| Message ID | 1489121599-23206-39-git-send-email-steve_longerbeam@mentor.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Mar 09, 2017 at 08:53:18PM -0800, Steve Longerbeam wrote: > From: Philipp Zabel <p.zabel@pengutronix.de> > > The csi_try_crop call in set_fmt should compare the cropping rectangle > to the currently set input format, not to the previous input format. Are we really sure that the cropping support is implemented correctly? I came across this while looking at what we're doing with the V4L2_SEL_FLAG_KEEP_CONFIG flag. Documentation/media/uapi/v4l/dev-subdev.rst defines the behaviour of the user API, and "Order of configuration and format propagation" says: The coordinates to a step always refer to the actual size of the previous step. The exception to this rule is the source compose rectangle, which refers to the sink compose bounds rectangle --- if it is supported by the hardware. 1. Sink pad format. The user configures the sink pad format. This format defines the parameters of the image the entity receives through the pad for further processing. 2. Sink pad actual crop selection. The sink pad crop defines the crop performed to the sink pad format. 3. Sink pad actual compose selection. The size of the sink pad compose rectangle defines the scaling ratio compared to the size of the sink pad crop rectangle. The location of the compose rectangle specifies the location of the actual sink compose rectangle in the sink compose bounds rectangle. 4. Source pad actual crop selection. Crop on the source pad defines crop performed to the image in the sink compose bounds rectangle. 5. Source pad format. The source pad format defines the output pixel format of the subdev, as well as the other parameters with the exception of the image width and height. Width and height are defined by the size of the source pad actual crop selection. Accessing any of the above rectangles not supported by the subdev will return ``EINVAL``. Any rectangle referring to a previous unsupported rectangle coordinates will instead refer to the previous supported rectangle. For example, if sink crop is not supported, the compose selection will refer to the sink pad format dimensions instead. Note step 3 above: scaling is defined by the ratio of the _sink_ crop rectangle to the _sink_ compose rectangle. So, lets say that the camera produces a 1280x720 image, and the sink pad format is configured with 1280x720. That's step 1. The sink crop operates within that rectangle, cropping it to an area. Let's say we're only interested in its centre, so we'd chose 640x360 with the top-left as 320,180. This is step 2. Then, if we want to down-scale by a factor of two, we'd set the sink compose selection to 320x180. This seems to be at odds with how the scaling is done in CSI at present: the selection implementations all reject attempts to configure the sink pad, instead only supporting crop rectangles on the source, and we use the source crop rectangle to define the down-scaling.
On 03/19/2017 08:22 AM, Russell King - ARM Linux wrote: > On Thu, Mar 09, 2017 at 08:53:18PM -0800, Steve Longerbeam wrote: >> From: Philipp Zabel <p.zabel@pengutronix.de> >> >> The csi_try_crop call in set_fmt should compare the cropping rectangle >> to the currently set input format, not to the previous input format. > Are we really sure that the cropping support is implemented correctly? > > I came across this while looking at what we're doing with the > V4L2_SEL_FLAG_KEEP_CONFIG flag. > > Documentation/media/uapi/v4l/dev-subdev.rst defines the behaviour of > the user API, and "Order of configuration and format propagation" says: > > The coordinates to a step always refer to the actual size of the > previous step. The exception to this rule is the source compose > rectangle, which refers to the sink compose bounds rectangle --- if it > is supported by the hardware. > > 1. Sink pad format. The user configures the sink pad format. This format > defines the parameters of the image the entity receives through the > pad for further processing. > > 2. Sink pad actual crop selection. The sink pad crop defines the crop > performed to the sink pad format. > > 3. Sink pad actual compose selection. The size of the sink pad compose > rectangle defines the scaling ratio compared to the size of the sink > pad crop rectangle. The location of the compose rectangle specifies > the location of the actual sink compose rectangle in the sink compose > bounds rectangle. > > 4. Source pad actual crop selection. Crop on the source pad defines crop > performed to the image in the sink compose bounds rectangle. > > 5. Source pad format. The source pad format defines the output pixel > format of the subdev, as well as the other parameters with the > exception of the image width and height. Width and height are defined > by the size of the source pad actual crop selection. > > Accessing any of the above rectangles not supported by the subdev will > return ``EINVAL``. Any rectangle referring to a previous unsupported > rectangle coordinates will instead refer to the previous supported > rectangle. For example, if sink crop is not supported, the compose > selection will refer to the sink pad format dimensions instead. > > Note step 3 above: scaling is defined by the ratio of the _sink_ crop > rectangle to the _sink_ compose rectangle. > > So, lets say that the camera produces a 1280x720 image, and the sink > pad format is configured with 1280x720. That's step 1. > > The sink crop operates within that rectangle, cropping it to an area. > Let's say we're only interested in its centre, so we'd chose 640x360 > with the top-left as 320,180. This is step 2. > > Then, if we want to down-scale by a factor of two, we'd set the sink > compose selection to 320x180. > > This seems to be at odds with how the scaling is done in CSI at > present: the selection implementations all reject attempts to > configure the sink pad, instead only supporting crop rectangles on > the source, Correct. Currently cropping is only supported at the source pad (step 4). Initially the CSI didn't support down-scaling, so step 3 is not supported, so the sink pad format/crop selection rectangle/crop compose rectangle are collapsed into the same sink pad format rectangle. Philipp later added support for /2 downscaling, but we didn't put this in the correct API, looks like this needs to move into the selection API at step 3 (sink pad compose rectangle). > and we use the source crop rectangle to define the > down-scaling. Yes. And maybe there is nothing wrong with that, because scaling is also defined by the source/sink _format_ ratios (if I'm not mistaken), so looking at this another way, we're just defining scaling in the CSI via another legal API. Steve
On Sun, 2017-03-19 at 12:08 -0700, Steve Longerbeam wrote: > > On 03/19/2017 08:22 AM, Russell King - ARM Linux wrote: > > On Thu, Mar 09, 2017 at 08:53:18PM -0800, Steve Longerbeam wrote: > >> From: Philipp Zabel <p.zabel@pengutronix.de> > >> > >> The csi_try_crop call in set_fmt should compare the cropping rectangle > >> to the currently set input format, not to the previous input format. > > Are we really sure that the cropping support is implemented correctly? > > > > I came across this while looking at what we're doing with the > > V4L2_SEL_FLAG_KEEP_CONFIG flag. > > > > Documentation/media/uapi/v4l/dev-subdev.rst defines the behaviour of > > the user API, and "Order of configuration and format propagation" says: > > > > The coordinates to a step always refer to the actual size of the > > previous step. The exception to this rule is the source compose > > rectangle, which refers to the sink compose bounds rectangle --- if it > > is supported by the hardware. > > > > 1. Sink pad format. The user configures the sink pad format. This format > > defines the parameters of the image the entity receives through the > > pad for further processing. > > > > 2. Sink pad actual crop selection. The sink pad crop defines the crop > > performed to the sink pad format. > > > > 3. Sink pad actual compose selection. The size of the sink pad compose > > rectangle defines the scaling ratio compared to the size of the sink > > pad crop rectangle. The location of the compose rectangle specifies > > the location of the actual sink compose rectangle in the sink compose > > bounds rectangle. > > > > 4. Source pad actual crop selection. Crop on the source pad defines crop > > performed to the image in the sink compose bounds rectangle. > > > > 5. Source pad format. The source pad format defines the output pixel > > format of the subdev, as well as the other parameters with the > > exception of the image width and height. Width and height are defined > > by the size of the source pad actual crop selection. > > > > Accessing any of the above rectangles not supported by the subdev will > > return ``EINVAL``. Any rectangle referring to a previous unsupported > > rectangle coordinates will instead refer to the previous supported > > rectangle. For example, if sink crop is not supported, the compose > > selection will refer to the sink pad format dimensions instead. > > > > Note step 3 above: scaling is defined by the ratio of the _sink_ crop > > rectangle to the _sink_ compose rectangle. The above paragraph suggests we skip any rectangles that are not supported. In our case that would be 3. and 4., since the CSI can't compose into a larger frame. I hadn't realised that the crop selection currently happens on the source pad. The hardware actually only supports cropping of the input (the crop rectangle we write into the window registers are before downscaling). So the crop rectangle should be moved to the sink pad. > > So, lets say that the camera produces a 1280x720 image, and the sink > > pad format is configured with 1280x720. That's step 1. > > > > The sink crop operates within that rectangle, cropping it to an area. > > Let's say we're only interested in its centre, so we'd chose 640x360 > > with the top-left as 320,180. This is step 2. >> > > Then, if we want to down-scale by a factor of two, we'd set the sink > > compose selection to 320x180. Except when composing is not supported. If the sink compose and source crop rectangles are not supported, the source pad format takes their place in determining the scaling output resolution. At least that's how I read the documentation. > > This seems to be at odds with how the scaling is done in CSI at > > present: the selection implementations all reject attempts to > > configure the sink pad, instead only supporting crop rectangles on > > the source, > > Correct. Currently cropping is only supported at the source pad > (step 4). > > Initially the CSI didn't support down-scaling, so step 3 is not supported, > so the sink pad format/crop selection rectangle/crop compose rectangle > are collapsed into the same sink pad format rectangle. > > Philipp later added support for /2 downscaling, but we didn't put this in > the correct API, looks like this needs to move into the selection API at > step 3 (sink pad compose rectangle). I am not sure about this. Wouldn't moving the input crop to the sink pad be enough? If we added support for the sink pad compose rectangle, that wouldn't actually allow to compose the CSI output into a larger frame. Since the subdevice can't compose, I'd leave the sink compose rectangle disabled. > > and we use the source crop rectangle to define the > > down-scaling. We use the source pad format to define the downscaling relative to the source crop rectangle (which is wrong, it should be relative to the sink crop rectangle). > Yes. And maybe there is nothing wrong with that, because scaling is also > defined by the source/sink _format_ ratios (if I'm not mistaken), so looking > at this another way, we're just defining scaling in the CSI via another > legal API. I didn't touch the crop rectangle at all, just setting the input resolution on the sink pad and the desired output resolution on the source pad should work. regards Philipp
On Mon, Mar 20, 2017 at 12:55:26PM +0100, Philipp Zabel wrote: > The above paragraph suggests we skip any rectangles that are not > supported. In our case that would be 3. and 4., since the CSI can't > compose into a larger frame. I hadn't realised that the crop selection > currently happens on the source pad. I'd recommend viewing the documentation in its post-processed version, because then you get the examples as pictures, and they say that a picture is worth 1000 words. See https://linuxtv.org/downloads/v4l-dvb-apis/uapi/v4l/dev-subdev.html There is almost an exact example of what we're trying to do - it's figure 4.6. Here, we have a sink pad with a cropping rectangle on the input, which is then scaled to a composition rectangle (there's no bounds rectangle, and it's specified that in such a case the top,left of the composition rectangle will always be 0,0 - see quote below). Where it differs is that the example also supports source cropping for two source pads. We don't support that. The same document says: Scaling support is optional. When supported by a subdev, the crop rectangle on the subdev's sink pad is scaled to the size configured using the :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the subdev supports scaling but not composing, the top and left values are not used and must always be set to zero. which in itself describes _exactly_ our hardware here as far as the cropping and scaling the hardware supports. > Except when composing is not supported. If the sink compose and source > crop rectangles are not supported, the source pad format takes their > place in determining the scaling output resolution. At least that's how > I read the documentation. This isn't how I read it. Scaling is the relationship between the size of the sink crop and sink compose rectangle. Composition requires that there be a composition bounds rectangle to define the composition space, and the top,left of the composition rectangle be adjustable to place the composition rectangle within that space. The above quoted paragraph from the documentation backs up my view in its final sentence - it doesn't say "if the subdev supports scaling but not composing, there is no composition rectangle" but says that there _is_ one but its top,left coordinates are always zero.
On Mon, 2017-03-20 at 12:08 +0000, Russell King - ARM Linux wrote: > On Mon, Mar 20, 2017 at 12:55:26PM +0100, Philipp Zabel wrote: > > The above paragraph suggests we skip any rectangles that are not > > supported. In our case that would be 3. and 4., since the CSI can't > > compose into a larger frame. I hadn't realised that the crop selection > > currently happens on the source pad. > > I'd recommend viewing the documentation in its post-processed version, > because then you get the examples as pictures, and they say that a > picture is worth 1000 words. See > > https://linuxtv.org/downloads/v4l-dvb-apis/uapi/v4l/dev-subdev.html > > There is almost an exact example of what we're trying to do - it's > figure 4.6. Here, we have a sink pad with a cropping rectangle on > the input, which is then scaled to a composition rectangle (there's > no bounds rectangle, and it's specified that in such a case the > top,left of the composition rectangle will always be 0,0 - see quote > below). > > Where it differs is that the example also supports source cropping > for two source pads. We don't support that. > > The same document says: > > Scaling support is optional. When supported by a subdev, the crop > rectangle on the subdev's sink pad is scaled to the size configured > using the > :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL > using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the > subdev supports scaling but not composing, the top and left values are > not used and must always be set to zero. Right, this sentence does imply that when scaling is supported, there must be a sink compose rectangle, even when composing is not. I have previously set up scaling like this: media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" Does this mean, it should work like this instead? media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" media-ctl --set-v4l2 "'ipu1_csi0':0[fmt:UYVY2X8/1920x1080@1/60,compose:(0,0)/960x540]" media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" I suppose setting the source pad format should not be allowed to modify the sink compose rectangle. regards Philipp
On Mon, Mar 20, 2017 at 03:00:51PM +0100, Philipp Zabel wrote: > On Mon, 2017-03-20 at 12:08 +0000, Russell King - ARM Linux wrote: > > The same document says: > > > > Scaling support is optional. When supported by a subdev, the crop > > rectangle on the subdev's sink pad is scaled to the size configured > > using the > > :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL > > using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the > > subdev supports scaling but not composing, the top and left values are > > not used and must always be set to zero. > > Right, this sentence does imply that when scaling is supported, there > must be a sink compose rectangle, even when composing is not. > > I have previously set up scaling like this: > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > Does this mean, it should work like this instead? > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':0[fmt:UYVY2X8/1920x1080@1/60,compose:(0,0)/960x540]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > I suppose setting the source pad format should not be allowed to modify > the sink compose rectangle. That is what I believe having read these documents several times, but we need v4l2 people to confirm. Note that setting the format on 'ipu1_csi0':0 should already be done by the previous media-ctl command, so it should be possible to simplify that to: media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" media-ctl --set-v4l2 "'ipu1_csi0':0[compose:(0,0)/960x540]" media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" I have tripped over a bug in media-ctl when specifying both a crop and compose rectangle - the --help output suggests that "," should be used to separate them. media-ctl rejects that, telling me the character at the "," should be "]". Replacing the "," with " " allows media-ctl to accept it and set both rectangles, so it sounds like a parser bug - I've not looked into this any further yet.
On Mon, Mar 20, 2017 at 02:17:05PM +0000, Russell King - ARM Linux wrote: > On Mon, Mar 20, 2017 at 03:00:51PM +0100, Philipp Zabel wrote: > > On Mon, 2017-03-20 at 12:08 +0000, Russell King - ARM Linux wrote: > > > The same document says: > > > > > > Scaling support is optional. When supported by a subdev, the crop > > > rectangle on the subdev's sink pad is scaled to the size configured > > > using the > > > :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL > > > using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the > > > subdev supports scaling but not composing, the top and left values are > > > not used and must always be set to zero. > > > > Right, this sentence does imply that when scaling is supported, there > > must be a sink compose rectangle, even when composing is not. > > > > I have previously set up scaling like this: > > > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > > > Does this mean, it should work like this instead? > > > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > > media-ctl --set-v4l2 "'ipu1_csi0':0[fmt:UYVY2X8/1920x1080@1/60,compose:(0,0)/960x540]" > > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > > > I suppose setting the source pad format should not be allowed to modify > > the sink compose rectangle. > > That is what I believe having read these documents several times, but > we need v4l2 people to confirm. > > Note that setting the format on 'ipu1_csi0':0 should already be done by > the previous media-ctl command, so it should be possible to simplify > that to: > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':0[compose:(0,0)/960x540]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" There is a slightly puzzling bit in the documentation. If we consider the CSI, and the sink compose rectangle size has to always match the the source output pad format size (since in hardware they are one of the same), then: Inside subdevs, the order of image processing steps will always be from the sink pad towards the source pad. This is also reflected in the order in which the configuration must be performed by the user: the changes made will be propagated to any subsequent stages. If this behaviour is not desired, the user must set ``V4L2_SEL_FLAG_KEEP_CONFIG`` flag. This ^^^^ flag causes no propagation of the changes are allowed in any ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ circumstances. This may also cause the accessed rectangle to be adjusted ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ by the driver, depending on the properties of the underlying hardware. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ presumably, this means if we get a request to change the source compose rectangle with V4L2_SEL_FLAG_KEEP_CONFIG set, we need to force its size to be the current output format size. I don't think we can do anything else - because the above makes it very clear that any following stages shall not be updated. The last sentence appears to give permission to do that. This also has implications when changing the sink crop - the sink crop (eg) must not be smaller than the sink compose, as we don't support scaling up in CSI. It seems to me that V4L2_SEL_FLAG_KEEP_CONFIG in practice changes the way validation of the request works. So, rather than validating the request against the upstream rectangle and propagating downstream, it needs to be validated against both the upstream and downstream rectangles instead. It seems there's many subtleties to this...
On Mon, 2017-03-20 at 14:17 +0000, Russell King - ARM Linux wrote: > On Mon, Mar 20, 2017 at 03:00:51PM +0100, Philipp Zabel wrote: > > On Mon, 2017-03-20 at 12:08 +0000, Russell King - ARM Linux wrote: > > > The same document says: > > > > > > Scaling support is optional. When supported by a subdev, the crop > > > rectangle on the subdev's sink pad is scaled to the size configured > > > using the > > > :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL > > > using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the > > > subdev supports scaling but not composing, the top and left values are > > > not used and must always be set to zero. > > > > Right, this sentence does imply that when scaling is supported, there > > must be a sink compose rectangle, even when composing is not. > > > > I have previously set up scaling like this: > > > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > > > Does this mean, it should work like this instead? > > > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > > media-ctl --set-v4l2 "'ipu1_csi0':0[fmt:UYVY2X8/1920x1080@1/60,compose:(0,0)/960x540]" > > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > > > I suppose setting the source pad format should not be allowed to modify > > the sink compose rectangle. > > That is what I believe having read these documents several times, but > we need v4l2 people to confirm. > > Note that setting the format on 'ipu1_csi0':0 should already be done by > the previous media-ctl command, so it should be possible to simplify > that to: > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':0[compose:(0,0)/960x540]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" Thanks, that works, too. > I have tripped over a bug in media-ctl when specifying both a crop and > compose rectangle - the --help output suggests that "," should be used > to separate them. media-ctl rejects that, telling me the character at > the "," should be "]". Replacing the "," with " " allows media-ctl to > accept it and set both rectangles, so it sounds like a parser bug - I've > not looked into this any further yet. I can confirm this. I don't see any place in v4l2_subdev_parse_pad_format that handles the "," separator. There's just whitespace skipping between the v4l2-properties. regards Philipp
On 03/20/2017 07:00 AM, Philipp Zabel wrote: > On Mon, 2017-03-20 at 12:08 +0000, Russell King - ARM Linux wrote: >> On Mon, Mar 20, 2017 at 12:55:26PM +0100, Philipp Zabel wrote: >>> The above paragraph suggests we skip any rectangles that are not >>> supported. In our case that would be 3. and 4., since the CSI can't >>> compose into a larger frame. I hadn't realised that the crop selection >>> currently happens on the source pad. >> I'd recommend viewing the documentation in its post-processed version, >> because then you get the examples as pictures, and they say that a >> picture is worth 1000 words. See >> >> https://linuxtv.org/downloads/v4l-dvb-apis/uapi/v4l/dev-subdev.html >> >> There is almost an exact example of what we're trying to do - it's >> figure 4.6. Here, we have a sink pad with a cropping rectangle on >> the input, which is then scaled to a composition rectangle (there's >> no bounds rectangle, and it's specified that in such a case the >> top,left of the composition rectangle will always be 0,0 - see quote >> below). >> >> Where it differs is that the example also supports source cropping >> for two source pads. We don't support that. >> >> The same document says: >> >> Scaling support is optional. When supported by a subdev, the crop >> rectangle on the subdev's sink pad is scaled to the size configured >> using the >> :ref:`VIDIOC_SUBDEV_S_SELECTION <VIDIOC_SUBDEV_G_SELECTION>` IOCTL >> using ``V4L2_SEL_TGT_COMPOSE`` selection target on the same pad. If the >> subdev supports scaling but not composing, the top and left values are >> not used and must always be set to zero. > Right, this sentence does imply that when scaling is supported, there > must be a sink compose rectangle, even when composing is not. Ok, this all makes consistent sense to me too. So: - the CSI hardware cropping rectangle should be specified via the sink pad crop selection. - the CSI hardware /2 downscaler should be specified via the sink pad compose selection. - the final source pad rectangle is the same as the sink pad compose rectangle. So that leaves only step 4 (source pad crop selection) as unsupported. Steve > I have previously set up scaling like this: > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > Does this mean, it should work like this instead? > > media-ctl --set-v4l2 "'ipu1_csi0_mux':2[fmt:UYVY2X8/1920x1080@1/60]" > media-ctl --set-v4l2 "'ipu1_csi0':0[fmt:UYVY2X8/1920x1080@1/60,compose:(0,0)/960x540]" > media-ctl --set-v4l2 "'ipu1_csi0':2[fmt:AYUV32/960x540@1/30]" > > I suppose setting the source pad format should not be allowed to modify > the sink compose rectangle. > > regards > Philipp >
diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c index e5105ec..cf070be 100644 --- a/drivers/staging/media/imx/imx-media-csi.c +++ b/drivers/staging/media/imx/imx-media-csi.c @@ -989,13 +989,11 @@ __csi_get_fmt(struct csi_priv *priv, struct v4l2_subdev_pad_config *cfg, static int csi_try_crop(struct csi_priv *priv, struct v4l2_rect *crop, struct v4l2_subdev_pad_config *cfg, - enum v4l2_subdev_format_whence which, + struct v4l2_mbus_framefmt *infmt, struct imx_media_subdev *sensor) { struct v4l2_of_endpoint *sensor_ep; - struct v4l2_mbus_framefmt *infmt; - infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, which); sensor_ep = &sensor->sensor_ep; crop->width = min_t(__u32, infmt->width, crop->width); @@ -1178,8 +1176,7 @@ static int csi_set_fmt(struct v4l2_subdev *sd, crop.top = 0; crop.width = sdformat->format.width; crop.height = sdformat->format.height; - ret = csi_try_crop(priv, &crop, cfg, - sdformat->which, sensor); + ret = csi_try_crop(priv, &crop, cfg, &sdformat->format, sensor); if (ret) goto out; @@ -1263,6 +1260,7 @@ static int csi_set_selection(struct v4l2_subdev *sd, struct v4l2_subdev_selection *sel) { struct csi_priv *priv = v4l2_get_subdevdata(sd); + struct v4l2_mbus_framefmt *infmt; struct imx_media_subdev *sensor; int ret = 0; @@ -1296,7 +1294,8 @@ static int csi_set_selection(struct v4l2_subdev *sd, goto out; } - ret = csi_try_crop(priv, &sel->r, cfg, sel->which, sensor); + infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, sel->which); + ret = csi_try_crop(priv, &sel->r, cfg, infmt, sensor); if (ret) goto out;