diff mbox series

[2/2] net: macb: Unregister nedev before MDIO bus in remove

Message ID 20231126141046.3505343-3-claudiu.beznea@tuxon.dev
State New
Headers
Series net: macb: Fixes for macb driver |

Commit Message

claudiu beznea Nov. 26, 2023, 2:10 p.m. UTC 
  unregister_netdev() calls the struct net_device_ops::ndo_stop function,
which in the case of the macb driver is macb_close(). macb_close() calls,
in turn, PHY-specific APIs (e.g., phy_detach()). The call trace is as
follows:

macb_close() ->
  phylink_disconnect_phy() ->
    phy_disconnect() ->
      phy_detach()

phy_detach() will remove associated sysfs files by calling
kernfs_remove_by_name_ns(), which will hit
"kernfs: cannot remove 'attached_dev', no directory" WARN(), which will
throw a stack trace too.

To avoid this, call unregiser_netdev() before mdiobus_unregister() and
mdiobus_free().

Fixes: 7897b071ac3b ("net: macb: convert to phylink")
Signed-off-by: Claudiu Beznea <claudiu.beznea@tuxon.dev>
---
 drivers/net/ethernet/cadence/macb_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

claudiu beznea Nov. 27, 2023, 6:11 a.m. UTC | #1 
On 26.11.2023 19:13, Andrew Lunn wrote:
>> diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
>> index cebae0f418f2..73d041af3de1 100644
>> --- a/drivers/net/ethernet/cadence/macb_main.c
>> +++ b/drivers/net/ethernet/cadence/macb_main.c
>> @@ -5165,11 +5165,11 @@ static void macb_remove(struct platform_device *pdev)
>>  
>>  	if (dev) {
>>  		bp = netdev_priv(dev);
>> +		unregister_netdev(dev);
>>  		phy_exit(bp->sgmii_phy);
>>  		mdiobus_unregister(bp->mii_bus);
>>  		mdiobus_free(bp->mii_bus);
>>  
>> -		unregister_netdev(dev);
>>  		tasklet_kill(&bp->hresp_err_tasklet);
> 
> 
> I don't know this driver...
> 
> What does this tasklet do? 

It handles bus errors that my happens while DMA fetches data from system
memory. It re-initializes the TX/RX, DMA buffers, clear interrupts,
stop/start all tx queues.

> Is it safe for it to run after the netdev
> is unregistered, and the PHY and the mdio bus is gone? 

Not really, as it accesses netdev specific data.

> Maybe this
> tasklet_kill should be after the interrupt is disabled, but before the
> netdev is unregistered?

That would be a better place, indeed.

Thank you,
Claudiu Beznea

> 
> If you have one bug here, there might be others.
> 
> 	Andrew
diff mbox series

Patch

diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index cebae0f418f2..73d041af3de1 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -5165,11 +5165,11 @@  static void macb_remove(struct platform_device *pdev)
 
 	if (dev) {
 		bp = netdev_priv(dev);
+		unregister_netdev(dev);
 		phy_exit(bp->sgmii_phy);
 		mdiobus_unregister(bp->mii_bus);
 		mdiobus_free(bp->mii_bus);
 
-		unregister_netdev(dev);
 		tasklet_kill(&bp->hresp_err_tasklet);
 		pm_runtime_disable(&pdev->dev);
 		pm_runtime_dont_use_autosuspend(&pdev->dev);