drm/radeon/kms: attempt to avoid copying data twice on coherent cards. (v3)

On coherent systems (not-AGP) the IB should be in cached memory so should
be just as fast, so we can avoid copying to temporary pages and just use it
directly.

provides minor speedups on rv530: gears ~1820->1860, ipers: 29.9->30.6,
but always good to use less CPU if we can.

v3: cleanup unneeded bits.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

drivers/gpu/drm/radeon/radeon_cs.c

@@ -278,11 +278,16 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
 				  p->chunks[p->chunk_ib_idx].length_dw);
 			return -EINVAL;
 		}
+		if ((p->rdev->flags & RADEON_IS_AGP)) {
 			p->chunks[p->chunk_ib_idx].kpage[0] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 			p->chunks[p->chunk_ib_idx].kpage[1] = kmalloc(PAGE_SIZE, GFP_KERNEL);
 			if (p->chunks[p->chunk_ib_idx].kpage[0] == NULL ||
-		    p->chunks[p->chunk_ib_idx].kpage[1] == NULL)
+			    p->chunks[p->chunk_ib_idx].kpage[1] == NULL) {
+				kfree(p->chunks[i].kpage[0]);
+				kfree(p->chunks[i].kpage[1]);
 				return -ENOMEM;
+			}
+		}
 		p->chunks[p->chunk_ib_idx].kpage_idx[0] = -1;
 		p->chunks[p->chunk_ib_idx].kpage_idx[1] = -1;
 		p->chunks[p->chunk_ib_idx].last_copied_page = -1;
@@ -323,9 +328,11 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 	kfree(parser->relocs_ptr);
 	for (i = 0; i < parser->nchunks; i++) {
 		kfree(parser->chunks[i].kdata);
+		if ((parser->rdev->flags & RADEON_IS_AGP)) {
 			kfree(parser->chunks[i].kpage[0]);
 			kfree(parser->chunks[i].kpage[1]);
 		}
+	}
 	kfree(parser->chunks);
 	kfree(parser->chunks_array);
 	radeon_ib_free(parser->rdev, &parser->ib);
@@ -573,6 +580,7 @@ int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 	struct radeon_cs_chunk *ibc = &p->chunks[p->chunk_ib_idx];
 	int i;
 	int size = PAGE_SIZE;
+	bool copy1 = (p->rdev->flags & RADEON_IS_AGP) ? false : true;
 
 	for (i = ibc->last_copied_page + 1; i < pg_idx; i++) {
 		if (DRM_COPY_FROM_USER(p->ib->ptr + (i * (PAGE_SIZE/4)),
@@ -583,14 +591,16 @@ int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 		}
 	}
 
-	new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
-
 	if (pg_idx == ibc->last_page_index) {
 		size = (ibc->length_dw * 4) % PAGE_SIZE;
 		if (size == 0)
 			size = PAGE_SIZE;
 	}
 
+	new_page = ibc->kpage_idx[0] < ibc->kpage_idx[1] ? 0 : 1;
+	if (copy1)
+		ibc->kpage[new_page] = p->ib->ptr + (pg_idx * (PAGE_SIZE / 4));
+
 	if (DRM_COPY_FROM_USER(ibc->kpage[new_page],
 			       ibc->user_ptr + (pg_idx * PAGE_SIZE),
 			       size)) {
@@ -598,7 +608,8 @@ int radeon_cs_update_pages(struct radeon_cs_parser *p, int pg_idx)
 		return 0;
 	}
 
-	/* copy to IB here */
+	/* copy to IB for non single case */
+	if (!copy1)
 		memcpy((void *)(p->ib->ptr+(pg_idx*(PAGE_SIZE/4))), ibc->kpage[new_page], size);
 
 	ibc->last_copied_page = pg_idx;