gpu: nvgpu: avoid using priv data for non-compressed buffer

Instead of allocating priv data for all external buffers, allocate
only on a demand basis for when compression is requested either in CDE
or via libnvrm_gpu.

This will allow allocators like nvidia-drm to use non-compressed
buffers without needing to avoid the core drm checks.
e.g. drm_gem_prime_import_dev that checks for
if (dma_buf->ops == &drm_gem_prime_dmabuf_ops)"

This patch also gets rid of optimization of dma_buf's attach/detach
calls. Now, nvgpu instead needs to call attach/detach for everytime
the dmabuf fd is imported.

Change-Id: Idefd269b32974106e85ff09e17ebc752b92f830c
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2372213
Tested-by: Yogish Kulkarni <yogishk@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/os/linux/cde.c

@@ -1140,10 +1140,10 @@ __releases(&l->cde_app->mutex)
 
 		nvgpu_log(g, gpu_dbg_cde, "surface=0x%p scatterBuffer=0x%p",
 			  surface, scatter_buffer);
-		sgt = nvgpu_mm_pin_privdata(dev_from_gk20a(g), compbits_scatter_buf,
+		sgt = nvgpu_mm_pin(dev_from_gk20a(g), compbits_scatter_buf,
 				   &attachment);
 		if (IS_ERR(sgt)) {
-			nvgpu_warn(g,
+			nvgpu_err(g,
 				   "mm_pin failed");
 			err = -EINVAL;
 			goto exit_unmap_surface;
@@ -1156,7 +1156,7 @@ __releases(&l->cde_app->mutex)
 #endif
 			if (err != 0) {
 				nvgpu_warn(g, "buffer access setup failed");
-				nvgpu_mm_unpin_privdata(dev_from_gk20a(g), compbits_scatter_buf,
+				nvgpu_mm_unpin(dev_from_gk20a(g), compbits_scatter_buf,
 					attachment, sgt);
 				goto exit_unmap_surface;
 			}
@@ -1172,7 +1172,7 @@ __releases(&l->cde_app->mutex)
 #endif
 			WARN_ON(err);
 
-			nvgpu_mm_unpin_privdata(dev_from_gk20a(g), compbits_scatter_buf,
+			nvgpu_mm_unpin(dev_from_gk20a(g), compbits_scatter_buf,
 				       attachment, sgt);
 			if (err)
 				goto exit_unmap_surface;

drivers/gpu/nvgpu/os/linux/comptags.c

@@ -47,13 +47,21 @@ int gk20a_alloc_or_get_comptags(struct gk20a *g,
 				struct gk20a_comptag_allocator *allocator,
 				struct gk20a_comptags *comptags)
 {
-	struct gk20a_dmabuf_priv *priv = gk20a_dma_buf_get_drvdata(
-						buf->dmabuf, buf->dev);
+	int ret = 0;
+
+	struct gk20a_dmabuf_priv *priv = NULL;
 	u32 offset;
 	int err;
 	u64 ctag_granularity;
 	u32 lines;
 
+	ret = gk20a_dmabuf_alloc_drvdata(buf->dmabuf, buf->dev);
+	if (ret) {
+		nvgpu_err(g, "error allocating comptags priv data");
+		return ret;
+	}
+
+	priv = gk20a_dma_buf_get_drvdata(buf->dmabuf, buf->dev);
 	if (!priv)
 		return -ENOSYS;
 

drivers/gpu/nvgpu/os/linux/dmabuf_priv.c

@@ -91,7 +91,7 @@ static void nvgpu_dma_buf_release(struct dma_buf *dmabuf)
 	dmabuf->ops->release(dmabuf);
 }
 
-int gk20a_dma_buf_set_drvdata(struct dma_buf *dmabuf, struct device *device,
+static int gk20a_dma_buf_set_drvdata(struct dma_buf *dmabuf, struct device *device,
 			struct gk20a_dmabuf_priv *priv)
 {
 	nvgpu_mutex_acquire(&priv->lock);
@@ -134,69 +134,40 @@ struct gk20a_dmabuf_priv *gk20a_dma_buf_get_drvdata(
 	return priv;
 }
 
-struct sg_table *nvgpu_mm_pin_privdata(struct device *dev,
+struct sg_table *nvgpu_mm_pin(struct device *dev,
 			struct dma_buf *dmabuf, struct dma_buf_attachment **attachment)
 {
 	struct gk20a *g = get_gk20a(dev);
-	struct gk20a_dmabuf_priv *priv = NULL;
+	struct dma_buf_attachment *attach = NULL;
+	struct sg_table *sgt = NULL;
 
-	priv = gk20a_dma_buf_get_drvdata(dmabuf, dev);
-	if (!priv) {
-		nvgpu_do_assert();
-		return ERR_PTR(-EINVAL);
-	}
-
-	nvgpu_mutex_acquire(&priv->lock);
-
-	if (priv->pin_count == 0) {
-		priv->attach = dma_buf_attach(dmabuf, dev);
-		if (IS_ERR(priv->attach)) {
-			nvgpu_mutex_release(&priv->lock);
+	attach = dma_buf_attach(dmabuf, dev);
+	if (IS_ERR(attach)) {
 		nvgpu_err(g, "Failed to attach dma_buf (err = %ld)!",
-				  PTR_ERR(priv->attach));
-			return ERR_CAST(priv->attach);
+				PTR_ERR(attach));
+		return ERR_CAST(attach);
 	}
 
-		priv->sgt = dma_buf_map_attachment(priv->attach,
-						   DMA_BIDIRECTIONAL);
-		if (IS_ERR(priv->sgt)) {
-			dma_buf_detach(dmabuf, priv->attach);
-			nvgpu_mutex_release(&priv->lock);
+	sgt = dma_buf_map_attachment(attach, DMA_BIDIRECTIONAL);
+	if (IS_ERR(sgt)) {
+		dma_buf_detach(dmabuf, attach);
 		nvgpu_err(g, "Failed to map attachment (err = %ld)!",
-				  PTR_ERR(priv->sgt));
-			return ERR_CAST(priv->sgt);
-		}
+				PTR_ERR(sgt));
+		return ERR_CAST(sgt);
 	}
 
-	priv->pin_count++;
-	nvgpu_mutex_release(&priv->lock);
-	*attachment = priv->attach;
-	return priv->sgt;
+	*attachment = attach;
+
+	return sgt;
 }
 
-void nvgpu_mm_unpin_privdata(struct device *dev,
+void nvgpu_mm_unpin(struct device *dev,
 			struct dma_buf *dmabuf,
 			struct dma_buf_attachment *attachment,
 			struct sg_table *sgt)
 {
-	struct gk20a_dmabuf_priv *priv = gk20a_dma_buf_get_drvdata(dmabuf, dev);
-	dma_addr_t dma_addr;
-
-	if (IS_ERR(priv) || !priv)
-		return;
-
-	nvgpu_mutex_acquire(&priv->lock);
-	nvgpu_assert(priv->sgt == sgt);
-	nvgpu_assert(priv->attach == attachment);
-	priv->pin_count--;
-	nvgpu_assert(priv->pin_count >= 0);
-	dma_addr = sg_dma_address(priv->sgt->sgl);
-	if (priv->pin_count == 0) {
-		dma_buf_unmap_attachment(priv->attach, priv->sgt,
-					 DMA_BIDIRECTIONAL);
-		dma_buf_detach(dmabuf, priv->attach);
-	}
-	nvgpu_mutex_release(&priv->lock);
+	dma_buf_unmap_attachment(attachment, sgt, DMA_BIDIRECTIONAL);
+	dma_buf_detach(dmabuf, attachment);
 }
 
 /* This function must be called after acquiring the global level

drivers/gpu/nvgpu/os/linux/dmabuf_priv.h

@@ -74,11 +74,6 @@ struct gk20a_dmabuf_priv {
 	struct gk20a_comptag_allocator *comptag_allocator;
 	struct gk20a_comptags comptags;
 
-	struct dma_buf_attachment *attach;
-	struct sg_table *sgt;
-
-	int pin_count;
-
 	struct nvgpu_list_node states;
 
 	u64 buffer_id;
@@ -102,11 +97,11 @@ struct gk20a_dmabuf_priv {
 	struct nvgpu_list_node list;
 };
 
-struct sg_table *nvgpu_mm_pin_privdata(struct device *dev,
+struct sg_table *nvgpu_mm_pin(struct device *dev,
 			struct dma_buf *dmabuf,
 			struct dma_buf_attachment **attachment);
 
-void nvgpu_mm_unpin_privdata(struct device *dev,
+void nvgpu_mm_unpin(struct device *dev,
 		struct dma_buf *dmabuf,
 		struct dma_buf_attachment *attachment,
 		struct sg_table *sgt);
@@ -118,8 +113,6 @@ int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev);
 int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g,
 			   u64 offset, struct gk20a_buffer_state **state);
 
-int gk20a_dma_buf_set_drvdata(struct dma_buf *dmabuf, struct device *device,
-			struct gk20a_dmabuf_priv *priv);
 void gk20a_dma_buf_priv_list_clear(struct nvgpu_os_linux *l);
 struct gk20a_dmabuf_priv *gk20a_dma_buf_get_drvdata(
 		struct dma_buf *dmabuf, struct device *device);

drivers/gpu/nvgpu/os/linux/linux-channel.c

@@ -433,14 +433,9 @@ int nvgpu_usermode_buf_from_dmabuf(struct gk20a *g, int dmabuf_fd,
 		goto put_dmabuf;
 	}
 
-	err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev);
-	if (err != 0) {
-		goto put_dmabuf;
-	}
-
-	sgt = nvgpu_mm_pin_privdata(dev, dmabuf, &attachment);
+	sgt = nvgpu_mm_pin(dev, dmabuf, &attachment);
 	if (IS_ERR(sgt)) {
-		nvgpu_warn(g, "Failed to pin dma_buf!");
+		nvgpu_err(g, "Failed to pin dma_buf!");
 		err = PTR_ERR(sgt);
 		goto put_dmabuf;
 	}
@@ -476,7 +471,7 @@ void nvgpu_os_channel_free_usermode_buffers(struct nvgpu_channel *c)
 	struct device *dev = dev_from_gk20a(g);
 
 	if (priv->usermode.gpfifo.dmabuf != NULL) {
-		nvgpu_mm_unpin_privdata(dev, priv->usermode.gpfifo.dmabuf,
+		nvgpu_mm_unpin(dev, priv->usermode.gpfifo.dmabuf,
 			       priv->usermode.gpfifo.attachment,
 			       priv->usermode.gpfifo.sgt);
 		dma_buf_put(priv->usermode.gpfifo.dmabuf);
@@ -484,7 +479,7 @@ void nvgpu_os_channel_free_usermode_buffers(struct nvgpu_channel *c)
 	}
 
 	if (priv->usermode.userd.dmabuf != NULL) {
-		nvgpu_mm_unpin_privdata(dev, priv->usermode.userd.dmabuf,
+		nvgpu_mm_unpin(dev, priv->usermode.userd.dmabuf,
 		       priv->usermode.userd.attachment,
 		       priv->usermode.userd.sgt);
 		dma_buf_put(priv->usermode.userd.dmabuf);
@@ -547,7 +542,7 @@ static int nvgpu_channel_alloc_usermode_buffers(struct nvgpu_channel *c,
 unmap_free_gpfifo:
 	nvgpu_dma_unmap_free(c->vm, &c->usermode_gpfifo);
 free_gpfifo:
-	nvgpu_mm_unpin_privdata(dev, priv->usermode.gpfifo.dmabuf,
+	nvgpu_mm_unpin(dev, priv->usermode.gpfifo.dmabuf,
 		       priv->usermode.gpfifo.attachment,
 		       priv->usermode.gpfifo.sgt);
 	dma_buf_put(priv->usermode.gpfifo.dmabuf);

drivers/gpu/nvgpu/os/linux/vm.c

@@ -175,7 +175,7 @@ struct nvgpu_mapped_buf *nvgpu_vm_find_mapping(struct vm_gk20a *vm,
 	 * the dmabuf doesn't support drvdata, prior SGT is unpinned as the
 	 * new SGT was pinned at the beginning of the current map call.
 	 */
-	nvgpu_mm_unpin_privdata(os_buf->dev, os_buf->dmabuf,
+	nvgpu_mm_unpin(os_buf->dev, os_buf->dmabuf,
 		       mapped_buffer->os_priv.attachment,
 		       mapped_buffer->os_priv.sgt);
 	dma_buf_put(os_buf->dmabuf);
@@ -205,7 +205,7 @@ int nvgpu_vm_map_linux(struct vm_gk20a *vm,
 	struct dma_buf_attachment *attachment;
 	int err = 0;
 
-	sgt = nvgpu_mm_pin_privdata(dev, dmabuf, &attachment);
+	sgt = nvgpu_mm_pin(dev, dmabuf, &attachment);
 	if (IS_ERR(sgt)) {
 		nvgpu_warn(g, "Failed to pin dma_buf!");
 		return PTR_ERR(sgt);
@@ -257,7 +257,7 @@ int nvgpu_vm_map_linux(struct vm_gk20a *vm,
 	return 0;
 
 clean_up:
-	nvgpu_mm_unpin_privdata(dev, dmabuf, attachment, sgt);
+	nvgpu_mm_unpin(dev, dmabuf, attachment, sgt);
 
 	return err;
 }
@@ -324,12 +324,6 @@ int nvgpu_vm_map_buffer(struct vm_gk20a *vm,
 		return -EINVAL;
 	}
 
-	err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev_from_vm(vm));
-	if (err) {
-		dma_buf_put(dmabuf);
-		return err;
-	}
-
 	err = nvgpu_vm_map_linux(vm, dmabuf, *map_addr,
 				 nvgpu_vm_translate_linux_flags(g, flags),
 				 page_size,
@@ -358,7 +352,7 @@ void nvgpu_vm_unmap_system(struct nvgpu_mapped_buf *mapped_buffer)
 {
 	struct vm_gk20a *vm = mapped_buffer->vm;
 
-	nvgpu_mm_unpin_privdata(dev_from_vm(vm), mapped_buffer->os_priv.dmabuf,
+	nvgpu_mm_unpin(dev_from_vm(vm), mapped_buffer->os_priv.dmabuf,
 			mapped_buffer->os_priv.attachment,
 			mapped_buffer->os_priv.sgt);