From 6dc080eeb2ba01973bfff0d79844d7a59e12542e Mon Sep 17 00:00:00 2001 From: Prateek Sood Date: Fri, 30 Nov 2018 20:40:56 +0530 Subject: [PATCH 1/5] sched/wait: Fix rcuwait_wake_up() ordering For some peculiar reason rcuwait_wake_up() has the right barrier in the comment, but not in the code. This mistake has been observed to cause a deadlock in the following situation: P1 P2 percpu_up_read() percpu_down_write() rcu_sync_is_idle() // false rcu_sync_enter() ... __percpu_up_read() [S] ,- __this_cpu_dec(*sem->read_count) | smp_rmb(); [L] | task = rcu_dereference(w->task) // NULL | | [S] w->task = current | smp_mb(); | [L] readers_active_check() // fail `-> Where the smp_rmb() (obviously) fails to constrain the store. [ peterz: Added changelog. ] Signed-off-by: Prateek Sood Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Andrea Parri Acked-by: Davidlohr Bueso Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Fixes: 8f95c90ceb54 ("sched/wait, RCU: Introduce rcuwait machinery") Link: https://lkml.kernel.org/r/1543590656-7157-1-git-send-email-prsood@codeaurora.org Signed-off-by: Ingo Molnar --- kernel/exit.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/exit.c b/kernel/exit.c index 284f2fe9a293..3fb7be001964 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -307,7 +307,7 @@ void rcuwait_wake_up(struct rcuwait *w) * MB (A) MB (B) * [L] cond [L] tsk */ - smp_rmb(); /* (B) */ + smp_mb(); /* (B) */ /* * Avoid using task_rcu_dereference() magic as long as we are careful, From e6018c0f5c996e61639adce6a0697391a2861916 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 17 Dec 2018 10:14:53 +0100 Subject: [PATCH 2/5] sched/wake_q: Document wake_q_add() The only guarantee provided by wake_q_add() is that a wakeup will happen after it, it does _NOT_ guarantee the wakeup will be delayed until the matching wake_up_q(). If wake_q_add() fails the cmpxchg() a concurrent wakeup is pending and that can happen at any time after the cmpxchg(). This means we should not rely on the wakeup happening at wake_q_up(), but should be ready for wake_q_add() to issue the wakeup. The delay; if provided (most likely); should only result in more efficient behaviour. Reported-by: Yongji Xie Signed-off-by: Peter Zijlstra (Intel) Cc: Davidlohr Bueso Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: Waiman Long Cc: Will Deacon Signed-off-by: Ingo Molnar --- include/linux/sched/wake_q.h | 6 +++++- kernel/sched/core.c | 12 ++++++++++++ 2 files changed, 17 insertions(+), 1 deletion(-) diff --git a/include/linux/sched/wake_q.h b/include/linux/sched/wake_q.h index 10b19a192b2d..545f37138057 100644 --- a/include/linux/sched/wake_q.h +++ b/include/linux/sched/wake_q.h @@ -24,9 +24,13 @@ * called near the end of a function. Otherwise, the list can be * re-initialized for later re-use by wake_q_init(). * - * Note that this can cause spurious wakeups. schedule() callers + * NOTE that this can cause spurious wakeups. schedule() callers * must ensure the call is done inside a loop, confirming that the * wakeup condition has in fact occurred. + * + * NOTE that there is no guarantee the wakeup will happen any later than the + * wake_q_add() location. Therefore task must be ready to be woken at the + * location of the wake_q_add(). */ #include diff --git a/kernel/sched/core.c b/kernel/sched/core.c index a674c7db2f29..cc814933f7d6 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -396,6 +396,18 @@ static bool set_nr_if_polling(struct task_struct *p) #endif #endif +/** + * wake_q_add() - queue a wakeup for 'later' waking. + * @head: the wake_q_head to add @task to + * @task: the task to queue for 'later' wakeup + * + * Queue a task for later wakeup, most likely by the wake_up_q() call in the + * same context, _HOWEVER_ this is not guaranteed, the wakeup can come + * instantly. + * + * This function must be used as-if it were wake_up_process(); IOW the task + * must be ready to be woken at this location. + */ void wake_q_add(struct wake_q_head *head, struct task_struct *task) { struct wake_q_node *node = &task->wake_q; From 4c4e3731564c8945ac5ac90fc2a1e1f21cb79c92 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 17 Dec 2018 10:14:53 +0100 Subject: [PATCH 3/5] sched/wake_q: Fix wakeup ordering for wake_q Notable cmpxchg() does not provide ordering when it fails, however wake_q_add() requires ordering in this specific case too. Without this it would be possible for the concurrent wakeup to not observe our prior state. Andrea Parri provided: C wake_up_q-wake_q_add { int next = 0; int y = 0; } P0(int *next, int *y) { int r0; /* in wake_up_q() */ WRITE_ONCE(*next, 1); /* node->next = NULL */ smp_mb(); /* implied by wake_up_process() */ r0 = READ_ONCE(*y); } P1(int *next, int *y) { int r1; /* in wake_q_add() */ WRITE_ONCE(*y, 1); /* wake_cond = true */ smp_mb__before_atomic(); r1 = cmpxchg_relaxed(next, 1, 2); } exists (0:r0=0 /\ 1:r1=0) This "exists" clause cannot be satisfied according to the LKMM: Test wake_up_q-wake_q_add Allowed States 3 0:r0=0; 1:r1=1; 0:r0=1; 1:r1=0; 0:r0=1; 1:r1=1; No Witnesses Positive: 0 Negative: 3 Condition exists (0:r0=0 /\ 1:r1=0) Observation wake_up_q-wake_q_add Never 0 3 Reported-by: Yongji Xie Signed-off-by: Peter Zijlstra (Intel) Cc: Davidlohr Bueso Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: Waiman Long Cc: Will Deacon Signed-off-by: Ingo Molnar --- kernel/sched/core.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index cc814933f7d6..d8d76a65cfdd 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -417,10 +417,11 @@ void wake_q_add(struct wake_q_head *head, struct task_struct *task) * its already queued (either by us or someone else) and will get the * wakeup due to that. * - * This cmpxchg() executes a full barrier, which pairs with the full - * barrier executed by the wakeup in wake_up_q(). + * In order to ensure that a pending wakeup will observe our pending + * state, even in the failed case, an explicit smp_mb() must be used. */ - if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) + smp_mb__before_atomic(); + if (cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL)) return; get_task_struct(task); From b061c38bef43406df8e73c5be06cbfacad5ee6ad Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 29 Nov 2018 14:44:49 +0100 Subject: [PATCH 4/5] futex: Fix (possible) missed wakeup We must not rely on wake_q_add() to delay the wakeup; in particular commit: 1d0dcb3ad9d3 ("futex: Implement lockless wakeups") moved wake_q_add() before smp_store_release(&q->lock_ptr, NULL), which could result in futex_wait() waking before observing ->lock_ptr == NULL and going back to sleep again. Signed-off-by: Peter Zijlstra (Intel) Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Fixes: 1d0dcb3ad9d3 ("futex: Implement lockless wakeups") Signed-off-by: Ingo Molnar --- kernel/futex.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) diff --git a/kernel/futex.c b/kernel/futex.c index be3bff2315ff..fdd312da0992 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1452,11 +1452,7 @@ static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) return; - /* - * Queue the task for later wakeup for after we've released - * the hb->lock. wake_q_add() grabs reference to p. - */ - wake_q_add(wake_q, p); + get_task_struct(p); __unqueue_futex(q); /* * The waiting task can free the futex_q as soon as q->lock_ptr = NULL @@ -1466,6 +1462,13 @@ static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) * plist_del in __unqueue_futex(). */ smp_store_release(&q->lock_ptr, NULL); + + /* + * Queue the task for later wakeup for after we've released + * the hb->lock. wake_q_add() grabs reference to p. + */ + wake_q_add(wake_q, p); + put_task_struct(p); } /* From e158488be27b157802753a59b336142dc0eb0380 Mon Sep 17 00:00:00 2001 From: Xie Yongji Date: Thu, 29 Nov 2018 20:50:30 +0800 Subject: [PATCH 5/5] locking/rwsem: Fix (possible) missed wakeup Because wake_q_add() can imply an immediate wakeup (cmpxchg failure case), we must not rely on the wakeup being delayed. However, commit: e38513905eea ("locking/rwsem: Rework zeroing reader waiter->task") relies on exactly that behaviour in that the wakeup must not happen until after we clear waiter->task. [ peterz: Added changelog. ] Signed-off-by: Xie Yongji Signed-off-by: Zhang Yu Signed-off-by: Peter Zijlstra (Intel) Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Fixes: e38513905eea ("locking/rwsem: Rework zeroing reader waiter->task") Link: https://lkml.kernel.org/r/1543495830-2644-1-git-send-email-xieyongji@baidu.com Signed-off-by: Ingo Molnar --- kernel/locking/rwsem-xadd.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/kernel/locking/rwsem-xadd.c b/kernel/locking/rwsem-xadd.c index 09b180063ee1..50d9af615dc4 100644 --- a/kernel/locking/rwsem-xadd.c +++ b/kernel/locking/rwsem-xadd.c @@ -198,15 +198,22 @@ static void __rwsem_mark_wake(struct rw_semaphore *sem, woken++; tsk = waiter->task; - wake_q_add(wake_q, tsk); + get_task_struct(tsk); list_del(&waiter->list); /* - * Ensure that the last operation is setting the reader + * Ensure calling get_task_struct() before setting the reader * waiter to nil such that rwsem_down_read_failed() cannot * race with do_exit() by always holding a reference count * to the task to wakeup. */ smp_store_release(&waiter->task, NULL); + /* + * Ensure issuing the wakeup (either by us or someone else) + * after setting the reader waiter to nil. + */ + wake_q_add(wake_q, tsk); + /* wake_q_add() already take the task ref */ + put_task_struct(tsk); } adjustment = woken * RWSEM_ACTIVE_READ_BIAS - adjustment;