[PATCH] media: nxp: imx8-isi: add m2m usage_count check in streamoff

[PATCH] media: nxp: imx8-isi: add m2m usage_count check in streamoff

[Laurentiu Palcu]

Currently, if streamon/streamoff calls are imbalanced we can end up
with a negative ISI m2m usage_count. When that happens, the next
streamon call will not enable the ISI m2m channel.

So, instead of throwing a warning in streamoff when usage_count drops
below 0, just make sure we don't get there.

Fixes: cf21f328fcafac ("media: nxp: Add i.MX8 ISI driver")
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@oss.nxp.com>
---
 drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
index 9745d6219a166..b71195a3ba256 100644
--- a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
+++ b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
@@ -575,6 +575,9 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
 
 	mutex_lock(&m2m->lock);
 
+	if (m2m->usage_count == 0)
+		goto unlock;
+
 	/*
 	 * If the last context is this one, reset it to make sure the device
 	 * will be reconfigured when streaming is restarted.
@@ -594,8 +597,7 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
 		mxc_isi_channel_release(m2m->pipe);
 	}
 
-	WARN_ON(m2m->usage_count < 0);
-
+unlock:
 	mutex_unlock(&m2m->lock);
 
 	return 0;
-- 
2.34.1

Re: [PATCH] media: nxp: imx8-isi: add m2m usage_count check in streamoff

[Jacopo Mondi]

Hi Laurentiu

On Fri, Sep 20, 2024 at 05:27:15PM GMT, Laurentiu Palcu wrote:
> Currently, if streamon/streamoff calls are imbalanced we can end up
> with a negative ISI m2m usage_count. When that happens, the next
> streamon call will not enable the ISI m2m channel.
>
> So, instead of throwing a warning in streamoff when usage_count drops
> below 0, just make sure we don't get there.

Isn't the whole purpose of the WARN() to highlight something's wrong
with userspace ? I think it's expected to have the same number of
streamon and streamoff calls, do you have any idea why it might not be
happening ?

Thanks
  j

>
> Fixes: cf21f328fcafac ("media: nxp: Add i.MX8 ISI driver")
> Signed-off-by: Laurentiu Palcu <laurentiu.palcu@oss.nxp.com>
> ---
>  drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> index 9745d6219a166..b71195a3ba256 100644
> --- a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> +++ b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> @@ -575,6 +575,9 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
>
>  	mutex_lock(&m2m->lock);
>
> +	if (m2m->usage_count == 0)
> +		goto unlock;
> +
>  	/*
>  	 * If the last context is this one, reset it to make sure the device
>  	 * will be reconfigured when streaming is restarted.
> @@ -594,8 +597,7 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
>  		mxc_isi_channel_release(m2m->pipe);
>  	}
>
> -	WARN_ON(m2m->usage_count < 0);
> -
> +unlock:
>  	mutex_unlock(&m2m->lock);
>
>  	return 0;
> --
> 2.34.1
>
>

Re: [PATCH] media: nxp: imx8-isi: add m2m usage_count check in streamoff

[Laurent Pinchart]

On Mon, Sep 23, 2024 at 07:50:18PM +0200, Jacopo Mondi wrote:
> On Fri, Sep 20, 2024 at 05:27:15PM GMT, Laurentiu Palcu wrote:
> > Currently, if streamon/streamoff calls are imbalanced we can end up
> > with a negative ISI m2m usage_count. When that happens, the next
> > streamon call will not enable the ISI m2m channel.
> >
> > So, instead of throwing a warning in streamoff when usage_count drops
> > below 0, just make sure we don't get there.
> 
> Isn't the whole purpose of the WARN() to highlight something's wrong
> with userspace ? I think it's expected to have the same number of
> streamon and streamoff calls, do you have any idea why it might not be
> happening ?

WARN() is much too strong to report userspace problems.

> > Fixes: cf21f328fcafac ("media: nxp: Add i.MX8 ISI driver")
> > Signed-off-by: Laurentiu Palcu <laurentiu.palcu@oss.nxp.com>
> > ---
> >  drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c | 6 ++++--
> >  1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > index 9745d6219a166..b71195a3ba256 100644
> > --- a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > +++ b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > @@ -575,6 +575,9 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
> >
> >  	mutex_lock(&m2m->lock);
> >
> > +	if (m2m->usage_count == 0)
> > +		goto unlock;

This isn't right. Userspace can still call VIDIOC_STREAMOFF while not
streaming, which will decrement the usage count when it shouldn't. The
right fix is to return from mxc_isi_m2m_streamoff() without decrementing
the usage count if the particular context the function is called for is
not streaming. You will possibly need to also modify
mxc_isi_m2m_streamon() to make sure the two functions won't race each
other.

Could you work on an improved patch ? Please let me know if you need
help, it's important to fix this issue.

> > +
> >  	/*
> >  	 * If the last context is this one, reset it to make sure the device
> >  	 * will be reconfigured when streaming is restarted.
> > @@ -594,8 +597,7 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
> >  		mxc_isi_channel_release(m2m->pipe);
> >  	}
> >
> > -	WARN_ON(m2m->usage_count < 0);
> > -
> > +unlock:
> >  	mutex_unlock(&m2m->lock);
> >
> >  	return 0;

-- 
Regards,

Laurent Pinchart

Re: [PATCH] media: nxp: imx8-isi: add m2m usage_count check in streamoff

[Laurentiu Palcu]

Hi Laurent,

On Tue, Sep 24, 2024 at 12:52:46AM +0300, Laurent Pinchart wrote:
> On Mon, Sep 23, 2024 at 07:50:18PM +0200, Jacopo Mondi wrote:
> > On Fri, Sep 20, 2024 at 05:27:15PM GMT, Laurentiu Palcu wrote:
> > > Currently, if streamon/streamoff calls are imbalanced we can end up
> > > with a negative ISI m2m usage_count. When that happens, the next
> > > streamon call will not enable the ISI m2m channel.
> > >
> > > So, instead of throwing a warning in streamoff when usage_count drops
> > > below 0, just make sure we don't get there.
> > 
> > Isn't the whole purpose of the WARN() to highlight something's wrong
> > with userspace ? I think it's expected to have the same number of
> > streamon and streamoff calls, do you have any idea why it might not be
> > happening ?
> 
> WARN() is much too strong to report userspace problems.
> 
> > > Fixes: cf21f328fcafac ("media: nxp: Add i.MX8 ISI driver")
> > > Signed-off-by: Laurentiu Palcu <laurentiu.palcu@oss.nxp.com>
> > > ---
> > >  drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c | 6 ++++--
> > >  1 file changed, 4 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > > index 9745d6219a166..b71195a3ba256 100644
> > > --- a/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > > +++ b/drivers/media/platform/nxp/imx8-isi/imx8-isi-m2m.c
> > > @@ -575,6 +575,9 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
> > >
> > >  	mutex_lock(&m2m->lock);
> > >
> > > +	if (m2m->usage_count == 0)
> > > +		goto unlock;
> 
> This isn't right. Userspace can still call VIDIOC_STREAMOFF while not
> streaming, which will decrement the usage count when it shouldn't.

Hmm, I didn't consider that... :/

> The right fix is to return from mxc_isi_m2m_streamoff() without
> decrementing the usage count if the particular context the function is
> called for is not streaming. You will possibly need to also modify
> mxc_isi_m2m_streamon() to make sure the two functions won't race each
> other.
> 
> Could you work on an improved patch ?

I'll give it a shot.

> Please let me know if you need help, it's important to fix this issue.

Thanks, I'll ping you on IRC if I get stuck.

Laurentiu

> 
> > > +
> > >  	/*
> > >  	 * If the last context is this one, reset it to make sure the device
> > >  	 * will be reconfigured when streaming is restarted.
> > > @@ -594,8 +597,7 @@ static int mxc_isi_m2m_streamoff(struct file *file, void *fh,
> > >  		mxc_isi_channel_release(m2m->pipe);
> > >  	}
> > >
> > > -	WARN_ON(m2m->usage_count < 0);
> > > -
> > > +unlock:
> > >  	mutex_unlock(&m2m->lock);
> > >
> > >  	return 0;
> 
> -- 
> Regards,
> 
> Laurent Pinchart