diff options
Diffstat (limited to 'kernel/sched/ext.c')
| -rw-r--r-- | kernel/sched/ext.c | 508 |
1 files changed, 278 insertions, 230 deletions
diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c index c09e3dc38c34..5900b06fd036 100644 --- a/kernel/sched/ext.c +++ b/kernel/sched/ext.c @@ -18,6 +18,12 @@ enum scx_consts { SCX_EXIT_DUMP_DFL_LEN = 32768, SCX_CPUPERF_ONE = SCHED_CAPACITY_SCALE, + + /* + * Iterating all tasks may take a while. Periodically drop + * scx_tasks_lock to avoid causing e.g. CSD and RCU stalls. + */ + SCX_OPS_TASK_ITER_BATCH = 32, }; enum scx_exit_kind { @@ -624,6 +630,10 @@ struct sched_ext_ops { /** * exit - Clean up after the BPF scheduler * @info: Exit info + * + * ops.exit() is also called on ops.init() failure, which is a bit + * unusual. This is to allow rich reporting through @info on how + * ops.init() failed. */ void (*exit)(struct scx_exit_info *info); @@ -691,6 +701,7 @@ enum scx_enq_flags { /* expose select ENQUEUE_* flags as enums */ SCX_ENQ_WAKEUP = ENQUEUE_WAKEUP, SCX_ENQ_HEAD = ENQUEUE_HEAD, + SCX_ENQ_CPU_SELECTED = ENQUEUE_RQ_SELECTED, /* high 32bits are SCX specific */ @@ -778,7 +789,6 @@ enum scx_tg_flags { }; enum scx_ops_enable_state { - SCX_OPS_PREPPING, SCX_OPS_ENABLING, SCX_OPS_ENABLED, SCX_OPS_DISABLING, @@ -786,7 +796,6 @@ enum scx_ops_enable_state { }; static const char *scx_ops_enable_state_str[] = { - [SCX_OPS_PREPPING] = "prepping", [SCX_OPS_ENABLING] = "enabling", [SCX_OPS_ENABLED] = "enabled", [SCX_OPS_DISABLING] = "disabling", @@ -854,6 +863,7 @@ DEFINE_STATIC_KEY_FALSE(__scx_ops_enabled); DEFINE_STATIC_PERCPU_RWSEM(scx_fork_rwsem); static atomic_t scx_ops_enable_state_var = ATOMIC_INIT(SCX_OPS_DISABLED); static atomic_t scx_ops_bypass_depth = ATOMIC_INIT(0); +static bool scx_ops_init_task_enabled; static bool scx_switching_all; DEFINE_STATIC_KEY_FALSE(__scx_switched_all); @@ -925,8 +935,15 @@ static unsigned long __percpu *scx_kick_cpus_pnt_seqs; */ static DEFINE_PER_CPU(struct task_struct *, direct_dispatch_task); -/* dispatch queues */ -static struct scx_dispatch_q __cacheline_aligned_in_smp scx_dsq_global; +/* + * Dispatch queues. + * + * The global DSQ (%SCX_DSQ_GLOBAL) is split per-node for scalability. This is + * to avoid live-locking in bypass mode where all tasks are dispatched to + * %SCX_DSQ_GLOBAL and all CPUs consume from it. If per-node split isn't + * sufficient, it can be further split. + */ +static struct scx_dispatch_q **global_dsqs; static const struct rhashtable_params dsq_hash_params = { .key_len = 8, @@ -1029,6 +1046,16 @@ static bool u32_before(u32 a, u32 b) return (s32)(a - b) < 0; } +static struct scx_dispatch_q *find_global_dsq(struct task_struct *p) +{ + return global_dsqs[cpu_to_node(task_cpu(p))]; +} + +static struct scx_dispatch_q *find_user_dsq(u64 dsq_id) +{ + return rhashtable_lookup_fast(&dsq_hash, &dsq_id, dsq_hash_params); +} + /* * scx_kf_mask enforcement. Some kfuncs can only be called from specific SCX * ops. When invoking SCX ops, SCX_CALL_OP[_RET]() should be used to indicate @@ -1252,86 +1279,104 @@ struct scx_task_iter { struct task_struct *locked; struct rq *rq; struct rq_flags rf; + u32 cnt; }; /** - * scx_task_iter_init - Initialize a task iterator + * scx_task_iter_start - Lock scx_tasks_lock and start a task iteration * @iter: iterator to init * - * Initialize @iter. Must be called with scx_tasks_lock held. Once initialized, - * @iter must eventually be exited with scx_task_iter_exit(). + * Initialize @iter and return with scx_tasks_lock held. Once initialized, @iter + * must eventually be stopped with scx_task_iter_stop(). * - * scx_tasks_lock may be released between this and the first next() call or - * between any two next() calls. If scx_tasks_lock is released between two - * next() calls, the caller is responsible for ensuring that the task being - * iterated remains accessible either through RCU read lock or obtaining a - * reference count. + * scx_tasks_lock and the rq lock may be released using scx_task_iter_unlock() + * between this and the first next() call or between any two next() calls. If + * the locks are released between two next() calls, the caller is responsible + * for ensuring that the task being iterated remains accessible either through + * RCU read lock or obtaining a reference count. * * All tasks which existed when the iteration started are guaranteed to be * visited as long as they still exist. */ -static void scx_task_iter_init(struct scx_task_iter *iter) +static void scx_task_iter_start(struct scx_task_iter *iter) { - lockdep_assert_held(&scx_tasks_lock); - BUILD_BUG_ON(__SCX_DSQ_ITER_ALL_FLAGS & ((1U << __SCX_DSQ_LNODE_PRIV_SHIFT) - 1)); + spin_lock_irq(&scx_tasks_lock); + iter->cursor = (struct sched_ext_entity){ .flags = SCX_TASK_CURSOR }; list_add(&iter->cursor.tasks_node, &scx_tasks); iter->locked = NULL; + iter->cnt = 0; +} + +static void __scx_task_iter_rq_unlock(struct scx_task_iter *iter) +{ + if (iter->locked) { + task_rq_unlock(iter->rq, iter->locked, &iter->rf); + iter->locked = NULL; + } } /** - * scx_task_iter_rq_unlock - Unlock rq locked by a task iterator - * @iter: iterator to unlock rq for + * scx_task_iter_unlock - Unlock rq and scx_tasks_lock held by a task iterator + * @iter: iterator to unlock * * If @iter is in the middle of a locked iteration, it may be locking the rq of - * the task currently being visited. Unlock the rq if so. This function can be - * safely called anytime during an iteration. + * the task currently being visited in addition to scx_tasks_lock. Unlock both. + * This function can be safely called anytime during an iteration. + */ +static void scx_task_iter_unlock(struct scx_task_iter *iter) +{ + __scx_task_iter_rq_unlock(iter); + spin_unlock_irq(&scx_tasks_lock); +} + +/** + * scx_task_iter_relock - Lock scx_tasks_lock released by scx_task_iter_unlock() + * @iter: iterator to re-lock * - * Returns %true if the rq @iter was locking is unlocked. %false if @iter was - * not locking an rq. + * Re-lock scx_tasks_lock unlocked by scx_task_iter_unlock(). Note that it + * doesn't re-lock the rq lock. Must be called before other iterator operations. */ -static bool scx_task_iter_rq_unlock(struct scx_task_iter *iter) +static void scx_task_iter_relock(struct scx_task_iter *iter) { - if (iter->locked) { - task_rq_unlock(iter->rq, iter->locked, &iter->rf); - iter->locked = NULL; - return true; - } else { - return false; - } + spin_lock_irq(&scx_tasks_lock); } /** - * scx_task_iter_exit - Exit a task iterator + * scx_task_iter_stop - Stop a task iteration and unlock scx_tasks_lock * @iter: iterator to exit * - * Exit a previously initialized @iter. Must be called with scx_tasks_lock held. - * If the iterator holds a task's rq lock, that rq lock is released. See - * scx_task_iter_init() for details. + * Exit a previously initialized @iter. Must be called with scx_tasks_lock held + * which is released on return. If the iterator holds a task's rq lock, that rq + * lock is also released. See scx_task_iter_start() for details. */ -static void scx_task_iter_exit(struct scx_task_iter *iter) +static void scx_task_iter_stop(struct scx_task_iter *iter) { - lockdep_assert_held(&scx_tasks_lock); - - scx_task_iter_rq_unlock(iter); list_del_init(&iter->cursor.tasks_node); + scx_task_iter_unlock(iter); } /** * scx_task_iter_next - Next task * @iter: iterator to walk * - * Visit the next task. See scx_task_iter_init() for details. + * Visit the next task. See scx_task_iter_start() for details. Locks are dropped + * and re-acquired every %SCX_OPS_TASK_ITER_BATCH iterations to avoid causing + * stalls by holding scx_tasks_lock for too long. */ static struct task_struct *scx_task_iter_next(struct scx_task_iter *iter) { struct list_head *cursor = &iter->cursor.tasks_node; struct sched_ext_entity *pos; - lockdep_assert_held(&scx_tasks_lock); + if (!(++iter->cnt % SCX_OPS_TASK_ITER_BATCH)) { + scx_task_iter_unlock(iter); + cond_resched(); + scx_task_iter_relock(iter); + } list_for_each_entry(pos, cursor, tasks_node) { if (&pos->tasks_node == &scx_tasks) @@ -1352,14 +1397,14 @@ static struct task_struct *scx_task_iter_next(struct scx_task_iter *iter) * @include_dead: Whether we should include dead tasks in the iteration * * Visit the non-idle task with its rq lock held. Allows callers to specify - * whether they would like to filter out dead tasks. See scx_task_iter_init() + * whether they would like to filter out dead tasks. See scx_task_iter_start() * for details. */ static struct task_struct *scx_task_iter_next_locked(struct scx_task_iter *iter) { struct task_struct *p; - scx_task_iter_rq_unlock(iter); + __scx_task_iter_rq_unlock(iter); while ((p = scx_task_iter_next(iter))) { /* @@ -1637,7 +1682,7 @@ static void dispatch_enqueue(struct scx_dispatch_q *dsq, struct task_struct *p, scx_ops_error("attempting to dispatch to a destroyed dsq"); /* fall back to the global dsq */ raw_spin_unlock(&dsq->lock); - dsq = &scx_dsq_global; + dsq = find_global_dsq(p); raw_spin_lock(&dsq->lock); } } @@ -1803,21 +1848,6 @@ static void dispatch_dequeue(struct rq *rq, struct task_struct *p) raw_spin_unlock(&dsq->lock); } -static struct scx_dispatch_q *find_user_dsq(u64 dsq_id) -{ - return rhashtable_lookup_fast(&dsq_hash, &dsq_id, dsq_hash_params); -} - -static struct scx_dispatch_q *find_non_local_dsq(u64 dsq_id) -{ - lockdep_assert(rcu_read_lock_any_held()); - - if (dsq_id == SCX_DSQ_GLOBAL) - return &scx_dsq_global; - else - return find_user_dsq(dsq_id); -} - static struct scx_dispatch_q *find_dsq_for_dispatch(struct rq *rq, u64 dsq_id, struct task_struct *p) { @@ -1830,16 +1860,20 @@ static struct scx_dispatch_q *find_dsq_for_dispatch(struct rq *rq, u64 dsq_id, s32 cpu = dsq_id & SCX_DSQ_LOCAL_CPU_MASK; if (!ops_cpu_valid(cpu, "in SCX_DSQ_LOCAL_ON dispatch verdict")) - return &scx_dsq_global; + return find_global_dsq(p); return &cpu_rq(cpu)->scx.local_dsq; } - dsq = find_non_local_dsq(dsq_id); + if (dsq_id == SCX_DSQ_GLOBAL) + dsq = find_global_dsq(p); + else + dsq = find_user_dsq(dsq_id); + if (unlikely(!dsq)) { scx_ops_error("non-existent DSQ 0x%llx for %s[%d]", dsq_id, p->comm, p->pid); - return &scx_dsq_global; + return find_global_dsq(p); } return dsq; @@ -2011,7 +2045,7 @@ local_norefill: global: touch_core_sched(rq, p); /* see the comment in local: */ p->scx.slice = SCX_SLICE_DFL; - dispatch_enqueue(&scx_dsq_global, p, enq_flags); + dispatch_enqueue(find_global_dsq(p), p, enq_flags); } static bool task_runnable(const struct task_struct *p) @@ -2357,6 +2391,7 @@ static bool consume_remote_task(struct rq *this_rq, struct task_struct *p, } } #else /* CONFIG_SMP */ +static inline void move_remote_task_to_local_dsq(struct task_struct *p, u64 enq_flags, struct rq *src_rq, struct rq *dst_rq) { WARN_ON_ONCE(1); } static inline bool task_can_run_on_remote_rq(struct task_struct *p, struct rq *rq, bool trigger_error) { return false; } static inline bool consume_remote_task(struct rq *this_rq, struct task_struct *p, struct scx_dispatch_q *dsq, struct rq *task_rq) { return false; } #endif /* CONFIG_SMP */ @@ -2396,6 +2431,13 @@ retry: return false; } +static bool consume_global_dsq(struct rq *rq) +{ + int node = cpu_to_node(cpu_of(rq)); + + return consume_dispatch_q(rq, global_dsqs[node]); +} + /** * dispatch_to_local_dsq - Dispatch a task to a local dsq * @rq: current rq which is locked @@ -2429,7 +2471,8 @@ static void dispatch_to_local_dsq(struct rq *rq, struct scx_dispatch_q *dst_dsq, #ifdef CONFIG_SMP if (unlikely(!task_can_run_on_remote_rq(p, dst_rq, true))) { - dispatch_enqueue(&scx_dsq_global, p, enq_flags | SCX_ENQ_CLEAR_OPSS); + dispatch_enqueue(find_global_dsq(p), p, + enq_flags | SCX_ENQ_CLEAR_OPSS); return; } @@ -2629,7 +2672,7 @@ static int balance_one(struct rq *rq, struct task_struct *prev) if (rq->scx.local_dsq.nr) goto has_tasks; - if (consume_dispatch_q(rq, &scx_dsq_global)) + if (consume_global_dsq(rq)) goto has_tasks; if (!SCX_HAS_OP(dispatch) || scx_rq_bypassing(rq) || !scx_rq_online(rq)) @@ -2654,7 +2697,7 @@ static int balance_one(struct rq *rq, struct task_struct *prev) if (rq->scx.local_dsq.nr) goto has_tasks; - if (consume_dispatch_q(rq, &scx_dsq_global)) + if (consume_global_dsq(rq)) goto has_tasks; /* @@ -2937,8 +2980,8 @@ static struct task_struct *pick_task_scx(struct rq *rq) if (unlikely(!p->scx.slice)) { if (!scx_rq_bypassing(rq) && !scx_warned_zero_slice) { - printk_deferred(KERN_WARNING "sched_ext: %s[%d] has zero slice in pick_next_task_scx()\n", - p->comm, p->pid); + printk_deferred(KERN_WARNING "sched_ext: %s[%d] has zero slice in %s()\n", + p->comm, p->pid, __func__); scx_warned_zero_slice = true; } p->scx.slice = SCX_SLICE_DFL; @@ -3043,11 +3086,6 @@ static s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, *found = false; - if (!static_branch_likely(&scx_builtin_idle_enabled)) { - scx_ops_error("built-in idle tracking is disabled"); - return prev_cpu; - } - /* * If WAKE_SYNC, the waker's local DSQ is empty, and the system is * under utilized, wake up @p to the local DSQ of the waker. Checking @@ -3058,22 +3096,13 @@ static s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, * there is an idle core elsewhere on the system. */ cpu = smp_processor_id(); - if ((wake_flags & SCX_WAKE_SYNC) && p->nr_cpus_allowed > 1 && + if ((wake_flags & SCX_WAKE_SYNC) && !cpumask_empty(idle_masks.cpu) && !(current->flags & PF_EXITING) && cpu_rq(cpu)->scx.local_dsq.nr == 0) { if (cpumask_test_cpu(cpu, p->cpus_ptr)) goto cpu_found; } - if (p->nr_cpus_allowed == 1) { - if (test_and_clear_cpu_idle(prev_cpu)) { - cpu = prev_cpu; - goto cpu_found; - } else { - return prev_cpu; - } - } - /* * If CPU has SMT, any wholly idle CPU is likely a better pick than * partially idle @prev_cpu. @@ -3121,7 +3150,7 @@ static int select_task_rq_scx(struct task_struct *p, int prev_cpu, int wake_flag if (unlikely(wake_flags & WF_EXEC)) return prev_cpu; - if (SCX_HAS_OP(select_cpu)) { + if (SCX_HAS_OP(select_cpu) && !scx_rq_bypassing(task_rq(p))) { s32 cpu; struct task_struct **ddsp_taskp; @@ -3186,7 +3215,7 @@ void __scx_update_idle(struct rq *rq, bool idle) { int cpu = cpu_of(rq); - if (SCX_HAS_OP(update_idle)) { + if (SCX_HAS_OP(update_idle) && !scx_rq_bypassing(rq)) { SCX_CALL_OP(SCX_KF_REST, update_idle, cpu_of(rq), idle); if (!static_branch_unlikely(&scx_builtin_idle_enabled)) return; @@ -3550,7 +3579,7 @@ int scx_fork(struct task_struct *p) { percpu_rwsem_assert_held(&scx_fork_rwsem); - if (scx_enabled()) + if (scx_ops_init_task_enabled) return scx_ops_init_task(p, task_group(p), true); else return 0; @@ -3558,7 +3587,7 @@ int scx_fork(struct task_struct *p) void scx_post_fork(struct task_struct *p) { - if (scx_enabled()) { + if (scx_ops_init_task_enabled) { scx_set_task_state(p, SCX_TASK_READY); /* @@ -3690,6 +3719,7 @@ bool scx_can_stop_tick(struct rq *rq) #ifdef CONFIG_EXT_GROUP_SCHED DEFINE_STATIC_PERCPU_RWSEM(scx_cgroup_rwsem); +static bool scx_cgroup_enabled; static bool cgroup_warned_missing_weight; static bool cgroup_warned_missing_idle; @@ -3709,8 +3739,7 @@ static void scx_cgroup_warn_missing_weight(struct task_group *tg) static void scx_cgroup_warn_missing_idle(struct task_group *tg) { - if (scx_ops_enable_state() == SCX_OPS_DISABLED || - cgroup_warned_missing_idle) + if (!scx_cgroup_enabled || cgroup_warned_missing_idle) return; if (!tg->idle) @@ -3731,15 +3760,18 @@ int scx_tg_online(struct task_group *tg) scx_cgroup_warn_missing_weight(tg); - if (SCX_HAS_OP(cgroup_init)) { - struct scx_cgroup_init_args args = { .weight = tg->scx_weight }; + if (scx_cgroup_enabled) { + if (SCX_HAS_OP(cgroup_init)) { + struct scx_cgroup_init_args args = + { .weight = tg->scx_weight }; - ret = SCX_CALL_OP_RET(SCX_KF_UNLOCKED, cgroup_init, - tg->css.cgroup, &args); - if (!ret) + ret = SCX_CALL_OP_RET(SCX_KF_UNLOCKED, cgroup_init, + tg->css.cgroup, &args); + if (ret) + ret = ops_sanitize_err("cgroup_init", ret); + } + if (ret == 0) tg->scx_flags |= SCX_TG_ONLINE | SCX_TG_INITED; - else - ret = ops_sanitize_err("cgroup_init", ret); } else { tg->scx_flags |= SCX_TG_ONLINE; } @@ -3770,7 +3802,7 @@ int scx_cgroup_can_attach(struct cgroup_taskset *tset) /* released in scx_finish/cancel_attach() */ percpu_down_read(&scx_cgroup_rwsem); - if (!scx_enabled()) + if (!scx_cgroup_enabled) return 0; cgroup_taskset_for_each(p, css, tset) { @@ -3813,7 +3845,7 @@ err: void scx_move_task(struct task_struct *p) { - if (!scx_enabled()) + if (!scx_cgroup_enabled) return; /* @@ -3849,7 +3881,7 @@ void scx_cgroup_cancel_attach(struct cgroup_taskset *tset) struct cgroup_subsys_state *css; struct task_struct *p; - if (!scx_enabled()) + if (!scx_cgroup_enabled) goto out_unlock; cgroup_taskset_for_each(p, css, tset) { @@ -3866,7 +3898,7 @@ void scx_group_set_weight(struct task_group *tg, unsigned long weight) { percpu_down_read(&scx_cgroup_rwsem); - if (tg->scx_weight != weight) { + if (scx_cgroup_enabled && tg->scx_weight != weight) { if (SCX_HAS_OP(cgroup_set_weight)) SCX_CALL_OP(SCX_KF_UNLOCKED, cgroup_set_weight, tg_cgrp(tg), weight); @@ -4038,6 +4070,8 @@ static void scx_cgroup_exit(void) percpu_rwsem_assert_held(&scx_cgroup_rwsem); + scx_cgroup_enabled = false; + /* * scx_tg_on/offline() are excluded through scx_cgroup_rwsem. If we walk * cgroups and exit all the inited ones, all online cgroups are exited. @@ -4104,6 +4138,7 @@ static int scx_cgroup_init(void) css->cgroup, &args); if (ret) { css_put(css); + scx_ops_error("ops.cgroup_init() failed (%d)", ret); return ret; } tg->scx_flags |= SCX_TG_INITED; @@ -4113,6 +4148,9 @@ static int scx_cgroup_init(void) } rcu_read_unlock(); + WARN_ON_ONCE(scx_cgroup_enabled); + scx_cgroup_enabled = true; + return 0; } @@ -4240,21 +4278,23 @@ bool task_should_scx(struct task_struct *p) * the DISABLING state and then cycling the queued tasks through dequeue/enqueue * to force global FIFO scheduling. * - * a. ops.enqueue() is ignored and tasks are queued in simple global FIFO order. - * %SCX_OPS_ENQ_LAST is also ignored. + * - ops.select_cpu() is ignored and the default select_cpu() is used. * - * b. ops.dispatch() is ignored. + * - ops.enqueue() is ignored and tasks are queued in simple global FIFO order. + * %SCX_OPS_ENQ_LAST is also ignored. * - * c. balance_scx() does not set %SCX_RQ_BAL_KEEP on non-zero slice as slice - * can't be trusted. Whenever a tick triggers, the running task is rotated to - * the tail of the queue with core_sched_at touched. + * - ops.dispatch() is ignored. * - * d. pick_next_task() suppresses zero slice warning. + * - balance_scx() does not set %SCX_RQ_BAL_KEEP on non-zero slice as slice + * can't be trusted. Whenever a tick triggers, the running task is rotated to + * the tail of the queue with core_sched_at touched. * - * e. scx_bpf_kick_cpu() is disabled to avoid irq_work malfunction during PM - * operations. + * - pick_next_task() suppresses zero slice warning. * - * f. scx_prio_less() reverts to the default core_sched_at order. + * - scx_bpf_kick_cpu() is disabled to avoid irq_work malfunction during PM + * operations. + * + * - scx_prio_less() reverts to the default core_sched_at order. */ static void scx_ops_bypass(bool bypass) { @@ -4324,7 +4364,7 @@ static void scx_ops_bypass(bool bypass) rq_unlock_irqrestore(rq, &rf); - /* kick to restore ticks */ + /* resched to restore ticks and idle state */ resched_cpu(cpu); } } @@ -4431,27 +4471,29 @@ static void scx_ops_disable_workfn(struct kthread_work *work) WRITE_ONCE(scx_switching_all, false); /* - * Avoid racing against fork and cgroup changes. See scx_ops_enable() - * for explanation on the locking order. + * Shut down cgroup support before tasks so that the cgroup attach path + * doesn't race against scx_ops_exit_task(). */ - percpu_down_write(&scx_fork_rwsem); - cpus_read_lock(); scx_cgroup_lock(); + scx_cgroup_exit(); + scx_cgroup_unlock(); - spin_lock_irq(&scx_tasks_lock); - scx_task_iter_init(&sti); /* * The BPF scheduler is going away. All tasks including %TASK_DEAD ones * must be switched out and exited synchronously. */ + percpu_down_write(&scx_fork_rwsem); + + scx_ops_init_task_enabled = false; + + scx_task_iter_start(&sti); while ((p = scx_task_iter_next_locked(&sti))) { const struct sched_class *old_class = p->sched_class; struct sched_enq_and_set_ctx ctx; sched_deq_and_put_task(p, DEQUEUE_SAVE | DEQUEUE_MOVE, &ctx); - p->scx.slice = min_t(u64, p->scx.slice, SCX_SLICE_DFL); - __setscheduler_prio(p, p->prio); + p->sched_class = __setscheduler_class(p, p->prio); check_class_changing(task_rq(p), p, old_class); sched_enq_and_set_task(&ctx); @@ -4459,25 +4501,19 @@ static void scx_ops_disable_workfn(struct kthread_work *work) check_class_changed(task_rq(p), p, old_class, p->prio); scx_ops_exit_task(p); } - scx_task_iter_exit(&sti); - spin_unlock_irq(&scx_tasks_lock); + scx_task_iter_stop(&sti); + percpu_up_write(&scx_fork_rwsem); /* no task is on scx, turn off all the switches and flush in-progress calls */ - static_branch_disable_cpuslocked(&__scx_ops_enabled); + static_branch_disable(&__scx_ops_enabled); for (i = SCX_OPI_BEGIN; i < SCX_OPI_END; i++) - static_branch_disable_cpuslocked(&scx_has_op[i]); - static_branch_disable_cpuslocked(&scx_ops_enq_last); - static_branch_disable_cpuslocked(&scx_ops_enq_exiting); - static_branch_disable_cpuslocked(&scx_ops_cpu_preempt); - static_branch_disable_cpuslocked(&scx_builtin_idle_enabled); + static_branch_disable(&scx_has_op[i]); + static_branch_disable(&scx_ops_enq_last); + static_branch_disable(&scx_ops_enq_exiting); + static_branch_disable(&scx_ops_cpu_preempt); + static_branch_disable(&scx_builtin_idle_enabled); synchronize_rcu(); - scx_cgroup_exit(); - - scx_cgroup_unlock(); - cpus_read_unlock(); - percpu_up_write(&scx_fork_rwsem); - if (ei->kind >= SCX_EXIT_ERROR) { pr_err("sched_ext: BPF scheduler \"%s\" disabled (%s)\n", scx_ops.name, ei->reason); @@ -4929,7 +4965,7 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) struct scx_task_iter sti; struct task_struct *p; unsigned long timeout; - int i, cpu, ret; + int i, cpu, node, ret; if (!cpumask_equal(housekeeping_cpumask(HK_TYPE_DOMAIN), cpu_possible_mask)) { @@ -4948,6 +4984,34 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) } } + if (!global_dsqs) { + struct scx_dispatch_q **dsqs; + + dsqs = kcalloc(nr_node_ids, sizeof(dsqs[0]), GFP_KERNEL); + if (!dsqs) { + ret = -ENOMEM; + goto err_unlock; + } + + for_each_node_state(node, N_POSSIBLE) { + struct scx_dispatch_q *dsq; + + dsq = kzalloc_node(sizeof(*dsq), GFP_KERNEL, node); + if (!dsq) { + for_each_node_state(node, N_POSSIBLE) + kfree(dsqs[node]); + kfree(dsqs); + ret = -ENOMEM; + goto err_unlock; + } + + init_dsq(dsq, SCX_DSQ_GLOBAL); + dsqs[node] = dsq; + } + + global_dsqs = dsqs; + } + if (scx_ops_enable_state() != SCX_OPS_DISABLED) { ret = -EBUSY; goto err_unlock; @@ -4971,12 +5035,12 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) } /* - * Set scx_ops, transition to PREPPING and clear exit info to arm the + * Set scx_ops, transition to ENABLING and clear exit info to arm the * disable path. Failure triggers full disabling from here on. */ scx_ops = *ops; - WARN_ON_ONCE(scx_ops_set_enable_state(SCX_OPS_PREPPING) != + WARN_ON_ONCE(scx_ops_set_enable_state(SCX_OPS_ENABLING) != SCX_OPS_DISABLED); atomic_set(&scx_exit_kind, SCX_EXIT_NONE); @@ -4997,7 +5061,9 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) ret = SCX_CALL_OP_RET(SCX_KF_UNLOCKED, init); if (ret) { ret = ops_sanitize_err("init", ret); - goto err_disable_unlock_cpus; + cpus_read_unlock(); + scx_ops_error("ops.init() failed (%d)", ret); + goto err_disable; } } @@ -5005,6 +5071,7 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) if (((void (**)(void))ops)[i]) static_branch_enable_cpuslocked(&scx_has_op[i]); + check_hotplug_seq(ops); cpus_read_unlock(); ret = validate_ops(ops); @@ -5032,57 +5099,40 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) scx_watchdog_timeout / 2); /* - * Lock out forks, cgroup on/offlining and moves before opening the - * floodgate so that they don't wander into the operations prematurely. - * - * We don't need to keep the CPUs stable but static_branch_*() requires - * cpus_read_lock() and scx_cgroup_rwsem must nest inside - * cpu_hotplug_lock because of the following dependency chain: - * - * cpu_hotplug_lock --> cgroup_threadgroup_rwsem --> scx_cgroup_rwsem - * - * So, we need to do cpus_read_lock() before scx_cgroup_lock() and use - * static_branch_*_cpuslocked(). - * - * Note that cpu_hotplug_lock must nest inside scx_fork_rwsem due to the - * following dependency chain: - * - * scx_fork_rwsem --> pernet_ops_rwsem --> cpu_hotplug_lock + * Once __scx_ops_enabled is set, %current can be switched to SCX + * anytime. This can lead to stalls as some BPF schedulers (e.g. + * userspace scheduling) may not function correctly before all tasks are + * switched. Init in bypass mode to guarantee forward progress. */ - percpu_down_write(&scx_fork_rwsem); - cpus_read_lock(); - scx_cgroup_lock(); - - check_hotplug_seq(ops); + scx_ops_bypass(true); for (i = SCX_OPI_NORMAL_BEGIN; i < SCX_OPI_NORMAL_END; i++) if (((void (**)(void))ops)[i]) - static_branch_enable_cpuslocked(&scx_has_op[i]); + static_branch_enable(&scx_has_op[i]); if (ops->flags & SCX_OPS_ENQ_LAST) - static_branch_enable_cpuslocked(&scx_ops_enq_last); + static_branch_enable(&scx_ops_enq_last); if (ops->flags & SCX_OPS_ENQ_EXITING) - static_branch_enable_cpuslocked(&scx_ops_enq_exiting); + static_branch_enable(&scx_ops_enq_exiting); if (scx_ops.cpu_acquire || scx_ops.cpu_release) - static_branch_enable_cpuslocked(&scx_ops_cpu_preempt); + static_branch_enable(&scx_ops_cpu_preempt); if (!ops->update_idle || (ops->flags & SCX_OPS_KEEP_BUILTIN_IDLE)) { reset_idle_masks(); - static_branch_enable_cpuslocked(&scx_builtin_idle_enabled); + static_branch_enable(&scx_builtin_idle_enabled); } else { - static_branch_disable_cpuslocked(&scx_builtin_idle_enabled); + static_branch_disable(&scx_builtin_idle_enabled); } /* - * All cgroups should be initialized before letting in tasks. cgroup - * on/offlining and task migrations are already locked out. + * Lock out forks, cgroup on/offlining and moves before opening the + * floodgate so that they don't wander into the operations prematurely. */ - ret = scx_cgroup_init(); - if (ret) - goto err_disable_unlock_all; + percpu_down_write(&scx_fork_rwsem); - static_branch_enable_cpuslocked(&__scx_ops_enabled); + WARN_ON_ONCE(scx_ops_init_task_enabled); + scx_ops_init_task_enabled = true; /* * Enable ops for every task. Fork is excluded by scx_fork_rwsem @@ -5090,10 +5140,19 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) * leaving as sched_ext_free() can handle both prepped and enabled * tasks. Prep all tasks first and then enable them with preemption * disabled. + * + * All cgroups should be initialized before scx_ops_init_task() so that + * the BPF scheduler can reliably track each task's cgroup membership + * from scx_ops_init_task(). Lock out cgroup on/offlining and task + * migrations while tasks are being initialized so that + * scx_cgroup_can_attach() never sees uninitialized tasks. */ - spin_lock_irq(&scx_tasks_lock); + scx_cgroup_lock(); + ret = scx_cgroup_init(); + if (ret) + goto err_disable_unlock_all; - scx_task_iter_init(&sti); + scx_task_iter_start(&sti); while ((p = scx_task_iter_next_locked(&sti))) { /* * @p may already be dead, have lost all its usages counts and @@ -5103,84 +5162,62 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link) if (!tryget_task_struct(p)) continue; - scx_task_iter_rq_unlock(&sti); - spin_unlock_irq(&scx_tasks_lock); + scx_task_iter_unlock(&sti); ret = scx_ops_init_task(p, task_group(p), false); if (ret) { put_task_struct(p); - spin_lock_irq(&scx_tasks_lock); - scx_task_iter_exit(&sti); - spin_unlock_irq(&scx_tasks_lock); - pr_err("sched_ext: ops.init_task() failed (%d) for %s[%d] while loading\n", - ret, p->comm, p->pid); + scx_task_iter_relock(&sti); + scx_task_iter_stop(&sti); + scx_ops_error("ops.init_task() failed (%d) for %s[%d]", + ret, p->comm, p->pid); goto err_disable_unlock_all; } + scx_set_task_state(p, SCX_TASK_READY); + put_task_struct(p); - spin_lock_irq(&scx_tasks_lock); + scx_task_iter_relock(&sti); } - scx_task_iter_exit(&sti); + scx_task_iter_stop(&sti); + scx_cgroup_unlock(); + percpu_up_write(&scx_fork_rwsem); /* - * All tasks are prepped but are still ops-disabled. Ensure that - * %current can't be scheduled out and switch everyone. - * preempt_disable() is necessary because we can't guarantee that - * %current won't be starved if scheduled out while switching. + * All tasks are READY. It's safe to turn on scx_enabled() and switch + * all eligible tasks. */ - preempt_disable(); - - /* - * From here on, the disable path must assume that tasks have ops - * enabled and need to be recovered. - * - * Transition to ENABLING fails iff the BPF scheduler has already - * triggered scx_bpf_error(). Returning an error code here would lose - * the recorded error information. Exit indicating success so that the - * error is notified through ops.exit() with all the details. - */ - if (!scx_ops_tryset_enable_state(SCX_OPS_ENABLING, SCX_OPS_PREPPING)) { - preempt_enable(); - spin_unlock_irq(&scx_tasks_lock); - WARN_ON_ONCE(atomic_read(&scx_exit_kind) == SCX_EXIT_NONE); - ret = 0; - goto err_disable_unlock_all; - } + WRITE_ONCE(scx_switching_all, !(ops->flags & SCX_OPS_SWITCH_PARTIAL)); + static_branch_enable(&__scx_ops_enabled); /* - * We're fully committed and can't fail. The PREPPED -> ENABLED + * We're fully committed and can't fail. The task READY -> ENABLED * transitions here are synchronized against sched_ext_free() through * scx_tasks_lock. */ - WRITE_ONCE(scx_switching_all, !(ops->flags & SCX_OPS_SWITCH_PARTIAL)); - - scx_task_iter_init(&sti); + percpu_down_write(&scx_fork_rwsem); + scx_task_iter_start(&sti); while ((p = scx_task_iter_next_locked(&sti))) { const struct sched_class *old_class = p->sched_class; struct sched_enq_and_set_ctx ctx; sched_deq_and_put_task(p, DEQUEUE_SAVE | DEQUEUE_MOVE, &ctx); - scx_set_task_state(p, SCX_TASK_READY); - __setscheduler_prio(p, p->prio); + p->scx.slice = SCX_SLICE_DFL; + p->sched_class = __setscheduler_class(p, p->prio); check_class_changing(task_rq(p), p, old_class); sched_enq_and_set_task(&ctx); check_class_changed(task_rq(p), p, old_class, p->prio); } - scx_task_iter_exit(&sti); - - spin_unlock_irq(&scx_tasks_lock); - preempt_enable(); - scx_cgroup_unlock(); - cpus_read_unlock(); + scx_task_iter_stop(&sti); percpu_up_write(&scx_fork_rwsem); - /* see above ENABLING transition for the explanation on exiting with 0 */ + scx_ops_bypass(false); + if (!scx_ops_tryset_enable_state(SCX_OPS_ENABLED, SCX_OPS_ENABLING)) { WARN_ON_ONCE(atomic_read(&scx_exit_kind) == SCX_EXIT_NONE); - ret = 0; goto err_disable; } @@ -5212,14 +5249,21 @@ err_unlock: err_disable_unlock_all: scx_cgroup_unlock(); percpu_up_write(&scx_fork_rwsem); -err_disable_unlock_cpus: - cpus_read_unlock(); + scx_ops_bypass(false); err_disable: mutex_unlock(&scx_ops_enable_mutex); - /* must be fully disabled before returning */ - scx_ops_disable(SCX_EXIT_ERROR); + /* + * Returning an error code here would not pass all the error information + * to userspace. Record errno using scx_ops_error() for cases + * scx_ops_error() wasn't already invoked and exit indicating success so + * that the error is notified through ops.exit() with all the details. + * + * Flush scx_ops_disable_work to ensure that error is reported before + * init completion. + */ + scx_ops_error("scx_ops_enable() failed (%d)", ret); kthread_flush_work(&scx_ops_disable_work); - return ret; + return 0; } @@ -5782,7 +5826,6 @@ void __init init_sched_ext_class(void) SCX_TG_ONLINE); BUG_ON(rhashtable_init(&dsq_hash, &dsq_hash_params)); - init_dsq(&scx_dsq_global, SCX_DSQ_GLOBAL); #ifdef CONFIG_SMP BUG_ON(!alloc_cpumask_var(&idle_masks.cpu, GFP_KERNEL)); BUG_ON(!alloc_cpumask_var(&idle_masks.smt, GFP_KERNEL)); @@ -5840,16 +5883,21 @@ __bpf_kfunc_start_defs(); __bpf_kfunc s32 scx_bpf_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags, bool *is_idle) { - if (!scx_kf_allowed(SCX_KF_SELECT_CPU)) { - *is_idle = false; - return prev_cpu; + if (!static_branch_likely(&scx_builtin_idle_enabled)) { + scx_ops_error("built-in idle tracking is disabled"); + goto prev_cpu; } + + if (!scx_kf_allowed(SCX_KF_SELECT_CPU)) + goto prev_cpu; + #ifdef CONFIG_SMP return scx_select_cpu_dfl(p, prev_cpu, wake_flags, is_idle); -#else +#endif + +prev_cpu: *is_idle = false; return prev_cpu; -#endif } __bpf_kfunc_end_defs(); @@ -6058,7 +6106,7 @@ static bool scx_dispatch_from_dsq(struct bpf_iter_scx_dsq_kern *kit, if (dst_dsq->id == SCX_DSQ_LOCAL) { dst_rq = container_of(dst_dsq, struct rq, scx.local_dsq); if (!task_can_run_on_remote_rq(p, dst_rq, true)) { - dst_dsq = &scx_dsq_global; + dst_dsq = find_global_dsq(p); dst_rq = src_rq; } } else { @@ -6175,7 +6223,7 @@ __bpf_kfunc bool scx_bpf_consume(u64 dsq_id) flush_dispatch_buf(dspc->rq); - dsq = find_non_local_dsq(dsq_id); + dsq = find_user_dsq(dsq_id); if (unlikely(!dsq)) { scx_ops_error("invalid DSQ ID 0x%016llx", dsq_id); return false; @@ -6496,7 +6544,7 @@ __bpf_kfunc s32 scx_bpf_dsq_nr_queued(u64 dsq_id) goto out; } } else { - dsq = find_non_local_dsq(dsq_id); + dsq = find_user_dsq(dsq_id); if (dsq) { ret = READ_ONCE(dsq->nr); goto out; @@ -6545,7 +6593,7 @@ __bpf_kfunc int bpf_iter_scx_dsq_new(struct bpf_iter_scx_dsq *it, u64 dsq_id, if (flags & ~__SCX_DSQ_ITER_USER_FLAGS) return -EINVAL; - kit->dsq = find_non_local_dsq(dsq_id); + kit->dsq = find_user_dsq(dsq_id); if (!kit->dsq) return -ENOENT; |