[PATCH 23/23] stash

[Matthew Auld]

Signed-off-by: Matthew Auld <matthew.auld at intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 68 ++++++++++++++++++++++++++++++-------
 drivers/gpu/drm/i915/i915_gem_gtt.h |  1 +
 2 files changed, 57 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 8a3dbb0cc521..9b276a18d01e 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -513,25 +513,67 @@ static void fill_page_dma_32(struct i915_address_space *vm,
 	fill_page_dma(vm, p, (u64)v << 32 | v);
 }
 
+static bool has_scratch_64K(struct i915_address_space *vm)
+{
+	return vm->scratch_page.order == get_order(I915_GTT_PAGE_SIZE_64K);
+}
+
 static int
 setup_scratch_page(struct i915_address_space *vm, gfp_t gfp)
 {
-	struct page *page;
+	struct page *page = NULL;
 	dma_addr_t addr;
+	int order;
 
-	page = alloc_page(gfp | __GFP_ZERO);
-	if (unlikely(!page))
-		return -ENOMEM;
+	/* If we want to support objects < 2M which use 64K pages, then we will
+	 * need to support a 64K scratch.
+	 *
+	 * XXX: make the scratch global throw it in i915->mm -- seems pointless
+	 * having a per-vm scratch, also allocating a properly aligned 64K
+	 * scratch page is maybe far too expensive?
+	 */
+	if (!i915_is_ggtt(vm) &&
+	    HAS_PAGE_SIZE(vm->i915, I915_GTT_PAGE_SIZE_64K)) {
+		order = get_order(I915_GTT_PAGE_SIZE_64K);
+		page = alloc_pages(order, gfp | __GFP_ZERO);
+		if (page) {
+			addr = dma_map_page(vm->dma, page, 0,
+					    I915_GTT_PAGE_SIZE_64K,
+					    PCI_DMA_BIDIRECTIONAL);
+			if (unlikely(dma_mapping_error(vm->dma, addr))) {
+				__free_pages(page, order);
+				page = NULL;
+			}
 
-	addr = dma_map_page(vm->dma, page, 0, PAGE_SIZE,
-			    PCI_DMA_BIDIRECTIONAL);
-	if (unlikely(dma_mapping_error(vm->dma, addr))) {
-		__free_page(page);
-		return -ENOMEM;
+			if (!IS_ALIGNED(addr, I915_GTT_PAGE_SIZE_64K)) {
+				dma_unmap_page(vm->dma, addr,
+					       I915_GTT_PAGE_SIZE_64K,
+					       PCI_DMA_BIDIRECTIONAL);
+				__free_pages(page, order);
+				page = NULL;
+				WARN_ON(1);
+			}
+		}
+	}
+
+	if (!page) {
+		order = 0;
+		page = alloc_page(gfp | __GFP_ZERO);
+		if (unlikely(!page))
+			return -ENOMEM;
+
+		addr = dma_map_page(vm->dma, page, 0, PAGE_SIZE,
+				    PCI_DMA_BIDIRECTIONAL);
+		if (unlikely(dma_mapping_error(vm->dma, addr))) {
+			__free_page(page);
+			return -ENOMEM;
+		}
 	}
 
 	vm->scratch_page.page = page;
 	vm->scratch_page.daddr = addr;
+	vm->scratch_page.order = order;
+
 	return 0;
 }
 
@@ -539,8 +581,9 @@ static void cleanup_scratch_page(struct i915_address_space *vm)
 {
 	struct i915_page_dma *p = &vm->scratch_page;
 
-	dma_unmap_page(vm->dma, p->daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	__free_page(p->page);
+	dma_unmap_page(vm->dma, p->daddr, BIT(p->order) << PAGE_SHIFT,
+		       PCI_DMA_BIDIRECTIONAL);
+	__free_pages(p->page, p->order);
 }
 
 static struct i915_page_table *alloc_pt(struct i915_address_space *vm)
@@ -1101,7 +1144,8 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
 		 */
 		if (maybe_64K) {
 			if (index == max ||
-			    (!iter->sg && IS_ALIGNED(vma->node.start +
+			    (has_scratch_64K(vma->vm) &&
+			     !iter->sg && IS_ALIGNED(vma->node.start +
 						     vma->node.size,
 						     I915_GTT_PAGE_SIZE_2M))) {
 				vaddr = kmap_atomic_px(pd);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 42be89d27193..62794c52a8da 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -213,6 +213,7 @@ struct i915_vma;
 
 struct i915_page_dma {
 	struct page *page;
+	int order;
 	union {
 		dma_addr_t daddr;
 
-- 
2.13.4