HACK: staging: ion: Try to cache mappings to avoid cache invalidations

Since 4.12, much later narrowed down to commit 2a55e7b5e544
("staging: android: ion: Call dma_map_sg for syncing and mapping"),
we have seen graphics performance issues on the HiKey960.

After chasing a number of other leads, I found that reverting
the ION code to 4.11-era got the majority of the graphics
performance back (there may yet be further EAS tweaks needed),
which lead me to the dma_map_sg change.

In talking w/ Laura and Liam, it was suspected that the extra
cache operations were causing the trouble. Additionally, I found
that part of the reason we didn't see this w/ the original
HiKey board is that its (proprietary blob) GL code uses ion_mmap
and ion_map_dma_buf is called very rarely, where as with
HiKey960, the (also proprietary blob) GL code calls
ion_map_dma_buf much more frequently via the kernel driver.

Anyway, with the cause of the performance regression isolated,
I've tried to find a way to improve the performance of the
current code.

This approach, which reworks my previous attempt, tries to
use the dma-attributes to avoid doing cache operations if
the current direction is the same.

NOTE: Technically, this probably isn't the right solution for
every ION user, thus the HACK prefix. However, for HiKey960 it
resolves a pretty big interactive performance issue, and I
think we should consider merging it so we can switch to 4.14
as the default kernel in the near term. Once a proper fix is
found upstream, we can backport it.

Change-Id: Ic7ce3dde5efe4c4dc7a1f3cdf138f0c572a2ba35
Signed-off-by: John Stultz <john.stultz@linaro.org>

1 files changed, 20 insertions, 4 deletions

diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
index 24cb666c9224..eea939be6306 100644
--- a/drivers/staging/android/ion/ion.c
+++ b/drivers/staging/android/ion/ion.c
@@ -219,6 +219,7 @@ struct ion_dma_buf_attachment {
 	struct device *dev;
 	struct sg_table *table;
 	struct list_head list;
+	enum dma_data_direction dir;
 };
 
 static int ion_dma_buf_attach(struct dma_buf *dmabuf, struct device *dev,
@@ -240,6 +241,7 @@ static int ion_dma_buf_attach(struct dma_buf *dmabuf, struct device *dev,
 
 	a->table = table;
 	a->dev = dev;
+	a->dir = DMA_NONE;
 	INIT_LIST_HEAD(&a->list);
 
 	attachment->priv = a;
@@ -270,12 +272,19 @@ static struct sg_table *ion_map_dma_buf(struct dma_buf_attachment *attachment,
 {
 	struct ion_dma_buf_attachment *a = attachment->priv;
 	struct sg_table *table;
+	unsigned long dma_attr = 0;
 
 	table = a->table;
 
-	if (!dma_map_sg(attachment->dev, table->sgl, table->nents,
-			direction))
-		return ERR_PTR(-ENOMEM);
+	if (a->dir == direction)
+		dma_attr = DMA_ATTR_SKIP_CPU_SYNC;
+
+	if (!dma_map_sg_attrs(attachment->dev, table->sgl, table->nents,
+			      direction,  dma_attr)) {
+		table = ERR_PTR(-ENOMEM);
+	} else {
+		a->dir = direction;
+	}
 
 	return table;
 }
@@ -284,7 +293,14 @@ static void ion_unmap_dma_buf(struct dma_buf_attachment *attachment,
 			      struct sg_table *table,
 			      enum dma_data_direction direction)
 {
-	dma_unmap_sg(attachment->dev, table->sgl, table->nents, direction);
+	struct ion_dma_buf_attachment *a = attachment->priv;
+	unsigned long dma_attr = 0;
+
+	if (direction == a->dir)
+		dma_attr = DMA_ATTR_SKIP_CPU_SYNC;
+
+	dma_unmap_sg_attrs(attachment->dev, table->sgl, table->nents,
+			   direction, dma_attr);
 }
 
 static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)