Messages in this thread Patch in this message | | | Date | Mon, 16 Feb 2015 09:21:13 +0100 | From | Peter Zijlstra <> | Subject | Re: [PATCH] sched/completion: completion_done() should serialize with complete() |
| |
On Thu, Feb 12, 2015 at 08:59:13PM +0100, Oleg Nesterov wrote: > Commit de30ec47302c "Remove unnecessary ->wait.lock serialization when > reading completion state" was not correct, without lock/unlock the code > like stop_machine_from_inactive_cpu() > > while (!completion_done()) > cpu_relax(); > > can return before complete() finishes its spin_unlock() which writes to > this memory. And spin_unlock_wait(). > > While at it, change try_wait_for_completion() to use READ_ONCE().
So I share Davidlohrs concern if we should not simply revert that change; but given we've now gone over it detail I suppose we should just keep the optimized version.
I did add a comment to your patch; and queued the below for sched/urgent.
--- Subject: sched/completion: completion_done() should serialize with complete() From: Oleg Nesterov <oleg@redhat.com> Date: Thu, 12 Feb 2015 20:59:13 +0100
Commit de30ec47302c "Remove unnecessary ->wait.lock serialization when reading completion state" was not correct, without lock/unlock the code like stop_machine_from_inactive_cpu()
while (!completion_done()) cpu_relax();
can return before complete() finishes its spin_unlock() which writes to this memory. And spin_unlock_wait().
While at it, change try_wait_for_completion() to use READ_ONCE().
Fixes: de30ec47302c ("sched/completion: Remove unnecessary ->wait.lock serialization when reading completion state") Cc: waiman.long@hp.com Cc: raghavendra.kt@linux.vnet.ibm.com Cc: dave@stgolabs.net Cc: Nicholas Mc Guire <der.herr@hofr.at> Cc: Linus Torvalds <torvalds@linux-foundation.org> Reported-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Tested-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Reported-by: Davidlohr Bueso <dave@stgolabs.net> Signed-off-by: Oleg Nesterov <oleg@redhat.com> [peterz: Add a comment with the barrier] Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: http://lkml.kernel.org/r/20150212195913.GA30430@redhat.com --- kernel/sched/completion.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-)
--- a/kernel/sched/completion.c +++ b/kernel/sched/completion.c @@ -274,7 +274,7 @@ bool try_wait_for_completion(struct comp * first without taking the lock so we can * return early in the blocking case. */ - if (!ACCESS_ONCE(x->done)) + if (!READ_ONCE(x->done)) return 0; spin_lock_irqsave(&x->wait.lock, flags); @@ -297,6 +297,21 @@ EXPORT_SYMBOL(try_wait_for_completion); */ bool completion_done(struct completion *x) { - return !!ACCESS_ONCE(x->done); + if (!READ_ONCE(x->done)) + return false; + + /* + * If ->done, we need to wait for complete() to release ->wait.lock + * otherwise we can end up freeing the completion before complete() + * is done referencing it. + * + * The RMB pairs with complete()'s RELEASE of ->wait.lock and orders + * the loads of ->done and ->wait.lock such that we cannot observe + * the lock before complete() acquires it while observing the ->done + * after it's acquired the lock. + */ + smp_rmb(); + spin_unlock_wait(&x->wait.lock); + return true; } EXPORT_SYMBOL(completion_done);
| |