Created attachment 107648 [details] dmesg ==System Environment== -------------------------- Regression: not sure, only reproduce on -nightly kernel. ==kernel== -------------------------- drm-intel-nightly/ea4bec8e96ea8b33b49a7892c1c7f20041a56da6 ==Bug detailed description== It fails on BYT with drm-intel-nightly kernel. it pass on drm-intel-queued, drm-intel-fixes, drm-fixes and drm-next kernel. igt/gem_persistent_relocs/forked-interruptible-thrashing and igt/gem_reloc_vs_gpu/forked-interruptible-faulting-reloc-thrashing also fail. output: IGT-Version: 1.8-g7f82289 (x86_64) (Linux: 3.17.0_drm-intel-nightly_ea4bec_20141009+ x86_64) Test assertion failure function do_test, file gem_persistent_relocs.c:260: Failed assertion: test == 0xdeadbeef Last errno: 4, Interrupted system call mismatch in buffer 0: 0x00000000 instead of 0xdeadbeef at offset 256 child 3 failed with exit status 99 Subtest forked-interruptible-faulting-reloc-thrashing: FAIL (451.180s) ==Reproduce steps== ---------------------------- 1. ./gem_persistent_relocs --run-subtest forked-interruptible-faulting-reloc-thrashing
It should be full-ppgtt related i.e. setting i915.enable_ppgtt=1 should make it pass again.
(In reply to Chris Wilson from comment #1) > It should be full-ppgtt related i.e. setting i915.enable_ppgtt=1 should make > it pass again. Yes, add i915.enable_ppgtt=1, it works well.
We decided against full PPGTT on pre-gen8.
At least, we can fix the test cases, making it do some platform checking before real testing. Then the case should return "NOT SUPPORT"/"NOT RUN", but not "FAIL". I think just leaving a "FAIL" result on pre-gen8 platform is not perfect. We can take it as a IGT issue, but I didn't find that component here.
It's an xfail. The test shouldn't be skipped, but we do expect it to fail until we can devise a w/a
Yi, I think the test is doing the right thing here like Chris says. It will fail as expected with full ppgtt (if/until we find a workaround for it) and will pass as expected in the default config, which is what we care about. Unless specifically requested by someone with a new full ppgtt workaround, you can skip running any full ppgtt enabled tests on pre-gen8.
Oops, I missed Chris's comments here previously. And thanks Chris and Jesse 's explanation. I see, so close the bug since everything seems good enough.
Closing old verified.
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.