[PATCH net] net: phy: clear phydev->devlink when the link is deleted

[PATCH net] net: phy: clear phydev->devlink when the link is deleted

[Wei Fang]

The phydev->devlink is not cleared when the link is deleted, so calling
phy_detach() again will cause a crash. For example, the link is deleted
when the network interface is set to down, then re-enable the network
interface and phy_attach_direct() will be called, but an error occurs
and jump to the error path, so phy_detach() is called again and cause a
crash, the crash log is as follows.

[   24.702421] Call trace:
[   24.704856]  device_link_put_kref+0x20/0x120
[   24.709124]  device_link_del+0x30/0x48
[   24.712864]  phy_detach+0x24/0x168
[   24.716261]  phy_attach_direct+0x168/0x3a4
[   24.720352]  phylink_fwnode_phy_connect+0xc8/0x14c
[   24.725140]  phylink_of_phy_connect+0x1c/0x34

Fixes: bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
---
 drivers/net/phy/phy_device.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
index cc1bfd22fb81..7d5e76a3db0e 100644
--- a/drivers/net/phy/phy_device.c
+++ b/drivers/net/phy/phy_device.c
@@ -1727,8 +1727,10 @@ void phy_detach(struct phy_device *phydev)
 	struct module *ndev_owner = NULL;
 	struct mii_bus *bus;
 
-	if (phydev->devlink)
+	if (phydev->devlink) {
 		device_link_del(phydev->devlink);
+		phydev->devlink = NULL;
+	}
 
 	if (phydev->sysfs_links) {
 		if (dev)
-- 
2.34.1

Re: [PATCH net] net: phy: clear phydev->devlink when the link is deleted

[Andrew Lunn]

On Thu, May 22, 2025 at 02:42:53PM +0800, Wei Fang wrote:
> The phydev->devlink is not cleared when the link is deleted, so calling
> phy_detach() again will cause a crash.

I would say crashing is correct. You have done something you should
not do, and the crash helped you find it. phy_attach() and
phy_detach() should always be in pairs.

    Andrew

---
pw-bot: cr

RE: [PATCH net] net: phy: clear phydev->devlink when the link is deleted

[Wei Fang]

> On Thu, May 22, 2025 at 02:42:53PM +0800, Wei Fang wrote:
> > The phydev->devlink is not cleared when the link is deleted, so calling
> > phy_detach() again will cause a crash.
> 
> I would say crashing is correct. You have done something you should
> not do, and the crash helped you find it. phy_attach() and
> phy_detach() should always be in pairs.
> 

phy_attach() and phy_detach() are called in pairs in my case. When
re-enabling the network port, if an error occurs in the phy_attach_direct(),
For example, if phy_init_hw() returns an error, it will jump to the error
path and call phy_detach(). Because phy_detach() did not clear the
phydev->devlink pointer when the network port was disabled,
device_link_del() will access a NULL pointer and cause a crash. And this
crash may cause the CPU to hang. I don't think it is reasonable to cause
the CPU to hang.

Re: [PATCH net] net: phy: clear phydev->devlink when the link is deleted

[Andrew Lunn]

On Thu, May 22, 2025 at 01:57:20PM +0000, Wei Fang wrote:
> > On Thu, May 22, 2025 at 02:42:53PM +0800, Wei Fang wrote:
> > > The phydev->devlink is not cleared when the link is deleted, so calling
> > > phy_detach() again will cause a crash.
> > 
> > I would say crashing is correct. You have done something you should
> > not do, and the crash helped you find it. phy_attach() and
> > phy_detach() should always be in pairs.
> > 
> 
> phy_attach() and phy_detach() are called in pairs in my case. When
> re-enabling the network port, if an error occurs in the phy_attach_direct(),
> For example, if phy_init_hw() returns an error, it will jump to the error
> path and call phy_detach(). Because phy_detach() did not clear the
> phydev->devlink pointer when the network port was disabled,
> device_link_del() will access a NULL pointer and cause a crash. And this
> crash may cause the CPU to hang. I don't think it is reasonable to cause
> the CPU to hang.

Ah, now i get it...

phy_attach_direct() runs, but fails early, before the call to:

phydev->devlink = device_link_add(dev->dev.parent, &phydev->mdio.dev,
						  DL_FLAG_PM_RUNTIME | DL_FLAG_STATELESS);

So it has its old value, from a previous attach/detach cycle. At the
error: label, it calls phy_detach(phydev), which does a
device_link_del(), using the old value...

Please improve your changelog message, add more details.

    Andrew

---
pw-bot: cr