On 12/1/2025 5:51 AM, Dave Airlie wrote:
> On Mon, 1 Dec 2025 at 11:04, Sinyuk, Konstantin
> <[email protected]> wrote:
>>
>> Hi Dave, Sima,
>>
>> We are excited to continue the upstreaming of the accel/habanalabs driver.
>> While maintaining the momentum with Gaudi2 fixes, we now introduce Gaudi3
>> support,
>> covering the development work done during the 2021-2024 period.
>>
>> This tag contains habanalabs driver changes for v6.19, continuing the
>> previous
>> upstream work from tags/drm-habanalabs-next-2025-05-25. It includes
>> improvements
>> in debug and visibility, general code cleanups, and the finalization of
>> Gaudi3
>> support, closing remaining gaps relative to Gaudi2. This is backed by
>> foundational features such as the MMU v3 architecture, ODP (On-Demand
>> Paging),
>> and enhanced security infrastructure.
>
> As per https://www.kernel.org/doc/html/latest/gpu/introduction.html
>
> "All feature work must be in the linux-next tree by the -rc6 release
> of the current release cycle, otherwise they must be postponed and
> can’t reach the next merge window. All patches must have landed in the
> drm-next tree by latest -rc7, but if your branch is not in linux-next
> then this must have happened by -rc6 already."
>
> This is new hardware support so I was seeing how nice I was feeling,
> but there were a few messy bits on initial review, so probably will
> have to wait for 6.20.
>
>> .clang-format | 179 +-
>
> I've no idea who owns .clang-format, but changes to it should not be
> happening via here, it looks like it regresses and changes a bunch of
> stuff unrelated to this.
>
> diff --git a/drivers/accel/habanalabs/Makefile
> b/drivers/accel/habanalabs/Makefile
> index 98510cdd5066..58c51f03858d 100644
> --- a/drivers/accel/habanalabs/Makefile
> +++ b/drivers/accel/habanalabs/Makefile
> @@ -3,11 +3,16 @@
> # Makefile for HabanaLabs AI accelerators driver
> index 98510cdd5066..58c51f03858d 100644
> --- a/drivers/accel/habanalabs/Makefile
> +++ b/drivers/accel/habanalabs/Makefile
> @@ -3,11 +3,16 @@
> # Makefile for HabanaLabs AI accelerators driver
> #
>
> +CONFIG_DRM_ACCEL_HABANALABS := m
>
> that seems suspect at best?
>
> +ifdef OFED_PATH
> +KBUILD_EXTRA_SYMBOLS := $(OFED_PATH)/Module.symvers
> +endif
> ???
>
> timeout = flags & HL_CS_FLAGS_CUSTOM_TIMEOUT
> - ? secs_to_jiffies(args->in.timeout)
> + ? msecs_to_jiffies(args->in.timeout * 1000)
> : hpriv->hdev->timeout_jiffies;
>
> There appears to be a bunch of reverts of upstream developed patches,
>
> 78cf56f8832a932ade20b8340a029ace14ac0e98
> Author: Easwar Hariharan <[email protected]>
> Date: Tue Feb 25 20:17:17 2025 +0000
>
> accel/habanalabs: convert timeouts to secs_to_jiffies()
>
> Please re-review this and make sure you aren't rewinding upstream work
> that never made it to your internal development tree.
>
> Dave.
Hi Dave,
Thank you for the detailed prompt review and for catching these issues early.
I am formally withdrawing this pull request to carefully rework the changes
and retarget for v6.20.
Corrective actions:
1. Remove all out-of-tree artifacts:
- Revert .clang-format changes
- Remove CONFIG_DRM_ACCEL_HABANALABS := m from Makefile
- Remove OFED_PATH dependencies
2. Properly rebase the Gaudi3 internal work:
- Ensure the ~6000+ internal commits are rebased in a way that preserves
upstream history and avoids undoing v6.13+ changes
- Make sure no upstream fixes are silently reverted by the Gaudi3 series
3. Preserve all upstream contributions, specifically:
- secs_to_jiffies() conversions (78cf56f8832a, c2cd5986ebcd)
- Any other v6.13+ changes that touched habanalabs
4. Audit all files for similar unintended reversions
5. Clean the commit history of internal development artifacts
6. Test thoroughly against current drm-next
7. Resubmit for v6.20 with proper timing (well before -rc6)
The Gaudi3 support represents ~300K lines, but that's no excuse for not
properly synchronizing with upstream. We'll get this right for v6.20.
Thank you for your patience.
Thanks,
Konstantin.
