Skip to contentSkip to conversation

ARM/dma-mapping: use kvcalloc for fallback memory allocation need

PATCH Superseded Deepak R Varma 1 participant 2 messages Jan 3, 2023 – Jan 22, 2023 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 ARM/dma-mapping: use kvcalloc for fallback memory allocation need

Current conditional determination of whether to use kzalloc or vzalloc has known issues such as "indefinite retry" when less than PAGE_SIZE memory is needed, but is unavailable. This LWN article [1] describes these issues in greater detail. Use helper function kvcalloc() instead which is more efficient in terms of performance and security.

[1] https://lwn.net/Articles/711653/

This patch proposal is based on following Coccinelle warning using the kvmalloc.cocci semantic patch.
arch/arm/mm/dma-mapping.c:858:28-29: WARNING opportunity for kvmalloc

The semantic patch suggests using kvzalloc() helper function, however, this patch proposes to use kvcalloc instead. kvcalloc() helper function uses 2-factor argument form which is better from a security perspective as described in the following KSPP project commit.

Commit 4e3fd7217105 ("wireguard: ratelimiter: use kvcalloc() instead of kvzalloc()")

---
Changes in v3:
1. Use updated maintainer list to send the patch. Suggested by "Russell King (Oracle)" <[email protected]>

Changes in v2:
1. Update patch subject to use kvcalloc
2. Use kvcalloc instead of kvzalloc helper function. Revise the patch proposal and the patch description accordingly.

 arch/arm/mm/dma-mapping.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
arch/arm/mm/dma-mapping.c +1 -5
index c135f6e37a00..35092ecd30e1 100644
@@ -851,14 +851,10 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 {
 	struct page **pages;
 	int count = size >> PAGE_SHIFT;
-	int array_size = count * sizeof(struct page *);
 	int i = 0;
 	int order_idx = 0;
 
-	if (array_size <= PAGE_SIZE)
-		pages = kzalloc(array_size, GFP_KERNEL);
-	else
-		pages = vzalloc(array_size);
+	pages = kvcalloc(count, sizeof(struct page *), GFP_KERNEL);
 	if (!pages)
 		return NULL;
 

--
2.34.1

#1 lore raw
Quoted 17 lines · Deepak R Varma

Current conditional determination of whether to use kzalloc or vzalloc has known issues such as "indefinite retry" when less than PAGE_SIZE memory is needed, but is unavailable. This LWN article [1] describes these issues in greater detail. Use helper function kvcalloc() instead which is more efficient in terms of performance and security.

[1] https://lwn.net/Articles/711653/

This patch proposal is based on following Coccinelle warning using the kvmalloc.cocci semantic patch.
arch/arm/mm/dma-mapping.c:858:28-29: WARNING opportunity for kvmalloc

The semantic patch suggests using kvzalloc() helper function, however, this patch proposes to use kvcalloc instead. kvcalloc() helper function uses 2-factor argument form which is better from a security perspective as described in the following KSPP project commit.

Quoted 7 lines

Commit 4e3fd7217105 ("wireguard: ratelimiter: use kvcalloc() instead of kvzalloc()")

---
Changes in v3:
1. Use updated maintainer list to send the patch. Suggested by "Russell King (Oracle)" <[email protected]>

Hello,
May I request a review and feedback comment on this patch proposal please?

Thank you,
./drv

Quoted 32 lines

Changes in v2:
1. Update patch subject to use kvcalloc
2. Use kvcalloc instead of kvzalloc helper function. Revise the patch proposal and the patch description accordingly.

 arch/arm/mm/dma-mapping.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
arch/arm/mm/dma-mapping.c +1 -6
index c135f6e37a00..35092ecd30e1 100644
@@ -851,14 +851,10 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 {
 	struct page **pages;
 	int count = size >> PAGE_SHIFT;
-	int array_size = count * sizeof(struct page *);
 	int i = 0;
 	int order_idx = 0;
 
-	if (array_size <= PAGE_SIZE)
-		pages = kzalloc(array_size, GFP_KERNEL);
-	else
-		pages = vzalloc(array_size);
+	pages = kvcalloc(count, sizeof(struct page *), GFP_KERNEL);
 	if (!pages)
 		return NULL;
 
--

2.34.1