ARM: dma-mapping: fix atomic allocation alignment

Commit Message

Aaro Koskinen Aug. 5, 2012, 5:32 p.m. UTC

The alignment mask is calculated incorrectly. Fixing the calculation
makes strange hangs/lockups disappear during the boot with Amstrad E3
and 3.6-rc1 kernel.

Signed-off-by: Aaro Koskinen <[e-mail]>
---
 arch/arm/mm/dma-mapping.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

Comments

Marek Szyprowski Aug. 7, 2012, 1:22 p.m. UTC | #1

Hi Aaro,

On Sunday, August 05, 2012 7:32 PM Aaro Koskinen wrote:

> The alignment mask is calculated incorrectly. Fixing the calculation
> makes strange hangs/lockups disappear during the boot with Amstrad E3
> and 3.6-rc1 kernel.
> 
> Signed-off-by: Aaro Koskinen <[e-mail]>

Again, thanks for spotting and fixing the issue. I've applied it to my fixes branch. 
I'm really sorry for introducing such stupid bugs together with my changes.

Best regards

Patch

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 2cc77b7..0e0466d 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -423,7 +423,7 @@  static void *__alloc_from_pool(size_t size, struct page **ret_page)
 	unsigned int pageno;
 	unsigned long flags;
 	void *ptr = NULL;
-	size_t align;
+	unsigned long align_mask;
 
 	if (!pool->vaddr) {
 		WARN(1, "coherent pool not initialised!\n");
@@ -435,11 +435,11 @@  static void *__alloc_from_pool(size_t size, struct page **ret_page)
 	 * small, so align them to their order in pages, minimum is a page
 	 * size. This helps reduce fragmentation of the DMA space.
 	 */
-	align = PAGE_SIZE << get_order(size);
+	align_mask = (1 << get_order(size)) - 1;
 
 	spin_lock_irqsave(&pool->lock, flags);
 	pageno = bitmap_find_next_zero_area(pool->bitmap, pool->nr_pages,
-					    0, count, (1 << align) - 1);
+					    0, count, align_mask);
 	if (pageno < pool->nr_pages) {
 		bitmap_set(pool->bitmap, pageno, count);
 		ptr = pool->vaddr + PAGE_SIZE * pageno;