Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Christian König
Am 23.01.20 um 16:39 schrieb William Lewis: Was the change to true from the only use of intr in amdgpu_bo_sync_wait_resv intentional?  If so, would it not make sense to remove the argument from the function signature while the API is changing? That's indeed a copy & paste typo. Thanks for

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread William Lewis
Was the change to true from the only use of intr in amdgpu_bo_sync_wait_resv intentional?  If so, would it not make sense to remove the argument from the function signature while the API is changing? On 1/23/20 8:21 AM, Christian König wrote: > If provided we only sync to the BOs reservation >

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Tom St Denis
I've tested piglit, unigine-heaven, and video playback on my navi10/raven1 system.  All fine (as far as the kernel is concerned).  You can add my Tested-by: Tom St Denis Tom On 2020-01-23 9:35 a.m., Christian König wrote: That is fixed by patch #2 in this series. Patch #4 is then

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Christian König
That is fixed by patch #2 in this series. Patch #4 is then re-applying the faulty synchronization cleanup. Regards, Christian. Am 23.01.20 um 15:25 schrieb Tom St Denis: On the tip of drm-next (as of this morning) I was still getting [  983.891264] [drm:amdgpu_gem_va_ioctl [amdgpu]] *ERROR*

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Tom St Denis
On the tip of drm-next (as of this morning) I was still getting [  983.891264] [drm:amdgpu_gem_va_ioctl [amdgpu]] *ERROR* Couldn't update BO_VA (-2) type errors.  So I'm looking for that to stop.  At least LLVM was fixed so I can run a full run of piglit without gfx hangs. Tom On

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Christian König
Thanks, please give them a full round. Took me a week to figure out that we accidentally pass in the reservation object as NULL for cleared BOs. Thanks, Christian. Am 23.01.20 um 15:22 schrieb Tom St Denis: Just applied these now, trying them out will report back in ~20 mins. On 2020-01-23

Re: [PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Tom St Denis
Just applied these now, trying them out will report back in ~20 mins. On 2020-01-23 9:21 a.m., Christian König wrote: If provided we only sync to the BOs reservation object and no longer to the root PD. v2: update comment, cleanup amdgpu_bo_sync_wait_resv v3: use correct reservation object

[PATCH 4/4] drm/amdgpu: rework synchronization of VM updates v3

2020-01-23 Thread Christian König
If provided we only sync to the BOs reservation object and no longer to the root PD. v2: update comment, cleanup amdgpu_bo_sync_wait_resv v3: use correct reservation object while clearing Signed-off-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 37