Multicam issues on Ubuntu v4l2 backend w/ 3.19 kernel

[aharmat]

I followed the installation instructions and was able to run the capture and enumerate examples. The multicam example did not work (got UVCIOC_CTRL_QUERY:UVC_SET_CUR error 5, Input/output error) with 2 cameras plugged in.

However, there is a bigger problem now. I started writing my own example code just to get the hang of things, with only this in main:

rs::context ctx;
std::cout << "Found "<< ctx.get_device_count() << " devices" << std::endl;`

When run with 0 or 1 camera plugged in, the result was as expected. When I had 2 cameras plugged in, the program entered an uninterruptible sleep, so I had to reboot the system. After the reboot, I can't even get a single camera to function correctly. c-capture reports UVCIOC_CTRL_QUERY:UVC_SET_CUR error 2, No such file or directory, and dmesg shows

[   79.060507] uvcvideo: Failed to set UVC commit control : -32 (exp. 34).
[   79.060514] uvcvideo 4-2:1.1: resume error -5
[   79.062588] uvcvideo: Failed to set UVC commit control : -32 (exp. 34).
[   79.062594] uvcvideo 4-2:1.3: resume error -5
[   79.063175] uvcvideo: Failed to query (GET_LEN) UVC control 1 on unit 2: -32 (exp. 2).

I have re-run the patch-uvcvideo-ko-generic.sh script followed by sudo modprobe uvcvideo, but the result is still the same. I'm using the 3.19.0-36-generic kernel.

Any idea what could have caused this?

Thanks,

Adam

[ddiakopoulos]

Hi Adam,

And you are on Ubuntu 14.04.03 correct? Did you also install the udev rules along with the connectivity workaround patch?

[aharmat]

Correct. I installed both the udev rules and connectivity workaround. Before everything broke, I was having good success getting camera connectivity with one camera - much better than with https://github.com/PercATI/RealSense_ROS. I have tested the realsense cameras on a different machine and they are working fine, I must have screwed up some configuration on my computer.

[ddiakopoulos]

We're seen some odd problems between minor patch versions... The last I tested with from Ubuntu was 3.19.00-33 not -36. Which is your other machine?

[aharmat]

The other machine was running kernel 3.13 with the https://github.com/PercATI/RealSense_ROS driver. Any idea what is going on with uvcvideo? I have booted with an older kernel but the "failed to set UVC commit control" error in dmesg still happens. When I can get a single realsense working again I will try the 3.19.0-33 kernel.

[ddiakopoulos]

Failed to commit control is usually the case that the udev rules aren't working correctly and the driver can't touch the necessary files to communicate with the camera. I've also seen this issue while accidentally running the odd command or script as root, which causes far reaching permissions errors for anything related to uvcvideo.

[aharmat]

Looking into the udev rules, it turned out that running usb-R200-in_udev was failing because /dswork.lock already existed. Deleting /dswork.lock fixed the problem related to uvcvideo.

[aharmat]

Installed kernel 3.19.0-33, single realsense works, two don't.
Symptoms:

1. c-enumerate shows two cameras with different serial numbers
2. Running cpp-multicam results in UVCIOC_CTRL_QUERY:UVC_SET_CUR error 2, No such file or directory or UVCIOC_CTRL_QUERY:UVC_SET_CUR error 5, Input/output error (not sure what causes one or the other message to appear). After this error is generated by cpp-multicam, c-enumerate no longer works (also displays the same error message).
3. dmesg log:

[  173.944938] uvcvideo: Failed to query (GET_DEF) UVC control 4 on unit 7: -32 (exp. 2).
[  173.945018] uvcvideo: Failed to query (GET_DEF) UVC control 5 on unit 7: -32 (exp. 1).
[  173.945098] uvcvideo: Failed to query (GET_CUR) UVC control 11 on unit 7: -32 (exp. 1).
[  173.945178] uvcvideo: Failed to query (GET_DEF) UVC control 8 on unit 7: -32 (exp. 2).
[  173.945258] uvcvideo: Failed to query (GET_DEF) UVC control 1 on unit 7: -32 (exp. 2).
[  184.831949] uvcvideo: Failed to set UVC commit control : -32 (exp. 34).
[  184.831956] uvcvideo 2-3:1.1: resume error -5
[  184.834042] uvcvideo: Failed to set UVC commit control : -32 (exp. 34).
[  184.834050] uvcvideo 2-3:1.3: resume error -5
[  184.918344] uvcvideo: Failed to query (GET_LEN) UVC control 1 on unit 2: -32 (exp. 2).

Hope this is useful, please let me know if you need any more information.

[ddiakopoulos]

It appears as through the R200 connectivity workaround isn't doing the right thing for multiple devices and failing to clean up after itself. Maybe you can try with the latest 4.2.6 stable kernel -- will test this myself shortly as well. In 4.2.xx kernels, no connectivity workaround is necessary.

Good find on the dswork.lock discovery. That script is one of the few things we didn't write for this project, but it's clearly the source of a couple issues here.

[ddiakopoulos]

Can confirm multicamera support is fundamentally broken on 3.19.0 kernels with the same issues you describe. Now testing on 4.2.6 stable.

[aharmat]

Any results from testing with the 4.2.6 kernel?

[ddiakopoulos]

Yes, 4.2.6 is fully stable (no connectivity patch necessary, 4+ simultaneous cameras working).

[ddiakopoulos]

Adam -- closing this out because I see you have another repo on Github where this functionality is seemingly working for one of our multi-cam compatible configurations.

[aharmat]

Thanks Dimitri, I got it working with the 4.2 kernel a couple of days ago.