From 3cac97cbb14aed00d83eb33d4613b0fe3aaea863 Mon Sep 17 00:00:00 2001 From: Lai Jiangshan Date: Sun, 6 Jul 2008 17:23:55 +0800 Subject: rcu classic: simplify the next pending batch use a batch number(rcp->pending) instead of a flag(rcp->next_pending) rcu_start_batch() need to change this flag, so mb()s is needed for memory-access safe. but(after this patch applied) rcu_start_batch() do not change this batch number(rcp->pending), rcp->pending is managed by __rcu_process_callbacks only, and troublesome mb()s are eliminated. And codes look simpler and clearer. Signed-off-by: Lai Jiangshan Cc: "Paul E. McKenney" Cc: Dipankar Sarma Cc: Gautham Shenoy Cc: Dhaval Giani Cc: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 22 ++++++++-------------- 1 file changed, 8 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 16eeeaa9d618..03726eb95193 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -60,12 +60,14 @@ EXPORT_SYMBOL_GPL(rcu_lock_map); static struct rcu_ctrlblk rcu_ctrlblk = { .cur = -300, .completed = -300, + .pending = -300, .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock), .cpumask = CPU_MASK_NONE, }; static struct rcu_ctrlblk rcu_bh_ctrlblk = { .cur = -300, .completed = -300, + .pending = -300, .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock), .cpumask = CPU_MASK_NONE, }; @@ -276,14 +278,8 @@ static void rcu_do_batch(struct rcu_data *rdp) */ static void rcu_start_batch(struct rcu_ctrlblk *rcp) { - if (rcp->next_pending && + if (rcp->cur != rcp->pending && rcp->completed == rcp->cur) { - rcp->next_pending = 0; - /* - * next_pending == 0 must be visible in - * __rcu_process_callbacks() before it can see new value of cur. - */ - smp_wmb(); rcp->cur++; /* @@ -441,16 +437,14 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, /* determine batch number */ rdp->batch = rcp->cur + 1; - /* see the comment and corresponding wmb() in - * the rcu_start_batch() - */ - smp_rmb(); - if (!rcp->next_pending) { + if (rcu_batch_after(rdp->batch, rcp->pending)) { /* and start it/schedule start if it's a new batch */ spin_lock(&rcp->lock); - rcp->next_pending = 1; - rcu_start_batch(rcp); + if (rcu_batch_after(rdp->batch, rcp->pending)) { + rcp->pending = rdp->batch; + rcu_start_batch(rcp); + } spin_unlock(&rcp->lock); } } -- cgit v1.2.3 From 5127bed588a2f8f3a1f732de2a8a190b7df5dce3 Mon Sep 17 00:00:00 2001 From: Lai Jiangshan Date: Sun, 6 Jul 2008 17:23:59 +0800 Subject: rcu classic: new algorithm for callbacks-processing(v2) This is v2, it's a little deference from v1 that I had send to lkml. use ACCESS_ONCE use rcu_batch_after/rcu_batch_before for batch # comparison. rcutorture test result: (hotplugs: do cpu-online/offline once per second) No CONFIG_NO_HZ: OK, 12hours No CONFIG_NO_HZ, hotplugs: OK, 12hours CONFIG_NO_HZ=y: OK, 24hours CONFIG_NO_HZ=y, hotplugs: Failed. (Failed also without my patch applied, exactly the same bug occurred, http://lkml.org/lkml/2008/7/3/24) v1's email thread: http://lkml.org/lkml/2008/6/2/539 v1's description: The code/algorithm of the implement of current callbacks-processing is very efficient and technical. But when I studied it and I found a disadvantage: In multi-CPU systems, when a new RCU callback is being queued(call_rcu[_bh]), this callback will be invoked after the grace period for the batch with batch number = rcp->cur+2 has completed very very likely in current implement. Actually, this callback can be invoked after the grace period for the batch with batch number = rcp->cur+1 has completed. The delay of invocation means that latency of synchronize_rcu() is extended. But more important thing is that the callbacks usually free memory, and these works are delayed too! it's necessary for reclaimer to free memory as soon as possible when left memory is few. A very simple way can solve this problem: a field(struct rcu_head::batch) is added to record the batch number for the RCU callback. And when a new RCU callback is being queued, we determine the batch number for this callback(head->batch = rcp->cur+1) and we move this callback to rdp->donelist if we find that head->batch <= rcp->completed when we process callbacks. This simple way reduces the wait time for invocation a lot. (about 2.5Grace Period -> 1.5Grace Period in average in multi-CPU systems) This is my algorithm. But I do not add any field for struct rcu_head in my implement. We just need to memorize the last 2 batches and their batch number, because these 2 batches include all entries that for whom the grace period hasn't completed. So we use a special linked-list rather than add a field. Please see the comment of struct rcu_data. Signed-off-by: Lai Jiangshan Cc: "Paul E. McKenney" Cc: Dipankar Sarma Cc: Gautham Shenoy Cc: Dhaval Giani Cc: Peter Zijlstra Signed-off-by: Ingo Molnar --- include/linux/rcuclassic.h | 26 +++++--- kernel/rcuclassic.c | 157 ++++++++++++++++++++++++++++----------------- 2 files changed, 114 insertions(+), 69 deletions(-) (limited to 'kernel') diff --git a/include/linux/rcuclassic.h b/include/linux/rcuclassic.h index c847e59c6006..04c728147be0 100644 --- a/include/linux/rcuclassic.h +++ b/include/linux/rcuclassic.h @@ -66,11 +66,7 @@ static inline int rcu_batch_after(long a, long b) return (a - b) > 0; } -/* - * Per-CPU data for Read-Copy UPdate. - * nxtlist - new callbacks are added here - * curlist - current batch for which quiescent cycle started if any - */ +/* Per-CPU data for Read-Copy UPdate. */ struct rcu_data { /* 1) quiescent state handling : */ long quiescbatch; /* Batch # for grace period */ @@ -78,12 +74,24 @@ struct rcu_data { int qs_pending; /* core waits for quiesc state */ /* 2) batch handling */ - long batch; /* Batch # for current RCU batch */ + /* + * if nxtlist is not NULL, then: + * batch: + * The batch # for the last entry of nxtlist + * [*nxttail[1], NULL = *nxttail[2]): + * Entries that batch # <= batch + * [*nxttail[0], *nxttail[1]): + * Entries that batch # <= batch - 1 + * [nxtlist, *nxttail[0]): + * Entries that batch # <= batch - 2 + * The grace period for these entries has completed, and + * the other grace-period-completed entries may be moved + * here temporarily in rcu_process_callbacks(). + */ + long batch; struct rcu_head *nxtlist; - struct rcu_head **nxttail; + struct rcu_head **nxttail[3]; long qlen; /* # of queued callbacks */ - struct rcu_head *curlist; - struct rcu_head **curtail; struct rcu_head *donelist; struct rcu_head **donetail; long blimit; /* Upper limit on a processed batch */ diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 03726eb95193..d3553ee55f64 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -120,6 +120,43 @@ static inline void force_quiescent_state(struct rcu_data *rdp, } #endif +static void __call_rcu(struct rcu_head *head, struct rcu_ctrlblk *rcp, + struct rcu_data *rdp) +{ + long batch; + smp_mb(); /* reads the most recently updated value of rcu->cur. */ + + /* + * Determine the batch number of this callback. + * + * Using ACCESS_ONCE to avoid the following error when gcc eliminates + * local variable "batch" and emits codes like this: + * 1) rdp->batch = rcp->cur + 1 # gets old value + * ...... + * 2)rcu_batch_after(rcp->cur + 1, rdp->batch) # gets new value + * then [*nxttail[0], *nxttail[1]) may contain callbacks + * that batch# = rdp->batch, see the comment of struct rcu_data. + */ + batch = ACCESS_ONCE(rcp->cur) + 1; + + if (rdp->nxtlist && rcu_batch_after(batch, rdp->batch)) { + /* process callbacks */ + rdp->nxttail[0] = rdp->nxttail[1]; + rdp->nxttail[1] = rdp->nxttail[2]; + if (rcu_batch_after(batch - 1, rdp->batch)) + rdp->nxttail[0] = rdp->nxttail[2]; + } + + rdp->batch = batch; + *rdp->nxttail[2] = head; + rdp->nxttail[2] = &head->next; + + if (unlikely(++rdp->qlen > qhimark)) { + rdp->blimit = INT_MAX; + force_quiescent_state(rdp, &rcu_ctrlblk); + } +} + /** * call_rcu - Queue an RCU callback for invocation after a grace period. * @head: structure to be used for queueing the RCU updates. @@ -135,18 +172,11 @@ void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) { unsigned long flags; - struct rcu_data *rdp; head->func = func; head->next = NULL; local_irq_save(flags); - rdp = &__get_cpu_var(rcu_data); - *rdp->nxttail = head; - rdp->nxttail = &head->next; - if (unlikely(++rdp->qlen > qhimark)) { - rdp->blimit = INT_MAX; - force_quiescent_state(rdp, &rcu_ctrlblk); - } + __call_rcu(head, &rcu_ctrlblk, &__get_cpu_var(rcu_data)); local_irq_restore(flags); } EXPORT_SYMBOL_GPL(call_rcu); @@ -171,20 +201,11 @@ void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) { unsigned long flags; - struct rcu_data *rdp; head->func = func; head->next = NULL; local_irq_save(flags); - rdp = &__get_cpu_var(rcu_bh_data); - *rdp->nxttail = head; - rdp->nxttail = &head->next; - - if (unlikely(++rdp->qlen > qhimark)) { - rdp->blimit = INT_MAX; - force_quiescent_state(rdp, &rcu_bh_ctrlblk); - } - + __call_rcu(head, &rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); local_irq_restore(flags); } EXPORT_SYMBOL_GPL(call_rcu_bh); @@ -213,12 +234,6 @@ EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); static inline void raise_rcu_softirq(void) { raise_softirq(RCU_SOFTIRQ); - /* - * The smp_mb() here is required to ensure that this cpu's - * __rcu_process_callbacks() reads the most recently updated - * value of rcu->cur. - */ - smp_mb(); } /* @@ -360,13 +375,15 @@ static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, * which is dead and hence not processing interrupts. */ static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, - struct rcu_head **tail) + struct rcu_head **tail, long batch) { - local_irq_disable(); - *this_rdp->nxttail = list; - if (list) - this_rdp->nxttail = tail; - local_irq_enable(); + if (list) { + local_irq_disable(); + this_rdp->batch = batch; + *this_rdp->nxttail[2] = list; + this_rdp->nxttail[2] = tail; + local_irq_enable(); + } } static void __rcu_offline_cpu(struct rcu_data *this_rdp, @@ -380,9 +397,9 @@ static void __rcu_offline_cpu(struct rcu_data *this_rdp, if (rcp->cur != rcp->completed) cpu_quiet(rdp->cpu, rcp); spin_unlock_bh(&rcp->lock); - rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail); - rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail); - rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail); + /* spin_lock implies smp_mb() */ + rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail, rcp->cur + 1); + rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail[2], rcp->cur + 1); local_irq_disable(); this_rdp->qlen += rdp->qlen; @@ -416,27 +433,37 @@ static void rcu_offline_cpu(int cpu) static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { - if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { - *rdp->donetail = rdp->curlist; - rdp->donetail = rdp->curtail; - rdp->curlist = NULL; - rdp->curtail = &rdp->curlist; - } - - if (rdp->nxtlist && !rdp->curlist) { + if (rdp->nxtlist) { local_irq_disable(); - rdp->curlist = rdp->nxtlist; - rdp->curtail = rdp->nxttail; - rdp->nxtlist = NULL; - rdp->nxttail = &rdp->nxtlist; - local_irq_enable(); /* - * start the next batch of callbacks + * move the other grace-period-completed entries to + * [rdp->nxtlist, *rdp->nxttail[0]) temporarily + */ + if (!rcu_batch_before(rcp->completed, rdp->batch)) + rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2]; + else if (!rcu_batch_before(rcp->completed, rdp->batch - 1)) + rdp->nxttail[0] = rdp->nxttail[1]; + + /* + * the grace period for entries in + * [rdp->nxtlist, *rdp->nxttail[0]) has completed and + * move these entries to donelist */ + if (rdp->nxttail[0] != &rdp->nxtlist) { + *rdp->donetail = rdp->nxtlist; + rdp->donetail = rdp->nxttail[0]; + rdp->nxtlist = *rdp->nxttail[0]; + *rdp->donetail = NULL; + + if (rdp->nxttail[1] == rdp->nxttail[0]) + rdp->nxttail[1] = &rdp->nxtlist; + if (rdp->nxttail[2] == rdp->nxttail[0]) + rdp->nxttail[2] = &rdp->nxtlist; + rdp->nxttail[0] = &rdp->nxtlist; + } - /* determine batch number */ - rdp->batch = rcp->cur + 1; + local_irq_enable(); if (rcu_batch_after(rdp->batch, rcp->pending)) { /* and start it/schedule start if it's a new batch */ @@ -462,15 +489,26 @@ static void rcu_process_callbacks(struct softirq_action *unused) static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { - /* This cpu has pending rcu entries and the grace period - * for them has completed. - */ - if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) - return 1; + if (rdp->nxtlist) { + /* + * This cpu has pending rcu entries and the grace period + * for them has completed. + */ + if (!rcu_batch_before(rcp->completed, rdp->batch)) + return 1; + if (!rcu_batch_before(rcp->completed, rdp->batch - 1) && + rdp->nxttail[0] != rdp->nxttail[1]) + return 1; + if (rdp->nxttail[0] != &rdp->nxtlist) + return 1; - /* This cpu has no pending entries, but there are new entries */ - if (!rdp->curlist && rdp->nxtlist) - return 1; + /* + * This cpu has pending rcu entries and the new batch + * for then hasn't been started nor scheduled start + */ + if (rcu_batch_after(rdp->batch, rcp->pending)) + return 1; + } /* This cpu has finished callbacks to invoke */ if (rdp->donelist) @@ -506,7 +544,7 @@ int rcu_needs_cpu(int cpu) struct rcu_data *rdp = &per_cpu(rcu_data, cpu); struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu); - return (!!rdp->curlist || !!rdp_bh->curlist || rcu_pending(cpu)); + return !!rdp->nxtlist || !!rdp_bh->nxtlist || rcu_pending(cpu); } void rcu_check_callbacks(int cpu, int user) @@ -553,8 +591,7 @@ static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { memset(rdp, 0, sizeof(*rdp)); - rdp->curtail = &rdp->curlist; - rdp->nxttail = &rdp->nxtlist; + rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2] = &rdp->nxtlist; rdp->donetail = &rdp->donelist; rdp->quiescbatch = rcp->completed; rdp->qs_pending = 0; -- cgit v1.2.3 From 67182ae1c42206e516f7efb292b745e826497b24 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 10 Aug 2008 18:35:38 -0700 Subject: rcu, debug: detect stalled grace periods this is a diagnostic patch for Classic RCU. The approach is to record a timestamp at the beginning of the grace period (in rcu_start_batch()), then have rcu_check_callbacks() complain if: 1. it is running on a CPU that has holding up grace periods for a long time (say one second). This will identify the culprit assuming that the culprit has not disabled hardware irqs, instruction execution, or some such. 2. it is running on a CPU that is not holding up grace periods, but grace periods have been held up for an even longer time (say two seconds). It is enabled via the default-off CONFIG_DEBUG_RCU_STALL kernel parameter. Rather than exponential backoff, it backs off to once per 30 seconds. My feeling upon thinking on it was that if you have stalled RCU grace periods for that long, a few extra printk() messages are probably the least of your worries... Signed-off-by: Paul E. McKenney Cc: Peter Zijlstra Cc: Yinghai Lu Cc: David Witbrodt Signed-off-by: Ingo Molnar --- include/linux/rcuclassic.h | 3 ++ kernel/rcuclassic.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++ lib/Kconfig.debug | 13 ++++++++ 3 files changed, 96 insertions(+) (limited to 'kernel') diff --git a/include/linux/rcuclassic.h b/include/linux/rcuclassic.h index 04c728147be0..16589958b40e 100644 --- a/include/linux/rcuclassic.h +++ b/include/linux/rcuclassic.h @@ -46,6 +46,9 @@ struct rcu_ctrlblk { long cur; /* Current batch number. */ long completed; /* Number of the last completed batch */ long pending; /* Number of the last pending batch */ +#ifdef CONFIG_DEBUG_RCU_STALL + unsigned long gp_check; /* Time grace period should end, in seconds. */ +#endif /* #ifdef CONFIG_DEBUG_RCU_STALL */ int signaled; diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index d4271146a9bd..d7ec731de75c 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -47,6 +47,7 @@ #include #include #include +#include #ifdef CONFIG_DEBUG_LOCK_ALLOC static struct lock_class_key rcu_lock_key; @@ -286,6 +287,81 @@ static void rcu_do_batch(struct rcu_data *rdp) * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace * period (if necessary). */ + +#ifdef CONFIG_DEBUG_RCU_STALL + +static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) +{ + rcp->gp_check = get_seconds() + 3; +} +static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) +{ + int cpu; + long delta; + + /* Only let one CPU complain about others per time interval. */ + + spin_lock(&rcp->lock); + delta = get_seconds() - rcp->gp_check; + if (delta < 2L || + cpus_empty(rcp->cpumask)) { + spin_unlock(&rcp->lock); + return; + rcp->gp_check = get_seconds() + 30; + } + spin_unlock(&rcp->lock); + + /* OK, time to rat on our buddy... */ + + printk(KERN_ERR "RCU detected CPU stalls:"); + for_each_cpu_mask(cpu, rcp->cpumask) + printk(" %d", cpu); + printk(" (detected by %d, t=%lu/%lu)\n", + smp_processor_id(), get_seconds(), rcp->gp_check); +} +static void print_cpu_stall(struct rcu_ctrlblk *rcp) +{ + printk(KERN_ERR "RCU detected CPU %d stall (t=%lu/%lu)\n", + smp_processor_id(), get_seconds(), rcp->gp_check); + dump_stack(); + spin_lock(&rcp->lock); + if ((long)(get_seconds() - rcp->gp_check) >= 0L) + rcp->gp_check = get_seconds() + 30; + spin_unlock(&rcp->lock); +} +static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, + struct rcu_data *rdp) +{ + long delta; + + delta = get_seconds() - rcp->gp_check; + if (cpu_isset(smp_processor_id(), rcp->cpumask) && delta >= 0L) { + + /* We haven't checked in, so go dump stack. */ + + print_cpu_stall(rcp); + + } else if (!cpus_empty(rcp->cpumask) && delta >= 2L) { + + /* They had two seconds to dump stack, so complain. */ + + print_other_cpu_stall(rcp); + + } +} + +#else /* #ifdef CONFIG_DEBUG_RCU_STALL */ + +static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) +{ +} +static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, + struct rcu_data *rdp) +{ +} + +#endif /* #else #ifdef CONFIG_DEBUG_RCU_STALL */ + /* * Register a new batch of callbacks, and start it up if there is currently no * active batch and the batch to be registered has not already occurred. @@ -296,6 +372,7 @@ static void rcu_start_batch(struct rcu_ctrlblk *rcp) if (rcp->cur != rcp->pending && rcp->completed == rcp->cur) { rcp->cur++; + record_gp_check_time(rcp); /* * Accessing nohz_cpu_mask before incrementing rcp->cur needs a @@ -489,6 +566,9 @@ static void rcu_process_callbacks(struct softirq_action *unused) static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + /* Check for CPU stalls, if enabled. */ + check_cpu_stall(rcp, rdp); + if (rdp->nxtlist) { /* * This cpu has pending rcu entries and the grace period diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index e1d4764435ed..2fb6d90bf1e6 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -597,6 +597,19 @@ config RCU_TORTURE_TEST_RUNNABLE Say N here if you want the RCU torture tests to start only after being manually enabled via /proc. +config RCU_CPU_STALL + bool "Check for stalled CPUs delaying RCU grace periods" + depends on CLASSIC_RCU + default n + help + This option causes RCU to printk information on which + CPUs are delaying the current grace period, but only when + the grace period extends for excessive time periods. + + Say Y if you want RCU to perform such checks. + + Say N if you are unsure. + config KPROBES_SANITY_TEST bool "Kprobes sanity tests" depends on DEBUG_KERNEL -- cgit v1.2.3 From 78635fc739b1254f3e0362ac430edbdd2cff01dc Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 11 Aug 2008 13:34:15 +0200 Subject: rcu, debug: detect stalled grace periods, cleanups small cleanups. Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 26 +++++++++++++------------- 1 file changed, 13 insertions(+), 13 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index d7ec731de75c..56b8712fe5aa 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -294,6 +294,7 @@ static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) { rcp->gp_check = get_seconds() + 3; } + static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) { int cpu; @@ -303,11 +304,9 @@ static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) spin_lock(&rcp->lock); delta = get_seconds() - rcp->gp_check; - if (delta < 2L || - cpus_empty(rcp->cpumask)) { + if (delta < 2L || cpus_empty(rcp->cpumask)) { spin_unlock(&rcp->lock); return; - rcp->gp_check = get_seconds() + 30; } spin_unlock(&rcp->lock); @@ -319,6 +318,7 @@ static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) printk(" (detected by %d, t=%lu/%lu)\n", smp_processor_id(), get_seconds(), rcp->gp_check); } + static void print_cpu_stall(struct rcu_ctrlblk *rcp) { printk(KERN_ERR "RCU detected CPU %d stall (t=%lu/%lu)\n", @@ -329,8 +329,8 @@ static void print_cpu_stall(struct rcu_ctrlblk *rcp) rcp->gp_check = get_seconds() + 30; spin_unlock(&rcp->lock); } -static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, - struct rcu_data *rdp) + +static void check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { long delta; @@ -341,12 +341,11 @@ static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, print_cpu_stall(rcp); - } else if (!cpus_empty(rcp->cpumask) && delta >= 2L) { - - /* They had two seconds to dump stack, so complain. */ - - print_other_cpu_stall(rcp); - + } else { + if (!cpus_empty(rcp->cpumask) && delta >= 2L) { + /* They had two seconds to dump stack, so complain. */ + print_other_cpu_stall(rcp); + } } } @@ -355,8 +354,9 @@ static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) { } -static inline void check_cpu_stall(struct rcu_ctrlblk *rcp, - struct rcu_data *rdp) + +static inline void +check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { } -- cgit v1.2.3 From 293a17ebc944c958e24e6ffbd1d5a49abdbf489e Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Tue, 12 Aug 2008 17:25:03 -0700 Subject: rcu: prevent console flood when one CPU sees another AWOL via RCU One small change needed to keep from flooding the console when one CPU notices that another is AWOL. Unless I am missing something subtle. Otherwise the cleanups look good! Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 56b8712fe5aa..dab2676d9d72 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -308,6 +308,7 @@ static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) spin_unlock(&rcp->lock); return; } + rcp->gp_check = get_seconds() + 30; spin_unlock(&rcp->lock); /* OK, time to rat on our buddy... */ -- cgit v1.2.3 From 1f7b94cd3d564901f9e04a8bc5832ae7bfd690a0 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Tue, 5 Aug 2008 09:21:44 -0700 Subject: rcu: classic RCU locking and memory-barrier cleanups This patch simplifies the locking and memory-barrier usage in the Classic RCU grace-period-detection mechanism, incorporating Lai Jiangshan's feedback from the earlier version (http://lkml.org/lkml/2008/8/1/400 and http://lkml.org/lkml/2008/8/3/43). Passed 10 hours of rcutorture concurrent with CPUs being put online and taken offline on a 128-hardware-thread Power machine. My apologies to whoever in the Eastern Hemisphere was planning to use this machine over the Western Hemisphere night, but it was sitting idle and... So this is ready for tip/core/rcu. This patch is in preparation for moving to a hierarchical algorithm to allow the very large SMP machines -- requested by some people at OLS, and there seem to have been a few recent patches in the 4096-CPU direction as well. The general idea is to move to a much more conservative concurrency design, then apply a hierarchy to reduce contention on the global lock by a few orders of magnitude (larger machines would see greater reductions). The reason for taking a conservative approach is that this code isn't on any fast path. Prototype in progress. This patch is against the linux-tip git tree (tip/core/rcu). If you wish to test this against 2.6.26, use the following set of patches: http://www.rdrop.com/users/paulmck/patches/2.6.26-ljsimp-1.patch http://www.rdrop.com/users/paulmck/patches/2.6.26-ljsimpfix-3.patch The first patch combines commits 5127bed588a2f8f3a1f732de2a8a190b7df5dce3 and 3cac97cbb14aed00d83eb33d4613b0fe3aaea863 from Lai Jiangshan , and the second patch contains my changes. Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 51 +++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 41 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index dab2676d9d72..5de126630b10 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -87,6 +87,7 @@ static void force_quiescent_state(struct rcu_data *rdp, int cpu; cpumask_t cpumask; set_need_resched(); + spin_lock(&rcp->lock); if (unlikely(!rcp->signaled)) { rcp->signaled = 1; /* @@ -112,6 +113,7 @@ static void force_quiescent_state(struct rcu_data *rdp, for_each_cpu_mask_nr(cpu, cpumask) smp_send_reschedule(cpu); } + spin_unlock(&rcp->lock); } #else static inline void force_quiescent_state(struct rcu_data *rdp, @@ -125,7 +127,9 @@ static void __call_rcu(struct rcu_head *head, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { long batch; - smp_mb(); /* reads the most recently updated value of rcu->cur. */ + + head->next = NULL; + smp_mb(); /* Read of rcu->cur must happen after any change by caller. */ /* * Determine the batch number of this callback. @@ -175,7 +179,6 @@ void call_rcu(struct rcu_head *head, unsigned long flags; head->func = func; - head->next = NULL; local_irq_save(flags); __call_rcu(head, &rcu_ctrlblk, &__get_cpu_var(rcu_data)); local_irq_restore(flags); @@ -204,7 +207,6 @@ void call_rcu_bh(struct rcu_head *head, unsigned long flags; head->func = func; - head->next = NULL; local_irq_save(flags); __call_rcu(head, &rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); local_irq_restore(flags); @@ -467,17 +469,17 @@ static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, static void __rcu_offline_cpu(struct rcu_data *this_rdp, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { - /* if the cpu going offline owns the grace period + /* + * if the cpu going offline owns the grace period * we can block indefinitely waiting for it, so flush * it here */ spin_lock_bh(&rcp->lock); if (rcp->cur != rcp->completed) cpu_quiet(rdp->cpu, rcp); - spin_unlock_bh(&rcp->lock); - /* spin_lock implies smp_mb() */ rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail, rcp->cur + 1); rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail[2], rcp->cur + 1); + spin_unlock_bh(&rcp->lock); local_irq_disable(); this_rdp->qlen += rdp->qlen; @@ -511,16 +513,19 @@ static void rcu_offline_cpu(int cpu) static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + long completed_snap; + if (rdp->nxtlist) { local_irq_disable(); + completed_snap = ACCESS_ONCE(rcp->completed); /* * move the other grace-period-completed entries to * [rdp->nxtlist, *rdp->nxttail[0]) temporarily */ - if (!rcu_batch_before(rcp->completed, rdp->batch)) + if (!rcu_batch_before(completed_snap, rdp->batch)) rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2]; - else if (!rcu_batch_before(rcp->completed, rdp->batch - 1)) + else if (!rcu_batch_before(completed_snap, rdp->batch - 1)) rdp->nxttail[0] = rdp->nxttail[1]; /* @@ -561,8 +566,24 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, static void rcu_process_callbacks(struct softirq_action *unused) { + /* + * Memory references from any prior RCU read-side critical sections + * executed by the interrupted code must be see before any RCU + * grace-period manupulations below. + */ + + smp_mb(); /* See above block comment. */ + __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); + + /* + * Memory references from any later RCU read-side critical sections + * executed by the interrupted code must be see after any RCU + * grace-period manupulations above. + */ + + smp_mb(); /* See above block comment. */ } static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) @@ -571,13 +592,15 @@ static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) check_cpu_stall(rcp, rdp); if (rdp->nxtlist) { + long completed_snap = ACCESS_ONCE(rcp->completed); + /* * This cpu has pending rcu entries and the grace period * for them has completed. */ - if (!rcu_batch_before(rcp->completed, rdp->batch)) + if (!rcu_batch_before(completed_snap, rdp->batch)) return 1; - if (!rcu_batch_before(rcp->completed, rdp->batch - 1) && + if (!rcu_batch_before(completed_snap, rdp->batch - 1) && rdp->nxttail[0] != rdp->nxttail[1]) return 1; if (rdp->nxttail[0] != &rdp->nxtlist) @@ -628,6 +651,12 @@ int rcu_needs_cpu(int cpu) return !!rdp->nxtlist || !!rdp_bh->nxtlist || rcu_pending(cpu); } +/* + * Top-level function driving RCU grace-period detection, normally + * invoked from the scheduler-clock interrupt. This function simply + * increments counters that are read only from softirq by this same + * CPU, so there are no memory barriers required. + */ void rcu_check_callbacks(int cpu, int user) { if (user || @@ -671,6 +700,7 @@ void rcu_check_callbacks(int cpu, int user) static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + spin_lock(&rcp->lock); memset(rdp, 0, sizeof(*rdp)); rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2] = &rdp->nxtlist; rdp->donetail = &rdp->donelist; @@ -678,6 +708,7 @@ static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, rdp->qs_pending = 0; rdp->cpu = cpu; rdp->blimit = blimit; + spin_unlock(&rcp->lock); } static void __cpuinit rcu_online_cpu(int cpu) -- cgit v1.2.3 From 5802294f1b1895ee19a3d0ae72805da453afb9de Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Wed, 30 Jul 2008 14:20:55 -0400 Subject: rcu: trace fix possible mem-leak In the initialization of the RCU trace module, if rcupreempt_debugfs_init() fails, we never free the the trace buffer. This patch frees the trace buffer in case the debugfs fails. Signed-off-by: Steven Rostedt Reviewed-by: "Paul E. McKenney" Signed-off-by: Ingo Molnar --- kernel/rcupreempt_trace.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/rcupreempt_trace.c b/kernel/rcupreempt_trace.c index 5edf82c34bbc..35c2d3360ecf 100644 --- a/kernel/rcupreempt_trace.c +++ b/kernel/rcupreempt_trace.c @@ -308,11 +308,16 @@ out: static int __init rcupreempt_trace_init(void) { + int ret; + mutex_init(&rcupreempt_trace_mutex); rcupreempt_trace_buf = kmalloc(RCUPREEMPT_TRACE_BUF_SIZE, GFP_KERNEL); if (!rcupreempt_trace_buf) return 1; - return rcupreempt_debugfs_init(); + ret = rcupreempt_debugfs_init(); + if (ret) + kfree(rcupreempt_trace_buf); + return ret; } static void __exit rcupreempt_trace_cleanup(void) -- cgit v1.2.3 From cd95851785bcfe95fdf73689e8ecb5a1c5959231 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 17 Aug 2008 07:37:15 -0700 Subject: rcu: fix classic RCU locking cleanup lockdep problem On Fri, Aug 15, 2008 at 04:24:30PM +0200, Ingo Molnar wrote: > > Paul, > > one of your two recent RCU patches caused this lockdep splat in -tip > testing: > > -------------------> > Brought up 2 CPUs > Total of 2 processors activated (6850.87 BogoMIPS). > PM: Adding info for No Bus:platform > khelper used greatest stack depth: 3124 bytes left > > ================================= > [ INFO: inconsistent lock state ] > 2.6.27-rc3-tip #1 > --------------------------------- > inconsistent {softirq-on-W} -> {in-softirq-W} usage. > ksoftirqd/0/4 [HC0[0]:SC1[1]:HE1:SE0] takes: > (&rcu_ctrlblk.lock){-+..}, at: [] __rcu_process_callbacks+0x1ac/0x1f0 > {softirq-on-W} state was registered at: > [] __lock_acquire+0x3f4/0x5b0 > [] lock_acquire+0x89/0xc0 > [] _spin_lock+0x3b/0x70 > [] rcu_init_percpu_data+0x29/0x80 > [] rcu_cpu_notify+0xaf/0xd0 > [] notifier_call_chain+0x2d/0x60 > [] __raw_notifier_call_chain+0x1e/0x30 > [] _cpu_up+0x79/0x110 > [] cpu_up+0x4d/0x70 > [] kernel_init+0xb1/0x200 > [] kernel_thread_helper+0x7/0x10 > [] 0xffffffff > irq event stamp: 14 > hardirqs last enabled at (14): [] trace_hardirqs_on+0xb/0x10 > hardirqs last disabled at (13): [] trace_hardirqs_off+0xb/0x10 > softirqs last enabled at (0): [] copy_process+0x276/0x1190 > softirqs last disabled at (11): [] call_on_stack+0x1a/0x30 > > other info that might help us debug this: > no locks held by ksoftirqd/0/4. > > stack backtrace: > Pid: 4, comm: ksoftirqd/0 Not tainted 2.6.27-rc3-tip #1 > [] print_usage_bug+0x16c/0x1b0 > [] mark_lock+0xa75/0xb10 > [] ? sched_clock+0x15/0x30 > [] __lock_acquire+0x3ad/0x5b0 > [] lock_acquire+0x89/0xc0 > [] ? __rcu_process_callbacks+0x1ac/0x1f0 > [] _spin_lock+0x3b/0x70 > [] ? __rcu_process_callbacks+0x1ac/0x1f0 > [] __rcu_process_callbacks+0x1ac/0x1f0 > [] rcu_process_callbacks+0x26/0x50 > [] __do_softirq+0x95/0x120 > [] ? __do_softirq+0x0/0x120 > [] call_on_stack+0x1a/0x30 > [] ? ksoftirqd+0x96/0x110 > [] ? ksoftirqd+0x0/0x110 > [] ? kthread+0x47/0x80 > [] ? kthread+0x0/0x80 > [] ? kernel_thread_helper+0x7/0x10 > ======================= > calling init_cpufreq_transition_notifier_list+0x0/0x20 > initcall init_cpufreq_transition_notifier_list+0x0/0x20 returned 0 after 0 msecs > calling net_ns_init+0x0/0x190 > net_namespace: 676 bytes > initcall net_ns_init+0x0/0x190 returned 0 after 0 msecs > calling cpufreq_tsc+0x0/0x20 > initcall cpufreq_tsc+0x0/0x20 returned 0 after 0 msecs > calling reboot_init+0x0/0x20 > initcall reboot_init+0x0/0x20 returned 0 after 0 msecs > calling print_banner+0x0/0x10 > Booting paravirtualized kernel on bare hardware > > <----------------------- > > my guess is on: > > commit 1f7b94cd3d564901f9e04a8bc5832ae7bfd690a0 > Author: Paul E. McKenney > Date: Tue Aug 5 09:21:44 2008 -0700 > > rcu: classic RCU locking and memory-barrier cleanups > > Ingo Fixes a problem detected by lockdep in which rcu->lock was acquired both in irq context and in process context, but without disabling from process context. Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 5de126630b10..fb1f1cc45142 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -700,7 +700,9 @@ void rcu_check_callbacks(int cpu, int user) static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { - spin_lock(&rcp->lock); + long flags; + + spin_lock_irqsave(&rcp->lock, flags); memset(rdp, 0, sizeof(*rdp)); rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2] = &rdp->nxtlist; rdp->donetail = &rdp->donelist; @@ -708,7 +710,7 @@ static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, rdp->qs_pending = 0; rdp->cpu = cpu; rdp->blimit = blimit; - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); } static void __cpuinit rcu_online_cpu(int cpu) -- cgit v1.2.3 From ded00a56e99555c3f4000ef3eebfd5fe0d574565 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 17 Aug 2008 12:50:36 -0700 Subject: rcu: remove redundant ACCESS_ONCE definition from rcupreempt.c Remove the redundant definition of ACCESS_ONCE() from rcupreempt.c in favor of the one in compiler.h. Also merge the comment header from rcupreempt.c's definition into that in compiler.h. Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- include/linux/compiler.h | 4 +++- kernel/rcupreempt.c | 8 -------- 2 files changed, 3 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/include/linux/compiler.h b/include/linux/compiler.h index c8bd2daf95ec..8322141ee480 100644 --- a/include/linux/compiler.h +++ b/include/linux/compiler.h @@ -190,7 +190,9 @@ extern void __chk_io_ptr(const volatile void __iomem *); * ACCESS_ONCE() in different C statements. * * This macro does absolutely -nothing- to prevent the CPU from reordering, - * merging, or refetching absolutely anything at any time. + * merging, or refetching absolutely anything at any time. Its main intended + * use is to mediate communication between process-level code and irq/NMI + * handlers, all running on the same CPU. */ #define ACCESS_ONCE(x) (*(volatile typeof(x) *)&(x)) diff --git a/kernel/rcupreempt.c b/kernel/rcupreempt.c index 27827931ca0d..ca4bbbe04aa4 100644 --- a/kernel/rcupreempt.c +++ b/kernel/rcupreempt.c @@ -58,14 +58,6 @@ #include #include -/* - * Macro that prevents the compiler from reordering accesses, but does - * absolutely -nothing- to prevent CPUs from reordering. This is used - * only to mediate communication between mainline code and hardware - * interrupt and NMI handlers. - */ -#define ACCESS_ONCE(x) (*(volatile typeof(x) *)&(x)) - /* * PREEMPT_RCU data structures. */ -- cgit v1.2.3 From eff9b713ee3540ddab862095aaf4b1511a6758bc Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Mon, 18 Aug 2008 17:51:08 -0700 Subject: rcu: fix locking cleanup fallout Given that the rcp->lock is now acquired from call_rcu(), which can be invoked from irq-disable regions, all acquisitions need to disable irqs. The following patch fixes this. Although I don't have any reason to believe that this is the cause of Yinghai's oops, it does need to be fixed. Signed-off-by: Paul E. McKenney Cc: Yinghai Lu Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 36 ++++++++++++++++++++++-------------- 1 file changed, 22 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index fb1f1cc45142..c6b6cf55f3e2 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -86,8 +86,10 @@ static void force_quiescent_state(struct rcu_data *rdp, { int cpu; cpumask_t cpumask; + unsigned long flags; + set_need_resched(); - spin_lock(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); if (unlikely(!rcp->signaled)) { rcp->signaled = 1; /* @@ -113,7 +115,7 @@ static void force_quiescent_state(struct rcu_data *rdp, for_each_cpu_mask_nr(cpu, cpumask) smp_send_reschedule(cpu); } - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); } #else static inline void force_quiescent_state(struct rcu_data *rdp, @@ -301,17 +303,18 @@ static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) { int cpu; long delta; + unsigned long flags; /* Only let one CPU complain about others per time interval. */ - spin_lock(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); delta = get_seconds() - rcp->gp_check; if (delta < 2L || cpus_empty(rcp->cpumask)) { spin_unlock(&rcp->lock); return; } rcp->gp_check = get_seconds() + 30; - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); /* OK, time to rat on our buddy... */ @@ -324,13 +327,15 @@ static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) static void print_cpu_stall(struct rcu_ctrlblk *rcp) { + unsigned long flags; + printk(KERN_ERR "RCU detected CPU %d stall (t=%lu/%lu)\n", smp_processor_id(), get_seconds(), rcp->gp_check); dump_stack(); - spin_lock(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); if ((long)(get_seconds() - rcp->gp_check) >= 0L) rcp->gp_check = get_seconds() + 30; - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); } static void check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) @@ -413,6 +418,8 @@ static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + unsigned long flags; + if (rdp->quiescbatch != rcp->cur) { /* start new grace period: */ rdp->qs_pending = 1; @@ -436,7 +443,7 @@ static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, return; rdp->qs_pending = 0; - spin_lock(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); /* * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync * during cpu startup. Ignore the quiescent state. @@ -444,7 +451,7 @@ static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, if (likely(rdp->quiescbatch == rcp->cur)) cpu_quiet(rdp->cpu, rcp); - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); } @@ -469,21 +476,22 @@ static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, static void __rcu_offline_cpu(struct rcu_data *this_rdp, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + unsigned long flags; + /* * if the cpu going offline owns the grace period * we can block indefinitely waiting for it, so flush * it here */ - spin_lock_bh(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); if (rcp->cur != rcp->completed) cpu_quiet(rdp->cpu, rcp); rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail, rcp->cur + 1); rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail[2], rcp->cur + 1); - spin_unlock_bh(&rcp->lock); + spin_unlock(&rcp->lock); - local_irq_disable(); this_rdp->qlen += rdp->qlen; - local_irq_enable(); + local_irq_restore(flags); } static void rcu_offline_cpu(int cpu) @@ -550,12 +558,12 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, if (rcu_batch_after(rdp->batch, rcp->pending)) { /* and start it/schedule start if it's a new batch */ - spin_lock(&rcp->lock); + spin_lock_irqsave(&rcp->lock, flags); if (rcu_batch_after(rdp->batch, rcp->pending)) { rcp->pending = rdp->batch; rcu_start_batch(rcp); } - spin_unlock(&rcp->lock); + spin_unlock_irqrestore(&rcp->lock, flags); } } -- cgit v1.2.3 From 0c925d79234fe77589d8ff3861f9f8bb9e7fc3f6 Mon Sep 17 00:00:00 2001 From: Hiroshi Shimamoto Date: Mon, 18 Aug 2008 21:49:51 -0700 Subject: rcuclassic: fix compilation NG fix: CC kernel/rcuclassic.o kernel/rcuclassic.c: In function '__rcu_process_callbacks': kernel/rcuclassic.c:561: error: 'flags' undeclared (first use in this function) kernel/rcuclassic.c:561: error: (Each undeclared identifier is reported only once kernel/rcuclassic.c:561: error: for each function it appears in.) Declare missing variable flags. Signed-off-by: Hiroshi Shimamoto Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index c6b6cf55f3e2..01e761a6b38c 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -557,6 +557,8 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, local_irq_enable(); if (rcu_batch_after(rdp->batch, rcp->pending)) { + unsigned long flags; + /* and start it/schedule start if it's a new batch */ spin_lock_irqsave(&rcp->lock, flags); if (rcu_batch_after(rdp->batch, rcp->pending)) { -- cgit v1.2.3 From 275a89bdd3868af3008852594d2e169eaf69441b Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Thu, 21 Aug 2008 06:14:55 -0700 Subject: rcu: use irq-safe locks Some earlier tip/core/rcu patches caused RCU to incorrectly enable irqs too early in boot. This caused Yinghai's repeated-kexec testing to hit oopses, presumably due to so that device interrupts left over from the prior kernel instance (which would oops the newly booting kernel before it got a chance to reset said devices). This patch therefore converts all the local_irq_disable()s in rcuclassic.c to local_irq_save(). Besides, I never did like local_irq_disable() anyway. ;-) Signed-off-by: Paul E. McKenney Signed-off-by: Yinghai Lu Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 01e761a6b38c..3f6918966bda 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -247,6 +247,7 @@ static inline void raise_rcu_softirq(void) */ static void rcu_do_batch(struct rcu_data *rdp) { + unsigned long flags; struct rcu_head *next, *list; int count = 0; @@ -261,9 +262,9 @@ static void rcu_do_batch(struct rcu_data *rdp) } rdp->donelist = list; - local_irq_disable(); + local_irq_save(flags); rdp->qlen -= count; - local_irq_enable(); + local_irq_restore(flags); if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) rdp->blimit = blimit; @@ -464,12 +465,14 @@ static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, struct rcu_head **tail, long batch) { + unsigned long flags; + if (list) { - local_irq_disable(); + local_irq_save(flags); this_rdp->batch = batch; *this_rdp->nxttail[2] = list; this_rdp->nxttail[2] = tail; - local_irq_enable(); + local_irq_restore(flags); } } @@ -521,10 +524,11 @@ static void rcu_offline_cpu(int cpu) static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { + unsigned long flags; long completed_snap; if (rdp->nxtlist) { - local_irq_disable(); + local_irq_save(flags); completed_snap = ACCESS_ONCE(rcp->completed); /* @@ -554,7 +558,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, rdp->nxttail[0] = &rdp->nxtlist; } - local_irq_enable(); + local_irq_restore(flags); if (rcu_batch_after(rdp->batch, rcp->pending)) { unsigned long flags; -- cgit v1.2.3 From 0cd418ddb1ee88df7d16d5df06cb2da68eceb9e4 Mon Sep 17 00:00:00 2001 From: Hiroshi Shimamoto Date: Mon, 18 Aug 2008 14:39:21 -0700 Subject: rcuclassic: fix compiler warning CC kernel/rcuclassic.o kernel/rcuclassic.c: In function 'rcu_init_percpu_data': kernel/rcuclassic.c:705: warning: comparison of distinct pointer types lacks a cast kernel/rcuclassic.c:713: warning: comparison of distinct pointer types lacks a cast flags should be unsigned long. Signed-off-by: Hiroshi Shimamoto Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 3f6918966bda..743cf0550ff4 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -714,7 +714,7 @@ void rcu_check_callbacks(int cpu, int user) static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { - long flags; + unsigned long flags; spin_lock_irqsave(&rcp->lock, flags); memset(rdp, 0, sizeof(*rdp)); -- cgit v1.2.3 From 3a72dc8eb5a7122fff439a22bd22486a4fff505c Mon Sep 17 00:00:00 2001 From: Harvey Harrison Date: Mon, 22 Sep 2008 14:55:46 -0700 Subject: rcu: fix sparse shadowed variable warning kernel/rcuclassic.c:564:18: warning: symbol 'flags' shadows an earlier one kernel/rcuclassic.c:527:16: originally declared here Signed-off-by: Harvey Harrison Acked-by: Paul E. McKenney Signed-off-by: Andrew Morton Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 743cf0550ff4..ed15128ca2c9 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -561,15 +561,15 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, local_irq_restore(flags); if (rcu_batch_after(rdp->batch, rcp->pending)) { - unsigned long flags; + unsigned long flags2; /* and start it/schedule start if it's a new batch */ - spin_lock_irqsave(&rcp->lock, flags); + spin_lock_irqsave(&rcp->lock, flags2); if (rcu_batch_after(rdp->batch, rcp->pending)) { rcp->pending = rdp->batch; rcu_start_batch(rcp); } - spin_unlock_irqrestore(&rcp->lock, flags); + spin_unlock_irqrestore(&rcp->lock, flags2); } } -- cgit v1.2.3 From 2133b5d7ff531bc15a923db4a6a50bf96c561be9 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Thu, 2 Oct 2008 16:06:39 -0700 Subject: rcu: RCU-based detection of stalled CPUs for Classic RCU This patch adds stalled-CPU detection to Classic RCU. This capability is enabled by a new config variable CONFIG_RCU_CPU_STALL_DETECTOR, which defaults disabled. This is a debugging feature to detect infinite loops in kernel code, not something that non-kernel-hackers would be expected to care about. This feature can detect looping CPUs in !PREEMPT builds and looping CPUs with preemption disabled in PREEMPT builds. This is essentially a port of this functionality from the treercu patch, replacing the stall debug patch that is already in tip/core/rcu (commit 67182ae1c4). The changes from the patch in tip/core/rcu include making the config variable name match that in treercu, changing from seconds to jiffies to avoid spurious warnings, and printing a boot message when this feature is enabled. Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- include/linux/rcuclassic.h | 12 +++- kernel/rcuclassic.c | 166 +++++++++++++++++++++++---------------------- lib/Kconfig.debug | 2 +- 3 files changed, 96 insertions(+), 84 deletions(-) (limited to 'kernel') diff --git a/include/linux/rcuclassic.h b/include/linux/rcuclassic.h index 29bf528c7dcc..5f89b62e6983 100644 --- a/include/linux/rcuclassic.h +++ b/include/linux/rcuclassic.h @@ -40,15 +40,21 @@ #include #include +#ifdef CONFIG_RCU_CPU_STALL_DETECTOR +#define RCU_SECONDS_TILL_STALL_CHECK ( 3 * HZ) /* for rcp->jiffies_stall */ +#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rcp->jiffies_stall */ +#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ /* Global control variables for rcupdate callback mechanism. */ struct rcu_ctrlblk { long cur; /* Current batch number. */ long completed; /* Number of the last completed batch */ long pending; /* Number of the last pending batch */ -#ifdef CONFIG_DEBUG_RCU_STALL - unsigned long gp_check; /* Time grace period should end, in seconds. */ -#endif /* #ifdef CONFIG_DEBUG_RCU_STALL */ +#ifdef CONFIG_RCU_CPU_STALL_DETECTOR + unsigned long gp_start; /* Time at which GP started in jiffies. */ + unsigned long jiffies_stall; + /* Time at which to check for CPU stalls. */ +#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ int signaled; diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index ed15128ca2c9..0d07e6e51578 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -164,6 +164,87 @@ static void __call_rcu(struct rcu_head *head, struct rcu_ctrlblk *rcp, } } +#ifdef CONFIG_RCU_CPU_STALL_DETECTOR + +static void record_gp_stall_check_time(struct rcu_ctrlblk *rcp) +{ + rcp->gp_start = jiffies; + rcp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; +} + +static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) +{ + int cpu; + long delta; + unsigned long flags; + + /* Only let one CPU complain about others per time interval. */ + + spin_lock_irqsave(&rcp->lock, flags); + delta = jiffies - rcp->jiffies_stall; + if (delta < 2 || rcp->cur != rcp->completed) { + spin_unlock_irqrestore(&rcp->lock, flags); + return; + } + rcp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; + spin_unlock_irqrestore(&rcp->lock, flags); + + /* OK, time to rat on our buddy... */ + + printk(KERN_ERR "RCU detected CPU stalls:"); + for_each_possible_cpu(cpu) { + if (cpu_isset(cpu, rcp->cpumask)) + printk(" %d", cpu); + } + printk(" (detected by %d, t=%ld jiffies)\n", + smp_processor_id(), (long)(jiffies - rcp->gp_start)); +} + +static void print_cpu_stall(struct rcu_ctrlblk *rcp) +{ + unsigned long flags; + + printk(KERN_ERR "RCU detected CPU %d stall (t=%lu/%lu jiffies)\n", + smp_processor_id(), jiffies, + jiffies - rcp->gp_start); + dump_stack(); + spin_lock_irqsave(&rcp->lock, flags); + if ((long)(jiffies - rcp->jiffies_stall) >= 0) + rcp->jiffies_stall = + jiffies + RCU_SECONDS_TILL_STALL_RECHECK; + spin_unlock_irqrestore(&rcp->lock, flags); + set_need_resched(); /* kick ourselves to get things going. */ +} + +static void check_cpu_stall(struct rcu_ctrlblk *rcp) +{ + long delta; + + delta = jiffies - rcp->jiffies_stall; + if (cpu_isset(smp_processor_id(), rcp->cpumask) && delta >= 0) { + + /* We haven't checked in, so go dump stack. */ + print_cpu_stall(rcp); + + } else if (rcp->cur != rcp->completed && delta >= 2) { + + /* They had two seconds to dump stack, so complain. */ + print_other_cpu_stall(rcp); + } +} + +#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ + +static void record_gp_stall_check_time(struct rcu_ctrlblk *rcp) +{ +} + +static void check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) +{ +} + +#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ + /** * call_rcu - Queue an RCU callback for invocation after a grace period. * @head: structure to be used for queueing the RCU updates. @@ -293,84 +374,6 @@ static void rcu_do_batch(struct rcu_data *rdp) * period (if necessary). */ -#ifdef CONFIG_DEBUG_RCU_STALL - -static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) -{ - rcp->gp_check = get_seconds() + 3; -} - -static void print_other_cpu_stall(struct rcu_ctrlblk *rcp) -{ - int cpu; - long delta; - unsigned long flags; - - /* Only let one CPU complain about others per time interval. */ - - spin_lock_irqsave(&rcp->lock, flags); - delta = get_seconds() - rcp->gp_check; - if (delta < 2L || cpus_empty(rcp->cpumask)) { - spin_unlock(&rcp->lock); - return; - } - rcp->gp_check = get_seconds() + 30; - spin_unlock_irqrestore(&rcp->lock, flags); - - /* OK, time to rat on our buddy... */ - - printk(KERN_ERR "RCU detected CPU stalls:"); - for_each_cpu_mask(cpu, rcp->cpumask) - printk(" %d", cpu); - printk(" (detected by %d, t=%lu/%lu)\n", - smp_processor_id(), get_seconds(), rcp->gp_check); -} - -static void print_cpu_stall(struct rcu_ctrlblk *rcp) -{ - unsigned long flags; - - printk(KERN_ERR "RCU detected CPU %d stall (t=%lu/%lu)\n", - smp_processor_id(), get_seconds(), rcp->gp_check); - dump_stack(); - spin_lock_irqsave(&rcp->lock, flags); - if ((long)(get_seconds() - rcp->gp_check) >= 0L) - rcp->gp_check = get_seconds() + 30; - spin_unlock_irqrestore(&rcp->lock, flags); -} - -static void check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) -{ - long delta; - - delta = get_seconds() - rcp->gp_check; - if (cpu_isset(smp_processor_id(), rcp->cpumask) && delta >= 0L) { - - /* We haven't checked in, so go dump stack. */ - - print_cpu_stall(rcp); - - } else { - if (!cpus_empty(rcp->cpumask) && delta >= 2L) { - /* They had two seconds to dump stack, so complain. */ - print_other_cpu_stall(rcp); - } - } -} - -#else /* #ifdef CONFIG_DEBUG_RCU_STALL */ - -static inline void record_gp_check_time(struct rcu_ctrlblk *rcp) -{ -} - -static inline void -check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) -{ -} - -#endif /* #else #ifdef CONFIG_DEBUG_RCU_STALL */ - /* * Register a new batch of callbacks, and start it up if there is currently no * active batch and the batch to be registered has not already occurred. @@ -381,7 +384,7 @@ static void rcu_start_batch(struct rcu_ctrlblk *rcp) if (rcp->cur != rcp->pending && rcp->completed == rcp->cur) { rcp->cur++; - record_gp_check_time(rcp); + record_gp_stall_check_time(rcp); /* * Accessing nohz_cpu_mask before incrementing rcp->cur needs a @@ -603,7 +606,7 @@ static void rcu_process_callbacks(struct softirq_action *unused) static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) { /* Check for CPU stalls, if enabled. */ - check_cpu_stall(rcp, rdp); + check_cpu_stall(rcp); if (rdp->nxtlist) { long completed_snap = ACCESS_ONCE(rcp->completed); @@ -769,6 +772,9 @@ static struct notifier_block __cpuinitdata rcu_nb = { */ void __init __rcu_init(void) { +#ifdef CONFIG_RCU_CPU_STALL_DETECTOR + printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); +#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long)smp_processor_id()); /* Register notifier for non-boot CPUs */ diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index ccede1aeab38..9fee969dd60e 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -597,7 +597,7 @@ config RCU_TORTURE_TEST_RUNNABLE Say N here if you want the RCU torture tests to start only after being manually enabled via /proc. -config RCU_CPU_STALL +config RCU_CPU_STALL_DETECTOR bool "Check for stalled CPUs delaying RCU grace periods" depends on CLASSIC_RCU default n -- cgit v1.2.3 From 2ec2b482b10a1ed3493c224f1893cddd3d33833b Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 3 Oct 2008 10:41:00 +0200 Subject: rcu: RCU-based detection of stalled CPUs for Classic RCU, fix fix the !CONFIG_RCU_CPU_STALL_DETECTOR path: kernel/rcuclassic.c: In function '__rcu_pending': kernel/rcuclassic.c:609: error: too few arguments to function 'check_cpu_stall' Signed-off-by: Ingo Molnar --- kernel/rcuclassic.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c index 0d07e6e51578..37f72e551542 100644 --- a/kernel/rcuclassic.c +++ b/kernel/rcuclassic.c @@ -239,7 +239,7 @@ static void record_gp_stall_check_time(struct rcu_ctrlblk *rcp) { } -static void check_cpu_stall(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) +static inline void check_cpu_stall(struct rcu_ctrlblk *rcp) { } -- cgit v1.2.3