Re: [BUG] via-rhine: NOHZ: local_softirq_pending 08

[Jamie Gloudon <jamie.gloudon@gmail.com>]

On Tue, Jan 29, 2013 at 10:53:28PM -0500, David Miller wrote:
> From: Francois Romieu <romieu@fr.zoreil.com>
> Date: Wed, 30 Jan 2013 01:04:32 +0100
> 
> > David Miller <davem@davemloft.net> :
> > [...]
> >> I don't think you can't keep using dev_kfree_skb_irq() there.
> >> 
> >> Francois any objections to this patch?
> > 
> > None.
> > 
> > I have almost surely broken via-rhine with NOHZ as I broke the r8169
> > driver before 7dbb491878a2c51d372a8890fa45a8ff80358af1. Something like
> > the patch below could be needed as well (quick shot before bed time
> > and ~24h delay if Jamie wants to try something now).
> 
> Good catch, Jamie can you test this combined patch for us?
> 
> ====================
> via-rhine: Fix bugs in NAPI support.
> 
> 1) rhine_tx() should use dev_kfree_skb() not dev_kfree_skb_irq()
> 
> 2) rhine_slow_event_task's NAPI triggering logic is racey, it
>    should just hit the interrupt mask register.  This is the
>    same as commit 7dbb491878a2c51d372a8890fa45a8ff80358af1
>    ("r8169: avoid NAPI scheduling delay.") made to fix the same
>    problem in the r8169 driver.  From Francois Romieu.
> 
> Reported-by: Jamie Gloudon <jamie.gloudon@gmail.com>
> Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> diff --git a/drivers/net/ethernet/via/via-rhine.c b/drivers/net/ethernet/via/via-rhine.c
> index 7992b3e..78ace59 100644
> --- a/drivers/net/ethernet/via/via-rhine.c
> +++ b/drivers/net/ethernet/via/via-rhine.c
> @@ -1801,7 +1801,7 @@ static void rhine_tx(struct net_device *dev)
>  					 rp->tx_skbuff[entry]->len,
>  					 PCI_DMA_TODEVICE);
>  		}
> -		dev_kfree_skb_irq(rp->tx_skbuff[entry]);
> +		dev_kfree_skb(rp->tx_skbuff[entry]);
>  		rp->tx_skbuff[entry] = NULL;
>  		entry = (++rp->dirty_tx) % TX_RING_SIZE;
>  	}
> @@ -2010,11 +2010,7 @@ static void rhine_slow_event_task(struct work_struct *work)
>  	if (intr_status & IntrPCIErr)
>  		netif_warn(rp, hw, dev, "PCI error\n");
>  
> -	napi_disable(&rp->napi);
> -	rhine_irq_disable(rp);
> -	/* Slow and safe. Consider __napi_schedule as a replacement ? */
> -	napi_enable(&rp->napi);
> -	napi_schedule(&rp->napi);
> +	iowrite16(RHINE_EVENT & 0xffff, rp->base + IntrEnable);
>  
>  out_unlock:
>  	mutex_unlock(&rp->task_lock);

I did some quick netperf test with both patches applied, everything
appears to be fine (no kernel panic or tput performance degradation).