diff mbox series

[net-next,RFC,2/5] net: ethernet: stmicro: stmmac: first disable all queues in release

Message ID 20220628013342.13581-3-ansuelsmth@gmail.com (mailing list archive)
State New, archived
Headers show
Series Add MTU change with stmmac interface running | expand

Commit Message

Christian Marangi June 28, 2022, 1:33 a.m. UTC
Disable all queues before tx_disable in stmmac_release to prevent a
corner case where packet may be still queued at the same time tx_disable
is called resulting in kernel panic if some packet still has to be
processed.

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

Comments

Jakub Kicinski June 29, 2022, 3:54 a.m. UTC | #1
On Tue, 28 Jun 2022 03:33:39 +0200 Christian Marangi wrote:
> +	stmmac_disable_all_queues(priv);
> +
> +	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
> +		hrtimer_cancel(&priv->tx_queue[chan].txtimer);

IIRC this hrtimer is to check for completions. Canceling it before
netif_tx_disable() looks odd, presumably until the queues are stopped
the timer can get scheduled again, no?

>  	netif_tx_disable(dev);
>  
>  	if (device_may_wakeup(priv->device))
> @@ -3764,11 +3769,6 @@ static int stmmac_release(struct net_device *dev)
>  	phylink_stop(priv->phylink);
>  	phylink_disconnect_phy(priv->phylink);
>  
> -	stmmac_disable_all_queues(priv);
> -
> -	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
> -		hrtimer_cancel(&priv->tx_queue[chan].txtimer);
Christian Marangi June 29, 2022, 10:44 a.m. UTC | #2
On Tue, Jun 28, 2022 at 08:54:35PM -0700, Jakub Kicinski wrote:
> On Tue, 28 Jun 2022 03:33:39 +0200 Christian Marangi wrote:
> > +	stmmac_disable_all_queues(priv);
> > +
> > +	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
> > +		hrtimer_cancel(&priv->tx_queue[chan].txtimer);
> 
> IIRC this hrtimer is to check for completions. Canceling it before
> netif_tx_disable() looks odd, presumably until the queues are stopped
> the timer can get scheduled again, no?
> 

My concern is that at the time hrtimer_cancel is called there may be
some packet that still has to be processed and this cause kernel panic
as stmmac_release free the descriptor (and tx_clean try to free garbage
pointer)

Bu honestly I put the hrtimer_cancel up to be extra safe, the main
problem here was disabling napi polling after tx_disable that I think
was wrong from the start.

> >  	netif_tx_disable(dev);
> >  
> >  	if (device_may_wakeup(priv->device))
> > @@ -3764,11 +3769,6 @@ static int stmmac_release(struct net_device *dev)
> >  	phylink_stop(priv->phylink);
> >  	phylink_disconnect_phy(priv->phylink);
> >  
> > -	stmmac_disable_all_queues(priv);
> > -
> > -	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
> > -		hrtimer_cancel(&priv->tx_queue[chan].txtimer);
diff mbox series

Patch

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index f861246de2e5..f4ba27c1c7e0 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -3756,6 +3756,11 @@  static int stmmac_release(struct net_device *dev)
 	struct stmmac_priv *priv = netdev_priv(dev);
 	u32 chan;
 
+	stmmac_disable_all_queues(priv);
+
+	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
+		hrtimer_cancel(&priv->tx_queue[chan].txtimer);
+
 	netif_tx_disable(dev);
 
 	if (device_may_wakeup(priv->device))
@@ -3764,11 +3769,6 @@  static int stmmac_release(struct net_device *dev)
 	phylink_stop(priv->phylink);
 	phylink_disconnect_phy(priv->phylink);
 
-	stmmac_disable_all_queues(priv);
-
-	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
-		hrtimer_cancel(&priv->tx_queue[chan].txtimer);
-
 	/* Free the IRQ lines */
 	stmmac_free_irq(dev, REQ_IRQ_ERR_ALL, 0);