From: Romain Perier Subject: [PATCH v3 2/2] crypto: marvell - Don't break chain for computable last ahash requests Date: Tue, 4 Oct 2016 14:57:20 +0200 Message-ID: <20161004125720.3347-3-romain.perier@free-electrons.com> References: <20161004125720.3347-1-romain.perier@free-electrons.com> Cc: "David S. Miller" , Herbert Xu , Thomas Petazzoni , Jason Cooper , Andrew Lunn , Sebastian Hesselbarth , Gregory Clement , Nadav Haklai , Ofer Heifetz , linux-crypto@vger.kernel.org, linux-arm-kernel@lists.infradead.org To: Boris Brezillon , Arnaud Ebalard Return-path: Received: from down.free-electrons.com ([37.187.137.238]:58699 "EHLO mail.free-electrons.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1753226AbcJDM5l (ORCPT ); Tue, 4 Oct 2016 08:57:41 -0400 In-Reply-To: <20161004125720.3347-1-romain.perier@free-electrons.com> Sender: linux-crypto-owner@vger.kernel.org List-ID: Currently, the driver breaks chain for all kind of hash requests in order to don't override intermediate states of partial ahash updates. However, some final ahash requests can be directly processed by the engine, and so without intermediate state. This is typically the case for most for the HMAC requests processed via IPSec. This commits adds a TDMA descriptor to copy context for these of requests into the "op" dma pool, then it allow to chain these requests at the DMA level. The 'complete' operation is also updated to retrieve the MAC digest from the right location. Signed-off-by: Romain Perier --- Changes in v3: - Copy the whole context back to RAM and not just the digest. Also fixed a rebase issue ^^ (whoops) Changes in v2: - Replaced BUG_ON by an error - Add a variable "break_chain", with "type" to break the chain with ahash requests. It improves code readability. drivers/crypto/marvell/hash.c | 79 +++++++++++++++++++++++++++++++++++-------- 1 file changed, 64 insertions(+), 15 deletions(-) diff --git a/drivers/crypto/marvell/hash.c b/drivers/crypto/marvell/hash.c index 9f28468..b36f196 100644 --- a/drivers/crypto/marvell/hash.c +++ b/drivers/crypto/marvell/hash.c @@ -312,24 +312,53 @@ static void mv_cesa_ahash_complete(struct crypto_async_request *req) int i; digsize = crypto_ahash_digestsize(crypto_ahash_reqtfm(ahashreq)); - for (i = 0; i < digsize / 4; i++) - creq->state[i] = readl_relaxed(engine->regs + CESA_IVDIG(i)); - if (creq->last_req) { + if (mv_cesa_req_get_type(&creq->base) == CESA_DMA_REQ && + !(creq->base.chain.last->flags & CESA_TDMA_BREAK_CHAIN)) { + struct mv_cesa_tdma_desc *tdma = NULL; + __le32 *data = NULL; + + for (tdma = creq->base.chain.first; tdma; tdma = tdma->next) { + u32 type = tdma->flags & CESA_TDMA_TYPE_MSK; + if (type == CESA_TDMA_RESULT) + break; + } + + if (!tdma) { + dev_err(cesa_dev->dev, "Failed to retrieve tdma " + "descriptor for outer data\n"); + return; + } + /* - * Hardware's MD5 digest is in little endian format, but - * SHA in big endian format + * Result is already in the correct endianess when the SA is + * used */ - if (creq->algo_le) { - __le32 *result = (void *)ahashreq->result; + data = tdma->op->ctx.hash.hash; + for (i = 0; i < digsize / 4; i++) + creq->state[i] = cpu_to_le32(data[i]); - for (i = 0; i < digsize / 4; i++) - result[i] = cpu_to_le32(creq->state[i]); - } else { - __be32 *result = (void *)ahashreq->result; + memcpy(ahashreq->result, data, digsize); + } else { + for (i = 0; i < digsize / 4; i++) + creq->state[i] = readl_relaxed(engine->regs + + CESA_IVDIG(i)); + if (creq->last_req) { + /* + * Hardware's MD5 digest is in little endian format, but + * SHA in big endian format + */ + if (creq->algo_le) { + __le32 *result = (void *)ahashreq->result; + + for (i = 0; i < digsize / 4; i++) + result[i] = cpu_to_le32(creq->state[i]); + } else { + __be32 *result = (void *)ahashreq->result; - for (i = 0; i < digsize / 4; i++) - result[i] = cpu_to_be32(creq->state[i]); + for (i = 0; i < digsize / 4; i++) + result[i] = cpu_to_be32(creq->state[i]); + } } } @@ -504,6 +533,12 @@ mv_cesa_ahash_dma_last_req(struct mv_cesa_tdma_chain *chain, CESA_SA_DESC_CFG_LAST_FRAG, CESA_SA_DESC_CFG_FRAG_MSK); + ret = mv_cesa_dma_add_result_op(chain, + CESA_SA_CFG_SRAM_OFFSET, + CESA_SA_DATA_SRAM_OFFSET, + CESA_TDMA_SRC_IN_SRAM, flags); + if (ret) + return ERR_PTR(-ENOMEM); return op; } @@ -564,6 +599,8 @@ static int mv_cesa_ahash_dma_req_init(struct ahash_request *req) struct mv_cesa_op_ctx *op = NULL; unsigned int frag_len; int ret; + u32 type; + bool break_chain = true; basereq->chain.first = NULL; basereq->chain.last = NULL; @@ -635,6 +672,16 @@ static int mv_cesa_ahash_dma_req_init(struct ahash_request *req) goto err_free_tdma; } + /* + * If results are copied via DMA, this means that this + * request can be directly processed by the engine, + * without partial updates. So we can chain it at the + * DMA level with other requests. + */ + type = basereq->chain.last->flags & CESA_TDMA_TYPE_MSK; + if (type == CESA_TDMA_RESULT) + break_chain = false; + if (op) { /* Add dummy desc to wait for crypto operation end */ ret = mv_cesa_dma_add_dummy_end(&basereq->chain, flags); @@ -648,8 +695,10 @@ static int mv_cesa_ahash_dma_req_init(struct ahash_request *req) else creq->cache_ptr = 0; - basereq->chain.last->flags |= (CESA_TDMA_END_OF_REQ | - CESA_TDMA_BREAK_CHAIN); + basereq->chain.last->flags |= CESA_TDMA_END_OF_REQ; + + if (break_chain) + basereq->chain.last->flags |= CESA_TDMA_BREAK_CHAIN; return 0; -- 2.9.3