linux/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c, branch v4.19

drm/amdkfd: Change the control stack MTYPE from UC to NC on GFX9

2018-09-20T15:25:17Z

CWSR fails on Raven if the control stack is MTYPE_UC, which is used for regular GART mappings. As a workaround we map it using MTYPE_NC. The MEC firmware expects the control stack at one page offset from the start of the MQD so it is part of the MQD allocation on GFXv9. AMDGPU added a memory allocation flag just for this purpose. Acked-by: Alex Deucher Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Felix Kuehling Signed-off-by: Alex Deucher

drm/amdgpu: Add kfd2kgd.set_compute_idle interface

2018-07-16T23:10:36Z

This allows automatic switching to the compute power profile depending on compute activity. Signed-off-by: Felix Kuehling Reviewed-by: Eric Huang Reviewed-by: Alex Deucher Signed-off-by: Oded Gabbay

drm/amdgpu: Enable the gpu reset from KFD

2018-07-12T02:32:57Z

Hook up the gpu_recover callback from KFD to amdgpu to enable handling of GPU hangs detected by KFD. Signed-off-by: Shaoyun Liu Reviewed-by: Felix Kuehling Signed-off-by: Felix Kuehling Acked-by: Christian König Signed-off-by: Oded Gabbay

drm/amdgpu: Call KFD reset handlers during GPU reset

2018-07-12T02:32:55Z

Signed-off-by: Shaoyun Liu Reviewed-by: Felix Kuehling Signed-off-by: Felix Kuehling Acked-by: Christian König Signed-off-by: Oded Gabbay

drm/amdgpu: allocate gart memory when it's required (v3)

2018-07-10T19:15:58Z

Instead of calling gart address space on every bo pin, allocates it on demand v2: fix error handling v3: drop the change on amdgpu_amdkfd_gpuvm.c, not needed. Signed-off-by: Junwei Zhang Acked-by: Felix Kuehling Reviewed-by: Christian König Signed-off-by: Alex Deucher

drm/amdgpu: separate gpu address from bo pin

2018-07-10T19:15:39Z

It could be got by amdgpu_bo_gpu_offset() if need Signed-off-by: Junwei Zhang Reviewed-by: Michel Dänzer Reviewed-by: Christian König Signed-off-by: Alex Deucher

Merge branch 'drm-next-4.18' of git://people.freedesktop.org/~agd5f/linux into drm-next

2018-06-15T01:32:29Z

Fixes for 4.18. Highlights: - Fixes for gfxoff on Raven - Remove an ATPX quirk now that the root cause is fixed - Runtime PM fixes - Vega20 register header update - Wattman fixes - Misc bug fixes Signed-off-by: Dave Airlie Link: https://patchwork.freedesktop.org/patch/msgid/20180614141428.2909-1-alexander.deucher@amd.com

drm/amdgpu: Fix NULL pointer when load kfd driver with PP block is disabled

2018-05-31T19:08:54Z

When PP block is disabled, return a fix value(100M) for mclk and sclk on bare-metal mode. This will cover the emulation mode as well. Signed-off-by: Shaoyun Liu Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher

drm/amdgpu: Use dev_info() to report amdkfd is not supported for this ASIC

2018-05-24T21:07:14Z

This is an important message, so it should be visible to users without having to enable extra debugging. Signed-off-by: Tom Stellard Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay

drm/amdgpu: conditionally compile amdgpu's amdkfd files

2018-05-18T19:18:16Z

In case CONFIG_HSA_AMD is not chosen, there is no need to compile amdkfd files that reside inside amdgpu dirver. In addition, because amdkfd depends on x86_64 architecture and amdgpu is not, compiling amdkfd files under i386 architecture can cause compiler errors and warnings. This patch modifies amdgpu's makefile to build amdkfd files only if CONFIG_HSA_AMD is chosen. The only file to be compiled unconditionally is amdgpu_amdkfd.c There are stub functions that are compiled only if amdkfd is not compiled. In that case, calls from amdgpu driver proper will go to those functions instead of the real functions. v2: instead of using function pointers, use stub functions v3: initialize kgd2kfd to NULL in case amdkfd is not compiled Reviewed-by: Felix Kuehling Signed-off-by: Oded Gabbay