Skip to contentSkip to conversation

net: fec: only stop PTP if it was initialized

PATCH Applied [email protected] 3 participants 3 messages Aug 26 – Aug 27

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 net: fec: only stop PTP if it was initialized

From: bui duc phuc <[email protected]>

fec_ptp_init() is only called when fep->bufdesc_ex is available. However, fec_probe() unconditionally calls fec_ptp_stop() on the failed_init path, and fec_drv_remove() unconditionally calls fec_ptp_stop() during device removal.

Check fep->bufdesc_ex before calling fec_ptp_stop() in both paths to avoid stopping PTP when it was not initialized.

Fixes 32cba57ba74b ("net: fec: introduce fec_ptp_stop and use in probe fail path")

---

Changes in v2:
- Handle fec_drv_remove() as suggested in Sashiko's review.
- Update the commit message.
- Add the Reviewed-by tag.

 drivers/net/ethernet/freescale/fec_main.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/ethernet/freescale/fec_main.c +4 -2
index ced4dbf8cd90..794ec427b0ee 100644
@@ -5457,7 +5457,8 @@ fec_probe(struct platform_device *pdev)
 failed_irq:
 	fec_enet_deinit(ndev);
 failed_init:
-	fec_ptp_stop(pdev);
+	if (fep->bufdesc_ex)
+		fec_ptp_stop(pdev);
 failed_reset:
 	pm_runtime_put_noidle(&pdev->dev);
 	pm_runtime_disable(&pdev->dev);
@@ -5499,7 +5500,8 @@ fec_drv_remove(struct platform_device *pdev)
 			ERR_PTR(ret));
 
 	cancel_work_sync(&fep->tx_timeout_work);
-	fec_ptp_stop(pdev);
+	if (fep->bufdesc_ex)
+		fec_ptp_stop(pdev);
 	unregister_netdev(ndev);
 	fec_enet_mii_remove(fep);
 	if (fep->reg_phy)
Signature · 1 line
2.43.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 7 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 7 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Hardware access with disabled clocks in fec_drv_remove() causing kernel panic
- [Critical] PTP clock ioctls access hardware directly without verifying clock status, leading to kernel panic
- [High] Race condition causing Use-After-Free during fec_drv_remove() watchdog teardown
- [High] Initialization order race in fec_probe() causing workqueue corruption
- [High] Race condition causing Use-After-Free during PTP timer teardown in fec_ptp_stop()
- [High] Use-After-Free in probe error path due to premature explicit freeing of network device
- [High] Concurrent fec_pps_interrupt execution and fec_ptp_stop causes Use-After-Free --

commit e0b7dcf388492f09f947c8b3bfed37b87cb8fc59
Author: bui duc phuc <[email protected]>

net: fec: only stop PTP if it was initialized

fec_ptp_init() is only called when fep->bufdesc_ex is available. However, fec_probe() unconditionally calls fec_ptp_stop() on the failed_init path, and fec_drv_remove() unconditionally calls fec_ptp_stop() during device removal.

Check fep->bufdesc_ex before calling fec_ptp_stop() in both paths to avoid stopping PTP when it was not initialized.

drivers/net/ethernet/freescale/fec_main.c fec_probe(struct platform_device *pdev) · line 5461

Quoted 11 lines
drivers/net/ethernet/freescale/fec_main.c +2 -1
index ced4dbf8cd90f..794ec427b0ee6 100644
@@ -5457,7 +5457,8 @@ fec_probe(struct platform_device *pdev)
 failed_irq:
 	fec_enet_deinit(ndev);
 failed_init:
-	fec_ptp_stop(pdev);
+	if (fep->bufdesc_ex)
+		fec_ptp_stop(pdev);

[Severity: High]
This is a pre-existing issue, but does the initialization order in fec_probe() create a race condition that corrupts the workqueue?

Currently, the device is registered and exposed to the system before the TX timeout work is initialized:

fec_probe() {
    ...
    ret = register_netdev(ndev);
    if (ret)
        goto failed_register;
    ...
    INIT_WORK(&fep->tx_timeout_work, fec_enet_timeout_work);
    ...
}

If a fast userspace actor brings the interface up and triggers a TX timeout after register_netdev() but before INIT_WORK(), .ndo_tx_timeout (fec_timeout) could call schedule_work() on an uninitialized work structure, potentially corrupting the workqueue. Should INIT_WORK() run before register_netdev()?

[Severity: High]
This is a pre-existing issue, but is there a use-after-free risk in the probe error path?

At the end of fec_probe():

failed_ioremap:

    free_netdev(ndev);
    return ret;

Since devm_request_irq() was used for the interrupt handler, the IRQ is not freed until after fec_probe() returns. Because fec_enet_init() writes to the MAC interrupt mask register to enable TX interrupts, a pending interrupt could fire during this window.

If fec_enet_interrupt() executes, it would dereference the already-freed ndev pointer. Would it be safer to free the netdev via devres, or manually disable interrupts/free the IRQ before freeing ndev?

[Severity: High]
This is a pre-existing issue, but does the timer teardown sequence in fec_ptp_stop() properly protect against concurrent ioctls?

fec_ptp_stop() {
    ...
    cancel_delayed_work_sync(&fep->time_keep);
    hrtimer_cancel(&fep->perout_timer);
    if (fep->ptp_clock)
        ptp_clock_unregister(fep->ptp_clock);
}

The PTP timers are cancelled before the PTP clock character device is unregistered. Could a concurrent ioctl (like PTP_PEROUT_REQUEST) invoke fec_ptp_enable() and re-arm the timers during unregistration?

If so, could the re-armed timers fire after the driver is removed and access freed memory? Should ptp_clock_unregister() be called first to block new ioctl requests?

[Severity: High]
This is a pre-existing issue, but does fec_ptp_stop() synchronize with a currently running PPS interrupt handler?

fec_ptp_stop() disables the PPS interrupt at the hardware level but does not call synchronize_irq(). It then unregisters the PTP clock:

    if (fep->ptp_clock)
        ptp_clock_unregister(fep->ptp_clock);

If a concurrent fec_pps_interrupt() is running on another CPU, could it dereference fep->ptp_clock via ptp_clock_event(fep->ptp_clock, &event) after the clock instance has been unregistered and freed?

[Severity: Critical]
This is a pre-existing issue, but do the PTP clock ioctls safely access hardware when the device is runtime-suspended?

Functions like fec_ptp_adjfine() perform direct register writes:

fec_ptp_adjfine() {
    ...
    tmp = readl(fep->hwp + FEC_ATIME_INC) & FEC_T_INC_MASK;
    tmp |= corr_ns << FEC_T_INC_CORR_OFFSET;
    writel(tmp, fep->hwp + FEC_ATIME_INC);
}

Unlike fec_ptp_gettime(), which checks fep->ptp_clk_on under a mutex, fec_ptp_adjfine() doesn't seem to verify if the underlying clocks are active. If an ioctl triggers this while the network interface is down and clocks are disabled, could this direct hardware access cause a synchronous abort and crash the system?

Quoted 3 lines
drivers/net/ethernet/freescale/fec_main.c +0 -0
 failed_reset:
 	pm_runtime_put_noidle(&pdev->dev);
 	pm_runtime_disable(&pdev->dev);

[ ... ]

drivers/net/ethernet/freescale/fec_main.c fec_drv_remove(struct platform_device *pdev) · line 5502

Quoted 4 lines
drivers/net/ethernet/freescale/fec_main.c +0 -0
@@ -5499,7 +5500,8 @@ fec_drv_remove(struct platform_device *pdev)
 			ERR_PTR(ret));
 
 	cancel_work_sync(&fep->tx_timeout_work);

[Severity: Critical]
This is a pre-existing issue, but does ignoring a runtime PM resume failure here lead to a kernel panic?

Just above this context, pm_runtime_get_sync() is called:

    ret = pm_runtime_get_sync(&pdev->dev);
    if (ret < 0)
        dev_err(&pdev->dev, ...);

If pm_runtime_get_sync() fails, the device clocks remain disabled. However, execution proceeds to fec_ptp_stop(), which eventually calls fec_ptp_enable_pps() and performs direct register writes (writel(0, fep->hwp + FEC_TCSR(fep->pps_channel))).

Could writing to the SoC registers while their backing clocks are disabled cause a synchronous external abort? Should the teardown path gracefully handle this resume failure?

drivers/net/ethernet/freescale/fec_main.c fec_drv_remove(struct platform_device *pdev) · line 5505

Quoted 4 lines
drivers/net/ethernet/freescale/fec_main.c +2 -1
-	fec_ptp_stop(pdev);
+	if (fep->bufdesc_ex)
+		fec_ptp_stop(pdev);
 	unregister_netdev(ndev);

[Severity: High]
This is a pre-existing issue, but does cancelling the timeout work before unregistering the network device leave a race condition window?

In the current sequence:

    cancel_work_sync(&fep->tx_timeout_work);
    if (fep->bufdesc_ex)
        fec_ptp_stop(pdev);
    unregister_netdev(ndev);

The network device is still fully active when cancel_work_sync() is called. Could the network core's watchdog timer fire right after the cancellation, invoke .ndo_tx_timeout, and re-queue the work item via schedule_work()?

If that happens, and the driver subsequently unregisters and frees the device, could the re-armed work item execute later and cause a use-after-free?

Quoted 2 lines
drivers/net/ethernet/freescale/fec_main.c +0 -0
 	fec_enet_mii_remove(fep);
 	if (fep->reg_phy)
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw
In reply to [email protected]
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:

Quoted 8 lines · On Wed, 26 Aug 2026 17:34:28 +0700 you wrote:

From: bui duc phuc <[email protected]>

fec_ptp_init() is only called when fep->bufdesc_ex is available. However, fec_probe() unconditionally calls fec_ptp_stop() on the failed_init path, and fec_drv_remove() unconditionally calls fec_ptp_stop() during device removal.

[...]

Here is the summary with links:
- [net,v2] net: fec: only stop PTP if it was initialized

    https://git.kernel.org/netdev/net/c/dd890ae29299

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html