequeued, its util_est should not be updated if its + * util_avg has not been updated in the meantime. + * This information is mapped into the MSB bit of util_est at dequeue time. + * Since max value of util_est for a task is 1024 (PELT util_avg for a task) + * it is safe to use MSB. + */ +#define UTIL_EST_WEIGHT_SHIFT 2 +#define UTIL_AVG_UNCHANGED 0x80000000 + struct sched_statistics { #ifdef CONFIG_SCHEDSTATS u64 wait_start; diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c index 168eecc209b4..8d5d98a5834d 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -684,8 +684,8 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) cfs_rq->avg.runnable_avg); SEQ_printf(m, " .%-30s: %lu\n", "util_avg", cfs_rq->avg.util_avg); - SEQ_printf(m, " .%-30s: %u\n", "util_est_enqueued", - cfs_rq->avg.util_est.enqueued); + SEQ_printf(m, " .%-30s: %u\n", "util_est", + cfs_rq->avg.util_est); SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", cfs_rq->removed.load_avg); SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", @@ -1075,8 +1075,7 @@ void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, P(se.avg.runnable_avg); P(se.avg.util_avg); P(se.avg.last_update_time); - P(se.avg.util_est.ewma); - PM(se.avg.util_est.enqueued, ~UTIL_AVG_UNCHANGED); + PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED); #endif #ifdef CONFIG_UCLAMP_TASK __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index e94d65da8d66..823dd76d0546 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4781,9 +4781,7 @@ static inline unsigned long task_runnable(struct task_struct *p) static inline unsigned long _task_util_est(struct task_struct *p) { - struct util_est ue = READ_ONCE(p->se.avg.util_est); - - return max(ue.ewma, (ue.enqueued & ~UTIL_AVG_UNCHANGED)); + return READ_ONCE(p->se.avg.util_est) & ~UTIL_AVG_UNCHANGED; } static inline unsigned long task_util_est(struct task_struct *p) @@ -4800,9 +4798,9 @@ static inline void util_est_enqueue(struct cfs_rq *cfs_rq, return; /* Update root cfs_rq's estimated utilization */ - enqueued = cfs_rq->avg.util_est.enqueued; + enqueued = cfs_rq->avg.util_est; enqueued += _task_util_est(p); - WRITE_ONCE(cfs_rq->avg.util_est.enqueued, enqueued); + WRITE_ONCE(cfs_rq->avg.util_est, enqueued); trace_sched_util_est_cfs_tp(cfs_rq); } @@ -4816,34 +4814,20 @@ static inline void util_est_dequeue(struct cfs_rq *cfs_rq, return; /* Update root cfs_rq's estimated utilization */ - enqueued = cfs_rq->avg.util_est.enqueued; + enqueued = cfs_rq->avg.util_est; enqueued -= min_t(unsigned int, enqueued, _task_util_est(p)); - WRITE_ONCE(cfs_rq->avg.util_est.enqueued, enqueued); + WRITE_ONCE(cfs_rq->avg.util_est, enqueued); trace_sched_util_est_cfs_tp(cfs_rq); } #define UTIL_EST_MARGIN (SCHED_CAPACITY_SCALE / 100) -/* - * Check if a (signed) value is within a specified (unsigned) margin, - * based on the observation that: - * - * abs(x) < y := (unsigned)(x + y - 1) < (2 * y - 1) - * - * NOTE: this only works when value + margin < INT_MAX. - */ -static inline bool within_margin(int value, int margin) -{ - return ((unsigned int)(value + margin - 1) < (2 * margin - 1)); -} - static inline void util_est_update(struct cfs_rq *cfs_rq, struct task_struct *p, bool task_sleep) { - long last_ewma_diff, last_enqueued_diff; - struct util_est ue; + unsigned int ewma, dequeued, last_ewma_diff; if (!sched_feat(UTIL_EST)) return; @@ -4855,23 +4839,25 @@ static inline void util_est_update(struct cfs_rq *cfs_rq, if (!task_sleep) return; + /* Get current estimate of utilization */ + ewma = READ_ONCE(p->se.avg.util_est); + /* * If the PELT values haven't changed since enqueue time, * skip the util_est update. */ - ue = p->se.avg.util_est; - if (ue.enqueued & UTIL_AVG_UNCHANGED) + if (ewma & UTIL_AVG_UNCHANGED) return; - last_enqueued_diff = ue.enqueued; + /* Get utilization at dequeue */ + dequeued = task_util(p); /* * Reset EWMA on utilization increases, the moving average is used only * to smooth utilization decreases. */ - ue.enqueued = task_util(p); - if (ue.ewma < ue.enqueued) { - ue.ewma = ue.enqueued; + if (ewma <= dequeued) { + ewma = dequeued; goto done; } @@ -4879,27 +4865,22 @@ static inline void util_est_update(struct cfs_rq *cfs_rq, * Skip update of task's estimated utilization when its members are * already ~1% close to its last activation value. */ - last_ewma_diff = ue.enqueued - ue.ewma; - last_enqueued_diff -= ue.enqueued; - if (within_margin(last_ewma_diff, UTIL_EST_MARGIN)) { - if (!within_margin(last_enqueued_diff, UTIL_EST_MARGIN)) - goto done; - - return; - } + last_ewma_diff = ewma - dequeued; + if (last_ewma_diff < UTIL_EST_MARGIN) + goto done; /* * To avoid overestimation of actual task utilization, skip updates if * we cannot grant there is idle time in this CPU. */ - if (task_util(p) > arch_scale_cpu_capacity(cpu_of(rq_of(cfs_rq)))) + if (dequeued > arch_scale_cpu_capacity(cpu_of(rq_of(cfs_rq)))) return; /* * To avoid underestimate of task utilization, skip updates of EWMA if * we cannot grant that thread got all CPU time it wanted. */ - if ((ue.enqueued + UTIL_EST_MARGIN) < task_runnable(p)) + if ((dequeued + UTIL_EST_MARGIN) < task_runnable(p)) goto done; @@ -4907,25 +4888,24 @@ static inline void util_est_update(struct cfs_rq *cfs_rq, * Update Task's estimated utilization * * When *p completes an activation we can consolidate another sample - * of the task size. This is done by storing the current PELT value - * as ue.enqueued and by using this value to update the Exponential - * Weighted Moving Average (EWMA): + * of the task size. This is done by using this value to update the + * Exponential Weighted Moving Average (EWMA): * * ewma(t) = w * task_util(p) + (1-w) * ewma(t-1) * = w * task_util(p) + ewma(t-1) - w * ewma(t-1) * = w * (task_util(p) - ewma(t-1)) + ewma(t-1) - * = w * ( last_ewma_diff ) + ewma(t-1) - * = w * (last_ewma_diff + ewma(t-1) / w) + * = w * ( -last_ewma_diff ) + ewma(t-1) + * = w * (-last_ewma_diff + ewma(t-1) / w) * * Where 'w' is the weight of new samples, which is configured to be * 0.25, thus making w=1/4 ( >>= UTIL_EST_WEIGHT_SHIFT) */ - ue.ewma <<= UTIL_EST_WEIGHT_SHIFT; - ue.ewma += last_ewma_diff; - ue.ewma >>= UTIL_EST_WEIGHT_SHIFT; + ewma <<= UTIL_EST_WEIGHT_SHIFT; + ewma -= last_ewma_diff; + ewma >>= UTIL_EST_WEIGHT_SHIFT; done: - ue.enqueued |= UTIL_AVG_UNCHANGED; - WRITE_ONCE(p->se.avg.util_est, ue); + ewma |= UTIL_AVG_UNCHANGED; + WRITE_ONCE(p->se.avg.util_est, ewma); trace_sched_util_est_se_tp(&p->se); } @@ -7653,16 +7633,16 @@ cpu_util(int cpu, struct task_struct *p, int dst_cpu, int boost) if (sched_feat(UTIL_EST)) { unsigned long util_est; - util_est = READ_ONCE(cfs_rq->avg.util_est.enqueued); + util_est = READ_ONCE(cfs_rq->avg.util_est); /* * During wake-up @p isn't enqueued yet and doesn't contribute - * to any cpu_rq(cpu)->cfs.avg.util_est.enqueued. + * to any cpu_rq(cpu)->cfs.avg.util_est. * If @dst_cpu == @cpu add it to "simulate" cpu_util after @p * has been enqueued. * * During exec (@dst_cpu = -1) @p is enqueued and does - * contribute to cpu_rq(cpu)->cfs.util_est.enqueued. + * contribute to cpu_rq(cpu)->cfs.util_est. * Remove it to "simulate" cpu_util without @p's contribution. * * Despite the task_on_rq_queued(@p) check there is still a diff --git a/kernel/sched/pelt.h b/kernel/sched/pelt.h index 3a0e0dc28721..9e1083465fbc 100644 --- a/kernel/sched/pelt.h +++ b/kernel/sched/pelt.h @@ -52,13 +52,13 @@ static inline void cfs_se_util_change(struct sched_avg *avg) return; /* Avoid store if the flag has been already reset */ - enqueued = avg->util_est.enqueued; + enqueued = avg->util_est; 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); + WRITE_ONCE(avg->util_est, enqueued); } static inline u64 rq_clock_pelt(struct rq *rq) From 3af7524b14198f5159a86692d57a9f28ec9375ce Mon Sep 17 00:00:00 2001 From: Pierre Gondois Date: Wed, 6 Dec 2023 10:00:43 +0100 Subject: [PATCH 28/30] sched/fair: Use all little CPUs for CPU-bound workloads Running N CPU-bound tasks on an N CPUs platform: - with asymmetric CPU capacity - not being a DynamIq system (i.e. having a PKG level sched domain without the SD_SHARE_PKG_RESOURCES flag set) .. might result in a task placement where two tasks run on a big CPU and none on a little CPU. This placement could be more optimal by using all CPUs. Testing platform: Juno-r2: - 2 big CPUs (1-2), maximum capacity of 1024 - 4 little CPUs (0,3-5), maximum capacity of 383 Testing workload ([1]): Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks is affine to a CPU, except for: - one little CPU which is left idle. - one big CPU which has 2 tasks affine. After the 100ms (step 2), remove the cpumask affinity. Behavior before the patch: During step 2, the load balancer running from the idle CPU tags sched domains as: - little CPUs: 'group_has_spare'. Cf. group_has_capacity() and group_is_overloaded(), 3 CPU-bound tasks run on a 4 CPUs sched-domain, and the idle CPU provides enough spare capacity regarding the imbalance_pct - big CPUs: 'group_overloaded'. Indeed, 3 tasks run on a 2 CPUs sched-domain, so the following path is used: group_is_overloaded() \-if (sgs->sum_nr_running <= sgs->group_weight) return true; The following path which would change the migration type to 'migrate_task' is not taken: calculate_imbalance() \-if (env->idle != CPU_NOT_IDLE && env->imbalance == 0) as the local group has some spare capacity, so the imbalance is not 0. The migration type requested is 'migrate_util' and the busiest runqueue is the big CPU's runqueue having 2 tasks (each having a utilization of 512). The idle little CPU cannot pull one of these task as its capacity is too small for the task. The following path is used: detach_tasks() \-case migrate_util: \-if (util > env->imbalance) goto next; After the patch: As the number of failed balancing attempts grows (with 'nr_balance_failed'), progressively make it easier to migrate a big task to the idling little CPU. A similar mechanism is used for the 'migrate_load' migration type. Improvement: Running the testing workload [1] with the step 2 representing a ~10s load for a big CPU: Before patch: ~19.3s After patch: ~18s (-6.7%) Similar issue reported at: https://lore.kernel.org/lkml/20230716014125.139577-1-qyousef@layalina.io/ Suggested-by: Vincent Guittot Signed-off-by: Pierre Gondois Signed-off-by: Ingo Molnar Reviewed-by: Vincent Guittot Reviewed-by: Dietmar Eggemann Acked-by: Qais Yousef Link: https://lore.kernel.org/r/20231206090043.634697-1-pierre.gondois@arm.com --- kernel/sched/fair.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 823dd76d0546..1d561b5aae2a 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -9071,7 +9071,7 @@ static int detach_tasks(struct lb_env *env) case migrate_util: util = task_util_est(p); - if (util > env->imbalance) + if (shr_bound(util, env->sd->nr_balance_failed) > env->imbalance) goto next; env->imbalance -= util; From fbb66ce0b1d670c72def736a13ac9176b860df4e Mon Sep 17 00:00:00 2001 From: Wang Jinchao Date: Thu, 14 Dec 2023 13:20:29 +0800 Subject: [PATCH 29/30] sched/fair: Remove unused 'next_buddy_marked' local variable in check_preempt_wakeup_fair() This variable became unused in: 5e963f2bd465 ("sched/fair: Commit to EEVDF") Signed-off-by: Wang Jinchao Signed-off-by: Ingo Molnar Reviewed-by: Vincent Guittot Link: https://lore.kernel.org/r/202312141319+0800-wangjinchao@xfusion.com --- kernel/sched/fair.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 1d561b5aae2a..9cc20855dc2b 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -8221,7 +8221,6 @@ static void check_preempt_wakeup_fair(struct rq *rq, struct task_struct *p, int struct task_struct *curr = rq->curr; struct sched_entity *se = &curr->se, *pse = &p->se; struct cfs_rq *cfs_rq = task_cfs_rq(curr); - int next_buddy_marked = 0; int cse_is_idle, pse_is_idle; if (unlikely(se == pse)) @@ -8238,7 +8237,6 @@ static void check_preempt_wakeup_fair(struct rq *rq, struct task_struct *p, int if (sched_feat(NEXT_BUDDY) && !(wake_flags & WF_FORK)) { set_next_buddy(pse); - next_buddy_marked = 1; } /* From f60a631ab9ed5df15e446269ea515f2b8948ba0c Mon Sep 17 00:00:00 2001 From: Vincent Guittot Date: Thu, 21 Dec 2023 17:40:14 +0100 Subject: [PATCH 30/30] sched/fair: Fix tg->load when offlining a CPU When a CPU is taken offline, the contribution of its cfs_rqs to task_groups' load may remain and will negatively impact the calculation of the share of the online CPUs. To fix this bug, clear the contribution of an offlining CPU to task groups' load and skip its contribution while it is inactive. Here's the reproducer of the anomaly, by Imran Khan: "So far I have encountered only one rather lengthy way of reproducing this issue, which is as follows: 1. Take a KVM guest (booted with 4 CPUs and can be scaled up to 124 CPUs) and create 2 custom cgroups: /sys/fs/cgroup/cpu/test_group_1 and /sys/fs/cgroup/ cpu/test_group_2 2. Assign a CPU intensive workload to each of these cgroups and start the workload. For my tests I am using following app: int main(int argc, char *argv[]) { unsigned long count, i, val; if (argc != 2) { printf("usage: ./a.out \n"); return 0; } count = strtoul(argv[1], NULL, 10); printf("Generating %lu random numbers \n", count); for (i = 0; i < count; i++) { val = rand(); val = val % 2; //usleep(1); } printf("Generated %lu random numbers \n", count); return 0; } Also since the system is booted with 4 CPUs, in order to completely load the system I am also launching 4 instances of same test app under: /sys/fs/cgroup/cpu/ 3. We can see that both of the cgroups get similar CPU time: # systemd-cgtop --depth 1 Path Tasks %CPU Memory Input/s Output/s / 659 - 5.5G - - /system.slice - - 5.7G - - /test_group_1 4 - - - - /test_group_2 3 - - - - /user.slice 31 - 56.5M - - Path Tasks %CPU Memory Input/s Output/s / 659 394.6 5.5G - - /test_group_2 3 65.7 - - - /user.slice 29 55.1 48.0M - - /test_group_1 4 47.3 - - - /system.slice - 2.2 5.7G - - Path Tasks %CPU Memory Input/s Output/s / 659 394.8 5.5G - - /test_group_1 4 62.9 - - - /user.slice 28 44.9 54.2M - - /test_group_2 3 44.7 - - - /system.slice - 0.9 5.7G - - Path Tasks %CPU Memory Input/s Output/s / 659 394.4 5.5G - - /test_group_2 3 58.8 - - - /test_group_1 4 51.9 - - - /user.slice 30 39.3 59.6M - - /system.slice - 1.9 5.7G - - Path Tasks %CPU Memory Input/s Output/s / 659 394.7 5.5G - - /test_group_1 4 60.9 - - - /test_group_2 3 57.9 - - - /user.slice 28 43.5 36.9M - - /system.slice - 3.0 5.7G - - Path Tasks %CPU Memory Input/s Output/s / 659 395.0 5.5G - - /test_group_1 4 66.8 - - - /test_group_2 3 56.3 - - - /user.slice 29 43.1 51.8M - - /system.slice - 0.7 5.7G - - 4. Now move systemd-udevd to one of these test groups, say test_group_1, and perform scale up to 124 CPUs followed by scale down back to 4 CPUs from the host side. 5. Run the same workload i.e 4 instances of CPU hogger under /sys/fs/cgroup/cpu and one instance of CPU hogger each in /sys/fs/cgroup/cpu/test_group_1 and /sys/fs/cgroup/test_group_2. It can be seen that test_group_1 (the one where systemd-udevd was moved) is getting much less CPU time than the test_group_2, even though at this point of time both of these groups have only CPU hogger running: # systemd-cgtop --depth 1 Path Tasks %CPU Memory Input/s Output/s / 1219 - 5.4G - - /system.slice - - 5.6G - - /test_group_1 4 - - - - /test_group_2 3 - - - - /user.slice 26 - 91.3M - - Path Tasks %CPU Memory Input/s Output/s / 1221 394.3 5.4G - - /test_group_2 3 82.7 - - - /test_group_1 4 14.3 - - - /system.slice - 0.8 5.6G - - /user.slice 26 0.4 91.2M - - Path Tasks %CPU Memory Input/s Output/s / 1221 394.6 5.4G - - /test_group_2 3 67.4 - - - /system.slice - 24.6 5.6G - - /test_group_1 4 12.5 - - - /user.slice 26 0.4 91.2M - - Path Tasks %CPU Memory Input/s Output/s / 1221 395.2 5.4G - - /test_group_2 3 60.9 - - - /system.slice - 27.9 5.6G - - /test_group_1 4 12.2 - - - /user.slice 26 0.4 91.2M - - Path Tasks %CPU Memory Input/s Output/s / 1221 395.2 5.4G - - /test_group_2 3 69.4 - - - /test_group_1 4 13.9 - - - /user.slice 28 1.6 92.0M - - /system.slice - 1.0 5.6G - - Path Tasks %CPU Memory Input/s Output/s / 1221 395.6 5.4G - - /test_group_2 3 59.3 - - - /test_group_1 4 14.1 - - - /user.slice 28 1.3 92.2M - - /system.slice - 0.7 5.6G - - Path Tasks %CPU Memory Input/s Output/s / 1221 395.5 5.4G - - /test_group_2 3 67.2 - - - /test_group_1 4 11.5 - - - /user.slice 28 1.3 92.5M - - /system.slice - 0.6 5.6G - - Path Tasks %CPU Memory Input/s Output/s / 1221 395.1 5.4G - - /test_group_2 3 76.8 - - - /test_group_1 4 12.9 - - - /user.slice 28 1.3 92.8M - - /system.slice - 1.2 5.6G - - From sched_debug data it can be seen that in bad case the load.weight of per-CPU sched entities corresponding to test_group_1 has reduced significantly and also load_avg of test_group_1 remains much higher than that of test_group_2, even though systemd-udevd stopped running long time back and at this point of time both cgroups just have the CPU hogger app as running entity." [ mingo: Added details from the original discussion, plus minor edits to the patch. ] Reported-by: Imran Khan Tested-by: Imran Khan Tested-by: Aaron Lu Signed-off-by: Vincent Guittot Signed-off-by: Ingo Molnar Reviewed-by: Imran Khan Cc: Peter Zijlstra Cc: Borislav Petkov Link: https://lore.kernel.org/r/20231223111545.62135-1-vincent.guittot@linaro.org --- kernel/sched/fair.c | 52 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 52 insertions(+) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index d7a3c63a2171..43c1216898cb 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4096,6 +4096,10 @@ static inline void update_tg_load_avg(struct cfs_rq *cfs_rq) if (cfs_rq->tg == &root_task_group) return; + /* rq has been offline and doesn't contribute to the share anymore: */ + if (!cpu_active(cpu_of(rq_of(cfs_rq)))) + return; + /* * For migration heavy workloads, access to tg->load_avg can be * unbound. Limit the update rate to at most once per ms. @@ -4112,6 +4116,49 @@ static inline void update_tg_load_avg(struct cfs_rq *cfs_rq) } } +static inline void clear_tg_load_avg(struct cfs_rq *cfs_rq) +{ + long delta; + u64 now; + + /* + * No need to update load_avg for root_task_group, as it is not used. + */ + if (cfs_rq->tg == &root_task_group) + return; + + now = sched_clock_cpu(cpu_of(rq_of(cfs_rq))); + delta = 0 - cfs_rq->tg_load_avg_contrib; + atomic_long_add(delta, &cfs_rq->tg->load_avg); + cfs_rq->tg_load_avg_contrib = 0; + cfs_rq->last_update_tg_load_avg = now; +} + +/* CPU offline callback: */ +static void __maybe_unused clear_tg_offline_cfs_rqs(struct rq *rq) +{ + struct task_group *tg; + + lockdep_assert_rq_held(rq); + + /* + * The rq clock has already been updated in + * set_rq_offline(), so we should skip updating + * the rq clock again in unthrottle_cfs_rq(). + */ + rq_clock_start_loop_update(rq); + + rcu_read_lock(); + list_for_each_entry_rcu(tg, &task_groups, list) { + struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)]; + + clear_tg_load_avg(cfs_rq); + } + rcu_read_unlock(); + + rq_clock_stop_loop_update(rq); +} + /* * Called within set_task_rq() right before setting a task's CPU. The * caller only guarantees p->pi_lock is held; no other assumptions, @@ -4408,6 +4455,8 @@ static inline bool skip_blocked_update(struct sched_entity *se) static inline void update_tg_load_avg(struct cfs_rq *cfs_rq) {} +static inline void clear_tg_offline_cfs_rqs(struct rq *rq) {} + static inline int propagate_entity_load_avg(struct sched_entity *se) { return 0; @@ -12413,6 +12462,9 @@ static void rq_offline_fair(struct rq *rq) /* Ensure any throttled groups are reachable by pick_next_task */ unthrottle_offline_cfs_rqs(rq); + + /* Ensure that we remove rq contribution to group share: */ + clear_tg_offline_cfs_rqs(rq); } #endif /* CONFIG_SMP */