Re: [CRYPTO] padlock: Fix alignment after aes_ctx rearrange

[Michal Ludvig]

Michal Ludvig wrote:
> Hi Herbert,
> 
> I just recently discovered that your patch that rearranges struct
> aes_ctx in padlock-aes.c breaks the alignment rules for xcrypt leading
> to GPF Oopses.
> 
> Note that *all* addresses passed to xcrypt must be 16-Bytes aligned for
> VIA C3 (including IV and Key - the latter one was not aligned and
> triggered this Oops).
> 
> As the rearrange patch made it to 2.6.18-rc1 it must be fixed before
> 2.6.18 is out. Attached is a patch.

Ehrm, ... now it is attached ;-)

Michal
Subject: padlock: Fix alignment after aes_ctx rearrange

Herbert's patch 82062c72cd643c99a9e1c231270acbab986fd23f 
in cryptodev-2.6 tree breaks alignment rules for PadLock 
xcrypt instruction leading to General protection Oopses.

This patch fixes the problem.

Signed-off-by: Michal Ludvig <michal@xxxxxxxx>

Index: linux-2.6.16.13-xenU/drivers/crypto/padlock-aes.c
===================================================================
--- linux-2.6.16.13-xenU.orig/drivers/crypto/padlock-aes.c
+++ linux-2.6.16.13-xenU/drivers/crypto/padlock-aes.c
@@ -59,6 +59,9 @@
 #define AES_EXTENDED_KEY_SIZE	64	/* in uint32_t units */
 #define AES_EXTENDED_KEY_SIZE_B	(AES_EXTENDED_KEY_SIZE * sizeof(uint32_t))
 
+/* Whenever making any changes to the following
+ * structure *make sure* you keep E, d_data
+ * and cword aligned on 16 Bytes boundaries!!! */
 struct aes_ctx {
 	struct {
 		struct cword encrypt;
@@ -66,8 +69,10 @@ struct aes_ctx {
 	} cword;
 	u32 *D;
 	int key_length;
-	u32 E[AES_EXTENDED_KEY_SIZE];
-	u32 d_data[AES_EXTENDED_KEY_SIZE];
+	u32 E[AES_EXTENDED_KEY_SIZE]
+		__attribute__ ((__aligned__(PADLOCK_ALIGNMENT)));
+	u32 d_data[AES_EXTENDED_KEY_SIZE]
+		__attribute__ ((__aligned__(PADLOCK_ALIGNMENT)));
 };
 
 /* ====== Key management routines ====== */