Patchwork via-rhine: Fix tx_timeout handling

login
register
mail settings
Submitter Richard Weinberger
Date July 19, 2013, 9:30 p.m.
Message ID <1374269428-6827-1-git-send-email-richard@nod.at>
Download mbox | patch
Permalink /patch/260357/
State Changes Requested
Delegated to: David Miller
Headers show

Comments

Richard Weinberger - July 19, 2013, 9:30 p.m.
rhine_reset_task() misses to call netif_stop_queue(),
this can lead to a crash if work is still scheduled while
we're resetting the tx queue.

Fixes:
[   93.591707] BUG: unable to handle kernel NULL pointer dereference at 0000004c
[   93.595514] IP: [<c119d10d>] rhine_napipoll+0x491/0x6e

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 drivers/net/ethernet/via/via-rhine.c | 1 +
 1 file changed, 1 insertion(+)
Ben Hutchings - July 21, 2013, 4:18 p.m.
On Fri, 2013-07-19 at 23:30 +0200, Richard Weinberger wrote:
> rhine_reset_task() misses to call netif_stop_queue(),
> this can lead to a crash if work is still scheduled while
> we're resetting the tx queue.
> 
> Fixes:
> [   93.591707] BUG: unable to handle kernel NULL pointer dereference at 0000004c
> [   93.595514] IP: [<c119d10d>] rhine_napipoll+0x491/0x6e
> 
> Signed-off-by: Richard Weinberger <richard@nod.at>
> ---
>  drivers/net/ethernet/via/via-rhine.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/net/ethernet/via/via-rhine.c b/drivers/net/ethernet/via/via-rhine.c
> index b75eb9e..57e1b40 100644
> --- a/drivers/net/ethernet/via/via-rhine.c
> +++ b/drivers/net/ethernet/via/via-rhine.c
> @@ -1615,6 +1615,7 @@ static void rhine_reset_task(struct work_struct *work)
>  		goto out_unlock;
>  
>  	napi_disable(&rp->napi);
> +	netif_stop_queue(dev);

This is not really fixing the bug because there is no synchronisation
with the TX scheduler.  You can call netif_tx_disable() instead to do
that.

(I also think that it is preferable to use
netif_device_{detach,attach}() to stop the queue during reconfiguration,
as this is independent of TX completions and the watchdog.)

Ben.

>  	spin_lock_bh(&rp->lock);
>  
>  	/* clear all descriptors */
Richard Weinberger - July 21, 2013, 4:32 p.m.
Am 21.07.2013 18:18, schrieb Ben Hutchings:
> On Fri, 2013-07-19 at 23:30 +0200, Richard Weinberger wrote:
>> rhine_reset_task() misses to call netif_stop_queue(),
>> this can lead to a crash if work is still scheduled while
>> we're resetting the tx queue.
>>
>> Fixes:
>> [   93.591707] BUG: unable to handle kernel NULL pointer dereference at 0000004c
>> [   93.595514] IP: [<c119d10d>] rhine_napipoll+0x491/0x6e
>>
>> Signed-off-by: Richard Weinberger <richard@nod.at>
>> ---
>>  drivers/net/ethernet/via/via-rhine.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/net/ethernet/via/via-rhine.c b/drivers/net/ethernet/via/via-rhine.c
>> index b75eb9e..57e1b40 100644
>> --- a/drivers/net/ethernet/via/via-rhine.c
>> +++ b/drivers/net/ethernet/via/via-rhine.c
>> @@ -1615,6 +1615,7 @@ static void rhine_reset_task(struct work_struct *work)
>>  		goto out_unlock;
>>  
>>  	napi_disable(&rp->napi);
>> +	netif_stop_queue(dev);
> 
> This is not really fixing the bug because there is no synchronisation
> with the TX scheduler.  You can call netif_tx_disable() instead to do
> that.

I guess other drivers suffer from that too.
A quick grep showed that not many drivers are using netif_tx_disable().

> (I also think that it is preferable to use
> netif_device_{detach,attach}() to stop the queue during reconfiguration,
> as this is independent of TX completions and the watchdog.)

So the correct down sequence is napi_disable() -> netif_tx_disable() -> netif_device_detach()?

Thanks,
//richard
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ben Hutchings - July 21, 2013, 8:24 p.m.
On Sun, 2013-07-21 at 18:32 +0200, Richard Weinberger wrote:
> Am 21.07.2013 18:18, schrieb Ben Hutchings:
> > On Fri, 2013-07-19 at 23:30 +0200, Richard Weinberger wrote:
> >> rhine_reset_task() misses to call netif_stop_queue(),
> >> this can lead to a crash if work is still scheduled while
> >> we're resetting the tx queue.
> >>
> >> Fixes:
> >> [   93.591707] BUG: unable to handle kernel NULL pointer dereference at 0000004c
> >> [   93.595514] IP: [<c119d10d>] rhine_napipoll+0x491/0x6e
> >>
> >> Signed-off-by: Richard Weinberger <richard@nod.at>
> >> ---
> >>  drivers/net/ethernet/via/via-rhine.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >>
> >> diff --git a/drivers/net/ethernet/via/via-rhine.c b/drivers/net/ethernet/via/via-rhine.c
> >> index b75eb9e..57e1b40 100644
> >> --- a/drivers/net/ethernet/via/via-rhine.c
> >> +++ b/drivers/net/ethernet/via/via-rhine.c
> >> @@ -1615,6 +1615,7 @@ static void rhine_reset_task(struct work_struct *work)
> >>  		goto out_unlock;
> >>  
> >>  	napi_disable(&rp->napi);
> >> +	netif_stop_queue(dev);
> > 
> > This is not really fixing the bug because there is no synchronisation
> > with the TX scheduler.  You can call netif_tx_disable() instead to do
> > that.
> 
> I guess other drivers suffer from that too.
> A quick grep showed that not many drivers are using netif_tx_disable().
> 
> > (I also think that it is preferable to use
> > netif_device_{detach,attach}() to stop the queue during reconfiguration,
> > as this is independent of TX completions and the watchdog.)

Actually, this is not independent of TX completions - netif_wake_queue()
will still start the TX scheduler while the device is not present, so
you have to avoid that.

> So the correct down sequence is napi_disable() -> netif_tx_disable() -> netif_device_detach()?

No, that's redundant.  You can do:

	napi_disable();
	netif_tx_lock_bh(); /* sync with TX scheduler */
	netif_device_detach();
	netif_tx_unlock_bh();

and then when the queue is ready to use again:

	netif_device_attach();
	napi_enable();

Ben.

Patch

diff --git a/drivers/net/ethernet/via/via-rhine.c b/drivers/net/ethernet/via/via-rhine.c
index b75eb9e..57e1b40 100644
--- a/drivers/net/ethernet/via/via-rhine.c
+++ b/drivers/net/ethernet/via/via-rhine.c
@@ -1615,6 +1615,7 @@  static void rhine_reset_task(struct work_struct *work)
 		goto out_unlock;
 
 	napi_disable(&rp->napi);
+	netif_stop_queue(dev);
 	spin_lock_bh(&rp->lock);
 
 	/* clear all descriptors */