Skip to contentSkip to conversation

watchdog: ni903x_wdt: Check ACPI_COMPANION() against NULL

PATCH Applied Rafael J. Wysocki 4 participants 4 messages May 12 – May 13

Subsystems (2)
lore
Share & tools
lore

Rafael J. Wysocki

author maintainer
#0 lore raw

PATCH 1/1 watchdog: ni903x_wdt: Check ACPI_COMPANION() against NULL

From: "Rafael J. Wysocki" <[email protected]>

Every platform driver can be forced to match a device that doesn't match its list of device IDs because of device_match_driver_override(), so platform drivers that rely on the existence of a device's ACPI companion object need to verify its presence.

Accordingly, add a requisite ACPI_COMPANION() check against NULL to the ni903x_wdt watchdog driver.

Fixes d37ec2fbab55 ("watchdog: ni903x_wdt: Convert to a platform driver")
---
 drivers/watchdog/ni903x_wdt.c |    7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/watchdog/ni903x_wdt.c +6 -1
@@ -183,9 +183,14 @@ static int ni903x_acpi_probe(struct plat
 	struct device *dev = &pdev->dev;
 	struct watchdog_device *wdd;
 	struct ni903x_wdt *wdt;
+	acpi_handle handle;
 	acpi_status status;
 	int ret;
 
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+
 	wdt = devm_kzalloc(dev, sizeof(*wdt), GFP_KERNEL);
 	if (!wdt)
 		return -ENOMEM;
@@ -193,7 +198,7 @@ static int ni903x_acpi_probe(struct plat
 	platform_set_drvdata(pdev, wdt);
 	wdt->dev = dev;
 
-	status = acpi_walk_resources(ACPI_HANDLE(dev), METHOD_NAME__CRS,
+	status = acpi_walk_resources(handle, METHOD_NAME__CRS,
 				     ni903x_resources, wdt);
 	if (ACPI_FAILURE(status) || wdt->io_base == 0) {
 		dev_err(dev, "failed to get resources\n");
Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Rafael J. Wysocki

Every platform driver can be forced to match a device that doesn't match its list of device IDs because of device_match_driver_override(), so platform drivers that rely on the existence of a device's ACPI companion object need to verify its presence.

Accordingly, add a requisite ACPI_COMPANION() check against NULL to the ni903x_wdt watchdog driver.

...

drivers/watchdog/ni903x_wdt.c static int ni903x_acpi_probe(struct plat · line 202

Quoted 3 lines
drivers/watchdog/ni903x_wdt.c +1 -1
-	status = acpi_walk_resources(ACPI_HANDLE(dev), METHOD_NAME__CRS,
+	status = acpi_walk_resources(handle, METHOD_NAME__CRS,
 				     ni903x_resources, wdt);

This smells like we can move to regular acpi_dev_*() resource APIs rather than custom walking via _CRS. But I haven't looked into the code.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
maintainer Applied
#2 lore raw
In reply to Rafael J. Wysocki
Quoted 7 lines · Rafael J. Wysocki

From: "Rafael J. Wysocki" <[email protected]>

Every platform driver can be forced to match a device that doesn't match its list of device IDs because of device_match_driver_override(), so platform drivers that rely on the existence of a device's ACPI companion object need to verify its presence.

Quoted 6 lines

Accordingly, add a requisite ACPI_COMPANION() check against NULL to the ni903x_wdt watchdog driver.

Fixes d37ec2fbab55 ("watchdog: ni903x_wdt: Convert to a platform driver")

Applied.

Thanks,
Guenter

#3 lore raw
In reply to Rafael J. Wysocki
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [High] The watchdog device is exposed to userspace before its hardware initialization is fully complete…

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] The watchdog device is exposed to userspace before its hardware initialization is fully complete in ni903x_acpi_probe.
- [High] The driver's removal function races with the watchdog core worker and userspace, potentially leaving the hardware running after driver unload. --

commit 6254d703c5994e457b8ddb7846a5094ab95cecda
Author: Rafael J. Wysocki <[email protected]>

watchdog: ni903x_wdt: Check ACPI_COMPANION() against NULL

This commit adds a check to ensure the ACPI companion object exists before proceeding with the ni903x_wdt watchdog driver probe.

Quoted 5 lines
drivers/watchdog/ni903x_wdt.c +0 -0
@@ -183,9 +183,14 @@ static int ni903x_acpi_probe(struct platform_device *pdev)
 	struct device *dev = &pdev->dev;

drivers/watchdog/ni903x_wdt.c static int ni903x_acpi_probe(struct platform_device *pdev) · line 193

Quoted 10 lines
drivers/watchdog/ni903x_wdt.c +5 -0
 	struct watchdog_device *wdd;
 	struct ni903x_wdt *wdt;
+	acpi_handle handle;
 	acpi_status status;
 	int ret;
 
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+

This isn't a regression introduced by this commit, but while reviewing the probe function modified here, I noticed a potential race condition.

In ni903x_acpi_probe(), the watchdog device is registered before the hardware initialization is fully complete:

ni903x_acpi_probe() {
    ...
    ret = watchdog_register_device(wdd);
    if (ret)
        return ret;

    outb(NIWD_CONTROL_RESET | NIWD_CONTROL_MODE,
         wdt->io_base + NIWD_CONTROL);
    ...
}

Since the character device becomes visible to userspace immediately upon registration, could a fast userspace process open the device and trigger ni903x_wdd_start() before the hardware is initialized?

If ni903x_wdd_start() configures the watchdog by writing to the control register, wouldn't the subsequent outb() in the probe thread clobber this configuration without preserving the bits set by userspace?

This might leave the watchdog in an unpredictable state and prevent recovery on a freeze.

There is also a similar pre-existing issue in the removal path.

If ni903x_acpi_remove() calls ni903x_wdd_stop() before watchdog_unregister_device(), the device remains registered with the watchdog core while stopping.

Could userspace or the watchdog core worker concurrently execute ni903x_wdd_ping() during this time?

If ni903x_wdd_ping() performs a read-modify-write on the control register, and ni903x_wdd_stop() writes a stop command to the same register, is it possible for the ping operation to write back the active state and re-enable the watchdog if the stop operation executes between the read and write?

If the driver then unloads while the hardware watchdog continues running with no driver to pet it, will the system spuriously reboot?

Would it be safer to call watchdog_unregister_device() before stopping the hardware, or perhaps use watchdog_stop_on_unregister()?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1