From: Matthew Brost <matthew.brost@intel.com>
To: intel-xe@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Subject: [PATCH v2] drm/xe: Do not blindly copy system CCS pages when shrinking
Date: Thu, 30 Apr 2026 18:19:07 -0700 [thread overview]
Message-ID: <20260501011907.2331654-1-matthew.brost@intel.com> (raw)
If the PAT index for a BO is configured without compression, copying CCS
pages during BO shrinking provides no value. Skip CCS copies for BOs that
have never been configured with compressed PAT indexes.
Once a BO is configured with a compressed PAT index, this becomes a
one-way transition that enables CCS copies. This heuristic is imprecise,
but simple and functional.
Assisted-by: Claude:claude-sonnet-4.6 # kernel doc
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
---
When running 10x tabs webgl aquarium, it was observed a ratio of ~6.5 to
1 of BOs that can possible enable compression (xe_bo_needs_ccs_pages
returns true) actually enable compression in the PAT index and overall
CCS copies are reduced by roughly 1/2 when the shrinker is triggered.
v2:
- Only skip CCS copy when moving to system placement (sashiko)
- Set ccs_used upon rebind (sashiko)
- Add helpers for ccs_used access + document locking
- Set ccs_used in kunit tests (CI)
---
drivers/gpu/drm/xe/tests/xe_bo.c | 2 ++
drivers/gpu/drm/xe/xe_bo.c | 3 +++
drivers/gpu/drm/xe/xe_bo.h | 35 ++++++++++++++++++++++++++++++++
drivers/gpu/drm/xe/xe_bo_types.h | 6 ++++++
drivers/gpu/drm/xe/xe_vm.c | 8 ++++++++
5 files changed, 54 insertions(+)
diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c b/drivers/gpu/drm/xe/tests/xe_bo.c
index 49c95ed67d7e..1686e735a2c7 100644
--- a/drivers/gpu/drm/xe/tests/xe_bo.c
+++ b/drivers/gpu/drm/xe/tests/xe_bo.c
@@ -149,6 +149,8 @@ static void ccs_test_run_tile(struct xe_device *xe, struct xe_tile *tile,
xe_bo_lock(bo, false);
+ xe_bo_set_ccs_used(bo);
+
kunit_info(test, "Verifying that CCS data is cleared on creation.\n");
ret = ccs_test_migrate(tile, bo, false, 0ULL, 0xdeadbeefdeadbeefULL,
test, exec);
diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 5ce60d161e09..765b5d345a7e 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -1105,6 +1105,9 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
flags |= XE_MIGRATE_CLEAR_FLAG_CCS_DATA;
fence = xe_migrate_clear(migrate, bo, new_mem, flags);
+ } else if (handle_system_ccs && new_mem->mem_type == XE_PL_SYSTEM &&
+ !xe_bo_is_ccs_used(bo)) {
+ fence = dma_fence_get_stub();
} else {
fence = xe_migrate_copy(migrate, bo, bo, old_mem, new_mem,
handle_system_ccs);
diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index 68dea7d25a6b..101dc7a36f9a 100644
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -514,4 +514,39 @@ static inline bool xe_bo_is_mem_type(struct xe_bo *bo, u32 mem_type)
xe_bo_assert_held(bo);
return bo->ttm.resource->mem_type == mem_type;
}
+
+/**
+ * xe_bo_set_ccs_used() - Mark the CCS region of a BO as used
+ * @bo: The &xe_bo whose CCS region is being marked as used
+ *
+ * Mark the CCS (Compression Control Surface) metadata region of @bo as in use.
+ * This is set when the BO is mapped with a PAT index that has compression
+ * enabled, indicating that the CCS region holds valid compression state and
+ * must be handled appropriately during migration (e.g. copied or cleared rather
+ * than skipped).
+ *
+ * Caller must hold the BO reservation lock.
+ */
+static inline void xe_bo_set_ccs_used(struct xe_bo *bo)
+{
+ xe_bo_assert_held(bo);
+ bo->ccs_used = true;
+}
+
+/**
+ * xe_bo_is_ccs_used() - Check whether the CCS region of a BO is in use
+ * @bo: The &xe_bo to check
+ *
+ * Return: true if the CCS (Compression Control Surface) metadata region of
+ * @bo has been marked as used (i.e. the BO was mapped with compression
+ * enabled), false otherwise.
+ *
+ * Caller must hold the BO reservation lock.
+ */
+static inline bool xe_bo_is_ccs_used(struct xe_bo *bo)
+{
+ xe_bo_assert_held(bo);
+ return bo->ccs_used;
+}
+
#endif
diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
index 9c199badd9b2..16cb97a5f2a6 100644
--- a/drivers/gpu/drm/xe/xe_bo_types.h
+++ b/drivers/gpu/drm/xe/xe_bo_types.h
@@ -88,6 +88,12 @@ struct xe_bo {
/** @ccs_cleared: true means that CCS region of BO is already cleared */
bool ccs_cleared;
+ /**
+ * @ccs_used: true means that CCS region of BO is used, protected by
+ * BO's dma-resv lock
+ */
+ bool ccs_used;
+
/** @bb_ccs: BB instructions of CCS read/write. Valid only for VF */
struct xe_mem_pool_node *bb_ccs[XE_SRIOV_VF_CCS_CTX_COUNT];
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index c3836f6eab35..7fbf3607b48b 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -743,11 +743,16 @@ static int xe_vm_ops_add_rebind(struct xe_vma_ops *vops, struct xe_vma *vma,
u8 tile_mask)
{
struct xe_vma_op *op;
+ struct xe_bo *bo = xe_vma_bo(vma);
op = kzalloc_obj(*op);
if (!op)
return -ENOMEM;
+ if (bo && xe_pat_index_get_comp_en(xe_vma_vm(vma)->xe,
+ vma->attr.pat_index))
+ xe_bo_set_ccs_used(bo);
+
xe_vm_populate_rebind(op, vma, tile_mask);
list_add_tail(&op->link, &vops->list);
xe_vma_ops_incr_pt_update_ops(vops, tile_mask, 1);
@@ -3054,6 +3059,9 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
if (flags.request_decompress)
err = xe_bo_decompress(bo);
+
+ if (xe_pat_index_get_comp_en(vm->xe, vma->attr.pat_index))
+ xe_bo_set_ccs_used(bo);
}
return err;
--
2.34.1
next reply other threads:[~2026-05-01 1:19 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-05-01 1:19 Matthew Brost [this message]
2026-05-04 23:41 ` Claude review: drm/xe: Do not blindly copy system CCS pages when shrinking Claude Code Review Bot
2026-05-04 23:41 ` Claude Code Review Bot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260501011907.2331654-1-matthew.brost@intel.com \
--to=matthew.brost@intel.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=intel-xe@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox