Skip to contentSkip to conversation

ARM: DMA: Fix kzalloc flags in __iommu_alloc_buffer()

PATCH RESEND In review Alexandre Courbot 3 participants 3 messages Feb 11, 2015 – Feb 13, 2015 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 ARM: DMA: Fix kzalloc flags in __iommu_alloc_buffer()

There doesn't seem to be any valid reason to allocate the pages array with the same flags as the buffer itself. Doing so can eventually lead to the following safeguard in mm/slab.c to be hit:

BUG_ON(flags & GFP_SLAB_BUG_MASK);

This happens when buffers are allocated with __GFP_DMA32 or __GFP_HIGHMEM.

Fix this by allocating the pages array with GFP_KERNEL to follow what is done elsewhere in this file. Using GFP_KERNEL in __iommu_alloc_buffer() is safe because atomic allocations are handled by __iommu_alloc_atomic().

---
 arch/arm/mm/dma-mapping.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mm/dma-mapping.c +1 -1
index 903dba0..170a116 100644
@@ -1106,7 +1106,7 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 	int i = 0;
 
 	if (array_size <= PAGE_SIZE)
-		pages = kzalloc(array_size, gfp);
+		pages = kzalloc(array_size, GFP_KERNEL);
 	else
 		pages = vzalloc(array_size);
 	if (!pages)
Signature · 1 line
2.3.0
Acked-by on v1 1/1
#1 lore raw
Quoted 5 lines · Alexandre Courbot

There doesn't seem to be any valid reason to allocate the pages array with the same flags as the buffer itself. Doing so can eventually lead to the following safeguard in mm/slab.c to be hit:

BUG_ON(flags & GFP_SLAB_BUG_MASK);

nit: I can't actually spot this BUG_ON in the kernel.

Quoted 31 lines

This happens when buffers are allocated with __GFP_DMA32 or __GFP_HIGHMEM.

Fix this by allocating the pages array with GFP_KERNEL to follow what is done elsewhere in this file. Using GFP_KERNEL in __iommu_alloc_buffer() is safe because atomic allocations are handled by __iommu_alloc_atomic().

---
 arch/arm/mm/dma-mapping.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mm/dma-mapping.c +1 -1
index 903dba0..170a116 100644
@@ -1106,7 +1106,7 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
 	int i = 0;
 
 	if (array_size <= PAGE_SIZE)
-		pages = kzalloc(array_size, gfp);
+		pages = kzalloc(array_size, GFP_KERNEL);
 	else
 		pages = vzalloc(array_size);
 	if (!pages)
Signature · 1 line
2.3.0

Looks sensible to me:

Will

#2 lore raw
Quoted 8 lines · Will Deacon
Quoted 5 lines · Alexandre Courbot

There doesn't seem to be any valid reason to allocate the pages array with the same flags as the buffer itself. Doing so can eventually lead to the following safeguard in mm/slab.c to be hit:

BUG_ON(flags & GFP_SLAB_BUG_MASK);

nit: I can't actually spot this BUG_ON in the kernel.

I have been trying to push this patch for so long that the line in question changed in the meantime. :) It is now

         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
                 pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK);
                 BUG();
         }

in cache_grow, line 2593 of mm/slab.c.

Quoted 35 lines
Quoted 31 lines

This happens when buffers are allocated with __GFP_DMA32 or __GFP_HIGHMEM.

Fix this by allocating the pages array with GFP_KERNEL to follow what is done elsewhere in this file. Using GFP_KERNEL in __iommu_alloc_buffer() is safe because atomic allocations are handled by __iommu_alloc_atomic().

---
  arch/arm/mm/dma-mapping.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mm/dma-mapping.c +1 -2
index 903dba0..170a116 100644
@@ -1106,7 +1106,7 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size,
  	int i = 0;
 
  	if (array_size <= PAGE_SIZE)
-		pages = kzalloc(array_size, gfp);
+		pages = kzalloc(array_size, GFP_KERNEL);
  	else
  		pages = vzalloc(array_size);
  	if (!pages)
--
Quoted 1 line

2.3.0

Looks sensible to me:

Quoted 1 line

Thanks! I will amend the commit message and resend.