Skip to contentSkip to conversation

OV5693 Kconfig missing a select VIDEO_V4L2_SUBDEV_API ?

DISCUSSION Hans de Goede 3 participants 10 messages Jun 14, 2023 – Jun 15, 2023 Media (V4L/DVB)

lore
Share & tools
lore
#0 lore raw

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Regards,

Hans

#1 lore raw

Hi,

Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Hmm, so while looking at the VIDEO_IMX290 entry in drivers/media/i2c/Kconfig I notice that the VIDEO_IMX274 entry directly above it seems to be missing "select V4L2_FWNODE".

I am starting to think that we need a V4L2_SENSOR_DRIVER_DEPS which can be selected which then in turn selects all the bits modern sensor drivers need ?

Regards,

Hans

#2 lore raw
In reply to Hans de Goede

Hi Hans,

Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Signature · 3 lines
Regards,

Sakari Ailus
#3 lore raw

Hi Sakari,

Quoted 17 lines · On 6/14/23 18:05, [email protected] wrote:

Hi Hans,

Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

and I stopped checking after the ov* drivers since I think the above list makes my point.

Regards,

Hans

#4 lore raw

Hi Hans,

Quoted 50 lines · Hans de Goede

Hi Sakari,

Quoted 17 lines · On 6/14/23 18:05, [email protected] wrote:

Hi Hans,

Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Also cc Hans and Laurent.

Signature · 3 lines
Regards,

Sakari Ailus
#5 lore raw
Quoted 61 lines · On Wed, Jun 14, 2023 at 08:30:16PM +0000, [email protected] wrote:
Quoted 46 lines · Hans de Goede
Quoted 15 lines · On 6/14/23 18:05, [email protected] wrote:
Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Quoted 5 lines

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Also cc Hans and Laurent.

I'm all for simplifying the current state and removing the need to get every Kconfig entry right by moving the dependencies to a common location.

Signature · 3 lines
Regards,

Laurent Pinchart
#6 lore raw
Quoted 66 lines · Laurent Pinchart
Quoted 65 lines · On Wed, Jun 14, 2023 at 08:30:16PM +0000, [email protected] wrote:
Quoted 46 lines · Hans de Goede
Quoted 15 lines · On 6/14/23 18:05, [email protected] wrote:
Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Quoted 5 lines
Quoted 1 line

Also cc Hans and Laurent.

I'm all for simplifying the current state and removing the need to get every Kconfig entry right by moving the dependencies to a common location.

Also, if sensor drivers are encouraged to use new APIs, but not all of them have been converted, I'd be fine selecting the new APIs unconditionally even if no selected sensor driver uses them.

Signature · 3 lines
Regards,

Laurent Pinchart
#7 lore raw

Hi Laurent,

Quoted 69 lines · Laurent Pinchart
Quoted 68 lines · Laurent Pinchart
Quoted 66 lines · On Wed, Jun 14, 2023 at 08:30:16PM +0000, [email protected] wrote:
Quoted 46 lines · Hans de Goede
Quoted 15 lines · On 6/14/23 18:05, [email protected] wrote:
Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Also cc Hans and Laurent.

Quoted 7 lines
Quoted 3 lines

I'm all for simplifying the current state and removing the need to get every Kconfig entry right by moving the dependencies to a common location.

Also, if sensor drivers are encouraged to use new APIs, but not all of them have been converted, I'd be fine selecting the new APIs unconditionally even if no selected sensor driver uses them.

There are old drivers that do not need these features and there has been discussion there may be a desire to disable features to make the kernel smaller (for e.g. OpenWRT), however I suspect in this case it's more likely drivers that are little used.

Similarly to sensor drivers needing V4L2_FWNODE, lens drivers will need V4L2_ASYNC.

I can post patches, not quite sure yet whether this approach is also better on stable kernels. Quite possibly so.

Signature · 3 lines
Kind regards,

Sakari Ailus
#8 lore raw

Hi Sakari,

Quoted 74 lines · On Thu, Jun 15, 2023 at 08:22:51AM +0000, [email protected] wrote:
Quoted 73 lines · Laurent Pinchart
Quoted 71 lines · Laurent Pinchart
Quoted 61 lines · On Wed, Jun 14, 2023 at 08:30:16PM +0000, [email protected] wrote:
Quoted 46 lines · Hans de Goede
Quoted 15 lines · On 6/14/23 18:05, [email protected] wrote:
Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Quoted 5 lines

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Also cc Hans and Laurent.

I'm all for simplifying the current state and removing the need to get every Kconfig entry right by moving the dependencies to a common location.

Quoted 8 lines
Quoted 3 lines

Also, if sensor drivers are encouraged to use new APIs, but not all of them have been converted, I'd be fine selecting the new APIs unconditionally even if no selected sensor driver uses them.

There are old drivers that do not need these features and there has been discussion there may be a desire to disable features to make the kernel smaller (for e.g. OpenWRT), however I suspect in this case it's more likely drivers that are little used.

Just to be clear, I didn't mean selecting V4L2_FWNODE, V4L2_ASYNC and VIDEO_V4L2_SUBDEV_API unconditionally, but when any sensor driver is selected. If a device has a camera, memory requirements increase significantly and I don't think enabling the above options will make a real difference.

Quoted 5 lines

Similarly to sensor drivers needing V4L2_FWNODE, lens drivers will need V4L2_ASYNC.

I can post patches, not quite sure yet whether this approach is also better on stable kernels. Quite possibly so.

Signature · 3 lines
Regards,

Laurent Pinchart
#9 lore raw

Hi Laurent,

Quoted 86 lines · Laurent Pinchart

Hi Sakari,

Quoted 74 lines · On Thu, Jun 15, 2023 at 08:22:51AM +0000, [email protected] wrote:
Quoted 73 lines · Laurent Pinchart
Quoted 71 lines · Laurent Pinchart
Quoted 61 lines · On Wed, Jun 14, 2023 at 08:30:16PM +0000, [email protected] wrote:
Quoted 46 lines · Hans de Goede
Quoted 15 lines · On 6/14/23 18:05, [email protected] wrote:
Quoted 6 lines · Hans de Goede

Hi All,

The ov5693 driver uses v4l2_subdev_get_try_crop() / v4l2_subdev_get_try_format() both of which are only defined if CONFIG_VIDEO_V4L2_SUBDEV_API=y .

Quoted 7 lines

Yet it does not do select VIDEO_V4L2_SUBDEV_API in its Kconfig bits ?

Note I've not seen any build errors because of this, I guess we somehow end up getting away with this...

But still I think the select should be added ?

Quoted 7 lines

I agree.

The reason there haven't been compile failures is that there's a vast number of sensor drivers that all select this so for a failure you'd need to select this one but none of the others.

I can send a fix.

Also see my follow-up email. If we're going to fix this we really should fix it properly. As mentioned in my folow-up email an intermediate Kconfig option might be best.

E.g. doing:

grep -l v4l2_async_register_subdev drivers/media/i2c/*.c

And comparing that to Kconfig finds the following Kconfig entries lacking a select V4L2_FWNODE / select V4l2_ASYNC

VIDEO_IMX208
VIDEO_IMX258
VIDEO_IMX274
VIDEO_IMX319
VIDEO_IMX355
VIDEO_OV6650
VIDEO_OV7740
VIDEO_OV9640
VIDEO_OV9650

Quoted 2 lines

and I stopped checking after the ov* drivers since I think the above list makes my point.

Yeah, sometimes difficult to find errors get repeated. Luckily it's "just" a compilation problem.

Using V4L2 fwnode and V4L2 sub-device APIs are still unrelated as such although in practice they do often happen together. There are still quite a few sensor drivers that don't need both of them. Some can be compiled with VIDEO_V4L2_SUBDEV_API disabled, too, but I'm not sure how useful that really is. The rest are probably not usable outside a very specific scope, such as I²C async matching used by a handful of receiver drivers (none use MC, thus no sub-device API either).

Quoted 5 lines

Perhaps we could group these in two classes where the common class has V4L2_FWNODE and VIDEO_V4L2_SUBDEV_API selected? I'm not sure having an intermediate, somewhat obscure, option would be helpful.

Also cc Hans and Laurent.

I'm all for simplifying the current state and removing the need to get every Kconfig entry right by moving the dependencies to a common location.

Quoted 8 lines
Quoted 3 lines

Also, if sensor drivers are encouraged to use new APIs, but not all of them have been converted, I'd be fine selecting the new APIs unconditionally even if no selected sensor driver uses them.

There are old drivers that do not need these features and there has been discussion there may be a desire to disable features to make the kernel smaller (for e.g. OpenWRT), however I suspect in this case it's more likely drivers that are little used.

Quoted 5 lines

Just to be clear, I didn't mean selecting V4L2_FWNODE, V4L2_ASYNC and VIDEO_V4L2_SUBDEV_API unconditionally, but when any sensor driver is selected. If a device has a camera, memory requirements increase significantly and I don't think enabling the above options will make a real difference.

How about something like this:

drivers/media/i2c/Kconfig +9 -3
index 256d55bb2b1da..8a6961aafef98 100644
@@ -25,8 +25,14 @@ config VIDEO_IR_I2C
 # V4L2 I2C drivers that are related with Camera support
 #
 
-menu "Camera sensor devices"
-	visible if MEDIA_CAMERA_SUPPORT
+menuconfig VIDEO_CAMERA_SENSOR
+	bool "Camera sensor devices"
+	depends on MEDIA_CAMERA_SUPPORT
+	select V4L2_FWNODE
+	select VIDEO_V4L2_SUBDEV_API
+	default y
+
+if VIDEO_CAMERA_SENSOR
 
 config VIDEO_APTINA_PLL
 	tristate
@@ -797,7 +803,7 @@ config VIDEO_ST_VGXY61
 source "drivers/media/i2c/ccs/Kconfig"
 source "drivers/media/i2c/et8ek8/Kconfig"
 
-endmenu
+endif # VIDEO_CAMERA_SENSOR
 
 menu "Lens drivers"
 	visible if MEDIA_CAMERA_SUPPORT
Signature · 3 lines
Regards,

Sakari Ailus