On 2018-07-18 06:46 PM, Christian König wrote: > Am 18.07.2018 um 18:44 schrieb Michel Dänzer: >> On 2018-07-18 06:17 PM, Michel Dänzer wrote: >>> On 2018-07-18 06:05 PM, Tom St Denis wrote: >>>> Hi Christian, >>>> >>>> This patch: >>>> >>>> [root at raven linux]# git bisect bad >>>> 90f362bdf0d0d06a126a5fd35b084436dd8250ad is the first bad commit >>>> commit 90f362bdf0d0d06a126a5fd35b084436dd8250ad >>>> Author: Christian König <christian.koenig at amd.com> >>>> Date:  Mon Jul 16 14:58:48 2018 +0200 >>>> >>>>     drm/amdgpu: change ring priority after pushing the job >>>> >>>>     Pushing a job can change the ring assignment of an entity. >>>> >>>>     Signed-off-by: Christian König <christian.koenig at amd.com> >>>>     Reviewed-by: Chunming Zhou <david1.zhou at amd.com> >>>> >>>> :040000 040000 9a09d3e9d055e4f5024019861c334ee9cc0bd11b >>>> 522540c31c3d1a4f4a6fbac75e985e9a7f7e93c9 M     drivers >>>> >>>> causes a KASAN while running piglit. It's reproduceable 100% of the >>>> time. The commit before this doesn't not cause a KASAN. >>> I got the attached KASAN use-after-free report while running piglit >>> today, could be the same? >> Sorry, that was for a different issue. Here's the KASAN report. > > No, problem. I've already know what's going wrong here. > > Accessing job after it is pushed to the scheduler is a bad idea, need to > cache the priority field locally. > > Give me an hour to get the kids into bed and I can fix it, Thanks, but no rush from my side, I won't be able to test it before tomorrow anyway (and it doesn't seem to cause any problems other than the KASAN report). -- Earthling Michel Dänzer | http://www.amd.com Libre software enthusiast | Mesa and X developer
