crypto: aesni-intel - fix unaligned cbc decrypt for x86-32
authorMathias Krause <minipli@googlemail.com>
Tue, 29 May 2012 23:43:08 +0000 (01:43 +0200)
committerHerbert Xu <herbert@gondor.apana.org.au>
Thu, 31 May 2012 10:53:22 +0000 (20:53 +1000)
The 32 bit variant of cbc(aes) decrypt is using instructions requiring
128 bit aligned memory locations but fails to ensure this constraint in
the code. Fix this by loading the data into intermediate registers with
load unaligned instructions.

This fixes reported general protection faults related to aesni.

References: https://bugzilla.kernel.org/show_bug.cgi?id=43223
Reported-by: Daniel <garkein@mailueberfall.de>
Cc: stable@kernel.org [v2.6.39+]
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/aesni-intel_asm.S

index be6d9e365a800a8569fb4915fccbfc50b0793a09..3470624d7835fa646b7c9f8dafa6acd3d9ed5f2f 100644 (file)
@@ -2460,10 +2460,12 @@ ENTRY(aesni_cbc_dec)
        pxor IN3, STATE4
        movaps IN4, IV
 #else
-       pxor (INP), STATE2
-       pxor 0x10(INP), STATE3
        pxor IN1, STATE4
        movaps IN2, IV
+       movups (INP), IN1
+       pxor IN1, STATE2
+       movups 0x10(INP), IN2
+       pxor IN2, STATE3
 #endif
        movups STATE1, (OUTP)
        movups STATE2, 0x10(OUTP)