Re: [PATCH v2 2/2] crypto: caam - add support for rfc4106(gcm(aes))

On 10/14/2014 12:17 PM, Tudor Ambarus wrote:

+static int rfc4106_setkey(struct crypto_aead *aead,
+			  const u8 *key, unsigned int keylen)
+{
+	struct caam_ctx *ctx = crypto_aead_ctx(aead);
+	struct device *jrdev = ctx->jrdev;
+	int ret = 0;
+
+	/*
+	 * The last four bytes of the key material are used as the salt value
+	 * in the nonce. Update the AES key length.
+	 */
+	if (keylen < 4)
+		return -EINVAL;
+	keylen -= 4;

The salt will not be copied in the ctx->key, so the descriptors will use 
as a salt whatever resides in memory after (updated) keylen bytes.

I will submit a new patch set in which I will update the AES key length 
after copying the key material in ctx->key.

tcrypt tests passed because they use salt values of zero.
Back-to-back tests passed because I used caam with rfc4106 accelerated 
on both boards.

+
+#ifdef DEBUG
+	print_hex_dump(KERN_ERR, "key in @"__stringify(__LINE__)": ",
+		       DUMP_PREFIX_ADDRESS, 16, 4, key, keylen, 1);
+#endif
+
+	memcpy(ctx->key, key, keylen);