media: sun6i-mipi-csi2: Propagate format to source pad in TRY mode
Subsystems (3)
PATCH 1/1 media: sun6i-mipi-csi2: Propagate format to source pad in TRY mode
sun6i-mipi-csi2 does not support any format conversion. So the format on sink and source pad must always match. This limitation is handled for ACTIVE state via mbus_format member of bridge private structure.
To enforce this limitation in TRY state, user space should only be able to set format on sink pad. The sink format must be propagated to source pad to ensure consistent behavior.
This also aligns the driver with userspace relying on media controller based format negotiation.
--- .../sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +15 -4
@@ -369,15 +369,26 @@ static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev,struct v4l2_mbus_framefmt *mbus_format = &format->format;struct mutex *lock = &csi2_dev->bridge.lock; +/* The format on the source pad always matches the sink pad. */ +if (format->pad !=SUN6I_MIPI_CSI2_PAD_SINK ) +return v4l2_subdev_get_fmt (subdev, state, format); +mutex_lock (lock);sun6i_mipi_csi2_mbus_format_prepare (mbus_format); -if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) - *v4l2_subdev_state_get_format (state, format->pad) = - *mbus_format; -else +if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) { +struct v4l2_mbus_framefmt *fmt; + + fmt =v4l2_subdev_state_get_format (state, format->pad); + *fmt = *mbus_format; + +/* Propagate the format to the source pad. */ + fmt =v4l2_subdev_state_get_format (state,SUN6I_MIPI_CSI2_PAD_SOURCE ); + *fmt = *mbus_format; + }else { csi2_dev->bridge.mbus_format = *mbus_format; + }mutex_unlock (lock);
Signature · 1 line
2.34.1
Hello Arash,
Thank you for the patch.
Quoted 3 lines · Arash Golgol
sun6i-mipi-csi2 does not support any format conversion. So the format on sink and source pad must always match. This limitation is handled for ACTIVE state via
mbus_formatmember of bridge private structure.
It's not even handled correctly though, the sun6i_mipi_csi2_set_fmt() function will accept setting the ACTIVE format on the source pad, when it shouldn't. You also fix this in your patch. This should be explained in the commit message.
Quoted 34 lines
To enforce this limitation in TRY state, user space should only be able to set format on sink pad. The sink format must be propagated to source pad to ensure consistent behavior.
This also aligns the driver with userspace relying on media controller based format negotiation.
Signed-off-by Arash Golgol <[email protected]>--- .../sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +6
-4@@ -369,15 +369,26 @@ static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev,struct v4l2_mbus_framefmt *mbus_format = &format->format;struct mutex *lock = &csi2_dev->bridge.lock; +/* The format on the source pad always matches the sink pad. */ +if (format->pad !=SUN6I_MIPI_CSI2_PAD_SINK ) +return v4l2_subdev_get_fmt (subdev, state, format); +mutex_lock (lock);sun6i_mipi_csi2_mbus_format_prepare (mbus_format); -if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) - *v4l2_subdev_state_get_format (state, format->pad) = - *mbus_format; -else +if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) { +struct v4l2_mbus_framefmt *fmt;
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev, · line 391
Quoted 10 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +9
-0+ + fmt =v4l2_subdev_state_get_format (state, format->pad); + *fmt = *mbus_format; + +/* Propagate the format to the source pad. */ + fmt =v4l2_subdev_state_get_format (state,SUN6I_MIPI_CSI2_PAD_SOURCE ); + *fmt = *mbus_format; + }else { csi2_dev->bridge.mbus_format = *mbus_format; + }
If you have a bit more time, could I ask you to take it one step further and convert the driver to framework-managed V4L2 subdev active ? This will simplify the implementation by dropping bridge.mbus_format.
You can find an example of the conversion in commit a2514b9a634a ("media: i2c: imx290: Use V4L2 subdev active state"). In a nutshell, you need to
- Call v4l2_subdev_init_finalize() at probe time and v4l2_subdev_cleanup() at remove time
- Drop the mutex (locking is now handled by the framework)
- Use v4l2_subdev_get_fmt() as the .get_fmt() handler
- Drop bridge.mbus_format and always use v4l2_subdev_get_pad_format() on the state passed to the subdev functions (which will be the ACTIVE or TRY state depending on the .which argument passed by userspace)
- In .s_stream(), access the state with v4l2_subdev_lock_and_get_active_state() as it's the only subdev operation that doesn't get the state as a parameter
You can ignore the control handler changes in commit a2514b9a634a, and you can leave subdev.state_lock NULL (the framework will then use an internal lock).
Quoted 3 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +0
-0mutex_unlock (lock);
Signature · 3 lines
Regards, Laurent Pinchart
Hi Laurent,
Thanks for the review and the detailed feedback.
Quoted 6 lines · Laurent Pinchart
Hello Arash,
Thank you for the patch.
On Mon, Jan 26, 2026 at 01:51:49PM +0330, Arash Golgol wrote:
Quoted 9 lines
Quoted 3 lines
sun6i-mipi-csi2 does not support any format conversion. So the format on sink and source pad must always match. This limitation is handled for ACTIVE state via
mbus_formatmember of bridge private structure.It's not even handled correctly though, the
sun6i_mipi_csi2_set_fmt()function will accept setting the ACTIVE format on the source pad, when it shouldn't. You also fix this in your patch. This should be explained in the commit message.
Yes, you're right.
Quoted 40 lines
Quoted 40 lines
To enforce this limitation in TRY state, user space should only be able to set format on sink pad. The sink format must be propagated to source pad to ensure consistent behavior.
This also aligns the driver with userspace relying on media controller based format negotiation.
Signed-off-by Arash Golgol <[email protected]>--- .../sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +12
-4@@ -369,15 +369,26 @@ static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev,struct v4l2_mbus_framefmt *mbus_format = &format->format;struct mutex *lock = &csi2_dev->bridge.lock; +/* The format on the source pad always matches the sink pad. */ +if (format->pad !=SUN6I_MIPI_CSI2_PAD_SINK ) +return v4l2_subdev_get_fmt (subdev, state, format); +mutex_lock (lock);sun6i_mipi_csi2_mbus_format_prepare (mbus_format); -if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) - *v4l2_subdev_state_get_format (state, format->pad) = - *mbus_format; -else +if (format->which ==V4L2_SUBDEV_FORMAT_TRY ) { +struct v4l2_mbus_framefmt *fmt; + + fmt =v4l2_subdev_state_get_format (state, format->pad); + *fmt = *mbus_format; + +/* Propagate the format to the source pad. */ + fmt =v4l2_subdev_state_get_format (state,SUN6I_MIPI_CSI2_PAD_SOURCE );
Quoted 9 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev, · line 391
Quoted 4 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +3
-0+ *fmt = *mbus_format; + }else { csi2_dev->bridge.mbus_format = *mbus_format; + }If you have a bit more time, could I ask you to take it one step further and convert the driver to framework-managed V4L2 subdev active ? This will simplify the implementation by dropping bridge.mbus_format.
Sure, I'll do that and submit a new patch converting the driver to framework-managed V4L2 subdev.
Quoted 15 lines
You can find an example of the conversion in commit
a2514b9a634a("media: i2c: imx290: Use V4L2 subdev active state"). In a nutshell, you need to- Call
v4l2_subdev_init_finalize()at probe time andv4l2_subdev_cleanup()at remove time
- Drop the mutex (locking is now handled by the framework)
- Usev4l2_subdev_get_fmt()as the .get_fmt() handler
- Drop bridge.mbus_format and always usev4l2_subdev_get_pad_format()on the state passed to the subdev functions (which will be the ACTIVE or TRY state depending on the .which argument passed by userspace)
- In .s_stream(), access the state withv4l2_subdev_lock_and_get_active_state()as it's the only subdev operation that doesn't get the state as a parameter
Quoted 4 lines
You can ignore the control handler changes in commit
a2514b9a634a, and you can leave subdev.state_lock NULL (the framework will then use an internal lock).
Thanks a lot for the example and roadmap of the changes.
Quoted 8 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c static int sun6i_mipi_csi2_set_fmt(struct v4l2_subdev *subdev, · line 394
Quoted 3 lines
drivers/media/platform/sunxi/sun6i-mipi-csi2/sun6i_mipi_csi2.c +0
-0mutex_unlock (lock);--
Regards,Laurent Pinchart
--
Regards,
Arash Golgol