starfive_wdt_pm_start() calls pm_runtime_get_sync() and returns its
error code directly on failure. Per Documentation/power/runtime_pm.rst,
pm_runtime_get_sync() does not drop the usage counter on error, so the
caller is expected to release the reference itself. The early return
here skips that, leaking a runtime PM usage count on wdd->parent on
every failed resume, which prevents the device from being suspended
correctly afterwards.

Fix it by calling pm_runtime_put_noidle() before returning, mirroring
the same balance the counterpart starfive_wdt_pm_stop() already
maintains with pm_runtime_put_sync().

v1 only balanced the pm_runtime_get_sync() failure path. As Guenter
Roeck and the Sashiko AI review both pointed out, the same leak
remains if the subsequent starfive_wdt_start() call fails: the
runtime PM reference acquired by pm_runtime_get_sync() would still be
held while returning an error, again leaving the device unable to
suspend. Drop the reference on that failure path too.

Signed-off-by: manushprajwal <manushprajwal555@gmail.com>
---
v2: Also drop the pm_runtime reference when starfive_wdt_start() fails,
    not just when pm_runtime_get_sync() fails, per Guenter Roeck and
    the Sashiko AI review on v1.

 drivers/watchdog/starfive-wdt.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/watchdog/starfive-wdt.c b/drivers/watchdog/starfive-wdt.c
index af55adc4a..b1c2d3e4f 100644
--- a/drivers/watchdog/starfive-wdt.c
+++ b/drivers/watchdog/starfive-wdt.c
@@ -373,8 +373,15 @@ static int starfive_wdt_pm_start(struct watchdog_device *wdd)
 	struct starfive_wdt *wdt = watchdog_get_drvdata(wdd);
 	int ret = pm_runtime_get_sync(wdd->parent);

-	if (ret < 0)
+	if (ret < 0) {
+		pm_runtime_put_noidle(wdd->parent);
 		return ret;
+	}

-	return starfive_wdt_start(wdt);
+	ret = starfive_wdt_start(wdt);
+	if (ret)
+		pm_runtime_put_noidle(wdd->parent);
+
+	return ret;
 }
--
2.46.2.windows.1

On Sun, Aug 23, 2026 at 09:14:59PM +0530, Manush Prajwal wrote:
> starfive_wdt_pm_start() calls pm_runtime_get_sync() and returns its
> error code directly on failure. Per Documentation/power/runtime_pm.rst,
> pm_runtime_get_sync() does not drop the usage counter on error, so the
> caller is expected to release the reference itself. The early return
> here skips that, leaking a runtime PM usage count on wdd->parent on
> every failed resume, which prevents the device from being suspended
> correctly afterwards.
> 
> Fix it by calling pm_runtime_put_noidle() before returning, mirroring
> the same balance the counterpart starfive_wdt_pm_stop() already
> maintains with pm_runtime_put_sync().
> 
> v1 only balanced the pm_runtime_get_sync() failure path. As Guenter
> Roeck and the Sashiko AI review both pointed out, the same leak
> remains if the subsequent starfive_wdt_start() call fails: the
> runtime PM reference acquired by pm_runtime_get_sync() would still be
> held while returning an error, again leaving the device unable to
> suspend. Drop the reference on that failure path too.
> 
> Signed-off-by: manushprajwal <manushprajwal555@gmail.com>

checkpatch says:

WARNING: From:/Signed-off-by: email name mismatch: 'From: Manush Prajwal <manushprajwal555@gmail.com>' != 'Signed-off-by: manushprajwal <manushprajwal555@gmail.com>'

... which means that I can not apply this patch. Please fix and resend.

Thanks,
Guenter

…
> Fix it by calling pm_runtime_put_noidle() before returning, mirroring
> the same balance the counterpart starfive_wdt_pm_stop() already
> maintains with pm_runtime_put_sync().
…

Can a bit of duplicate source code be avoided for the exception handling?
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/coding-style.rst?h=v7.2#n526

Regards,
Markus