]> pilppa.com Git - linux-2.6-omap-h63xx.git/commit
[NET_SCHED]: Revert "avoid transmit softirq on watchdog wakeup" optimization
authorPatrick McHardy <kaber@trash.net>
Sun, 15 Jul 2007 03:49:26 +0000 (20:49 -0700)
committerDavid S. Miller <davem@davemloft.net>
Sun, 15 Jul 2007 03:49:26 +0000 (20:49 -0700)
commit0621ed2e4edbe2f6f83dafbf85eecefae7aaf2e8
tree30e005117719da029a238b88c742b0a7e516ad50
parent59eecdfb166f6846ae356ddc744abed5820ad965
[NET_SCHED]: Revert "avoid transmit softirq on watchdog wakeup" optimization

As noticed by Ranko Zivojnovic <ranko@spidernet.net>, calling qdisc_run
from the timer handler can result in deadlock:

> CPU#0
>
> qdisc_watchdog() fires and gets dev->queue_lock
> qdisc_run()...qdisc_restart()...
> -> releases dev->queue_lock and enters dev_hard_start_xmit()
>
> CPU#1
>
> tc del qdisc dev ...
> qdisc_graft()...dev_graft_qdisc()...dev_deactivate()...
> -> grabs dev->queue_lock ...
>
> qdisc_reset()...{cbq,hfsc,htb,netem,tbf}_reset()...qdisc_watchdog_cancel()...
> -> hrtimer_cancel() - waiting for the qdisc_watchdog() to exit, while still
>         holding dev->queue_lock
>
> CPU#0
>
> dev_hard_start_xmit() returns ...
> -> wants to get dev->queue_lock(!)
>
> DEADLOCK!

The entire optimization is a bit questionable IMO, it moves potentially
large parts of NET_TX_SOFTIRQ work to TIMER_SOFTIRQ/HRTIMER_SOFTIRQ,
which kind of defeats the separation of them.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Ranko Zivojnovic <ranko@spidernet.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_api.c