Created attachment 106020 [details] dmesg System Environment: -------------------------- Platform: IVB Libdrm: (master)libdrm-2.4.56-21-g8c2ae1695913990cbe6c6a8aa294f354eba33bd0 Mesa: (master)4ea1565bbc49df79d4c39ba5491b0a83c2679a05 server: (master)xorg-server-1.16.0-152-g3a51418b2db353519a1779cf3cebbcc9afba2520 Xf86_video_intel:(master)2.99.916-2-ge0f7e9fc2f0b39b9e939ff48edea29950f125420 Libva: (master)e0d25ece01e7aba819c910e98c4fb4706cdab055 Libva_intel_driver:(master)b18479e4fdd7af7cf2840542ef19dbe9114fdeaf Kernel: (drm-intel-nightly)4a3d32734bdcef6813b31f06a58430436e98711e Bug detailed description: --------------------------- Clean boot system, run xinit, X is no response, screen is black and dmesg reports "*ERROR* pin & fence failed". Fail rate: drm-intel-nightly: 2/10 drm-intel-next-queued: 0/10 drm-intel-fixes: 0/10 drm-fixes: 0/10 drm-next: 0/10 Reproduce steps: ------------------------- 1. clean boot system 2. xinit
Created attachment 106021 [details] Xorg.0.log
You have several GPU hangs during boot... One of those hangs occurs during the modeset which ends up triggering an -EIO and failure. Can you please attach the error state whilst I try to sell Daniel on the virtues of my patch?
A bisect would be useful, or you can just try i915.enable_ppgtt=1, or you can try my shotgun at git://people.freedesktop.org/~ickle/linux-2.6 requests (i.e. the requests branch and http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?h=requests&id=7c7dd4183286496be6c24580b6bf347ba9162d5f in particular)
Created attachment 106025 [details] i915_error_state
Dies inside the first BLT. The PP_DIR_BASE and GFX_MODE look correct, so the LRI worked correctly. However, ACTHD is looping through a ppgtt that is all zero. A coherency problem? A TLB issue? PDEs?
(In reply to comment #3) > A bisect would be useful, or you can just try i915.enable_ppgtt=1, or you > can try my shotgun at > > git://people.freedesktop.org/~ickle/linux-2.6 requests > > (i.e. the requests branch and > http://cgit.freedesktop.org/~ickle/linux-2.6/commit/ > ?h=requests&id=7c7dd4183286496be6c24580b6bf347ba9162d5f in particular) Add i915.enable_ppgtt=1, run 10 cycles, it works well. About bisect, I am not sure which branch is broken, up to now, I only reproduce it on -nightly branch.
Don't worry about the bisect, it is pretty clear this is full-ppgtt fallout.
I tripped over the golden render state, but that should only affect haswell. I haven't had any misfortune with ivb yet.
# cat /sys/module/i915/parameters/enable_ppgtt 1 Default enable_ppgtt is 1, this issue goes away, decreasing priority.
No full PPGTT on pre-gen8.
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.