On Tue, Oct 16, 2012 at 12:28 AM, Luís Picciochi Oliveira <[email protected]> wrote: > On 15/10/2012, Daniel Vetter <[email protected]> wrote: >> gem_cs_tlb is harmless, the testcase itself isn't that robust and >> fails an internal self-check once a while when other things have >> happened - it should work if you run it alone. The >> gem_set_tiling_vs_pwrite is more unsettling - does that reliably fail >> with the little backport and reliably work on a kernel without that >> backport? Might be we miss another workaround. >> >> I'll double-check these tests on my own i915g meanwhile. > > More test results: > - without any of the patches: > gem_set_tiling_vs_pwrite passes > > - with: > c9c4b6f6c283 drm/i915: fix swizzle detection for gen3 > gem_set_tiling_vs_pwrite FAILs > > - with: > 7dd490658627 drm/i915: Mark untiled BLT commands as fenced on gen2/3 > (the first patch on that commit + the fix proposed on this thread) > gem_set_tiling_vs_pwrite passes > 13 other tests fail > > - with: > same as above > + 15a13bbdffb0 drm/i915: clear fencing tracking state when retiring requests > gem_set_tiling_vs_pwrite passes > 13 other tests fail > > > Unfortunately, the c9c4b6f6c283 commit (that we can conclude is the > one making gem_set_tiling_vs_pwrite fail), is needed to fix the > corrupted images and text. > You can see the output from the running all tests with the other > patches applied at [1].
c07496fa61f4c5cb2addd1c57f6b should fix this testcase regression I hope. At least it fixed an intermediate regression with this testcase for me on my i915G. The issue is with bit17 swizzling, which c9c4b6f6c283 fixes, and which has been the reason behind image corruptions when swapping graphics buffers in&out papered over another, much older bug. If I'm right, can you please submit that patch for the stable team, too? Current userspace doesn't really use this path, but we (ab)use it a lot in the testcases to check consistency. Yours, Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
