Re: via_rhine kernel crashes in 2.6.32

[Jarek Poplawski <jarkao2@gmail.com>]

On 22-12-2009 16:26, Roger Luethi wrote:
> On Tue, 22 Dec 2009 20:00:45 +0500, Andrey Rahmatullin wrote:
>> On Tue, Dec 22, 2009 at 01:38:17PM +0000, Jarek Poplawski wrote:
>>>>> It looks like napi_disable() should be illegal in ndo_tx_timeout().
>>>>> Here is a patch which moves most of the timeout work to a workqueue,
>>>>> similarly to tg3 etc. It should prevent at least one of reported
>>>>> bugs. Alas I can't even check-compile it at the moment, so let me
>>>>> know on any problems.
>>>> It seems I needlessly changed locking btw, so here it is again.
>>> Hmm... On the other hand, it definitely needs at least _bh now...
>> I've tried this patch. There are lots of "Transmit timed out", but no
>> crashes.
> 
> ACK. Looks like you guys tracked down the crashing and fixed it (thanks!).
> I suspect we shouldn't have to reset due to timeouts that often, but that's
> another story.

BTW, it seems a change in 2.6.31 might trigger these timeouts more
often than before. Andrey, could you try if this matters here?

Thanks,
Jarek P.

--- (on top of net-2.6 with the previous "Fix scheduling..." patch)

diff -Nurp a/drivers/net/via-rhine.c b/drivers/net/via-rhine.c
--- a/drivers/net/via-rhine.c	2009-12-23 09:28:25.000000000 +0000
+++ b/drivers/net/via-rhine.c	2009-12-23 09:33:57.000000000 +0000
@@ -1226,6 +1226,7 @@ static void rhine_tx_timeout(struct net_
 	       mdio_read(dev, rp->mii_if.phy_id, MII_BMSR));
 
 	schedule_work(&rp->reset_task);
+	netdev_get_tx_queue(dev, 0)->trans_start = jiffies;
 }
 
 static netdev_tx_t rhine_start_tx(struct sk_buff *skb,