[PATCH 0/8] KFD SDMA support for GFX7 and GFX8 v2

Felix Kuehling felix.kuehling at amd.com
Thu Nov 2 15:22:39 UTC 2017


I've given up on targeting a specific kernel release. This whole
upstreaming process is way more work than I expected, and is taking much
longer than I thought when I started. If I'm making this harder for
myself than necessary, please let me know. But I'm quite sure that this
is just the cost of turning our messy history into something sensible
and reviewable.

For example, most of the SDMA work was originally done after adding dGPU
support. Also, much of it was done on GFX8 first and later ported back
to GFX7. However, to make more sense for upstreaming and to avoid
duplicate reviews, I reordered the history to fix GFX7 first and
introduce GFX8 support in its final, fixed form.

In that sense, SDMA support is a recursive mini-instance of what I'm
doing with dGPU support. Many of the fixes and cleanups I have already
upstreamed were originally done on top of (or rather in the middle of)
dGPU support. I had to disentangle all of that to upstream the changes
that are applicable to CZ and KV, so that I can eventually upstream dGPU
support in its final form, skipping all the broken intermediate steps,
with all the hind-sight we have gained in the mean time.

Looking at the work that's left to do, and the likely feedback I'll get
when I get to the dGPU memory management parts, this is going to take
months, not weeks, until most of KFD is upstream. Especially if I also
want to rebase/upstream our libhsakmt and open-source KFDTest. Maybe I
can delegate some of this ...


On 2017-11-02 11:04 AM, Oded Gabbay wrote:
> Thanks!
> Taken to -next, but it will only be for 4.16 I'm afraid
> On Thu, Nov 2, 2017 at 1:21 AM, Felix Kuehling <Felix.Kuehling at amd.com> wrote:
>> This patch series fixes SDMA user mode queue support for GFX7 and adds
>> support for GFX8.
>> v2: Rebased. radeon_kfd.c doesn't exist any more.
>> Felix Kuehling (5):
>>   drm/amdgpu: Correct SDMA load/unload sequence on HWS disabled mode
>>   drm/amdkfd: Fix SDMA oversubsription handling
>>   drm/amd: Update kgd_kfd interface for resuming SDMA queues
>>   drm/amdgpu: Add support for resuming SDMA queues w/o HWS
>>   drm/amdkfd: Use ASIC-specific SDMA MQD type
>> Philip Cox (2):
>>   drm/amdgpu: Implement amdgpu SDMA functions for VI
>>   drm/amdkfd: Implement amdkfd SDMA functions for VI
>> shaoyunl (1):
>>   drm/amdkfd: Correct SDMA ring buffer size
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c  |  73 ++++++++++----
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c  | 100 +++++++++++++++----
>>  drivers/gpu/drm/amd/amdgpu/vid.h                   |   2 +
>>  drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c   |  21 ++--
>>  drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c    | 108 ++++++++++++++++++++-
>>  drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |   2 -
>>  .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c |  18 ++++
>>  drivers/gpu/drm/amd/include/kgd_kfd_interface.h    |   3 +-
>>  drivers/gpu/drm/amd/include/vi_structs.h           |   2 +
>>  9 files changed, 277 insertions(+), 52 deletions(-)
>> --
>> 2.7.4

More information about the amd-gfx mailing list