From: Thomas Gleixner Date: Fri, 26 Oct 2012 18:50:54 +0100 Subject: [PATCH 236/353] sched: Add support for lazy preemption Origin: https://git.kernel.org/cgit/linux/kernel/git/rt/linux-stable-rt.git/commit?id=091e3b619256c38a1ca8760bea7e44d40b465fef It has become an obsession to mitigate the determinism vs. throughput loss of RT. Looking at the mainline semantics of preemption points gives a hint why RT sucks throughput wise for ordinary SCHED_OTHER tasks. One major issue is the wakeup of tasks which are right away preempting the waking task while the waking task holds a lock on which the woken task will block right after having preempted the wakee. In mainline this is prevented due to the implicit preemption disable of spin/rw_lock held regions. On RT this is not possible due to the fully preemptible nature of sleeping spinlocks. Though for a SCHED_OTHER task preempting another SCHED_OTHER task this is really not a correctness issue. RT folks are concerned about SCHED_FIFO/RR tasks preemption and not about the purely fairness driven SCHED_OTHER preemption latencies. So I introduced a lazy preemption mechanism which only applies to SCHED_OTHER tasks preempting another SCHED_OTHER task. Aside of the existing preempt_count each tasks sports now a preempt_lazy_count which is manipulated on lock acquiry and release. This is slightly incorrect as for lazyness reasons I coupled this on migrate_disable/enable so some other mechanisms get the same treatment (e.g. get_cpu_light). Now on the scheduler side instead of setting NEED_RESCHED this sets NEED_RESCHED_LAZY in case of a SCHED_OTHER/SCHED_OTHER preemption and therefor allows to exit the waking task the lock held region before the woken task preempts. That also works better for cross CPU wakeups as the other side can stay in the adaptive spinning loop. For RT class preemption there is no change. This simply sets NEED_RESCHED and forgoes the lazy preemption counter. Initial test do not expose any observable latency increasement, but history shows that I've been proven wrong before :) The lazy preemption mode is per default on, but with CONFIG_SCHED_DEBUG enabled it can be disabled via: # echo NO_PREEMPT_LAZY >/sys/kernel/debug/sched_features and reenabled via # echo PREEMPT_LAZY >/sys/kernel/debug/sched_features The test results so far are very machine and workload dependent, but there is a clear trend that it enhances the non RT workload performance. Signed-off-by: Thomas Gleixner --- include/linux/preempt.h | 35 ++++++++++++++- include/linux/sched.h | 38 +++++++++++++++++ include/linux/thread_info.h | 12 +++++- include/linux/trace_events.h | 1 + kernel/Kconfig.preempt | 6 +++ kernel/cpu.c | 2 + kernel/sched/core.c | 83 +++++++++++++++++++++++++++++++++++- kernel/sched/fair.c | 16 +++---- kernel/sched/features.h | 3 ++ kernel/sched/sched.h | 9 ++++ kernel/trace/trace.c | 36 +++++++++------- kernel/trace/trace.h | 2 + kernel/trace/trace_output.c | 14 +++++- 13 files changed, 228 insertions(+), 29 deletions(-) diff --git a/include/linux/preempt.h b/include/linux/preempt.h index ed8413e7140f..9c74a019bf57 100644 --- a/include/linux/preempt.h +++ b/include/linux/preempt.h @@ -180,6 +180,20 @@ extern void preempt_count_sub(int val); #define preempt_count_inc() preempt_count_add(1) #define preempt_count_dec() preempt_count_sub(1) +#ifdef CONFIG_PREEMPT_LAZY +#define add_preempt_lazy_count(val) do { preempt_lazy_count() += (val); } while (0) +#define sub_preempt_lazy_count(val) do { preempt_lazy_count() -= (val); } while (0) +#define inc_preempt_lazy_count() add_preempt_lazy_count(1) +#define dec_preempt_lazy_count() sub_preempt_lazy_count(1) +#define preempt_lazy_count() (current_thread_info()->preempt_lazy_count) +#else +#define add_preempt_lazy_count(val) do { } while (0) +#define sub_preempt_lazy_count(val) do { } while (0) +#define inc_preempt_lazy_count() do { } while (0) +#define dec_preempt_lazy_count() do { } while (0) +#define preempt_lazy_count() (0) +#endif + #ifdef CONFIG_PREEMPT_COUNT #define preempt_disable() \ @@ -188,6 +202,12 @@ do { \ barrier(); \ } while (0) +#define preempt_lazy_disable() \ +do { \ + inc_preempt_lazy_count(); \ + barrier(); \ +} while (0) + #define sched_preempt_enable_no_resched() \ do { \ barrier(); \ @@ -250,6 +270,13 @@ do { \ __preempt_schedule(); \ } while (0) +#define preempt_lazy_enable() \ +do { \ + dec_preempt_lazy_count(); \ + barrier(); \ + preempt_check_resched(); \ +} while (0) + #else /* !CONFIG_PREEMPT */ #define preempt_enable() \ do { \ @@ -257,6 +284,12 @@ do { \ preempt_count_dec(); \ } while (0) +#define preempt_lazy_enable() \ +do { \ + dec_preempt_lazy_count(); \ + barrier(); \ +} while (0) + #define preempt_enable_notrace() \ do { \ barrier(); \ @@ -323,7 +356,7 @@ do { \ } while (0) #define preempt_fold_need_resched() \ do { \ - if (tif_need_resched()) \ + if (tif_need_resched_now()) \ set_preempt_need_resched(); \ } while (0) diff --git a/include/linux/sched.h b/include/linux/sched.h index 19eb4838cf03..00b307a4156c 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1737,6 +1737,44 @@ static inline int test_tsk_need_resched(struct task_struct *tsk) return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED)); } +#ifdef CONFIG_PREEMPT_LAZY +static inline void set_tsk_need_resched_lazy(struct task_struct *tsk) +{ + set_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY); +} + +static inline void clear_tsk_need_resched_lazy(struct task_struct *tsk) +{ + clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY); +} + +static inline int test_tsk_need_resched_lazy(struct task_struct *tsk) +{ + return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED_LAZY)); +} + +static inline int need_resched_lazy(void) +{ + return test_thread_flag(TIF_NEED_RESCHED_LAZY); +} + +static inline int need_resched_now(void) +{ + return test_thread_flag(TIF_NEED_RESCHED); +} + +#else +static inline void clear_tsk_need_resched_lazy(struct task_struct *tsk) { } +static inline int need_resched_lazy(void) { return 0; } + +static inline int need_resched_now(void) +{ + return test_thread_flag(TIF_NEED_RESCHED); +} + +#endif + + static inline bool __task_is_stopped_or_traced(struct task_struct *task) { if (task->state & (__TASK_STOPPED | __TASK_TRACED)) diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h index 62dbecfe9132..57a3d99dc067 100644 --- a/include/linux/thread_info.h +++ b/include/linux/thread_info.h @@ -110,7 +110,17 @@ static inline int test_ti_thread_flag(struct thread_info *ti, int flag) #define test_thread_flag(flag) \ test_ti_thread_flag(current_thread_info(), flag) -#define tif_need_resched() test_thread_flag(TIF_NEED_RESCHED) +#ifdef CONFIG_PREEMPT_LAZY +#define tif_need_resched() (test_thread_flag(TIF_NEED_RESCHED) || \ + test_thread_flag(TIF_NEED_RESCHED_LAZY)) +#define tif_need_resched_now() (test_thread_flag(TIF_NEED_RESCHED)) +#define tif_need_resched_lazy() test_thread_flag(TIF_NEED_RESCHED_LAZY)) + +#else +#define tif_need_resched() test_thread_flag(TIF_NEED_RESCHED) +#define tif_need_resched_now() test_thread_flag(TIF_NEED_RESCHED) +#define tif_need_resched_lazy() 0 +#endif #ifndef CONFIG_HAVE_ARCH_WITHIN_STACK_FRAMES static inline int arch_within_stack_frames(const void * const stack, diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h index 1cc4d2da954c..72864a11cec0 100644 --- a/include/linux/trace_events.h +++ b/include/linux/trace_events.h @@ -64,6 +64,7 @@ struct trace_entry { int pid; unsigned short migrate_disable; unsigned short padding; + unsigned char preempt_lazy_count; }; #define TRACE_EVENT_TYPE_MAX \ diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt index 907d72b3ba95..306567f72a3e 100644 --- a/kernel/Kconfig.preempt +++ b/kernel/Kconfig.preempt @@ -6,6 +6,12 @@ config PREEMPT_RT_BASE bool select PREEMPT +config HAVE_PREEMPT_LAZY + bool + +config PREEMPT_LAZY + def_bool y if HAVE_PREEMPT_LAZY && PREEMPT_RT_FULL + choice prompt "Preemption Model" default PREEMPT_NONE diff --git a/kernel/cpu.c b/kernel/cpu.c index 6ae5b18bf3a5..396394ebbc5b 100644 --- a/kernel/cpu.c +++ b/kernel/cpu.c @@ -307,11 +307,13 @@ void pin_current_cpu(void) return; } cpu = smp_processor_id(); + preempt_lazy_enable(); preempt_enable(); __read_rt_lock(cpuhp_pin); preempt_disable(); + preempt_lazy_disable(); if (cpu != smp_processor_id()) { __read_rt_unlock(cpuhp_pin); goto again; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 41219ea235eb..e581ce1edc12 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -493,6 +493,48 @@ void resched_curr(struct rq *rq) trace_sched_wake_idle_without_ipi(cpu); } +#ifdef CONFIG_PREEMPT_LAZY + +static int tsk_is_polling(struct task_struct *p) +{ +#ifdef TIF_POLLING_NRFLAG + return test_tsk_thread_flag(p, TIF_POLLING_NRFLAG); +#else + return 0; +#endif +} + +void resched_curr_lazy(struct rq *rq) +{ + struct task_struct *curr = rq->curr; + int cpu; + + if (!sched_feat(PREEMPT_LAZY)) { + resched_curr(rq); + return; + } + + lockdep_assert_held(&rq->lock); + + if (test_tsk_need_resched(curr)) + return; + + if (test_tsk_need_resched_lazy(curr)) + return; + + set_tsk_need_resched_lazy(curr); + + cpu = cpu_of(rq); + if (cpu == smp_processor_id()) + return; + + /* NEED_RESCHED_LAZY must be visible before we test polling */ + smp_mb(); + if (!tsk_is_polling(curr)) + smp_send_reschedule(cpu); +} +#endif + void resched_cpu(int cpu) { struct rq *rq = cpu_rq(cpu); @@ -2412,6 +2454,9 @@ int sched_fork(unsigned long clone_flags, struct task_struct *p) p->on_cpu = 0; #endif init_task_preempt_count(p); +#ifdef CONFIG_HAVE_PREEMPT_LAZY + task_thread_info(p)->preempt_lazy_count = 0; +#endif #ifdef CONFIG_SMP plist_node_init(&p->pushable_tasks, MAX_PRIO); RB_CLEAR_NODE(&p->pushable_dl_tasks); @@ -3520,6 +3565,7 @@ static void __sched notrace __schedule(bool preempt) next = pick_next_task(rq, prev, &rf); clear_tsk_need_resched(prev); + clear_tsk_need_resched_lazy(prev); clear_preempt_need_resched(); if (likely(prev != next)) { @@ -3700,6 +3746,30 @@ static void __sched notrace preempt_schedule_common(void) } while (need_resched()); } +#ifdef CONFIG_PREEMPT_LAZY +/* + * If TIF_NEED_RESCHED is then we allow to be scheduled away since this is + * set by a RT task. Oterwise we try to avoid beeing scheduled out as long as + * preempt_lazy_count counter >0. + */ +static __always_inline int preemptible_lazy(void) +{ + if (test_thread_flag(TIF_NEED_RESCHED)) + return 1; + if (current_thread_info()->preempt_lazy_count) + return 0; + return 1; +} + +#else + +static inline int preemptible_lazy(void) +{ + return 1; +} + +#endif + #ifdef CONFIG_PREEMPT /* * this is the entry point to schedule() from in-kernel preemption @@ -3714,7 +3784,8 @@ asmlinkage __visible void __sched notrace preempt_schedule(void) */ if (likely(!preemptible())) return; - + if (!preemptible_lazy()) + return; preempt_schedule_common(); } NOKPROBE_SYMBOL(preempt_schedule); @@ -3741,6 +3812,9 @@ asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) if (likely(!preemptible())) return; + if (!preemptible_lazy()) + return; + do { /* * Because the function tracer can trace preempt_count_sub() @@ -5508,7 +5582,9 @@ void init_idle(struct task_struct *idle, int cpu) /* Set the preempt count _outside_ the spinlocks! */ init_idle_preempt_count(idle, cpu); - +#ifdef CONFIG_HAVE_PREEMPT_LAZY + task_thread_info(idle)->preempt_lazy_count = 0; +#endif /* * The idle tasks have their own, simple scheduling class: */ @@ -7225,6 +7301,7 @@ void migrate_disable(void) } preempt_disable(); + preempt_lazy_disable(); pin_current_cpu(); migrate_disable_update_cpus_allowed(p); @@ -7292,6 +7369,7 @@ void migrate_enable(void) arg.dest_cpu = dest_cpu; unpin_current_cpu(); + preempt_lazy_enable(); preempt_enable(); stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); tlb_migrate_finish(p->mm); @@ -7300,6 +7378,7 @@ void migrate_enable(void) } } unpin_current_cpu(); + preempt_lazy_enable(); preempt_enable(); } EXPORT_SYMBOL(migrate_enable); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index bd9a375c45f4..430248f46f72 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4151,7 +4151,7 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) ideal_runtime = sched_slice(cfs_rq, curr); delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; if (delta_exec > ideal_runtime) { - resched_curr(rq_of(cfs_rq)); + resched_curr_lazy(rq_of(cfs_rq)); /* * The current task ran long enough, ensure it doesn't get * re-elected due to buddy favours. @@ -4175,7 +4175,7 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) return; if (delta > ideal_runtime) - resched_curr(rq_of(cfs_rq)); + resched_curr_lazy(rq_of(cfs_rq)); } static void @@ -4317,7 +4317,7 @@ entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) * validating it and just reschedule. */ if (queued) { - resched_curr(rq_of(cfs_rq)); + resched_curr_lazy(rq_of(cfs_rq)); return; } /* @@ -4451,7 +4451,7 @@ static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) * hierarchy can be throttled */ if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr)) - resched_curr(rq_of(cfs_rq)); + resched_curr_lazy(rq_of(cfs_rq)); } static __always_inline @@ -5159,7 +5159,7 @@ static void hrtick_start_fair(struct rq *rq, struct task_struct *p) if (delta < 0) { if (rq->curr == p) - resched_curr(rq); + resched_curr_lazy(rq); return; } hrtick_start(rq, delta); @@ -6750,7 +6750,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_ return; preempt: - resched_curr(rq); + resched_curr_lazy(rq); /* * Only set the backward buddy when the current task is still * on the rq. This can happen when a wakeup gets interleaved @@ -9884,7 +9884,7 @@ static void task_fork_fair(struct task_struct *p) * 'current' within the tree based on its new key value. */ swap(curr->vruntime, se->vruntime); - resched_curr(rq); + resched_curr_lazy(rq); } se->vruntime -= cfs_rq->min_vruntime; @@ -9908,7 +9908,7 @@ prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio) */ if (rq->curr == p) { if (p->prio > oldprio) - resched_curr(rq); + resched_curr_lazy(rq); } else check_preempt_curr(rq, p, 0); } diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 68de18405857..12a12be6770b 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -48,6 +48,9 @@ SCHED_FEAT(NONTASK_CAPACITY, true) #ifdef CONFIG_PREEMPT_RT_FULL SCHED_FEAT(TTWU_QUEUE, false) +# ifdef CONFIG_PREEMPT_LAZY +SCHED_FEAT(PREEMPT_LAZY, true) +# endif #else /* diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 637c408fb2dc..87a05bb90124 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -1643,6 +1643,15 @@ extern void reweight_task(struct task_struct *p, int prio); extern void resched_curr(struct rq *rq); extern void resched_cpu(int cpu); +#ifdef CONFIG_PREEMPT_LAZY +extern void resched_curr_lazy(struct rq *rq); +#else +static inline void resched_curr_lazy(struct rq *rq) +{ + resched_curr(rq); +} +#endif + extern struct rt_bandwidth def_rt_bandwidth; extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index b1c82b1dc3a6..d137601eed0e 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2153,6 +2153,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, struct task_struct *tsk = current; entry->preempt_count = pc & 0xff; + entry->preempt_lazy_count = preempt_lazy_count(); entry->pid = (tsk) ? tsk->pid : 0; entry->flags = #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT @@ -2163,7 +2164,8 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) | ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) | ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) | - (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) | + (tif_need_resched_now() ? TRACE_FLAG_NEED_RESCHED : 0) | + (need_resched_lazy() ? TRACE_FLAG_NEED_RESCHED_LAZY : 0) | (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0); entry->migrate_disable = (tsk) ? __migrate_disabled(tsk) & 0xFF : 0; @@ -3364,15 +3366,17 @@ get_total_entries(struct trace_buffer *buf, static void print_lat_help_header(struct seq_file *m) { - seq_puts(m, "# _------=> CPU# \n" - "# / _-----=> irqs-off \n" - "# | / _----=> need-resched \n" - "# || / _---=> hardirq/softirq \n" - "# ||| / _--=> preempt-depth \n" - "# |||| / _--=> migrate-disable\n" - "# ||||| / delay \n" - "# cmd pid |||||| time | caller \n" - "# \\ / ||||| \\ | / \n"); + seq_puts(m, "# _--------=> CPU# \n" + "# / _-------=> irqs-off \n" + "# | / _------=> need-resched \n" + "# || / _-----=> need-resched_lazy \n" + "# ||| / _----=> hardirq/softirq \n" + "# |||| / _---=> preempt-depth \n" + "# ||||| / _--=> preempt-lazy-depth\n" + "# |||||| / _-=> migrate-disable \n" + "# ||||||| / delay \n" + "# cmd pid |||||||| time | caller \n" + "# \\ / |||||||| \\ | / \n"); } static void print_event_info(struct trace_buffer *buf, struct seq_file *m) @@ -3410,15 +3414,17 @@ static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file tgid ? tgid_space : space); seq_printf(m, "# %s / _----=> need-resched\n", tgid ? tgid_space : space); - seq_printf(m, "# %s| / _---=> hardirq/softirq\n", + seq_printf(m, "# %s| / _---=> need-resched_lazy\n", tgid ? tgid_space : space); - seq_printf(m, "# %s|| / _--=> preempt-depth\n", + seq_printf(m, "# %s|| / _--=> hardirq/softirq\n", tgid ? tgid_space : space); - seq_printf(m, "# %s||| / delay\n", + seq_printf(m, "# %s||| / preempt-depth\n", tgid ? tgid_space : space); - seq_printf(m, "# TASK-PID %sCPU# |||| TIMESTAMP FUNCTION\n", + seq_printf(m, "# %s|||| / delay\n", + tgid ? tgid_space : space); + seq_printf(m, "# TASK-PID %sCPU# ||||| TIMESTAMP FUNCTION\n", tgid ? " TGID " : space); - seq_printf(m, "# | | %s | |||| | |\n", + seq_printf(m, "# | | %s | ||||| | |\n", tgid ? " | " : space); } diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index 0923d1b18d1f..80a2d3f56c35 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -127,6 +127,7 @@ struct kretprobe_trace_entry_head { * NEED_RESCHED - reschedule is requested * HARDIRQ - inside an interrupt handler * SOFTIRQ - inside a softirq handler + * NEED_RESCHED_LAZY - lazy reschedule is requested */ enum trace_flag_type { TRACE_FLAG_IRQS_OFF = 0x01, @@ -136,6 +137,7 @@ enum trace_flag_type { TRACE_FLAG_SOFTIRQ = 0x10, TRACE_FLAG_PREEMPT_RESCHED = 0x20, TRACE_FLAG_NMI = 0x40, + TRACE_FLAG_NEED_RESCHED_LAZY = 0x80, }; #define TRACE_BUF_SIZE 1024 diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 9255100cf9fa..4424a658434e 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -448,6 +448,7 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry) { char hardsoft_irq; char need_resched; + char need_resched_lazy; char irqs_off; int hardirq; int softirq; @@ -478,6 +479,9 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry) break; } + need_resched_lazy = + (entry->flags & TRACE_FLAG_NEED_RESCHED_LAZY) ? 'L' : '.'; + hardsoft_irq = (nmi && hardirq) ? 'Z' : nmi ? 'z' : @@ -486,14 +490,20 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry) softirq ? 's' : '.' ; - trace_seq_printf(s, "%c%c%c", - irqs_off, need_resched, hardsoft_irq); + trace_seq_printf(s, "%c%c%c%c", + irqs_off, need_resched, need_resched_lazy, + hardsoft_irq); if (entry->preempt_count) trace_seq_printf(s, "%x", entry->preempt_count); else trace_seq_putc(s, '.'); + if (entry->preempt_lazy_count) + trace_seq_printf(s, "%x", entry->preempt_lazy_count); + else + trace_seq_putc(s, '.'); + if (entry->migrate_disable) trace_seq_printf(s, "%x", entry->migrate_disable); else