From: David Miller <davem@xxxxxxxxxxxxx> Date: Tue, 30 Sep 2014 22:57:12 -0400 (EDT) > I can reproduce this AES failure as well as the unaligned accesses. I've figured out the nature of the AES failure. The code tries to optimize the case of processing many blocks at a time, by preloading the KEY material into the FPU registers then running the encrypt/decrypt routine over and over again reusing those precooked registers. Somehow it appears that the pre-cooked key registers get spammed upon. Even if we get an intervening memcpy() via the blkcipher*() calls, we fully support recursive FPU usage in the kernel on sparc64 so it should "just work". The test case passes if I change the ->ecb_encrypt() call in ctr_crypt_final() to ->encrypt() (which hand loads the key into the FPU before performing a single block encrypt). Anyways I'll dig further and fix this. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
