[PATCH net v2 RESEND] net: phy: nxp-tja11xx: cancel registration work on remove

[PATCH net v2 RESEND] net: phy: nxp-tja11xx: cancel registration work on remove

[Hongyan Xu]

tja1102_p0_probe() schedules work to register the second port. That work
uses the Port 0 private data and phydev. The private data is
devm-allocated, but the driver does not cancel the work before detach.

If the device is removed before the worker runs, devres can free the
private data while the pending work still dereferences it.

Store the Port 0 private data in phydev->priv and add a remove callback.
The callback waits for the registration work before devres can free it.

This issue was found by a static analysis tool.

Fixes: 6a64d3cdc5ef ("net: phy: tja11xx: add delayed registration of TJA1102 PHY1")
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Hongyan Xu <getshell@seu.edu.cn>
---
RESEND:
- resend v2 as a standalone thread as required by netdev
- no code changes

v2: https://lore.kernel.org/netdev/20260729091621.1726-1-getshell@seu.edu.cn/
- mark the patch for the net tree as suggested by Andrew Lunn
- add the Fixes tag
- no code changes

v1: https://lore.kernel.org/netdev/20260728123423.781-1-getshell@seu.edu.cn/

 drivers/net/phy/nxp-tja11xx.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/net/phy/nxp-tja11xx.c b/drivers/net/phy/nxp-tja11xx.c
index 07e94a247..68062a4c5 100644
--- a/drivers/net/phy/nxp-tja11xx.c
+++ b/drivers/net/phy/nxp-tja11xx.c
@@ -620,6 +620,7 @@ static int tja1102_p0_probe(struct phy_device *phydev)
 		return -ENOMEM;
 
 	priv->phydev = phydev;
+	phydev->priv = priv;
 	INIT_WORK(&priv->phy_register_work, tja1102_p1_register);
 
 	ret = tja11xx_hwmon_register(phydev, priv);
@@ -631,6 +632,13 @@ static int tja1102_p0_probe(struct phy_device *phydev)
 	return 0;
 }
 
+static void tja1102_p0_remove(struct phy_device *phydev)
+{
+	struct tja11xx_priv *priv = phydev->priv;
+
+	cancel_work_sync(&priv->phy_register_work);
+}
+
 static int tja1102_match_phy_device(struct phy_device *phydev, bool port0)
 {
 	int ret;
@@ -847,6 +855,7 @@ static struct phy_driver tja11xx_driver[] = {
 		.features       = PHY_BASIC_T1_FEATURES,
 		.flags          = PHY_POLL_CABLE_TEST,
 		.probe		= tja1102_p0_probe,
+		.remove		= tja1102_p0_remove,
 		.soft_reset	= tja11xx_soft_reset,
 		.config_aneg	= tja11xx_config_aneg,
 		.config_init	= tja11xx_config_init,
-- 
2.50.1.windows.1

Re: [PATCH net v2 RESEND] net: phy: nxp-tja11xx: cancel registration work on remove

[Andrew Lunn]

On Wed, Jul 29, 2026 at 05:27:14PM +0800, Hongyan Xu wrote:
> tja1102_p0_probe() schedules work to register the second port. That work
> uses the Port 0 private data and phydev. The private data is
> devm-allocated, but the driver does not cancel the work before detach.
> 
> If the device is removed before the worker runs, devres can free the
> private data while the pending work still dereferences it.
> 
> Store the Port 0 private data in phydev->priv and add a remove callback.
> The callback waits for the registration work before devres can free it.
> 
> This issue was found by a static analysis tool.
> 
> Fixes: 6a64d3cdc5ef ("net: phy: tja11xx: add delayed registration of TJA1102 PHY1")

I'm not sure this is worth a fixes tag. Does it bother people?

Have you managed to trigger it? It would need a script looking for the
first instance to pop into exists, and then unbinding the driver as
soon as it is created.

To me, this is yet another theoretical problem which never happens in
real life.

     Andrew