Skip to contentSkip to conversation

dma-contiguous: two bug fixes for dma_{alloc,free}_contiguous()

PATCH Superseded Nicolin Chen 2 participants 6 messages Jul 25, 2019 – Jul 26, 2019 DMA mapping helpers

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover dma-contiguous: two bug fixes for dma_{alloc,free}_contiguous()

There are two obvious bugs in these two functions. So having two patches to fix them.

Nicolin Chen (2):
dma-contiguous: do not overwrite align in dma_alloc_contiguous()
dma-contiguous: page-align the size in dma_free_contiguous()

 kernel/dma/contiguous.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
Signature · 6 lines
2.17.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
#1 lore raw

PATCH 1/2 dma-contiguous: do not overwrite align in dma_alloc_contiguous()

The dma_alloc_contiguous() limits align at CONFIG_CMA_ALIGNMENT for cma_alloc() however it does not restore it for the fallback routine. This will result in a size mismatch between the allocation and free when running in the fallback routines, if the align is larger than CONFIG_CMA_ALIGNMENT.

This patch adds a cma_align to take care of cma_alloc() and prevent the align from being overwritten.

Fixes fdaeec198ada ("dma-contiguous: add dma_{alloc,free}_contiguous() helpers")
---
 kernel/dma/contiguous.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
kernel/dma/contiguous.c +5 -4
index bfc0c17f2a3d..fa8cd0f0512e 100644
@@ -233,6 +233,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	int node = dev ? dev_to_node(dev) : NUMA_NO_NODE;
 	size_t count = PAGE_ALIGN(size) >> PAGE_SHIFT;
 	size_t align = get_order(PAGE_ALIGN(size));
+	size_t cma_align = CONFIG_CMA_ALIGNMENT;
 	struct page *page = NULL;
 	struct cma *cma = NULL;
 
@@ -241,11 +242,11 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	else if (count > 1)
 		cma = dma_contiguous_default_area;
 
+	cma_align = min_t(size_t, align, cma_align);
+
 	/* CMA can be used only in the context which permits sleeping */
-	if (cma && gfpflags_allow_blocking(gfp)) {
-		align = min_t(size_t, align, CONFIG_CMA_ALIGNMENT);
-		page = cma_alloc(cma, count, align, gfp & __GFP_NOWARN);
-	}
+	if (cma && gfpflags_allow_blocking(gfp))
+		page = cma_alloc(cma, count, cma_align, gfp & __GFP_NOWARN);
 
 	/* Fallback allocation of normal pages */
 	if (!page)
Signature · 6 lines
2.17.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
#2 lore raw

Re: [PATCH 1/2] dma-contiguous: do not overwrite align in dma_alloc_contiguous()

Quoted 31 lines · Nicolin Chen

The dma_alloc_contiguous() limits align at CONFIG_CMA_ALIGNMENT for cma_alloc() however it does not restore it for the fallback routine. This will result in a size mismatch between the allocation and free when running in the fallback routines, if the align is larger than CONFIG_CMA_ALIGNMENT.

This patch adds a cma_align to take care of cma_alloc() and prevent the align from being overwritten.

Fixes fdaeec198ada ("dma-contiguous: add dma_{alloc,free}_contiguous() helpers")
---
 kernel/dma/contiguous.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
kernel/dma/contiguous.c +1 -0
index bfc0c17f2a3d..fa8cd0f0512e 100644
@@ -233,6 +233,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	int node = dev ? dev_to_node(dev) : NUMA_NO_NODE;
 	size_t count = PAGE_ALIGN(size) >> PAGE_SHIFT;
 	size_t align = get_order(PAGE_ALIGN(size));
+	size_t cma_align = CONFIG_CMA_ALIGNMENT;
 	struct page *page = NULL;
 	struct cma *cma = NULL;
 
@@ -241,11 +242,11 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	else if (count > 1)
 		cma = dma_contiguous_default_area;

kernel/dma/contiguous.c struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) · line 249

Quoted 10 lines
kernel/dma/contiguous.c +4 -4
 
+	cma_align = min_t(size_t, align, cma_align);
+
 	/* CMA can be used only in the context which permits sleeping */
-	if (cma && gfpflags_allow_blocking(gfp)) {
-		align = min_t(size_t, align, CONFIG_CMA_ALIGNMENT);
-		page = cma_alloc(cma, count, align, gfp & __GFP_NOWARN);
-	}
+	if (cma && gfpflags_allow_blocking(gfp))
+		page = cma_alloc(cma, count, cma_align, gfp & __GFP_NOWARN);

Shouldn't cma_align be confined to the block guarded by "if (cma && gfpflags_allow_blocking(gfp))" so that we can optimize it away for configurations that do not support CMA?

#3 lore raw

Re: [PATCH 1/2] dma-contiguous: do not overwrite align in dma_alloc_contiguous()

Quoted 33 lines · Christoph Hellwig
Quoted 32 lines · Nicolin Chen

The dma_alloc_contiguous() limits align at CONFIG_CMA_ALIGNMENT for cma_alloc() however it does not restore it for the fallback routine. This will result in a size mismatch between the allocation and free when running in the fallback routines, if the align is larger than CONFIG_CMA_ALIGNMENT.

This patch adds a cma_align to take care of cma_alloc() and prevent the align from being overwritten.

Fixes fdaeec198ada ("dma-contiguous: add dma_{alloc,free}_contiguous() helpers")
---
 kernel/dma/contiguous.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
kernel/dma/contiguous.c +1 -0
index bfc0c17f2a3d..fa8cd0f0512e 100644
@@ -233,6 +233,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	int node = dev ? dev_to_node(dev) : NUMA_NO_NODE;
 	size_t count = PAGE_ALIGN(size) >> PAGE_SHIFT;
 	size_t align = get_order(PAGE_ALIGN(size));
+	size_t cma_align = CONFIG_CMA_ALIGNMENT;
 	struct page *page = NULL;
 	struct cma *cma = NULL;
 
@@ -241,11 +242,11 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
 	else if (count > 1)
 		cma = dma_contiguous_default_area;
 
Quoted 13 lines

kernel/dma/contiguous.c struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) · line 249

Quoted 9 lines
kernel/dma/contiguous.c +4 -4
+	cma_align = min_t(size_t, align, cma_align);
+
 	/* CMA can be used only in the context which permits sleeping */
-	if (cma && gfpflags_allow_blocking(gfp)) {
-		align = min_t(size_t, align, CONFIG_CMA_ALIGNMENT);
-		page = cma_alloc(cma, count, align, gfp & __GFP_NOWARN);
-	}
+	if (cma && gfpflags_allow_blocking(gfp))
+		page = cma_alloc(cma, count, cma_align, gfp & __GFP_NOWARN);

Shouldn't cma_align be confined to the block guarded by "if (cma && gfpflags_allow_blocking(gfp))" so that we can optimize it away for configurations that do not support CMA?

Had my local 1st version doing just like that but then wanted to simplify the statement within that if-condition so redid in this way. Will change it back as you suggested. Thanks
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#4 lore raw

PATCH 2/2 dma-contiguous: page-align the size in dma_free_contiguous()

In reply to Nicolin Chen

According to the original dma_direct_alloc_pages() code:

{
	unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;

	if (!dma_release_from_contiguous(dev, page, count))
		__free_pages(page, get_order(size));
}

The count parameter for dma_release_from_contiguous() was page aligned before the right-shifting operation, while the new API dma_free_contiguous() forgets to have PAGE_ALIGN() at the size.

So this patch simply adds it to prevent any corner case.

Fixes fdaeec198ada ("dma-contiguous: add dma_{alloc,free}_contiguous() helpers")
---
 kernel/dma/contiguous.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/dma/contiguous.c +2 -1
index fa8cd0f0512e..5735a9166866 100644
@@ -267,7 +267,8 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp)
  */
 void dma_free_contiguous(struct device *dev, struct page *page, size_t size)
 {
-	if (!cma_release(dev_get_cma_area(dev), page, size >> PAGE_SHIFT))
+	if (!cma_release(dev_get_cma_area(dev), page,
+			 PAGE_ALIGN(size) >> PAGE_SHIFT))
 		__free_pages(page, get_order(size));
 }
 
Signature · 6 lines
2.17.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Reviewed-by on v1 2/2
#5 lore raw

Re: [PATCH 2/2] dma-contiguous: page-align the size in dma_free_contiguous()

Quoted 8 lines · Nicolin Chen

According to the original dma_direct_alloc_pages() code:

{
	unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT;

	if (!dma_release_from_contiguous(dev, page, count))
		__free_pages(page, get_order(size));
}
Quoted 8 lines

The count parameter for dma_release_from_contiguous() was page aligned before the right-shifting operation, while the new API dma_free_contiguous() forgets to have PAGE_ALIGN() at the size.

So this patch simply adds it to prevent any corner case.

Fixes fdaeec198ada ("dma-contiguous: add dma_{alloc,free}_contiguous() helpers")

Looks good,

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu