[PATCH] drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear()

Thomas Hellström thomas.hellstrom at linux.intel.com
Mon Aug 12 08:55:46 UTC 2024


On Sat, 2024-08-10 at 00:23 +0200, Nirmoy Das wrote:
> Extracting this from
> https://patchwork.freedesktop.org/series/136277/ as 
> there is a regression in performance number on drm-tip.
> 
> I think it will take some time to find out what is going on. This
> patch 
> can be applied independently sending it separately.

Hi, Nirmoy.

What's the regression and do we have a Fixes: tag for this patch?
/Thomas


> 
> 
> Regards,
> 
> Nirmoy
> 
> On 8/10/2024 12:03 AM, Nirmoy Das wrote:
> > Parameterize clearing ccs and bo data in xe_migrate_clear() which 
> > higher
> > layers can utilize. This patch will be used later on when doing bo
> > data
> > clear for igfx as well.
> > 
> > v2: Replace multiple params with flags in xe_migrate_clear (Matt B)
> > v3: s/CLEAR_BO_DATA_FLAG_*/XE_MIGRATE_CLEAR_FLAG_* and move to
> >      xe_migrate.h. other nits(Matt B)
> > 
> > Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray at intel.com>
> > Cc: Matthew Auld <matthew.auld at intel.com>
> > Cc: Matthew Brost <matthew.brost at intel.com>
> > Cc: "Thomas Hellström" <thomas.hellstrom at linux.intel.com>
> > Signed-off-by: Nirmoy Das <nirmoy.das at intel.com>
> > Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar at intel.com>
> > Reviewed-by: Matthew Auld <matthew.auld at intel.com>
> > ---
> >   drivers/gpu/drm/xe/tests/xe_bo.c      |  3 ++-
> >   drivers/gpu/drm/xe/tests/xe_migrate.c | 12 ++++++++----
> >   drivers/gpu/drm/xe/xe_bo.c            | 12 ++++++++++--
> >   drivers/gpu/drm/xe/xe_migrate.c       | 27 +++++++++++++++++++---
> > -----
> >   drivers/gpu/drm/xe/xe_migrate.h       |  7 ++++++-
> >   5 files changed, 45 insertions(+), 16 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c
> > b/drivers/gpu/drm/xe/tests/xe_bo.c
> > index 1768483da1b7..df9fd907edd4 100644
> > --- a/drivers/gpu/drm/xe/tests/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/tests/xe_bo.c
> > @@ -36,7 +36,8 @@ static int ccs_test_migrate(struct xe_tile *tile,
> > struct xe_bo *bo,
> >   
> >   	/* Optionally clear bo *and* CCS data in VRAM. */
> >   	if (clear) {
> > -		fence = xe_migrate_clear(tile->migrate, bo, bo-
> > >ttm.resource);
> > +		fence = xe_migrate_clear(tile->migrate, bo, bo-
> > >ttm.resource,
> > +					
> > XE_MIGRATE_CLEAR_FLAG_FULL);
> >   		if (IS_ERR(fence)) {
> >   			KUNIT_FAIL(test, "Failed to submit bo
> > clear.\n");
> >   			return PTR_ERR(fence);
> > diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c
> > b/drivers/gpu/drm/xe/tests/xe_migrate.c
> > index 4344a1724029..47ae9d0b8864 100644
> > --- a/drivers/gpu/drm/xe/tests/xe_migrate.c
> > +++ b/drivers/gpu/drm/xe/tests/xe_migrate.c
> > @@ -105,7 +105,8 @@ static void test_copy(struct xe_migrate *m,
> > struct xe_bo *bo,
> >   	}
> >   
> >   	xe_map_memset(xe, &remote->vmap, 0, 0xd0, remote->size);
> > -	fence = xe_migrate_clear(m, remote, remote->ttm.resource);
> > +	fence = xe_migrate_clear(m, remote, remote->ttm.resource,
> > +				 XE_MIGRATE_CLEAR_FLAG_FULL);
> >   	if (!sanity_fence_failed(xe, fence, big ? "Clearing remote
> > big bo" :
> >   				 "Clearing remote small bo",
> > test)) {
> >   		retval = xe_map_rd(xe, &remote->vmap, 0, u64);
> > @@ -279,7 +280,8 @@ static void xe_migrate_sanity_test(struct
> > xe_migrate *m, struct kunit *test)
> >   	kunit_info(test, "Clearing small buffer object\n");
> >   	xe_map_memset(xe, &tiny->vmap, 0, 0x22, tiny->size);
> >   	expected = 0;
> > -	fence = xe_migrate_clear(m, tiny, tiny->ttm.resource);
> > +	fence = xe_migrate_clear(m, tiny, tiny->ttm.resource,
> > +				 XE_MIGRATE_CLEAR_FLAG_FULL);
> >   	if (sanity_fence_failed(xe, fence, "Clearing small bo",
> > test))
> >   		goto out;
> >   
> > @@ -300,7 +302,8 @@ static void xe_migrate_sanity_test(struct
> > xe_migrate *m, struct kunit *test)
> >   	kunit_info(test, "Clearing big buffer object\n");
> >   	xe_map_memset(xe, &big->vmap, 0, 0x11, big->size);
> >   	expected = 0;
> > -	fence = xe_migrate_clear(m, big, big->ttm.resource);
> > +	fence = xe_migrate_clear(m, big, big->ttm.resource,
> > +				 XE_MIGRATE_CLEAR_FLAG_FULL);
> >   	if (sanity_fence_failed(xe, fence, "Clearing big bo",
> > test))
> >   		goto out;
> >   
> > @@ -603,7 +606,8 @@ static void test_clear(struct xe_device *xe,
> > struct xe_tile *tile,
> >   
> >   	kunit_info(test, "Clear vram buffer object\n");
> >   	expected = 0x0000000000000000;
> > -	fence = xe_migrate_clear(tile->migrate, vram_bo, vram_bo-
> > >ttm.resource);
> > +	fence = xe_migrate_clear(tile->migrate, vram_bo, vram_bo-
> > >ttm.resource,
> > +				 XE_MIGRATE_CLEAR_FLAG_FULL);
> >   	if (sanity_fence_failed(xe, fence, "Clear vram_bo", test))
> >   		return;
> >   	dma_fence_put(fence);
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index 3295bc92d7aa..56a089aa3916 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -793,8 +793,16 @@ static int xe_bo_move(struct ttm_buffer_object
> > *ttm_bo, bool evict,
> >   			}
> >   		}
> >   	} else {
> > -		if (move_lacks_source)
> > -			fence = xe_migrate_clear(migrate, bo,
> > new_mem);
> > +		if (move_lacks_source) {
> > +			u32 flags = 0;
> > +
> > +			if (mem_type_is_vram(new_mem->mem_type))
> > +				flags |=
> > XE_MIGRATE_CLEAR_FLAG_FULL;
> > +			else if (handle_system_ccs)
> > +				flags |=
> > XE_MIGRATE_CLEAR_FLAG_CCS_DATA;
> > +
> > +			fence = xe_migrate_clear(migrate, bo,
> > new_mem, flags);
> > +		}
> >   		else
> >   			fence = xe_migrate_copy(migrate, bo, bo,
> > old_mem,
> >   						new_mem,
> > handle_system_ccs);
> > diff --git a/drivers/gpu/drm/xe/xe_migrate.c
> > b/drivers/gpu/drm/xe/xe_migrate.c
> > index 6f24aaf58252..a2d0ce3c59bf 100644
> > --- a/drivers/gpu/drm/xe/xe_migrate.c
> > +++ b/drivers/gpu/drm/xe/xe_migrate.c
> > @@ -1037,9 +1037,11 @@ static void emit_clear(struct xe_gt *gt,
> > struct xe_bb *bb, u64 src_ofs,
> >    * @m: The migration context.
> >    * @bo: The buffer object @dst is currently bound to.
> >    * @dst: The dst TTM resource to be cleared.
> > + * @clear_flags: flags to specify which data to clear: CCS, BO, or
> > both.
> >    *
> > - * Clear the contents of @dst to zero. On flat CCS devices,
> > - * the CCS metadata is cleared to zero as well on VRAM
> > destinations.
> > + * Clear the contents of @dst to zero when
> > XE_MIGRATE_CLEAR_FLAG_BO_DATA is set.
> > + * On flat CCS devices, the CCS metadata is cleared to zero with
> > XE_MIGRATE_CLEAR_FLAG_CCS_DATA.
> > + * Set XE_MIGRATE_CLEAR_FLAG_FULL to clear bo as well as CCS
> > metadata.
> >    * TODO: Eliminate the @bo argument.
> >    *
> >    * Return: Pointer to a dma_fence representing the last clear
> > batch, or
> > @@ -1048,18 +1050,27 @@ static void emit_clear(struct xe_gt *gt,
> > struct xe_bb *bb, u64 src_ofs,
> >    */
> >   struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> >   				   struct xe_bo *bo,
> > -				   struct ttm_resource *dst)
> > +				   struct ttm_resource *dst,
> > +				   u32 clear_flags)
> >   {
> >   	bool clear_vram = mem_type_is_vram(dst->mem_type);
> > +	bool clear_bo_data = XE_MIGRATE_CLEAR_FLAG_BO_DATA &
> > clear_flags;
> > +	bool clear_ccs = XE_MIGRATE_CLEAR_FLAG_CCS_DATA &
> > clear_flags;
> >   	struct xe_gt *gt = m->tile->primary_gt;
> >   	struct xe_device *xe = gt_to_xe(gt);
> > -	bool clear_system_ccs = (xe_bo_needs_ccs_pages(bo) &&
> > !IS_DGFX(xe)) ? true : false;
> > +	bool clear_only_system_ccs = false;
> >   	struct dma_fence *fence = NULL;
> >   	u64 size = bo->size;
> >   	struct xe_res_cursor src_it;
> >   	struct ttm_resource *src = dst;
> >   	int err;
> >   
> > +	if (WARN_ON(!clear_bo_data && !clear_ccs))
> > +		return NULL;
> > +
> > +	if (!clear_bo_data && clear_ccs && !IS_DGFX(xe))
> > +		clear_only_system_ccs = true;
> > +
> >   	if (!clear_vram)
> >   		xe_res_first_sg(xe_bo_sg(bo), 0, bo->size,
> > &src_it);
> >   	else
> > @@ -1085,7 +1096,7 @@ struct dma_fence *xe_migrate_clear(struct
> > xe_migrate *m,
> >   		batch_size = 2 +
> >   			pte_update_size(m, pte_flags, src,
> > &src_it,
> >   					&clear_L0, &clear_L0_ofs,
> > &clear_L0_pt,
> > -					clear_system_ccs ? 0 :
> > emit_clear_cmd_len(gt), 0,
> > +					clear_bo_data ?
> > emit_clear_cmd_len(gt) : 0, 0,
> >   					avail_pts);
> >   
> >   		if (xe_migrate_needs_ccs_emit(xe))
> > @@ -1107,13 +1118,13 @@ struct dma_fence *xe_migrate_clear(struct
> > xe_migrate *m,
> >   		if (clear_vram &&
> > xe_migrate_allow_identity(clear_L0, &src_it))
> >   			xe_res_next(&src_it, clear_L0);
> >   		else
> > -			emit_pte(m, bb, clear_L0_pt, clear_vram,
> > clear_system_ccs,
> > +			emit_pte(m, bb, clear_L0_pt, clear_vram,
> > clear_only_system_ccs,
> >   				 &src_it, clear_L0, dst);
> >   
> >   		bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
> >   		update_idx = bb->len;
> >   
> > -		if (!clear_system_ccs)
> > +		if (clear_bo_data)
> >   			emit_clear(gt, bb, clear_L0_ofs, clear_L0,
> > XE_PAGE_SIZE, clear_vram);
> >   
> >   		if (xe_migrate_needs_ccs_emit(xe)) {
> > @@ -1172,7 +1183,7 @@ struct dma_fence *xe_migrate_clear(struct
> > xe_migrate *m,
> >   		return ERR_PTR(err);
> >   	}
> >   
> > -	if (clear_system_ccs)
> > +	if (clear_ccs)
> >   		bo->ccs_cleared = true;
> >   
> >   	return fence;
> > diff --git a/drivers/gpu/drm/xe/xe_migrate.h
> > b/drivers/gpu/drm/xe/xe_migrate.h
> > index 453e0ecf5034..7929cc2425e8 100644
> > --- a/drivers/gpu/drm/xe/xe_migrate.h
> > +++ b/drivers/gpu/drm/xe/xe_migrate.h
> > @@ -102,9 +102,14 @@ struct dma_fence *xe_migrate_copy(struct
> > xe_migrate *m,
> >   				  struct ttm_resource *dst,
> >   				  bool copy_only_ccs);
> >   
> > +#define XE_MIGRATE_CLEAR_FLAG_BO_DATA		BIT(0)
> > +#define XE_MIGRATE_CLEAR_FLAG_CCS_DATA		BIT(1)
> > +#define
> > XE_MIGRATE_CLEAR_FLAG_FULL	(XE_MIGRATE_CLEAR_FLAG_BO_DATA | \
> > +					XE_MIGRATE_CLEAR_FLAG_CCS_
> > DATA)
> >   struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
> >   				   struct xe_bo *bo,
> > -				   struct ttm_resource *dst);
> > +				   struct ttm_resource *dst,
> > +				   u32 clear_flags);
> >   
> >   struct xe_vm *xe_migrate_get_vm(struct xe_migrate *m);
> >   



More information about the Intel-xe mailing list