From 095777c417db142970adeb776fa0cb10810b8122 Mon Sep 17 00:00:00 2001 From: Davidlohr Bueso Date: Wed, 22 Jul 2015 14:07:27 -0700 Subject: locktorture: Support rtmutex torturing Real time mutexes is one of the few general primitives that we do not have in locktorture. Address this -- a few considerations: o To spice things up, enable competing thread(s) to become rt, such that we can stress different prio boosting paths in the rtmutex code. Introduce a ->task_boost callback, only used by rtmutex-torturer. Tasks will boost/deboost around every 50k (arbitrarily) lock/unlock operations. o Hold times are similar to what we have for other locks: only occasionally having longer hold times (per ~200k ops). So we roughly do two full rt boost+deboosting ops with short hold times. Signed-off-by: Davidlohr Bueso Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/locktorture.c | 114 ++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 112 insertions(+), 2 deletions(-) (limited to 'kernel/locking') diff --git a/kernel/locking/locktorture.c b/kernel/locking/locktorture.c index 32244186f1f2..e1ca7a2fae91 100644 --- a/kernel/locking/locktorture.c +++ b/kernel/locking/locktorture.c @@ -17,12 +17,14 @@ * * Copyright (C) IBM Corporation, 2014 * - * Author: Paul E. McKenney + * Authors: Paul E. McKenney + * Davidlohr Bueso * Based on kernel/rcu/torture.c. */ #include #include #include +#include #include #include #include @@ -91,11 +93,13 @@ struct lock_torture_ops { void (*init)(void); int (*writelock)(void); void (*write_delay)(struct torture_random_state *trsp); + void (*task_boost)(struct torture_random_state *trsp); void (*writeunlock)(void); int (*readlock)(void); void (*read_delay)(struct torture_random_state *trsp); void (*readunlock)(void); - unsigned long flags; + + unsigned long flags; /* for irq spinlocks */ const char *name; }; @@ -139,9 +143,15 @@ static void torture_lock_busted_write_unlock(void) /* BUGGY, do not use in real life!!! */ } +static void torture_boost_dummy(struct torture_random_state *trsp) +{ + /* Only rtmutexes care about priority */ +} + static struct lock_torture_ops lock_busted_ops = { .writelock = torture_lock_busted_write_lock, .write_delay = torture_lock_busted_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_lock_busted_write_unlock, .readlock = NULL, .read_delay = NULL, @@ -185,6 +195,7 @@ static void torture_spin_lock_write_unlock(void) __releases(torture_spinlock) static struct lock_torture_ops spin_lock_ops = { .writelock = torture_spin_lock_write_lock, .write_delay = torture_spin_lock_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_spin_lock_write_unlock, .readlock = NULL, .read_delay = NULL, @@ -211,6 +222,7 @@ __releases(torture_spinlock) static struct lock_torture_ops spin_lock_irq_ops = { .writelock = torture_spin_lock_write_lock_irq, .write_delay = torture_spin_lock_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_lock_spin_write_unlock_irq, .readlock = NULL, .read_delay = NULL, @@ -275,6 +287,7 @@ static void torture_rwlock_read_unlock(void) __releases(torture_rwlock) static struct lock_torture_ops rw_lock_ops = { .writelock = torture_rwlock_write_lock, .write_delay = torture_rwlock_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_rwlock_write_unlock, .readlock = torture_rwlock_read_lock, .read_delay = torture_rwlock_read_delay, @@ -315,6 +328,7 @@ __releases(torture_rwlock) static struct lock_torture_ops rw_lock_irq_ops = { .writelock = torture_rwlock_write_lock_irq, .write_delay = torture_rwlock_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_rwlock_write_unlock_irq, .readlock = torture_rwlock_read_lock_irq, .read_delay = torture_rwlock_read_delay, @@ -354,6 +368,7 @@ static void torture_mutex_unlock(void) __releases(torture_mutex) static struct lock_torture_ops mutex_lock_ops = { .writelock = torture_mutex_lock, .write_delay = torture_mutex_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_mutex_unlock, .readlock = NULL, .read_delay = NULL, @@ -361,6 +376,90 @@ static struct lock_torture_ops mutex_lock_ops = { .name = "mutex_lock" }; +#ifdef CONFIG_RT_MUTEXES +static DEFINE_RT_MUTEX(torture_rtmutex); + +static int torture_rtmutex_lock(void) __acquires(torture_rtmutex) +{ + rt_mutex_lock(&torture_rtmutex); + return 0; +} + +static void torture_rtmutex_boost(struct torture_random_state *trsp) +{ + int policy; + struct sched_param param; + const unsigned int factor = 50000; /* yes, quite arbitrary */ + + if (!rt_task(current)) { + /* + * (1) Boost priority once every ~50k operations. When the + * task tries to take the lock, the rtmutex it will account + * for the new priority, and do any corresponding pi-dance. + */ + if (!(torture_random(trsp) % + (cxt.nrealwriters_stress * factor))) { + policy = SCHED_FIFO; + param.sched_priority = MAX_RT_PRIO - 1; + } else /* common case, do nothing */ + return; + } else { + /* + * The task will remain boosted for another ~500k operations, + * then restored back to its original prio, and so forth. + * + * When @trsp is nil, we want to force-reset the task for + * stopping the kthread. + */ + if (!trsp || !(torture_random(trsp) % + (cxt.nrealwriters_stress * factor * 2))) { + policy = SCHED_NORMAL; + param.sched_priority = 0; + } else /* common case, do nothing */ + return; + } + + sched_setscheduler_nocheck(current, policy, ¶m); +} + +static void torture_rtmutex_delay(struct torture_random_state *trsp) +{ + const unsigned long shortdelay_us = 2; + const unsigned long longdelay_ms = 100; + + /* + * We want a short delay mostly to emulate likely code, and + * we want a long delay occasionally to force massive contention. + */ + if (!(torture_random(trsp) % + (cxt.nrealwriters_stress * 2000 * longdelay_ms))) + mdelay(longdelay_ms); + if (!(torture_random(trsp) % + (cxt.nrealwriters_stress * 2 * shortdelay_us))) + udelay(shortdelay_us); +#ifdef CONFIG_PREEMPT + if (!(torture_random(trsp) % (cxt.nrealwriters_stress * 20000))) + preempt_schedule(); /* Allow test to be preempted. */ +#endif +} + +static void torture_rtmutex_unlock(void) __releases(torture_rtmutex) +{ + rt_mutex_unlock(&torture_rtmutex); +} + +static struct lock_torture_ops rtmutex_lock_ops = { + .writelock = torture_rtmutex_lock, + .write_delay = torture_rtmutex_delay, + .task_boost = torture_rtmutex_boost, + .writeunlock = torture_rtmutex_unlock, + .readlock = NULL, + .read_delay = NULL, + .readunlock = NULL, + .name = "rtmutex_lock" +}; +#endif + static DECLARE_RWSEM(torture_rwsem); static int torture_rwsem_down_write(void) __acquires(torture_rwsem) { @@ -419,6 +518,7 @@ static void torture_rwsem_up_read(void) __releases(torture_rwsem) static struct lock_torture_ops rwsem_lock_ops = { .writelock = torture_rwsem_down_write, .write_delay = torture_rwsem_write_delay, + .task_boost = torture_boost_dummy, .writeunlock = torture_rwsem_up_write, .readlock = torture_rwsem_down_read, .read_delay = torture_rwsem_read_delay, @@ -442,6 +542,7 @@ static int lock_torture_writer(void *arg) if ((torture_random(&rand) & 0xfffff) == 0) schedule_timeout_uninterruptible(1); + cxt.cur_ops->task_boost(&rand); cxt.cur_ops->writelock(); if (WARN_ON_ONCE(lock_is_write_held)) lwsp->n_lock_fail++; @@ -456,6 +557,8 @@ static int lock_torture_writer(void *arg) stutter_wait("lock_torture_writer"); } while (!torture_must_stop()); + + cxt.cur_ops->task_boost(NULL); /* reset prio */ torture_kthread_stopping("lock_torture_writer"); return 0; } @@ -642,6 +745,9 @@ static int __init lock_torture_init(void) &spin_lock_ops, &spin_lock_irq_ops, &rw_lock_ops, &rw_lock_irq_ops, &mutex_lock_ops, +#ifdef CONFIG_RT_MUTEXES + &rtmutex_lock_ops, +#endif &rwsem_lock_ops, }; @@ -676,6 +782,10 @@ static int __init lock_torture_init(void) if (strncmp(torture_type, "mutex", 5) == 0) cxt.debug_lock = true; #endif +#ifdef CONFIG_DEBUG_RT_MUTEXES + if (strncmp(torture_type, "rtmutex", 7) == 0) + cxt.debug_lock = true; +#endif #ifdef CONFIG_DEBUG_SPINLOCK if ((strncmp(torture_type, "spin", 4) == 0) || (strncmp(torture_type, "rw_lock", 7) == 0)) -- cgit v1.2.3 From 302707fd7cd341a23dbc448a335d432ad0069c20 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Mon, 31 Aug 2015 20:21:59 -0700 Subject: locking/percpu-rwsem: Export symbols for locktorture This commit exports percpu_down_read(), percpu_down_write(), __percpu_init_rwsem(), percpu_up_read(), and percpu_up_write() to allow locktorture to test them when built as a module. Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/percpu-rwsem.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'kernel/locking') diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c index f32567254867..e2621fbbcbd1 100644 --- a/kernel/locking/percpu-rwsem.c +++ b/kernel/locking/percpu-rwsem.c @@ -22,6 +22,7 @@ int __percpu_init_rwsem(struct percpu_rw_semaphore *brw, init_waitqueue_head(&brw->write_waitq); return 0; } +EXPORT_SYMBOL_GPL(__percpu_init_rwsem); void percpu_free_rwsem(struct percpu_rw_semaphore *brw) { @@ -87,6 +88,7 @@ void percpu_down_read(struct percpu_rw_semaphore *brw) /* avoid up_read()->rwsem_release() */ __up_read(&brw->rw_sem); } +EXPORT_SYMBOL_GPL(percpu_down_read); int percpu_down_read_trylock(struct percpu_rw_semaphore *brw) { @@ -112,6 +114,7 @@ void percpu_up_read(struct percpu_rw_semaphore *brw) if (atomic_dec_and_test(&brw->slow_read_ctr)) wake_up_all(&brw->write_waitq); } +EXPORT_SYMBOL_GPL(percpu_up_read); static int clear_fast_ctr(struct percpu_rw_semaphore *brw) { @@ -163,6 +166,7 @@ void percpu_down_write(struct percpu_rw_semaphore *brw) /* wait for all readers to complete their percpu_up_read() */ wait_event(brw->write_waitq, !atomic_read(&brw->slow_read_ctr)); } +EXPORT_SYMBOL_GPL(percpu_down_write); void percpu_up_write(struct percpu_rw_semaphore *brw) { @@ -176,3 +180,4 @@ void percpu_up_write(struct percpu_rw_semaphore *brw) /* the last writer unblocks update_fast_ctr() */ atomic_dec(&brw->write_ctr); } +EXPORT_SYMBOL_GPL(percpu_up_write); -- cgit v1.2.3 From 617783dd99704331e22636388c932450e02ee636 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sat, 29 Aug 2015 14:46:29 -0700 Subject: locktorture: Add torture tests for percpu_rwsem This commit adds percpu_rwsem tests based on the earlier rwsem tests. Signed-off-by: Paul E. McKenney Cc: Oleg Nesterov Cc: Davidlohr Bueso Reviewed-by: Josh Triplett --- kernel/locking/locktorture.c | 44 ++++++++++++++++++++++ .../selftests/rcutorture/configs/lock/CFLIST | 1 + .../selftests/rcutorture/configs/lock/LOCK06 | 6 +++ .../selftests/rcutorture/configs/lock/LOCK06.boot | 1 + 4 files changed, 52 insertions(+) create mode 100644 tools/testing/selftests/rcutorture/configs/lock/LOCK06 create mode 100644 tools/testing/selftests/rcutorture/configs/lock/LOCK06.boot (limited to 'kernel/locking') diff --git a/kernel/locking/locktorture.c b/kernel/locking/locktorture.c index e1ca7a2fae91..8545e12598ce 100644 --- a/kernel/locking/locktorture.c +++ b/kernel/locking/locktorture.c @@ -36,6 +36,7 @@ #include #include #include +#include #include MODULE_LICENSE("GPL"); @@ -526,6 +527,48 @@ static struct lock_torture_ops rwsem_lock_ops = { .name = "rwsem_lock" }; +#include +static struct percpu_rw_semaphore pcpu_rwsem; + +void torture_percpu_rwsem_init(void) +{ + BUG_ON(percpu_init_rwsem(&pcpu_rwsem)); +} + +static int torture_percpu_rwsem_down_write(void) __acquires(pcpu_rwsem) +{ + percpu_down_write(&pcpu_rwsem); + return 0; +} + +static void torture_percpu_rwsem_up_write(void) __releases(pcpu_rwsem) +{ + percpu_up_write(&pcpu_rwsem); +} + +static int torture_percpu_rwsem_down_read(void) __acquires(pcpu_rwsem) +{ + percpu_down_read(&pcpu_rwsem); + return 0; +} + +static void torture_percpu_rwsem_up_read(void) __releases(pcpu_rwsem) +{ + percpu_up_read(&pcpu_rwsem); +} + +static struct lock_torture_ops percpu_rwsem_lock_ops = { + .init = torture_percpu_rwsem_init, + .writelock = torture_percpu_rwsem_down_write, + .write_delay = torture_rwsem_write_delay, + .task_boost = torture_boost_dummy, + .writeunlock = torture_percpu_rwsem_up_write, + .readlock = torture_percpu_rwsem_down_read, + .read_delay = torture_rwsem_read_delay, + .readunlock = torture_percpu_rwsem_up_read, + .name = "percpu_rwsem_lock" +}; + /* * Lock torture writer kthread. Repeatedly acquires and releases * the lock, checking for duplicate acquisitions. @@ -749,6 +792,7 @@ static int __init lock_torture_init(void) &rtmutex_lock_ops, #endif &rwsem_lock_ops, + &percpu_rwsem_lock_ops, }; if (!torture_init_begin(torture_type, verbose, &torture_runnable)) diff --git a/tools/testing/selftests/rcutorture/configs/lock/CFLIST b/tools/testing/selftests/rcutorture/configs/lock/CFLIST index 6ed32794eaa1..b9611c523723 100644 --- a/tools/testing/selftests/rcutorture/configs/lock/CFLIST +++ b/tools/testing/selftests/rcutorture/configs/lock/CFLIST @@ -3,3 +3,4 @@ LOCK02 LOCK03 LOCK04 LOCK05 +LOCK06 diff --git a/tools/testing/selftests/rcutorture/configs/lock/LOCK06 b/tools/testing/selftests/rcutorture/configs/lock/LOCK06 new file mode 100644 index 000000000000..1d1da1477fc3 --- /dev/null +++ b/tools/testing/selftests/rcutorture/configs/lock/LOCK06 @@ -0,0 +1,6 @@ +CONFIG_SMP=y +CONFIG_NR_CPUS=4 +CONFIG_HOTPLUG_CPU=y +CONFIG_PREEMPT_NONE=n +CONFIG_PREEMPT_VOLUNTARY=n +CONFIG_PREEMPT=y diff --git a/tools/testing/selftests/rcutorture/configs/lock/LOCK06.boot b/tools/testing/selftests/rcutorture/configs/lock/LOCK06.boot new file mode 100644 index 000000000000..f92219cd4ad9 --- /dev/null +++ b/tools/testing/selftests/rcutorture/configs/lock/LOCK06.boot @@ -0,0 +1 @@ +locktorture.torture_type=percpu_rwsem_lock -- cgit v1.2.3 From 95b19f684c61ffc9b039e02c5d1113c2d8cd7105 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Fri, 21 Aug 2015 19:42:55 +0200 Subject: locking/percpu-rwsem: Make percpu_free_rwsem() after kzalloc() safe This is the temporary ugly hack which will be reverted later. We only need it to ensure that the next patch will not break "change sb_writers to use percpu_rw_semaphore" patches routed via the VFS tree. The alloc_super()->destroy_super() error path assumes that it is safe to call percpu_free_rwsem() after kzalloc() without percpu_init_rwsem(), so let's not disappoint it. Signed-off-by: Oleg Nesterov Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/percpu-rwsem.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'kernel/locking') diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c index e2621fbbcbd1..9529a30ec57b 100644 --- a/kernel/locking/percpu-rwsem.c +++ b/kernel/locking/percpu-rwsem.c @@ -26,6 +26,13 @@ EXPORT_SYMBOL_GPL(__percpu_init_rwsem); void percpu_free_rwsem(struct percpu_rw_semaphore *brw) { + /* + * XXX: temporary kludge. The error path in alloc_super() + * assumes that percpu_free_rwsem() is safe after kzalloc(). + */ + if (!brw->fast_read_ctr) + return; + free_percpu(brw->fast_read_ctr); brw->fast_read_ctr = NULL; /* catch use after free bugs */ } -- cgit v1.2.3 From 001dac627ff37433d5528ffb0d897cd19c2b1e43 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Fri, 21 Aug 2015 19:42:57 +0200 Subject: locking/percpu-rwsem: Make use of the rcu_sync infrastructure Currently down_write/up_write calls synchronize_sched_expedited() twice, which is evil. Change this code to rely on rcu-sync primitives. This avoids the _expedited "big hammer", and this can be faster in the contended case or even in the case when a single thread does down_write/up_write in a loop. Of course, a single down_write() will take more time, but otoh it will be much more friendly to the whole system. To simplify the review this patch doesn't update the comments, fixed by the next change. Signed-off-by: Oleg Nesterov Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- include/linux/percpu-rwsem.h | 3 ++- kernel/locking/percpu-rwsem.c | 18 +++++++----------- 2 files changed, 9 insertions(+), 12 deletions(-) (limited to 'kernel/locking') diff --git a/include/linux/percpu-rwsem.h b/include/linux/percpu-rwsem.h index 834c4e52cb2d..c2fa3ecb0dce 100644 --- a/include/linux/percpu-rwsem.h +++ b/include/linux/percpu-rwsem.h @@ -5,11 +5,12 @@ #include #include #include +#include #include struct percpu_rw_semaphore { + struct rcu_sync rss; unsigned int __percpu *fast_read_ctr; - atomic_t write_ctr; struct rw_semaphore rw_sem; atomic_t slow_read_ctr; wait_queue_head_t write_waitq; diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c index 9529a30ec57b..183a71151ac0 100644 --- a/kernel/locking/percpu-rwsem.c +++ b/kernel/locking/percpu-rwsem.c @@ -17,7 +17,7 @@ int __percpu_init_rwsem(struct percpu_rw_semaphore *brw, /* ->rw_sem represents the whole percpu_rw_semaphore for lockdep */ __init_rwsem(&brw->rw_sem, name, rwsem_key); - atomic_set(&brw->write_ctr, 0); + rcu_sync_init(&brw->rss, RCU_SCHED_SYNC); atomic_set(&brw->slow_read_ctr, 0); init_waitqueue_head(&brw->write_waitq); return 0; @@ -33,6 +33,7 @@ void percpu_free_rwsem(struct percpu_rw_semaphore *brw) if (!brw->fast_read_ctr) return; + rcu_sync_dtor(&brw->rss); free_percpu(brw->fast_read_ctr); brw->fast_read_ctr = NULL; /* catch use after free bugs */ } @@ -62,13 +63,12 @@ void percpu_free_rwsem(struct percpu_rw_semaphore *brw) */ static bool update_fast_ctr(struct percpu_rw_semaphore *brw, unsigned int val) { - bool success = false; + bool success; preempt_disable(); - if (likely(!atomic_read(&brw->write_ctr))) { + success = rcu_sync_is_idle(&brw->rss); + if (likely(success)) __this_cpu_add(*brw->fast_read_ctr, val); - success = true; - } preempt_enable(); return success; @@ -149,8 +149,6 @@ static int clear_fast_ctr(struct percpu_rw_semaphore *brw) */ void percpu_down_write(struct percpu_rw_semaphore *brw) { - /* tell update_fast_ctr() there is a pending writer */ - atomic_inc(&brw->write_ctr); /* * 1. Ensures that write_ctr != 0 is visible to any down_read/up_read * so that update_fast_ctr() can't succeed. @@ -162,7 +160,7 @@ void percpu_down_write(struct percpu_rw_semaphore *brw) * fast-path, it executes a full memory barrier before we return. * See R_W case in the comment above update_fast_ctr(). */ - synchronize_sched_expedited(); + rcu_sync_enter(&brw->rss); /* exclude other writers, and block the new readers completely */ down_write(&brw->rw_sem); @@ -183,8 +181,6 @@ void percpu_up_write(struct percpu_rw_semaphore *brw) * Insert the barrier before the next fast-path in down_read, * see W_R case in the comment above update_fast_ctr(). */ - synchronize_sched_expedited(); - /* the last writer unblocks update_fast_ctr() */ - atomic_dec(&brw->write_ctr); + rcu_sync_exit(&brw->rss); } EXPORT_SYMBOL_GPL(percpu_up_write); -- cgit v1.2.3 From f324a76324c97e81a6ba66a8efac20cdbffd759e Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Fri, 21 Aug 2015 19:43:00 +0200 Subject: locking/percpu-rwsem: Fix the comments outdated by rcu_sync Update the comments broken by the previous change. Signed-off-by: Oleg Nesterov Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/percpu-rwsem.c | 50 ++++++++++--------------------------------- 1 file changed, 11 insertions(+), 39 deletions(-) (limited to 'kernel/locking') diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c index 183a71151ac0..02a726dd9adc 100644 --- a/kernel/locking/percpu-rwsem.c +++ b/kernel/locking/percpu-rwsem.c @@ -39,27 +39,12 @@ void percpu_free_rwsem(struct percpu_rw_semaphore *brw) } /* - * This is the fast-path for down_read/up_read, it only needs to ensure - * there is no pending writer (atomic_read(write_ctr) == 0) and inc/dec the - * fast per-cpu counter. The writer uses synchronize_sched_expedited() to - * serialize with the preempt-disabled section below. - * - * The nontrivial part is that we should guarantee acquire/release semantics - * in case when - * - * R_W: down_write() comes after up_read(), the writer should see all - * changes done by the reader - * or - * W_R: down_read() comes after up_write(), the reader should see all - * changes done by the writer + * This is the fast-path for down_read/up_read. If it succeeds we rely + * on the barriers provided by rcu_sync_enter/exit; see the comments in + * percpu_down_write() and percpu_up_write(). * * If this helper fails the callers rely on the normal rw_semaphore and * atomic_dec_and_test(), so in this case we have the necessary barriers. - * - * But if it succeeds we do not have any barriers, atomic_read(write_ctr) or - * __this_cpu_add() below can be reordered with any LOAD/STORE done by the - * reader inside the critical section. See the comments in down_write and - * up_write below. */ static bool update_fast_ctr(struct percpu_rw_semaphore *brw, unsigned int val) { @@ -136,29 +121,15 @@ static int clear_fast_ctr(struct percpu_rw_semaphore *brw) return sum; } -/* - * A writer increments ->write_ctr to force the readers to switch to the - * slow mode, note the atomic_read() check in update_fast_ctr(). - * - * After that the readers can only inc/dec the slow ->slow_read_ctr counter, - * ->fast_read_ctr is stable. Once the writer moves its sum into the slow - * counter it represents the number of active readers. - * - * Finally the writer takes ->rw_sem for writing and blocks the new readers, - * then waits until the slow counter becomes zero. - */ void percpu_down_write(struct percpu_rw_semaphore *brw) { /* - * 1. Ensures that write_ctr != 0 is visible to any down_read/up_read - * so that update_fast_ctr() can't succeed. - * - * 2. Ensures we see the result of every previous this_cpu_add() in - * update_fast_ctr(). + * Make rcu_sync_is_idle() == F and thus disable the fast-path in + * percpu_down_read() and percpu_up_read(), and wait for gp pass. * - * 3. Ensures that if any reader has exited its critical section via - * fast-path, it executes a full memory barrier before we return. - * See R_W case in the comment above update_fast_ctr(). + * The latter synchronises us with the preceding readers which used + * the fast-past, so we can not miss the result of __this_cpu_add() + * or anything else inside their criticial sections. */ rcu_sync_enter(&brw->rss); @@ -178,8 +149,9 @@ void percpu_up_write(struct percpu_rw_semaphore *brw) /* release the lock, but the readers can't use the fast-path */ up_write(&brw->rw_sem); /* - * Insert the barrier before the next fast-path in down_read, - * see W_R case in the comment above update_fast_ctr(). + * Enable the fast-path in percpu_down_read() and percpu_up_read() + * but only after another gp pass; this adds the necessary barrier + * to ensure the reader can't miss the changes done by us. */ rcu_sync_exit(&brw->rss); } -- cgit v1.2.3 From cc5f730b41506d37a5c2826b2e801d0a59853d11 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Fri, 21 Aug 2015 19:43:03 +0200 Subject: locking/percpu-rwsem: Clean up the lockdep annotations in percpu_down_read() Based on Peter Zijlstra's earlier patch. Change percpu_down_read() to use __down_read(), this way we can do rwsem_acquire_read() unconditionally at the start to make this code more symmetric and clean. Originally-From: Peter Zijlstra Signed-off-by: Oleg Nesterov Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/percpu-rwsem.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'kernel/locking') diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c index 02a726dd9adc..f231e0bb311c 100644 --- a/kernel/locking/percpu-rwsem.c +++ b/kernel/locking/percpu-rwsem.c @@ -70,14 +70,14 @@ static bool update_fast_ctr(struct percpu_rw_semaphore *brw, unsigned int val) void percpu_down_read(struct percpu_rw_semaphore *brw) { might_sleep(); - if (likely(update_fast_ctr(brw, +1))) { - rwsem_acquire_read(&brw->rw_sem.dep_map, 0, 0, _RET_IP_); + rwsem_acquire_read(&brw->rw_sem.dep_map, 0, 0, _RET_IP_); + + if (likely(update_fast_ctr(brw, +1))) return; - } - down_read(&brw->rw_sem); + /* Avoid rwsem_acquire_read() and rwsem_release() */ + __down_read(&brw->rw_sem); atomic_inc(&brw->slow_read_ctr); - /* avoid up_read()->rwsem_release() */ __up_read(&brw->rw_sem); } EXPORT_SYMBOL_GPL(percpu_down_read); -- cgit v1.2.3 From a36a99618b1adb2d6ca0b7e08e3a656a04e477fe Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 30 Aug 2015 20:01:48 -0700 Subject: locktorture: Fix module unwind when bad torture_type specified The locktorture module has a list of torture types, and specifying a type not on this list is supposed to cleanly fail the module load. Unfortunately, the "fail" happens without the "cleanly". This commit therefore adds the needed clean-up after an incorrect torture_type. Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett --- kernel/locking/locktorture.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel/locking') diff --git a/kernel/locking/locktorture.c b/kernel/locking/locktorture.c index 32244186f1f2..820852f69858 100644 --- a/kernel/locking/locktorture.c +++ b/kernel/locking/locktorture.c @@ -661,11 +661,11 @@ static int __init lock_torture_init(void) for (i = 0; i < ARRAY_SIZE(torture_ops); i++) pr_alert(" %s", torture_ops[i]->name); pr_alert("\n"); - torture_init_end(); - return -EINVAL; + firsterr = -EINVAL; + goto unwind; } if (cxt.cur_ops->init) - cxt.cur_ops->init(); /* no "goto unwind" prior to this point!!! */ + cxt.cur_ops->init(); if (nwriters_stress >= 0) cxt.nrealwriters_stress = nwriters_stress; -- cgit v1.2.3