rt2x00: Move watchdog work to kernel work_queue

The watchdog function must run on a work_queue
which is independent of any other work inside rt2x00.

The main reasons, being that a broken work on the mac80211
work_queue can otherwise prevent the watchdog to run (while
in fact the watchdog could fix the issue). And on the other
hand because the watchdog relies on the completion of the
completion handlers for RX/TX which for the USB case, occur
on the mac80211 workqueue.

This fixes some "Queue %d failed to flush" errors, which were
caused by the watchdog function waiting on the completion
handler which was scheduled to run right after the watchdog.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Acked-by: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

1 files changed, 5 insertions, 0 deletions

diff --git a/drivers/net/wireless/rt2x00/rt2x00.h b/drivers/net/wireless/rt2x00/rt2x00.h
index 2322c84..94fe589 100644
--- a/drivers/net/wireless/rt2x00/rt2x00.h
+++ b/drivers/net/wireless/rt2x00/rt2x00.h
@@ -338,6 +338,11 @@ struct link {
 
 	/*
 	 * Work structure for scheduling periodic watchdog monitoring.
+	 * This work must be scheduled on the kernel workqueue, while
+	 * all other work structures must be queued on the mac80211
+	 * workqueue. This guarantees that the watchdog can schedule
+	 * other work structures and wait for their completion in order
+	 * to bring the device/driver back into the desired state.
 	 */
 	struct delayed_work watchdog_work;
 };