[PATCH] sky2: Avoid race in sky2_change_mtu

[PATCH] sky2: Avoid race in sky2_change_mtu

[Mike McCormack]

netif_stop_queue does not ensure all in-progress transmits are complete,
 so use netif_tx_disable() instead.

Make sure NAPI polls are disabled, otherwise NAPI might trigger a TX
 restart between when we stop the queue and NAPI is disabled.

Signed-off-by: Mike McCormack <mikem@ring3k.org>
---
 drivers/net/sky2.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/net/sky2.c b/drivers/net/sky2.c
index 088c797..b839bae 100644
--- a/drivers/net/sky2.c
+++ b/drivers/net/sky2.c
@@ -2236,8 +2236,8 @@ static int sky2_change_mtu(struct net_device *dev, int new_mtu)
 	sky2_write32(hw, B0_IMSK, 0);
 
 	dev->trans_start = jiffies;	/* prevent tx timeout */
-	netif_stop_queue(dev);
 	napi_disable(&hw->napi);
+	netif_tx_disable(dev);
 
 	synchronize_irq(hw->pdev->irq);
 
-- 
1.5.6.5

Re: [PATCH] sky2: Avoid race in sky2_change_mtu

[David Miller]

From: Mike McCormack <mikem@ring3k.org>
Date: Mon, 03 May 2010 23:18:56 +0900

> netif_stop_queue does not ensure all in-progress transmits are complete,
>  so use netif_tx_disable() instead.
> 
> Make sure NAPI polls are disabled, otherwise NAPI might trigger a TX
>  restart between when we stop the queue and NAPI is disabled.
> 
> Signed-off-by: Mike McCormack <mikem@ring3k.org>

This looks quite reasonable, Stephen please review.

Re: [PATCH] sky2: Avoid race in sky2_change_mtu

[Stephen Hemminger]

On Mon, 03 May 2010 23:18:56 +0900
Mike McCormack <mikem@ring3k.org> wrote:

> netif_stop_queue does not ensure all in-progress transmits are complete,
>  so use netif_tx_disable() instead.
> 
> Make sure NAPI polls are disabled, otherwise NAPI might trigger a TX
>  restart between when we stop the queue and NAPI is disabled.

How could NAPI trigger a TX restart? Restart is a timer, not controlled
by NAPI. There is a different race that your patch isn't covering.

If a large MTU send is in progress while mtu is changing, the changes
to the GMAC store and forward (sky2_set_tx_stfwd) are not synchronized
with hardware queued transmits.

So I think a there needs to be another spinloop.

--- a/drivers/net/sky2.c	2010-05-03 16:19:58.828106182 -0700
+++ b/drivers/net/sky2.c	2010-05-03 16:24:37.850232475 -0700
@@ -2274,8 +2274,12 @@ static int sky2_change_mtu(struct net_de
 	imask = sky2_read32(hw, B0_IMSK);
 	sky2_write32(hw, B0_IMSK, 0);
 
-	dev->trans_start = jiffies;	/* prevent tx timeout */
 	netif_stop_queue(dev);
+
+	/* Wait for ongoing Tx to complete */
+	while (sky2->tx_cons != sky2->tx_prod)
+		udelay(10);
+
 	napi_disable(&hw->napi);
 
 	synchronize_irq(hw->pdev->irq);

Re: [PATCH] sky2: Avoid race in sky2_change_mtu

[David Miller]

From: Stephen Hemminger <shemminger@vyatta.com>
Date: Mon, 3 May 2010 16:26:34 -0700

> On Mon, 03 May 2010 23:18:56 +0900
> Mike McCormack <mikem@ring3k.org> wrote:
> 
>> netif_stop_queue does not ensure all in-progress transmits are complete,
>>  so use netif_tx_disable() instead.
>> 
>> Make sure NAPI polls are disabled, otherwise NAPI might trigger a TX
>>  restart between when we stop the queue and NAPI is disabled.
> 
> How could NAPI trigger a TX restart? Restart is a timer, not controlled
> by NAPI. There is a different race that your patch isn't covering.

I think he meant to say that NAPI can trigger a TX queue wake, which
indeed it can, and it will if there is space in the TX queue and
we happen to see the TXIDLE status at this time.