Skip to contentSkip to conversation

net: altera_tse: fix init sequence to avoid races with register_netdev

PATCH In review Maxime Chevallier 2 participants 3 messages Jun 11, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: altera_tse: fix init sequence to avoid races with register_netdev

As reported here[1], the init sequence in altera_tse can be racy should any operation on the registered netdev happen after netdev registration but before phylink initialization.

Fix the registering order to avoid such races by making register_netdev the last step of the probing sequence.

[1]: https://lore.kernel.org/netdev/ZH9XK5yEGyoDMIs%[email protected]/

Fixes fef2998203e1 ("net: altera: tse: convert to phylink")

---
This patch targets net-next as it fixes a commit that is in net-next too.

 drivers/net/ethernet/altera/altera_tse_main.c | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)
drivers/net/ethernet/altera/altera_tse_main.c +6 -8
index 2e15800e5310..54f1b5ad84ce 100644
@@ -1395,12 +1395,6 @@ static int altera_tse_probe(struct platform_device *pdev)
 	spin_lock_init(&priv->rxdma_irq_lock);
 
 	netif_carrier_off(ndev);
-	ret = register_netdev(ndev);
-	if (ret) {
-		dev_err(&pdev->dev, "failed to register TSE net device\n");
-		goto err_register_netdev;
-	}
-
 	platform_set_drvdata(pdev, ndev);
 
 	priv->revision = ioread32(&priv->mac_dev->megacore_revision);
@@ -1449,12 +1443,16 @@ static int altera_tse_probe(struct platform_device *pdev)
 		goto err_init_phylink;
 	}
 
+	ret = register_netdev(ndev);
+	if (ret) {
+		dev_err(&pdev->dev, "failed to register TSE net device\n");
+		return ret;
+	}
+
 	return 0;
 err_init_phylink:
 	lynx_pcs_destroy(priv->pcs);
 err_init_pcs:
-	unregister_netdev(ndev);
-err_register_netdev:
 	netif_napi_del(&priv->napi);
 	altera_tse_mdio_destroy(ndev);
 err_free_netdev:
Signature · 1 line
2.40.1
#1 lore raw
Quoted 7 lines · Maxime Chevallier

As reported here[1], the init sequence in altera_tse can be racy should any operation on the registered netdev happen after netdev registration but before phylink initialization.

Fix the registering order to avoid such races by making register_netdev the last step of the probing sequence.

Quoted 7 lines

[1]: https://lore.kernel.org/netdev/ZH9XK5yEGyoDMIs%[email protected]/

Fixes fef2998203e1 ("net: altera: tse: convert to phylink")

---
This patch targets net-next as it fixes a commit that is in net-next too.

While it fixes the order in which stuff is registered, it introduces a new bug - register_netdev() is what atomically allocated a netdev name, and you use the netdev name when creating the PCS MII bus:

	snprintf(mrc.name, MII_BUS_ID_SIZE, "%s-pcs-mii", ndev->name);

This needs to change, because this will end up being "eth%d-pcs-mii".

I am at a loss why you didn't realise this (or in fact recognise that there are other issues) given that I sent you three patches fixing this mess.

Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
#2 lore raw

Hi,

Quoted 20 lines · On Sun, 11 Jun 2023 16:42:26 +0100 "Russell King (Oracle)"…
Quoted 7 lines · Maxime Chevallier

As reported here[1], the init sequence in altera_tse can be racy should any operation on the registered netdev happen after netdev registration but before phylink initialization.

Fix the registering order to avoid such races by making register_netdev the last step of the probing sequence.

Quoted 7 lines

[1]: https://lore.kernel.org/netdev/ZH9XK5yEGyoDMIs%[email protected]/

Fixes fef2998203e1 ("net: altera: tse: convert to phylink")

---
This patch targets net-next as it fixes a commit that is in net-next too.

While it fixes the order in which stuff is registered, it introduces a new bug - register_netdev() is what atomically allocated a netdev name, and you use the netdev name when creating the PCS MII bus:

Quoted 8 lines
	snprintf(mrc.name, MII_BUS_ID_SIZE, "%s-pcs-mii", ndev->name);

This needs to change, because this will end up being "eth%d-pcs-mii".

I am at a loss why you didn't realise this (or in fact recognise that there are other issues) given that I sent you three patches fixing this mess.

Hmpf your patches went un-noticed, sorry... I'll base later work on that.

I will also followup with a similar series for dwmac-socfpga

Best regards,

Maxime