[PATCH v6 0/3] drm/xe/oa: Wa_14026633728

[PATCH v6 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v6 though v2: Change to Patch 1 (see changelog in Patch 1)

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 123 ++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   4 +-
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 70 insertions(+), 58 deletions(-)

-- 
2.54.0

[PATCH 1/3] drm/xe/oa: Use xe_map layer

[Ashutosh Dixit]

OA code should have used xe_map layer to begin with. In CRI, the OA buffer
can be located both in system and device memory. For these reasons, move OA
code to use the xe_map layer when accessing the OA buffer.

v2: Use xe_map layer and put_user in xe_oa_copy_to_user (Umesh)
v3: To avoid performance impact in v2, revert to v1 but move
    xe_map_copy_to_user() to xe_map.h
v4: Use bounce buffer and copy_to_user in xe_oa_copy_to_user
v5: Fix offsets in oa_timestamp() and oa_timestamp_clear() (Umesh)
v6: Rename head/tail in helper args to report_offset (Umesh)

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c       | 95 +++++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h |  4 +-
 2 files changed, 58 insertions(+), 41 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 6337e671c97ae..1fff0e8e9e78e 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -213,32 +213,40 @@ static u32 xe_oa_hw_tail_read(struct xe_oa_stream *stream)
 #define oa_report_header_64bit(__s) \
 	((__s)->oa_buffer.format->header == HDR_64_BIT)
 
-static u64 oa_report_id(struct xe_oa_stream *stream, void *report)
+static u64 oa_report_id(struct xe_oa_stream *stream, u32 report_offset)
 {
-	return oa_report_header_64bit(stream) ? *(u64 *)report : *(u32 *)report;
+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
+
+	return oa_report_header_64bit(stream) ?
+		xe_map_rd(stream->oa->xe, map, report_offset, u64) :
+		xe_map_rd(stream->oa->xe, map, report_offset, u32);
 }
 
-static void oa_report_id_clear(struct xe_oa_stream *stream, u32 *report)
+static void oa_report_id_clear(struct xe_oa_stream *stream, u32 report_offset)
 {
-	if (oa_report_header_64bit(stream))
-		*(u64 *)report = 0;
-	else
-		*report = 0;
+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
+
+	oa_report_header_64bit(stream) ?
+		xe_map_wr(stream->oa->xe, map, report_offset, u64, 0) :
+		xe_map_wr(stream->oa->xe, map, report_offset, u32, 0);
 }
 
-static u64 oa_timestamp(struct xe_oa_stream *stream, void *report)
+static u64 oa_timestamp(struct xe_oa_stream *stream, u32 report_offset)
 {
+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
+
 	return oa_report_header_64bit(stream) ?
-		*((u64 *)report + 1) :
-		*((u32 *)report + 1);
+		xe_map_rd(stream->oa->xe, map, report_offset + 8, u64) :
+		xe_map_rd(stream->oa->xe, map, report_offset + 4, u32);
 }
 
-static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 *report)
+static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
 {
-	if (oa_report_header_64bit(stream))
-		*(u64 *)&report[2] = 0;
-	else
-		report[1] = 0;
+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
+
+	oa_report_header_64bit(stream) ?
+		xe_map_wr(stream->oa->xe, map, report_offset + 8, u64, 0) :
+		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
 }
 
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
@@ -275,9 +283,7 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 	 * they were written.  If not : (╯°□°）╯︵ ┻━┻
 	 */
 	while (xe_oa_circ_diff(stream, tail, stream->oa_buffer.tail) >= report_size) {
-		void *report = stream->oa_buffer.vaddr + tail;
-
-		if (oa_report_id(stream, report) || oa_timestamp(stream, report))
+		if (oa_report_id(stream, tail) || oa_timestamp(stream, tail))
 			break;
 
 		tail = xe_oa_circ_diff(stream, tail, report_size);
@@ -311,30 +317,35 @@ static enum hrtimer_restart xe_oa_poll_check_timer_cb(struct hrtimer *hrtimer)
 	return HRTIMER_RESTART;
 }
 
+static unsigned long
+xe_oa_copy_to_user(struct xe_oa_stream *stream, void __user *dst, u32 report_offset, u32 len)
+{
+	xe_map_memcpy_from(stream->oa->xe, stream->oa_buffer.bounce,
+			   &stream->oa_buffer.bo->vmap, report_offset, len);
+	return copy_to_user(dst, stream->oa_buffer.bounce, len);
+}
+
 static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
-			       size_t count, size_t *offset, const u8 *report)
+			       size_t count, size_t *offset, u32 report_offset)
 {
 	int report_size = stream->oa_buffer.format->size;
 	int report_size_partial;
-	u8 *oa_buf_end;
 
 	if ((count - *offset) < report_size)
 		return -ENOSPC;
 
 	buf += *offset;
 
-	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
-	report_size_partial = oa_buf_end - report;
+	report_size_partial = stream->oa_buffer.circ_size - report_offset;
 
 	if (report_size_partial < report_size) {
-		if (copy_to_user(buf, report, report_size_partial))
+		if (xe_oa_copy_to_user(stream, buf, report_offset, report_size_partial))
 			return -EFAULT;
 		buf += report_size_partial;
 
-		if (copy_to_user(buf, stream->oa_buffer.vaddr,
-				 report_size - report_size_partial))
+		if (xe_oa_copy_to_user(stream, buf, 0, report_size - report_size_partial))
 			return -EFAULT;
-	} else if (copy_to_user(buf, report, report_size)) {
+	} else if (xe_oa_copy_to_user(stream, buf, report_offset, report_size)) {
 		return -EFAULT;
 	}
 
@@ -347,7 +358,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 				size_t count, size_t *offset)
 {
 	int report_size = stream->oa_buffer.format->size;
-	u8 *oa_buf_base = stream->oa_buffer.vaddr;
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
 	size_t start_offset = *offset;
 	unsigned long flags;
@@ -364,26 +374,24 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
 
 	for (; xe_oa_circ_diff(stream, tail, head);
 	     head = xe_oa_circ_incr(stream, head, report_size)) {
-		u8 *report = oa_buf_base + head;
-
-		ret = xe_oa_append_report(stream, buf, count, offset, report);
+		ret = xe_oa_append_report(stream, buf, count, offset, head);
 		if (ret)
 			break;
 
 		if (!(stream->oa_buffer.circ_size % report_size)) {
 			/* Clear out report id and timestamp to detect unlanded reports */
-			oa_report_id_clear(stream, (void *)report);
-			oa_timestamp_clear(stream, (void *)report);
+			oa_report_id_clear(stream, head);
+			oa_timestamp_clear(stream, head);
 		} else {
-			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
-			u32 part = oa_buf_end - report;
+			struct iosys_map *map = &stream->oa_buffer.bo->vmap;
+			u32 part = stream->oa_buffer.circ_size - head;
 
 			/* Zero out the entire report */
 			if (report_size <= part) {
-				memset(report, 0, report_size);
+				xe_map_memset(stream->oa->xe, map, head, 0, report_size);
 			} else {
-				memset(report, 0, part);
-				memset(oa_buf_base, 0, report_size - part);
+				xe_map_memset(stream->oa->xe, map, head, 0, part);
+				xe_map_memset(stream->oa->xe, map, 0, 0, report_size - part);
 			}
 		}
 	}
@@ -436,7 +444,8 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
 
 	/* Zero out the OA buffer since we rely on zero report id and timestamp fields */
-	memset(stream->oa_buffer.vaddr, 0, xe_bo_size(stream->oa_buffer.bo));
+	xe_map_memset(stream->oa->xe, &stream->oa_buffer.bo->vmap, 0, 0,
+		      xe_bo_size(stream->oa_buffer.bo));
 }
 
 static u32 __format_to_oactrl(const struct xe_oa_format *format, int counter_sel_mask)
@@ -699,6 +708,7 @@ static int num_lri_dwords(int num_regs)
 static void xe_oa_free_oa_buffer(struct xe_oa_stream *stream)
 {
 	xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
+	kfree(stream->oa_buffer.bounce);
 }
 
 static void xe_oa_free_configs(struct xe_oa_stream *stream)
@@ -889,9 +899,16 @@ static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 		return PTR_ERR(bo);
 
 	stream->oa_buffer.bo = bo;
+
 	/* mmap implementation requires OA buffer to be in system memory */
 	xe_assert(stream->oa->xe, bo->vmap.is_iomem == 0);
-	stream->oa_buffer.vaddr = bo->vmap.vaddr;
+
+	stream->oa_buffer.bounce = kmalloc(stream->oa_buffer.format->size, GFP_KERNEL);
+	if (!stream->oa_buffer.bounce) {
+		xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
+		return -ENOMEM;
+	}
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/xe/xe_oa_types.h b/drivers/gpu/drm/xe/xe_oa_types.h
index 8906c3084b5f8..3d9ec8490899c 100644
--- a/drivers/gpu/drm/xe/xe_oa_types.h
+++ b/drivers/gpu/drm/xe/xe_oa_types.h
@@ -164,8 +164,8 @@ struct xe_oa_buffer {
 	/** @bo: xe_bo backing the OA buffer */
 	struct xe_bo *bo;
 
-	/** @vaddr: mapped vaddr of the OA buffer */
-	u8 *vaddr;
+	/** @bounce: bounce buffer used with xe_map layer */
+	void *bounce;
 
 	/** @ptr_lock: Lock protecting reads/writes to head/tail pointers */
 	spinlock_t ptr_lock;
-- 
2.54.0

Re: [PATCH 1/3] drm/xe/oa: Use xe_map layer

[Umesh Nerlige Ramappa]

On Mon, Apr 27, 2026 at 12:02:21PM -0700, Ashutosh Dixit wrote:
>OA code should have used xe_map layer to begin with. In CRI, the OA buffer
>can be located both in system and device memory. For these reasons, move OA
>code to use the xe_map layer when accessing the OA buffer.
>
>v2: Use xe_map layer and put_user in xe_oa_copy_to_user (Umesh)
>v3: To avoid performance impact in v2, revert to v1 but move
>    xe_map_copy_to_user() to xe_map.h
>v4: Use bounce buffer and copy_to_user in xe_oa_copy_to_user
>v5: Fix offsets in oa_timestamp() and oa_timestamp_clear() (Umesh)
>v6: Rename head/tail in helper args to report_offset (Umesh)

Thanks for renaming. My comments are in the v5 version of the series.

With an assert/check in the xe_oa_copy_to_user(), this should be

Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Umesh

>
>Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
>---
> drivers/gpu/drm/xe/xe_oa.c       | 95 +++++++++++++++++++-------------
> drivers/gpu/drm/xe/xe_oa_types.h |  4 +-
> 2 files changed, 58 insertions(+), 41 deletions(-)
>
>diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
>index 6337e671c97ae..1fff0e8e9e78e 100644
>--- a/drivers/gpu/drm/xe/xe_oa.c
>+++ b/drivers/gpu/drm/xe/xe_oa.c
>@@ -213,32 +213,40 @@ static u32 xe_oa_hw_tail_read(struct xe_oa_stream *stream)
> #define oa_report_header_64bit(__s) \
> 	((__s)->oa_buffer.format->header == HDR_64_BIT)
>
>-static u64 oa_report_id(struct xe_oa_stream *stream, void *report)
>+static u64 oa_report_id(struct xe_oa_stream *stream, u32 report_offset)
> {
>-	return oa_report_header_64bit(stream) ? *(u64 *)report : *(u32 *)report;
>+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
>+
>+	return oa_report_header_64bit(stream) ?
>+		xe_map_rd(stream->oa->xe, map, report_offset, u64) :
>+		xe_map_rd(stream->oa->xe, map, report_offset, u32);
> }
>
>-static void oa_report_id_clear(struct xe_oa_stream *stream, u32 *report)
>+static void oa_report_id_clear(struct xe_oa_stream *stream, u32 report_offset)
> {
>-	if (oa_report_header_64bit(stream))
>-		*(u64 *)report = 0;
>-	else
>-		*report = 0;
>+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
>+
>+	oa_report_header_64bit(stream) ?
>+		xe_map_wr(stream->oa->xe, map, report_offset, u64, 0) :
>+		xe_map_wr(stream->oa->xe, map, report_offset, u32, 0);
> }
>
>-static u64 oa_timestamp(struct xe_oa_stream *stream, void *report)
>+static u64 oa_timestamp(struct xe_oa_stream *stream, u32 report_offset)
> {
>+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
>+
> 	return oa_report_header_64bit(stream) ?
>-		*((u64 *)report + 1) :
>-		*((u32 *)report + 1);
>+		xe_map_rd(stream->oa->xe, map, report_offset + 8, u64) :
>+		xe_map_rd(stream->oa->xe, map, report_offset + 4, u32);
> }
>
>-static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 *report)
>+static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
> {
>-	if (oa_report_header_64bit(stream))
>-		*(u64 *)&report[2] = 0;
>-	else
>-		report[1] = 0;
>+	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
>+
>+	oa_report_header_64bit(stream) ?
>+		xe_map_wr(stream->oa->xe, map, report_offset + 8, u64, 0) :
>+		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
> }
>
> static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
>@@ -275,9 +283,7 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
> 	 * they were written.  If not : (╯°□°）╯︵ ┻━┻
> 	 */
> 	while (xe_oa_circ_diff(stream, tail, stream->oa_buffer.tail) >= report_size) {
>-		void *report = stream->oa_buffer.vaddr + tail;
>-
>-		if (oa_report_id(stream, report) || oa_timestamp(stream, report))
>+		if (oa_report_id(stream, tail) || oa_timestamp(stream, tail))
> 			break;
>
> 		tail = xe_oa_circ_diff(stream, tail, report_size);
>@@ -311,30 +317,35 @@ static enum hrtimer_restart xe_oa_poll_check_timer_cb(struct hrtimer *hrtimer)
> 	return HRTIMER_RESTART;
> }
>
>+static unsigned long
>+xe_oa_copy_to_user(struct xe_oa_stream *stream, void __user *dst, u32 report_offset, u32 len)
>+{
>+	xe_map_memcpy_from(stream->oa->xe, stream->oa_buffer.bounce,
>+			   &stream->oa_buffer.bo->vmap, report_offset, len);
>+	return copy_to_user(dst, stream->oa_buffer.bounce, len);
>+}
>+
> static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
>-			       size_t count, size_t *offset, const u8 *report)
>+			       size_t count, size_t *offset, u32 report_offset)
> {
> 	int report_size = stream->oa_buffer.format->size;
> 	int report_size_partial;
>-	u8 *oa_buf_end;
>
> 	if ((count - *offset) < report_size)
> 		return -ENOSPC;
>
> 	buf += *offset;
>
>-	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
>-	report_size_partial = oa_buf_end - report;
>+	report_size_partial = stream->oa_buffer.circ_size - report_offset;
>
> 	if (report_size_partial < report_size) {
>-		if (copy_to_user(buf, report, report_size_partial))
>+		if (xe_oa_copy_to_user(stream, buf, report_offset, report_size_partial))
> 			return -EFAULT;
> 		buf += report_size_partial;
>
>-		if (copy_to_user(buf, stream->oa_buffer.vaddr,
>-				 report_size - report_size_partial))
>+		if (xe_oa_copy_to_user(stream, buf, 0, report_size - report_size_partial))
> 			return -EFAULT;
>-	} else if (copy_to_user(buf, report, report_size)) {
>+	} else if (xe_oa_copy_to_user(stream, buf, report_offset, report_size)) {
> 		return -EFAULT;
> 	}
>
>@@ -347,7 +358,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
> 				size_t count, size_t *offset)
> {
> 	int report_size = stream->oa_buffer.format->size;
>-	u8 *oa_buf_base = stream->oa_buffer.vaddr;
> 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
> 	size_t start_offset = *offset;
> 	unsigned long flags;
>@@ -364,26 +374,24 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
>
> 	for (; xe_oa_circ_diff(stream, tail, head);
> 	     head = xe_oa_circ_incr(stream, head, report_size)) {
>-		u8 *report = oa_buf_base + head;
>-
>-		ret = xe_oa_append_report(stream, buf, count, offset, report);
>+		ret = xe_oa_append_report(stream, buf, count, offset, head);
> 		if (ret)
> 			break;
>
> 		if (!(stream->oa_buffer.circ_size % report_size)) {
> 			/* Clear out report id and timestamp to detect unlanded reports */
>-			oa_report_id_clear(stream, (void *)report);
>-			oa_timestamp_clear(stream, (void *)report);
>+			oa_report_id_clear(stream, head);
>+			oa_timestamp_clear(stream, head);
> 		} else {
>-			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
>-			u32 part = oa_buf_end - report;
>+			struct iosys_map *map = &stream->oa_buffer.bo->vmap;
>+			u32 part = stream->oa_buffer.circ_size - head;
>
> 			/* Zero out the entire report */
> 			if (report_size <= part) {
>-				memset(report, 0, report_size);
>+				xe_map_memset(stream->oa->xe, map, head, 0, report_size);
> 			} else {
>-				memset(report, 0, part);
>-				memset(oa_buf_base, 0, report_size - part);
>+				xe_map_memset(stream->oa->xe, map, head, 0, part);
>+				xe_map_memset(stream->oa->xe, map, 0, 0, report_size - part);
> 			}
> 		}
> 	}
>@@ -436,7 +444,8 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
> 	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
>
> 	/* Zero out the OA buffer since we rely on zero report id and timestamp fields */
>-	memset(stream->oa_buffer.vaddr, 0, xe_bo_size(stream->oa_buffer.bo));
>+	xe_map_memset(stream->oa->xe, &stream->oa_buffer.bo->vmap, 0, 0,
>+		      xe_bo_size(stream->oa_buffer.bo));
> }
>
> static u32 __format_to_oactrl(const struct xe_oa_format *format, int counter_sel_mask)
>@@ -699,6 +708,7 @@ static int num_lri_dwords(int num_regs)
> static void xe_oa_free_oa_buffer(struct xe_oa_stream *stream)
> {
> 	xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
>+	kfree(stream->oa_buffer.bounce);
> }
>
> static void xe_oa_free_configs(struct xe_oa_stream *stream)
>@@ -889,9 +899,16 @@ static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
> 		return PTR_ERR(bo);
>
> 	stream->oa_buffer.bo = bo;
>+
> 	/* mmap implementation requires OA buffer to be in system memory */
> 	xe_assert(stream->oa->xe, bo->vmap.is_iomem == 0);
>-	stream->oa_buffer.vaddr = bo->vmap.vaddr;
>+
>+	stream->oa_buffer.bounce = kmalloc(stream->oa_buffer.format->size, GFP_KERNEL);
>+	if (!stream->oa_buffer.bounce) {
>+		xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
>+		return -ENOMEM;
>+	}
>+
> 	return 0;
> }
>
>diff --git a/drivers/gpu/drm/xe/xe_oa_types.h b/drivers/gpu/drm/xe/xe_oa_types.h
>index 8906c3084b5f8..3d9ec8490899c 100644
>--- a/drivers/gpu/drm/xe/xe_oa_types.h
>+++ b/drivers/gpu/drm/xe/xe_oa_types.h
>@@ -164,8 +164,8 @@ struct xe_oa_buffer {
> 	/** @bo: xe_bo backing the OA buffer */
> 	struct xe_bo *bo;
>
>-	/** @vaddr: mapped vaddr of the OA buffer */
>-	u8 *vaddr;
>+	/** @bounce: bounce buffer used with xe_map layer */
>+	void *bounce;
>
> 	/** @ptr_lock: Lock protecting reads/writes to head/tail pointers */
> 	spinlock_t ptr_lock;
>-- 
>2.54.0
>

Re: [PATCH 1/3] drm/xe/oa: Use xe_map layer

[Dixit, Ashutosh]

On Mon, 27 Apr 2026 12:34:24 -0700, Umesh Nerlige Ramappa wrote:
>
> On Mon, Apr 27, 2026 at 12:02:21PM -0700, Ashutosh Dixit wrote:
> > OA code should have used xe_map layer to begin with. In CRI, the OA buffer
> > can be located both in system and device memory. For these reasons, move OA
> > code to use the xe_map layer when accessing the OA buffer.
> >
> > v2: Use xe_map layer and put_user in xe_oa_copy_to_user (Umesh)
> > v3: To avoid performance impact in v2, revert to v1 but move
> >    xe_map_copy_to_user() to xe_map.h
> > v4: Use bounce buffer and copy_to_user in xe_oa_copy_to_user
> > v5: Fix offsets in oa_timestamp() and oa_timestamp_clear() (Umesh)
> > v6: Rename head/tail in helper args to report_offset (Umesh)
>
> Thanks for renaming. My comments are in the v5 version of the series.
>
> With an assert/check in the xe_oa_copy_to_user(), this should be

Add in v7. The only remaining comment is the one about the test, will make
a note of it.

>
> Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Thanks.
--
Ashutosh


>
> >
> > Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
> > ---
> > drivers/gpu/drm/xe/xe_oa.c       | 95 +++++++++++++++++++-------------
> > drivers/gpu/drm/xe/xe_oa_types.h |  4 +-
> > 2 files changed, 58 insertions(+), 41 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
> > index 6337e671c97ae..1fff0e8e9e78e 100644
> > --- a/drivers/gpu/drm/xe/xe_oa.c
> > +++ b/drivers/gpu/drm/xe/xe_oa.c
> > @@ -213,32 +213,40 @@ static u32 xe_oa_hw_tail_read(struct xe_oa_stream *stream)
> > #define oa_report_header_64bit(__s) \
> >	((__s)->oa_buffer.format->header == HDR_64_BIT)
> >
> > -static u64 oa_report_id(struct xe_oa_stream *stream, void *report)
> > +static u64 oa_report_id(struct xe_oa_stream *stream, u32 report_offset)
> > {
> > -	return oa_report_header_64bit(stream) ? *(u64 *)report : *(u32 *)report;
> > +	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
> > +
> > +	return oa_report_header_64bit(stream) ?
> > +		xe_map_rd(stream->oa->xe, map, report_offset, u64) :
> > +		xe_map_rd(stream->oa->xe, map, report_offset, u32);
> > }
> >
> > -static void oa_report_id_clear(struct xe_oa_stream *stream, u32 *report)
> > +static void oa_report_id_clear(struct xe_oa_stream *stream, u32 report_offset)
> > {
> > -	if (oa_report_header_64bit(stream))
> > -		*(u64 *)report = 0;
> > -	else
> > -		*report = 0;
> > +	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
> > +
> > +	oa_report_header_64bit(stream) ?
> > +		xe_map_wr(stream->oa->xe, map, report_offset, u64, 0) :
> > +		xe_map_wr(stream->oa->xe, map, report_offset, u32, 0);
> > }
> >
> > -static u64 oa_timestamp(struct xe_oa_stream *stream, void *report)
> > +static u64 oa_timestamp(struct xe_oa_stream *stream, u32 report_offset)
> > {
> > +	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
> > +
> >	return oa_report_header_64bit(stream) ?
> > -		*((u64 *)report + 1) :
> > -		*((u32 *)report + 1);
> > +		xe_map_rd(stream->oa->xe, map, report_offset + 8, u64) :
> > +		xe_map_rd(stream->oa->xe, map, report_offset + 4, u32);
> > }
> >
> > -static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 *report)
> > +static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
> > {
> > -	if (oa_report_header_64bit(stream))
> > -		*(u64 *)&report[2] = 0;
> > -	else
> > -		report[1] = 0;
> > +	struct iosys_map *map = &stream->oa_buffer.bo->vmap;
> > +
> > +	oa_report_header_64bit(stream) ?
> > +		xe_map_wr(stream->oa->xe, map, report_offset + 8, u64, 0) :
> > +		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
> > }
> >
> > static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
> > @@ -275,9 +283,7 @@ static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
> >	 * they were written.  If not : (╯°□°）╯︵ ┻━┻
> >	 */
> >	while (xe_oa_circ_diff(stream, tail, stream->oa_buffer.tail) >= report_size) {
> > -		void *report = stream->oa_buffer.vaddr + tail;
> > -
> > -		if (oa_report_id(stream, report) || oa_timestamp(stream, report))
> > +		if (oa_report_id(stream, tail) || oa_timestamp(stream, tail))
> >			break;
> >
> >		tail = xe_oa_circ_diff(stream, tail, report_size);
> > @@ -311,30 +317,35 @@ static enum hrtimer_restart xe_oa_poll_check_timer_cb(struct hrtimer *hrtimer)
> >	return HRTIMER_RESTART;
> > }
> >
> > +static unsigned long
> > +xe_oa_copy_to_user(struct xe_oa_stream *stream, void __user *dst, u32 report_offset, u32 len)
> > +{
> > +	xe_map_memcpy_from(stream->oa->xe, stream->oa_buffer.bounce,
> > +			   &stream->oa_buffer.bo->vmap, report_offset, len);
> > +	return copy_to_user(dst, stream->oa_buffer.bounce, len);
> > +}
> > +
> > static int xe_oa_append_report(struct xe_oa_stream *stream, char __user *buf,
> > -			       size_t count, size_t *offset, const u8 *report)
> > +			       size_t count, size_t *offset, u32 report_offset)
> > {
> >	int report_size = stream->oa_buffer.format->size;
> >	int report_size_partial;
> > -	u8 *oa_buf_end;
> >
> >	if ((count - *offset) < report_size)
> >		return -ENOSPC;
> >
> >	buf += *offset;
> >
> > -	oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
> > -	report_size_partial = oa_buf_end - report;
> > +	report_size_partial = stream->oa_buffer.circ_size - report_offset;
> >
> >	if (report_size_partial < report_size) {
> > -		if (copy_to_user(buf, report, report_size_partial))
> > +		if (xe_oa_copy_to_user(stream, buf, report_offset, report_size_partial))
> >			return -EFAULT;
> >		buf += report_size_partial;
> >
> > -		if (copy_to_user(buf, stream->oa_buffer.vaddr,
> > -				 report_size - report_size_partial))
> > +		if (xe_oa_copy_to_user(stream, buf, 0, report_size - report_size_partial))
> >			return -EFAULT;
> > -	} else if (copy_to_user(buf, report, report_size)) {
> > +	} else if (xe_oa_copy_to_user(stream, buf, report_offset, report_size)) {
> >		return -EFAULT;
> >	}
> >
> > @@ -347,7 +358,6 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
> >				size_t count, size_t *offset)
> > {
> >	int report_size = stream->oa_buffer.format->size;
> > -	u8 *oa_buf_base = stream->oa_buffer.vaddr;
> >	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
> >	size_t start_offset = *offset;
> >	unsigned long flags;
> > @@ -364,26 +374,24 @@ static int xe_oa_append_reports(struct xe_oa_stream *stream, char __user *buf,
> >
> >	for (; xe_oa_circ_diff(stream, tail, head);
> >	     head = xe_oa_circ_incr(stream, head, report_size)) {
> > -		u8 *report = oa_buf_base + head;
> > -
> > -		ret = xe_oa_append_report(stream, buf, count, offset, report);
> > +		ret = xe_oa_append_report(stream, buf, count, offset, head);
> >		if (ret)
> >			break;
> >
> >		if (!(stream->oa_buffer.circ_size % report_size)) {
> >			/* Clear out report id and timestamp to detect unlanded reports */
> > -			oa_report_id_clear(stream, (void *)report);
> > -			oa_timestamp_clear(stream, (void *)report);
> > +			oa_report_id_clear(stream, head);
> > +			oa_timestamp_clear(stream, head);
> >		} else {
> > -			u8 *oa_buf_end = stream->oa_buffer.vaddr + stream->oa_buffer.circ_size;
> > -			u32 part = oa_buf_end - report;
> > +			struct iosys_map *map = &stream->oa_buffer.bo->vmap;
> > +			u32 part = stream->oa_buffer.circ_size - head;
> >
> >			/* Zero out the entire report */
> >			if (report_size <= part) {
> > -				memset(report, 0, report_size);
> > +				xe_map_memset(stream->oa->xe, map, head, 0, report_size);
> >			} else {
> > -				memset(report, 0, part);
> > -				memset(oa_buf_base, 0, report_size - part);
> > +				xe_map_memset(stream->oa->xe, map, head, 0, part);
> > +				xe_map_memset(stream->oa->xe, map, 0, 0, report_size - part);
> >			}
> >		}
> >	}
> > @@ -436,7 +444,8 @@ static void xe_oa_init_oa_buffer(struct xe_oa_stream *stream)
> >	spin_unlock_irqrestore(&stream->oa_buffer.ptr_lock, flags);
> >
> >	/* Zero out the OA buffer since we rely on zero report id and timestamp fields */
> > -	memset(stream->oa_buffer.vaddr, 0, xe_bo_size(stream->oa_buffer.bo));
> > +	xe_map_memset(stream->oa->xe, &stream->oa_buffer.bo->vmap, 0, 0,
> > +		      xe_bo_size(stream->oa_buffer.bo));
> > }
> >
> > static u32 __format_to_oactrl(const struct xe_oa_format *format, int counter_sel_mask)
> > @@ -699,6 +708,7 @@ static int num_lri_dwords(int num_regs)
> > static void xe_oa_free_oa_buffer(struct xe_oa_stream *stream)
> > {
> >	xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
> > +	kfree(stream->oa_buffer.bounce);
> > }
> >
> > static void xe_oa_free_configs(struct xe_oa_stream *stream)
> > @@ -889,9 +899,16 @@ static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
> >		return PTR_ERR(bo);
> >
> >	stream->oa_buffer.bo = bo;
> > +
> >	/* mmap implementation requires OA buffer to be in system memory */
> >	xe_assert(stream->oa->xe, bo->vmap.is_iomem == 0);
> > -	stream->oa_buffer.vaddr = bo->vmap.vaddr;
> > +
> > +	stream->oa_buffer.bounce = kmalloc(stream->oa_buffer.format->size, GFP_KERNEL);
> > +	if (!stream->oa_buffer.bounce) {
> > +		xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
> > +		return -ENOMEM;
> > +	}
> > +
> >	return 0;
> > }
> >
> > diff --git a/drivers/gpu/drm/xe/xe_oa_types.h b/drivers/gpu/drm/xe/xe_oa_types.h
> > index 8906c3084b5f8..3d9ec8490899c 100644
> > --- a/drivers/gpu/drm/xe/xe_oa_types.h
> > +++ b/drivers/gpu/drm/xe/xe_oa_types.h
> > @@ -164,8 +164,8 @@ struct xe_oa_buffer {
> >	/** @bo: xe_bo backing the OA buffer */
> >	struct xe_bo *bo;
> >
> > -	/** @vaddr: mapped vaddr of the OA buffer */
> > -	u8 *vaddr;
> > +	/** @bounce: bounce buffer used with xe_map layer */
> > +	void *bounce;
> >
> >	/** @ptr_lock: Lock protecting reads/writes to head/tail pointers */
> >	spinlock_t ptr_lock;
> > --
> > 2.54.0
> >

[PATCH 2/3] drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap

[Ashutosh Dixit]

OA buffer mmap can currently only mmap OA buffer in system memory. CRI
MERTOA buffer can be located in device memory. Switch OA buffer mmap to
using drm_gem_mmap_obj, which can handle mmap's of both system and device
memory buffers.

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c | 20 +++-----------------
 1 file changed, 3 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 1fff0e8e9e78e..d3994ae8bc82d 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -9,6 +9,7 @@
 #include <linux/poll.h>
 
 #include <drm/drm_drv.h>
+#include <drm/drm_gem.h>
 #include <drm/drm_managed.h>
 #include <drm/drm_syncobj.h>
 #include <uapi/drm/xe_drm.h>
@@ -900,9 +901,6 @@ static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 
 	stream->oa_buffer.bo = bo;
 
-	/* mmap implementation requires OA buffer to be in system memory */
-	xe_assert(stream->oa->xe, bo->vmap.is_iomem == 0);
-
 	stream->oa_buffer.bounce = kmalloc(stream->oa_buffer.format->size, GFP_KERNEL);
 	if (!stream->oa_buffer.bounce) {
 		xe_bo_unpin_map_no_vm(stream->oa_buffer.bo);
@@ -1690,8 +1688,6 @@ static int xe_oa_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	struct xe_oa_stream *stream = file->private_data;
 	struct xe_bo *bo = stream->oa_buffer.bo;
-	unsigned long start = vma->vm_start;
-	int i, ret;
 
 	if (xe_observation_paranoid && !perfmon_capable()) {
 		drm_dbg(&stream->oa->xe->drm, "Insufficient privilege to map OA buffer\n");
@@ -1699,7 +1695,7 @@ static int xe_oa_mmap(struct file *file, struct vm_area_struct *vma)
 	}
 
 	/* Can mmap the entire OA buffer or nothing (no partial OA buffer mmaps) */
-	if (vma->vm_end - vma->vm_start != xe_bo_size(stream->oa_buffer.bo)) {
+	if (vma->vm_end - vma->vm_start != xe_bo_size(bo)) {
 		drm_dbg(&stream->oa->xe->drm, "Wrong mmap size, must be OA buffer size\n");
 		return -EINVAL;
 	}
@@ -1715,17 +1711,7 @@ static int xe_oa_mmap(struct file *file, struct vm_area_struct *vma)
 	vm_flags_mod(vma, VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP | VM_DONTCOPY,
 		     VM_MAYWRITE | VM_MAYEXEC);
 
-	xe_assert(stream->oa->xe, bo->ttm.ttm->num_pages == vma_pages(vma));
-	for (i = 0; i < bo->ttm.ttm->num_pages; i++) {
-		ret = remap_pfn_range(vma, start, page_to_pfn(bo->ttm.ttm->pages[i]),
-				      PAGE_SIZE, vma->vm_page_prot);
-		if (ret)
-			break;
-
-		start += PAGE_SIZE;
-	}
-
-	return ret;
+	return drm_gem_mmap_obj(&bo->ttm.base, xe_bo_size(bo), vma);
 }
 
 static const struct file_operations xe_oa_fops = {
-- 
2.54.0

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index d3994ae8bc82d..b988b02c661c7 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
 		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -891,11 +896,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.54.0

[PATCH v8 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v7 though v2: Change to Patch 1 (see changelog in Patch 1)
v8: Change to Patch 3 (see changelog in Patch 3)

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_device_wa_oob.rules |   1 +
 drivers/gpu/drm/xe/xe_oa.c                | 126 ++++++++++++----------
 drivers/gpu/drm/xe/xe_oa_types.h          |   4 +-
 3 files changed, 73 insertions(+), 58 deletions(-)

-- 
2.54.0

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

v2: Move WA to xe_device_wa_oob.rules and use XE_DEVICE_WA() (Matt Roper)

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_device_wa_oob.rules |  1 +
 drivers/gpu/drm/xe/xe_oa.c                | 11 ++++++++++-
 2 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_device_wa_oob.rules b/drivers/gpu/drm/xe/xe_device_wa_oob.rules
index d129cddb6ead4..92371c4905290 100644
--- a/drivers/gpu/drm/xe/xe_device_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_device_wa_oob.rules
@@ -4,3 +4,4 @@
 22019338487_display	PLATFORM(LUNARLAKE)
 14022085890	SUBPLATFORM(BATTLEMAGE, G21)
 14026539277	PLATFORM(NOVALAKE_P), PLATFORM_STEP(A0, B0)
+14026633728	PLATFORM(CRESCENTISLAND)
diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 83797c9479f7c..5de5bf19240a2 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -15,6 +15,7 @@
 #include <uapi/drm/xe_drm.h>
 
 #include <generated/xe_wa_oob.h>
+#include <generated/xe_device_wa_oob.h>
 
 #include "abi/guc_actions_slpc_abi.h"
 #include "instructions/xe_mi_commands.h"
@@ -250,6 +251,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
 		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_DEVICE_WA(s->oa->xe, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -893,11 +899,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
-- 
2.54.0

[PATCH v7 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v7 though v2: Change to Patch 1 (see changelog in Patch 1)

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 125 ++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   4 +-
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 72 insertions(+), 58 deletions(-)

-- 
2.54.0

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 83797c9479f7c..d905302f3f151 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
 		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -893,11 +898,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.54.0

Re: [PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Matt Roper]

On Mon, Apr 27, 2026 at 01:26:55PM -0700, Ashutosh Dixit wrote:
> CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
> not system memory (which is the default for OA buffers).
> 
> Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
> Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
>  drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
>  2 files changed, 10 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
> index 83797c9479f7c..d905302f3f151 100644
> --- a/drivers/gpu/drm/xe/xe_oa.c
> +++ b/drivers/gpu/drm/xe/xe_oa.c
> @@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
>  		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
>  }
>  
> +static bool mert_wa_14026633728(struct xe_oa_stream *s)
> +{
> +	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
> +}
> +
>  static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
>  {
>  	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
> @@ -893,11 +898,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
>  
>  static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
>  {
> +	u32 vram = mert_wa_14026633728(stream) ?
> +		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
> +		XE_BO_FLAG_SYSTEM;
>  	struct xe_bo *bo;
>  
>  	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
>  				       size, ttm_bo_type_kernel,
> -				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
> +				       vram | XE_BO_FLAG_GGTT, false);
>  	if (IS_ERR(bo))
>  		return PTR_ERR(bo);
>  
> diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
> index f8a185103b805..a7c1bd9bcb943 100644
> --- a/drivers/gpu/drm/xe/xe_wa_oob.rules
> +++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
> @@ -65,3 +65,4 @@
>  
>  14025883347	MEDIA_VERSION_RANGE(1301, 3503)
>  		GRAPHICS_VERSION_RANGE(2004, 3005)
> +14026633728	PLATFORM(CRESCENTISLAND)

Is the underlying defect here tied to the graphics IP or the
platform/SoC? If graphics IP, this should key off the graphics version
instead of platform).  If platform, we should treat it as a device
workaround rather than a GT workaround. I.e., definition in
xe_device_wa_oob.rules and use XE_DEVICE_WA() at the check sites.


Matt

> -- 
> 2.54.0
> 

-- 
Matt Roper
Graphics Software Engineer
Linux GPU Platform Enablement
Intel Corporation

Re: [PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Dixit, Ashutosh]

On Mon, 27 Apr 2026 13:52:41 -0700, Matt Roper wrote:
>

Hi Matt,

> On Mon, Apr 27, 2026 at 01:26:55PM -0700, Ashutosh Dixit wrote:
> > CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
> > not system memory (which is the default for OA buffers).
> >
> > Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
> > Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
> >  drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
> >  2 files changed, 10 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
> > index 83797c9479f7c..d905302f3f151 100644
> > --- a/drivers/gpu/drm/xe/xe_oa.c
> > +++ b/drivers/gpu/drm/xe/xe_oa.c
> > @@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 report_offset)
> >		xe_map_wr(stream->oa->xe, map, report_offset + 4, u32, 0);
> >  }
> >
> > +static bool mert_wa_14026633728(struct xe_oa_stream *s)
> > +{
> > +	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
> > +}
> > +
> >  static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
> >  {
> >	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
> > @@ -893,11 +898,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
> >
> >  static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
> >  {
> > +	u32 vram = mert_wa_14026633728(stream) ?
> > +		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
> > +		XE_BO_FLAG_SYSTEM;
> >	struct xe_bo *bo;
> >
> >	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
> >				       size, ttm_bo_type_kernel,
> > -				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
> > +				       vram | XE_BO_FLAG_GGTT, false);
> >	if (IS_ERR(bo))
> >		return PTR_ERR(bo);
> >
> > diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
> > index f8a185103b805..a7c1bd9bcb943 100644
> > --- a/drivers/gpu/drm/xe/xe_wa_oob.rules
> > +++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
> > @@ -65,3 +65,4 @@
> >
> >  14025883347	MEDIA_VERSION_RANGE(1301, 3503)
> >		GRAPHICS_VERSION_RANGE(2004, 3005)
> > +14026633728	PLATFORM(CRESCENTISLAND)
>
> Is the underlying defect here tied to the graphics IP or the
> platform/SoC? If graphics IP, this should key off the graphics version
> instead of platform).  If platform, we should treat it as a device
> workaround rather than a GT workaround. I.e., definition in
> xe_device_wa_oob.rules and use XE_DEVICE_WA() at the check sites.

Because the WA is in MERT I'd think it would be a device WA. So I've
changed it to a device WA as suggested here in v8.

Thanks.
--
Ashutosh

[PATCH v5 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v5 though v2: Change to Patch 1 (see changelog in Patch 1)

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 123 ++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   4 +-
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 70 insertions(+), 58 deletions(-)

-- 
2.54.0

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 5ccf63c807920..2a9b0d9884237 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
 		xe_map_wr(stream->oa->xe, map, head + 4, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -891,11 +896,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.54.0

[PATCH v4 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v4: Change to Patch 1 again (see changelog in Patch 1)
v3: Change to Patch 1 again (see changelog in Patch 1)
v2: Change to Patch 1

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 123 ++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   4 +-
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 70 insertions(+), 58 deletions(-)

-- 
2.48.1

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 5b8a9bd26770c..6516e933b865c 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
 		xe_map_wr(stream->oa->xe, map, head + 1, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -891,11 +896,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.48.1

[PATCH v3 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v3: Change to Patch 1 again (see changelog in Patch 1)
v2: Change to Patch 1

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_map.h        |  14 ++++
 drivers/gpu/drm/xe/xe_oa.c         | 109 ++++++++++++++---------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   3 -
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 4 files changed, 68 insertions(+), 59 deletions(-)

-- 
2.48.1

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 7ada6e2a0d955..7610b2d744066 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
 		xe_map_wr(stream->oa->xe, map, head + 1, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -884,11 +889,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.48.1

[PATCH v2 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

v2: Change to Patch 1

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 127 ++++++++++++++++-------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   3 -
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 72 insertions(+), 59 deletions(-)

-- 
2.48.1

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index b24aae055c7df..7b31cf0a29ce1 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
 		xe_map_wr(stream->oa->xe, map, head + 1, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -902,11 +907,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.48.1

[PATCH 0/3] drm/xe/oa: Wa_14026633728

[Ashutosh Dixit]

Ashutosh Dixit (3):
  drm/xe/oa: Use xe_map layer
  drm/xe/oa: Use drm_gem_mmap_obj for OA buffer mmap
  drm/xe/oa: Implement Wa_14026633728

 drivers/gpu/drm/xe/xe_oa.c         | 117 +++++++++++++++--------------
 drivers/gpu/drm/xe/xe_oa_types.h   |   3 -
 drivers/gpu/drm/xe/xe_wa_oob.rules |   1 +
 3 files changed, 62 insertions(+), 59 deletions(-)

-- 
2.48.1

[PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Ashutosh Dixit]

CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
not system memory (which is the default for OA buffers).

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
---
 drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
 drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 0e65141e57ee8..f65cec62a0d2d 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
 		xe_map_wr(stream->oa->xe, map, head + 1, u32, 0);
 }
 
+static bool mert_wa_14026633728(struct xe_oa_stream *s)
+{
+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
+}
+
 static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
 {
 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
@@ -892,11 +897,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
 
 static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
 {
+	u32 vram = mert_wa_14026633728(stream) ?
+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
+		XE_BO_FLAG_SYSTEM;
 	struct xe_bo *bo;
 
 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
 				       size, ttm_bo_type_kernel,
-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
+				       vram | XE_BO_FLAG_GGTT, false);
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index f8a185103b805..a7c1bd9bcb943 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -65,3 +65,4 @@
 
 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
 		GRAPHICS_VERSION_RANGE(2004, 3005)
+14026633728	PLATFORM(CRESCENTISLAND)
-- 
2.48.1

Re: [PATCH 3/3] drm/xe/oa: Implement Wa_14026633728

[Umesh Nerlige Ramappa]

On Mon, Apr 06, 2026 at 08:02:19PM -0700, Ashutosh Dixit wrote:
>CRI Wa_14026633728 requires MERTOA buffer to be allocated in device memory,
>not system memory (which is the default for OA buffers).
>
>Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

LGTM,

Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

>---
> drivers/gpu/drm/xe/xe_oa.c         | 10 +++++++++-
> drivers/gpu/drm/xe/xe_wa_oob.rules |  1 +
> 2 files changed, 10 insertions(+), 1 deletion(-)
>
>diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
>index 0e65141e57ee8..f65cec62a0d2d 100644
>--- a/drivers/gpu/drm/xe/xe_oa.c
>+++ b/drivers/gpu/drm/xe/xe_oa.c
>@@ -250,6 +250,11 @@ static void oa_timestamp_clear(struct xe_oa_stream *stream, u32 head)
> 		xe_map_wr(stream->oa->xe, map, head + 1, u32, 0);
> }
>
>+static bool mert_wa_14026633728(struct xe_oa_stream *s)
>+{
>+	return s->oa_unit->type == DRM_XE_OA_UNIT_TYPE_MERT && XE_GT_WA(s->gt, 14026633728);
>+}
>+
> static bool xe_oa_buffer_check_unlocked(struct xe_oa_stream *stream)
> {
> 	u32 gtt_offset = xe_bo_ggtt_addr(stream->oa_buffer.bo);
>@@ -892,11 +897,14 @@ static void xe_oa_stream_destroy(struct xe_oa_stream *stream)
>
> static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size)
> {
>+	u32 vram = mert_wa_14026633728(stream) ?
>+		XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(stream->oa->xe)) :
>+		XE_BO_FLAG_SYSTEM;
> 	struct xe_bo *bo;
>
> 	bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile,
> 				       size, ttm_bo_type_kernel,
>-				       XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false);
>+				       vram | XE_BO_FLAG_GGTT, false);
> 	if (IS_ERR(bo))
> 		return PTR_ERR(bo);
>
>diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
>index f8a185103b805..a7c1bd9bcb943 100644
>--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
>+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
>@@ -65,3 +65,4 @@
>
> 14025883347	MEDIA_VERSION_RANGE(1301, 3503)
> 		GRAPHICS_VERSION_RANGE(2004, 3005)
>+14026633728	PLATFORM(CRESCENTISLAND)
>-- 
>2.48.1
>