Hi Gertjan
On Mon, Mar 05, 2012 at 08:54:37PM +0100, Gertjan van Wingerde wrote:
> There are more places in the rt2x00 code that call upon rt2x00queue_pause_queue and rt2x00queue_unpause_queue. Shouldn't these places be protected with tx_lock as well?
Hmm, good question. Perhaps there are possible races between other usage
of pause/unpause, but they are not obvious for me.
Seems there are more bugs there, i.e on rt2x00queue_stop_queue, we first
clear QUEUE_STARTED then call pause_queue, which will just return with
that bit cleared.
On rt2x00queue_flush_queue(), we first call pause queue, then
->kick_queue, which will cause to call txdone and unpause queue.
So, it's hard to tell for me right now if other paces needs serialization,
apparently related area needs some more detailed review,
> Or better, shouldn't the locking be moved inside the pause / unpause functions?
Thats a bit more complication, because we need lock for TX queues only
and it has to be taken before test_and_{set,clear}_bit(QUEUE_PAUSED, &queue->flags),
otherwise we still can race.
I believe this patch is simplest possible approach to solve the problem,
at least for now.
Stanislaw
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
