Must omap34xxcam be a module?

Must omap34xxcam be a module?

[Michael Jones]

Hi Laurent & Sakari,

Regarding the omap3camera/devel branch:

In v4l2-common.c:v4l2_i2c_new_subdev_board(), request_module() is called to ensure that the sensor driver is already registered before registering the sensor device.  When I compile-in both my sensor driver and omap34xxcam with the kernel, this call to request_module() fails, and indeed omap34xxcam is initialized before my sensor driver, causing the omap34xxcam device registration to fail. When I leave omap34xxcam compiled-in and try to just let it load the sensor module when needed on bootup, request_module() fails.  I haven't managed to track down why that is.  When I compile both omap34xxcam and my sensor driver as modules, and load them after boot-up, registration succeeds.

Is it neccessary for omap34xxcam and its subdevices to be modules?  How are you guys building these?

Full disclosure: my sensor is actually an SPI device, but the v4l2_spi_new_subdev() function I'm actually using seems to be _very_ analogous to its I2C counterpart, so I'm assuming SPI is not responsible.

thanks,
Michael

MATRIX VISION GmbH, Talstrasse 16, DE-71570 Oppenweiler
Registergericht: Amtsgericht Stuttgart, HRB 271090
Geschaeftsfuehrer: Gerhard Thullner, Werner Armingeon, Uwe Furtner, Hans-Joachim Reich

Re: Must omap34xxcam be a module?

[Laurent Pinchart]

Hi Michael,

On Friday 13 August 2010 16:43:13 Michael Jones wrote:
> Hi Laurent & Sakari,
> 
> Regarding the omap3camera/devel branch:
> 
> In v4l2-common.c:v4l2_i2c_new_subdev_board(), request_module() is called to
> ensure that the sensor driver is already registered before registering the
> sensor device.  When I compile-in both my sensor driver and omap34xxcam
> with the kernel, this call to request_module() fails, and indeed
> omap34xxcam is initialized before my sensor driver, causing the
> omap34xxcam device registration to fail.
>
> When I leave omap34xxcam compiled-in and try to just let it load the sensor
> module when needed on bootup, request_module() fails.  I haven't managed to
> track down why that is.

That's because userspace isn't available yet when the omap34xxcam driver is 
initialized, so there's no way to load a module at that time.

> When I compile both omap34xxcam and my sensor driver as modules, and
> load them after boot-up, registration succeeds.
> 
> Is it neccessary for omap34xxcam and its subdevices to be modules?  How are
> you guys building these?

At the moment it's indeed necessary. It's a V4L2 core issue, not specific to 
omap34xxcam. I'm not aware of plans to fix this, but a proposal is welcome :-)

> Full disclosure: my sensor is actually an SPI device, but the
> v4l2_spi_new_subdev() function I'm actually using seems to be _very_
> analogous to its I2C counterpart, so I'm assuming SPI is not responsible.

The same issue exists with I2C, SPI is not responsible.

-- 
Regards,

Laurent Pinchart