[PATCH v2 0/2] [media] videobuf2-dc: Add support for cacheable MMAP

[PATCH v2 0/2] [media] videobuf2-dc: Add support for cacheable MMAP

[Thierry Escande]

Hi,

This series adds support for cacheable MMAP in DMA coherent allocator.

The first patch moves the vb2_dc_get_base_sgt() function above mmap
callbacks for calls introduced by the second patch. This avoids a
forward declaration.

Changes in v2:
- Put function move in a separate patch
- Added comments

Heng-Ruey Hsu (1):
  [media] videobuf2-dc: Support cacheable MMAP

Thierry Escande (1):
  [media] videobuf2-dc: Move vb2_dc_get_base_sgt() above mmap callbacks

 drivers/media/v4l2-core/videobuf2-dma-contig.c | 62 +++++++++++++++++---------
 1 file changed, 40 insertions(+), 22 deletions(-)

-- 
2.7.4

[PATCH v2 1/2] [media] videobuf2-dc: Move vb2_dc_get_base_sgt() above mmap callbacks

[Thierry Escande]

This patch moves vb2_dc_get_base_sgt() function above mmap buffers
callbacks, particularly vb2_dc_alloc() and vb2_dc_mmap() from where it
will be called for cacheable MMAP support introduced in the next patch.

Signed-off-by: Thierry Escande <thierry.escande@collabora.com>
---
 drivers/media/v4l2-core/videobuf2-dma-contig.c | 44 +++++++++++++-------------
 1 file changed, 22 insertions(+), 22 deletions(-)

diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c b/drivers/media/v4l2-core/videobuf2-dma-contig.c
index a44e383..0d9665d 100644
--- a/drivers/media/v4l2-core/videobuf2-dma-contig.c
+++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c
@@ -61,6 +61,28 @@ static unsigned long vb2_dc_get_contiguous_size(struct sg_table *sgt)
 	return size;
 }
 
+static struct sg_table *vb2_dc_get_base_sgt(struct vb2_dc_buf *buf)
+{
+	int ret;
+	struct sg_table *sgt;
+
+	sgt = kmalloc(sizeof(*sgt), GFP_KERNEL);
+	if (!sgt) {
+		dev_err(buf->dev, "failed to alloc sg table\n");
+		return NULL;
+	}
+
+	ret = dma_get_sgtable_attrs(buf->dev, sgt, buf->cookie, buf->dma_addr,
+		buf->size, buf->attrs);
+	if (ret < 0) {
+		dev_err(buf->dev, "failed to get scatterlist from DMA API\n");
+		kfree(sgt);
+		return NULL;
+	}
+
+	return sgt;
+}
+
 /*********************************************/
 /*         callbacks for all buffers         */
 /*********************************************/
@@ -363,28 +385,6 @@ static struct dma_buf_ops vb2_dc_dmabuf_ops = {
 	.release = vb2_dc_dmabuf_ops_release,
 };
 
-static struct sg_table *vb2_dc_get_base_sgt(struct vb2_dc_buf *buf)
-{
-	int ret;
-	struct sg_table *sgt;
-
-	sgt = kmalloc(sizeof(*sgt), GFP_KERNEL);
-	if (!sgt) {
-		dev_err(buf->dev, "failed to alloc sg table\n");
-		return NULL;
-	}
-
-	ret = dma_get_sgtable_attrs(buf->dev, sgt, buf->cookie, buf->dma_addr,
-		buf->size, buf->attrs);
-	if (ret < 0) {
-		dev_err(buf->dev, "failed to get scatterlist from DMA API\n");
-		kfree(sgt);
-		return NULL;
-	}
-
-	return sgt;
-}
-
 static struct dma_buf *vb2_dc_get_dmabuf(void *buf_priv, unsigned long flags)
 {
 	struct vb2_dc_buf *buf = buf_priv;
-- 
2.7.4

[PATCH v2 2/2] [media] videobuf2-dc: Support cacheable MMAP

[Thierry Escande]

From: Heng-Ruey Hsu <henryhsu@chromium.org>

DMA allocations for MMAP type are uncached by default. But for
some cases, CPU has to access the buffers. ie: memcpy for format
converter. Supporting cacheable MMAP improves huge performance.

This patch enables cacheable memory for DMA coherent allocator in mmap
buffer allocation if non-consistent DMA attribute is set and kernel
mapping is present. Even if userspace doesn't mmap the buffer, sync
still should be happening if kernel mapping is present.
If not done in allocation, it is enabled when memory is mapped from
userspace when no kernel mapping is present and non-consistent DMA
attribute set.

Signed-off-by: Heng-Ruey Hsu <henryhsu@chromium.org>
Tested-by: Heng-ruey Hsu <henryhsu@chromium.org>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Thierry Escande <thierry.escande@collabora.com>
---
 drivers/media/v4l2-core/videobuf2-dma-contig.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c b/drivers/media/v4l2-core/videobuf2-dma-contig.c
index 0d9665d..1f7649d 100644
--- a/drivers/media/v4l2-core/videobuf2-dma-contig.c
+++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c
@@ -151,6 +151,10 @@ static void vb2_dc_put(void *buf_priv)
 		sg_free_table(buf->sgt_base);
 		kfree(buf->sgt_base);
 	}
+	if (buf->dma_sgt) {
+		sg_free_table(buf->dma_sgt);
+		kfree(buf->dma_sgt);
+	}
 	dma_free_attrs(buf->dev, buf->size, buf->cookie, buf->dma_addr,
 		       buf->attrs);
 	put_device(buf->dev);
@@ -192,6 +196,14 @@ static void *vb2_dc_alloc(struct device *dev, unsigned long attrs,
 	buf->handler.put = vb2_dc_put;
 	buf->handler.arg = buf;
 
+	/*
+	 * Enable cacheable memory. Even if userspace doesn't mmap the buffer,
+	 * sync still should be happening if kernel mapping is present.
+	 */
+	if (!(buf->attrs & DMA_ATTR_NO_KERNEL_MAPPING) &&
+	    buf->attrs & DMA_ATTR_NON_CONSISTENT)
+		buf->dma_sgt = vb2_dc_get_base_sgt(buf);
+
 	atomic_inc(&buf->refcount);
 
 	return buf;
@@ -227,6 +239,12 @@ static int vb2_dc_mmap(void *buf_priv, struct vm_area_struct *vma)
 
 	vma->vm_ops->open(vma);
 
+	/* Enable cacheable memory if not enabled in allocation. */
+	if (!buf->dma_sgt &&
+	    buf->attrs & DMA_ATTR_NO_KERNEL_MAPPING &&
+	    buf->attrs & DMA_ATTR_NON_CONSISTENT)
+		buf->dma_sgt = vb2_dc_get_base_sgt(buf);
+
 	pr_debug("%s: mapped dma addr 0x%08lx at 0x%08lx, size %ld\n",
 		__func__, (unsigned long)buf->dma_addr, vma->vm_start,
 		buf->size);
-- 
2.7.4