From: Kamil Konieczny Subject: Re: [PATCH v2 3/4] crypto: inside-secure - only update the result buffer when provided Date: Thu, 30 Nov 2017 10:19:26 +0100 Message-ID: References: <20171128154236.19192-1-antoine.tenart@free-electrons.com> <20171128154236.19192-4-antoine.tenart@free-electrons.com> Mime-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Cc: thomas.petazzoni@free-electrons.com, gregory.clement@free-electrons.com, miquel.raynal@free-electrons.com, oferh@marvell.com, igall@marvell.com, nadavh@marvell.com, linux-crypto@vger.kernel.org To: Antoine Tenart , herbert@gondor.apana.org.au, davem@davemloft.net Return-path: Received: from mailout2.w1.samsung.com ([210.118.77.12]:36508 "EHLO mailout2.w1.samsung.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751157AbdK3JTb (ORCPT ); Thu, 30 Nov 2017 04:19:31 -0500 In-reply-to: <20171128154236.19192-4-antoine.tenart@free-electrons.com> Content-language: en-US Sender: linux-crypto-owner@vger.kernel.org List-ID: Hi Antoine, On 28.11.2017 16:42, Antoine Tenart wrote: > The patch fixes the ahash support by only updating the result buffer > when provided. Otherwise the driver could crash with NULL pointer > exceptions, because the ahash caller isn't required to supply a result > buffer on all calls. Can you point to bug crush report ? > Fixes: 1b44c5a60c13 ("crypto: inside-secure - add SafeXcel EIP197 crypto engine driver") > Signed-off-by: Antoine Tenart > --- > drivers/crypto/inside-secure/safexcel_hash.c | 7 ++++++- > 1 file changed, 6 insertions(+), 1 deletion(-) > > diff --git a/drivers/crypto/inside-secure/safexcel_hash.c b/drivers/crypto/inside-secure/safexcel_hash.c > index 6135c9f5742c..424f4c5d4d25 100644 > --- a/drivers/crypto/inside-secure/safexcel_hash.c > +++ b/drivers/crypto/inside-secure/safexcel_hash.c > @@ -150,7 +150,12 @@ static int safexcel_handle_req_result(struct safexcel_crypto_priv *priv, int rin > > if (sreq->finish) > result_sz = crypto_ahash_digestsize(ahash); > - memcpy(sreq->state, areq->result, result_sz); > + > + /* The called isn't required to supply a result buffer. Updated it only > + * when provided. > + */ > + if (areq->result) > + memcpy(sreq->state, areq->result, result_sz); > > dma_unmap_sg(priv->dev, areq->src, > sg_nents_for_len(areq->src, areq->nbytes), DMA_TO_DEVICE); > can the driver get request for final/finup/digest with null req->result ? If yes (?), such checks can be done before any hardware processing, saving time, for example: int hash_final(struct ahash_request *areq) { if (!areq->result) return -EINVAL; /* normal processing here */ } -- Best regards, Kamil Konieczny Samsung R&D Institute Poland