Messages in this thread Patch in this message | | | Date | Wed, 4 Dec 2013 22:57:31 -0800 | From | Simon Kirby <> | Subject | Re: [PATCH] mutexes: Add CONFIG_DEBUG_MUTEX_FASTPATH=y debug variant to debug SMP races |
| |
On Tue, Dec 03, 2013 at 09:52:33AM +0100, Ingo Molnar wrote:
> Indeed: this comes from mutex->count being separate from > mutex->wait_lock, and this should affect every architecture that has a > mutex->count fast-path implemented (essentially every architecture > that matters). > > Such bugs should also magically go away with mutex debugging enabled.
Confirmed: I ran the reproducer with CONFIG_DEBUG_MUTEXES for a few hours, and never got a single poison overwritten notice.
> I'd expect such bugs to be more prominent with unlucky object > size/alignment: if mutex->count lies on a separate cache line from > mutex->wait_lock. > > Side note: this might be a valid light weight debugging technique, we > could add padding between the two fields to force them into separate > cache lines, without slowing it down. > > Simon, would you be willing to try the fairly trivial patch below? > Please enable CONFIG_DEBUG_MUTEX_FASTPATH=y. Does your kernel fail > faster that way?
I didn't see much of a change other than the incremented poison byte is now further in due to the padding, and it shows up in kmalloc-256.
I also tried with Linus' udelay() suggestion, below. With this, there were many occurrences per second.
Simon-
diff --git a/kernel/mutex.c b/kernel/mutex.c index d24105b..f65e735 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -25,6 +25,7 @@ #include <linux/spinlock.h> #include <linux/interrupt.h> #include <linux/debug_locks.h> +#include <linux/delay.h> /* * In the DEBUG case we are using the "NULL fastpath" for mutexes, @@ -740,6 +741,11 @@ __mutex_unlock_common_slowpath(atomic_t *lock_count, int nested) wake_up_process(waiter->task); } + /* udelay a bit if the spinlock isn't contended */ + if (lock->wait_lock.rlock.raw_lock.tickets.head + 1 == + lock->wait_lock.rlock.raw_lock.tickets.tail) + udelay(1); + spin_unlock_mutex(&lock->wait_lock, flags); }
| |