From d689fe222a858c767cb8594faf280048e532b53f Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Wed, 13 Nov 2013 21:01:57 +0100 Subject: NOHZ: Check for nohz active instead of nohz enabled RCU and the fine grained idle time accounting functions check tick_nohz_enabled. But that variable is merily telling that NOHZ has been enabled in the config and not been disabled on the command line. But it does not tell anything about nohz being active. That's what all this should check for. Matthew reported, that the idle accounting on his old P1 machine showed bogus values, when he enabled NOHZ in the config and did not disable it on the kernel command line. The reason is that his machine uses (refined) jiffies as a clocksource which explains why the "fine" grained accounting went into lala land, because it depends on when the system goes and leaves idle relative to the jiffies increment. Provide a tick_nohz_active indicator and let RCU and the accounting code use this instead of tick_nohz_enable. Reported-and-tested-by: Matthew Whitehead Signed-off-by: Thomas Gleixner Reviewed-by: Steven Rostedt Reviewed-by: Paul E. McKenney Cc: john.stultz@linaro.org Cc: mwhitehe@redhat.com Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1311132052240.30673@ionos.tec.linutronix.de --- kernel/rcu/tree_plugin.h | 4 ++-- kernel/time/tick-sched.c | 21 +++++++++------------ 2 files changed, 11 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h index 6abb03dff5c0..08a765232432 100644 --- a/kernel/rcu/tree_plugin.h +++ b/kernel/rcu/tree_plugin.h @@ -1632,7 +1632,7 @@ module_param(rcu_idle_gp_delay, int, 0644); static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; module_param(rcu_idle_lazy_gp_delay, int, 0644); -extern int tick_nohz_enabled; +extern int tick_nohz_active; /* * Try to advance callbacks for all flavors of RCU on the current CPU, but @@ -1729,7 +1729,7 @@ static void rcu_prepare_for_idle(int cpu) int tne; /* Handle nohz enablement switches conservatively. */ - tne = ACCESS_ONCE(tick_nohz_enabled); + tne = ACCESS_ONCE(tick_nohz_active); if (tne != rdtp->tick_nohz_enabled_snap) { if (rcu_cpu_has_callbacks(cpu, NULL)) invoke_rcu_core(); /* force nohz to see update. */ diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index 3612fc77f834..a12df5abde0b 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -361,8 +361,8 @@ void __init tick_nohz_init(void) /* * NO HZ enabled ? */ -int tick_nohz_enabled __read_mostly = 1; - +static int tick_nohz_enabled __read_mostly = 1; +int tick_nohz_active __read_mostly; /* * Enable / Disable tickless mode */ @@ -465,7 +465,7 @@ u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); ktime_t now, idle; - if (!tick_nohz_enabled) + if (!tick_nohz_active) return -1; now = ktime_get(); @@ -506,7 +506,7 @@ u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); ktime_t now, iowait; - if (!tick_nohz_enabled) + if (!tick_nohz_active) return -1; now = ktime_get(); @@ -799,11 +799,6 @@ void tick_nohz_idle_enter(void) local_irq_disable(); ts = &__get_cpu_var(tick_cpu_sched); - /* - * set ts->inidle unconditionally. even if the system did not - * switch to nohz mode the cpu frequency governers rely on the - * update of the idle time accounting in tick_nohz_start_idle(). - */ ts->inidle = 1; __tick_nohz_idle_enter(ts); @@ -973,7 +968,7 @@ static void tick_nohz_switch_to_nohz(void) struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); ktime_t next; - if (!tick_nohz_enabled) + if (!tick_nohz_active) return; local_irq_disable(); @@ -981,7 +976,7 @@ static void tick_nohz_switch_to_nohz(void) local_irq_enable(); return; } - + tick_nohz_active = 1; ts->nohz_mode = NOHZ_MODE_LOWRES; /* @@ -1139,8 +1134,10 @@ void tick_setup_sched_timer(void) } #ifdef CONFIG_NO_HZ_COMMON - if (tick_nohz_enabled) + if (tick_nohz_enabled) { ts->nohz_mode = NOHZ_MODE_HIGHRES; + tick_nohz_active = 1; + } #endif } #endif /* HIGH_RES_TIMERS */ -- cgit v1.2.3 From da554eba2e68c8ec051977db5ee1f42d384a01ed Mon Sep 17 00:00:00 2001 From: Joe Perches Date: Fri, 15 Nov 2013 14:15:31 -0800 Subject: timer: Convert kmalloc_node(...GFP_ZERO...) to kzalloc_node(...) Use the helper function instead of __GFP_ZERO. Signed-off-by: Joe Perches Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/timer.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/timer.c b/kernel/timer.c index 6582b82fa966..accfd241b9e5 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -1518,9 +1518,8 @@ static int init_timers_cpu(int cpu) /* * The APs use this path later in boot */ - base = kmalloc_node(sizeof(*base), - GFP_KERNEL | __GFP_ZERO, - cpu_to_node(cpu)); + base = kzalloc_node(sizeof(*base), GFP_KERNEL, + cpu_to_node(cpu)); if (!base) return -ENOMEM; -- cgit v1.2.3 From 050ded1bbaea3331745cf2782315f5bc2582d083 Mon Sep 17 00:00:00 2001 From: Andrew Morton Date: Fri, 15 Nov 2013 14:15:33 -0800 Subject: tick: Document tick_do_timer_cpu Taken straight from a tglx email ;) Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/time/tick-common.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) (limited to 'kernel') diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c index 64522ecdfe0e..162b03ab0ad2 100644 --- a/kernel/time/tick-common.c +++ b/kernel/time/tick-common.c @@ -33,6 +33,21 @@ DEFINE_PER_CPU(struct tick_device, tick_cpu_device); */ ktime_t tick_next_period; ktime_t tick_period; + +/* + * tick_do_timer_cpu is a timer core internal variable which holds the CPU NR + * which is responsible for calling do_timer(), i.e. the timekeeping stuff. This + * variable has two functions: + * + * 1) Prevent a thundering herd issue of a gazillion of CPUs trying to grab the + * timekeeping lock all at once. Only the CPU which is assigned to do the + * update is handling it. + * + * 2) Hand off the duty in the NOHZ idle case by setting the value to + * TICK_DO_TIMER_NONE, i.e. a non existing CPU. So the next cpu which looks + * at it will take over and keep the time keeping alive. The handover + * procedure also covers cpu hotplug. + */ int tick_do_timer_cpu __read_mostly = TICK_DO_TIMER_BOOT; /* -- cgit v1.2.3 From d5b5f391d434c5cc8bcb1ab2d759738797b85f52 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 14 Nov 2013 16:23:04 +0100 Subject: ftrace, perf: Avoid infinite event generation loop Vince's perf-trinity fuzzer found yet another 'interesting' problem. When we sample the irq_work_exit tracepoint with period==1 (or PERF_SAMPLE_PERIOD) and we add an fasync SIGNAL handler we create an infinite event generation loop: ,-> | irq_work_exit() -> | trace_irq_work_exit() -> | ... | __perf_event_overflow() -> (due to fasync) | irq_work_queue() -> (irq_work_list must be empty) '--------- arch_irq_work_raise() Similar things can happen due to regular poll() wakeups if we exceed the ring-buffer wakeup watermark, or have an event_limit. To avoid this, dis-allow sampling this particular tracepoint. In order to achieve this, create a special perf_perm function pointer for each event and call this (when set) on trying to create a tracepoint perf event. [ roasted: use expr... to allow for ',' in your expression ] Reported-by: Vince Weaver Tested-by: Vince Weaver Signed-off-by: Peter Zijlstra Cc: Steven Rostedt Cc: Dave Jones Cc: Frederic Weisbecker Link: http://lkml.kernel.org/r/20131114152304.GC5364@laptop.programming.kicks-ass.net Signed-off-by: Ingo Molnar --- arch/x86/include/asm/trace/irq_vectors.h | 11 +++++++++++ include/linux/ftrace_event.h | 16 ++++++++++++++++ include/linux/tracepoint.h | 4 ++++ include/trace/ftrace.h | 7 +++++++ kernel/trace/trace_event_perf.c | 6 ++++++ 5 files changed, 44 insertions(+) (limited to 'kernel') diff --git a/arch/x86/include/asm/trace/irq_vectors.h b/arch/x86/include/asm/trace/irq_vectors.h index 2874df24e7a4..4cab890007a7 100644 --- a/arch/x86/include/asm/trace/irq_vectors.h +++ b/arch/x86/include/asm/trace/irq_vectors.h @@ -71,6 +71,17 @@ DEFINE_IRQ_VECTOR_EVENT(x86_platform_ipi); */ DEFINE_IRQ_VECTOR_EVENT(irq_work); +/* + * We must dis-allow sampling irq_work_exit() because perf event sampling + * itself can cause irq_work, which would lead to an infinite loop; + * + * 1) irq_work_exit happens + * 2) generates perf sample + * 3) generates irq_work + * 4) goto 1 + */ +TRACE_EVENT_PERF_PERM(irq_work_exit, is_sampling_event(p_event) ? -EPERM : 0); + /* * call_function - called when entering/exiting a call function interrupt * vector handler diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 9abbe630c456..8c9b7a1c4138 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -248,6 +248,9 @@ struct ftrace_event_call { #ifdef CONFIG_PERF_EVENTS int perf_refcount; struct hlist_head __percpu *perf_events; + + int (*perf_perm)(struct ftrace_event_call *, + struct perf_event *); #endif }; @@ -317,6 +320,19 @@ struct ftrace_event_file { } \ early_initcall(trace_init_flags_##name); +#define __TRACE_EVENT_PERF_PERM(name, expr...) \ + static int perf_perm_##name(struct ftrace_event_call *tp_event, \ + struct perf_event *p_event) \ + { \ + return ({ expr; }); \ + } \ + static int __init trace_init_perf_perm_##name(void) \ + { \ + event_##name.perf_perm = &perf_perm_##name; \ + return 0; \ + } \ + early_initcall(trace_init_perf_perm_##name); + #define PERF_MAX_TRACE_SIZE 2048 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */ diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h index ebeab360d851..f16dc0a40049 100644 --- a/include/linux/tracepoint.h +++ b/include/linux/tracepoint.h @@ -267,6 +267,8 @@ static inline void tracepoint_synchronize_unregister(void) #define TRACE_EVENT_FLAGS(event, flag) +#define TRACE_EVENT_PERF_PERM(event, expr...) + #endif /* DECLARE_TRACE */ #ifndef TRACE_EVENT @@ -399,4 +401,6 @@ static inline void tracepoint_synchronize_unregister(void) #define TRACE_EVENT_FLAGS(event, flag) +#define TRACE_EVENT_PERF_PERM(event, expr...) + #endif /* ifdef TRACE_EVENT (see note above) */ diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 52594b20179e..6b852f60f8ae 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -90,6 +90,10 @@ #define TRACE_EVENT_FLAGS(name, value) \ __TRACE_EVENT_FLAGS(name, value) +#undef TRACE_EVENT_PERF_PERM +#define TRACE_EVENT_PERF_PERM(name, expr...) \ + __TRACE_EVENT_PERF_PERM(name, expr) + #include TRACE_INCLUDE(TRACE_INCLUDE_FILE) @@ -140,6 +144,9 @@ #undef TRACE_EVENT_FLAGS #define TRACE_EVENT_FLAGS(event, flag) +#undef TRACE_EVENT_PERF_PERM +#define TRACE_EVENT_PERF_PERM(event, expr...) + #include TRACE_INCLUDE(TRACE_INCLUDE_FILE) /* diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c index 78e27e3b52ac..630889f68b1d 100644 --- a/kernel/trace/trace_event_perf.c +++ b/kernel/trace/trace_event_perf.c @@ -24,6 +24,12 @@ static int total_ref_count; static int perf_trace_event_perm(struct ftrace_event_call *tp_event, struct perf_event *p_event) { + if (tp_event->perf_perm) { + int ret = tp_event->perf_perm(tp_event, p_event); + if (ret) + return ret; + } + /* The ftrace function trace is allowed only for root. */ if (ftrace_event_is_function(tp_event) && perf_paranoid_tracepoint_raw() && !capable(CAP_SYS_ADMIN)) -- cgit v1.2.3 From 06db0b21712f878b808480ef31097637013bbf0f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Sep 2013 13:14:47 +0200 Subject: perf: Remove fragile swevent hlist optimization Currently we only allocate a single cpu hashtable for per-cpu swevents; do away with this optimization for it is fragile in the face of things like perf_pmu_migrate_context(). The easiest thing is to make sure all CPUs are consistent wrt state. Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20130913111447.GN31370@twins.programming.kicks-ass.net Signed-off-by: Ingo Molnar --- kernel/events/core.c | 8 -------- 1 file changed, 8 deletions(-) (limited to 'kernel') diff --git a/kernel/events/core.c b/kernel/events/core.c index d724e7757cd1..72348dc192c1 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5680,11 +5680,6 @@ static void swevent_hlist_put(struct perf_event *event) { int cpu; - if (event->cpu != -1) { - swevent_hlist_put_cpu(event, event->cpu); - return; - } - for_each_possible_cpu(cpu) swevent_hlist_put_cpu(event, cpu); } @@ -5718,9 +5713,6 @@ static int swevent_hlist_get(struct perf_event *event) int err; int cpu, failed_cpu; - if (event->cpu != -1) - return swevent_hlist_get_cpu(event, event->cpu); - get_online_cpus(); for_each_possible_cpu(cpu) { err = swevent_hlist_get_cpu(event, cpu); -- cgit v1.2.3 From 0022cedd4a7d8a87841351e2b018bb6794cf2e67 Mon Sep 17 00:00:00 2001 From: Vince Weaver Date: Fri, 15 Nov 2013 12:39:45 -0500 Subject: perf/trace: Properly use u64 to hold event_id The 64-bit attr.config value for perf trace events was being copied into an "int" before doing a comparison, meaning the top 32 bits were being truncated. As far as I can tell this didn't cause any errors, but it did mean it was possible to create valid aliases for all the tracepoint ids which I don't think was intended. (For example, 0xffffffff00000018 and 0x18 both enable the same tracepoint). Signed-off-by: Vince Weaver Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1311151236100.11932@vincent-weaver-1.um.maine.edu Signed-off-by: Ingo Molnar --- kernel/trace/trace_event_perf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c index 630889f68b1d..e854f420e033 100644 --- a/kernel/trace/trace_event_perf.c +++ b/kernel/trace/trace_event_perf.c @@ -179,7 +179,7 @@ static int perf_trace_event_init(struct ftrace_event_call *tp_event, int perf_trace_init(struct perf_event *p_event) { struct ftrace_event_call *tp_event; - int event_id = p_event->attr.config; + u64 event_id = p_event->attr.config; int ret = -EINVAL; mutex_lock(&event_mutex); -- cgit v1.2.3 From 9abf24d465180f5f2eb26a43545348262f16b771 Mon Sep 17 00:00:00 2001 From: Srikar Dronamraju Date: Tue, 12 Nov 2013 22:11:26 +0530 Subject: sched: Check sched_domain before computing group power After commit 863bffc80898 ("sched/fair: Fix group power_orig computation"), we can dereference rq->sd before it is set. Fix this by falling back to power_of() in this case and add a comment explaining things. Signed-off-by: Srikar Dronamraju [ Added comment and tweaked patch. ] Signed-off-by: Peter Zijlstra Cc: mikey@neuling.org Link: http://lkml.kernel.org/r/20131113151718.GN21461@twins.programming.kicks-ass.net Signed-off-by: Ingo Molnar --- kernel/sched/fair.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index e8b652ebe027..fd773ade1a31 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -5379,10 +5379,31 @@ void update_group_power(struct sched_domain *sd, int cpu) */ for_each_cpu(cpu, sched_group_cpus(sdg)) { - struct sched_group *sg = cpu_rq(cpu)->sd->groups; + struct sched_group_power *sgp; + struct rq *rq = cpu_rq(cpu); - power_orig += sg->sgp->power_orig; - power += sg->sgp->power; + /* + * build_sched_domains() -> init_sched_groups_power() + * gets here before we've attached the domains to the + * runqueues. + * + * Use power_of(), which is set irrespective of domains + * in update_cpu_power(). + * + * This avoids power/power_orig from being 0 and + * causing divide-by-zero issues on boot. + * + * Runtime updates will correct power_orig. + */ + if (unlikely(!rq->sd)) { + power_orig += power_of(cpu); + power += power_of(cpu); + continue; + } + + sgp = rq->sd->groups->sgp; + power_orig += sgp->power_orig; + power += sgp->power; } } else { /* -- cgit v1.2.3 From 42eb088ed246a5a817bb45a8b32fe234cf1c0f8b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 19 Nov 2013 16:41:49 +0100 Subject: sched: Avoid NULL dereference on sd_busy Commit 37dc6b50cee9 ("sched: Remove unnecessary iteration over sched domains to update nr_busy_cpus") forgot to clear 'sd_busy' under some conditions leading to a possible NULL deref in set_cpu_sd_state_idle(). Reported-by: Anton Blanchard Cc: Preeti U Murthy Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20131118113701.GF3866@twins.programming.kicks-ass.net Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched/core.c b/kernel/sched/core.c index c1808606ee5f..a1591ca7eb5a 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4910,8 +4910,9 @@ static void update_top_cache_domain(int cpu) if (sd) { id = cpumask_first(sched_domain_span(sd)); size = cpumask_weight(sched_domain_span(sd)); - rcu_assign_pointer(per_cpu(sd_busy, cpu), sd->parent); + sd = sd->parent; /* sd_busy */ } + rcu_assign_pointer(per_cpu(sd_busy, cpu), sd); rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); per_cpu(sd_llc_size, cpu) = size; -- cgit v1.2.3 From 0515973ffb16c2852a1bb1df2ca1456556faaaa5 Mon Sep 17 00:00:00 2001 From: Shigeru Yoshida Date: Sun, 17 Nov 2013 12:12:36 +0900 Subject: sched: Fix a trivial typo in comments Fix a trivial typo in rq_attach_root(). Signed-off-by: Shigeru Yoshida Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20131117.121236.1990617639803941055.shigeru.yoshida@gmail.com Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched/core.c b/kernel/sched/core.c index a1591ca7eb5a..718730dd0480 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4762,7 +4762,7 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd) cpumask_clear_cpu(rq->cpu, old_rd->span); /* - * If we dont want to free the old_rt yet then + * If we dont want to free the old_rd yet then * set old_rd to NULL to skip the freeing later * in this function: */ -- cgit v1.2.3 From 4be77398ac9d948773116b6be4a3c91b3d6ea18c Mon Sep 17 00:00:00 2001 From: Martin Schwidefsky Date: Fri, 22 Nov 2013 11:44:51 -0800 Subject: time: Fix 1ns/tick drift w/ GENERIC_TIME_VSYSCALL_OLD Since commit 1e75fa8be9f (time: Condense timekeeper.xtime into xtime_sec - merged in v3.6), there has been an problem with the error accounting in the timekeeping code, such that when truncating to nanoseconds, we round up to the next nsec, but the balancing adjustment to the ntp_error value was dropped. This causes 1ns per tick drift forward of the clock. In 3.7, this logic was isolated to only GENERIC_TIME_VSYSCALL_OLD architectures (s390, ia64, powerpc). The fix is simply to balance the accounting and to subtract the added nanosecond from ntp_error. This allows the internal long-term clock steering to keep the clock accurate. While this fix removes the regression added in 1e75fa8be9f, the ideal solution is to move away from GENERIC_TIME_VSYSCALL_OLD and use the new VSYSCALL method, which avoids entirely the nanosecond granular rounding, and the resulting short-term clock adjustment oscillation needed to keep long term accurate time. [ jstultz: Many thanks to Martin for his efforts identifying this subtle bug, and providing the fix. ] Originally-from: Martin Schwidefsky Cc: Tony Luck Cc: Paul Mackerras Cc: Benjamin Herrenschmidt Cc: Andy Lutomirski Cc: Paul Turner Cc: Steven Rostedt Cc: Richard Cochran Cc: Prarit Bhargava Cc: Fenghua Yu Cc: Thomas Gleixner Cc: stable #v3.6+ Link: http://lkml.kernel.org/r/1385149491-20307-1-git-send-email-john.stultz@linaro.org Signed-off-by: John Stultz Signed-off-by: Thomas Gleixner --- kernel/time/timekeeping.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 3abf53418b67..87b4f00284c9 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -1347,7 +1347,7 @@ static inline void old_vsyscall_fixup(struct timekeeper *tk) tk->xtime_nsec -= remainder; tk->xtime_nsec += 1ULL << tk->shift; tk->ntp_error += remainder << tk->ntp_error_shift; - + tk->ntp_error -= (1ULL << tk->shift) << tk->ntp_error_shift; } #else #define old_vsyscall_fixup(tk) -- cgit v1.2.3 From e5fca243abae1445afbfceebda5f08462ef869d3 Mon Sep 17 00:00:00 2001 From: Tejun Heo Date: Fri, 22 Nov 2013 17:14:39 -0500 Subject: cgroup: use a dedicated workqueue for cgroup destruction Since be44562613851 ("cgroup: remove synchronize_rcu() from cgroup_diput()"), cgroup destruction path makes use of workqueue. css freeing is performed from a work item from that point on and a later commit, ea15f8ccdb430 ("cgroup: split cgroup destruction into two steps"), moves css offlining to workqueue too. As cgroup destruction isn't depended upon for memory reclaim, the destruction work items were put on the system_wq; unfortunately, some controller may block in the destruction path for considerable duration while holding cgroup_mutex. As large part of destruction path is synchronized through cgroup_mutex, when combined with high rate of cgroup removals, this has potential to fill up system_wq's max_active of 256. Also, it turns out that memcg's css destruction path ends up queueing and waiting for work items on system_wq through work_on_cpu(). If such operation happens while system_wq is fully occupied by cgroup destruction work items, work_on_cpu() can't make forward progress because system_wq is full and other destruction work items on system_wq can't make forward progress because the work item waiting for work_on_cpu() is holding cgroup_mutex, leading to deadlock. This can be fixed by queueing destruction work items on a separate workqueue. This patch creates a dedicated workqueue - cgroup_destroy_wq - for this purpose. As these work items shouldn't have inter-dependencies and mostly serialized by cgroup_mutex anyway, giving high concurrency level doesn't buy anything and the workqueue's @max_active is set to 1 so that destruction work items are executed one by one on each CPU. Hugh Dickins: Because cgroup_init() is run before init_workqueues(), cgroup_destroy_wq can't be allocated from cgroup_init(). Do it from a separate core_initcall(). In the future, we probably want to reorder so that workqueue init happens before cgroup_init(). Signed-off-by: Tejun Heo Reported-by: Hugh Dickins Reported-by: Shawn Bohrer Link: http://lkml.kernel.org/r/20131111220626.GA7509@sbohrermbp13-local.rgmadvisors.com Link: http://lkml.kernel.org/g/alpine.LNX.2.00.1310301606080.2333@eggly.anvils Cc: stable@vger.kernel.org # v3.9+ --- kernel/cgroup.c | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/cgroup.c b/kernel/cgroup.c index 4c62513fe19f..a7b98ee35ef7 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -89,6 +89,14 @@ static DEFINE_MUTEX(cgroup_mutex); static DEFINE_MUTEX(cgroup_root_mutex); +/* + * cgroup destruction makes heavy use of work items and there can be a lot + * of concurrent destructions. Use a separate workqueue so that cgroup + * destruction work items don't end up filling up max_active of system_wq + * which may lead to deadlock. + */ +static struct workqueue_struct *cgroup_destroy_wq; + /* * Generate an array of cgroup subsystem pointers. At boot time, this is * populated with the built in subsystems, and modular subsystems are @@ -871,7 +879,7 @@ static void cgroup_free_rcu(struct rcu_head *head) struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); - schedule_work(&cgrp->destroy_work); + queue_work(cgroup_destroy_wq, &cgrp->destroy_work); } static void cgroup_diput(struct dentry *dentry, struct inode *inode) @@ -4249,7 +4257,7 @@ static void css_free_rcu_fn(struct rcu_head *rcu_head) * css_put(). dput() requires process context which we don't have. */ INIT_WORK(&css->destroy_work, css_free_work_fn); - schedule_work(&css->destroy_work); + queue_work(cgroup_destroy_wq, &css->destroy_work); } static void css_release(struct percpu_ref *ref) @@ -4539,7 +4547,7 @@ static void css_killed_ref_fn(struct percpu_ref *ref) container_of(ref, struct cgroup_subsys_state, refcnt); INIT_WORK(&css->destroy_work, css_killed_work_fn); - schedule_work(&css->destroy_work); + queue_work(cgroup_destroy_wq, &css->destroy_work); } /** @@ -5063,6 +5071,22 @@ out: return err; } +static int __init cgroup_wq_init(void) +{ + /* + * There isn't much point in executing destruction path in + * parallel. Good chunk is serialized with cgroup_mutex anyway. + * Use 1 for @max_active. + * + * We would prefer to do this in cgroup_init() above, but that + * is called before init_workqueues(): so leave this until after. + */ + cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); + BUG_ON(!cgroup_destroy_wq); + return 0; +} +core_initcall(cgroup_wq_init); + /* * proc_cgroup_show() * - Print task's cgroup paths into seq_file, one line for each hierarchy -- cgit v1.2.3 From 91151228065354a050fd0d190aefdd662a0580aa Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Thu, 14 Nov 2013 12:56:18 +0100 Subject: workqueue: swap set_cpus_allowed_ptr() and PF_NO_SETAFFINITY Move the setting of PF_NO_SETAFFINITY up before set_cpus_allowed() in create_worker(). Otherwise userland can change ->cpus_allowed in between. Signed-off-by: Oleg Nesterov Signed-off-by: Tejun Heo --- kernel/workqueue.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/workqueue.c b/kernel/workqueue.c index 987293d03ebc..f8942429268b 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -1736,16 +1736,17 @@ static struct worker *create_worker(struct worker_pool *pool) if (IS_ERR(worker->task)) goto fail; + set_user_nice(worker->task, pool->attrs->nice); + + /* prevent userland from meddling with cpumask of workqueue workers */ + worker->task->flags |= PF_NO_SETAFFINITY; + /* * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any * online CPUs. It'll be re-applied when any of the CPUs come up. */ - set_user_nice(worker->task, pool->attrs->nice); set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); - /* prevent userland from meddling with cpumask of workqueue workers */ - worker->task->flags |= PF_NO_SETAFFINITY; - /* * The caller is responsible for ensuring %POOL_DISASSOCIATED * remains stable across this function. See the comments above the -- cgit v1.2.3 From 8a2b75384444488fc4f2cbb9f0921b6a0794838f Mon Sep 17 00:00:00 2001 From: Tejun Heo Date: Thu, 5 Sep 2013 12:30:04 -0400 Subject: workqueue: fix ordered workqueues in NUMA setups An ordered workqueue implements execution ordering by using single pool_workqueue with max_active == 1. On a given pool_workqueue, work items are processed in FIFO order and limiting max_active to 1 enforces the queued work items to be processed one by one. Unfortunately, 4c16bd327c ("workqueue: implement NUMA affinity for unbound workqueues") accidentally broke this guarantee by applying NUMA affinity to ordered workqueues too. On NUMA setups, an ordered workqueue would end up with separate pool_workqueues for different nodes. Each pool_workqueue still limits max_active to 1 but multiple work items may be executed concurrently and out of order depending on which node they are queued to. Fix it by using dedicated ordered_wq_attrs[] when creating ordered workqueues. The new attrs match the unbound ones except that no_numa is always set thus forcing all NUMA nodes to share the default pool_workqueue. While at it, add sanity check in workqueue creation path which verifies that an ordered workqueues has only the default pool_workqueue. Signed-off-by: Tejun Heo Reported-by: Libin Cc: stable@vger.kernel.org Cc: Lai Jiangshan --- kernel/workqueue.c | 24 ++++++++++++++++++++++-- 1 file changed, 22 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/workqueue.c b/kernel/workqueue.c index f8942429268b..bbb5e9832d85 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -305,6 +305,9 @@ static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); /* I: attributes used when instantiating standard unbound pools on demand */ static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; +/* I: attributes used when instantiating ordered pools on demand */ +static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS]; + struct workqueue_struct *system_wq __read_mostly; EXPORT_SYMBOL(system_wq); struct workqueue_struct *system_highpri_wq __read_mostly; @@ -4107,7 +4110,7 @@ out_unlock: static int alloc_and_link_pwqs(struct workqueue_struct *wq) { bool highpri = wq->flags & WQ_HIGHPRI; - int cpu; + int cpu, ret; if (!(wq->flags & WQ_UNBOUND)) { wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); @@ -4127,6 +4130,13 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq) mutex_unlock(&wq->mutex); } return 0; + } else if (wq->flags & __WQ_ORDERED) { + ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); + /* there should only be single pwq for ordering guarantee */ + WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || + wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), + "ordering guarantee broken for workqueue %s\n", wq->name); + return ret; } else { return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); } @@ -5052,13 +5062,23 @@ static int __init init_workqueues(void) } } - /* create default unbound wq attrs */ + /* create default unbound and ordered wq attrs */ for (i = 0; i < NR_STD_WORKER_POOLS; i++) { struct workqueue_attrs *attrs; BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); attrs->nice = std_nice[i]; unbound_std_wq_attrs[i] = attrs; + + /* + * An ordered wq should have only one pwq as ordering is + * guaranteed by max_active which is enforced by pwqs. + * Turn off NUMA so that dfl_pwq is used for all nodes. + */ + BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); + attrs->nice = std_nice[i]; + attrs->no_numa = true; + ordered_wq_attrs[i] = attrs; } system_wq = alloc_workqueue("events", 0, 0); -- cgit v1.2.3 From 9ef28a73ff6a1598d6f915973c282fe28291f800 Mon Sep 17 00:00:00 2001 From: Li Bin Date: Mon, 9 Sep 2013 13:13:58 +0800 Subject: workqueue: fix comment typo for __queue_work() It seems the "dying" should be "draining" here. Signed-off-by: Li Bin Signed-off-by: Tejun Heo --- kernel/workqueue.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/workqueue.c b/kernel/workqueue.c index bbb5e9832d85..73bdf3c1f9b7 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -1323,7 +1323,7 @@ static void __queue_work(int cpu, struct workqueue_struct *wq, debug_work_activate(work); - /* if dying, only works from the same workqueue are allowed */ + /* if draining, only works from the same workqueue are allowed */ if (unlikely(wq->flags & __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq))) return; -- cgit v1.2.3 From 4e8b22bd1a37447712f1b1d96352fc53b463c6b3 Mon Sep 17 00:00:00 2001 From: Li Bin Date: Tue, 10 Sep 2013 09:52:35 +0800 Subject: workqueue: fix pool ID allocation leakage and remove BUILD_BUG_ON() in init_workqueues When one work starts execution, the high bits of work's data contain pool ID. It can represent a maximum of WORK_OFFQ_POOL_NONE. Pool ID is assigned WORK_OFFQ_POOL_NONE when the work being initialized indicating that no pool is associated and get_work_pool() uses it to check the associated pool. So if worker_pool_assign_id() assigns a ID greater than or equal WORK_OFFQ_POOL_NONE to a pool, it triggers leakage, and it may break the non-reentrance guarantee. This patch fix this issue by modifying the worker_pool_assign_id() function calling idr_alloc() by setting @end param WORK_OFFQ_POOL_NONE. Furthermore, in the current implementation, the BUILD_BUG_ON() in init_workqueues makes no sense. The number of worker pools needed cannot be determined at compile time, because the number of backing pools for UNBOUND workqueues is dynamic based on the assigned custom attributes. So remove it. tj: Minor comment and indentation updates. Signed-off-by: Li Bin Signed-off-by: Tejun Heo --- kernel/workqueue.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/workqueue.c b/kernel/workqueue.c index 73bdf3c1f9b7..c66912be990f 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -521,14 +521,21 @@ static inline void debug_work_activate(struct work_struct *work) { } static inline void debug_work_deactivate(struct work_struct *work) { } #endif -/* allocate ID and assign it to @pool */ +/** + * worker_pool_assign_id - allocate ID and assing it to @pool + * @pool: the pool pointer of interest + * + * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned + * successfully, -errno on failure. + */ static int worker_pool_assign_id(struct worker_pool *pool) { int ret; lockdep_assert_held(&wq_pool_mutex); - ret = idr_alloc(&worker_pool_idr, pool, 0, 0, GFP_KERNEL); + ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE, + GFP_KERNEL); if (ret >= 0) { pool->id = ret; return 0; @@ -5020,10 +5027,6 @@ static int __init init_workqueues(void) int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; int i, cpu; - /* make sure we have enough bits for OFFQ pool ID */ - BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < - WORK_CPU_END * NR_STD_WORKER_POOLS); - WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); -- cgit v1.2.3 From ac01810c9d2814238f08a227062e66a35a0e1ea2 Mon Sep 17 00:00:00 2001 From: Laxman Dewangan Date: Mon, 25 Nov 2013 19:39:47 +0530 Subject: irq: Enable all irqs unconditionally in irq_resume When the system enters suspend, it disables all interrupts in suspend_device_irqs(), including the interrupts marked EARLY_RESUME. On the resume side things are different. The EARLY_RESUME interrupts are reenabled in sys_core_ops->resume and the non EARLY_RESUME interrupts are reenabled in the normal system resume path. When suspend_noirq() failed or suspend is aborted for any other reason, we might omit the resume side call to sys_core_ops->resume() and therefor the interrupts marked EARLY_RESUME are not reenabled and stay disabled forever. To solve this, enable all irqs unconditionally in irq_resume() regardless whether interrupts marked EARLY_RESUMEhave been already enabled or not. This might try to reenable already enabled interrupts in the non failure case, but the only affected platform is XEN and it has been confirmed that it does not cause any side effects. [ tglx: Massaged changelog. ] Signed-off-by: Laxman Dewangan Acked-by-and-tested-by: Konrad Rzeszutek Wilk Acked-by: Heiko Stuebner Reviewed-by: Pavel Machek Cc: Cc: Cc: Cc: Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/1385388587-16442-1-git-send-email-ldewangan@nvidia.com Signed-off-by: Thomas Gleixner --- kernel/irq/pm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c index cb228bf21760..abcd6ca86cb7 100644 --- a/kernel/irq/pm.c +++ b/kernel/irq/pm.c @@ -50,7 +50,7 @@ static void resume_irqs(bool want_early) bool is_early = desc->action && desc->action->flags & IRQF_EARLY_RESUME; - if (is_early != want_early) + if (!is_early && want_early) continue; raw_spin_lock_irqsave(&desc->lock, flags); -- cgit v1.2.3 From 8a56d7761d2d041ae5e8215d20b4167d8aa93f51 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" Date: Mon, 25 Nov 2013 20:59:46 -0500 Subject: ftrace: Fix function graph with loading of modules Commit 8c4f3c3fa9681 "ftrace: Check module functions being traced on reload" fixed module loading and unloading with respect to function tracing, but it missed the function graph tracer. If you perform the following # cd /sys/kernel/debug/tracing # echo function_graph > current_tracer # modprobe nfsd # echo nop > current_tracer You'll get the following oops message: ------------[ cut here ]------------ WARNING: CPU: 2 PID: 2910 at /linux.git/kernel/trace/ftrace.c:1640 __ftrace_hash_rec_update.part.35+0x168/0x1b9() Modules linked in: nfsd exportfs nfs_acl lockd ipt_MASQUERADE sunrpc ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables uinput snd_hda_codec_idt CPU: 2 PID: 2910 Comm: bash Not tainted 3.13.0-rc1-test #7 Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./To be filled by O.E.M., BIOS SDBLI944.86P 05/08/2007 0000000000000668 ffff8800787efcf8 ffffffff814fe193 ffff88007d500000 0000000000000000 ffff8800787efd38 ffffffff8103b80a 0000000000000668 ffffffff810b2b9a ffffffff81a48370 0000000000000001 ffff880037aea000 Call Trace: [] dump_stack+0x4f/0x7c [] warn_slowpath_common+0x81/0x9b [] ? __ftrace_hash_rec_update.part.35+0x168/0x1b9 [] warn_slowpath_null+0x1a/0x1c [] __ftrace_hash_rec_update.part.35+0x168/0x1b9 [] ? __mutex_lock_slowpath+0x364/0x364 [] ftrace_shutdown+0xd7/0x12b [] unregister_ftrace_graph+0x49/0x78 [] graph_trace_reset+0xe/0x10 [] tracing_set_tracer+0xa7/0x26a [] tracing_set_trace_write+0x8b/0xbd [] ? ftrace_return_to_handler+0xb2/0xde [] ? __sb_end_write+0x5e/0x5e [] vfs_write+0xab/0xf6 [] ftrace_graph_caller+0x85/0x85 [] SyS_write+0x59/0x82 [] ftrace_graph_caller+0x85/0x85 [] system_call_fastpath+0x16/0x1b ---[ end trace 940358030751eafb ]--- The above mentioned commit didn't go far enough. Well, it covered the function tracer by adding checks in __register_ftrace_function(). The problem is that the function graph tracer circumvents that (for a slight efficiency gain when function graph trace is running with a function tracer. The gain was not worth this). The problem came with ftrace_startup() which should always be called after __register_ftrace_function(), if you want this bug to be completely fixed. Anyway, this solution moves __register_ftrace_function() inside of ftrace_startup() and removes the need to call them both. Reported-by: Dave Wysochanski Fixes: ed926f9b35cd ("ftrace: Use counters to enable functions to trace") Cc: stable@vger.kernel.org # 3.0+ Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 64 ++++++++++++++++++++++++++++----------------------- 1 file changed, 35 insertions(+), 29 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 22fa55696760..0e9f9eaade2f 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -367,9 +367,6 @@ static int remove_ftrace_list_ops(struct ftrace_ops **list, static int __register_ftrace_function(struct ftrace_ops *ops) { - if (unlikely(ftrace_disabled)) - return -ENODEV; - if (FTRACE_WARN_ON(ops == &global_ops)) return -EINVAL; @@ -428,9 +425,6 @@ static int __unregister_ftrace_function(struct ftrace_ops *ops) { int ret; - if (ftrace_disabled) - return -ENODEV; - if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED))) return -EBUSY; @@ -2088,10 +2082,15 @@ static void ftrace_startup_enable(int command) static int ftrace_startup(struct ftrace_ops *ops, int command) { bool hash_enable = true; + int ret; if (unlikely(ftrace_disabled)) return -ENODEV; + ret = __register_ftrace_function(ops); + if (ret) + return ret; + ftrace_start_up++; command |= FTRACE_UPDATE_CALLS; @@ -2113,12 +2112,17 @@ static int ftrace_startup(struct ftrace_ops *ops, int command) return 0; } -static void ftrace_shutdown(struct ftrace_ops *ops, int command) +static int ftrace_shutdown(struct ftrace_ops *ops, int command) { bool hash_disable = true; + int ret; if (unlikely(ftrace_disabled)) - return; + return -ENODEV; + + ret = __unregister_ftrace_function(ops); + if (ret) + return ret; ftrace_start_up--; /* @@ -2153,9 +2157,10 @@ static void ftrace_shutdown(struct ftrace_ops *ops, int command) } if (!command || !ftrace_enabled) - return; + return 0; ftrace_run_update_code(command); + return 0; } static void ftrace_startup_sysctl(void) @@ -3060,16 +3065,13 @@ static void __enable_ftrace_function_probe(void) if (i == FTRACE_FUNC_HASHSIZE) return; - ret = __register_ftrace_function(&trace_probe_ops); - if (!ret) - ret = ftrace_startup(&trace_probe_ops, 0); + ret = ftrace_startup(&trace_probe_ops, 0); ftrace_probe_registered = 1; } static void __disable_ftrace_function_probe(void) { - int ret; int i; if (!ftrace_probe_registered) @@ -3082,9 +3084,7 @@ static void __disable_ftrace_function_probe(void) } /* no more funcs left */ - ret = __unregister_ftrace_function(&trace_probe_ops); - if (!ret) - ftrace_shutdown(&trace_probe_ops, 0); + ftrace_shutdown(&trace_probe_ops, 0); ftrace_probe_registered = 0; } @@ -4366,12 +4366,15 @@ core_initcall(ftrace_nodyn_init); static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; } static inline void ftrace_startup_enable(int command) { } /* Keep as macros so we do not need to define the commands */ -# define ftrace_startup(ops, command) \ - ({ \ - (ops)->flags |= FTRACE_OPS_FL_ENABLED; \ - 0; \ +# define ftrace_startup(ops, command) \ + ({ \ + int ___ret = __register_ftrace_function(ops); \ + if (!___ret) \ + (ops)->flags |= FTRACE_OPS_FL_ENABLED; \ + ___ret; \ }) -# define ftrace_shutdown(ops, command) do { } while (0) +# define ftrace_shutdown(ops, command) __unregister_ftrace_function(ops) + # define ftrace_startup_sysctl() do { } while (0) # define ftrace_shutdown_sysctl() do { } while (0) @@ -4780,9 +4783,7 @@ int register_ftrace_function(struct ftrace_ops *ops) mutex_lock(&ftrace_lock); - ret = __register_ftrace_function(ops); - if (!ret) - ret = ftrace_startup(ops, 0); + ret = ftrace_startup(ops, 0); mutex_unlock(&ftrace_lock); @@ -4801,9 +4802,7 @@ int unregister_ftrace_function(struct ftrace_ops *ops) int ret; mutex_lock(&ftrace_lock); - ret = __unregister_ftrace_function(ops); - if (!ret) - ftrace_shutdown(ops, 0); + ret = ftrace_shutdown(ops, 0); mutex_unlock(&ftrace_lock); return ret; @@ -4997,6 +4996,13 @@ ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state, return NOTIFY_DONE; } +/* Just a place holder for function graph */ +static struct ftrace_ops fgraph_ops __read_mostly = { + .func = ftrace_stub, + .flags = FTRACE_OPS_FL_STUB | FTRACE_OPS_FL_GLOBAL | + FTRACE_OPS_FL_RECURSION_SAFE, +}; + int register_ftrace_graph(trace_func_graph_ret_t retfunc, trace_func_graph_ent_t entryfunc) { @@ -5023,7 +5029,7 @@ int register_ftrace_graph(trace_func_graph_ret_t retfunc, ftrace_graph_return = retfunc; ftrace_graph_entry = entryfunc; - ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET); + ret = ftrace_startup(&fgraph_ops, FTRACE_START_FUNC_RET); out: mutex_unlock(&ftrace_lock); @@ -5040,7 +5046,7 @@ void unregister_ftrace_graph(void) ftrace_graph_active--; ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub; ftrace_graph_entry = ftrace_graph_entry_stub; - ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET); + ftrace_shutdown(&fgraph_ops, FTRACE_STOP_FUNC_RET); unregister_pm_notifier(&ftrace_suspend_notifier); unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL); -- cgit v1.2.3 From 32e475d76a3e40879cd9ee4f69b19615062280d7 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Thu, 21 Nov 2013 12:41:44 +0100 Subject: sched: Expose preempt_schedule_irq() Tony reported that aa0d53260596 ("ia64: Use preempt_schedule_irq") broke PREEMPT=n builds on ia64. Ok, wrapped my brain around it. I tripped over the magic asm foo which has a single need_resched check and schedule point for both sys call return and interrupt return. So you need the schedule_preempt_irq() for kernel preemption from interrupt return while on a normal syscall preemption a schedule would be sufficient. But using schedule_preempt_irq() is not harmful here in any way. It just sets the preempt_active bit also in cases where it would not be required. Even on preempt=n kernels adding the preempt_active bit is completely harmless. So instead of having an extra function, moving the existing one out of the ifdef PREEMPT looks like the sanest thing to do. It would also allow getting rid of various other sti/schedule/cli asm magic in other archs. Reported-and-Tested-by: Tony Luck Fixes: aa0d53260596 ("ia64: Use preempt_schedule_irq") Signed-off-by: Thomas Gleixner [slightly edited Changelog] Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1311211230030.30673@ionos.tec.linutronix.de Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 718730dd0480..e85cda20ab2b 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -2660,6 +2660,7 @@ asmlinkage void __sched notrace preempt_schedule(void) } while (need_resched()); } EXPORT_SYMBOL(preempt_schedule); +#endif /* CONFIG_PREEMPT */ /* * this is the entry point to schedule() from kernel preemption @@ -2693,8 +2694,6 @@ asmlinkage void __sched preempt_schedule_irq(void) exception_exit(prev_state); } -#endif /* CONFIG_PREEMPT */ - int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, void *key) { -- cgit v1.2.3 From 0fc0287c9ed1ffd3706f8b4d9b314aa102ef1245 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 26 Nov 2013 15:03:41 +0100 Subject: cpuset: Fix memory allocator deadlock Juri hit the below lockdep report: [ 4.303391] ====================================================== [ 4.303392] [ INFO: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected ] [ 4.303394] 3.12.0-dl-peterz+ #144 Not tainted [ 4.303395] ------------------------------------------------------ [ 4.303397] kworker/u4:3/689 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire: [ 4.303399] (&p->mems_allowed_seq){+.+...}, at: [] new_slab+0x6c/0x290 [ 4.303417] [ 4.303417] and this task is already holding: [ 4.303418] (&(&q->__queue_lock)->rlock){..-...}, at: [] blk_execute_rq_nowait+0x5b/0x100 [ 4.303431] which would create a new lock dependency: [ 4.303432] (&(&q->__queue_lock)->rlock){..-...} -> (&p->mems_allowed_seq){+.+...} [ 4.303436] [ 4.303898] the dependencies between the lock to be acquired and SOFTIRQ-irq-unsafe lock: [ 4.303918] -> (&p->mems_allowed_seq){+.+...} ops: 2762 { [ 4.303922] HARDIRQ-ON-W at: [ 4.303923] [] __lock_acquire+0x65a/0x1ff0 [ 4.303926] [] lock_acquire+0x93/0x140 [ 4.303929] [] kthreadd+0x86/0x180 [ 4.303931] [] ret_from_fork+0x7c/0xb0 [ 4.303933] SOFTIRQ-ON-W at: [ 4.303933] [] __lock_acquire+0x68c/0x1ff0 [ 4.303935] [] lock_acquire+0x93/0x140 [ 4.303940] [] kthreadd+0x86/0x180 [ 4.303955] [] ret_from_fork+0x7c/0xb0 [ 4.303959] INITIAL USE at: [ 4.303960] [] __lock_acquire+0x344/0x1ff0 [ 4.303963] [] lock_acquire+0x93/0x140 [ 4.303966] [] kthreadd+0x86/0x180 [ 4.303969] [] ret_from_fork+0x7c/0xb0 [ 4.303972] } Which reports that we take mems_allowed_seq with interrupts enabled. A little digging found that this can only be from cpuset_change_task_nodemask(). This is an actual deadlock because an interrupt doing an allocation will hit get_mems_allowed()->...->__read_seqcount_begin(), which will spin forever waiting for the write side to complete. Cc: John Stultz Cc: Mel Gorman Reported-by: Juri Lelli Signed-off-by: Peter Zijlstra Tested-by: Juri Lelli Acked-by: Li Zefan Acked-by: Mel Gorman Signed-off-by: Tejun Heo Cc: stable@vger.kernel.org --- kernel/cpuset.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/cpuset.c b/kernel/cpuset.c index 6bf981e13c43..4772034b4b17 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -1033,8 +1033,10 @@ static void cpuset_change_task_nodemask(struct task_struct *tsk, need_loop = task_has_mempolicy(tsk) || !nodes_intersects(*newmems, tsk->mems_allowed); - if (need_loop) + if (need_loop) { + local_irq_disable(); write_seqcount_begin(&tsk->mems_allowed_seq); + } nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); @@ -1042,8 +1044,10 @@ static void cpuset_change_task_nodemask(struct task_struct *tsk, mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); tsk->mems_allowed = *newmems; - if (need_loop) + if (need_loop) { write_seqcount_end(&tsk->mems_allowed_seq); + local_irq_enable(); + } task_unlock(tsk); } -- cgit v1.2.3 From e605b36575e896edd8161534550c9ea021b03bc0 Mon Sep 17 00:00:00 2001 From: Tejun Heo Date: Wed, 27 Nov 2013 18:16:21 -0500 Subject: cgroup: fix cgroup_subsys_state leak for seq_files If a cgroup file implements either read_map() or read_seq_string(), such file is served using seq_file by overriding file->f_op to cgroup_seqfile_operations, which also overrides the release method to single_release() from cgroup_file_release(). Because cgroup_file_open() didn't use to acquire any resources, this used to be fine, but since f7d58818ba42 ("cgroup: pin cgroup_subsys_state when opening a cgroupfs file"), cgroup_file_open() pins the css (cgroup_subsys_state) which is put by cgroup_file_release(). The patch forgot to update the release path for seq_files and each open/release cycle leaks a css reference. Fix it by updating cgroup_file_release() to also handle seq_files and using it for seq_file release path too. Signed-off-by: Tejun Heo Cc: stable@vger.kernel.org # v3.12 --- kernel/cgroup.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/cgroup.c b/kernel/cgroup.c index a7b98ee35ef7..8b729c278b64 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -199,6 +199,7 @@ static void cgroup_destroy_css_killed(struct cgroup *cgrp); static int cgroup_destroy_locked(struct cgroup *cgrp); static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], bool is_add); +static int cgroup_file_release(struct inode *inode, struct file *file); /** * cgroup_css - obtain a cgroup's css for the specified subsystem @@ -2429,7 +2430,7 @@ static const struct file_operations cgroup_seqfile_operations = { .read = seq_read, .write = cgroup_file_write, .llseek = seq_lseek, - .release = single_release, + .release = cgroup_file_release, }; static int cgroup_file_open(struct inode *inode, struct file *file) @@ -2490,6 +2491,8 @@ static int cgroup_file_release(struct inode *inode, struct file *file) ret = cft->release(inode, file); if (css->ss) css_put(css); + if (file->f_op == &cgroup_seqfile_operations) + single_release(inode, file); return ret; } -- cgit v1.2.3 From 5ecbe3c3c690b5ab493c730c317475287a9e8b45 Mon Sep 17 00:00:00 2001 From: Helge Deller Date: Thu, 28 Nov 2013 09:16:33 +0100 Subject: kernel/extable: fix address-checks for core_kernel and init areas The init_kernel_text() and core_kernel_text() functions should not include the labels _einittext and _etext when checking if an address is inside the .text or .init sections. Signed-off-by: Helge Deller Signed-off-by: Linus Torvalds --- kernel/extable.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/extable.c b/kernel/extable.c index 832cb28105bb..763faf037ec1 100644 --- a/kernel/extable.c +++ b/kernel/extable.c @@ -61,7 +61,7 @@ const struct exception_table_entry *search_exception_tables(unsigned long addr) static inline int init_kernel_text(unsigned long addr) { if (addr >= (unsigned long)_sinittext && - addr <= (unsigned long)_einittext) + addr < (unsigned long)_einittext) return 1; return 0; } @@ -69,7 +69,7 @@ static inline int init_kernel_text(unsigned long addr) int core_kernel_text(unsigned long addr) { if (addr >= (unsigned long)_stext && - addr <= (unsigned long)_etext) + addr < (unsigned long)_etext) return 1; if (system_state == SYSTEM_BOOTING && -- cgit v1.2.3 From 0e576acbc1d9600cf2d9b4a141a2554639959d50 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 29 Nov 2013 12:18:13 +0100 Subject: nohz: Fix another inconsistency between CONFIG_NO_HZ=n and nohz=off If CONFIG_NO_HZ=n tick_nohz_get_sleep_length() returns NSEC_PER_SEC/HZ. If CONFIG_NO_HZ=y and the nohz functionality is disabled via the command line option "nohz=off" or not enabled due to missing hardware support, then tick_nohz_get_sleep_length() returns 0. That happens because ts->sleep_length is never set in that case. Set it to NSEC_PER_SEC/HZ when the NOHZ mode is inactive. Reported-by: Michal Hocko Reported-by: Borislav Petkov Signed-off-by: Thomas Gleixner --- kernel/time/tick-sched.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index a12df5abde0b..ea20f7d1ac2c 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -711,8 +711,10 @@ static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) return false; } - if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) + if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) { + ts->sleep_length = (ktime_t) { .tv64 = NSEC_PER_SEC/HZ }; return false; + } if (need_resched()) return false; -- cgit v1.2.3 From 3ccb01239201af06a07482ec686b14cd148102a5 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 3 Dec 2013 12:41:20 -0500 Subject: tracing: Only run synchronize_sched() at instance deletion time It has been reported that boot up with FTRACE_SELFTEST enabled can take a very long time. There can be stalls of over a minute. This was tracked down to the synchronize_sched() called when a system call event is disabled. As the self tests enable and disable thousands of events, this makes the synchronize_sched() get called thousands of times. The synchornize_sched() was added with d562aff93bfb53 "tracing: Add support for SOFT_DISABLE to syscall events" which caused this regression (added in 3.13-rc1). The synchronize_sched() is to protect against the events being accessed when a tracer instance is being deleted. When an instance is being deleted all the events associated to it are unregistered. The synchronize_sched() makes sure that no more users are running when it finishes. Instead of calling synchronize_sched() for all syscall events, we only need to call it once, after the events are unregistered and before the instance is deleted. The event_mutex is held during this action to prevent new users from enabling events. Link: http://lkml.kernel.org/r/20131203124120.427b9661@gandalf.local.home Reported-by: Petr Mladek Acked-by: Tom Zanussi Acked-by: Petr Mladek Tested-by: Petr Mladek Signed-off-by: Steven Rostedt --- kernel/trace/trace_events.c | 3 +++ kernel/trace/trace_syscalls.c | 10 ---------- 2 files changed, 3 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index f919a2e21bf3..a11800ae96de 100644 --- a/kernel/trace/trace_events.c +++ b/kernel/trace/trace_events.c @@ -2314,6 +2314,9 @@ int event_trace_del_tracer(struct trace_array *tr) /* Disable any running events */ __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0); + /* Access to events are within rcu_read_lock_sched() */ + synchronize_sched(); + down_write(&trace_event_sem); __trace_remove_event_dirs(tr); debugfs_remove_recursive(tr->event_dir); diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c index e4b6d11bdf78..ea90eb5f6f17 100644 --- a/kernel/trace/trace_syscalls.c +++ b/kernel/trace/trace_syscalls.c @@ -431,11 +431,6 @@ static void unreg_event_syscall_enter(struct ftrace_event_file *file, if (!tr->sys_refcount_enter) unregister_trace_sys_enter(ftrace_syscall_enter, tr); mutex_unlock(&syscall_trace_lock); - /* - * Callers expect the event to be completely disabled on - * return, so wait for current handlers to finish. - */ - synchronize_sched(); } static int reg_event_syscall_exit(struct ftrace_event_file *file, @@ -474,11 +469,6 @@ static void unreg_event_syscall_exit(struct ftrace_event_file *file, if (!tr->sys_refcount_exit) unregister_trace_sys_exit(ftrace_syscall_exit, tr); mutex_unlock(&syscall_trace_lock); - /* - * Callers expect the event to be completely disabled on - * return, so wait for current handlers to finish. - */ - synchronize_sched(); } static int __init init_syscall_trace(struct ftrace_event_call *call) -- cgit v1.2.3 From 8e8339a3a1069141985daaa2521ba304509ddecd Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 11 Dec 2013 11:09:53 +0100 Subject: sched: Initialize power_orig for overlapping groups Yinghai reported that he saw a /0 in sg_capacity on his EX parts. Make sure to always initialize power_orig now that we actually use it. Ideally build_sched_domains() -> init_sched_groups_power() would also initialize this; but for some yet unexplained reason some setups seem to miss updates there. Reported-by: Yinghai Lu Tested-by: Yinghai Lu Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-l8ng2m9uml6fhibln8wqpom7@git.kernel.org Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/sched/core.c b/kernel/sched/core.c index e85cda20ab2b..19af58f3a261 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -5112,6 +5112,7 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu) * die on a /0 trap. */ sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); + sg->sgp->power_orig = sg->sgp->power; /* * Make sure the first group of this domain contains the -- cgit v1.2.3 From 9dbdb155532395ba000c5d5d187658b0e17e529f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 18 Nov 2013 18:27:06 +0100 Subject: sched/fair: Rework sched_fair time accounting Christian suffers from a bad BIOS that wrecks his i5's TSC sync. This results in him occasionally seeing time going backwards - which crashes the scheduler ... Most of our time accounting can actually handle that except the most common one; the tick time update of sched_fair. There is a further problem with that code; previously we assumed that because we get a tick every TICK_NSEC our time delta could never exceed 32bits and math was simpler. However, ever since Frederic managed to get NO_HZ_FULL merged; this is no longer the case since now a task can run for a long time indeed without getting a tick. It only takes about ~4.2 seconds to overflow our u32 in nanoseconds. This means we not only need to better deal with time going backwards; but also means we need to be able to deal with large deltas. This patch reworks the entire code and uses mul_u64_u32_shr() as proposed by Andy a long while ago. We express our virtual time scale factor in a u32 multiplier and shift right and the 32bit mul_u64_u32_shr() implementation reduces to a single 32x32->64 multiply if the time delta is still short (common case). For 64bit a 64x64->128 multiply can be used if ARCH_SUPPORTS_INT128. Reported-and-Tested-by: Christian Engelmayer Signed-off-by: Peter Zijlstra Cc: fweisbec@gmail.com Cc: Paul Turner Cc: Stanislaw Gruszka Cc: Andy Lutomirski Cc: Linus Torvalds Cc: Andrew Morton Link: http://lkml.kernel.org/r/20131118172706.GI3866@twins.programming.kicks-ass.net Signed-off-by: Ingo Molnar --- include/linux/sched.h | 3 +- kernel/sched/fair.c | 144 ++++++++++++++++++++++---------------------------- 2 files changed, 66 insertions(+), 81 deletions(-) (limited to 'kernel') diff --git a/include/linux/sched.h b/include/linux/sched.h index 96d674ba3876..53f97eb8dbc7 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -930,7 +930,8 @@ struct pipe_inode_info; struct uts_namespace; struct load_weight { - unsigned long weight, inv_weight; + unsigned long weight; + u32 inv_weight; }; struct sched_avg { diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index fd773ade1a31..9030da7bcb15 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -178,59 +178,61 @@ void sched_init_granularity(void) update_sysctl(); } -#if BITS_PER_LONG == 32 -# define WMULT_CONST (~0UL) -#else -# define WMULT_CONST (1UL << 32) -#endif - +#define WMULT_CONST (~0U) #define WMULT_SHIFT 32 -/* - * Shift right and round: - */ -#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) +static void __update_inv_weight(struct load_weight *lw) +{ + unsigned long w; + + if (likely(lw->inv_weight)) + return; + + w = scale_load_down(lw->weight); + + if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST)) + lw->inv_weight = 1; + else if (unlikely(!w)) + lw->inv_weight = WMULT_CONST; + else + lw->inv_weight = WMULT_CONST / w; +} /* - * delta *= weight / lw + * delta_exec * weight / lw.weight + * OR + * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT + * + * Either weight := NICE_0_LOAD and lw \e prio_to_wmult[], in which case + * we're guaranteed shift stays positive because inv_weight is guaranteed to + * fit 32 bits, and NICE_0_LOAD gives another 10 bits; therefore shift >= 22. + * + * Or, weight =< lw.weight (because lw.weight is the runqueue weight), thus + * weight/lw.weight <= 1, and therefore our shift will also be positive. */ -static unsigned long -calc_delta_mine(unsigned long delta_exec, unsigned long weight, - struct load_weight *lw) +static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw) { - u64 tmp; + u64 fact = scale_load_down(weight); + int shift = WMULT_SHIFT; - /* - * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched - * entities since MIN_SHARES = 2. Treat weight as 1 if less than - * 2^SCHED_LOAD_RESOLUTION. - */ - if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION))) - tmp = (u64)delta_exec * scale_load_down(weight); - else - tmp = (u64)delta_exec; + __update_inv_weight(lw); - if (!lw->inv_weight) { - unsigned long w = scale_load_down(lw->weight); - - if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST)) - lw->inv_weight = 1; - else if (unlikely(!w)) - lw->inv_weight = WMULT_CONST; - else - lw->inv_weight = WMULT_CONST / w; + if (unlikely(fact >> 32)) { + while (fact >> 32) { + fact >>= 1; + shift--; + } } - /* - * Check whether we'd overflow the 64-bit multiplication: - */ - if (unlikely(tmp > WMULT_CONST)) - tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, - WMULT_SHIFT/2); - else - tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); + /* hint to use a 32x32->64 mul */ + fact = (u64)(u32)fact * lw->inv_weight; + + while (fact >> 32) { + fact >>= 1; + shift--; + } - return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); + return mul_u64_u32_shr(delta_exec, fact, shift); } @@ -443,7 +445,7 @@ find_matching_se(struct sched_entity **se, struct sched_entity **pse) #endif /* CONFIG_FAIR_GROUP_SCHED */ static __always_inline -void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec); +void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec); /************************************************************** * Scheduling class tree data structure manipulation methods: @@ -612,11 +614,10 @@ int sched_proc_update_handler(struct ctl_table *table, int write, /* * delta /= w */ -static inline unsigned long -calc_delta_fair(unsigned long delta, struct sched_entity *se) +static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se) { if (unlikely(se->load.weight != NICE_0_LOAD)) - delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load); + delta = __calc_delta(delta, NICE_0_LOAD, &se->load); return delta; } @@ -665,7 +666,7 @@ static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) update_load_add(&lw, se->load.weight); load = &lw; } - slice = calc_delta_mine(slice, se->load.weight, load); + slice = __calc_delta(slice, se->load.weight, load); } return slice; } @@ -703,47 +704,32 @@ void init_task_runnable_average(struct task_struct *p) #endif /* - * Update the current task's runtime statistics. Skip current tasks that - * are not in our scheduling class. + * Update the current task's runtime statistics. */ -static inline void -__update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, - unsigned long delta_exec) -{ - unsigned long delta_exec_weighted; - - schedstat_set(curr->statistics.exec_max, - max((u64)delta_exec, curr->statistics.exec_max)); - - curr->sum_exec_runtime += delta_exec; - schedstat_add(cfs_rq, exec_clock, delta_exec); - delta_exec_weighted = calc_delta_fair(delta_exec, curr); - - curr->vruntime += delta_exec_weighted; - update_min_vruntime(cfs_rq); -} - static void update_curr(struct cfs_rq *cfs_rq) { struct sched_entity *curr = cfs_rq->curr; u64 now = rq_clock_task(rq_of(cfs_rq)); - unsigned long delta_exec; + u64 delta_exec; if (unlikely(!curr)) return; - /* - * Get the amount of time the current task was running - * since the last time we changed load (this cannot - * overflow on 32 bits): - */ - delta_exec = (unsigned long)(now - curr->exec_start); - if (!delta_exec) + delta_exec = now - curr->exec_start; + if (unlikely((s64)delta_exec <= 0)) return; - __update_curr(cfs_rq, curr, delta_exec); curr->exec_start = now; + schedstat_set(curr->statistics.exec_max, + max(delta_exec, curr->statistics.exec_max)); + + curr->sum_exec_runtime += delta_exec; + schedstat_add(cfs_rq, exec_clock, delta_exec); + + curr->vruntime += calc_delta_fair(delta_exec, curr); + update_min_vruntime(cfs_rq); + if (entity_is_task(curr)) { struct task_struct *curtask = task_of(curr); @@ -3015,8 +3001,7 @@ static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq) } } -static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, - unsigned long delta_exec) +static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) { /* dock delta_exec before expiring quota (as it could span periods) */ cfs_rq->runtime_remaining -= delta_exec; @@ -3034,7 +3019,7 @@ static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, } static __always_inline -void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec) +void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) { if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled) return; @@ -3574,8 +3559,7 @@ static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq) return rq_clock_task(rq_of(cfs_rq)); } -static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, - unsigned long delta_exec) {} +static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {} static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq) {} static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {} static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {} -- cgit v1.2.3 From 5d4cf996cf134e8ddb4f906b8197feb9267c2b77 Mon Sep 17 00:00:00 2001 From: Mel Gorman Date: Tue, 17 Dec 2013 09:21:25 +0000 Subject: sched: Assign correct scheduling domain to 'sd_llc' Commit 42eb088e (sched: Avoid NULL dereference on sd_busy) corrected a NULL dereference on sd_busy but the fix also altered what scheduling domain it used for the 'sd_llc' percpu variable. One impact of this is that a task selecting a runqueue may consider idle CPUs that are not cache siblings as candidates for running. Tasks are then running on CPUs that are not cache hot. This was found through bisection where ebizzy threads were not seeing equal performance and it looked like a scheduling fairness issue. This patch mitigates but does not completely fix the problem on all machines tested implying there may be an additional bug or a common root cause. Here are the average range of performance seen by individual ebizzy threads. It was tested on top of candidate patches related to x86 TLB range flushing. 4-core machine 3.13.0-rc3 3.13.0-rc3 vanilla fixsd-v3r3 Mean 1 0.00 ( 0.00%) 0.00 ( 0.00%) Mean 2 0.34 ( 0.00%) 0.10 ( 70.59%) Mean 3 1.29 ( 0.00%) 0.93 ( 27.91%) Mean 4 7.08 ( 0.00%) 0.77 ( 89.12%) Mean 5 193.54 ( 0.00%) 2.14 ( 98.89%) Mean 6 151.12 ( 0.00%) 2.06 ( 98.64%) Mean 7 115.38 ( 0.00%) 2.04 ( 98.23%) Mean 8 108.65 ( 0.00%) 1.92 ( 98.23%) 8-core machine Mean 1 0.00 ( 0.00%) 0.00 ( 0.00%) Mean 2 0.40 ( 0.00%) 0.21 ( 47.50%) Mean 3 23.73 ( 0.00%) 0.89 ( 96.25%) Mean 4 12.79 ( 0.00%) 1.04 ( 91.87%) Mean 5 13.08 ( 0.00%) 2.42 ( 81.50%) Mean 6 23.21 ( 0.00%) 69.46 (-199.27%) Mean 7 15.85 ( 0.00%) 101.72 (-541.77%) Mean 8 109.37 ( 0.00%) 19.13 ( 82.51%) Mean 12 124.84 ( 0.00%) 28.62 ( 77.07%) Mean 16 113.50 ( 0.00%) 24.16 ( 78.71%) It's eliminated for one machine and reduced for another. Signed-off-by: Mel Gorman Signed-off-by: Peter Zijlstra Cc: Alex Shi Cc: Andrew Morton Cc: Fengguang Wu Cc: H Peter Anvin Cc: Linus Torvalds Link: http://lkml.kernel.org/r/20131217092124.GV11295@suse.de Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 19af58f3a261..a88f4a485c5e 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4902,6 +4902,7 @@ DEFINE_PER_CPU(struct sched_domain *, sd_asym); static void update_top_cache_domain(int cpu) { struct sched_domain *sd; + struct sched_domain *busy_sd = NULL; int id = cpu; int size = 1; @@ -4909,9 +4910,9 @@ static void update_top_cache_domain(int cpu) if (sd) { id = cpumask_first(sched_domain_span(sd)); size = cpumask_weight(sched_domain_span(sd)); - sd = sd->parent; /* sd_busy */ + busy_sd = sd->parent; /* sd_busy */ } - rcu_assign_pointer(per_cpu(sd_busy, cpu), sd); + rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); per_cpu(sd_llc_size, cpu) = size; -- cgit v1.2.3 From 757dfcaa41844595964f1220f1d33182dae49976 Mon Sep 17 00:00:00 2001 From: Kirill Tkhai Date: Wed, 27 Nov 2013 19:59:13 +0400 Subject: sched/rt: Fix rq's cpupri leak while enqueue/dequeue child RT entities This patch touches the RT group scheduling case. Functions inc_rt_prio_smp() and dec_rt_prio_smp() change (global) rq's priority, while rt_rq passed to them may be not the top-level rt_rq. This is wrong, because changing of priority on a child level does not guarantee that the priority is the highest all over the rq. So, this leak makes RT balancing unusable. The short example: the task having the highest priority among all rq's RT tasks (no one other task has the same priority) are waking on a throttle rt_rq. The rq's cpupri is set to the task's priority equivalent, but real rq->rt.highest_prio.curr is less. The patch below fixes the problem. Signed-off-by: Kirill Tkhai Signed-off-by: Peter Zijlstra CC: Steven Rostedt CC: stable@vger.kernel.org Link: http://lkml.kernel.org/r/49231385567953@web4m.yandex.ru Signed-off-by: Ingo Molnar --- kernel/sched/rt.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'kernel') diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 7d57275fc396..1c4065575fa2 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -901,6 +901,13 @@ inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) { struct rq *rq = rq_of_rt_rq(rt_rq); +#ifdef CONFIG_RT_GROUP_SCHED + /* + * Change rq's cpupri only if rt_rq is the top queue. + */ + if (&rq->rt != rt_rq) + return; +#endif if (rq->online && prio < prev_prio) cpupri_set(&rq->rd->cpupri, rq->cpu, prio); } @@ -910,6 +917,13 @@ dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) { struct rq *rq = rq_of_rt_rq(rt_rq); +#ifdef CONFIG_RT_GROUP_SCHED + /* + * Change rq's cpupri only if rt_rq is the top queue. + */ + if (&rq->rt != rt_rq) + return; +#endif if (rq->online && rt_rq->highest_prio.curr != prev_prio) cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); } -- cgit v1.2.3