This bug is the mesa component of https://bugs.freedesktop.org/show_bug.cgi?id=90854. We're getting hangs, which the kernel fails to recover from. The hangs only occur on more recent SKL steppings. The older steppings don't seem to have an issue. With all the latest hang fixes for mesa, sometimes the test does complete for me, but if I run it in a loop, it will hang eventually. Here is an example of a command which will hang: bin/fbo-depth-array depth-clear -auto -fbo
Created attachment 116680 [details] Error sate from mesa 10.6 with push constant fix
If I make the FBO size 4k square instead of 8k, the test doesn't hang.
Reproduced with last setup: The system hangs during the test ext_texture_array@fbo-depth-array. Setup: ------- Hardware Platform: SKY LAKE Y A0 CPU : Intel(R) Core(TM) m3-6Y30 CPU @ 0.8GHz 4MB (family: 6, model: 78 stepping: 3) MCP : SKL-Y D1 2+2 (ou ULX-D1) QDF : QYV3 CPU : SKL D0 Chipset PCH: Sunrise Point LP C1 CRB : SKY LAKE Y LPDDR3 RVP3 CRB FAB2 Reworks : All Mandatories + FBS02 & FBS03, O-06 Software Linux : Ubuntu 14.04 LTS 64 bits BIOS : SKLSE2R1.R00.X085.B02.150601337 ME FW : 11.0.0.1149 Ksc (EC FW): 1.15 Kernel 4.1-0 (drm-intel-nightly-2015-06-27) Mesa: mesa-10.5.8 (master) 24b043aab73ce066ded6e4bc93f589008dfc8484 Xf86_video_intel: 2.99.917 (master) baec802b21387d04aebb10ac29e719a1800c5aa0 Libdrm: libdrm-2.4.61 (master) 203983f842a889b279698fdea46e83ee4450a1db libva: libva-1.6.0.pre1 (master) 0f88a645ab3cea69d63371189e53cd465ab95a20 intel-driver: 1.6.0.pre1 (master) f3f74ea23601750078215fad04dde6748364b88d xorg: 1.17.99 Xserver: xorg-server-1.17.2 (master) 2123f7682d522619f101b05fb75efa75dabbe371 Piglit: (master) 107318d835dbbf51af55c62abb2aee154822a4c7
As indicated by Ben in comment 2, I think the failure is related to the fact that it is creating a really large colour buffer (8192x8192) and has nothing to do with the depth/stencil buffer. I can replicate a hang by running the really simple test below: https://raw.githubusercontent.com/bpeel/piglit/big-window/tests/fbo/big-window.c If I run it a few times (fewer than 20) then it consistently hangs. This doesn't use a depth or stencil buffer at all and just does a glClear on the large colour buffer. The width that it is using (8192) is a little bit special because that is the point at which Mesa will switch to using a linear buffer instead of tiling (see brw_tex_layout.c line 678). If I comment out that section so that it uses y-tiling anyway then it doesn't hang any more.
This seems fixed now? I cannot reproduce using drm-intel-nightly (e1561a5) with Mesa 10.6.4 or Mesa 11.
Bug was in kernel. No change in mesa.
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.