Re: [PATCH v5 03/10] media: rcar-csi2: Move {enable|disable}_streams() calls

[Niklas Söderlund]

Hello,

On 2026-03-18 22:54:35 +0200, Laurent Pinchart wrote:
> On Wed, Mar 11, 2026 at 03:53:16PM +0200, Tomi Valkeinen wrote:
> > With multiple streams the operation to enable the CSI-2 hardware and to
> > call {enable|disable}_streams() on upstream subdev will need to be
> > handled separately.
> > 
> > Prepare for that by moving {enable|disable}_streams() calls out from
> > rcsi2_start() and rcsi2_stop().
> > 
> > On Gen3, a side effect of this change is that if the sink side devices
> > call .enable_streams() on rcar-csi2 multiple times, the second call will
> > fail. This is because we always use stream ID 0, so the second call
> > would attempt to enable the same stream again, leading to an error. In
> > other words, a normal single-stream setup continues to work, but trying
> > to use the current driver's custom VC based routing will fail.
> 
> I assume this gets addressed later in the series.
> 
> > 
> > On Gen4, this doesn't matter as the rcar-isp behaves in a similar way as
> > described above, and thus rcar-csi2 will only get a single
> > .enable_streams() call.
> > 
> > Signed-off-by: Tomi Valkeinen <tomi.valkeinen+renesas@xxxxxxxxxxxxxxxx>
> > ---
> >  drivers/media/platform/renesas/rcar-csi2.c | 25 +++++++++++++++----------
> >  1 file changed, 15 insertions(+), 10 deletions(-)
> > 
> > diff --git a/drivers/media/platform/renesas/rcar-csi2.c b/drivers/media/platform/renesas/rcar-csi2.c
> > index 7305cc4a04cb..158fa447e668 100644
> > --- a/drivers/media/platform/renesas/rcar-csi2.c
> > +++ b/drivers/media/platform/renesas/rcar-csi2.c
> > @@ -1822,20 +1822,12 @@ static int rcsi2_start(struct rcar_csi2 *priv, struct v4l2_subdev_state *state)
> >  		return ret;
> >  	}
> >  
> > -	ret = v4l2_subdev_enable_streams(priv->remote, priv->remote_pad,
> > -					 BIT_ULL(0));
> > -	if (ret) {
> > -		rcsi2_enter_standby(priv);
> > -		return ret;
> > -	}
> > -
> >  	return 0;
> >  }
> >  
> >  static void rcsi2_stop(struct rcar_csi2 *priv)
> >  {
> >  	rcsi2_enter_standby(priv);
> > -	v4l2_subdev_disable_streams(priv->remote, priv->remote_pad, BIT_ULL(0));
> >  }
> >  
> >  static int rcsi2_enable_streams(struct v4l2_subdev *sd,
> > @@ -1857,6 +1849,14 @@ static int rcsi2_enable_streams(struct v4l2_subdev *sd,
> >  			return ret;
> >  	}
> >  
> > +	ret = v4l2_subdev_enable_streams(priv->remote, priv->remote_pad,
> > +					 BIT_ULL(0));
> > +	if (ret) {
> > +		if (priv->stream_count == 0)
> > +			rcsi2_stop(priv);
> > +		return ret;
> > +	}
> > +
> >  	priv->stream_count += 1;
> >  
> >  	return ret;
> > @@ -1867,7 +1867,7 @@ static int rcsi2_disable_streams(struct v4l2_subdev *sd,
> >  				 u32 source_pad, u64 source_streams_mask)
> >  {
> >  	struct rcar_csi2 *priv = sd_to_csi2(sd);
> > -	int ret = 0;
> > +	int ret;
> >  
> >  	if (source_streams_mask != 1)
> >  		return -EINVAL;
> > @@ -1878,9 +1878,14 @@ static int rcsi2_disable_streams(struct v4l2_subdev *sd,
> >  	if (priv->stream_count == 1)
> >  		rcsi2_stop(priv);
> >  
> > +	ret = v4l2_subdev_disable_streams(priv->remote, priv->remote_pad,
> > +					  BIT_ULL(0));
> > +	if (ret)
> > +		return ret;
> > +
> >  	priv->stream_count -= 1;
> >  
> > -	return ret;
> > +	return 0;
> >  }
> 
> rcsi2_irq_thread() also calls rcsi2_stop(), followed by rcsi2_start().
> This is to handle errors reported by the AFIFO_OF, ERRSOTHS and
> ERRSOTSYNCHS interrupts. If the source isn't restarted, such an attempt
> to recover from errors will likely fail. On the other hand, restarting
> the source will likely not lead to great results either.
> 
> Error handling was introduced in
> 
> commit 4ab44ff0841b9a825f9875623d24809d29e37a10
> Author: Niklas Söderlund <niklas.soderlund+renesas@xxxxxxxxxxxx>
> Date:   Thu Apr 11 16:30:58 2019 -0400
> 
>     media: rcar-csi2: restart CSI-2 link if error is detected
> 
>     Restart the CSI-2 link if the CSI-2 receiver detects an error during
>     reception. The driver did nothing when a link error happened and the
>     data flow simply stopped without the user knowing why.
> 
>     Change the driver to try and recover from errors by restarting the link
>     and informing the user that something is not right. For obvious reasons
>     it's not possible to recover from all errors (video source disconnected
>     for example) but in such cases the user is at least informed of the
>     error and the same behavior of the stopped data flow is retained.
> 
> Niklas, do you recall anything about the errors you saw ?

IIRC, the spark for the work was a user report where the wish was to 
handle detectable errors somehow instead of just silently fail. The 
first try was this commit where the driver tries to recover. However as 
you point out it might not be the optimal solution.

Instead I proposed a way to signal to user-space that there where a link 
error and let it deal with it [1]. As seen on the date from that work it 
kind of stalled. I still plan to resume that work once all the cleanups 
of the R-Car VIN pipeline are done.

1.  https://lore.kernel.org/linux-renesas-soc/20211108160220.767586-1-niklas.soderlund%2Brenesas@xxxxxxxxxxxx/

> 
> >  
> >  static int rcsi2_set_pad_format(struct v4l2_subdev *sd,
> > 
> 
> -- 
> Regards,
> 
> Laurent Pinchart

-- 
Kind Regards,
Niklas Söderlund