Skip to contentSkip to conversation

Patch "drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change" has been added to the 7.2-stable tree

PATCH [email protected] 2 participants 2 messages Oct 21

Subsystems (2)
lore
Share & tools
lore

[email protected]

author Applied
#0 lore raw

This 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.

Fixes 536a2ead3a8f ("drm/xe/bo: Add purgeable shrinker state helpers")
Cc Arvind Yadav <[email protected]>
Cc Matthew Brost <[email protected]>
Cc Himal Prasad Ghimiray <[email protected]>
Cc Rodrigo Vivi <[email protected]>
Cc Matthew Auld <[email protected]>
Cc <[email protected]> # v7.1+
Assisted-by LLM
(cherry picked from commit afe6bc012034123fd838328668e609bf57e46bd3)
---
 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_shrinker
 		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;

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

#1 lore raw
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() --

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
index 7ed76349075fd..2d8283f6010a6 100644
@@ -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