scsi: use dma_get_cache_alignment() as minimum DMA alignment

commit 90addc6b3c9cda0146fbd62a08e234c2b224a80c upstream.

In non-coherent DMA mode, kernel uses cache flushing operations to
maintain I/O coherency, so scsi's block queue should be aligned to the
value returned by dma_get_cache_alignment().  Otherwise, If a DMA buffer
and a kernel structure share a same cache line, and if the kernel
structure has dirty data, cache_invalidate (no writeback) will cause
data corruption.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
[hch: rebased and updated the comment and changelog]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index f6a464a..823f0e9 100644 (file)
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1688,11 +1688,13 @@ struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
                q->limits.cluster = 0;
 
        /*
-        * set a reasonable default alignment on word boundaries: the
-        * host and device may alter it using
-        * blk_queue_update_dma_alignment() later.
+        * Set a reasonable default alignment:  The larger of 32-byte (dword),
+        * which is a common minimum for HBAs, and the minimum DMA alignment,
+        * which is set by the platform.
+        *
+        * Devices that require a bigger alignment can increase it later.
         */
-       blk_queue_dma_alignment(q, 0x03);
+       blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
 
        return q;
 }