drm/amdgpu: Move to gtt before cpu accesses dma buf.

To improve cpu read performance. This is implemented for APUs currently.

v2: Adapt to change https://lists.freedesktop.org/archives/amd-gfx/2017-October/015174.html
v3: Adapt to change "forward begin_cpu_access callback to drivers"
v4: Instead of v3, reuse drm_gem dmabuf_ops here. Also some minor fixes as suggested.
v5: only set dma_buf ops when it is valid (Samuel)
Signed-off-by: Samuel Li <Samuel.Li@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -418,6 +418,8 @@ amdgpu_gem_prime_import_sg_table(struct drm_device *dev,
 struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
 					struct drm_gem_object *gobj,
 					int flags);
+struct drm_gem_object *amdgpu_gem_prime_import(struct drm_device *dev,
+					    struct dma_buf *dma_buf);
 int amdgpu_gem_prime_pin(struct drm_gem_object *obj);
 void amdgpu_gem_prime_unpin(struct drm_gem_object *obj);
 struct reservation_object *amdgpu_gem_prime_res_obj(struct drm_gem_object *);

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -870,7 +870,7 @@ static struct drm_driver kms_driver = {
 	.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
 	.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
 	.gem_prime_export = amdgpu_gem_prime_export,
-	.gem_prime_import = drm_gem_prime_import,
+	.gem_prime_import = amdgpu_gem_prime_import,
 	.gem_prime_pin = amdgpu_gem_prime_pin,
 	.gem_prime_unpin = amdgpu_gem_prime_unpin,
 	.gem_prime_res_obj = amdgpu_gem_prime_res_obj,

drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c

@@ -26,6 +26,7 @@
 #include <drm/drmP.h>
 
 #include "amdgpu.h"
+#include "amdgpu_display.h"
 #include <drm/amdgpu_drm.h>
 #include <linux/dma-buf.h>
 
@@ -164,6 +165,50 @@ struct reservation_object *amdgpu_gem_prime_res_obj(struct drm_gem_object *obj)
 	return bo->tbo.resv;
 }
 
+static int amdgpu_gem_begin_cpu_access(struct dma_buf *dma_buf,
+				       enum dma_data_direction direction)
+{
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(dma_buf->priv);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct ttm_operation_ctx ctx = { true, false };
+	u32 domain = amdgpu_display_framebuffer_domains(adev);
+	int ret;
+	bool reads = (direction == DMA_BIDIRECTIONAL ||
+		      direction == DMA_FROM_DEVICE);
+
+	if (!reads || !(domain & AMDGPU_GEM_DOMAIN_GTT))
+		return 0;
+
+	/* move to gtt */
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (!bo->pin_count && (bo->allowed_domains & AMDGPU_GEM_DOMAIN_GTT)) {
+		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+		ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	}
+
+	amdgpu_bo_unreserve(bo);
+	return ret;
+}
+
+static const struct dma_buf_ops amdgpu_dmabuf_ops = {
+	.attach = drm_gem_map_attach,
+	.detach = drm_gem_map_detach,
+	.map_dma_buf = drm_gem_map_dma_buf,
+	.unmap_dma_buf = drm_gem_unmap_dma_buf,
+	.release = drm_gem_dmabuf_release,
+	.begin_cpu_access = amdgpu_gem_begin_cpu_access,
+	.map = drm_gem_dmabuf_kmap,
+	.map_atomic = drm_gem_dmabuf_kmap_atomic,
+	.unmap = drm_gem_dmabuf_kunmap,
+	.unmap_atomic = drm_gem_dmabuf_kunmap_atomic,
+	.mmap = drm_gem_dmabuf_mmap,
+	.vmap = drm_gem_dmabuf_vmap,
+	.vunmap = drm_gem_dmabuf_vunmap,
+};
+
 struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
 					struct drm_gem_object *gobj,
 					int flags)
@@ -176,7 +221,30 @@ struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
 		return ERR_PTR(-EPERM);
 
 	buf = drm_gem_prime_export(dev, gobj, flags);
-	if (!IS_ERR(buf))
+	if (!IS_ERR(buf)) {
 		buf->file->f_mapping = dev->anon_inode->i_mapping;
+		buf->ops = &amdgpu_dmabuf_ops;
+	}
+
 	return buf;
 }
+
+struct drm_gem_object *amdgpu_gem_prime_import(struct drm_device *dev,
+					    struct dma_buf *dma_buf)
+{
+	struct drm_gem_object *obj;
+
+	if (dma_buf->ops == &amdgpu_dmabuf_ops) {
+		obj = dma_buf->priv;
+		if (obj->dev == dev) {
+			/*
+			 * Importing dmabuf exported from out own gem increases
+			 * refcount on gem itself instead of f_count of dmabuf.
+			 */
+			drm_gem_object_get(obj);
+			return obj;
+		}
+	}
+
+	return drm_gem_prime_import(dev, dma_buf);
+}