camera-streamer breaks udev monitoring

[Thugmek]

I have a weird error with camera-streamer on Raspberry Pi with my custom-built Yocto image. When I stream ov5647 camera with camera-streamer --camera-path=/base/soc/i2c0mux/i2c@1/ov5647@36 --camera-type=libcamera --camera-format=YUYV --camera-width=1296 --camera-height=972 --http-port=8081, it completely breaks udev monitoring. No events are captured by udevadm monitor --udev, no new devices are listed in /dev. Once I terminate streaming process, udev starts to work again (but actions performed during camera streaming, eg. USB inserts, are lost). If I stream USB camera with camera-streamer --camera-path=/dev/v4l/by-id/usb-046d_C270_HD_WEBCAM_200901010001-video-index0 --camera-type=v4l2 --camera-format=MJPEG --camera-width=1280 --camera-height=720 --http-port=8081, everything works fine. If I create a service with the same command, I use to stream ov5674 camera, udev monitor works fine. But when I make python script to monitor udevand switch cameras, udev monitor doesn't work inside this service and I won't get the actual filesystem state, when I runls /dev` inside the service. It seems that the camera streamer breaks udev in some local context. I've been stuck on this for a few days and I'm not able to investigate, what's going on here. Do you have any ideas?

[kbingham]

This is interesting. libcamera uses udev internally to monitor for events (i.e. camera hotplug events) ... but I wouldn't have expected that to block the main udev monitor.

Might be worth looking through https://git.libcamera.org/libcamera/libcamera.git/tree/src/libcamera/device_enumerator_udev.cpp and the api's used to see if there are any suspicious calls there that are documented to block the system. If so we'd have to look deeper into how we manage it.

[Thugmek]

By further investigation, I have found this error log in journalctl:

Sep 12 14:03:14 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup context command completion code 0x11.
Sep 12 14:03:14 MichalYocto kernel: usb 1-1.1: hub failed to enable device, error -22
Sep 12 14:03:14 MichalYocto kernel: usb 1-1.1: new high-speed USB device number 20 using xhci_hcd
Sep 12 14:03:14 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup context command completion code 0x11.
Sep 12 14:03:14 MichalYocto kernel: usb 1-1.1: hub failed to enable device, error -22
Sep 12 14:03:14 MichalYocto kernel: usb 1-1-port1: attempt power cycle
Sep 12 14:03:15 MichalYocto kernel: usb 1-1.1: new high-speed USB device number 21 using xhci_hcd
Sep 12 14:03:15 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup address command completion code 0x11.
Sep 12 14:03:15 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup address command completion code 0x11.
Sep 12 14:03:15 MichalYocto kernel: usb 1-1.1: device not accepting address 21, error -22
Sep 12 14:03:15 MichalYocto kernel: usb 1-1.1: new high-speed USB device number 22 using xhci_hcd
Sep 12 14:03:15 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup address command completion code 0x11.
Sep 12 14:03:16 MichalYocto kernel: xhci_hcd 0000:01:00.0: ERROR: unexpected setup address command completion code 0x11.
Sep 12 14:03:16 MichalYocto kernel: usb 1-1.1: device not accepting address 22, error -22
Sep 12 14:03:16 MichalYocto kernel: usb 1-1-port1: unable to enumerate USB device

It happens every time I plug in something to USB (Flashdisk, USB webcam, wireless mouse dongle).

[Thugmek]

Well, I have found quite weird connection between gpu_mem in /boot/config.txt and this behaviour. Currently, I have default value set to 256MB. As I lowered this value, there were still less and less errors. 64MB is not enough, 128MB is too much. There is sweet spot at 96MB, where ov5674 still works, but no USB error are present in kernel logs 😄 But it's not solution for my problem, because now USB camera hasn't enough memory to work. Trap next to trap...