Re: [PATCH v4 4/4] media: platform: Add Renesas RZ/G2L CRU driver

[Sakari Ailus]

Hi Prabhakar,

On Mon, Oct 31, 2022 at 12:40:28PM +0000, Lad, Prabhakar wrote:
> Hi Sakari,
> 
> On Mon, Oct 31, 2022 at 8:19 AM Sakari Ailus
> <sakari.ailus@xxxxxxxxxxxxxxx> wrote:
> >
> > Hi Prabhakar,
> >
> > On Sun, Oct 30, 2022 at 10:32:43PM +0000, Lad, Prabhakar wrote:
> > > Hi Sakari,
> > >
> > > On Fri, Oct 28, 2022 at 1:22 PM Sakari Ailus
> > > <sakari.ailus@xxxxxxxxxxxxxxx> wrote:
> > > >
> > > > Hi Prabhakar,
> > > >
> > > > On Thu, Oct 27, 2022 at 08:04:40PM +0100, Lad, Prabhakar wrote:
> > > > ...
> > > > > > > +static int rzg2l_cru_ip_s_stream(struct v4l2_subdev *sd, int enable)
> > > > > > > +{
> > > > > > > +     struct rzg2l_cru_dev *cru;
> > > > > > > +     int ret;
> > > > > > > +
> > > > > > > +     cru = v4l2_get_subdevdata(sd);
> > > > > > > +
> > > > > > > +     if (!cru->is_csi)
> > > > > > > +             return -EINVAL;
> > > > > > > +
> > > > > > > +     ret = v4l2_subdev_call(cru->ip.remote, video, s_stream, enable);
> > > > > >
> > > > > > It's up to the driver how call pre_streamon() and post_streamoff(), as long
> > > > > > as it takes place on both sides of s_stream().
> > > > > >
> > > > > > In other words, as it seems your device doesn't need anything special, you
> > > > > > could waive implemeting the callbacks yourself and call pre_streamon() and
> > > > > > post_streamoff() here.
> > > > > >
> > > > > Here the cru->ip.remote = CSI, in the rzg2l_cru_set_stream(1) where we
> > > > > are calling pre_streamon()/post_streamoff() callbacks the subdev is
> > > > > CRU-IP. So the calls from rzg2l_cru_set_stream() land into
> > > > > rzg2l_cru_ip_pre_streamon() and rzg2l_cru_ip_post_streamoff() which
> > > > > are calling pre_streamon/post_streamoff for the CSI subdev.
> > > >
> > > > Again, you should call the source sub-device's pre_streamon and
> > > > post_streamoff from the s_stream handler (not from
> > > > rzg2l_cru_ip_pre_streamon or rzg2l_cru_ip_post_streamoff).
> > > >
> > > > Starting streaming takes place link by link. This allows a driver to omit
> > > > implementing pre_streamon and post_streamon callbacks if it doesn't need
> > > > them.
> > > >
> > > Thank you for the explanation that makes sense now to me.
> > >
> > > Now with this approach the initialization sequence of CSI + CRU won't
> > > align as per the HW manual. Unfortunately I'll have to switch back on
> > > exporting the functions. I hope that's okay?
> >
> > It is not.
> >
> > What exactly would you like to do that you can't with the
> > pre_streamon/post_streamoff callbacks called from s_stream?
> >
> The initialization sequence for MIPI CSI [0]. As per [0] we need to
> initialize the CSI2 dphy first then setup the AXI (part of CRU driver)
> and then later MIPI CSI2 link (part of csi driver) and lastly turn on
> clock and link (in the cru driver).
> 
> So as per the current implementation we have the below:
> 1] CRU IP subdev is calling pre_stream for the CSI2 subdev in its
> pre_stream on callback - This is where the CSI2 DPHY is initialized
> 2] Later in the flow we initialize the AXI part - ie part of
> rzg2l_cru_set_stream
> 3] We call s_stream in rzg2l_cru_set_stream - This lands into CSI2
> subdev to initialize the MIPI CSI2 Link
> 4] In the rzg2l_cru_set_stream we setup up the vclk and enable link reception
> 
> [0] https://ibb.co/QpHNkLh

How is this changed by calling the pre_streamon/post_streamoff callbacks
from s_stream? I understand the pipeline device-wise is:

	... -> CSI-2 -> CRU

I wonder if it would be faster to discuss this on #linux-media.

-- 
Kind regards,

Sakari Ailus