Patch "drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change" has been added to the 7.2-stable tree
Subsystems (2)
[email protected]
AppliedThis is a note to let you know that I've just added the patch titled
drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change
to the 7.2-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
drm-xe-don-t-transfer-shrinker-pages-for-pinned-bos-on-purgeable-state-change.patch
and it can be found in the queue-7.2 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.
From ffbfd3f98f4393e91dffce3a46ccb7cf5329cf2f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= <[email protected]>
Date: Fri, 9 Oct 2026 09:00:53 +0200
Subject: drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
From: Thomas Hellström <[email protected]>
commit ffbfd3f98f4393e91dffce3a46ccb7cf5329cf2f upstream.
Pinned bos have their pages removed from the shrinker's shrinkable / purgeable counts on pin, and added back on unpin according to the bo's purgeable state at that time.
However, xe_bo_set_purgeable_shrinker() unconditionally transfers pages between the shrinkable and purgeable counts. If the purgeable state of a pinned bo changes, for example by userspace madvising DONTNEED on a VMA of a bo that is also pinned as a scanout buffer, this underflows one count and, after unpin, leaves the other one permanently inflated.
Only update the purgeable flag for pinned bos, and leave the page accounting to unpin.
--- drivers/gpu/drm/xe/xe_bo.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_bo.c +4 -1
@@ -938,6 +938,8 @@ static int xe_bo_move_notify(struct xe_b* * Transfers pages between shrinkable and purgeable buckets when the BO * purgeable state changes. Called automatically from xe_bo_set_purgeable_state(). +* For pinned BOs, whose pages are not accounted, only the purgeable flag is +* updated, so that the correct bucket is used when the BO is unpinned. */ static void xe_bo_set_purgeable_shrinker (struct xe_bo *bo,enum xe_madv_purgeable_state new_state)
@@ -954,7 +956,8 @@ static void xe_bo_set_purgeable_shrinkerreturn ; xe_tt = container_of(tt, struct xe_ttm_tt, ttm); - tt_pages = tt->num_pages; +/* Pinned bos are not accounted. Only update the purgeable flag. */ + tt_pages =xe_bo_is_pinned (bo) ?0 : tt->num_pages;if (!xe_tt->purgeable && new_state ==XE_MADV_PURGEABLE_DONTNEED ) { xe_tt->purgeable = true;
Patches currently in stable-queue which might be from [email protected] are
queue-7.2/drm-xe-bump-lru-for-external-objects-on-exec-and-rebind.patch queue-7.2/drm-xe-don-t-transfer-shrinker-pages-for-pinned-bos-on-purgeable-state-change.patch
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] drm/xe: shrinker page accounting corruption in xe_bo_set_purgeable_shrinker() --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] drm/xe: shrinker page accounting corruption in xe_bo_set_purgeable_shrinker() --
commit 38fe9b0b7cfb46e5ebc7efb56c8bdf33363c1017
Author: [email protected] <[email protected]>
Patch "drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change" has been added to the 7.2-stable tree
This commit updates xe_bo_set_purgeable_shrinker() in drivers/gpu/drm/xe/xe_bo.c to set tt_pages to 0 when xe_bo_is_pinned(bo) is true, updating only xe_tt->purgeable without transferring pages between shrinker buckets. However, 7.2-stable does not include upstream commit 6a117a2bb7ca65 ("drm/xe: Fix shrinker accounting double-subtraction on nested external pins"), which changed scanout framebuffer pinning to use xe_bo_pin_external().
Quoted 14 lines
drivers/gpu/drm/xe/xe_bo.c +2
-0@@ -838,6 +838,8 @@ static int xe_bo_move_notify(struct xe_bo *bo,* * Transfers pages between shrinkable and purgeable buckets when the BO * purgeable state changes. Called automatically from xe_bo_set_purgeable_state(). +* For pinned BOs, whose pages are not accounted, only the purgeable flag is +* updated, so that the correct bucket is used when the BO is unpinned. */ static void xe_bo_set_purgeable_shrinker (struct xe_bo *bo,enum xe_madv_purgeable_state new_state)
drivers/gpu/drm/xe/xe_bo.c static void xe_bo_set_purgeable_shrinker(struct xe_bo *bo, · line 863
Quoted 9 lines
drivers/gpu/drm/xe/xe_bo.c +2
-1@@ -854,7 +856,8 @@ static void xe_bo_set_purgeable_shrinker(struct xe_bo *bo,return ; xe_tt = container_of(tt, struct xe_ttm_tt, ttm); - tt_pages = tt->num_pages; +/* Pinned bos are not accounted. Only update the purgeable flag. */ + tt_pages =xe_bo_is_pinned (bo) ?0 : tt->num_pages;if (!xe_tt->purgeable && new_state ==XE_MADV_PURGEABLE_DONTNEED ) { xe_tt->purgeable = true;
[Severity: High]
Does setting tt_pages = 0 when xe_bo_is_pinned(bo) is true in xe_bo_set_purgeable_shrinker() in drivers/gpu/drm/xe/xe_bo.c introduce a shrinker page accounting regression on 7.2-stable when a scanout framebuffer buffer object changes purgeable state while pinned?
In 7.2-stable (which lacks upstream commit 6a117a2bb7ca65, "drm/xe: Fix shrinker accounting double-subtraction on nested external pins"), scanout framebuffers are pinned and unpinned directly via ttm_bo_pin() and ttm_bo_unpin() without calling xe_ttm_tt_account_subtract() or xe_ttm_tt_account_add():
drivers/gpu/drm/xe/display/xe_fb_pin.c:__xe_pin_fb_vma() {
...
if (IS_DGFX(xe))
ret = xe_bo_migrate(bo, XE_PL_VRAM0, NULL, &exec);
else
ret = xe_bo_validate(bo, NULL, true, &exec);
drm_exec_retry_on_contention(&exec);
xe_validation_retry_on_oom(&ctx, &ret);
if (!ret)
ttm_bo_pin(&bo->ttm);
...
}drivers/gpu/drm/xe/display/xe_fb_pin.c:__xe_unpin_fb_vma() {
...
ttm_bo_reserve(&vma->bo->ttm, false, false, NULL);
ttm_bo_unpin(&vma->bo->ttm);
ttm_bo_unreserve(&vma->bo->ttm);
...
}When a populated user buffer object has its pages accounted in shrinker->shrinkable_pages via xe_ttm_tt_populate() -> xe_ttm_tt_account_add() and is then pinned for scanout via __xe_pin_fb_vma() -> ttm_bo_pin(&bo->ttm), its pages remain accounted in shrinker->shrinkable_pages while bo->ttm.pin_count > 0.
If userspace then updates the purgeable state via madvise_purgeable() -> xe_bo_willneed_put_locked() -> xe_bo_set_purgeable_state() -> xe_bo_set_purgeable_shrinker(), xe_bo_is_pinned(bo) is true, so xe_bo_set_purgeable_shrinker() sets tt_pages = 0 and updates xe_tt->purgeable = true without moving tt->num_pages from shrinker->shrinkable_pages to shrinker->purgeable_pages.
After __xe_unpin_fb_vma() -> ttm_bo_unpin(&vma->bo->ttm) drops bo->ttm.pin_count to 0 without re-adding pages, subsequent unpopulation in xe_ttm_tt_unpopulate(), purging in xe_bo_shrink_purge(), or another state change while unpinned in xe_bo_set_purgeable_shrinker() subtracts from the bucket selected by xe_tt->purgeable:
drivers/gpu/drm/xe/xe_bo.c:xe_ttm_tt_account_subtract() {
struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
if (xe_tt->purgeable)
xe_shrinker_mod_pages(xe->mem.shrinker, 0, -(long)tt->num_pages);
else
xe_shrinker_mod_pages(xe->mem.shrinker, -(long)tt->num_pages, 0);
}Because xe_tt->purgeable was flipped while the pages stayed in shrinker->shrinkable_pages, wouldn't this subtract tt->num_pages from shrinker->purgeable_pages instead, underflowing shrinker->purgeable_pages to a negative value, leaving shrinker->shrinkable_pages permanently inflated, and causing unsigned long wrap-around in xe_shrinker_count() in drivers/gpu/drm/xe/xe_shrinker.c:
drivers/gpu/drm/xe/xe_shrinker.c:xe_shrinker_count() {
...
if (can_backup)
num_pages = min_t(unsigned long, num_pages, shrinker->shrinkable_pages);
else
num_pages = 0;
num_pages += shrinker->purgeable_pages;
...
}when shrinker->purgeable_pages is negative?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/2026102151-compound-fondness-145b@gregkh?part=1