Hi Jacopo,
Thanks for your work.
On 2024-05-03 17:51:22 +0200, Jacopo Mondi wrote:
> The adv748x-csi2 driver configures the CSI-2 transmitter to
> automatically infer the image stream format from the connected
> frontend (HDMI or AFE).
>
> Setting a new format on the subdevice hence does not actually control
> the CSI-2 output format, but it's only there for the purpose of
> pipeline validation.
>
> However, there is currently no validation that the supplied media bus
> code is valid and supported by the device.
>
> With the introduction of enum_mbus_codes a list of supported format is
> now available, use it to validate that the supplied format is correct
> and use the default YUYV8 one if that's not the case.
With the update tests for the change in patch 4 I hit multiple issues
with this patch for CVBS capture.
>
> Signed-off-by: Jacopo Mondi <jacopo.mondi@xxxxxxxxxxxxxxxx>
> ---
> drivers/media/i2c/adv748x/adv748x-csi2.c | 27 +++++++++++++++++++++++-
> 1 file changed, 26 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/media/i2c/adv748x/adv748x-csi2.c b/drivers/media/i2c/adv748x/adv748x-csi2.c
> index 219417b319a6..1aae6467ca62 100644
> --- a/drivers/media/i2c/adv748x/adv748x-csi2.c
> +++ b/drivers/media/i2c/adv748x/adv748x-csi2.c
> @@ -215,6 +215,23 @@ static int adv748x_csi2_get_format(struct v4l2_subdev *sd,
> return 0;
> }
>
> +static int adv748x_csi2_is_fmt_supported(struct adv748x_csi2 *tx,
> + unsigned int code)
> +{
> + const unsigned int *codes = is_txa(tx) ?
> + adv748x_csi2_txa_fmts :
> + adv748x_csi2_txb_fmts;
> + size_t num_fmts = is_txa(tx) ? ARRAY_SIZE(adv748x_csi2_txa_fmts)
> + : ARRAY_SIZE(adv748x_csi2_txb_fmts);
> +
> + for (unsigned int i = 0; i < num_fmts; i++) {
> + if (codes[i] == code)
> + return 0;
> + }
> +
> + return -EINVAL;
> +}
> +
> static int adv748x_csi2_set_format(struct v4l2_subdev *sd,
> struct v4l2_subdev_state *sd_state,
> struct v4l2_subdev_format *sdformat)
> @@ -222,7 +239,15 @@ static int adv748x_csi2_set_format(struct v4l2_subdev *sd,
> struct adv748x_csi2 *tx = adv748x_sd_to_csi2(sd);
> struct adv748x_state *state = tx->state;
> struct v4l2_mbus_framefmt *mbusformat;
> - int ret = 0;
> + int ret;
> +
> + /*
> + * Make sure the format is supported, if not default it to
> + * YUYV8 as it's supported by both TXes.
> + */
> + ret = adv748x_csi2_is_fmt_supported(tx, sdformat->format.code);
> + if (ret)
> + sdformat->format.code = MEDIA_BUS_FMT_YUYV8_1X16;
If adv748x_csi2_is_fmt_supported() returns non-zero you default to
MEDIA_BUS_FMT_YUYV8_1X16, which is fine. But the non-zero return code is
propagated at the end of this function and to user-space falling the
IOCTL.
Fixing that I hit another issue that kind of shows we need this format
validation ;-)
The TXB entity only supports MEDIA_BUS_FMT_YUYV8_1X16 formats, the AFE
entity only outputs MEDIA_BUS_FMT_UYVY8_2X8... So with format validation
in place it becomes impossible to connect AFE to TXB and breaking CBVS
capture on Gen3. As a hack I added MEDIA_BUS_FMT_UYVY8_2X8 support to
TXB and I can again capture CVBS with patch 1-8 applied.
>
> mbusformat = adv748x_csi2_get_pad_format(sd, sd_state, sdformat->pad,
> sdformat->which);
> --
> 2.44.0
>
--
Kind Regards,
Niklas Söderlund
