[PATCH 00/24] KFD fixes, cleanups, features to make the runtime happy

Wed Aug 16 07:51:12 UTC 2017

> For testing KV with current user mode stack, please use amdgpu. I
> don't expect this to work with radeon and I'm not planning to spend
> any effort on making radeon work with a current user mode stack.
In this case any objection to just rip out radeon support?

Christian.

Am 16.08.2017 um 04:59 schrieb Felix Kuehling:
> I merged the last two patch series and made a number of updates as
> discussed. I also rebased it on Alex's drm-next-4.14-wip branch. All
> patches applied cleanly without having to resolve any conflicts.
>
> I'm also publishing my branch with this patch series on GitHub to make
> it easier to test and apply the changes. This KFD branch requires the
> corresponding Thunk branch from GitHub, which uses a different ioctl
> ABI from the last ROCm release. The rest of the user mode stack
> should be OK on top of that Thunk.
>
> https://github.com/RadeonOpenCompute/ROCK-Kernel-Driver/tree/fkxamd/drm-next-wip
> https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/fxkamd/drm-next-wip
>
> On CZ, hsaconformance passes. OpenCL tests can be run with the current
> ROCm OpenCL stack. I ran the SHOC benchmark as an example.
>
> On KV, the current ROCm OpenCL driver doesn't work. This is a
> limitation of the OpenCL driver. I'm trying to find out how hard it
> would be to change that. In the mean time, some hsaconformance tests
> can be run on KV. It passes most tests up to
> 162:code_recursive_kernel_function, where it hangs.
>
> For testing KV with current user mode stack, please use amdgpu. I
> don't expect this to work with radeon and I'm not planning to spend
> any effort on making radeon work with a current user mode stack.
>
> Dropped patches:
>    * drm/amdkfd: Fix double Mutex lock order
>
> Added patches:
>    * drm/radeon: Return dword offsets of address watch registers
>    * drm/amdgpu: Program SH_STATIC_MEM_CONFIG globally, not per-VMID
>
> Reordered patch sequence:
>    I moved "drm/amdgpu: Remove hard-coded assumptions about compute pipes"
>    just after "drm/amdkfd: Fix allocated_queues bitmap initialization"
>    because these two patches fix related problems (how queues are shared
>    between KFD and KGD).
>
> Updated patches:
>    * drm/amdkfd: Clean up KFD style errors and warnings v2
>    * drm/amdkfd: Fix goto usage v2
>    * drm/amdkfd: Handle remaining BUG_ONs more gracefully v2
>    * drm/amdkfd: Add more error printing to help bringup v2
>    * drm/amdgpu: Add kgd/kfd interface to support scratch memory v2
>    * drm/amdkfd: Adding new IOCTL for scratch memory v2
>    * drm/amdgpu: Add kgd kfd interface get_tile_config() v2
>    * drm/amdkfd: Implement image tiling mode support v2
>
> Patches that still need a Reviewed-by or Acked-by:
>    * drm/radeon: Return dword offsets of address watch registers
>    * drm/amdkfd: Remove bogus divide-by-sizeof(uint32_t)
>    * drm/amdkfd: Clean up KFD style errors and warnings v2
>    * drm/amdkfd: Handle remaining BUG_ONs more gracefully v2
>    * drm/amdkfd: Add more error printing to help bringup v2
>    * drm/amdgpu: Program SH_STATIC_MEM_CONFIG globally, not per-VMID
>    * drm/amdgpu: Add kgd/kfd interface to support scratch memory v2
>    * drm/amdkfd: Adding new IOCTL for scratch memory v2
>
> ***
>
> Felix Kuehling (13):
>    drm/amdkfd: Fix typo in dbgdev_wave_reset_wavefronts
>    drm/radeon: Return dword offsets of address watch registers
>    drm/amdkfd: Remove bogus divide-by-sizeof(uint32_t)
>    drm/amdkfd: Fix allocated_queues bitmap initialization
>    drm/amdgpu: Remove hard-coded assumptions about compute pipes
>    drm/amdkfd: Remove BUG_ONs for NULL pointer arguments
>    drm/amdkfd: Fix doorbell initialization and finalization
>    drm/amdkfd: Allocate gtt_sa_bitmap in long units
>    drm/amdkfd: Handle remaining BUG_ONs more gracefully v2
>    drm/amdkfd: Update PM4 packet headers
>    drm/amdgpu: Disable GFX PG on CZ
>    drm/amd: Update MEC HQD loading code for KFD
>    drm/amdgpu: Program SH_STATIC_MEM_CONFIG globally, not per-VMID
>
> Jay Cornwall (1):
>    drm/amdkfd: Clamp EOP queue size correctly on Gfx8
>
> Kent Russell (5):
>    drm/amdkfd: Clean up KFD style errors and warnings v2
>    drm/amdkfd: Consolidate and clean up log commands
>    drm/amdkfd: Change x==NULL/false references to !x
>    drm/amdkfd: Fix goto usage v2
>    drm/amdkfd: Remove usage of alloc(sizeof(struct...
>
> Moses Reuben (2):
>    drm/amdgpu: Add kgd/kfd interface to support scratch memory v2
>    drm/amdkfd: Adding new IOCTL for scratch memory v2
>
> Yong Zhao (3):
>    drm/amdkfd: Add more error printing to help bringup v2
>    drm/amdgpu: Add kgd kfd interface get_tile_config() v2
>    drm/amdkfd: Implement image tiling mode support v2
>
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c         |   4 +-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h         |  16 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c  | 189 ++++++++++--
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c  | 227 ++++++++++++--
>   drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c              |   2 +-
>   drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c              |   3 +-
>   drivers/gpu/drm/amd/amdgpu/vi.c                    |   3 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_chardev.c           | 183 ++++++++----
>   drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c            | 102 +++----
>   drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c            |  21 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h            |  27 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_device.c            | 123 ++++----
>   .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c  | 316 ++++++++------------
>   .../drm/amd/amdkfd/kfd_device_queue_manager_cik.c  |   8 +-
>   .../drm/amd/amdkfd/kfd_device_queue_manager_vi.c   |   8 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c          |  40 +--
>   drivers/gpu/drm/amd/amdkfd/kfd_events.c            |  33 +--
>   drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c       |   2 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c         |   2 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c      |  63 ++--
>   drivers/gpu/drm/amd/amdkfd/kfd_module.c            |  10 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h       |   3 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c   |  62 ++--
>   drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c    |  46 +--
>   drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c    | 294 ++++++------------
>   drivers/gpu/drm/amd/amdkfd/kfd_pasid.c             |   7 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers.h       | 330 +++------------------
>   drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_vi.h    | 140 ++++++++-
>   drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |  32 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c           |  25 +-
>   .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c |  71 ++---
>   drivers/gpu/drm/amd/amdkfd/kfd_queue.c             |  12 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_topology.c          |  46 +--
>   drivers/gpu/drm/amd/include/kgd_kfd_interface.h    |  30 +-
>   drivers/gpu/drm/radeon/radeon_kfd.c                |  15 +-
>   include/uapi/linux/kfd_ioctl.h                     |  37 ++-
>   36 files changed, 1280 insertions(+), 1252 deletions(-)
>