Intel gpu memory corruption

Intel gpu memory corruption

[Baltazár Radics]

Hello!

My laptop (ThinkPad T460) seems to have a memory corruption issue that
only occures when the gpu is in use (it has `Intel Corporation Skylake
GT2 [HD Graphics 520] (rev 07)` as reported by lspci).

I haven't been able to reproduce the corruption with standard memory
testing utilities like lenovo's builtin hardware diagnostic tool,
memtest86+, or even the user-space program memtester when it's the only
thing running.

However, running memtester alongside vkmark for example can reproduce
it quite consistently. It will always be a single address for a given
instance of memtester, but looking into /proc/[pid]/pagemap revealed
that seemingly it's always the same hardware address.

With this information, I think I managed to stop it from happening by
appending `memmap=4K$0x1F9D7C000` to my kernel commandline to stop that
address from being allocated. Since then I haven't been able to catch
it with memtester, but I did have a crash that kinda resembled the ones
I had earlier. Many processes segfaulted and I had some `Bad swap file
entry` errors in my dmesg.

I haven't been able to do testing on other OSes yet, but since none of
the regular memtests have found any issues, I'm fairly certain this is
not a hardware issue with my ram. Could still be a hardware issue with
the gpu itself, but for now I'm guessing this is a gpu driver bug.

Is there anything else I can test to confirm that this is i915's fault,
and if so, anything I can do to help track down the bug?

Thanks!

Intel GPU memory corruption

[Baltazár Radics]

Hello!

I have a Lenovo Thinkpad T460 with an integrated `Intel Corporation
Skylake GT2 [HD Graphics 520]` GPU. Whenever there's intensive usage of
it, after a while I experience memory corruption. Under normal usage,
this results in random programs segfaulting for example.

I have done several memory tests, including the Lenovo hardware
diagnostics included with my BIOS and the userspace utility `memtester`
as well. When the GPU is not used, no tests have found any issues, so
I'm fairly certain my actual RAM sticks are fine.

However, when I run e.g. `vkmark` or `mpv` alongside `memtester`, it
will report issues after one or two loops. During the time I ran it,
all failures happened at the same offset.

Based on this, I suspect the video driver is the culprit somehow, but
I'm still not entirely certain it's not a hardware issue. If anyone has
ideas on how to track this down, I'd be happy to do some tests.