enable_irq() and disable_irq() are reference counted, so we must make sure
that each enable_irq() is always paired with a single disable_irq(). If we
call disable_irq() twice followed by just a single enable_irq(), the IRQ
will remain disabled forever.

For the error handling path in qcom_q6v5_wait_for_start(), disable_irq()
will end up being called twice, because disable_irq() also happens in
qcom_q6v5_unprepare() when rolling back the call to qcom_q6v5_prepare().

Fix this by dropping disable_irq() in qcom_q6v5_wait_for_start(). Since
qcom_q6v5_prepare() is the function that calls enable_irq(), it makes more
sense to have the rollback handled always by qcom_q6v5_unprepare().

Fixes: 3b415c8fb263 ("remoteproc: q6v5: Extract common resource handling")
Signed-off-by: Stephan Gerhold <stephan.gerhold@linaro.org>
---
 drivers/remoteproc/qcom_q6v5.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/drivers/remoteproc/qcom_q6v5.c b/drivers/remoteproc/qcom_q6v5.c
index 4ee5e67a9f03f5f766f04396b9a3e45f77293764..769c6d6d6a731672eca9f960b05c68f6d4d77af2 100644
--- a/drivers/remoteproc/qcom_q6v5.c
+++ b/drivers/remoteproc/qcom_q6v5.c
@@ -156,9 +156,6 @@ int qcom_q6v5_wait_for_start(struct qcom_q6v5 *q6v5, int timeout)
 	int ret;
 
 	ret = wait_for_completion_timeout(&q6v5->start_done, timeout);
-	if (!ret)
-		disable_irq(q6v5->handover_irq);
-
 	return !ret ? -ETIMEDOUT : 0;
 }
 EXPORT_SYMBOL_GPL(qcom_q6v5_wait_for_start);

-- 
2.50.1

On Tue, Aug 19, 2025 at 01:08:02PM +0200, Stephan Gerhold wrote:
> enable_irq() and disable_irq() are reference counted, so we must make sure
> that each enable_irq() is always paired with a single disable_irq(). If we
> call disable_irq() twice followed by just a single enable_irq(), the IRQ
> will remain disabled forever.
> 
> For the error handling path in qcom_q6v5_wait_for_start(), disable_irq()
> will end up being called twice, because disable_irq() also happens in
> qcom_q6v5_unprepare() when rolling back the call to qcom_q6v5_prepare().
> 
> Fix this by dropping disable_irq() in qcom_q6v5_wait_for_start(). Since
> qcom_q6v5_prepare() is the function that calls enable_irq(), it makes more
> sense to have the rollback handled always by qcom_q6v5_unprepare().
> 
> Fixes: 3b415c8fb263 ("remoteproc: q6v5: Extract common resource handling")

Didn't earlier versions also have the same behaviour?

> Signed-off-by: Stephan Gerhold <stephan.gerhold@linaro.org>
> ---
>  drivers/remoteproc/qcom_q6v5.c | 3 ---
>  1 file changed, 3 deletions(-)
> 
> diff --git a/drivers/remoteproc/qcom_q6v5.c b/drivers/remoteproc/qcom_q6v5.c
> index 4ee5e67a9f03f5f766f04396b9a3e45f77293764..769c6d6d6a731672eca9f960b05c68f6d4d77af2 100644
> --- a/drivers/remoteproc/qcom_q6v5.c
> +++ b/drivers/remoteproc/qcom_q6v5.c
> @@ -156,9 +156,6 @@ int qcom_q6v5_wait_for_start(struct qcom_q6v5 *q6v5, int timeout)
>  	int ret;
>  
>  	ret = wait_for_completion_timeout(&q6v5->start_done, timeout);
> -	if (!ret)
> -		disable_irq(q6v5->handover_irq);
> -
>  	return !ret ? -ETIMEDOUT : 0;
>  }
>  EXPORT_SYMBOL_GPL(qcom_q6v5_wait_for_start);
> 
> -- 
> 2.50.1
> 

-- 
With best wishes
Dmitry

On Tue, Aug 19, 2025 at 02:44:26PM +0300, Dmitry Baryshkov wrote:
> On Tue, Aug 19, 2025 at 01:08:02PM +0200, Stephan Gerhold wrote:
> > enable_irq() and disable_irq() are reference counted, so we must make sure
> > that each enable_irq() is always paired with a single disable_irq(). If we
> > call disable_irq() twice followed by just a single enable_irq(), the IRQ
> > will remain disabled forever.
> > 
> > For the error handling path in qcom_q6v5_wait_for_start(), disable_irq()
> > will end up being called twice, because disable_irq() also happens in
> > qcom_q6v5_unprepare() when rolling back the call to qcom_q6v5_prepare().
> > 
> > Fix this by dropping disable_irq() in qcom_q6v5_wait_for_start(). Since
> > qcom_q6v5_prepare() is the function that calls enable_irq(), it makes more
> > sense to have the rollback handled always by qcom_q6v5_unprepare().
> > 
> > Fixes: 3b415c8fb263 ("remoteproc: q6v5: Extract common resource handling")
> 
> Didn't earlier versions also have the same behaviour?
> 

I don't think so. The "extracted common resource handling" came from
qcom_q6v5_pil.c, but q6v5_start() just had most of this code inline in a
single function [1]. The handling of enable_irq()/disable_irq() through
the goto labels looks correct there.

Thanks,
Stephan

[1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/remoteproc/qcom_q6v5_pil.c?id=0e622e80191e75c99b6ecc265c140a37d81e7a63#n795

On Tue, Aug 19, 2025 at 04:55:09PM +0200, Stephan Gerhold wrote:
> On Tue, Aug 19, 2025 at 02:44:26PM +0300, Dmitry Baryshkov wrote:
> > On Tue, Aug 19, 2025 at 01:08:02PM +0200, Stephan Gerhold wrote:
> > > enable_irq() and disable_irq() are reference counted, so we must make sure
> > > that each enable_irq() is always paired with a single disable_irq(). If we
> > > call disable_irq() twice followed by just a single enable_irq(), the IRQ
> > > will remain disabled forever.
> > > 
> > > For the error handling path in qcom_q6v5_wait_for_start(), disable_irq()
> > > will end up being called twice, because disable_irq() also happens in
> > > qcom_q6v5_unprepare() when rolling back the call to qcom_q6v5_prepare().
> > > 
> > > Fix this by dropping disable_irq() in qcom_q6v5_wait_for_start(). Since
> > > qcom_q6v5_prepare() is the function that calls enable_irq(), it makes more
> > > sense to have the rollback handled always by qcom_q6v5_unprepare().
> > > 
> > > Fixes: 3b415c8fb263 ("remoteproc: q6v5: Extract common resource handling")
> > 
> > Didn't earlier versions also have the same behaviour?
> > 
> 
> I don't think so. The "extracted common resource handling" came from
> qcom_q6v5_pil.c, but q6v5_start() just had most of this code inline in a
> single function [1]. The handling of enable_irq()/disable_irq() through
> the goto labels looks correct there.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>


-- 
With best wishes
Dmitry