On Thu, Oct 08, 2020 at 11:12:37AM +0100, Chris Wilson wrote:

Quoting Ville Syrjälä (2020-10-08 11:04:22)

...

On Thu, Oct 08, 2020 at 10:54:36AM +0100, Chris Wilson wrote:

...

The GPU is trashing the low pages of its reserved memory upon reset. If we are using this memory for ringbuffers, then we will dutiful resubmit the trashed rings after the reset causing further resets, and worse. We must exclude this range from our own use. The value of 128KiB was found by empirical measurement on gen9.

Signed-off-by: Chris Wilson chris@chris-wilson.co.uk Cc: stable@vger.kernel.org

---

v2 comes with a selftest to see how widespread the issue is

Do we need something to make sure FBC isn't scribbling into stolen during the test?

igt runs the tests with disable_display=1, that is still being honoured right?

It just marks all connectors as disconnected now. If coming straight from boot presumaly whatever the BIOS lit up could still be on at that point. Though I guess we would have typically done a module reload for this? rmmod should shut things down IIRC. Also wouldn't think the BIOS would even enables FBC.

I did think about looking up the address to see if the drm_mm_node is in use to try and filter out such users. For starters, I just want to confirm that CI is seeing what I'm seeing. -Chris