A number of test failures have been observed when running ES2-CTS.functional.fragment_ops.blend.* tests on IVB with the 4.13 kernel. Setting INTEL_DEBUG=reemit (again, on 4.13) causes these tests to pass consistently. In order to reproduce this, we had to loop these tests and run multiple loops at the same time on the system. It took a handful of iterations for the failure to happen. These failures do not occur when the kernel version is rolled back to 4.9. No other hardware platforms were tested.
Hi, Have you been able to bisect this?
Bunch of things: - needs bisect (yes will take time, but it's mostly machine time) - might also be a mesa bug, since 4.9->4.13 enables a bunch of things in mesa afaik
I can confirm Clayton's report, and bisected it way back to the 4.11-rc1 timeframe: b7048ea12fbb2724ee0cd30752d4fac43cab0651 is the first bad commit commit b7048ea12fbb2724ee0cd30752d4fac43cab0651 Author: Ville Syrjälä <ville.syrjala@linux.intel.com> Date: Wed Mar 15 16:31:58 2017 +0200 drm/i915: Do .init_clock_gating() earlier to avoid it clobbering watermarks To reproduce, do: $ while true; do ./glcts -n ES2-CTS.functional.fragment_ops.blend.*; done |& grep Fail in 3-4 terminals simultaneously. Failures should show up very quickly. FWIW, this is using the teamforge CTS repo, 'updated-cts' branch, built with DEQP_TARGET=x11. (It's old, we're in the process of updating to the modern ES CTS, but not quite there yet...)
Ville and Chris tell me this was fixed by 6ac43272768c ("drm/i915: Move init_clock_gating() back to where it was") but it hasn't made it back to stable yet.
Resolved in drm-tip
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.