Thread (5 messages) 5 messages, 3 authors, 2016-03-22

[PATCH] arm64/dma-mapping: Add DMA_ATTR_ALLOC_SINGLE_PAGES support

From: Will Deacon <hidden>
Date: 2016-03-21 18:01:30
Also in: linux-iommu, linux-mediatek, lkml

On Thu, Mar 03, 2016 at 02:54:26AM +0800, Yong Wu wrote:
quoted hunk ↗ jump to hunk
Sometimes it is not worth for the iommu allocating big chunks.
Here we enable DMA_ATTR_ALLOC_SINGLE_PAGES which could help avoid to
allocate big chunks while iommu allocating buffer.

More information about this attribute, please check Doug's commit[1].

[1]: https://lkml.org/lkml/2016/1/11/720

Cc: Robin Murphy <robin.murphy@arm.com>
Suggested-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Yong Wu <yong.wu@mediatek.com>
---

Our video drivers may soon use this.

 arch/arm64/mm/dma-mapping.c |  4 ++--
 drivers/iommu/dma-iommu.c   | 14 ++++++++++----
 include/linux/dma-iommu.h   |  4 ++--
 3 files changed, 14 insertions(+), 8 deletions(-)
diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
index 331c4ca..3225e3ca 100644
--- a/arch/arm64/mm/dma-mapping.c
+++ b/arch/arm64/mm/dma-mapping.c
@@ -562,8 +562,8 @@ static void *__iommu_alloc_attrs(struct device *dev, size_t size,
 		struct page **pages;
 		pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, coherent);
 
-		pages = iommu_dma_alloc(dev, iosize, gfp, ioprot, handle,
-					flush_page);
+		pages = iommu_dma_alloc(dev, iosize, gfp, ioprot, attrs,
+					handle, flush_page);
 		if (!pages)
 			return NULL;
 
diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 72d6182..3569cb6 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -190,7 +190,8 @@ static void __iommu_dma_free_pages(struct page **pages, int count)
 	kvfree(pages);
 }
 
-static struct page **__iommu_dma_alloc_pages(unsigned int count, gfp_t gfp)
+static struct page **__iommu_dma_alloc_pages(unsigned int count, gfp_t gfp,
+					     struct dma_attrs *attrs)
 {
 	struct page **pages;
 	unsigned int i = 0, array_size = count * sizeof(*pages);
@@ -203,6 +204,10 @@ static struct page **__iommu_dma_alloc_pages(unsigned int count, gfp_t gfp)
 	if (!pages)
 		return NULL;
 
+	/* Go straight to 4K chunks if caller says it's OK. */
+	if (dma_get_attr(DMA_ATTR_ALLOC_SINGLE_PAGES, attrs))
+		order = 0;
I have a slight snag with this, in that you don't consult the IOMMU
pgsize_bitmap at any point, and assume that it can map pages at the
same granularity as the CPU. The documentation for
DMA_ATTR_ALLOC_SINGLE_PAGES seems to be weaker than that.

Will
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help