From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
Fixes: 6ffcb7971486 ("gpio: sysfs: use per-gpio locking") Cc: stable@vger.kernel.org Signed-off-by: Bartosz Golaszewski bartosz.golaszewski@linaro.org --- drivers/gpio/gpiolib-sysfs.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/drivers/gpio/gpiolib-sysfs.c b/drivers/gpio/gpiolib-sysfs.c index 4a3aa09dad9d..cd3381a4bc93 100644 --- a/drivers/gpio/gpiolib-sysfs.c +++ b/drivers/gpio/gpiolib-sysfs.c @@ -713,6 +713,7 @@ void gpiod_unexport(struct gpio_desc *desc) }
put_device(dev); + mutex_destroy(&data->mutex); kfree(data); } EXPORT_SYMBOL_GPL(gpiod_unexport);
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
You can't just make shit up and then pretend to fix it...
Fixes: 6ffcb7971486 ("gpio: sysfs: use per-gpio locking") Cc: stable@vger.kernel.org Signed-off-by: Bartosz Golaszewski bartosz.golaszewski@linaro.org
Johan
On Fri, May 16, 2025 at 1:42 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
No, there's no leak but with lock debugging it still warns if the mutex is locked when it's being destroyed so the change still makes sense with a modified commit message.
You can't just make shit up and then pretend to fix it...
There's no need for this kind of comment. You made your point clear in the first sentence.
Bartosz
On Fri, May 16, 2025 at 02:32:54PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 1:42 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
No, there's no leak but with lock debugging it still warns if the mutex is locked when it's being destroyed so the change still makes sense with a modified commit message.
You can't just make shit up and then pretend to fix it...
There's no need for this kind of comment. You made your point clear in the first sentence.
Your claim that there's "a resource leak with mutex debugging enabled" is is quite specific. Now I had to go check that no one had changed something in ways they shouldn't have recently. But mutex_destroy() still works as it always has, which you should have verified yourself before sending a "fix" tagged for stable backport based on a hunch.
Johan
On Fri, May 16, 2025 at 6:58 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 02:32:54PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 1:42 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
No, there's no leak but with lock debugging it still warns if the mutex is locked when it's being destroyed so the change still makes sense with a modified commit message.
You can't just make shit up and then pretend to fix it...
There's no need for this kind of comment. You made your point clear in the first sentence.
Your claim that there's "a resource leak with mutex debugging enabled" is is quite specific. Now I had to go check that no one had changed something in ways they shouldn't have recently. But mutex_destroy() still works as it always has, which you should have verified yourself before sending a "fix" tagged for stable backport based on a hunch.
Yes, I admitted that the commit message was wrong. And yes, it sometimes happens that we get copied on crappy patches. However, unlike what your comment suggests, I don't go around the kernel, "making sh*t up" just to add a "Fixes: Johan's commit". I had this as part of a bigger rework I have in progress[1] (discussed previously here[2]) and figured that with the series growing in size, I'll at least get the fix upstream before v6.16-rc1.
I should have given the patch more than 10 seconds of thought for sure but your immediate hostility is uncalled for. Please try to assume good faith a bit more.
Bartosz
[1] https://github.com/brgl/linux/tree/b4/gpio-sysfs-chip-export [2] https://lore.kernel.org/all/CAMRc=McUCeZcU6co1aN54rTudo+JfPjjForu4iKQ5npwXk6...
On Mon, May 19, 2025 at 02:18:15PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 6:58 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 02:32:54PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 1:42 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
No, there's no leak but with lock debugging it still warns if the mutex is locked when it's being destroyed so the change still makes sense with a modified commit message.
You can't just make shit up and then pretend to fix it...
There's no need for this kind of comment. You made your point clear in the first sentence.
Your claim that there's "a resource leak with mutex debugging enabled" is is quite specific. Now I had to go check that no one had changed something in ways they shouldn't have recently. But mutex_destroy() still works as it always has, which you should have verified yourself before sending a "fix" tagged for stable backport based on a hunch.
Yes, I admitted that the commit message was wrong. And yes, it sometimes happens that we get copied on crappy patches. However, unlike what your comment suggests, I don't go around the kernel, "making sh*t up" just to add a "Fixes: Johan's commit". I had this as part of a bigger rework I have in progress[1] (discussed previously here[2]) and figured that with the series growing in size, I'll at least get the fix upstream before v6.16-rc1.
But it is not a fix. It is based on a misunderstanding that you should have caught yourself by just looking at the code before posting.
Sure, mutex_destroy() is an odd bird, but you still need to verify your guesses before posting patches based on them. It's that lazy attitude (and violation of the stable kernel policy) that I'm criticising.
Johan
On Mon, May 19, 2025 at 2:42 PM Johan Hovold johan@kernel.org wrote:
On Mon, May 19, 2025 at 02:18:15PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 6:58 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 02:32:54PM +0200, Bartosz Golaszewski wrote:
On Fri, May 16, 2025 at 1:42 PM Johan Hovold johan@kernel.org wrote:
On Fri, May 16, 2025 at 12:40:23PM +0200, Bartosz Golaszewski wrote:
From: Bartosz Golaszewski bartosz.golaszewski@linaro.org
We initialize the data->mutex in gpiod_export() but lack the corresponding mutex_destroy() in gpiod_unexport() causing a resource leak with mutex debugging enabled. Add the call right before kfreeing the GPIO data.
No, there's no resource leak and it's perfectly fine not to call mutex_destroy().
No, there's no leak but with lock debugging it still warns if the mutex is locked when it's being destroyed so the change still makes sense with a modified commit message.
You can't just make shit up and then pretend to fix it...
There's no need for this kind of comment. You made your point clear in the first sentence.
Your claim that there's "a resource leak with mutex debugging enabled" is is quite specific. Now I had to go check that no one had changed something in ways they shouldn't have recently. But mutex_destroy() still works as it always has, which you should have verified yourself before sending a "fix" tagged for stable backport based on a hunch.
Yes, I admitted that the commit message was wrong. And yes, it sometimes happens that we get copied on crappy patches. However, unlike what your comment suggests, I don't go around the kernel, "making sh*t up" just to add a "Fixes: Johan's commit". I had this as part of a bigger rework I have in progress[1] (discussed previously here[2]) and figured that with the series growing in size, I'll at least get the fix upstream before v6.16-rc1.
But it is not a fix. It is based on a misunderstanding that you should have caught yourself by just looking at the code before posting.
Noted, I'll drop the Fixes: tag and submit it as part of the rest of the rework.
Sure, mutex_destroy() is an odd bird, but you still need to verify your guesses before posting patches based on them. It's that lazy attitude (and violation of the stable kernel policy) that I'm criticising.
IMO what you just wrote here is a way better way of expressing your criticism than what you led with.
Bartosz
linux-stable-mirror@lists.linaro.org
-
Bartosz Golaszewski -
Johan Hovold