On Mon, 13 Apr 2026 10:05:45 +0200, Michal Pecio wrote:
> Question: can you kill it by starting some video application to set
> the camera up, closing it and then running this loop?
> 
> while :; do v4l2-ctl -d /dev/video0 --stream-mmap --stream-count=1; done

Hi again,

Any chance you could try it? And also the attached test program:

cc -lusb-1.0 hammerint.c -o hammerint
sudo ./hammerint 1532 0e05 0 85

Initial arguments are VID:PID of the device, next is the number of
an interface containing some interrupt endpoint and then the endpoint
address (including 8_ if IN).

I tried with a variety of SuperSpeed devices (UVC, NICs, hubs) and this
reliably breaks ASMedia HCs within seconds. If the same is the case on
Intel then it's a bigger problem than just UVC.

The video streaming loop breaks even more controllers. I have some
general idea how the streaming case could be dealt with, but not so
much the interrupt one. Maybe rate limiting. I found that avoiding
Set TR Dequeue to Link TRBs reduces failure rate, but not to zero.

Long ago I also looked at the issued command sequences and I haven't
noticed obvious errors or spec violations. Looks like a HW bug.

Regards,
Michal