On Sun, 2006-07-30 at 08:36 +0400, Oleg Nesterov wrote:
> Another question, task_blocks_on_rt_mutex() does get_task_struct() and checks
> owner->pi_blocked_on != NULL under owner->pi_lock. Why ? RT_MUTEX_HAS_WAITERS
> bit is set, we are holding ->wait_lock, so the 'owner' can't go away until
> we drop ->wait_lock. I think we can drop owner->pi_lock right after
> __rt_mutex_adjust_prio(owner), we can't miss owner->pi_blocked_on != NULL
> if it was true before we take owner->pi_lock, and this is the case we should
> worry about, yes?
No.
We hold lock->wait_lock. The owner of this lock can be blocked itself,
which makes it necessary to do the chain walk. The indicator is
owner->pi_blocked_on. This field is only protected by owner->pi_lock.
If we look at this field outside of owner->pi_lock, then we might miss a
chain walk.
CPU 0 CPU 1
lock lock->wait_lock
block_on_rt_mutex()
lock current->pi_lock
current->pi_blocked_on = waiter(lock)
unlock current->pi_lock
boost = owner->pi_blocked_on
owner blocks on lock2
lock owner->pi_lock
owner->pi_blocked_on = waiter(lock2)
unlock owner->pi_lock
lock owner->pi_lock
enqueue waiter
adjust prio
unlock owner->pi_lock
unlock lock->wait_lock
if boost
do_chain_walk()
->pi_blocked_on is protected by ->pi_lock and has to be read/modified
under this lock. That way we can not miss a chain walk:
CPU 0 CPU 1
lock lock->wait_lock
block_on_rt_mutex()
lock current->pi_lock
current->pi_blocked_on = waiter(lock)
unlock current->pi_lock
owner blocks on lock2
lock owner->pi_lock
owner->pi_blocked_on = waiter(lock2)
unlock owner->pi_lock
lock owner->pi_lock
enqueue waiter
adjust prio
boost = owner->pi_blocked_on
unlock owner->pi_lock
unlock lock->wait_lock
if boost
do_chain_walk()
CPU 0 CPU 1
lock lock->wait_lock
block_on_rt_mutex()
lock current->pi_lock
current->pi_blocked_on = waiter(lock)
unlock current->pi_lock
lock owner->pi_lock
enqueue waiter
adjust prio of owner
boost = owner->pi_blocked_on
unlock owner->pi_lock
unlock lock->wait_lock
owner blocks on lock2
lock owner->pi_lock
owner->pi_blocked_on = waiter(lock2)
unlock owner->pi_lock
if boost
do_chain_walk()
owner propagates the priority to owner(lock2)
get_task_struct() is there for a different reason. We have to protect
the gap where we hold _no_ lock in rt_mutex_adjust_prio_chain:
retry:
lock task->pi_lock
block = task->pi_blocked_on->lock
if (! trylock block->wait_lock) {
unlock task->pi_lock
-> task could go away right here, if we do not hold a reference
cpu_relax()
goto retry
}
tglx
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
