crypto: sha-mb - SHA1 multibuffer job manager and glue code
authorTim Chen <tim.c.chen@linux.intel.com>
Thu, 31 Jul 2014 17:30:03 +0000 (10:30 -0700)
committerHerbert Xu <herbert@gondor.apana.org.au>
Mon, 25 Aug 2014 12:32:30 +0000 (20:32 +0800)
This patch introduces the multi-buffer job manager which is responsible
for submitting scatter-gather buffers from several SHA1 jobs to the
multi-buffer algorithm.  It also contains the flush routine to that's
called by the crypto daemon to complete the job when no new jobs arrive
before the deadline of maximum latency of a SHA1 crypto job.

The SHA1 multi-buffer crypto algorithm is defined and initialized in
this patch.

Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/Makefile
arch/x86/crypto/sha-mb/Makefile [new file with mode: 0644]
arch/x86/crypto/sha-mb/sha1_mb.c [new file with mode: 0644]

index d551165a3159022bcafda949a4573c5183732654..fd0f848938ccd81a165c9dff570e108c84271c4d 100644 (file)
@@ -26,6 +26,7 @@ obj-$(CONFIG_CRYPTO_GHASH_CLMUL_NI_INTEL) += ghash-clmulni-intel.o
 
 obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
 obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
+obj-$(CONFIG_CRYPTO_SHA1_MB) += sha-mb/
 obj-$(CONFIG_CRYPTO_CRC32_PCLMUL) += crc32-pclmul.o
 obj-$(CONFIG_CRYPTO_SHA256_SSSE3) += sha256-ssse3.o
 obj-$(CONFIG_CRYPTO_SHA512_SSSE3) += sha512-ssse3.o
diff --git a/arch/x86/crypto/sha-mb/Makefile b/arch/x86/crypto/sha-mb/Makefile
new file mode 100644 (file)
index 0000000..2f87563
--- /dev/null
@@ -0,0 +1,11 @@
+#
+# Arch-specific CryptoAPI modules.
+#
+
+avx2_supported := $(call as-instr,vpgatherdd %ymm0$(comma)(%eax$(comma)%ymm1\
+                                $(comma)4)$(comma)%ymm2,yes,no)
+ifeq ($(avx2_supported),yes)
+       obj-$(CONFIG_CRYPTO_SHA1_MB) += sha1-mb.o
+       sha1-mb-y := sha1_mb.o sha1_mb_mgr_flush_avx2.o \
+            sha1_mb_mgr_init_avx2.o sha1_mb_mgr_submit_avx2.o sha1_x8_avx2.o
+endif
diff --git a/arch/x86/crypto/sha-mb/sha1_mb.c b/arch/x86/crypto/sha-mb/sha1_mb.c
new file mode 100644 (file)
index 0000000..e27ff31
--- /dev/null
@@ -0,0 +1,935 @@
+/*
+ * Multi buffer SHA1 algorithm Glue Code
+ *
+ * This file is provided under a dual BSD/GPLv2 license.  When using or
+ * redistributing this file, you may do so under either license.
+ *
+ * GPL LICENSE SUMMARY
+ *
+ *  Copyright(c) 2014 Intel Corporation.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of version 2 of the GNU General Public License as
+ *  published by the Free Software Foundation.
+ *
+ *  This program is distributed in the hope that it will be useful, but
+ *  WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ *  General Public License for more details.
+ *
+ *  Contact Information:
+ *     Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ *  BSD LICENSE
+ *
+ *  Copyright(c) 2014 Intel Corporation.
+ *
+ *  Redistribution and use in source and binary forms, with or without
+ *  modification, are permitted provided that the following conditions
+ *  are met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above copyright
+ *      notice, this list of conditions and the following disclaimer in
+ *      the documentation and/or other materials provided with the
+ *      distribution.
+ *    * Neither the name of Intel Corporation nor the names of its
+ *      contributors may be used to endorse or promote products derived
+ *      from this software without specific prior written permission.
+ *
+ *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#define pr_fmt(fmt)    KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <linux/list.h>
+#include <crypto/scatterwalk.h>
+#include <crypto/sha.h>
+#include <crypto/mcryptd.h>
+#include <crypto/crypto_wq.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+#include <linux/hardirq.h>
+#include <asm/fpu-internal.h>
+#include "sha_mb_ctx.h"
+
+#define FLUSH_INTERVAL 1000 /* in usec */
+
+struct mcryptd_alg_state sha1_mb_alg_state;
+
+struct sha1_mb_ctx {
+       struct mcryptd_ahash *mcryptd_tfm;
+};
+
+static inline struct mcryptd_hash_request_ctx *cast_hash_to_mcryptd_ctx(struct sha1_hash_ctx *hash_ctx)
+{
+       struct shash_desc *desc;
+
+       desc = container_of((void *) hash_ctx, struct shash_desc, __ctx);
+       return container_of(desc, struct mcryptd_hash_request_ctx, desc);
+}
+
+static inline struct ahash_request *cast_mcryptd_ctx_to_req(struct mcryptd_hash_request_ctx *ctx)
+{
+       return container_of((void *) ctx, struct ahash_request, __ctx);
+}
+
+static void req_ctx_init(struct mcryptd_hash_request_ctx *rctx,
+                               struct shash_desc *desc)
+{
+       rctx->flag = HASH_UPDATE;
+}
+
+asmlinkage void (*sha1_job_mgr_init)(struct sha1_mb_mgr *state);
+asmlinkage struct job_sha1* (*sha1_job_mgr_submit)(struct sha1_mb_mgr *state,
+                                                         struct job_sha1 *job);
+asmlinkage struct job_sha1* (*sha1_job_mgr_flush)(struct sha1_mb_mgr *state);
+asmlinkage struct job_sha1* (*sha1_job_mgr_get_comp_job)(struct sha1_mb_mgr *state);
+
+inline void sha1_init_digest(uint32_t *digest)
+{
+       static const uint32_t initial_digest[SHA1_DIGEST_LENGTH] = {SHA1_H0,
+                                       SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 };
+       memcpy(digest, initial_digest, sizeof(initial_digest));
+}
+
+inline uint32_t sha1_pad(uint8_t padblock[SHA1_BLOCK_SIZE * 2],
+                        uint32_t total_len)
+{
+       uint32_t i = total_len & (SHA1_BLOCK_SIZE - 1);
+
+       memset(&padblock[i], 0, SHA1_BLOCK_SIZE);
+       padblock[i] = 0x80;
+
+       i += ((SHA1_BLOCK_SIZE - 1) &
+             (0 - (total_len + SHA1_PADLENGTHFIELD_SIZE + 1)))
+            + 1 + SHA1_PADLENGTHFIELD_SIZE;
+
+#if SHA1_PADLENGTHFIELD_SIZE == 16
+       *((uint64_t *) &padblock[i - 16]) = 0;
+#endif
+
+       *((uint64_t *) &padblock[i - 8]) = cpu_to_be64(total_len << 3);
+
+       /* Number of extra blocks to hash */
+       return i >> SHA1_LOG2_BLOCK_SIZE;
+}
+
+static struct sha1_hash_ctx *sha1_ctx_mgr_resubmit(struct sha1_ctx_mgr *mgr, struct sha1_hash_ctx *ctx)
+{
+       while (ctx) {
+               if (ctx->status & HASH_CTX_STS_COMPLETE) {
+                       /* Clear PROCESSING bit */
+                       ctx->status = HASH_CTX_STS_COMPLETE;
+                       return ctx;
+               }
+
+               /*
+                * If the extra blocks are empty, begin hashing what remains
+                * in the user's buffer.
+                */
+               if (ctx->partial_block_buffer_length == 0 &&
+                   ctx->incoming_buffer_length) {
+
+                       const void *buffer = ctx->incoming_buffer;
+                       uint32_t len = ctx->incoming_buffer_length;
+                       uint32_t copy_len;
+
+                       /*
+                        * Only entire blocks can be hashed.
+                        * Copy remainder to extra blocks buffer.
+                        */
+                       copy_len = len & (SHA1_BLOCK_SIZE-1);
+
+                       if (copy_len) {
+                               len -= copy_len;
+                               memcpy(ctx->partial_block_buffer,
+                                      ((const char *) buffer + len),
+                                      copy_len);
+                               ctx->partial_block_buffer_length = copy_len;
+                       }
+
+                       ctx->incoming_buffer_length = 0;
+
+                       /* len should be a multiple of the block size now */
+                       assert((len % SHA1_BLOCK_SIZE) == 0);
+
+                       /* Set len to the number of blocks to be hashed */
+                       len >>= SHA1_LOG2_BLOCK_SIZE;
+
+                       if (len) {
+
+                               ctx->job.buffer = (uint8_t *) buffer;
+                               ctx->job.len = len;
+                               ctx = (struct sha1_hash_ctx *) sha1_job_mgr_submit(&mgr->mgr,
+                                                                                 &ctx->job);
+                               continue;
+                       }
+               }
+
+               /*
+                * If the extra blocks are not empty, then we are
+                * either on the last block(s) or we need more
+                * user input before continuing.
+                */
+               if (ctx->status & HASH_CTX_STS_LAST) {
+
+                       uint8_t *buf = ctx->partial_block_buffer;
+                       uint32_t n_extra_blocks = sha1_pad(buf, ctx->total_length);
+
+                       ctx->status = (HASH_CTX_STS_PROCESSING |
+                                      HASH_CTX_STS_COMPLETE);
+                       ctx->job.buffer = buf;
+                       ctx->job.len = (uint32_t) n_extra_blocks;
+                       ctx = (struct sha1_hash_ctx *) sha1_job_mgr_submit(&mgr->mgr, &ctx->job);
+                       continue;
+               }
+
+               if (ctx)
+                       ctx->status = HASH_CTX_STS_IDLE;
+               return ctx;
+       }
+
+       return NULL;
+}
+
+struct sha1_hash_ctx *sha1_ctx_mgr_get_comp_ctx(struct sha1_ctx_mgr *mgr)
+{
+       /*
+        * If get_comp_job returns NULL, there are no jobs complete.
+        * If get_comp_job returns a job, verify that it is safe to return to the user.
+        * If it is not ready, resubmit the job to finish processing.
+        * If sha1_ctx_mgr_resubmit returned a job, it is ready to be returned.
+        * Otherwise, all jobs currently being managed by the hash_ctx_mgr still need processing.
+        */
+       struct sha1_hash_ctx *ctx;
+
+       ctx = (struct sha1_hash_ctx *) sha1_job_mgr_get_comp_job(&mgr->mgr);
+       return sha1_ctx_mgr_resubmit(mgr, ctx);
+}
+
+void sha1_ctx_mgr_init(struct sha1_ctx_mgr *mgr)
+{
+       sha1_job_mgr_init(&mgr->mgr);
+}
+
+struct sha1_hash_ctx *sha1_ctx_mgr_submit(struct sha1_ctx_mgr *mgr,
+                                         struct sha1_hash_ctx *ctx,
+                                         const void *buffer,
+                                         uint32_t len,
+                                         int flags)
+{
+       if (flags & (~HASH_ENTIRE)) {
+               /* User should not pass anything other than FIRST, UPDATE, or LAST */
+               ctx->error = HASH_CTX_ERROR_INVALID_FLAGS;
+               return ctx;
+       }
+
+       if (ctx->status & HASH_CTX_STS_PROCESSING) {
+               /* Cannot submit to a currently processing job. */
+               ctx->error = HASH_CTX_ERROR_ALREADY_PROCESSING;
+               return ctx;
+       }
+
+       if ((ctx->status & HASH_CTX_STS_COMPLETE) && !(flags & HASH_FIRST)) {
+               /* Cannot update a finished job. */
+               ctx->error = HASH_CTX_ERROR_ALREADY_COMPLETED;
+               return ctx;
+       }
+
+
+       if (flags & HASH_FIRST) {
+               /* Init digest */
+               sha1_init_digest(ctx->job.result_digest);
+
+               /* Reset byte counter */
+               ctx->total_length = 0;
+
+               /* Clear extra blocks */
+               ctx->partial_block_buffer_length = 0;
+       }
+
+       /* If we made it here, there were no errors during this call to submit */
+       ctx->error = HASH_CTX_ERROR_NONE;
+
+       /* Store buffer ptr info from user */
+       ctx->incoming_buffer = buffer;
+       ctx->incoming_buffer_length = len;
+
+       /* Store the user's request flags and mark this ctx as currently being processed. */
+       ctx->status = (flags & HASH_LAST) ?
+                       (HASH_CTX_STS_PROCESSING | HASH_CTX_STS_LAST) :
+                       HASH_CTX_STS_PROCESSING;
+
+       /* Advance byte counter */
+       ctx->total_length += len;
+
+       /*
+        * If there is anything currently buffered in the extra blocks,
+        * append to it until it contains a whole block.
+        * Or if the user's buffer contains less than a whole block,
+        * append as much as possible to the extra block.
+        */
+       if ((ctx->partial_block_buffer_length) | (len < SHA1_BLOCK_SIZE)) {
+               /* Compute how many bytes to copy from user buffer into extra block */
+               uint32_t copy_len = SHA1_BLOCK_SIZE - ctx->partial_block_buffer_length;
+               if (len < copy_len)
+                       copy_len = len;
+
+               if (copy_len) {
+                       /* Copy and update relevant pointers and counters */
+                       memcpy(&ctx->partial_block_buffer[ctx->partial_block_buffer_length],
+                               buffer, copy_len);
+
+                       ctx->partial_block_buffer_length += copy_len;
+                       ctx->incoming_buffer = (const void *)((const char *)buffer + copy_len);
+                       ctx->incoming_buffer_length = len - copy_len;
+               }
+
+               /* The extra block should never contain more than 1 block here */
+               assert(ctx->partial_block_buffer_length <= SHA1_BLOCK_SIZE);
+
+               /* If the extra block buffer contains exactly 1 block, it can be hashed. */
+               if (ctx->partial_block_buffer_length >= SHA1_BLOCK_SIZE) {
+                       ctx->partial_block_buffer_length = 0;
+
+                       ctx->job.buffer = ctx->partial_block_buffer;
+                       ctx->job.len = 1;
+                       ctx = (struct sha1_hash_ctx *) sha1_job_mgr_submit(&mgr->mgr, &ctx->job);
+               }
+       }
+
+       return sha1_ctx_mgr_resubmit(mgr, ctx);
+}
+
+struct sha1_hash_ctx *sha1_ctx_mgr_flush(struct sha1_ctx_mgr *mgr)
+{
+       struct sha1_hash_ctx *ctx;
+
+       while (1) {
+               ctx = (struct sha1_hash_ctx *) sha1_job_mgr_flush(&mgr->mgr);
+
+               /* If flush returned 0, there are no more jobs in flight. */
+               if (!ctx)
+                       return NULL;
+
+               /*
+                * If flush returned a job, resubmit the job to finish processing.
+                */
+               ctx = sha1_ctx_mgr_resubmit(mgr, ctx);
+
+               /*
+                * If sha1_ctx_mgr_resubmit returned a job, it is ready to be returned.
+                * Otherwise, all jobs currently being managed by the sha1_ctx_mgr
+                * still need processing. Loop.
+                */
+               if (ctx)
+                       return ctx;
+       }
+}
+
+static int sha1_mb_init(struct shash_desc *desc)
+{
+       struct sha1_hash_ctx *sctx = shash_desc_ctx(desc);
+
+       hash_ctx_init(sctx);
+       sctx->job.result_digest[0] = SHA1_H0;
+       sctx->job.result_digest[1] = SHA1_H1;
+       sctx->job.result_digest[2] = SHA1_H2;
+       sctx->job.result_digest[3] = SHA1_H3;
+       sctx->job.result_digest[4] = SHA1_H4;
+       sctx->total_length = 0;
+       sctx->partial_block_buffer_length = 0;
+       sctx->status = HASH_CTX_STS_IDLE;
+
+       return 0;
+}
+
+static int sha1_mb_set_results(struct mcryptd_hash_request_ctx *rctx)
+{
+       int     i;
+       struct  sha1_hash_ctx *sctx = shash_desc_ctx(&rctx->desc);
+       __be32  *dst = (__be32 *) rctx->out;
+
+       for (i = 0; i < 5; ++i)
+               dst[i] = cpu_to_be32(sctx->job.result_digest[i]);
+
+       return 0;
+}
+
+static int sha_finish_walk(struct mcryptd_hash_request_ctx **ret_rctx,
+                       struct mcryptd_alg_cstate *cstate, bool flush)
+{
+       int     flag = HASH_UPDATE;
+       int     nbytes, err = 0;
+       struct mcryptd_hash_request_ctx *rctx = *ret_rctx;
+       struct sha1_hash_ctx *sha_ctx;
+
+       /* more work ? */
+       while (!(rctx->flag & HASH_DONE)) {
+               nbytes = crypto_ahash_walk_done(&rctx->walk, 0);
+               if (nbytes < 0) {
+                       err = nbytes;
+                       goto out;
+               }
+               /* check if the walk is done */
+               if (crypto_ahash_walk_last(&rctx->walk)) {
+                       rctx->flag |= HASH_DONE;
+                       if (rctx->flag & HASH_FINAL)
+                               flag |= HASH_LAST;
+
+               }
+               sha_ctx = (struct sha1_hash_ctx *) shash_desc_ctx(&rctx->desc);
+               kernel_fpu_begin();
+               sha_ctx = sha1_ctx_mgr_submit(cstate->mgr, sha_ctx, rctx->walk.data, nbytes, flag);
+               if (!sha_ctx) {
+                       if (flush)
+                               sha_ctx = sha1_ctx_mgr_flush(cstate->mgr);
+               }
+               kernel_fpu_end();
+               if (sha_ctx)
+                       rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+               else {
+                       rctx = NULL;
+                       goto out;
+               }
+       }
+
+       /* copy the results */
+       if (rctx->flag & HASH_FINAL)
+               sha1_mb_set_results(rctx);
+
+out:
+       *ret_rctx = rctx;
+       return err;
+}
+
+static int sha_complete_job(struct mcryptd_hash_request_ctx *rctx,
+                           struct mcryptd_alg_cstate *cstate,
+                           int err)
+{
+       struct ahash_request *req = cast_mcryptd_ctx_to_req(rctx);
+       struct sha1_hash_ctx *sha_ctx;
+       struct mcryptd_hash_request_ctx *req_ctx;
+       int ret;
+
+       /* remove from work list */
+       spin_lock(&cstate->work_lock);
+       list_del(&rctx->waiter);
+       spin_unlock(&cstate->work_lock);
+
+       if (irqs_disabled())
+               rctx->complete(&req->base, err);
+       else {
+               local_bh_disable();
+               rctx->complete(&req->base, err);
+               local_bh_enable();
+       }
+
+       /* check to see if there are other jobs that are done */
+       sha_ctx = sha1_ctx_mgr_get_comp_ctx(cstate->mgr);
+       while (sha_ctx) {
+               req_ctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+               ret = sha_finish_walk(&req_ctx, cstate, false);
+               if (req_ctx) {
+                       spin_lock(&cstate->work_lock);
+                       list_del(&req_ctx->waiter);
+                       spin_unlock(&cstate->work_lock);
+
+                       req = cast_mcryptd_ctx_to_req(req_ctx);
+                       if (irqs_disabled())
+                               rctx->complete(&req->base, ret);
+                       else {
+                               local_bh_disable();
+                               rctx->complete(&req->base, ret);
+                               local_bh_enable();
+                       }
+               }
+               sha_ctx = sha1_ctx_mgr_get_comp_ctx(cstate->mgr);
+       }
+
+       return 0;
+}
+
+static void sha1_mb_add_list(struct mcryptd_hash_request_ctx *rctx,
+                            struct mcryptd_alg_cstate *cstate)
+{
+       unsigned long next_flush;
+       unsigned long delay = usecs_to_jiffies(FLUSH_INTERVAL);
+
+       /* initialize tag */
+       rctx->tag.arrival = jiffies;    /* tag the arrival time */
+       rctx->tag.seq_num = cstate->next_seq_num++;
+       next_flush = rctx->tag.arrival + delay;
+       rctx->tag.expire = next_flush;
+
+       spin_lock(&cstate->work_lock);
+       list_add_tail(&rctx->waiter, &cstate->work_list);
+       spin_unlock(&cstate->work_lock);
+
+       mcryptd_arm_flusher(cstate, delay);
+}
+
+static int sha1_mb_update(struct shash_desc *desc, const u8 *data,
+                         unsigned int len)
+{
+       struct mcryptd_hash_request_ctx *rctx =
+                       container_of(desc, struct mcryptd_hash_request_ctx, desc);
+       struct mcryptd_alg_cstate *cstate =
+                               this_cpu_ptr(sha1_mb_alg_state.alg_cstate);
+
+       struct ahash_request *req = cast_mcryptd_ctx_to_req(rctx);
+       struct sha1_hash_ctx *sha_ctx;
+       int ret = 0, nbytes;
+
+
+       /* sanity check */
+       if (rctx->tag.cpu != smp_processor_id()) {
+               pr_err("mcryptd error: cpu clash\n");
+               goto done;
+       }
+
+       /* need to init context */
+       req_ctx_init(rctx, desc);
+
+       nbytes = crypto_ahash_walk_first(req, &rctx->walk);
+
+       if (nbytes < 0) {
+               ret = nbytes;
+               goto done;
+       }
+
+       if (crypto_ahash_walk_last(&rctx->walk))
+               rctx->flag |= HASH_DONE;
+
+       /* submit */
+       sha_ctx = (struct sha1_hash_ctx *) shash_desc_ctx(desc);
+       sha1_mb_add_list(rctx, cstate);
+       kernel_fpu_begin();
+       sha_ctx = sha1_ctx_mgr_submit(cstate->mgr, sha_ctx, rctx->walk.data, nbytes, HASH_UPDATE);
+       kernel_fpu_end();
+
+       /* check if anything is returned */
+       if (!sha_ctx)
+               return -EINPROGRESS;
+
+       if (sha_ctx->error) {
+               ret = sha_ctx->error;
+               rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+               goto done;
+       }
+
+       rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+       ret = sha_finish_walk(&rctx, cstate, false);
+
+       if (!rctx)
+               return -EINPROGRESS;
+done:
+       sha_complete_job(rctx, cstate, ret);
+       return ret;
+}
+
+static int sha1_mb_finup(struct shash_desc *desc, const u8 *data,
+                            unsigned int len, u8 *out)
+{
+       struct mcryptd_hash_request_ctx *rctx =
+                       container_of(desc, struct mcryptd_hash_request_ctx, desc);
+       struct mcryptd_alg_cstate *cstate =
+                               this_cpu_ptr(sha1_mb_alg_state.alg_cstate);
+
+       struct ahash_request *req = cast_mcryptd_ctx_to_req(rctx);
+       struct sha1_hash_ctx *sha_ctx;
+       int ret = 0, flag = HASH_UPDATE, nbytes;
+
+       /* sanity check */
+       if (rctx->tag.cpu != smp_processor_id()) {
+               pr_err("mcryptd error: cpu clash\n");
+               goto done;
+       }
+
+       /* need to init context */
+       req_ctx_init(rctx, desc);
+
+       nbytes = crypto_ahash_walk_first(req, &rctx->walk);
+
+       if (nbytes < 0) {
+               ret = nbytes;
+               goto done;
+       }
+
+       if (crypto_ahash_walk_last(&rctx->walk)) {
+               rctx->flag |= HASH_DONE;
+               flag = HASH_LAST;
+       }
+       rctx->out = out;
+
+       /* submit */
+       rctx->flag |= HASH_FINAL;
+       sha_ctx = (struct sha1_hash_ctx *) shash_desc_ctx(desc);
+       sha1_mb_add_list(rctx, cstate);
+
+       kernel_fpu_begin();
+       sha_ctx = sha1_ctx_mgr_submit(cstate->mgr, sha_ctx, rctx->walk.data, nbytes, flag);
+       kernel_fpu_end();
+
+       /* check if anything is returned */
+       if (!sha_ctx)
+               return -EINPROGRESS;
+
+       if (sha_ctx->error) {
+               ret = sha_ctx->error;
+               goto done;
+       }
+
+       rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+       ret = sha_finish_walk(&rctx, cstate, false);
+       if (!rctx)
+               return -EINPROGRESS;
+done:
+       sha_complete_job(rctx, cstate, ret);
+       return ret;
+}
+
+static int sha1_mb_final(struct shash_desc *desc, u8 *out)
+{
+       struct mcryptd_hash_request_ctx *rctx =
+                       container_of(desc, struct mcryptd_hash_request_ctx, desc);
+       struct mcryptd_alg_cstate *cstate =
+                               this_cpu_ptr(sha1_mb_alg_state.alg_cstate);
+
+       struct sha1_hash_ctx *sha_ctx;
+       int ret = 0;
+       u8 data;
+
+       /* sanity check */
+       if (rctx->tag.cpu != smp_processor_id()) {
+               pr_err("mcryptd error: cpu clash\n");
+               goto done;
+       }
+
+       /* need to init context */
+       req_ctx_init(rctx, desc);
+
+       rctx->out = out;
+       rctx->flag |= HASH_DONE | HASH_FINAL;
+
+       sha_ctx = (struct sha1_hash_ctx *) shash_desc_ctx(desc);
+       /* flag HASH_FINAL and 0 data size */
+       sha1_mb_add_list(rctx, cstate);
+       kernel_fpu_begin();
+       sha_ctx = sha1_ctx_mgr_submit(cstate->mgr, sha_ctx, &data, 0, HASH_LAST);
+       kernel_fpu_end();
+
+       /* check if anything is returned */
+       if (!sha_ctx)
+               return -EINPROGRESS;
+
+       if (sha_ctx->error) {
+               ret = sha_ctx->error;
+               rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+               goto done;
+       }
+
+       rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+       ret = sha_finish_walk(&rctx, cstate, false);
+       if (!rctx)
+               return -EINPROGRESS;
+done:
+       sha_complete_job(rctx, cstate, ret);
+       return ret;
+}
+
+static int sha1_mb_export(struct shash_desc *desc, void *out)
+{
+       struct sha1_hash_ctx *sctx = shash_desc_ctx(desc);
+
+       memcpy(out, sctx, sizeof(*sctx));
+
+       return 0;
+}
+
+static int sha1_mb_import(struct shash_desc *desc, const void *in)
+{
+       struct sha1_hash_ctx *sctx = shash_desc_ctx(desc);
+
+       memcpy(sctx, in, sizeof(*sctx));
+
+       return 0;
+}
+
+
+static struct shash_alg sha1_mb_shash_alg = {
+       .digestsize     =       SHA1_DIGEST_SIZE,
+       .init           =       sha1_mb_init,
+       .update         =       sha1_mb_update,
+       .final          =       sha1_mb_final,
+       .finup          =       sha1_mb_finup,
+       .export         =       sha1_mb_export,
+       .import         =       sha1_mb_import,
+       .descsize       =       sizeof(struct sha1_hash_ctx),
+       .statesize      =       sizeof(struct sha1_hash_ctx),
+       .base           =       {
+               .cra_name        = "__sha1-mb",
+               .cra_driver_name = "__intel_sha1-mb",
+               .cra_priority    = 100,
+               /*
+                * use ASYNC flag as some buffers in multi-buffer
+                * algo may not have completed before hashing thread sleep
+                */
+               .cra_flags       = CRYPTO_ALG_TYPE_SHASH | CRYPTO_ALG_ASYNC,
+               .cra_blocksize   = SHA1_BLOCK_SIZE,
+               .cra_module      = THIS_MODULE,
+               .cra_list        = LIST_HEAD_INIT(sha1_mb_shash_alg.base.cra_list),
+       }
+};
+
+static int sha1_mb_async_init(struct ahash_request *req)
+{
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct sha1_mb_ctx *ctx = crypto_ahash_ctx(tfm);
+       struct ahash_request *mcryptd_req = ahash_request_ctx(req);
+       struct mcryptd_ahash *mcryptd_tfm = ctx->mcryptd_tfm;
+
+       memcpy(mcryptd_req, req, sizeof(*req));
+       ahash_request_set_tfm(mcryptd_req, &mcryptd_tfm->base);
+       return crypto_ahash_init(mcryptd_req);
+}
+
+static int sha1_mb_async_update(struct ahash_request *req)
+{
+       struct ahash_request *mcryptd_req = ahash_request_ctx(req);
+
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct sha1_mb_ctx *ctx = crypto_ahash_ctx(tfm);
+       struct mcryptd_ahash *mcryptd_tfm = ctx->mcryptd_tfm;
+
+       memcpy(mcryptd_req, req, sizeof(*req));
+       ahash_request_set_tfm(mcryptd_req, &mcryptd_tfm->base);
+       return crypto_ahash_update(mcryptd_req);
+}
+
+static int sha1_mb_async_finup(struct ahash_request *req)
+{
+       struct ahash_request *mcryptd_req = ahash_request_ctx(req);
+
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct sha1_mb_ctx *ctx = crypto_ahash_ctx(tfm);
+       struct mcryptd_ahash *mcryptd_tfm = ctx->mcryptd_tfm;
+
+       memcpy(mcryptd_req, req, sizeof(*req));
+       ahash_request_set_tfm(mcryptd_req, &mcryptd_tfm->base);
+       return crypto_ahash_finup(mcryptd_req);
+}
+
+static int sha1_mb_async_final(struct ahash_request *req)
+{
+       struct ahash_request *mcryptd_req = ahash_request_ctx(req);
+
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct sha1_mb_ctx *ctx = crypto_ahash_ctx(tfm);
+       struct mcryptd_ahash *mcryptd_tfm = ctx->mcryptd_tfm;
+
+       memcpy(mcryptd_req, req, sizeof(*req));
+       ahash_request_set_tfm(mcryptd_req, &mcryptd_tfm->base);
+       return crypto_ahash_final(mcryptd_req);
+}
+
+int sha1_mb_async_digest(struct ahash_request *req)
+{
+       struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+       struct sha1_mb_ctx *ctx = crypto_ahash_ctx(tfm);
+       struct ahash_request *mcryptd_req = ahash_request_ctx(req);
+       struct mcryptd_ahash *mcryptd_tfm = ctx->mcryptd_tfm;
+
+       memcpy(mcryptd_req, req, sizeof(*req));
+       ahash_request_set_tfm(mcryptd_req, &mcryptd_tfm->base);
+       return crypto_ahash_digest(mcryptd_req);
+}
+
+static int sha1_mb_async_init_tfm(struct crypto_tfm *tfm)
+{
+       struct mcryptd_ahash *mcryptd_tfm;
+       struct sha1_mb_ctx *ctx = crypto_tfm_ctx(tfm);
+       struct mcryptd_hash_ctx *mctx;
+
+       mcryptd_tfm = mcryptd_alloc_ahash("__intel_sha1-mb", 0, 0);
+       if (IS_ERR(mcryptd_tfm))
+               return PTR_ERR(mcryptd_tfm);
+       mctx = crypto_ahash_ctx(&mcryptd_tfm->base);
+       mctx->alg_state = &sha1_mb_alg_state;
+       ctx->mcryptd_tfm = mcryptd_tfm;
+       crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
+                                sizeof(struct ahash_request) +
+                                crypto_ahash_reqsize(&mcryptd_tfm->base));
+
+       return 0;
+}
+
+static void sha1_mb_async_exit_tfm(struct crypto_tfm *tfm)
+{
+       struct sha1_mb_ctx *ctx = crypto_tfm_ctx(tfm);
+
+       mcryptd_free_ahash(ctx->mcryptd_tfm);
+}
+
+static struct ahash_alg sha1_mb_async_alg = {
+       .init           = sha1_mb_async_init,
+       .update         = sha1_mb_async_update,
+       .final          = sha1_mb_async_final,
+       .finup          = sha1_mb_async_finup,
+       .digest         = sha1_mb_async_digest,
+       .halg = {
+               .digestsize     = SHA1_DIGEST_SIZE,
+               .base = {
+                       .cra_name               = "sha1",
+                       .cra_driver_name        = "sha1_mb",
+                       .cra_priority           = 200,
+                       .cra_flags              = CRYPTO_ALG_TYPE_AHASH | CRYPTO_ALG_ASYNC,
+                       .cra_blocksize          = SHA1_BLOCK_SIZE,
+                       .cra_type               = &crypto_ahash_type,
+                       .cra_module             = THIS_MODULE,
+                       .cra_list               = LIST_HEAD_INIT(sha1_mb_async_alg.halg.base.cra_list),
+                       .cra_init               = sha1_mb_async_init_tfm,
+                       .cra_exit               = sha1_mb_async_exit_tfm,
+                       .cra_ctxsize            = sizeof(struct sha1_mb_ctx),
+                       .cra_alignmask          = 0,
+               },
+       },
+};
+
+unsigned long sha1_mb_flusher(struct mcryptd_alg_cstate *cstate)
+{
+       struct mcryptd_hash_request_ctx *rctx;
+       unsigned long cur_time;
+       unsigned long next_flush = 0;
+       struct sha1_hash_ctx *sha_ctx;
+
+
+       cur_time = jiffies;
+
+       while (!list_empty(&cstate->work_list)) {
+               rctx = list_entry(cstate->work_list.next,
+                               struct mcryptd_hash_request_ctx, waiter);
+               if time_before(cur_time, rctx->tag.expire)
+                       break;
+               kernel_fpu_begin();
+               sha_ctx = (struct sha1_hash_ctx *) sha1_ctx_mgr_flush(cstate->mgr);
+               kernel_fpu_end();
+               if (!sha_ctx) {
+                       pr_err("sha1_mb error: nothing got flushed for non-empty list\n");
+                       break;
+               }
+               rctx = cast_hash_to_mcryptd_ctx(sha_ctx);
+               sha_finish_walk(&rctx, cstate, true);
+               sha_complete_job(rctx, cstate, 0);
+       }
+
+       if (!list_empty(&cstate->work_list)) {
+               rctx = list_entry(cstate->work_list.next,
+                               struct mcryptd_hash_request_ctx, waiter);
+               /* get the hash context and then flush time */
+               next_flush = rctx->tag.expire;
+               mcryptd_arm_flusher(cstate, get_delay(next_flush));
+       }
+       return next_flush;
+}
+
+static int __init sha1_mb_mod_init(void)
+{
+
+       int cpu;
+       int err;
+       struct mcryptd_alg_cstate *cpu_state;
+
+       /* check for dependent cpu features */
+       if (!boot_cpu_has(X86_FEATURE_AVX2) ||
+           !boot_cpu_has(X86_FEATURE_BMI2))
+               return -ENODEV;
+
+       /* initialize multibuffer structures */
+       sha1_mb_alg_state.alg_cstate = alloc_percpu(struct mcryptd_alg_cstate);
+
+       sha1_job_mgr_init = sha1_mb_mgr_init_avx2;
+       sha1_job_mgr_submit = sha1_mb_mgr_submit_avx2;
+       sha1_job_mgr_flush = sha1_mb_mgr_flush_avx2;
+       sha1_job_mgr_get_comp_job = sha1_mb_mgr_get_comp_job_avx2;
+
+       if (!sha1_mb_alg_state.alg_cstate)
+               return -ENOMEM;
+       for_each_possible_cpu(cpu) {
+               cpu_state = per_cpu_ptr(sha1_mb_alg_state.alg_cstate, cpu);
+               cpu_state->next_flush = 0;
+               cpu_state->next_seq_num = 0;
+               cpu_state->flusher_engaged = false;
+               INIT_DELAYED_WORK(&cpu_state->flush, mcryptd_flusher);
+               cpu_state->cpu = cpu;
+               cpu_state->alg_state = &sha1_mb_alg_state;
+               cpu_state->mgr = (struct sha1_ctx_mgr *) kzalloc(sizeof(struct sha1_ctx_mgr), GFP_KERNEL);
+               if (!cpu_state->mgr)
+                       goto err2;
+               sha1_ctx_mgr_init(cpu_state->mgr);
+               INIT_LIST_HEAD(&cpu_state->work_list);
+               spin_lock_init(&cpu_state->work_lock);
+       }
+       sha1_mb_alg_state.flusher = &sha1_mb_flusher;
+
+       err = crypto_register_shash(&sha1_mb_shash_alg);
+       if (err)
+               goto err2;
+       err = crypto_register_ahash(&sha1_mb_async_alg);
+       if (err)
+               goto err1;
+
+
+       return 0;
+err1:
+       crypto_unregister_shash(&sha1_mb_shash_alg);
+err2:
+       for_each_possible_cpu(cpu) {
+               cpu_state = per_cpu_ptr(sha1_mb_alg_state.alg_cstate, cpu);
+               kfree(cpu_state->mgr);
+       }
+       free_percpu(sha1_mb_alg_state.alg_cstate);
+       return -ENODEV;
+}
+
+static void __exit sha1_mb_mod_fini(void)
+{
+       int cpu;
+       struct mcryptd_alg_cstate *cpu_state;
+
+       crypto_unregister_ahash(&sha1_mb_async_alg);
+       crypto_unregister_shash(&sha1_mb_shash_alg);
+       for_each_possible_cpu(cpu) {
+               cpu_state = per_cpu_ptr(sha1_mb_alg_state.alg_cstate, cpu);
+               kfree(cpu_state->mgr);
+       }
+       free_percpu(sha1_mb_alg_state.alg_cstate);
+}
+
+module_init(sha1_mb_mod_init);
+module_exit(sha1_mb_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA1 Secure Hash Algorithm, multi buffer accelerated");
+
+MODULE_ALIAS("sha1");