From 41d2e494937715d3150e5c75d01f0e75ae899337 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 13 Nov 2009 17:05:44 +0100 Subject: [PATCH 1/3] hrtimer: Tune hrtimer_interrupt hang logic The hrtimer_interrupt hang logic adjusts min_delta_ns based on the execution time of the hrtimer callbacks. This is error-prone for virtual machines, where a guest vcpu can be scheduled out during the execution of the callbacks (and the callbacks themselves can do operations that translate to blocking operations in the hypervisor), which in can lead to large min_delta_ns rendering the system unusable. Replace the current heuristics with something more reliable. Allow the interrupt code to try 3 times to catch up with the lost time. If that fails use the total time spent in the interrupt handler to defer the next timer interrupt so the system can catch up with other things which got delayed. Limit that deferment to 100ms. The retry events and the maximum time spent in the interrupt handler are recorded and exposed via /proc/timer_list Inspired by a patch from Marcelo. Reported-by: Michael Tokarev Signed-off-by: Thomas Gleixner Tested-by: Marcelo Tosatti Cc: kvm@vger.kernel.org --- include/linux/hrtimer.h | 13 ++++-- kernel/hrtimer.c | 97 +++++++++++++++++++++++----------------- kernel/time/timer_list.c | 5 ++- 3 files changed, 70 insertions(+), 45 deletions(-) diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h index 9bace4b9f4fe..040b6796ab4d 100644 --- a/include/linux/hrtimer.h +++ b/include/linux/hrtimer.h @@ -162,10 +162,11 @@ struct hrtimer_clock_base { * @expires_next: absolute time of the next event which was scheduled * via clock_set_next_event() * @hres_active: State of high resolution mode - * @check_clocks: Indictator, when set evaluate time source and clock - * event devices whether high resolution mode can be - * activated. - * @nr_events: Total number of timer interrupt events + * @hang_detected: The last hrtimer interrupt detected a hang + * @nr_events: Total number of hrtimer interrupt events + * @nr_retries: Total number of hrtimer interrupt retries + * @nr_hangs: Total number of hrtimer interrupt hangs + * @max_hang_time: Maximum time spent in hrtimer_interrupt */ struct hrtimer_cpu_base { spinlock_t lock; @@ -173,7 +174,11 @@ struct hrtimer_cpu_base { #ifdef CONFIG_HIGH_RES_TIMERS ktime_t expires_next; int hres_active; + int hang_detected; unsigned long nr_events; + unsigned long nr_retries; + unsigned long nr_hangs; + ktime_t max_hang_time; #endif }; diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index ede527708123..931a4d99bc55 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -557,7 +557,7 @@ hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) static int hrtimer_reprogram(struct hrtimer *timer, struct hrtimer_clock_base *base) { - ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; + struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); int res; @@ -582,7 +582,16 @@ static int hrtimer_reprogram(struct hrtimer *timer, if (expires.tv64 < 0) return -ETIME; - if (expires.tv64 >= expires_next->tv64) + if (expires.tv64 >= cpu_base->expires_next.tv64) + return 0; + + /* + * If a hang was detected in the last timer interrupt then we + * do not schedule a timer which is earlier than the expiry + * which we enforced in the hang detection. We want the system + * to make progress. + */ + if (cpu_base->hang_detected) return 0; /* @@ -590,7 +599,7 @@ static int hrtimer_reprogram(struct hrtimer *timer, */ res = tick_program_event(expires, 0); if (!IS_ERR_VALUE(res)) - *expires_next = expires; + cpu_base->expires_next = expires; return res; } @@ -1217,30 +1226,6 @@ static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) #ifdef CONFIG_HIGH_RES_TIMERS -static int force_clock_reprogram; - -/* - * After 5 iteration's attempts, we consider that hrtimer_interrupt() - * is hanging, which could happen with something that slows the interrupt - * such as the tracing. Then we force the clock reprogramming for each future - * hrtimer interrupts to avoid infinite loops and use the min_delta_ns - * threshold that we will overwrite. - * The next tick event will be scheduled to 3 times we currently spend on - * hrtimer_interrupt(). This gives a good compromise, the cpus will spend - * 1/4 of their time to process the hrtimer interrupts. This is enough to - * let it running without serious starvation. - */ - -static inline void -hrtimer_interrupt_hanging(struct clock_event_device *dev, - ktime_t try_time) -{ - force_clock_reprogram = 1; - dev->min_delta_ns = (unsigned long)try_time.tv64 * 3; - printk(KERN_WARNING "hrtimer: interrupt too slow, " - "forcing clock min delta to %llu ns\n", - (unsigned long long) dev->min_delta_ns); -} /* * High resolution timer interrupt * Called with interrupts disabled @@ -1249,21 +1234,15 @@ void hrtimer_interrupt(struct clock_event_device *dev) { struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); struct hrtimer_clock_base *base; - ktime_t expires_next, now; - int nr_retries = 0; - int i; + ktime_t expires_next, now, entry_time, delta; + int i, retries = 0; BUG_ON(!cpu_base->hres_active); cpu_base->nr_events++; dev->next_event.tv64 = KTIME_MAX; - retry: - /* 5 retries is enough to notice a hang */ - if (!(++nr_retries % 5)) - hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now)); - - now = ktime_get(); - + entry_time = now = ktime_get(); +retry: expires_next.tv64 = KTIME_MAX; spin_lock(&cpu_base->lock); @@ -1325,10 +1304,48 @@ void hrtimer_interrupt(struct clock_event_device *dev) spin_unlock(&cpu_base->lock); /* Reprogramming necessary ? */ - if (expires_next.tv64 != KTIME_MAX) { - if (tick_program_event(expires_next, force_clock_reprogram)) - goto retry; + if (expires_next.tv64 == KTIME_MAX || + !tick_program_event(expires_next, 0)) { + cpu_base->hang_detected = 0; + return; } + + /* + * The next timer was already expired due to: + * - tracing + * - long lasting callbacks + * - being scheduled away when running in a VM + * + * We need to prevent that we loop forever in the hrtimer + * interrupt routine. We give it 3 attempts to avoid + * overreacting on some spurious event. + */ + now = ktime_get(); + cpu_base->nr_retries++; + if (++retries < 3) + goto retry; + /* + * Give the system a chance to do something else than looping + * here. We stored the entry time, so we know exactly how long + * we spent here. We schedule the next event this amount of + * time away. + */ + cpu_base->nr_hangs++; + cpu_base->hang_detected = 1; + delta = ktime_sub(now, entry_time); + if (delta.tv64 > cpu_base->max_hang_time.tv64) + cpu_base->max_hang_time = delta; + /* + * Limit it to a sensible value as we enforce a longer + * delay. Give the CPU at least 100ms to catch up. + */ + if (delta.tv64 > 100 * NSEC_PER_MSEC) + expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); + else + expires_next = ktime_add(now, delta); + tick_program_event(expires_next, 1); + printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", + ktime_to_ns(delta)); } /* diff --git a/kernel/time/timer_list.c b/kernel/time/timer_list.c index 665c76edbf17..9d80db4747d4 100644 --- a/kernel/time/timer_list.c +++ b/kernel/time/timer_list.c @@ -150,6 +150,9 @@ static void print_cpu(struct seq_file *m, int cpu, u64 now) P_ns(expires_next); P(hres_active); P(nr_events); + P(nr_retries); + P(nr_hangs); + P_ns(max_hang_time); #endif #undef P #undef P_ns @@ -254,7 +257,7 @@ static int timer_list_show(struct seq_file *m, void *v) u64 now = ktime_to_ns(ktime_get()); int cpu; - SEQ_printf(m, "Timer List Version: v0.4\n"); + SEQ_printf(m, "Timer List Version: v0.5\n"); SEQ_printf(m, "HRTIMER_MAX_CLOCK_BASES: %d\n", HRTIMER_MAX_CLOCK_BASES); SEQ_printf(m, "now at %Ld nsecs\n", (unsigned long long)now); From 5f201907dfe4ad42c44006ddfcec00ed12e59497 Mon Sep 17 00:00:00 2001 From: Heiko Carstens Date: Thu, 10 Dec 2009 10:56:29 +0100 Subject: [PATCH 2/3] hrtimer: move timer stats helper functions to hrtimer.c There is no reason to make timer_stats_hrtimer_set_start_info and friends visible to the rest of the kernel. So move all of them to hrtimer.c. Also make timer_stats_hrtimer_set_start_info a static inline function so it gets inlined and we avoid another function call. Based on a patch by Thomas Gleixner. Signed-off-by: Heiko Carstens LKML-Reference: <20091210095629.GC4144@osiris.boeblingen.de.ibm.com> Signed-off-by: Thomas Gleixner --- include/linux/hrtimer.h | 43 ----------------------------------------- kernel/hrtimer.c | 26 ++++++++++++++++++++----- 2 files changed, 21 insertions(+), 48 deletions(-) diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h index 040b6796ab4d..af634e95871d 100644 --- a/include/linux/hrtimer.h +++ b/include/linux/hrtimer.h @@ -440,47 +440,4 @@ extern u64 ktime_divns(const ktime_t kt, s64 div); /* Show pending timers: */ extern void sysrq_timer_list_show(void); -/* - * Timer-statistics info: - */ -#ifdef CONFIG_TIMER_STATS - -extern void timer_stats_update_stats(void *timer, pid_t pid, void *startf, - void *timerf, char *comm, - unsigned int timer_flag); - -static inline void timer_stats_account_hrtimer(struct hrtimer *timer) -{ - if (likely(!timer_stats_active)) - return; - timer_stats_update_stats(timer, timer->start_pid, timer->start_site, - timer->function, timer->start_comm, 0); -} - -extern void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, - void *addr); - -static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) -{ - __timer_stats_hrtimer_set_start_info(timer, __builtin_return_address(0)); -} - -static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) -{ - timer->start_site = NULL; -} -#else -static inline void timer_stats_account_hrtimer(struct hrtimer *timer) -{ -} - -static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) -{ -} - -static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) -{ -} -#endif - #endif diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index 931a4d99bc55..d2f9239dc6ba 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -756,17 +756,33 @@ static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } #endif /* CONFIG_HIGH_RES_TIMERS */ -#ifdef CONFIG_TIMER_STATS -void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) +static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) { +#ifdef CONFIG_TIMER_STATS if (timer->start_site) return; - - timer->start_site = addr; + timer->start_site = __builtin_return_address(0); memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); timer->start_pid = current->pid; -} #endif +} + +static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) +{ +#ifdef CONFIG_TIMER_STATS + timer->start_site = NULL; +#endif +} + +static inline void timer_stats_account_hrtimer(struct hrtimer *timer) +{ +#ifdef CONFIG_TIMER_STATS + if (likely(!timer_stats_active)) + return; + timer_stats_update_stats(timer, timer->start_pid, timer->start_site, + timer->function, timer->start_comm, 0); +#endif +} /* * Counterpart to lock_hrtimer_base above: From e9c0748b687aa70179a9e6d8ffc24b2874fe350b Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Thu, 10 Dec 2009 13:23:19 +0100 Subject: [PATCH 3/3] itimer: Fix the itimer trace print format Compiling powerpc64 results in: include/trace/events/timer.h:279: warning: format '%lu' expects type 'long unsigned int', but argument 4 has type 'cputime_t' .... cputime_t on power is u64, which triggers the above warning. Cast the cputime_t to unsigned long long and fix the print format string. That works on both 32 and 64 bit architectures. While at it change the print format for long variables from %lu to %ld. Signed-off-by: Thomas Gleixner Cc: Xiao Guangrong --- include/trace/events/timer.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/include/trace/events/timer.h b/include/trace/events/timer.h index e5ce87a0498d..9496b965d62a 100644 --- a/include/trace/events/timer.h +++ b/include/trace/events/timer.h @@ -301,8 +301,8 @@ TRACE_EVENT(itimer_state, __entry->interval_usec = value->it_interval.tv_usec; ), - TP_printk("which=%d expires=%lu it_value=%lu.%lu it_interval=%lu.%lu", - __entry->which, __entry->expires, + TP_printk("which=%d expires=%llu it_value=%ld.%ld it_interval=%ld.%ld", + __entry->which, (unsigned long long)__entry->expires, __entry->value_sec, __entry->value_usec, __entry->interval_sec, __entry->interval_usec) ); @@ -331,8 +331,8 @@ TRACE_EVENT(itimer_expire, __entry->pid = pid_nr(pid); ), - TP_printk("which=%d pid=%d now=%lu", __entry->which, - (int) __entry->pid, __entry->now) + TP_printk("which=%d pid=%d now=%llu", __entry->which, + (int) __entry->pid, (unsigned long long)__entry->now) ); #endif /* _TRACE_TIMER_H */