[PATCH v3 0/3] Flat-CCS eviction enhancements

[PATCH v3 0/3] Flat-CCS eviction enhancements

[Ramalingam C]

Flat-CCS eviction enhancements

v3:
  Incorporated the review suggestions [Matt]

Ramalingam C (3):
  drm/i915/gt: BUG_ON unexpected NULL at scatterlist walking
  drm/i915/gt: optimize the ccs_sz calculation per chunk
  drm/i915/gt: Document the eviction of the Flat-CCS objects

 drivers/gpu/drm/i915/gt/intel_migrate.c | 73 ++++++++++++++-----------
 1 file changed, 40 insertions(+), 33 deletions(-)

-- 
2.20.1

[PATCH v3 1/3] drm/i915/gt: BUG_ON unexpected NULL at scatterlist walking

[Ramalingam C]

While locating the start of ccs scatterlist in smem scatterlist, that has
to be the size of lmem obj size + corresponding ccs data size, report bug
if scatterlist terminate before that length.

v2:
  s/GEM_BUG_ON/BUG_ON with more commenting [Matt]
v3:
  Converted GEM_BUG_ON into BUG_ON with more documentation [Matt]

Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com> (v1)
---
 drivers/gpu/drm/i915/gt/intel_migrate.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
index 9d552f30b627..168d17b6f48a 100644
--- a/drivers/gpu/drm/i915/gt/intel_migrate.c
+++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
@@ -687,6 +687,16 @@ static void get_ccs_sg_sgt(struct sgt_dma *it, u32 bytes_to_cpy)
 		bytes_to_cpy -= len;
 
 		it->sg = __sg_next(it->sg);
+
+		/*
+		 * On Flat-CCS capable platform when we back the lmem pages with
+		 * smem pages we add extra pages at the end of the smem
+		 * scatterlist, to store the ccs data corresponding to the lmem
+		 * pages. get_ccs_sg_sgt() is called to get the pointer for the
+		 * start of the extra pages added at the end of smem scatterlist.
+		 * So scatterlist can't end at or before bytes_to_cpy.
+		 */
+		BUG_ON(!it->sg);
 		it->dma = sg_dma_address(it->sg);
 		it->max = it->dma + sg_dma_len(it->sg);
 	} while (bytes_to_cpy);
@@ -748,8 +758,10 @@ intel_context_migrate_copy(struct intel_context *ce,
 		 * Need to fix it.
 		 */
 		ccs_bytes_to_cpy = src_sz != dst_sz ? GET_CCS_BYTES(i915, bytes_to_cpy) : 0;
-		if (ccs_bytes_to_cpy)
+		if (ccs_bytes_to_cpy) {
+			WARN_ON(abs(src_sz - dst_sz) < ccs_bytes_to_cpy);
 			get_ccs_sg_sgt(&it_ccs, bytes_to_cpy);
+		}
 	}
 
 	src_offset = 0;
-- 
2.20.1

[PATCH v3 2/3] drm/i915/gt: optimize the ccs_sz calculation per chunk

[Ramalingam C]

Calculate the ccs_sz that needs to be emitted based on the src
and dst pages emitted per chunk. And handle the return value of emit_pte
for the ccs pages.

v2:
  ccs_sz moved to the reduced scope [Matt]

Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/gt/intel_migrate.c | 36 +++++++++----------------
 1 file changed, 13 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
index 168d17b6f48a..fc6975e55fae 100644
--- a/drivers/gpu/drm/i915/gt/intel_migrate.c
+++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
@@ -647,17 +647,9 @@ static int scatter_list_length(struct scatterlist *sg)
 
 static void
 calculate_chunk_sz(struct drm_i915_private *i915, bool src_is_lmem,
-		   int *src_sz, int *ccs_sz, u32 bytes_to_cpy,
-		   u32 ccs_bytes_to_cpy)
+		   int *src_sz, u32 bytes_to_cpy, u32 ccs_bytes_to_cpy)
 {
 	if (ccs_bytes_to_cpy) {
-		/*
-		 * We can only copy the ccs data corresponding to
-		 * the CHUNK_SZ of lmem which is
-		 * GET_CCS_BYTES(i915, CHUNK_SZ))
-		 */
-		*ccs_sz = min_t(int, ccs_bytes_to_cpy, GET_CCS_BYTES(i915, CHUNK_SZ));
-
 		if (!src_is_lmem)
 			/*
 			 * When CHUNK_SZ is passed all the pages upto CHUNK_SZ
@@ -717,10 +709,10 @@ intel_context_migrate_copy(struct intel_context *ce,
 	struct drm_i915_private *i915 = ce->engine->i915;
 	u32 ccs_bytes_to_cpy = 0, bytes_to_cpy;
 	enum i915_cache_level ccs_cache_level;
-	int src_sz, dst_sz, ccs_sz;
 	u32 src_offset, dst_offset;
 	u8 src_access, dst_access;
 	struct i915_request *rq;
+	int src_sz, dst_sz;
 	bool ccs_is_src;
 	int err;
 
@@ -803,7 +795,7 @@ intel_context_migrate_copy(struct intel_context *ce,
 		if (err)
 			goto out_rq;
 
-		calculate_chunk_sz(i915, src_is_lmem, &src_sz, &ccs_sz,
+		calculate_chunk_sz(i915, src_is_lmem, &src_sz,
 				   bytes_to_cpy, ccs_bytes_to_cpy);
 
 		len = emit_pte(rq, &it_src, src_cache_level, src_is_lmem,
@@ -837,37 +829,35 @@ intel_context_migrate_copy(struct intel_context *ce,
 		bytes_to_cpy -= len;
 
 		if (ccs_bytes_to_cpy) {
+			int ccs_sz;
+
 			err = rq->engine->emit_flush(rq, EMIT_INVALIDATE);
 			if (err)
 				goto out_rq;
 
+			ccs_sz = GET_CCS_BYTES(i915, len);
 			err = emit_pte(rq, &it_ccs, ccs_cache_level, false,
 				       ccs_is_src ? src_offset : dst_offset,
 				       ccs_sz);
+			if (err < 0)
+				goto out_rq;
+			if (err < ccs_sz) {
+				err = -EINVAL;
+				goto out_rq;
+			}
 
 			err = rq->engine->emit_flush(rq, EMIT_INVALIDATE);
 			if (err)
 				goto out_rq;
 
-			/*
-			 * Using max of src_sz and dst_sz, as we need to
-			 * pass the lmem size corresponding to the ccs
-			 * blocks we need to handle.
-			 */
-			ccs_sz = max_t(int, ccs_is_src ? ccs_sz : src_sz,
-				       ccs_is_src ? dst_sz : ccs_sz);
-
 			err = emit_copy_ccs(rq, dst_offset, dst_access,
-					    src_offset, src_access, ccs_sz);
+					    src_offset, src_access, len);
 			if (err)
 				goto out_rq;
 
 			err = rq->engine->emit_flush(rq, EMIT_INVALIDATE);
 			if (err)
 				goto out_rq;
-
-			/* Converting back to ccs bytes */
-			ccs_sz = GET_CCS_BYTES(rq->engine->i915, ccs_sz);
 			ccs_bytes_to_cpy -= ccs_sz;
 		}
 
-- 
2.20.1

[PATCH v3 3/3] drm/i915/gt: Document the eviction of the Flat-CCS objects

[Ramalingam C]

Capture the eviction details for Flat-CCS capable, lmem objects.

v2:
  Fix the Flat-ccs capbility of lmem obj with smem residency
  possibility [Thomas]
v3:
  Fixed the suggestions [Matt]

Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
cc: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/i915/gt/intel_migrate.c | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
index fc6975e55fae..509955885b93 100644
--- a/drivers/gpu/drm/i915/gt/intel_migrate.c
+++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
@@ -485,16 +485,21 @@ static bool wa_1209644611_applies(int ver, u32 size)
  * And CCS data can be copied in and out of CCS region through
  * XY_CTRL_SURF_COPY_BLT. CPU can't access the CCS data directly.
  *
- * When we exhaust the lmem, if the object's placements support smem, then we can
- * directly decompress the compressed lmem object into smem and start using it
- * from smem itself.
+ * I915 supports Flat-CCS on lmem only objects. When an objects has smem in
+ * its preference list, on memory pressure, i915 needs to migrate the lmem
+ * content into smem. If the lmem object is Flat-CCS compressed by userspace,
+ * then i915 needs to decompress it. But I915 lack the required information
+ * for such decompression. Hence I915 supports Flat-CCS only on lmem only objects.
  *
- * But when we need to swapout the compressed lmem object into a smem region
- * though objects' placement doesn't support smem, then we copy the lmem content
- * as it is into smem region along with ccs data (using XY_CTRL_SURF_COPY_BLT).
- * When the object is referred, lmem content will be swaped in along with
- * restoration of the CCS data (using XY_CTRL_SURF_COPY_BLT) at corresponding
- * location.
+ * When we exhaust the lmem, Flat-CCS capable objects' lmem backing memory can
+ * be temporarily evicted to smem, along with the auxiliary CCS state, where
+ * it can be potentially swapped-out at a later point, if required.
+ * If userspace later touches the evicted pages, then we always move
+ * the backing memory back to lmem, which includes restoring the saved CCS state,
+ * and potentially performing any required swap-in.
+ *
+ * For the migration of the lmem objects with smem in placement list, such as
+ * {lmem, smem}, objects are treated as non Flat-CCS capable objects.
  */
 
 static inline u32 *i915_flush_dw(u32 *cmd, u32 flags)
-- 
2.20.1

Re: [PATCH v3 1/3] drm/i915/gt: BUG_ON unexpected NULL at scatterlist walking

[Jani Nikula]

On Mon, 02 May 2022, Ramalingam C <ramalingam.c@intel.com> wrote:
> While locating the start of ccs scatterlist in smem scatterlist, that has
> to be the size of lmem obj size + corresponding ccs data size, report bug
> if scatterlist terminate before that length.
>
> v2:
>   s/GEM_BUG_ON/BUG_ON with more commenting [Matt]
> v3:
>   Converted GEM_BUG_ON into BUG_ON with more documentation [Matt]
>
> Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com> (v1)
> ---
>  drivers/gpu/drm/i915/gt/intel_migrate.c | 14 +++++++++++++-
>  1 file changed, 13 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
> index 9d552f30b627..168d17b6f48a 100644
> --- a/drivers/gpu/drm/i915/gt/intel_migrate.c
> +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
> @@ -687,6 +687,16 @@ static void get_ccs_sg_sgt(struct sgt_dma *it, u32 bytes_to_cpy)
>  		bytes_to_cpy -= len;
>  
>  		it->sg = __sg_next(it->sg);
> +
> +		/*
> +		 * On Flat-CCS capable platform when we back the lmem pages with
> +		 * smem pages we add extra pages at the end of the smem
> +		 * scatterlist, to store the ccs data corresponding to the lmem
> +		 * pages. get_ccs_sg_sgt() is called to get the pointer for the
> +		 * start of the extra pages added at the end of smem scatterlist.
> +		 * So scatterlist can't end at or before bytes_to_cpy.
> +		 */
> +		BUG_ON(!it->sg);

Why would you have to bring the entire kernel down in this case? Why not
just let it oops on the NULL pointer dereference?

I'd prefer nuking *all* of the current BUG/BUG_ON in the driver, and not
add any single one back.


BR,
Jani.


>  		it->dma = sg_dma_address(it->sg);
>  		it->max = it->dma + sg_dma_len(it->sg);
>  	} while (bytes_to_cpy);
> @@ -748,8 +758,10 @@ intel_context_migrate_copy(struct intel_context *ce,
>  		 * Need to fix it.
>  		 */
>  		ccs_bytes_to_cpy = src_sz != dst_sz ? GET_CCS_BYTES(i915, bytes_to_cpy) : 0;
> -		if (ccs_bytes_to_cpy)
> +		if (ccs_bytes_to_cpy) {
> +			WARN_ON(abs(src_sz - dst_sz) < ccs_bytes_to_cpy);
>  			get_ccs_sg_sgt(&it_ccs, bytes_to_cpy);
> +		}
>  	}
>  
>  	src_offset = 0;

-- 
Jani Nikula, Intel Open Source Graphics Center