drm/xe: Fix suspend handling and allow pinned external VRAM BOs #235
Conversation
CI Bug Log - changes from xkb-6.14-189-92fbf0b89afa05508181d2ca1503e4ea096c5382_BAT -> xkb-pr-6.14-235-403-f578951d7f4920e8ab139159bee8363aee9faed8_BATSummarySUCCESS No regressions found. Participating hosts (5 -> 5)No changes in participating hosts Known issuesHere are the changes found in xkb-pr-6.14-235-403-f578951d7f4920e8ab139159bee8363aee9faed8_BAT that come from known issues: IGT changesIssues hit
Build changes
igt-xkb-82-4e12718e2: 4e12718e2 |
ayazahma
left a comment
ayazahma
left a comment
There was a problem hiding this comment.
why these are in sriov ?
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= thomas.hellstrom@linux.intel.com
Date: Thu, 18 Sep 2025 11:22:06 +0200
Subject: drm/xe: Pre-allocate system memory for pinned external bos in
the pm notfier
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
does patches was taken from linux-next or downloaded from link
can you please suggest how I can forward
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= thomas.hellstrom@linux.intel.com Date: Thu, 18 Sep 2025 11:22:06 +0200 Subject: drm/xe: Pre-allocate system memory for pinned external bos in the pm notfier MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit
this changes I have taken from https://gitlab.freedesktop.org/drm/tip.git |
we have request for I have raised single commit for each series |
| Tested-by: Matthew Brost <matthew.brost@intel.com> | ||
| Reviewed-by: Matthew Brost <matthew.brost@intel.com> | ||
| Link: https://patch.msgid.link/20251217093441.5073-1-thomas.hellstrom@linux.intel.com | ||
| (cherry picked from commit 425fe550fb513b567bd6d01f397d274092a9c274 linux-next) |
There was a problem hiding this comment.
recheck placement, do we have multiple sign-off of same person?
There was a problem hiding this comment.
commit it self has twice
from tip:-
commit fe3ccd24138fd391ae8e32289d492c85f67770fc
Author: Thomas Hellström thomas.hellstrom@linux.intel.com
Date: Wed Dec 17 10:34:41 2025 +0100
drm/xe: Drop preempt-fences when destroying imported dma-bufs.
When imported dma-bufs are destroyed, TTM is not fully
individualizing the dma-resv, but it *is* copying the fences that
need to be waited for before declaring idle. So in the case where
the bo->resv != bo->_resv we can still drop the preempt-fences, but
make sure we do that on bo->_resv which contains the fence-pointer
copy.
In the case where the copying fails, bo->_resv will typically not
contain any fences pointers at all, so there will be nothing to
drop. In that case, TTM would have ensured all fences that would
have been copied are signaled, including any remaining preempt
fences.
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Fixes: fa0af721bd1f ("drm/ttm: test private resv obj on release/destroy")
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: <stable@vger.kernel.org> # v6.16+
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Tested-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20251217093441.5073-1-thomas.hellstrom@linux.intel.com
(cherry picked from commit 425fe550fb513b567bd6d01f397d274092a9c274)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
| @@ -0,0 +1,50 @@ | |||
| From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 | |||
| From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com> | |||
There was a problem hiding this comment.
check this line, why UTF?
There was a problem hiding this comment.
from drm tip itself it is commits same, it is not my local change
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= thomas.hellstrom@linux.intel.com
Date: Thu, 18 Sep 2025 11:22:07 +0200
Fix double copying of pinned kernel BOs on suspend and copy pinned external BOs in the correct pass. Pre-allocate system memory for pinned external BOs in the PM notifier. Allow pinning of P2P dma-bufs in VRAM with proper TTM cgroup accounting. Signed-off-by: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
Test fences in the private dma_resv object instead of checking a pointer that may reference a shared dma_resv. This only affects imported BOs with an SG table, since their dma_resv pointer is not replaced on release. Signed-off-by: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
When destroying imported dma-bufs, drop preempt fences from the private
dma-resv object rather than a potentially shared one.
TTM copies the fences needed for idle detection into the private resv,
so preempt fences must be removed from bo->_resv. If the copy fails,
no fences are present to drop and TTM guarantees they are already
signaled.
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Fixes: fa0af721bd1f ("drm/ttm: test private resv obj on release/destroy")
Signed-off-by: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
4 participants
Fix double copying of pinned kernel BOs on suspend and copy pinned
external BOs in the correct pass.
Pre-allocate system memory for pinned external BOs in the PM notifier.
Allow pinning of P2P dma-bufs in VRAM with proper TTM cgroup accounting.
Signed-off-by: Bommu Krishnaiah krishnaiah.bommu@intel.com