Catastrophic shmem leak on kernel 6.19.5

[RealityAnomaly]

Over the course of 3-4 days, available shared memory will fill up and eventually exhaust all system memory, causing the VMs to start swapping and killing programs. The shared memory can't be traced to any programs or tmpfs i.e. it is truly leaked. The issue started occurring after I updated this module to the kernel 6.19.5 branch.

dmesg (including OOM information): https://gist.github.com/RealityAnomaly/b51fcc2714aed9e7298eddd3da9276dc

[bbaa-bbaa]

Based on the dmesg, it looks like this is from a Guest VM.
Is the memory leak only happening in the Guest VM?
What kernel version is your Host running? Do the host logs show anything interesting?

Since you're using the same version as the original kernel tree, this probably isn't a backporting issue.
However, when applying the new code, it's possible that some code paths from the newer version got mixed into the wrong #if branches. Could you try building and testing the original kernel code directly from here?

bbaa-bbaa/intel-mainline-tracking@bd7a13a

[RealityAnomaly]

Thanks for the quick response. Yes, the memory leak is only happening in guest VMs, the shared memory amount on the host is normal and it has remained stable for multiple weeks. It is running the exact same kernel as the VM.

The memory leak seems to somehow be tied to application usage in the guest VMs. Specifically, it seems to be caused by browsers. In the other two VMs I have VFs assigned to, the leak happens much more slowly.

I don't see anything concerning in the host dmesg that's related to i915.

Will see if I can test the mainline tracking code.