Re: [PATCH net-next v3] net: phy: Fix suspicious rcu_dereference usage

On 21/01/2025 13:01, Kory Maincent wrote:

On Tue, 21 Jan 2025 11:34:48 +0000
Paul Barker [off-list ref] wrote:

quoted

On 21/01/2025 09:38, Kory Maincent wrote:

quoted

On Mon, 20 Jan 2025 11:12:28 -0800
Jakub Kicinski [off-list ref] wrote:
  

quoted

On Mon, 20 Jan 2025 15:19:25 +0100 Kory Maincent wrote:  

 [...]  

quoted

quoted

I maintain that ravb is buggy, plenty of drivers take rtnl_lock 
from the .suspend callback. We need _some_ write protection here,
the patch as is only silences a legitimate warning.  

Indeed if the suspend path is buggy we should fix it. Still there is lots of
ethernet drivers calling phy_disconnect without rtnl (IIUC) if probe return
an error or in the remove path. What should we do about it?

About ravb suspend, I don't have the board, Claudiu could you try this
instead of the current fix:

diff --git a/drivers/net/ethernet/renesas/ravb_main.c

b/drivers/net/ethernet/renesas/ravb_main.c index bc395294a32d..c9a0d2d6f371
100644 --- a/drivers/net/ethernet/renesas/ravb_main.c

+++ b/drivers/net/ethernet/renesas/ravb_main.c

@@ -3215,15 +3215,22 @@ static int ravb_suspend(struct device *dev)
        if (!netif_running(ndev))
                goto reset_assert;
 
+       rtnl_lock();
        netif_device_detach(ndev);
 
-       if (priv->wol_enabled)
-               return ravb_wol_setup(ndev);
+       if (priv->wol_enabled) {
+               ret = ravb_wol_setup(ndev);
+               rtnl_unlock();
+               return ret;
+       }
 
        ret = ravb_close(ndev);
-       if (ret)
+       if (ret) {
+               rtnl_unlock();
                return ret;
+       }
 
+       rtnl_unlock();
        ret = pm_runtime_force_suspend(&priv->pdev->dev);
        if (ret)
                return ret;

Regards,  

(Cc'ing Niklas and Sergey as this relates to the ravb driver)

Yes, thanks.

quoted

Why do we need to hold the rtnl mutex across the calls to
netif_device_detach() and ravb_wol_setup()?

My reading of Documentation/networking/netdevices.rst is that the rtnl
mutex is held when the net subsystem calls the driver's ndo_stop method,
which in our case is ravb_close(). So, we should take the rtnl mutex
when we call ravb_close() directly, in both ravb_suspend() and
ravb_wol_restore(). That would ensure that we do not call
phy_disconnect() without holding the rtnl mutex and should fix this
issue.

Not sure about it. For example ravb_ptp_stop() called in ravb_wol_setup() won't
be protected by the rtnl lock.

ravb_ptp_stop() modifies a couple of device registers and calls
ptp_clock_unregister(). I don't see anything to suggest that this
requires the rtnl lock to be held, unless I am missing something.

I don't know about netif_device_detach(). It doesn't seems to be the case as
there is lots of driver using it without holding rtnl lock.

netif_device_detach() clears the present flag from the link state and
stops all TX queues, so I don't think the rtnl lock needs to be held to
call this.

Indeed we should add the rtnl lock also in the resume path. 

quoted

Commit 35f7cad1743e ("net: Add the possibility to support a selected
hwtstamp in netdevice") may have unearthed the issue, but the fixes tag
should point to the commits adding those unlocked calls to ravb_close().

The current patch was on phy_device.c that's why the fixes tag does not point to
a ravb commit, it will change.

Ack. Thanks for digging in to this!

-- 
Paul Barker