diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
index a29e694..cd8c7c2 100644
--- a/drivers/block/cciss.c
+++ b/drivers/block/cciss.c
@@ -4496,7 +4496,7 @@
 	 * boundary. Given that we use pci_alloc_consistent() to allocate an
 	 * array of them, the size must be a multiple of 8 bytes.
 	 */
-	BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
+	BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
 
 	printk(KERN_INFO DRIVER_NAME "\n");
 
diff --git a/drivers/block/cciss_cmd.h b/drivers/block/cciss_cmd.h
index 515c9f0..e624ff9 100644
--- a/drivers/block/cciss_cmd.h
+++ b/drivers/block/cciss_cmd.h
@@ -167,10 +167,13 @@
 #define CMD_MSG_TIMEOUT 0x05
 #define CMD_MSG_STALE	0xff
 
-/* This structure needs to be divisible by 8 for new
- * indexing method.  PAD_32 and PAD_64 can be adjusted
- * independently as needed for 32-bit and 64-bits systems.
+/* This structure needs to be divisible by COMMANDLIST_ALIGNMENT
+ * because low bits of the address are used to to indicate that
+ * whether the tag contains an index or an address.  PAD_32 and
+ * PAD_64 can be adjusted independently as needed for 32-bit
+ * and 64-bits systems.
  */
+#define COMMANDLIST_ALIGNMENT (8)
 #define IS_64_BIT ((sizeof(long) - 4)/4)
 #define IS_32_BIT (!IS_64_BIT)
 #define PAD_32 (0)
diff --git a/drivers/block/cciss_scsi.c b/drivers/block/cciss_scsi.c
index 5d0e46d..f203606 100644
--- a/drivers/block/cciss_scsi.c
+++ b/drivers/block/cciss_scsi.c
@@ -93,11 +93,15 @@
 };
 
 #pragma pack(1)
+
+#define SCSI_PAD_32 4
+#define SCSI_PAD_64 4
+
 struct cciss_scsi_cmd_stack_elem_t {
 	CommandList_struct cmd;
 	ErrorInfo_struct Err;
 	__u32 busaddr;
-	__u32 pad;
+	u8 pad[IS_32_BIT * SCSI_PAD_32 + IS_64_BIT * SCSI_PAD_64];
 };
 
 #pragma pack()
@@ -202,9 +206,9 @@
 	stk = &sa->cmd_stack; 
 	size = sizeof(struct cciss_scsi_cmd_stack_elem_t) * CMD_STACK_SIZE;
 
-	// pci_alloc_consistent guarantees 32-bit DMA address will
-	// be used
-
+	/* Check alignment, see cciss_cmd.h near CommandList_struct def. */
+	BUILD_BUG_ON((sizeof(*stk->pool) % COMMANDLIST_ALIGNMENT) != 0);
+	/* pci_alloc_consistent guarantees 32-bit DMA address will be used */
 	stk->pool = (struct cciss_scsi_cmd_stack_elem_t *)
 		pci_alloc_consistent(hba[ctlr]->pdev, size, &stk->cmd_pool_handle);