On Tue, Apr 30, 2024 at 01:51:13PM +0200, Jacopo Mondi wrote: > Hi Niklas > thanks for the quick reply > > On Tue, Apr 30, 2024 at 01:17:18PM +0200, Niklas Söderlund wrote: > > Hi Jacopo, > > > > Thanks for this work! I'm happy that we finally can get this upstream. I > > will test and review this in the coming days. As there appears to be a > > few dependencies to get this to work do you have a branch covering > > everything needed? > > Sure > https://git.kernel.org/pub/scm/linux/kernel/git/jmondi/linux.git/ > jmondi/renesas-drivers-2024-04-23-v6.9-rc5/multistream > > I should have added it to the cover letter > > > On 2024-04-30 12:39:36 +0200, Jacopo Mondi wrote: > > > Hello this series implements multi-stream support for R-Car CSI-2 and the > > > two CSI-2 devices that are available on Renesas dev-boards which are mainline > > > supported (the ADV748x HDMI decoder and the MAX9286 GMSL deserializer). > > > > You also need to update the MAX96712 driver in staging else V3U and Gen4 > > will break. > > > > Ouch, right. I can do that but won't be able to test. Would you ? > > > > The net gain in feature is: > > > - ADV748x: Select on which MIPI CSI-2 VC to send the stream on > > > - MAX9286: Paves the way for run-time selection of active links > > > > > > Both features are realized by controlling the devices routing tables. > > > The max9286 routing support is not implemented, but these patches are needed > > > in order to maintain the current version compatible with R-Car CSI-2. > > > > > > The R-Car CSI-2 driver needs to be updated in order to support multiplexed > > > transmitters, in particular: > > > > > > - Use LINK_FREQ to compute the lane bandwidth > > > - Support subdev_active state > > > - Configure DT and VC handling by using the remote's frame_desc > > > > > > A separate fix for VIN sits at the top of the series to re-enable YUYV capture > > > operations. > > > > > > The series needs to activate streams in v4l2-subdev to work correctly: > > > > > > --- a/drivers/media/v4l2-core/v4l2-subdev.c > > > +++ b/drivers/media/v4l2-core/v4l2-subdev.c > > > @@ -32,7 +32,7 @@ > > > * 'v4l2_subdev_enable_streams_api' to 1 below. > > > */ > > > > > > -static bool v4l2_subdev_enable_streams_api; > > > +static bool v4l2_subdev_enable_streams_api = 1; > > > #endif > > > > I'm not up to date with the latest state of the multiple streams API. > > Can this be runtime configured or will this break capture if the kernel > > is not patched to enable this? > > I'm afraid the latter. I really hope in the next kernel release we can > get rid of this protection and enable streams by default. I hope it will be done for v6.12 or possibly v6.11. > > > > > > Test scripts, based on vin-tests are available at: > > > https://git.uk.ideasonboard.com/jmondi/vin-tests > > > > > > A note for Sakari: patch > > > "media: max9286: Implement support for LINK_FREQ" > > > > > > Could actually use your new > > > "media: v4l: Support passing sub-device argument to v4l2_get_link_freq()" > > > > > > I kept it as it is for this version waiting for your patch to get in. > > > > I like this idea. > > > > > > > > A note for Niklas: > > > I think this series paves the way to actually moving the control of which > > > VC goes to which VIN using a routing table instead of using link enablement. > > > > > > Currently the setup is as it follows: routing within R-Car CSI-2 is fixed, sink > > > stream 0/X goes to source stream X+1/0 and each source pad represents a VC. > > > Which VC gets sent to which VIN is controlled by link enablement. > > > > > > A more natural way of handling this would be to make the routing table > > > within CSI-2 configurable and have media-links immutable between the CSI-2 > > > source pad. A CSI-2 source pad would then represent a 'channel' between > > > CSI-2 and VIN and not the VC which is sent on it. > > > > > > This setup would better represent the current implementation, with VIN's VCSEL > > > fixed channel routing and with the CSI-2 routing table controlling which VC gets > > > sent on which channel. What do you think ? > > > > That could be useful. Maybe we can even switch to use immutable links > > and use the routing instead greatly simplifying the VIN driver as this > > together with the other VIN series on the list we can drop the ugly > > group concept ;-) > > I had the same idea, but I'm not yet sure how this would work when a > CSI-2 can be routed to multiple VIN groups... > > > But this series is already quiet large, I think we can work on this > > on-top in a new series. > > Exactly, I didn't want to pile too many things. max9286 routing would > be next in line, I kept it out to avoid sending 30 patches in one go > (and to send this out earlier to get feedbacks) > > > > Jacopo Mondi (19): > > > media: adv748x: Add support for active state > > > media: adv748x: Add flags to adv748x_subdev_init() > > > media: adv748x: Use V4L2 streams > > > media: adv748x: Propagate format to opposite stream > > > media: adv748x: Implement set_routing() > > > media: adv748x: Use routes to configure VC > > > media: adv748x: Implement .get_frame_desc() > > > media: max9286: Add support for subdev active state > > > media: max9286: Fix enum_mbus_code > > > media: max9286: Use frame interval from subdev state > > > media: max9286: Use V4L2 Streams > > > media: max9286: Implement .get_frame_desc() > > > media: max9286: Implement support for LINK_FREQ > > > media: max9286: Implement .get_mbus_config() > > > media: rcar-csi2: Add support for multiplexed streams > > > media: rcar-csi2: Support multiplexed transmitters > > > media: rcar-csi2: Store format in the subdev state > > > media: rcar-csi2: Implement set_routing > > > media: rcar-vin: Fix YUYV8_1X16 handling for CSI-2 > > > > > > drivers/media/i2c/adv748x/adv748x-afe.c | 2 +- > > > drivers/media/i2c/adv748x/adv748x-core.c | 12 +- > > > drivers/media/i2c/adv748x/adv748x-csi2.c | 251 +++++++--- > > > drivers/media/i2c/adv748x/adv748x-hdmi.c | 2 +- > > > drivers/media/i2c/adv748x/adv748x.h | 4 +- > > > drivers/media/i2c/max9286.c | 457 ++++++++++++------ > > > drivers/media/platform/renesas/rcar-csi2.c | 442 ++++++++++++----- > > > .../platform/renesas/rcar-vin/rcar-dma.c | 16 +- > > > 8 files changed, 834 insertions(+), 352 deletions(-) -- Regards, Laurent Pinchart
