raspberrypi / linux

Describe the bug

Use rpi-update to update to the latest 6.6.18 kernel and test the ov64a40 camera, and find that the camera cannot be recognized normally.

Steps to reproduce the behaviour

sudo rpi-update
Modify /boot/firmware/config.txt and add dtoverlay=ov64a40
reboot

Device (s)

Raspberry Pi 4 Mod. B

System

Raspberry Pi reference 2023-12-05
Generated using pi-gen, https://github.com/RPi-Distro/pi-gen, 70cd6f2a1e34d07f5cba7047aea5b92457372e05, stage4

Feb 14 2024 14:57:03
Copyright (c) 2012 Broadcom
version 409a5df2480fbb1bd36b3d48743ec5a714616f44 (clean) (release) (start)

Linux raspberrypi 6.6.18-v8+ #1736 SMP PREEMPT Fri Feb 23 15:37:38 GMT 2024 aarch64 GNU/Linux

Logs

[ 9.136847] vc4-drm gpu: [drm] fb0: vc4drmfb frame buffer device
[ 9.235295] ov64a40 10-0036: sub-device fwnode is an endpoint!
[ 9.235358] ov64a40 10-0036: failed to register sensor sub-device: -22
[ 9.248915] ov64a40: probe of 10-0036 failed with error -22

Additional context

Hi @jmondi @kbingham

After a simple debugging, I found that sd.fwnode was artificially set as an endpoint,

linux/drivers/media/i2c/ov64a40.c

Lines 3517 to 3518 in f381965

    
           /* Register the subdev on the endpoint, so don't put it yet. */ 
        
           ov64a40->sd.fwnode = endpoint;

but v4l2-async clearly requires that the fwnode of subdev cannot be an endpoint.

linux/drivers/media/v4l2-core/v4l2-async.c

Lines 799 to 812 in f381965

    
           	/* 
        
           	 * No reference taken. The reference is held by the device (struct 
        
           	 * v4l2_subdev.dev), and async sub-device does not exist independently 
        
           	 * of the device at any point of time. 
        
           	 * 
        
           	 * The async sub-device shall always be registered for its device node, 
        
           	 * not the endpoint node. 
        
           	 */ 
        
           	if (!sd->fwnode && sd->dev) { 
        
           		sd->fwnode = dev_fwnode(sd->dev); 
        
           	} else if (fwnode_graph_is_endpoint(sd->fwnode)) { 
        
           		dev_warn(sd->dev, "sub-device fwnode is an endpoint!\n"); 
        
           		return -EINVAL; 
        
           	}

I am not sure why it was set as an endpoint in the first place.
Any ideas?

That's because in v6.1 (for which we sent the pull request to RPi kernel) the v4l2-async framework accepted endpoints, and at the time the consensus was to go in that direction
https://elixir.bootlin.com/linux/v6.1/source/drivers/media/v4l2-core/v4l2-async.c

Later on endpoint matching has been removed (see 1029939b3782 ("media: v4l: async: Simplify async sub-device fwnode matching")) and drivers got updated.

In facts, you can see that the version of the driver in upstream, doesn't set the fwnode
https://elixir.bootlin.com/linux/v6.8-rc6/source/drivers/media/i2c/ov64a40.c#L3510

When moving from v6.1 to v6.6 the matching on endpoint should have been dropped. I can send a PR for this.

Pull request sent #5998

@ArduCAM please test and confirm the PR (now merged) fixes the issue!