6by9

I hadn't created the internal merge request for the patch as I hadn't had a chance to test. Based on your report of it working, I have now created that,...

And I wouldn't recommend you pointing users at test releases from issues/PRs as they will become out of date fairly quickly and hence potentially cause other issues.

Forum thread https://forums.raspberrypi.com/viewtopic.php?t=365356 If a subject has already been discussed on the forums, please include a link if you raise a Github issue. It saves going over the same ground...

> The github build artefacts are only kept for a limited time (maybe 1 month). 90 days, which is the maximum that Github allows https://github.com/raspberrypi/linux/blob/rpi-6.6.y/.github/workflows/kernel-build.yml#L108

Investigating. It looks like it fails and is still holding sm_state->lock, so `cat /sys/kernel/debug/vcsm-cma` blocks.

#6703 fixes one issue. Allocating and freeing kernel IDs took the spinlock, but looking up the value didn't. Another thread importing or freeing could therefore corrupt the idr whilst a...

Definitely timing related as adding logging reduces the rate of reproduction :-/ So it is almost certainly a thread safety issue, but I'm not seeing it at the moment.

The logging output is leaving me confused at the moment. I've tweaked the test case to have a separate dmabuf with a different size per thread, so that I can...

Not wasting my time at all - more eyes on code is always a good thing. I still remember the hours wasted much earlier in my career trying to find...

> I traced further into `vc_sm_cma_vchi_send_msg`. It allocates a cmd_blk via [vc_vchi_cmd_create](https://github.com/raspberrypi/linux/blob/rpi-6.12.y/drivers/staging/vc04_services/vc-sm-cma/vc_sm_cma_vchi.c#L426) and control reaches up to the [wait_for_completion_interruptible](https://github.com/raspberrypi/linux/blob/rpi-6.12.y/drivers/staging/vc04_services/vc-sm-cma/vc_sm_cma_vchi.c#L446) line. Then the callback event triggers and causes the NULL-deref. After...