crypto: omap-sham - fix closing of hash with separate finalize call

Currently there is an interesting corner case failure with omap-sham
driver, if the finalize call is done separately with no data, but
all previous data has already been processed. In this case, it is not
possible to close the hash with the hardware without providing any data,
so we get incorrect results. Fix this by adjusting the size of data
sent to the hardware crypto engine in case the non-final data size falls
on the block size boundary, by reducing the amount of data sent by one
full block. This makes it sure that we always have some data available
for the finalize call and we can close the hash properly.

Signed-off-by: Tero Kristo <t-kristo@ti.com>
Reported-by: Aparna Balasubramanian <aparnab@ti.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

drivers/crypto/omap-sham.c

@@ -751,7 +751,10 @@ static int omap_sham_align_sgs(struct scatterlist *sg,
 	if (final)
 		new_len = DIV_ROUND_UP(new_len, bs) * bs;
 	else
-		new_len = new_len / bs * bs;
+		new_len = (new_len - 1) / bs * bs;
+
+	if (nbytes != new_len)
+		list_ok = false;
 
 	while (nbytes > 0 && sg_tmp) {
 		n++;
@@ -847,6 +850,8 @@ static int omap_sham_prepare_request(struct ahash_request *req, bool update)
 			xmit_len = DIV_ROUND_UP(xmit_len, bs) * bs;
 		else
 			xmit_len = xmit_len / bs * bs;
+	} else if (!final) {
+		xmit_len -= bs;
 	}
 
 	hash_later = rctx->total - xmit_len;
@@ -1138,7 +1143,7 @@ retry:
 	ctx = ahash_request_ctx(req);
 
 	err = omap_sham_prepare_request(req, ctx->op == OP_UPDATE);
-	if (err)
+	if (err || !ctx->total)
 		goto err1;
 
 	dev_dbg(dd->dev, "handling new req, op: %lu, nbytes: %d\n",