From e01288155222bafac9da4c09e925e28077fce407 Mon Sep 17 00:00:00 2001 From: "yang.yang29@zte.com.cn" Date: Mon, 12 Jun 2023 15:11:35 +0800 Subject: [PATCH 1/4] drm/i915/gt: Queue and wait for the irq_work item. commit f244bf6346a0af29def0e50b00ebef9cb19dba0d upstream. Disabling interrupts and invoking the irq_work function directly breaks on PREEMPT_RT. PREEMPT_RT does not invoke all irq_work from hardirq context because some of the user have spinlock_t locking in the callback function. These locks are then turned into a sleeping locks which can not be acquired with disabled interrupts. Using irq_work_queue() has the benefit that the irqwork will be invoked in the regular context. In general there is "no" delay between enqueuing the callback and its invocation because the interrupt is raised right away on architectures which support it (which includes x86). Use irq_work_queue() + irq_work_sync() instead invoking the callback directly. Reported-by: Clark Williams Signed-off-by: Sebastian Andrzej Siewior Reviewed-by: Maarten Lankhorst Signed-off-by: Steven Rostedt (VMware) --- drivers/gpu/drm/i915/gt/intel_breadcrumbs.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c index 0040b4765a54..3f4f854786f2 100644 --- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c +++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c @@ -342,10 +342,9 @@ void intel_breadcrumbs_park(struct intel_breadcrumbs *b) /* Kick the work once more to drain the signalers */ irq_work_sync(&b->irq_work); while (unlikely(READ_ONCE(b->irq_armed))) { - local_irq_disable(); - signal_irq_work(&b->irq_work); - local_irq_enable(); + irq_work_queue(&b->irq_work); cond_resched(); + irq_work_sync(&b->irq_work); } GEM_BUG_ON(!list_empty(&b->signalers)); } -- Gitee From 475115be2d93a4dd974c5c2fcea8171893b236a3 Mon Sep 17 00:00:00 2001 From: "yang.yang29@zte.com.cn" Date: Mon, 12 Jun 2023 15:16:17 +0800 Subject: [PATCH 2/4] irq_work: Allow irq_work_sync() to sleep if irq_work() no IRQ support. commit 36e7e978dfa0ab131c4d510c5887c245ca7ec453 upstream. irq_work() triggers instantly an interrupt if supported by the architecture. Otherwise the work will be processed on the next timer tick. In worst case irq_work_sync() could spin up to a jiffy. irq_work_sync() is usually used in tear down context which is fully preemptible. Based on review irq_work_sync() is invoked from preemptible context and there is one waiter at a time. This qualifies it to use rcuwait for synchronisation. Let irq_work_sync() synchronize with rcuwait if the architecture processes irqwork via the timer tick. Signed-off-by: Sebastian Andrzej Siewior Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20211006111852.1514359-3-bigeasy@linutronix.de Signed-off-by: Steven Rostedt (VMware) --- include/linux/irq_work.h | 10 +++++++++- kernel/irq_work.c | 10 ++++++++++ 2 files changed, 19 insertions(+), 1 deletion(-) diff --git a/include/linux/irq_work.h b/include/linux/irq_work.h index 7d181ced3746..94f446fa7b45 100644 --- a/include/linux/irq_work.h +++ b/include/linux/irq_work.h @@ -4,6 +4,7 @@ #include #include +#include /* * An entry can be in one of four states: @@ -23,6 +24,7 @@ struct irq_work { }; }; void (*func)(struct irq_work *); + struct rcuwait irqwait; CK_KABI_RESERVE(1) }; @@ -32,13 +34,19 @@ void init_irq_work(struct irq_work *work, void (*func)(struct irq_work *)) { atomic_set(&work->flags, 0); work->func = func; + rcuwait_init(&work->irqwait); } #define DEFINE_IRQ_WORK(name, _f) struct irq_work name = { \ .flags = ATOMIC_INIT(0), \ - .func = (_f) \ + .func = (_f), \ + .irqwait = __RCUWAIT_INITIALIZER(irqwait), \ } +static inline bool irq_work_is_busy(struct irq_work *work) +{ + return atomic_read(&work->flags) & IRQ_WORK_BUSY; +} bool irq_work_queue(struct irq_work *work); bool irq_work_queue_on(struct irq_work *work, int cpu); diff --git a/kernel/irq_work.c b/kernel/irq_work.c index 8183d30e1bb1..8969aff790e2 100644 --- a/kernel/irq_work.c +++ b/kernel/irq_work.c @@ -165,6 +165,9 @@ void irq_work_single(void *arg) */ flags &= ~IRQ_WORK_PENDING; (void)atomic_cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY); + + if (!arch_irq_work_has_interrupt()) + rcuwait_wake_up(&work->irqwait); } static void irq_work_run_list(struct llist_head *list) @@ -231,6 +234,13 @@ void irq_work_tick_soft(void) void irq_work_sync(struct irq_work *work) { lockdep_assert_irqs_enabled(); + might_sleep(); + + if (!arch_irq_work_has_interrupt()) { + rcuwait_wait_event(&work->irqwait, !irq_work_is_busy(work), + TASK_UNINTERRUPTIBLE); + return; + } while (atomic_read(&work->flags) & IRQ_WORK_BUSY) cpu_relax(); -- Gitee From 7a0a03784b8728f8c96da0a0c35095325647df5e Mon Sep 17 00:00:00 2001 From: "yang.yang29@zte.com.cn" Date: Mon, 12 Jun 2023 15:16:40 +0800 Subject: [PATCH 3/4] irq_work: Handle some irq_work in a per-CPU thread on PREEMPT_RT commit 4c1b7809adc54cf0908ecb84d80b4e142dda254f upstream. The irq_work callback is invoked in hard IRQ context. By default all callbacks are scheduled for invocation right away (given supported by the architecture) except for the ones marked IRQ_WORK_LAZY which are delayed until the next timer-tick. While looking over the callbacks, some of them may acquire locks (spinlock_t, rwlock_t) which are transformed into sleeping locks on PREEMPT_RT and must not be acquired in hard IRQ context. Changing the locks into locks which could be acquired in this context will lead to other problems such as increased latencies if everything in the chain has IRQ-off locks. This will not solve all the issues as one callback has been noticed which invoked kref_put() and its callback invokes kfree() and this can not be invoked in hardirq context. Some callbacks are required to be invoked in hardirq context even on PREEMPT_RT to work properly. This includes for instance the NO_HZ callback which needs to be able to observe the idle context. The callbacks which require to be run in hardirq have already been marked. Use this information to split the callbacks onto the two lists on PREEMPT_RT: - lazy_list Work items which are not marked with IRQ_WORK_HARD_IRQ will be added to this list. Callbacks on this list will be invoked from a per-CPU thread. The handler here may acquire sleeping locks such as spinlock_t and invoke kfree(). - raised_list Work items which are marked with IRQ_WORK_HARD_IRQ will be added to this list. They will be invoked in hardirq context and must not acquire any sleeping locks. The wake up of the per-CPU thread occurs from irq_work handler/ hardirq context. The thread runs with lowest RT priority to ensure it runs before any SCHED_OTHER tasks do. [bigeasy: melt tglx's irq_work_tick_soft() which splits irq_work_tick() into a hard and soft variant. Collected fixes over time from Steven Rostedt and Mike Galbraith. Move to per-CPU threads instead of softirq as suggested by PeterZ.] Signed-off-by: Sebastian Andrzej Siewior Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20211007092646.uhshe3ut2wkrcfzv@linutronix.de Signed-off-by: Steven Rostedt (VMware) --- include/linux/irq_work.h | 16 +++-- kernel/irq_work.c | 131 ++++++++++++++++++++++++++++----------- kernel/time/timer.c | 2 - 3 files changed, 106 insertions(+), 43 deletions(-) diff --git a/include/linux/irq_work.h b/include/linux/irq_work.h index 94f446fa7b45..d1f89a964b8c 100644 --- a/include/linux/irq_work.h +++ b/include/linux/irq_work.h @@ -43,6 +43,16 @@ void init_irq_work(struct irq_work *work, void (*func)(struct irq_work *)) .irqwait = __RCUWAIT_INITIALIZER(irqwait), \ } +#define __IRQ_WORK_INIT(_func, _flags) (struct irq_work){ \ + .flags = ATOMIC_INIT(_flags), \ + .func = (_func), \ + .irqwait = __RCUWAIT_INITIALIZER(irqwait), \ +} + +#define IRQ_WORK_INIT(_func) __IRQ_WORK_INIT(_func, 0) +#define IRQ_WORK_INIT_LAZY(_func) __IRQ_WORK_INIT(_func, IRQ_WORK_LAZY) +#define IRQ_WORK_INIT_HARD(_func) __IRQ_WORK_INIT(_func, IRQ_WORK_HARD_IRQ) + static inline bool irq_work_is_busy(struct irq_work *work) { return atomic_read(&work->flags) & IRQ_WORK_BUSY; @@ -66,10 +76,4 @@ static inline void irq_work_run(void) { } static inline void irq_work_single(void *arg) { } #endif -#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT) -void irq_work_tick_soft(void); -#else -static inline void irq_work_tick_soft(void) { } -#endif - #endif /* _LINUX_IRQ_WORK_H */ diff --git a/kernel/irq_work.c b/kernel/irq_work.c index 8969aff790e2..03d09d779ee1 100644 --- a/kernel/irq_work.c +++ b/kernel/irq_work.c @@ -18,12 +18,37 @@ #include #include #include +#include #include #include static DEFINE_PER_CPU(struct llist_head, raised_list); static DEFINE_PER_CPU(struct llist_head, lazy_list); +static DEFINE_PER_CPU(struct task_struct *, irq_workd); + +static void wake_irq_workd(void) +{ + struct task_struct *tsk = __this_cpu_read(irq_workd); + + if (!llist_empty(this_cpu_ptr(&lazy_list)) && tsk) + wake_up_process(tsk); +} + +#ifdef CONFIG_SMP +static void irq_work_wake(struct irq_work *entry) +{ + wake_irq_workd(); +} + +static DEFINE_PER_CPU(struct irq_work, irq_work_wakeup) = + IRQ_WORK_INIT_HARD(irq_work_wake); +#endif + +static int irq_workd_should_run(unsigned int cpu) +{ + return !llist_empty(this_cpu_ptr(&lazy_list)); +} /* * Claim the entry so that no one else will poke at it. @@ -54,20 +79,28 @@ void __weak arch_irq_work_raise(void) static void __irq_work_queue_local(struct irq_work *work) { struct llist_head *list; - bool lazy_work, realtime = IS_ENABLED(CONFIG_PREEMPT_RT); - - lazy_work = atomic_read(&work->flags) & IRQ_WORK_LAZY; - - /* If the work is "lazy", handle it from next tick if any */ - if (lazy_work || (realtime && !(atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ))) + bool rt_lazy_work = false; + bool lazy_work = false; + int work_flags; + + work_flags = atomic_read(&work->flags); + if (work_flags & IRQ_WORK_LAZY) + lazy_work = true; + else if (IS_ENABLED(CONFIG_PREEMPT_RT) && + !(work_flags & IRQ_WORK_HARD_IRQ)) + rt_lazy_work = true; + + if (lazy_work || rt_lazy_work) list = this_cpu_ptr(&lazy_list); else list = this_cpu_ptr(&raised_list); - if (llist_add(&work->llnode, list)) { - if (!lazy_work || tick_nohz_tick_stopped()) - arch_irq_work_raise(); - } + if (!llist_add(&work->llnode, list)) + return; + + /* If the work is "lazy", handle it from next tick if any */ + if (!lazy_work || tick_nohz_tick_stopped()) + arch_irq_work_raise(); } /* Enqueue the irq work @work on the current CPU */ @@ -110,15 +143,27 @@ bool irq_work_queue_on(struct irq_work *work, int cpu) /* Arch remote IPI send/receive backend aren't NMI safe */ WARN_ON_ONCE(in_nmi()); - if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ)) { - if (llist_add(&work->llnode, &per_cpu(lazy_list, cpu))) - arch_send_call_function_single_ipi(cpu); - } else { - __smp_call_single_queue(cpu, &work->llnode); + /* + * On PREEMPT_RT the items which are not marked as + * IRQ_WORK_HARD_IRQ are added to the lazy list and a HARD work + * item is used on the remote CPU to wake the thread. + */ + if (IS_ENABLED(CONFIG_PREEMPT_RT) && + !(atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ)) { + + if (!llist_add(&work->llnode, &per_cpu(lazy_list, cpu))) + goto out; + + work = &per_cpu(irq_work_wakeup, cpu); + if (!irq_work_claim(work)) + goto out; } + + __smp_call_single_queue(cpu, &work->llnode); } else { __irq_work_queue_local(work); } +out: preempt_enable(); return true; @@ -175,12 +220,13 @@ static void irq_work_run_list(struct llist_head *list) struct irq_work *work, *tmp; struct llist_node *llnode; -#ifndef CONFIG_PREEMPT_RT /* - * nort: On RT IRQ-work may run in SOFTIRQ context. + * On PREEMPT_RT IRQ-work which is not marked as HARD will be processed + * in a per-CPU thread in preemptible context. Only the items which are + * marked as IRQ_WORK_HARD_IRQ will be processed in hardirq context. */ - BUG_ON(!irqs_disabled()); -#endif + BUG_ON(!irqs_disabled() && !IS_ENABLED(CONFIG_PREEMPT_RT)); + if (llist_empty(list)) return; @@ -196,16 +242,10 @@ static void irq_work_run_list(struct llist_head *list) void irq_work_run(void) { irq_work_run_list(this_cpu_ptr(&raised_list)); - if (IS_ENABLED(CONFIG_PREEMPT_RT)) { - /* - * NOTE: we raise softirq via IPI for safety, - * and execute in irq_work_tick() to move the - * overhead from hard to soft irq context. - */ - if (!llist_empty(this_cpu_ptr(&lazy_list))) - raise_softirq(TIMER_SOFTIRQ); - } else + if (!IS_ENABLED(CONFIG_PREEMPT_RT)) irq_work_run_list(this_cpu_ptr(&lazy_list)); + else + wake_irq_workd(); } EXPORT_SYMBOL_GPL(irq_work_run); @@ -218,15 +258,10 @@ void irq_work_tick(void) if (!IS_ENABLED(CONFIG_PREEMPT_RT)) irq_work_run_list(this_cpu_ptr(&lazy_list)); + else + wake_irq_workd(); } -#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT) -void irq_work_tick_soft(void) -{ - irq_work_run_list(this_cpu_ptr(&lazy_list)); -} -#endif - /* * Synchronize against the irq_work @entry, ensures the entry is not * currently in use. @@ -246,3 +281,29 @@ void irq_work_sync(struct irq_work *work) cpu_relax(); } EXPORT_SYMBOL_GPL(irq_work_sync); + +static void run_irq_workd(unsigned int cpu) +{ + irq_work_run_list(this_cpu_ptr(&lazy_list)); +} + +static void irq_workd_setup(unsigned int cpu) +{ + sched_set_fifo_low(current); +} + +static struct smp_hotplug_thread irqwork_threads = { + .store = &irq_workd, + .setup = irq_workd_setup, + .thread_should_run = irq_workd_should_run, + .thread_fn = run_irq_workd, + .thread_comm = "irq_work/%u", +}; + +static __init int irq_work_init_threads(void) +{ + if (IS_ENABLED(CONFIG_PREEMPT_RT)) + BUG_ON(smpboot_register_percpu_thread(&irqwork_threads)); + return 0; +} +early_initcall(irq_work_init_threads); diff --git a/kernel/time/timer.c b/kernel/time/timer.c index 1cad0efd635c..a4fdc7cfb723 100644 --- a/kernel/time/timer.c +++ b/kernel/time/timer.c @@ -1770,8 +1770,6 @@ static __latent_entropy void run_timer_softirq(struct softirq_action *h) { struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); - irq_work_tick_soft(); - __run_timers(base); if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); -- Gitee From 3500b1b185f42dfe9cdea7e996cbe29ae4b4b474 Mon Sep 17 00:00:00 2001 From: "yang.yang29@zte.com.cn" Date: Mon, 12 Jun 2023 15:16:57 +0800 Subject: [PATCH 4/4] irq_work: Also rcuwait for !IRQ_WORK_HARD_IRQ on PREEMPT_RT commit 4fd14c237c345d597345595857e12605a8f3e1a2 upstream. On PREEMPT_RT most items are processed as LAZY via softirq context. Avoid to spin-wait for them because irq_work_sync() could have higher priority and not allow the irq-work to be completed. Wait additionally for !IRQ_WORK_HARD_IRQ irq_work items on PREEMPT_RT. Signed-off-by: Sebastian Andrzej Siewior Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20211006111852.1514359-5-bigeasy@linutronix.de Signed-off-by: Steven Rostedt (VMware) --- include/linux/irq_work.h | 5 +++++ kernel/irq_work.c | 6 ++++-- 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/include/linux/irq_work.h b/include/linux/irq_work.h index d1f89a964b8c..d0f84d197bd8 100644 --- a/include/linux/irq_work.h +++ b/include/linux/irq_work.h @@ -58,6 +58,11 @@ static inline bool irq_work_is_busy(struct irq_work *work) return atomic_read(&work->flags) & IRQ_WORK_BUSY; } +static inline bool irq_work_is_hard(struct irq_work *work) +{ + return atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ; +} + bool irq_work_queue(struct irq_work *work); bool irq_work_queue_on(struct irq_work *work, int cpu); diff --git a/kernel/irq_work.c b/kernel/irq_work.c index 03d09d779ee1..cbec10c32ead 100644 --- a/kernel/irq_work.c +++ b/kernel/irq_work.c @@ -211,7 +211,8 @@ void irq_work_single(void *arg) flags &= ~IRQ_WORK_PENDING; (void)atomic_cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY); - if (!arch_irq_work_has_interrupt()) + if ((IS_ENABLED(CONFIG_PREEMPT_RT) && !irq_work_is_hard(work)) || + !arch_irq_work_has_interrupt()) rcuwait_wake_up(&work->irqwait); } @@ -271,7 +272,8 @@ void irq_work_sync(struct irq_work *work) lockdep_assert_irqs_enabled(); might_sleep(); - if (!arch_irq_work_has_interrupt()) { + if ((IS_ENABLED(CONFIG_PREEMPT_RT) && !irq_work_is_hard(work)) || + !arch_irq_work_has_interrupt()) { rcuwait_wait_event(&work->irqwait, !irq_work_is_busy(work), TASK_UNINTERRUPTIBLE); return; -- Gitee