* [PATCH v4 0/7] crypto: SHA1 multibuffer implementation
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
@ 2014-07-11 20:32 ` Tim Chen
2014-07-11 20:32 ` [PATCH v4 1/7] crypto: SHA1 multibuffer crypto hash infrastructure Tim Chen
` (6 subsequent siblings)
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:32 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
Herbert,
I've updated my implementation from v3 to flush the jobs early when cpu goes to idle.
The flush routine was moved out of the notifier path to the crypto
thread. To check that there's no other jobs running, I've added the
nr_running_cpu function to obtain the information. I've also cleaned
up the shash_ahash_mcryptd_digest to mcryptd.c as you have suggested.
Please note also that a separate bug fix to the crypto scatter gather list walk
for the null string needs to be incorporated, which I encountered during
my testing: http://marc.info/?l=linux-crypto-vger&m=140503429412699&w=2
In this patch series, we introduce the multi-buffer crypto algorithm on
x86_64 and apply it to SHA1 hash computation. The multi-buffer technique
takes advantage of the 8 data lanes in the AVX2 registers and allows
computation to be performed on data from multiple jobs in parallel.
This allows us to parallelize computations when data inter-dependency in
a single crypto job prevents us to fully parallelize our computations.
The algorithm can be extended to other hashing and encryption schemes
in the future.
On multi-buffer SHA1 computation with AVX2, we see throughput increase
up to 2.2x over the existing x86_64 single buffer AVX2 algorithm.
The multi-buffer crypto algorithm is described in the following paper:
Processing Multiple Buffers in Parallel to Increase Performance on
Intel® Architecture Processors
http://www.intel.com/content/www/us/en/communications/communications-ia-multi-buffer-paper.html
The outline of the algorithm is sketched below:
Any driver requesting the crypto service will place an async crypto
request on the workqueue. The multi-buffer crypto daemon will pull
request from work queue and put each request in an empty data lane for
multi-buffer crypto computation. When all the empty lanes are filled,
computation will commence on the jobs in parallel and the job with the
shortest remaining buffer will get completed and be returned. To prevent
prolonged stall when there is no new jobs arriving, we will flush a
crypto job if it has not been completed after a maximum allowable delay,
or when cpu becomes idle and cpu cycles become available.
The multi-buffer algorithm necessitates mapping multiple scatter gather
buffers to linear addresses simultaneously. The crypto daemon may need
to sleep and yield the cpu to work on something else from time to time.
We made a change to not use kmap_atomic to do scatter-gather buffer
mapping and take advantage of the fact that we can directly translate
address the buffer's address to its linear address with x86_64.
To accommodate the fragmented nature of scatter-gather, we will keep
submitting the next scatter-buffer fragment for a job for multi-buffer
computation until a job is completed and no more buffer fragments remain.
At that time we will pull a new job to fill the now empty data slot.
We call a get_completed_job function to check whether there are other
jobs that have been completed when we job when we have no new job arrival
to prevent extraneous delay in returning any completed jobs.
The multi-buffer algorithm should be used for cases where crypto jobs
submissions are at a reasonable high rate. For low crypto job submission
rate, this algorithm will not be beneficial. The reason is at low rate,
we do not fill out the data lanes before flushing the jobs instead of
processing them with all the data lanes full. We will miss the benefit
of parallel computation, and adding delay to the processing of the crypto
job at the same time. Some tuning of the maximum latency parameter may
be needed to get the best performance.
Note that the tcrypt SHA1 speed test, we wait for a previous job to
be completed before submitting a new job. Hence this is not a valid
test for multi-buffer algorithm as it requires multiple outstanding jobs
submitted to fill the all data lanes to be effective (i.e. 8 outstanding
jobs for the AVX2 case).
Feedbacks and testings will be most welcomed.
Tim Chen
Change log:
v4
1. Move the early flush of jobs when cpu becomes idle to crypto thread.
2. Move shash_ahash_mcryptd_digest to mcryptd.c
v3
1. Add notifier to multi-buffer algorithm to flush job when the cpu
goes to idle to take advantage of available cpu cycles.
2. Clean up of error messages.
http://marc.info/?l=linux-crypto-vger&m=140252063401632&w=2
v2
1. Change the sha1 crypto walk to use the new crypto_ahash_walk
interface for proper kmap.
2. Drop the hack that map buffer in crypto_hash_walk without kmap_atomic
as the new crypto_ahash_walk interface is merged.
3. Reorganize some
of the mcryptd hash interface code from ahash.c to mcryptd.c
http://marc.info/?l=linux-crypto-vger&m=140088627927559&w=2
v1
refer to: http://www.spinics.net/lists/linux-crypto/msg10993.html
Tim Chen (7):
crypto: SHA1 multibuffer crypto hash infrastructure
crypto: SHA1 multibuffer algorithm data structures
crypto: SHA1 multibuffer submit and flush routines for AVX2
crypto: SHA1 multibuffer crypto computation (x8 AVX2)
crypto: SHA1 multibuffer scheduler
sched: add function nr_running_cpu to expose number of tasks running
on cpu
crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha-mb/Makefile | 11 +
arch/x86/crypto/sha-mb/sha1_mb.c | 1004 ++++++++++++++++++++++
arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S | 287 +++++++
arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S | 327 +++++++
arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c | 64 ++
arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S | 228 +++++
arch/x86/crypto/sha-mb/sha1_x8_avx2.S | 472 ++++++++++
arch/x86/crypto/sha-mb/sha_mb_ctx.h | 136 +++
arch/x86/crypto/sha-mb/sha_mb_mgr.h | 110 +++
crypto/Kconfig | 30 +
crypto/Makefile | 1 +
crypto/mcryptd.c | 622 ++++++++++++++
include/crypto/internal/hash.h | 9 +
include/crypto/mcryptd.h | 110 +++
include/linux/sched.h | 1 +
kernel/sched/core.c | 8 +
17 files changed, 3422 insertions(+)
create mode 100644 arch/x86/crypto/sha-mb/Makefile
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb.c
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S
create mode 100644 arch/x86/crypto/sha-mb/sha1_x8_avx2.S
create mode 100644 arch/x86/crypto/sha-mb/sha_mb_ctx.h
create mode 100644 arch/x86/crypto/sha-mb/sha_mb_mgr.h
create mode 100644 crypto/mcryptd.c
create mode 100644 include/crypto/mcryptd.h
--
1.7.11.7
^ permalink raw reply [flat|nested] 35+ messages in thread
* [PATCH v4 1/7] crypto: SHA1 multibuffer crypto hash infrastructure
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
2014-07-11 20:32 ` [PATCH v4 0/7] crypto: SHA1 multibuffer implementation Tim Chen
@ 2014-07-11 20:32 ` Tim Chen
2014-07-11 20:32 ` [PATCH v4 2/7] crypto: SHA1 multibuffer algorithm data structures Tim Chen
` (5 subsequent siblings)
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:32 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch introduces the multi-buffer crypto daemon which is responsible
for submitting crypto jobs in a work queue to the responsible multi-buffer
crypto algorithm. The idea of the multi-buffer algorihtm is to put
data streams from multiple jobs in a wide (AVX2) register and then
take advantage of SIMD instructions to do crypto computation on several
buffers simultaneously.
The multi-buffer crypto daemon is also responsbile for flushing the
remaining buffers to complete the computation if no new buffers arrive
for a while.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
crypto/Kconfig | 30 ++
crypto/Makefile | 1 +
crypto/mcryptd.c | 622 +++++++++++++++++++++++++++++++++++++++++
include/crypto/internal/hash.h | 9 +
include/crypto/mcryptd.h | 109 ++++++++
5 files changed, 771 insertions(+)
create mode 100644 crypto/mcryptd.c
create mode 100644 include/crypto/mcryptd.h
diff --git a/crypto/Kconfig b/crypto/Kconfig
index ce4012a..5d3c711 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -157,6 +157,20 @@ config CRYPTO_CRYPTD
converts an arbitrary synchronous software crypto algorithm
into an asynchronous algorithm that executes in a kernel thread.
+config CRYPTO_MCRYPTD
+ tristate "Software async multi-buffer crypto daemon"
+ select CRYPTO_BLKCIPHER
+ select CRYPTO_HASH
+ select CRYPTO_MANAGER
+ select CRYPTO_WORKQUEUE
+ help
+ This is a generic software asynchronous crypto daemon that
+ provides the kernel thread to assist multi-buffer crypto
+ algorithms for submitting jobs and flushing jobs in multi-buffer
+ crypto algorithms. Multi-buffer crypto algorithms are executed
+ in the context of this kernel thread and drivers can post
+ their crypto request asyncrhously and process by this daemon.
+
config CRYPTO_AUTHENC
tristate "Authenc support"
select CRYPTO_AEAD
@@ -547,6 +561,22 @@ config CRYPTO_SHA1_PPC
This is the powerpc hardware accelerated implementation of the
SHA-1 secure hash standard (FIPS 180-1/DFIPS 180-2).
+config CRYPTO_SHA1_MB
+ tristate "SHA1 digest algorithm (x86_64 Multi-Buffer, Experimental)"
+ depends on X86 && 64BIT
+ select CRYPTO_SHA1
+ select CRYPTO_HASH
+ select CRYPTO_MCRYPTD
+ help
+ SHA-1 secure hash standard (FIPS 180-1/DFIPS 180-2) implemented
+ using multi-buffer technique. This algorithm computes on
+ multiple data lanes concurrently with SIMD instructions for
+ better throughput. It should not be enabled by default but
+ used when there is significant amount of work to keep the keep
+ the data lanes filled to get performance benefit. If the data
+ lanes remain unfilled, a flush operation will be initiated to
+ process the crypto jobs, adding a slight latency.
+
config CRYPTO_SHA256
tristate "SHA224 and SHA256 digest algorithm"
select CRYPTO_HASH
diff --git a/crypto/Makefile b/crypto/Makefile
index 38e64231..27dbbad 100644
--- a/crypto/Makefile
+++ b/crypto/Makefile
@@ -60,6 +60,7 @@ obj-$(CONFIG_CRYPTO_GCM) += gcm.o
obj-$(CONFIG_CRYPTO_CCM) += ccm.o
obj-$(CONFIG_CRYPTO_PCRYPT) += pcrypt.o
obj-$(CONFIG_CRYPTO_CRYPTD) += cryptd.o
+obj-$(CONFIG_CRYPTO_MCRYPTD) += mcryptd.o
obj-$(CONFIG_CRYPTO_DES) += des_generic.o
obj-$(CONFIG_CRYPTO_FCRYPT) += fcrypt.o
obj-$(CONFIG_CRYPTO_BLOWFISH) += blowfish_generic.o
diff --git a/crypto/mcryptd.c b/crypto/mcryptd.c
new file mode 100644
index 0000000..39fa72b
--- /dev/null
+++ b/crypto/mcryptd.c
@@ -0,0 +1,622 @@
+/*
+ * Software multibuffer async crypto daemon.
+ *
+ * Copyright (c) 2014 Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * Adapted from crypto daemon.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/internal/hash.h>
+#include <crypto/internal/aead.h>
+#include <crypto/mcryptd.h>
+#include <crypto/crypto_wq.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/scatterlist.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/hardirq.h>
+
+#define MCRYPTD_MAX_CPU_QLEN 100
+#define MCRYPTD_BATCH 9
+
+static void *mcryptd_alloc_instance(struct crypto_alg *alg, unsigned int head,
+ unsigned int tail);
+
+struct hashd_instance_ctx {
+ struct crypto_shash_spawn spawn;
+ struct mcryptd_queue *queue;
+};
+
+static void mcryptd_queue_worker(struct work_struct *work);
+
+static int mcryptd_init_queue(struct mcryptd_queue *queue,
+ unsigned int max_cpu_qlen)
+{
+ int cpu;
+ struct mcryptd_cpu_queue *cpu_queue;
+
+ queue->cpu_queue = alloc_percpu(struct mcryptd_cpu_queue);
+ pr_debug("mqueue:%p mcryptd_cpu_queue %p\n", queue, queue->cpu_queue);
+ if (!queue->cpu_queue)
+ return -ENOMEM;
+ for_each_possible_cpu(cpu) {
+ cpu_queue = per_cpu_ptr(queue->cpu_queue, cpu);
+ pr_debug("cpu_queue #%d %p\n", cpu, queue->cpu_queue);
+ crypto_init_queue(&cpu_queue->queue, max_cpu_qlen);
+ INIT_WORK(&cpu_queue->work, mcryptd_queue_worker);
+ }
+ return 0;
+}
+
+static void mcryptd_fini_queue(struct mcryptd_queue *queue)
+{
+ int cpu;
+ struct mcryptd_cpu_queue *cpu_queue;
+
+ for_each_possible_cpu(cpu) {
+ cpu_queue = per_cpu_ptr(queue->cpu_queue, cpu);
+ BUG_ON(cpu_queue->queue.qlen);
+ }
+ free_percpu(queue->cpu_queue);
+}
+
+static int mcryptd_enqueue_request(struct mcryptd_queue *queue,
+ struct crypto_async_request *request,
+ struct mcryptd_hash_request_ctx *rctx)
+{
+ int cpu, err;
+ struct mcryptd_cpu_queue *cpu_queue;
+
+ cpu = get_cpu();
+ cpu_queue = this_cpu_ptr(queue->cpu_queue);
+ rctx->tag.cpu = cpu;
+
+ err = crypto_enqueue_request(&cpu_queue->queue, request);
+ pr_debug("enqueue request: cpu %d cpu_queue %p request %p\n",
+ cpu, cpu_queue, request);
+ queue_work_on(cpu, kcrypto_wq, &cpu_queue->work);
+ put_cpu();
+
+ return err;
+}
+
+/* Called in workqueue context, do one real cryption work (via
+ * req->complete) and reschedule itself if there are more work to
+ * do. */
+static void mcryptd_queue_worker(struct work_struct *work)
+{
+ struct mcryptd_cpu_queue *cpu_queue;
+ struct crypto_async_request *req, *backlog;
+ int i;
+
+ /*
+ * Need to loop through more than once for multi-buffer to
+ * be effective.
+ */
+
+ cpu_queue = container_of(work, struct mcryptd_cpu_queue, work);
+ for (i = 0; i < MCRYPTD_BATCH; ++i) {
+ /*
+ * preempt_disable/enable is used to prevent
+ * being preempted by mcryptd_enqueue_request()
+ */
+ local_bh_disable();
+ preempt_disable();
+ backlog = crypto_get_backlog(&cpu_queue->queue);
+ req = crypto_dequeue_request(&cpu_queue->queue);
+ preempt_enable();
+ local_bh_enable();
+
+ if (!req)
+ return;
+
+ if (backlog)
+ backlog->complete(backlog, -EINPROGRESS);
+ req->complete(req, 0);
+ if (!cpu_queue->queue.qlen)
+ return;
+ }
+ if (cpu_queue->queue.qlen)
+ queue_work(kcrypto_wq, &cpu_queue->work);
+}
+
+void mcryptd_flusher(struct work_struct *__work)
+{
+ struct mcryptd_alg_cstate *alg_cpu_state;
+ struct mcryptd_alg_state *alg_state;
+
+ alg_cpu_state = container_of(to_delayed_work(__work),
+ struct mcryptd_alg_cstate, flush);
+ alg_state = alg_cpu_state->alg_state;
+ if (alg_cpu_state->cpu != smp_processor_id())
+ pr_debug("mcryptd error: work on cpu %d, should be cpu %d\n",
+ smp_processor_id(), alg_cpu_state->cpu);
+
+ alg_state->flusher(alg_cpu_state);
+}
+EXPORT_SYMBOL_GPL(mcryptd_flusher);
+
+static inline struct mcryptd_queue *mcryptd_get_queue(struct crypto_tfm *tfm)
+{
+ struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
+ struct mcryptd_instance_ctx *ictx = crypto_instance_ctx(inst);
+
+ return ictx->queue;
+}
+
+static void *mcryptd_alloc_instance(struct crypto_alg *alg, unsigned int head,
+ unsigned int tail)
+{
+ char *p;
+ struct crypto_instance *inst;
+ int err;
+
+ p = kzalloc(head + sizeof(*inst) + tail, GFP_KERNEL);
+ if (!p)
+ return ERR_PTR(-ENOMEM);
+
+ inst = (void *)(p + head);
+
+ err = -ENAMETOOLONG;
+ if (snprintf(inst->alg.cra_driver_name, CRYPTO_MAX_ALG_NAME,
+ "mcryptd(%s)", alg->cra_driver_name) >= CRYPTO_MAX_ALG_NAME)
+ goto out_free_inst;
+
+ memcpy(inst->alg.cra_name, alg->cra_name, CRYPTO_MAX_ALG_NAME);
+
+ inst->alg.cra_priority = alg->cra_priority + 50;
+ inst->alg.cra_blocksize = alg->cra_blocksize;
+ inst->alg.cra_alignmask = alg->cra_alignmask;
+
+out:
+ return p;
+
+out_free_inst:
+ kfree(p);
+ p = ERR_PTR(err);
+ goto out;
+}
+
+static int mcryptd_hash_init_tfm(struct crypto_tfm *tfm)
+{
+ struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
+ struct hashd_instance_ctx *ictx = crypto_instance_ctx(inst);
+ struct crypto_shash_spawn *spawn = &ictx->spawn;
+ struct mcryptd_hash_ctx *ctx = crypto_tfm_ctx(tfm);
+ struct crypto_shash *hash;
+
+ hash = crypto_spawn_shash(spawn);
+ if (IS_ERR(hash))
+ return PTR_ERR(hash);
+
+ ctx->child = hash;
+ crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
+ sizeof(struct mcryptd_hash_request_ctx) +
+ crypto_shash_descsize(hash));
+ return 0;
+}
+
+static void mcryptd_hash_exit_tfm(struct crypto_tfm *tfm)
+{
+ struct mcryptd_hash_ctx *ctx = crypto_tfm_ctx(tfm);
+
+ crypto_free_shash(ctx->child);
+}
+
+static int mcryptd_hash_setkey(struct crypto_ahash *parent,
+ const u8 *key, unsigned int keylen)
+{
+ struct mcryptd_hash_ctx *ctx = crypto_ahash_ctx(parent);
+ struct crypto_shash *child = ctx->child;
+ int err;
+
+ crypto_shash_clear_flags(child, CRYPTO_TFM_REQ_MASK);
+ crypto_shash_set_flags(child, crypto_ahash_get_flags(parent) &
+ CRYPTO_TFM_REQ_MASK);
+ err = crypto_shash_setkey(child, key, keylen);
+ crypto_ahash_set_flags(parent, crypto_shash_get_flags(child) &
+ CRYPTO_TFM_RES_MASK);
+ return err;
+}
+
+static int mcryptd_hash_enqueue(struct ahash_request *req,
+ crypto_completion_t complete)
+{
+ int ret;
+
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+ struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
+ struct mcryptd_queue *queue =
+ mcryptd_get_queue(crypto_ahash_tfm(tfm));
+
+ rctx->complete = req->base.complete;
+ req->base.complete = complete;
+
+ ret = mcryptd_enqueue_request(queue, &req->base, rctx);
+
+ return ret;
+}
+
+static void mcryptd_hash_init(struct crypto_async_request *req_async, int err)
+{
+ struct mcryptd_hash_ctx *ctx = crypto_tfm_ctx(req_async->tfm);
+ struct crypto_shash *child = ctx->child;
+ struct ahash_request *req = ahash_request_cast(req_async);
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+ struct shash_desc *desc = &rctx->desc;
+
+ if (unlikely(err == -EINPROGRESS))
+ goto out;
+
+ desc->tfm = child;
+ desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
+
+ err = crypto_shash_init(desc);
+
+ req->base.complete = rctx->complete;
+
+out:
+ local_bh_disable();
+ rctx->complete(&req->base, err);
+ local_bh_enable();
+}
+
+static int mcryptd_hash_init_enqueue(struct ahash_request *req)
+{
+ return mcryptd_hash_enqueue(req, mcryptd_hash_init);
+}
+
+static void mcryptd_hash_update(struct crypto_async_request *req_async, int err)
+{
+ struct ahash_request *req = ahash_request_cast(req_async);
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+
+ if (unlikely(err == -EINPROGRESS))
+ goto out;
+
+ err = shash_ahash_mcryptd_update(req, &rctx->desc);
+ if (err) {
+ req->base.complete = rctx->complete;
+ goto out;
+ }
+
+ return;
+out:
+ local_bh_disable();
+ rctx->complete(&req->base, err);
+ local_bh_enable();
+}
+
+static int mcryptd_hash_update_enqueue(struct ahash_request *req)
+{
+ return mcryptd_hash_enqueue(req, mcryptd_hash_update);
+}
+
+static void mcryptd_hash_final(struct crypto_async_request *req_async, int err)
+{
+ struct ahash_request *req = ahash_request_cast(req_async);
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+
+ if (unlikely(err == -EINPROGRESS))
+ goto out;
+
+ err = shash_ahash_mcryptd_final(req, &rctx->desc);
+ if (err) {
+ req->base.complete = rctx->complete;
+ goto out;
+ }
+
+ return;
+out:
+ local_bh_disable();
+ rctx->complete(&req->base, err);
+ local_bh_enable();
+}
+
+static int mcryptd_hash_final_enqueue(struct ahash_request *req)
+{
+ return mcryptd_hash_enqueue(req, mcryptd_hash_final);
+}
+
+static void mcryptd_hash_finup(struct crypto_async_request *req_async, int err)
+{
+ struct ahash_request *req = ahash_request_cast(req_async);
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+
+ if (unlikely(err == -EINPROGRESS))
+ goto out;
+
+ err = shash_ahash_mcryptd_finup(req, &rctx->desc);
+
+ if (err) {
+ req->base.complete = rctx->complete;
+ goto out;
+ }
+
+ return;
+out:
+ local_bh_disable();
+ rctx->complete(&req->base, err);
+ local_bh_enable();
+}
+
+static int mcryptd_hash_finup_enqueue(struct ahash_request *req)
+{
+ return mcryptd_hash_enqueue(req, mcryptd_hash_finup);
+}
+
+static void mcryptd_hash_digest(struct crypto_async_request *req_async, int err)
+{
+ struct mcryptd_hash_ctx *ctx = crypto_tfm_ctx(req_async->tfm);
+ struct crypto_shash *child = ctx->child;
+ struct ahash_request *req = ahash_request_cast(req_async);
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+ struct shash_desc *desc = &rctx->desc;
+
+ if (unlikely(err == -EINPROGRESS))
+ goto out;
+
+ desc->tfm = child;
+ desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP; /* check this again */
+
+ err = shash_ahash_mcryptd_digest(req, desc);
+
+ if (err) {
+ req->base.complete = rctx->complete;
+ goto out;
+ }
+
+ return;
+out:
+ local_bh_disable();
+ rctx->complete(&req->base, err);
+ local_bh_enable();
+}
+
+static int mcryptd_hash_digest_enqueue(struct ahash_request *req)
+{
+ return mcryptd_hash_enqueue(req, mcryptd_hash_digest);
+}
+
+static int mcryptd_hash_export(struct ahash_request *req, void *out)
+{
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+
+ return crypto_shash_export(&rctx->desc, out);
+}
+
+static int mcryptd_hash_import(struct ahash_request *req, const void *in)
+{
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+
+ return crypto_shash_import(&rctx->desc, in);
+}
+
+static int mcryptd_create_hash(struct crypto_template *tmpl, struct rtattr **tb,
+ struct mcryptd_queue *queue)
+{
+ struct hashd_instance_ctx *ctx;
+ struct ahash_instance *inst;
+ struct shash_alg *salg;
+ struct crypto_alg *alg;
+ int err;
+
+ salg = shash_attr_alg(tb[1], 0, 0);
+ if (IS_ERR(salg))
+ return PTR_ERR(salg);
+
+ alg = &salg->base;
+ pr_debug("crypto: mcryptd hash alg: %s\n", alg->cra_name);
+ inst = mcryptd_alloc_instance(alg, ahash_instance_headroom(),
+ sizeof(*ctx));
+ err = PTR_ERR(inst);
+ if (IS_ERR(inst))
+ goto out_put_alg;
+
+ ctx = ahash_instance_ctx(inst);
+ ctx->queue = queue;
+
+ err = crypto_init_shash_spawn(&ctx->spawn, salg,
+ ahash_crypto_instance(inst));
+ if (err)
+ goto out_free_inst;
+
+ inst->alg.halg.base.cra_flags = CRYPTO_ALG_ASYNC;
+
+ inst->alg.halg.digestsize = salg->digestsize;
+ inst->alg.halg.base.cra_ctxsize = sizeof(struct mcryptd_hash_ctx);
+
+ inst->alg.halg.base.cra_init = mcryptd_hash_init_tfm;
+ inst->alg.halg.base.cra_exit = mcryptd_hash_exit_tfm;
+
+ inst->alg.init = mcryptd_hash_init_enqueue;
+ inst->alg.update = mcryptd_hash_update_enqueue;
+ inst->alg.final = mcryptd_hash_final_enqueue;
+ inst->alg.finup = mcryptd_hash_finup_enqueue;
+ inst->alg.export = mcryptd_hash_export;
+ inst->alg.import = mcryptd_hash_import;
+ inst->alg.setkey = mcryptd_hash_setkey;
+ inst->alg.digest = mcryptd_hash_digest_enqueue;
+
+ err = ahash_register_instance(tmpl, inst);
+ if (err) {
+ crypto_drop_shash(&ctx->spawn);
+out_free_inst:
+ kfree(inst);
+ }
+
+out_put_alg:
+ crypto_mod_put(alg);
+ return err;
+}
+
+static struct mcryptd_queue mqueue;
+
+static int mcryptd_create(struct crypto_template *tmpl, struct rtattr **tb)
+{
+ struct crypto_attr_type *algt;
+
+ algt = crypto_get_attr_type(tb);
+ if (IS_ERR(algt))
+ return PTR_ERR(algt);
+
+ switch (algt->type & algt->mask & CRYPTO_ALG_TYPE_MASK) {
+ case CRYPTO_ALG_TYPE_DIGEST:
+ return mcryptd_create_hash(tmpl, tb, &mqueue);
+ break;
+ }
+
+ return -EINVAL;
+}
+
+static void mcryptd_free(struct crypto_instance *inst)
+{
+ struct mcryptd_instance_ctx *ctx = crypto_instance_ctx(inst);
+ struct hashd_instance_ctx *hctx = crypto_instance_ctx(inst);
+
+ switch (inst->alg.cra_flags & CRYPTO_ALG_TYPE_MASK) {
+ case CRYPTO_ALG_TYPE_AHASH:
+ crypto_drop_shash(&hctx->spawn);
+ kfree(ahash_instance(inst));
+ return;
+ default:
+ crypto_drop_spawn(&ctx->spawn);
+ kfree(inst);
+ }
+}
+
+static struct crypto_template mcryptd_tmpl = {
+ .name = "mcryptd",
+ .create = mcryptd_create,
+ .free = mcryptd_free,
+ .module = THIS_MODULE,
+};
+
+struct mcryptd_ahash *mcryptd_alloc_ahash(const char *alg_name,
+ u32 type, u32 mask)
+{
+ char mcryptd_alg_name[CRYPTO_MAX_ALG_NAME];
+ struct crypto_ahash *tfm;
+
+ if (snprintf(mcryptd_alg_name, CRYPTO_MAX_ALG_NAME,
+ "mcryptd(%s)", alg_name) >= CRYPTO_MAX_ALG_NAME)
+ return ERR_PTR(-EINVAL);
+ tfm = crypto_alloc_ahash(mcryptd_alg_name, type, mask);
+ if (IS_ERR(tfm))
+ return ERR_CAST(tfm);
+ if (tfm->base.__crt_alg->cra_module != THIS_MODULE) {
+ crypto_free_ahash(tfm);
+ return ERR_PTR(-EINVAL);
+ }
+
+ return __mcryptd_ahash_cast(tfm);
+}
+EXPORT_SYMBOL_GPL(mcryptd_alloc_ahash);
+
+int shash_ahash_mcryptd_digest(struct ahash_request *req,
+ struct shash_desc *desc)
+{
+ int err;
+
+ err = crypto_shash_init(desc) ?:
+ shash_ahash_mcryptd_finup(req, desc);
+
+ return err;
+}
+EXPORT_SYMBOL_GPL(shash_ahash_mcryptd_digest);
+
+int shash_ahash_mcryptd_update(struct ahash_request *req,
+ struct shash_desc *desc)
+{
+ struct crypto_shash *tfm = desc->tfm;
+ struct shash_alg *shash = crypto_shash_alg(tfm);
+
+ /* alignment is to be done by multi-buffer crypto algorithm if needed */
+
+ return shash->update(desc, NULL, 0);
+}
+EXPORT_SYMBOL_GPL(shash_ahash_mcryptd_update);
+
+int shash_ahash_mcryptd_finup(struct ahash_request *req,
+ struct shash_desc *desc)
+{
+ struct crypto_shash *tfm = desc->tfm;
+ struct shash_alg *shash = crypto_shash_alg(tfm);
+
+ /* alignment is to be done by multi-buffer crypto algorithm if needed */
+
+ return shash->finup(desc, NULL, 0, req->result);
+}
+EXPORT_SYMBOL_GPL(shash_ahash_mcryptd_finup);
+
+int shash_ahash_mcryptd_final(struct ahash_request *req,
+ struct shash_desc *desc)
+{
+ struct crypto_shash *tfm = desc->tfm;
+ struct shash_alg *shash = crypto_shash_alg(tfm);
+
+ /* alignment is to be done by multi-buffer crypto algorithm if needed */
+
+ return shash->final(desc, req->result);
+}
+EXPORT_SYMBOL_GPL(shash_ahash_mcryptd_final);
+
+struct crypto_shash *mcryptd_ahash_child(struct mcryptd_ahash *tfm)
+{
+ struct mcryptd_hash_ctx *ctx = crypto_ahash_ctx(&tfm->base);
+
+ return ctx->child;
+}
+EXPORT_SYMBOL_GPL(mcryptd_ahash_child);
+
+struct shash_desc *mcryptd_shash_desc(struct ahash_request *req)
+{
+ struct mcryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
+ return &rctx->desc;
+}
+EXPORT_SYMBOL_GPL(mcryptd_shash_desc);
+
+void mcryptd_free_ahash(struct mcryptd_ahash *tfm)
+{
+ crypto_free_ahash(&tfm->base);
+}
+EXPORT_SYMBOL_GPL(mcryptd_free_ahash);
+
+
+static int __init mcryptd_init(void)
+{
+ int err;
+
+ err = mcryptd_init_queue(&mqueue, MCRYPTD_MAX_CPU_QLEN);
+ if (err)
+ return err;
+
+ err = crypto_register_template(&mcryptd_tmpl);
+ if (err)
+ mcryptd_fini_queue(&mqueue);
+
+ return err;
+}
+
+static void __exit mcryptd_exit(void)
+{
+ mcryptd_fini_queue(&mqueue);
+ crypto_unregister_template(&mcryptd_tmpl);
+}
+
+subsys_initcall(mcryptd_init);
+module_exit(mcryptd_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Software async multibuffer crypto daemon");
diff --git a/include/crypto/internal/hash.h b/include/crypto/internal/hash.h
index 9b6f32a..3b4af1d 100644
--- a/include/crypto/internal/hash.h
+++ b/include/crypto/internal/hash.h
@@ -117,6 +117,15 @@ int shash_ahash_update(struct ahash_request *req, struct shash_desc *desc);
int shash_ahash_finup(struct ahash_request *req, struct shash_desc *desc);
int shash_ahash_digest(struct ahash_request *req, struct shash_desc *desc);
+int shash_ahash_mcryptd_update(struct ahash_request *req,
+ struct shash_desc *desc);
+int shash_ahash_mcryptd_final(struct ahash_request *req,
+ struct shash_desc *desc);
+int shash_ahash_mcryptd_finup(struct ahash_request *req,
+ struct shash_desc *desc);
+int shash_ahash_mcryptd_digest(struct ahash_request *req,
+ struct shash_desc *desc);
+
int crypto_init_shash_ops_async(struct crypto_tfm *tfm);
static inline void *crypto_ahash_ctx(struct crypto_ahash *tfm)
diff --git a/include/crypto/mcryptd.h b/include/crypto/mcryptd.h
new file mode 100644
index 0000000..b2b9055
--- /dev/null
+++ b/include/crypto/mcryptd.h
@@ -0,0 +1,109 @@
+/*
+ * Software async multibuffer crypto daemon headers
+ *
+ * Author:
+ * Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * Copyright (c) 2014, Intel Corporation.
+ */
+
+#ifndef _CRYPTO_MCRYPT_H
+#define _CRYPTO_MCRYPT_H
+
+#include <linux/crypto.h>
+#include <linux/kernel.h>
+#include <crypto/hash.h>
+
+struct mcryptd_ahash {
+ struct crypto_ahash base;
+};
+
+static inline struct mcryptd_ahash *__mcryptd_ahash_cast(
+ struct crypto_ahash *tfm)
+{
+ return (struct mcryptd_ahash *)tfm;
+}
+
+struct mcryptd_cpu_queue {
+ struct crypto_queue queue;
+ struct work_struct work;
+};
+
+struct mcryptd_queue {
+ struct mcryptd_cpu_queue __percpu *cpu_queue;
+};
+
+struct mcryptd_instance_ctx {
+ struct crypto_spawn spawn;
+ struct mcryptd_queue *queue;
+};
+
+struct mcryptd_hash_ctx {
+ struct crypto_shash *child;
+ struct mcryptd_alg_state *alg_state;
+};
+
+struct mcryptd_tag {
+ /* seq number of request */
+ unsigned seq_num;
+ /* arrival time of request */
+ unsigned long arrival;
+ unsigned long expire;
+ int cpu;
+};
+
+struct mcryptd_hash_request_ctx {
+ struct list_head waiter;
+ crypto_completion_t complete;
+ struct mcryptd_tag tag;
+ struct crypto_hash_walk walk;
+ u8 *out;
+ int flag;
+ struct shash_desc desc;
+};
+
+struct mcryptd_ahash *mcryptd_alloc_ahash(const char *alg_name,
+ u32 type, u32 mask);
+struct crypto_shash *mcryptd_ahash_child(struct mcryptd_ahash *tfm);
+struct shash_desc *mcryptd_shash_desc(struct ahash_request *req);
+void mcryptd_free_ahash(struct mcryptd_ahash *tfm);
+void mcryptd_flusher(struct work_struct *work);
+
+enum mcryptd_req_type {
+ MCRYPTD_NONE,
+ MCRYPTD_UPDATE,
+ MCRYPTD_FINUP,
+ MCRYPTD_DIGEST,
+ MCRYPTD_FINAL
+};
+
+struct mcryptd_alg_cstate {
+ unsigned long next_flush;
+ unsigned next_seq_num;
+ bool flusher_engaged;
+ struct delayed_work flush;
+ int cpu;
+ struct mcryptd_alg_state *alg_state;
+ void *mgr;
+ spinlock_t work_lock;
+ struct list_head work_list;
+};
+
+struct mcryptd_alg_state {
+ struct mcryptd_alg_cstate __percpu *alg_cstate;
+ unsigned long (*flusher)(struct mcryptd_alg_cstate *cstate);
+};
+
+/* return delay in jiffies from current time */
+static inline unsigned long get_delay(unsigned long t)
+{
+ long delay;
+
+ delay = (long) t - (long) jiffies;
+ if (delay <= 0)
+ return 0;
+ else
+ return (unsigned long) delay;
+}
+
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 2/7] crypto: SHA1 multibuffer algorithm data structures
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
2014-07-11 20:32 ` [PATCH v4 0/7] crypto: SHA1 multibuffer implementation Tim Chen
2014-07-11 20:32 ` [PATCH v4 1/7] crypto: SHA1 multibuffer crypto hash infrastructure Tim Chen
@ 2014-07-11 20:32 ` Tim Chen
2014-07-11 20:32 ` [PATCH v4 3/7] crypto: SHA1 multibuffer submit and flush routines for AVX2 Tim Chen
` (4 subsequent siblings)
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:32 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch introduces the data structures and prototypes of functions
needed for computing SHA1 hash using multi-buffer. Included are the
structures of the multi-buffer SHA1 job, job scheduler in C and x86
assembly.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S | 287 ++++++++++++++++++++++++
arch/x86/crypto/sha-mb/sha_mb_ctx.h | 136 +++++++++++
arch/x86/crypto/sha-mb/sha_mb_mgr.h | 110 +++++++++
3 files changed, 533 insertions(+)
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S
create mode 100644 arch/x86/crypto/sha-mb/sha_mb_ctx.h
create mode 100644 arch/x86/crypto/sha-mb/sha_mb_mgr.h
diff --git a/arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S b/arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S
new file mode 100644
index 0000000..86688c6
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S
@@ -0,0 +1,287 @@
+/*
+ * Header file for multi buffer SHA1 algorithm data structure
+ *
+ * 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:
+ * James Guilford <james.guilford@intel.com>
+ * 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.
+ */
+
+# Macros for defining data structures
+
+# Usage example
+
+#START_FIELDS # JOB_AES
+### name size align
+#FIELD _plaintext, 8, 8 # pointer to plaintext
+#FIELD _ciphertext, 8, 8 # pointer to ciphertext
+#FIELD _IV, 16, 8 # IV
+#FIELD _keys, 8, 8 # pointer to keys
+#FIELD _len, 4, 4 # length in bytes
+#FIELD _status, 4, 4 # status enumeration
+#FIELD _user_data, 8, 8 # pointer to user data
+#UNION _union, size1, align1, \
+# size2, align2, \
+# size3, align3, \
+# ...
+#END_FIELDS
+#%assign _JOB_AES_size _FIELD_OFFSET
+#%assign _JOB_AES_align _STRUCT_ALIGN
+
+#########################################################################
+
+# Alternate "struc-like" syntax:
+# STRUCT job_aes2
+# RES_Q .plaintext, 1
+# RES_Q .ciphertext, 1
+# RES_DQ .IV, 1
+# RES_B .nested, _JOB_AES_SIZE, _JOB_AES_ALIGN
+# RES_U .union, size1, align1, \
+# size2, align2, \
+# ...
+# ENDSTRUCT
+# # Following only needed if nesting
+# %assign job_aes2_size _FIELD_OFFSET
+# %assign job_aes2_align _STRUCT_ALIGN
+#
+# RES_* macros take a name, a count and an optional alignment.
+# The count in in terms of the base size of the macro, and the
+# default alignment is the base size.
+# The macros are:
+# Macro Base size
+# RES_B 1
+# RES_W 2
+# RES_D 4
+# RES_Q 8
+# RES_DQ 16
+# RES_Y 32
+# RES_Z 64
+#
+# RES_U defines a union. It's arguments are a name and two or more
+# pairs of "size, alignment"
+#
+# The two assigns are only needed if this structure is being nested
+# within another. Even if the assigns are not done, one can still use
+# STRUCT_NAME_size as the size of the structure.
+#
+# Note that for nesting, you still need to assign to STRUCT_NAME_size.
+#
+# The differences between this and using "struc" directly are that each
+# type is implicitly aligned to its natural length (although this can be
+# over-ridden with an explicit third parameter), and that the structure
+# is padded at the end to its overall alignment.
+#
+
+#########################################################################
+
+#ifndef _SHA1_MB_MGR_DATASTRUCT_ASM_
+#define _SHA1_MB_MGR_DATASTRUCT_ASM_
+
+## START_FIELDS
+.macro START_FIELDS
+ _FIELD_OFFSET = 0
+ _STRUCT_ALIGN = 0
+.endm
+
+## FIELD name size align
+.macro FIELD name size align
+ _FIELD_OFFSET = (_FIELD_OFFSET + (\align) - 1) & (~ ((\align)-1))
+ \name = _FIELD_OFFSET
+ _FIELD_OFFSET = _FIELD_OFFSET + (\size)
+.if (\align > _STRUCT_ALIGN)
+ _STRUCT_ALIGN = \align
+.endif
+.endm
+
+## END_FIELDS
+.macro END_FIELDS
+ _FIELD_OFFSET = (_FIELD_OFFSET + _STRUCT_ALIGN-1) & (~ (_STRUCT_ALIGN-1))
+.endm
+
+########################################################################
+
+.macro STRUCT p1
+START_FIELDS
+.struc \p1
+.endm
+
+.macro ENDSTRUCT
+ tmp = _FIELD_OFFSET
+ END_FIELDS
+ tmp = (_FIELD_OFFSET - %%tmp)
+.if (tmp > 0)
+ .lcomm tmp
+.endif
+.endstruc
+.endm
+
+## RES_int name size align
+.macro RES_int p1 p2 p3
+ name = \p1
+ size = \p2
+ align = .\p3
+
+ _FIELD_OFFSET = (_FIELD_OFFSET + (align) - 1) & (~ ((align)-1))
+.align align
+.lcomm name size
+ _FIELD_OFFSET = _FIELD_OFFSET + (size)
+.if (align > _STRUCT_ALIGN)
+ _STRUCT_ALIGN = align
+.endif
+.endm
+
+
+
+# macro RES_B name, size [, align]
+.macro RES_B _name, _size, _align=1
+RES_int _name _size _align
+.endm
+
+# macro RES_W name, size [, align]
+.macro RES_W _name, _size, _align=2
+RES_int _name 2*(_size) _align
+.endm
+
+# macro RES_D name, size [, align]
+.macro RES_D _name, _size, _align=4
+RES_int _name 4*(_size) _align
+.endm
+
+# macro RES_Q name, size [, align]
+.macro RES_Q _name, _size, _align=8
+RES_int _name 8*(_size) _align
+.endm
+
+# macro RES_DQ name, size [, align]
+.macro RES_DQ _name, _size, _align=16
+RES_int _name 16*(_size) _align
+.endm
+
+# macro RES_Y name, size [, align]
+.macro RES_Y _name, _size, _align=32
+RES_int _name 32*(_size) _align
+.endm
+
+# macro RES_Z name, size [, align]
+.macro RES_Z _name, _size, _align=64
+RES_int _name 64*(_size) _align
+.endm
+
+
+#endif
+
+########################################################################
+#### Define constants
+########################################################################
+
+########################################################################
+#### Define SHA1 Out Of Order Data Structures
+########################################################################
+
+START_FIELDS # LANE_DATA
+### name size align
+FIELD _job_in_lane, 8, 8 # pointer to job object
+END_FIELDS
+
+_LANE_DATA_size = _FIELD_OFFSET
+_LANE_DATA_align = _STRUCT_ALIGN
+
+########################################################################
+
+START_FIELDS # SHA1_ARGS_X8
+### name size align
+FIELD _digest, 4*5*8, 16 # transposed digest
+FIELD _data_ptr, 8*8, 8 # array of pointers to data
+END_FIELDS
+
+_SHA1_ARGS_X4_size = _FIELD_OFFSET
+_SHA1_ARGS_X4_align = _STRUCT_ALIGN
+_SHA1_ARGS_X8_size = _FIELD_OFFSET
+_SHA1_ARGS_X8_align = _STRUCT_ALIGN
+
+########################################################################
+
+START_FIELDS # MB_MGR
+### name size align
+FIELD _args, _SHA1_ARGS_X4_size, _SHA1_ARGS_X4_align
+FIELD _lens, 4*8, 8
+FIELD _unused_lanes, 8, 8
+FIELD _ldata, _LANE_DATA_size*8, _LANE_DATA_align
+END_FIELDS
+
+_MB_MGR_size = _FIELD_OFFSET
+_MB_MGR_align = _STRUCT_ALIGN
+
+_args_digest = _args + _digest
+_args_data_ptr = _args + _data_ptr
+
+
+########################################################################
+#### Define constants
+########################################################################
+
+#define STS_UNKNOWN 0
+#define STS_BEING_PROCESSED 1
+#define STS_COMPLETED 2
+
+########################################################################
+#### Define JOB_SHA1 structure
+########################################################################
+
+START_FIELDS # JOB_SHA1
+
+### name size align
+FIELD _buffer, 8, 8 # pointer to buffer
+FIELD _len, 4, 4 # length in bytes
+FIELD _result_digest, 5*4, 32 # Digest (output)
+FIELD _status, 4, 4
+FIELD _user_data, 8, 8
+END_FIELDS
+
+_JOB_SHA1_size = _FIELD_OFFSET
+_JOB_SHA1_align = _STRUCT_ALIGN
diff --git a/arch/x86/crypto/sha-mb/sha_mb_ctx.h b/arch/x86/crypto/sha-mb/sha_mb_ctx.h
new file mode 100644
index 0000000..e36069d
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha_mb_ctx.h
@@ -0,0 +1,136 @@
+/*
+ * Header file for multi buffer SHA context
+ *
+ * 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.
+ */
+
+#ifndef _SHA_MB_CTX_INTERNAL_H
+#define _SHA_MB_CTX_INTERNAL_H
+
+#include "sha_mb_mgr.h"
+
+#define HASH_UPDATE 0x00
+#define HASH_FIRST 0x01
+#define HASH_LAST 0x02
+#define HASH_ENTIRE 0x03
+#define HASH_DONE 0x04
+#define HASH_FINAL 0x08
+
+#define HASH_CTX_STS_IDLE 0x00
+#define HASH_CTX_STS_PROCESSING 0x01
+#define HASH_CTX_STS_LAST 0x02
+#define HASH_CTX_STS_COMPLETE 0x04
+
+enum hash_ctx_error {
+ HASH_CTX_ERROR_NONE = 0,
+ HASH_CTX_ERROR_INVALID_FLAGS = -1,
+ HASH_CTX_ERROR_ALREADY_PROCESSING = -2,
+ HASH_CTX_ERROR_ALREADY_COMPLETED = -3,
+
+#ifdef HASH_CTX_DEBUG
+ HASH_CTX_ERROR_DEBUG_DIGEST_MISMATCH = -4,
+#endif
+};
+
+
+#define hash_ctx_user_data(ctx) ((ctx)->user_data)
+#define hash_ctx_digest(ctx) ((ctx)->job.result_digest)
+#define hash_ctx_processing(ctx) ((ctx)->status & HASH_CTX_STS_PROCESSING)
+#define hash_ctx_complete(ctx) ((ctx)->status == HASH_CTX_STS_COMPLETE)
+#define hash_ctx_status(ctx) ((ctx)->status)
+#define hash_ctx_error(ctx) ((ctx)->error)
+#define hash_ctx_init(ctx) \
+ do { \
+ (ctx)->error = HASH_CTX_ERROR_NONE; \
+ (ctx)->status = HASH_CTX_STS_COMPLETE; \
+ } while (0)
+
+
+/* Hash Constants and Typedefs */
+#define SHA1_DIGEST_LENGTH 5
+#define SHA1_LOG2_BLOCK_SIZE 6
+
+#define SHA1_PADLENGTHFIELD_SIZE 8
+
+#ifdef SHA_MB_DEBUG
+#define assert(expr) \
+do { \
+ if (unlikely(!(expr))) { \
+ printk(KERN_ERR "Assertion failed! %s,%s,%s,line=%d\n", \
+ #expr, __FILE__, __func__, __LINE__); \
+ } \
+} while (0)
+#else
+#define assert(expr) do {} while (0)
+#endif
+
+struct sha1_ctx_mgr {
+ struct sha1_mb_mgr mgr;
+};
+
+/* typedef struct sha1_ctx_mgr sha1_ctx_mgr; */
+
+struct sha1_hash_ctx {
+ /* Must be at struct offset 0 */
+ struct job_sha1 job;
+ /* status flag */
+ int status;
+ /* error flag */
+ int error;
+
+ uint32_t total_length;
+ const void *incoming_buffer;
+ uint32_t incoming_buffer_length;
+ uint8_t partial_block_buffer[SHA1_BLOCK_SIZE * 2];
+ uint32_t partial_block_buffer_length;
+ void *user_data;
+};
+
+#endif
diff --git a/arch/x86/crypto/sha-mb/sha_mb_mgr.h b/arch/x86/crypto/sha-mb/sha_mb_mgr.h
new file mode 100644
index 0000000..08ad1a9
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha_mb_mgr.h
@@ -0,0 +1,110 @@
+/*
+ * Header file for multi buffer SHA1 algorithm manager
+ *
+ * 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:
+ * James Guilford <james.guilford@intel.com>
+ * 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.
+ */
+#ifndef __SHA_MB_MGR_H
+#define __SHA_MB_MGR_H
+
+
+#include <linux/types.h>
+
+#define NUM_SHA1_DIGEST_WORDS 5
+
+enum job_sts { STS_UNKNOWN = 0,
+ STS_BEING_PROCESSED = 1,
+ STS_COMPLETED = 2,
+ STS_INTERNAL_ERROR = 3,
+ STS_ERROR = 4
+};
+
+struct job_sha1 {
+ u8 *buffer;
+ u32 len;
+ u32 result_digest[NUM_SHA1_DIGEST_WORDS] __aligned(32);
+ enum job_sts status;
+ void *user_data;
+};
+
+/* SHA1 out-of-order scheduler */
+
+/* typedef uint32_t sha1_digest_array[5][8]; */
+
+struct sha1_args_x8 {
+ uint32_t digest[5][8];
+ uint8_t *data_ptr[8];
+};
+
+struct sha1_lane_data {
+ struct job_sha1 *job_in_lane;
+};
+
+struct sha1_mb_mgr {
+ struct sha1_args_x8 args;
+
+ uint32_t lens[8];
+
+ /* each byte is index (0...7) of unused lanes */
+ uint64_t unused_lanes;
+ /* byte 4 is set to FF as a flag */
+ struct sha1_lane_data ldata[8];
+};
+
+
+#define SHA1_MB_MGR_NUM_LANES_AVX2 8
+
+void sha1_mb_mgr_init_avx2(struct sha1_mb_mgr *state);
+struct job_sha1 *sha1_mb_mgr_submit_avx2(struct sha1_mb_mgr *state,
+ struct job_sha1 *job);
+struct job_sha1 *sha1_mb_mgr_flush_avx2(struct sha1_mb_mgr *state);
+struct job_sha1 *sha1_mb_mgr_get_comp_job_avx2(struct sha1_mb_mgr *state);
+
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 3/7] crypto: SHA1 multibuffer submit and flush routines for AVX2
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
` (2 preceding siblings ...)
2014-07-11 20:32 ` [PATCH v4 2/7] crypto: SHA1 multibuffer algorithm data structures Tim Chen
@ 2014-07-11 20:32 ` Tim Chen
2014-07-11 20:32 ` [PATCH v4 4/7] crypto: SHA1 multibuffer crypto computation (x8 AVX2) Tim Chen
` (3 subsequent siblings)
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:32 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch introduces the routines used to submit and flush buffers
belonging to SHA1 crypto jobs to the SHA1 multibuffer algorithm. It is
implemented mostly in assembly optimized with AVX2 instructions.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S | 327 +++++++++++++++++++++++
arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c | 64 +++++
arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S | 228 ++++++++++++++++
3 files changed, 619 insertions(+)
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S
diff --git a/arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S b/arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S
new file mode 100644
index 0000000..85c4e1c
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S
@@ -0,0 +1,327 @@
+/*
+ * Flush routine for SHA1 multibuffer
+ *
+ * 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:
+ * James Guilford <james.guilford@intel.com>
+ * 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.
+ */
+#include <linux/linkage.h>
+#include "sha1_mb_mgr_datastruct.S"
+
+
+.extern sha1_x8_avx2
+
+# LINUX register definitions
+#define arg1 %rdi
+#define arg2 %rsi
+
+# Common definitions
+#define state arg1
+#define job arg2
+#define len2 arg2
+
+# idx must be a register not clobbered by sha1_x8_avx2
+#define idx %r8
+#define DWORD_idx %r8d
+
+#define unused_lanes %rbx
+#define lane_data %rbx
+#define tmp2 %rbx
+#define tmp2_w %ebx
+
+#define job_rax %rax
+#define tmp1 %rax
+#define size_offset %rax
+#define tmp %rax
+#define start_offset %rax
+
+#define tmp3 %arg1
+
+#define extra_blocks %arg2
+#define p %arg2
+
+
+# STACK_SPACE needs to be an odd multiple of 8
+_XMM_SAVE_SIZE = 10*16
+_GPR_SAVE_SIZE = 8*8
+_ALIGN_SIZE = 8
+
+_XMM_SAVE = 0
+_GPR_SAVE = _XMM_SAVE + _XMM_SAVE_SIZE
+STACK_SPACE = _GPR_SAVE + _GPR_SAVE_SIZE + _ALIGN_SIZE
+
+.macro LABEL prefix n
+\prefix\n\():
+.endm
+
+.macro JNE_SKIP i
+jne skip_\i
+.endm
+
+.altmacro
+.macro SET_OFFSET _offset
+offset = \_offset
+.endm
+.noaltmacro
+
+# JOB* sha1_mb_mgr_flush_avx2(MB_MGR *state)
+# arg 1 : rcx : state
+ENTRY(sha1_mb_mgr_flush_avx2)
+ mov %rsp, %r10
+ sub $STACK_SPACE, %rsp
+ and $~31, %rsp
+ mov %rbx, _GPR_SAVE(%rsp)
+ mov %r10, _GPR_SAVE+8*1(%rsp) #save rsp
+ mov %rbp, _GPR_SAVE+8*3(%rsp)
+ mov %r12, _GPR_SAVE+8*4(%rsp)
+ mov %r13, _GPR_SAVE+8*5(%rsp)
+ mov %r14, _GPR_SAVE+8*6(%rsp)
+ mov %r15, _GPR_SAVE+8*7(%rsp)
+
+ # If bit (32+3) is set, then all lanes are empty
+ mov _unused_lanes(state), unused_lanes
+ bt $32+3, unused_lanes
+ jc return_null
+
+ # find a lane with a non-null job
+ xor idx, idx
+ offset = (_ldata + 1 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne one(%rip), idx
+ offset = (_ldata + 2 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne two(%rip), idx
+ offset = (_ldata + 3 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne three(%rip), idx
+ offset = (_ldata + 4 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne four(%rip), idx
+ offset = (_ldata + 5 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne five(%rip), idx
+ offset = (_ldata + 6 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne six(%rip), idx
+ offset = (_ldata + 7 * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+ cmovne seven(%rip), idx
+
+ # copy idx to empty lanes
+copy_lane_data:
+ offset = (_args + _data_ptr)
+ mov offset(state,idx,8), tmp
+
+ I = 0
+.rep 8
+ offset = (_ldata + I * _LANE_DATA_size + _job_in_lane)
+ cmpq $0, offset(state)
+.altmacro
+ JNE_SKIP %I
+ offset = (_args + _data_ptr + 8*I)
+ mov tmp, offset(state)
+ offset = (_lens + 4*I)
+ movl $0xFFFFFFFF, offset(state)
+LABEL skip_ %I
+ I = (I+1)
+.noaltmacro
+.endr
+
+ # Find min length
+ vmovdqa _lens+0*16(state), %xmm0
+ vmovdqa _lens+1*16(state), %xmm1
+
+ vpminud %xmm1, %xmm0, %xmm2 # xmm2 has {D,C,B,A}
+ vpalignr $8, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,D,C}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has {x,x,E,F}
+ vpalignr $4, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,x,E}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has min value in low dword
+
+ vmovd %xmm2, DWORD_idx
+ mov idx, len2
+ and $0xF, idx
+ shr $4, len2
+ jz len_is_0
+
+ vpand clear_low_nibble(%rip), %xmm2, %xmm2
+ vpshufd $0, %xmm2, %xmm2
+
+ vpsubd %xmm2, %xmm0, %xmm0
+ vpsubd %xmm2, %xmm1, %xmm1
+
+ vmovdqa %xmm0, _lens+0*16(state)
+ vmovdqa %xmm1, _lens+1*16(state)
+
+ # "state" and "args" are the same address, arg1
+ # len is arg2
+ call sha1_x8_avx2
+ # state and idx are intact
+
+
+len_is_0:
+ # process completed job "idx"
+ imul $_LANE_DATA_size, idx, lane_data
+ lea _ldata(state, lane_data), lane_data
+
+ mov _job_in_lane(lane_data), job_rax
+ movq $0, _job_in_lane(lane_data)
+ movl $STS_COMPLETED, _status(job_rax)
+ mov _unused_lanes(state), unused_lanes
+ shl $4, unused_lanes
+ or idx, unused_lanes
+ mov unused_lanes, _unused_lanes(state)
+
+ movl $0xFFFFFFFF, _lens(state, idx, 4)
+
+ vmovd _args_digest(state , idx, 4) , %xmm0
+ vpinsrd $1, _args_digest+1*32(state, idx, 4), %xmm0, %xmm0
+ vpinsrd $2, _args_digest+2*32(state, idx, 4), %xmm0, %xmm0
+ vpinsrd $3, _args_digest+3*32(state, idx, 4), %xmm0, %xmm0
+ movl _args_digest+4*32(state, idx, 4), tmp2_w
+
+ vmovdqu %xmm0, _result_digest(job_rax)
+ offset = (_result_digest + 1*16)
+ mov tmp2_w, offset(job_rax)
+
+return:
+
+ mov _GPR_SAVE(%rsp), %rbx
+ mov _GPR_SAVE+8*1(%rsp), %r10 #saved rsp
+ mov _GPR_SAVE+8*3(%rsp), %rbp
+ mov _GPR_SAVE+8*4(%rsp), %r12
+ mov _GPR_SAVE+8*5(%rsp), %r13
+ mov _GPR_SAVE+8*6(%rsp), %r14
+ mov _GPR_SAVE+8*7(%rsp), %r15
+ mov %r10, %rsp
+
+ ret
+
+return_null:
+ xor job_rax, job_rax
+ jmp return
+ENDPROC(sha1_mb_mgr_flush_avx2)
+
+
+#################################################################
+
+.align 16
+ENTRY(sha1_mb_mgr_get_comp_job_avx2)
+ push %rbx
+
+ ## if bit 32+3 is set, then all lanes are empty
+ mov _unused_lanes(state), unused_lanes
+ bt $(32+3), unused_lanes
+ jc .return_null
+
+ # Find min length
+ vmovdqa _lens(state), %xmm0
+ vmovdqa _lens+1*16(state), %xmm1
+
+ vpminud %xmm1, %xmm0, %xmm2 # xmm2 has {D,C,B,A}
+ vpalignr $8, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,D,C}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has {x,x,E,F}
+ vpalignr $4, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,x,E}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has min value in low dword
+
+ vmovd %xmm2, DWORD_idx
+ test $~0xF, idx
+ jnz .return_null
+
+ # process completed job "idx"
+ imul $_LANE_DATA_size, idx, lane_data
+ lea _ldata(state, lane_data), lane_data
+
+ mov _job_in_lane(lane_data), job_rax
+ movq $0, _job_in_lane(lane_data)
+ movl $STS_COMPLETED, _status(job_rax)
+ mov _unused_lanes(state), unused_lanes
+ shl $4, unused_lanes
+ or idx, unused_lanes
+ mov unused_lanes, _unused_lanes(state)
+
+ movl $0xFFFFFFFF, _lens(state, idx, 4)
+
+ vmovd _args_digest(state, idx, 4), %xmm0
+ vpinsrd $1, _args_digest+1*32(state, idx, 4), %xmm0, %xmm0
+ vpinsrd $2, _args_digest+2*32(state, idx, 4), %xmm0, %xmm0
+ vpinsrd $3, _args_digest+3*32(state, idx, 4), %xmm0, %xmm0
+ movl _args_digest+4*32(state, idx, 4), tmp2_w
+
+ vmovdqu %xmm0, _result_digest(job_rax)
+ movl tmp2_w, _result_digest+1*16(job_rax)
+
+ pop %rbx
+
+ ret
+
+.return_null:
+ xor job_rax, job_rax
+ pop %rbx
+ ret
+ENDPROC(sha1_mb_mgr_get_comp_job_avx2)
+
+.data
+
+.align 16
+clear_low_nibble:
+.octa 0x000000000000000000000000FFFFFFF0
+one:
+.quad 1
+two:
+.quad 2
+three:
+.quad 3
+four:
+.quad 4
+five:
+.quad 5
+six:
+.quad 6
+seven:
+.quad 7
diff --git a/arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c b/arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c
new file mode 100644
index 0000000..4ca7e16
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c
@@ -0,0 +1,64 @@
+/*
+ * Initialization code for multi buffer SHA1 algorithm for AVX2
+ *
+ * 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.
+ */
+
+#include "sha_mb_mgr.h"
+
+void sha1_mb_mgr_init_avx2(struct sha1_mb_mgr *state)
+{
+ unsigned int j;
+ state->unused_lanes = 0xF76543210;
+ for (j = 0; j < 8; j++) {
+ state->lens[j] = 0xFFFFFFFF;
+ state->ldata[j].job_in_lane = NULL;
+ }
+}
diff --git a/arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S b/arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S
new file mode 100644
index 0000000..2ab9560
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S
@@ -0,0 +1,228 @@
+/*
+ * Buffer submit code for multi buffer SHA1 algorithm
+ *
+ * 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:
+ * James Guilford <james.guilford@intel.com>
+ * 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.
+ */
+
+#include <linux/linkage.h>
+#include "sha1_mb_mgr_datastruct.S"
+
+
+.extern sha1_x8_avx
+
+# LINUX register definitions
+arg1 = %rdi
+arg2 = %rsi
+size_offset = %rcx
+tmp2 = %rcx
+extra_blocks = %rdx
+
+# Common definitions
+#define state arg1
+#define job %rsi
+#define len2 arg2
+#define p2 arg2
+
+# idx must be a register not clobberred by sha1_x8_avx2
+idx = %r8
+DWORD_idx = %r8d
+last_len = %r8
+
+p = %r11
+start_offset = %r11
+
+unused_lanes = %rbx
+BYTE_unused_lanes = %bl
+
+job_rax = %rax
+len = %rax
+DWORD_len = %eax
+
+lane = %rbp
+tmp3 = %rbp
+
+tmp = %r9
+DWORD_tmp = %r9d
+
+lane_data = %r10
+
+# STACK_SPACE needs to be an odd multiple of 8
+STACK_SPACE = 8*8 + 16*10 + 8
+
+# JOB* submit_mb_mgr_submit_avx2(MB_MGR *state, job_sha1 *job)
+# arg 1 : rcx : state
+# arg 2 : rdx : job
+ENTRY(sha1_mb_mgr_submit_avx2)
+
+ mov %rsp, %r10
+ sub $STACK_SPACE, %rsp
+ and $~31, %rsp
+
+ mov %rbx, (%rsp)
+ mov %r10, 8*2(%rsp) #save old rsp
+ mov %rbp, 8*3(%rsp)
+ mov %r12, 8*4(%rsp)
+ mov %r13, 8*5(%rsp)
+ mov %r14, 8*6(%rsp)
+ mov %r15, 8*7(%rsp)
+
+ mov _unused_lanes(state), unused_lanes
+ mov unused_lanes, lane
+ and $0xF, lane
+ shr $4, unused_lanes
+ imul $_LANE_DATA_size, lane, lane_data
+ movl $STS_BEING_PROCESSED, _status(job)
+ lea _ldata(state, lane_data), lane_data
+ mov unused_lanes, _unused_lanes(state)
+ movl _len(job), DWORD_len
+
+ mov job, _job_in_lane(lane_data)
+ shl $4, len
+ or lane, len
+
+ movl DWORD_len, _lens(state , lane, 4)
+
+ # Load digest words from result_digest
+ vmovdqu _result_digest(job), %xmm0
+ mov _result_digest+1*16(job), DWORD_tmp
+ vmovd %xmm0, _args_digest(state, lane, 4)
+ vpextrd $1, %xmm0, _args_digest+1*32(state , lane, 4)
+ vpextrd $2, %xmm0, _args_digest+2*32(state , lane, 4)
+ vpextrd $3, %xmm0, _args_digest+3*32(state , lane, 4)
+ movl DWORD_tmp, _args_digest+4*32(state , lane, 4)
+
+ mov _buffer(job), p
+ mov p, _args_data_ptr(state, lane, 8)
+
+ cmp $0xF, unused_lanes
+ jne return_null
+
+start_loop:
+ # Find min length
+ vmovdqa _lens(state), %xmm0
+ vmovdqa _lens+1*16(state), %xmm1
+
+ vpminud %xmm1, %xmm0, %xmm2 # xmm2 has {D,C,B,A}
+ vpalignr $8, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,D,C}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has {x,x,E,F}
+ vpalignr $4, %xmm2, %xmm3, %xmm3 # xmm3 has {x,x,x,E}
+ vpminud %xmm3, %xmm2, %xmm2 # xmm2 has min value in low dword
+
+ vmovd %xmm2, DWORD_idx
+ mov idx, len2
+ and $0xF, idx
+ shr $4, len2
+ jz len_is_0
+
+ vpand clear_low_nibble(%rip), %xmm2, %xmm2
+ vpshufd $0, %xmm2, %xmm2
+
+ vpsubd %xmm2, %xmm0, %xmm0
+ vpsubd %xmm2, %xmm1, %xmm1
+
+ vmovdqa %xmm0, _lens + 0*16(state)
+ vmovdqa %xmm1, _lens + 1*16(state)
+
+
+ # "state" and "args" are the same address, arg1
+ # len is arg2
+ call sha1_x8_avx2
+
+ # state and idx are intact
+
+len_is_0:
+ # process completed job "idx"
+ imul $_LANE_DATA_size, idx, lane_data
+ lea _ldata(state, lane_data), lane_data
+
+ mov _job_in_lane(lane_data), job_rax
+ mov _unused_lanes(state), unused_lanes
+ movq $0, _job_in_lane(lane_data)
+ movl $STS_COMPLETED, _status(job_rax)
+ shl $4, unused_lanes
+ or idx, unused_lanes
+ mov unused_lanes, _unused_lanes(state)
+
+ movl $0xFFFFFFFF, _lens(state, idx, 4)
+
+ vmovd _args_digest(state, idx, 4), %xmm0
+ vpinsrd $1, _args_digest+1*32(state , idx, 4), %xmm0, %xmm0
+ vpinsrd $2, _args_digest+2*32(state , idx, 4), %xmm0, %xmm0
+ vpinsrd $3, _args_digest+3*32(state , idx, 4), %xmm0, %xmm0
+ movl 4*32(state, idx, 4), DWORD_tmp
+
+ vmovdqu %xmm0, _result_digest(job_rax)
+ movl DWORD_tmp, _result_digest+1*16(job_rax)
+
+return:
+
+ mov (%rsp), %rbx
+ mov 8*2(%rsp), %r10 #save old rsp
+ mov 8*3(%rsp), %rbp
+ mov 8*4(%rsp), %r12
+ mov 8*5(%rsp), %r13
+ mov 8*6(%rsp), %r14
+ mov 8*7(%rsp), %r15
+ mov %r10, %rsp
+
+ ret
+
+return_null:
+ xor job_rax, job_rax
+ jmp return
+
+ENDPROC(sha1_mb_mgr_submit_avx2)
+
+.data
+
+.align 16
+clear_low_nibble:
+ .octa 0x000000000000000000000000FFFFFFF0
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 4/7] crypto: SHA1 multibuffer crypto computation (x8 AVX2)
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
` (3 preceding siblings ...)
2014-07-11 20:32 ` [PATCH v4 3/7] crypto: SHA1 multibuffer submit and flush routines for AVX2 Tim Chen
@ 2014-07-11 20:32 ` Tim Chen
2014-07-11 20:33 ` [PATCH v4 5/7] crypto: SHA1 multibuffer scheduler Tim Chen
` (2 subsequent siblings)
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:32 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch introduces the assembly routines to do SHA1 computation on
buffers belonging to serveral jobs at once. The assembly routines are
optimized with AVX2 instructions that have 8 data lanes and using AVX2
registers.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha-mb/sha1_x8_avx2.S | 472 ++++++++++++++++++++++++++++++++++
1 file changed, 472 insertions(+)
create mode 100644 arch/x86/crypto/sha-mb/sha1_x8_avx2.S
diff --git a/arch/x86/crypto/sha-mb/sha1_x8_avx2.S b/arch/x86/crypto/sha-mb/sha1_x8_avx2.S
new file mode 100644
index 0000000..8e1b477
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_x8_avx2.S
@@ -0,0 +1,472 @@
+/*
+ * Multi-buffer SHA1 algorithm hash compute routine
+ *
+ * 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:
+ * James Guilford <james.guilford@intel.com>
+ * 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.
+ */
+
+#include <linux/linkage.h>
+#include "sha1_mb_mgr_datastruct.S"
+
+## code to compute oct SHA1 using SSE-256
+## outer calling routine takes care of save and restore of XMM registers
+
+## Function clobbers: rax, rcx, rdx, rbx, rsi, rdi, r9-r15# ymm0-15
+##
+## Linux clobbers: rax rbx rcx rdx rsi r9 r10 r11 r12 r13 r14 r15
+## Linux preserves: rdi rbp r8
+##
+## clobbers ymm0-15
+
+
+# TRANSPOSE8 r0, r1, r2, r3, r4, r5, r6, r7, t0, t1
+# "transpose" data in {r0...r7} using temps {t0...t1}
+# Input looks like: {r0 r1 r2 r3 r4 r5 r6 r7}
+# r0 = {a7 a6 a5 a4 a3 a2 a1 a0}
+# r1 = {b7 b6 b5 b4 b3 b2 b1 b0}
+# r2 = {c7 c6 c5 c4 c3 c2 c1 c0}
+# r3 = {d7 d6 d5 d4 d3 d2 d1 d0}
+# r4 = {e7 e6 e5 e4 e3 e2 e1 e0}
+# r5 = {f7 f6 f5 f4 f3 f2 f1 f0}
+# r6 = {g7 g6 g5 g4 g3 g2 g1 g0}
+# r7 = {h7 h6 h5 h4 h3 h2 h1 h0}
+#
+# Output looks like: {r0 r1 r2 r3 r4 r5 r6 r7}
+# r0 = {h0 g0 f0 e0 d0 c0 b0 a0}
+# r1 = {h1 g1 f1 e1 d1 c1 b1 a1}
+# r2 = {h2 g2 f2 e2 d2 c2 b2 a2}
+# r3 = {h3 g3 f3 e3 d3 c3 b3 a3}
+# r4 = {h4 g4 f4 e4 d4 c4 b4 a4}
+# r5 = {h5 g5 f5 e5 d5 c5 b5 a5}
+# r6 = {h6 g6 f6 e6 d6 c6 b6 a6}
+# r7 = {h7 g7 f7 e7 d7 c7 b7 a7}
+#
+
+.macro TRANSPOSE8 r0 r1 r2 r3 r4 r5 r6 r7 t0 t1
+ # process top half (r0..r3) {a...d}
+ vshufps $0x44, \r1, \r0, \t0 # t0 = {b5 b4 a5 a4 b1 b0 a1 a0}
+ vshufps $0xEE, \r1, \r0, \r0 # r0 = {b7 b6 a7 a6 b3 b2 a3 a2}
+ vshufps $0x44, \r3, \r2, \t1 # t1 = {d5 d4 c5 c4 d1 d0 c1 c0}
+ vshufps $0xEE, \r3, \r2, \r2 # r2 = {d7 d6 c7 c6 d3 d2 c3 c2}
+ vshufps $0xDD, \t1, \t0, \r3 # r3 = {d5 c5 b5 a5 d1 c1 b1 a1}
+ vshufps $0x88, \r2, \r0, \r1 # r1 = {d6 c6 b6 a6 d2 c2 b2 a2}
+ vshufps $0xDD, \r2, \r0, \r0 # r0 = {d7 c7 b7 a7 d3 c3 b3 a3}
+ vshufps $0x88, \t1, \t0, \t0 # t0 = {d4 c4 b4 a4 d0 c0 b0 a0}
+
+ # use r2 in place of t0
+ # process bottom half (r4..r7) {e...h}
+ vshufps $0x44, \r5, \r4, \r2 # r2 = {f5 f4 e5 e4 f1 f0 e1 e0}
+ vshufps $0xEE, \r5, \r4, \r4 # r4 = {f7 f6 e7 e6 f3 f2 e3 e2}
+ vshufps $0x44, \r7, \r6, \t1 # t1 = {h5 h4 g5 g4 h1 h0 g1 g0}
+ vshufps $0xEE, \r7, \r6, \r6 # r6 = {h7 h6 g7 g6 h3 h2 g3 g2}
+ vshufps $0xDD, \t1, \r2, \r7 # r7 = {h5 g5 f5 e5 h1 g1 f1 e1}
+ vshufps $0x88, \r6, \r4, \r5 # r5 = {h6 g6 f6 e6 h2 g2 f2 e2}
+ vshufps $0xDD, \r6, \r4, \r4 # r4 = {h7 g7 f7 e7 h3 g3 f3 e3}
+ vshufps $0x88, \t1, \r2, \t1 # t1 = {h4 g4 f4 e4 h0 g0 f0 e0}
+
+ vperm2f128 $0x13, \r1, \r5, \r6 # h6...a6
+ vperm2f128 $0x02, \r1, \r5, \r2 # h2...a2
+ vperm2f128 $0x13, \r3, \r7, \r5 # h5...a5
+ vperm2f128 $0x02, \r3, \r7, \r1 # h1...a1
+ vperm2f128 $0x13, \r0, \r4, \r7 # h7...a7
+ vperm2f128 $0x02, \r0, \r4, \r3 # h3...a3
+ vperm2f128 $0x13, \t0, \t1, \r4 # h4...a4
+ vperm2f128 $0x02, \t0, \t1, \r0 # h0...a0
+
+.endm
+##
+## Magic functions defined in FIPS 180-1
+##
+# macro MAGIC_F0 F,B,C,D,T ## F = (D ^ (B & (C ^ D)))
+.macro MAGIC_F0 regF regB regC regD regT
+ vpxor \regD, \regC, \regF
+ vpand \regB, \regF, \regF
+ vpxor \regD, \regF, \regF
+.endm
+
+# macro MAGIC_F1 F,B,C,D,T ## F = (B ^ C ^ D)
+.macro MAGIC_F1 regF regB regC regD regT
+ vpxor \regC, \regD, \regF
+ vpxor \regB, \regF, \regF
+.endm
+
+# macro MAGIC_F2 F,B,C,D,T ## F = ((B & C) | (B & D) | (C & D))
+.macro MAGIC_F2 regF regB regC regD regT
+ vpor \regC, \regB, \regF
+ vpand \regC, \regB, \regT
+ vpand \regD, \regF, \regF
+ vpor \regT, \regF, \regF
+.endm
+
+# macro MAGIC_F3 F,B,C,D,T ## F = (B ^ C ^ D)
+.macro MAGIC_F3 regF regB regC regD regT
+ MAGIC_F1 \regF,\regB,\regC,\regD,\regT
+.endm
+
+# PROLD reg, imm, tmp
+.macro PROLD reg imm tmp
+ vpsrld $(32-\imm), \reg, \tmp
+ vpslld $\imm, \reg, \reg
+ vpor \tmp, \reg, \reg
+.endm
+
+.macro PROLD_nd reg imm tmp src
+ vpsrld $(32-\imm), \src, \tmp
+ vpslld $\imm, \src, \reg
+ vpor \tmp, \reg, \reg
+.endm
+
+.macro SHA1_STEP_00_15 regA regB regC regD regE regT regF memW immCNT MAGIC
+ vpaddd \immCNT, \regE, \regE
+ vpaddd \memW*32(%rsp), \regE, \regE
+ PROLD_nd \regT, 5, \regF, \regA
+ vpaddd \regT, \regE, \regE
+ \MAGIC \regF, \regB, \regC, \regD, \regT
+ PROLD \regB, 30, \regT
+ vpaddd \regF, \regE, \regE
+.endm
+
+.macro SHA1_STEP_16_79 regA regB regC regD regE regT regF memW immCNT MAGIC
+ vpaddd \immCNT, \regE, \regE
+ offset = ((\memW - 14) & 15) * 32
+ vmovdqu offset(%rsp), W14
+ vpxor W14, W16, W16
+ offset = ((\memW - 8) & 15) * 32
+ vpxor offset(%rsp), W16, W16
+ offset = ((\memW - 3) & 15) * 32
+ vpxor offset(%rsp), W16, W16
+ vpsrld $(32-1), W16, \regF
+ vpslld $1, W16, W16
+ vpor W16, \regF, \regF
+
+ ROTATE_W
+
+ offset = ((\memW - 0) & 15) * 32
+ vmovdqu \regF, offset(%rsp)
+ vpaddd \regF, \regE, \regE
+ PROLD_nd \regT, 5, \regF, \regA
+ vpaddd \regT, \regE, \regE
+ \MAGIC \regF,\regB,\regC,\regD,\regT ## FUN = MAGIC_Fi(B,C,D)
+ PROLD \regB,30, \regT
+ vpaddd \regF, \regE, \regE
+.endm
+
+########################################################################
+########################################################################
+########################################################################
+
+## FRAMESZ plus pushes must be an odd multiple of 8
+YMM_SAVE = (15-15)*32
+FRAMESZ = 32*16 + YMM_SAVE
+_YMM = FRAMESZ - YMM_SAVE
+
+#define VMOVPS vmovups
+
+IDX = %rax
+inp0 = %r9
+inp1 = %r10
+inp2 = %r11
+inp3 = %r12
+inp4 = %r13
+inp5 = %r14
+inp6 = %r15
+inp7 = %rcx
+arg1 = %rdi
+arg2 = %rsi
+RSP_SAVE = %rdx
+
+# ymm0 A
+# ymm1 B
+# ymm2 C
+# ymm3 D
+# ymm4 E
+# ymm5 F AA
+# ymm6 T0 BB
+# ymm7 T1 CC
+# ymm8 T2 DD
+# ymm9 T3 EE
+# ymm10 T4 TMP
+# ymm11 T5 FUN
+# ymm12 T6 K
+# ymm13 T7 W14
+# ymm14 T8 W15
+# ymm15 T9 W16
+
+
+A = %ymm0
+B = %ymm1
+C = %ymm2
+D = %ymm3
+E = %ymm4
+F = %ymm5
+T0 = %ymm6
+T1 = %ymm7
+T2 = %ymm8
+T3 = %ymm9
+T4 = %ymm10
+T5 = %ymm11
+T6 = %ymm12
+T7 = %ymm13
+T8 = %ymm14
+T9 = %ymm15
+
+AA = %ymm5
+BB = %ymm6
+CC = %ymm7
+DD = %ymm8
+EE = %ymm9
+TMP = %ymm10
+FUN = %ymm11
+K = %ymm12
+W14 = %ymm13
+W15 = %ymm14
+W16 = %ymm15
+
+.macro ROTATE_ARGS
+ TMP_ = E
+ E = D
+ D = C
+ C = B
+ B = A
+ A = TMP_
+.endm
+
+.macro ROTATE_W
+TMP_ = W16
+W16 = W15
+W15 = W14
+W14 = TMP_
+.endm
+
+# 8 streams x 5 32bit words per digest x 4 bytes per word
+#define DIGEST_SIZE (8*5*4)
+
+.align 32
+
+# void sha1_x8_avx2(void **input_data, UINT128 *digest, UINT32 size)
+# arg 1 : pointer to array[4] of pointer to input data
+# arg 2 : size (in blocks) ;; assumed to be >= 1
+#
+ENTRY(sha1_x8_avx2)
+
+ push RSP_SAVE
+
+ #save rsp
+ mov %rsp, RSP_SAVE
+ sub $FRAMESZ, %rsp
+
+ #align rsp to 32 Bytes
+ and $~0x1F, %rsp
+
+ ## Initialize digests
+ vmovdqu 0*32(arg1), A
+ vmovdqu 1*32(arg1), B
+ vmovdqu 2*32(arg1), C
+ vmovdqu 3*32(arg1), D
+ vmovdqu 4*32(arg1), E
+
+ ## transpose input onto stack
+ mov _data_ptr+0*8(arg1),inp0
+ mov _data_ptr+1*8(arg1),inp1
+ mov _data_ptr+2*8(arg1),inp2
+ mov _data_ptr+3*8(arg1),inp3
+ mov _data_ptr+4*8(arg1),inp4
+ mov _data_ptr+5*8(arg1),inp5
+ mov _data_ptr+6*8(arg1),inp6
+ mov _data_ptr+7*8(arg1),inp7
+
+ xor IDX, IDX
+lloop:
+ vmovdqu PSHUFFLE_BYTE_FLIP_MASK(%rip), F
+ I=0
+.rep 2
+ VMOVPS (inp0, IDX), T0
+ VMOVPS (inp1, IDX), T1
+ VMOVPS (inp2, IDX), T2
+ VMOVPS (inp3, IDX), T3
+ VMOVPS (inp4, IDX), T4
+ VMOVPS (inp5, IDX), T5
+ VMOVPS (inp6, IDX), T6
+ VMOVPS (inp7, IDX), T7
+
+ TRANSPOSE8 T0, T1, T2, T3, T4, T5, T6, T7, T8, T9
+ vpshufb F, T0, T0
+ vmovdqu T0, (I*8)*32(%rsp)
+ vpshufb F, T1, T1
+ vmovdqu T1, (I*8+1)*32(%rsp)
+ vpshufb F, T2, T2
+ vmovdqu T2, (I*8+2)*32(%rsp)
+ vpshufb F, T3, T3
+ vmovdqu T3, (I*8+3)*32(%rsp)
+ vpshufb F, T4, T4
+ vmovdqu T4, (I*8+4)*32(%rsp)
+ vpshufb F, T5, T5
+ vmovdqu T5, (I*8+5)*32(%rsp)
+ vpshufb F, T6, T6
+ vmovdqu T6, (I*8+6)*32(%rsp)
+ vpshufb F, T7, T7
+ vmovdqu T7, (I*8+7)*32(%rsp)
+ add $32, IDX
+ I = (I+1)
+.endr
+ # save old digests
+ vmovdqu A,AA
+ vmovdqu B,BB
+ vmovdqu C,CC
+ vmovdqu D,DD
+ vmovdqu E,EE
+
+##
+## perform 0-79 steps
+##
+ vmovdqu K00_19(%rip), K
+## do rounds 0...15
+ I = 0
+.rep 16
+ SHA1_STEP_00_15 A,B,C,D,E, TMP,FUN, I, K, MAGIC_F0
+ ROTATE_ARGS
+ I = (I+1)
+.endr
+
+## do rounds 16...19
+ vmovdqu ((16 - 16) & 15) * 32 (%rsp), W16
+ vmovdqu ((16 - 15) & 15) * 32 (%rsp), W15
+.rep 4
+ SHA1_STEP_16_79 A,B,C,D,E, TMP,FUN, I, K, MAGIC_F0
+ ROTATE_ARGS
+ I = (I+1)
+.endr
+
+## do rounds 20...39
+ vmovdqu K20_39(%rip), K
+.rep 20
+ SHA1_STEP_16_79 A,B,C,D,E, TMP,FUN, I, K, MAGIC_F1
+ ROTATE_ARGS
+ I = (I+1)
+.endr
+
+## do rounds 40...59
+ vmovdqu K40_59(%rip), K
+.rep 20
+ SHA1_STEP_16_79 A,B,C,D,E, TMP,FUN, I, K, MAGIC_F2
+ ROTATE_ARGS
+ I = (I+1)
+.endr
+
+## do rounds 60...79
+ vmovdqu K60_79(%rip), K
+.rep 20
+ SHA1_STEP_16_79 A,B,C,D,E, TMP,FUN, I, K, MAGIC_F3
+ ROTATE_ARGS
+ I = (I+1)
+.endr
+
+ vpaddd AA,A,A
+ vpaddd BB,B,B
+ vpaddd CC,C,C
+ vpaddd DD,D,D
+ vpaddd EE,E,E
+
+ sub $1, arg2
+ jne lloop
+
+ # write out digests
+ vmovdqu A, 0*32(arg1)
+ vmovdqu B, 1*32(arg1)
+ vmovdqu C, 2*32(arg1)
+ vmovdqu D, 3*32(arg1)
+ vmovdqu E, 4*32(arg1)
+
+ # update input pointers
+ add IDX, inp0
+ add IDX, inp1
+ add IDX, inp2
+ add IDX, inp3
+ add IDX, inp4
+ add IDX, inp5
+ add IDX, inp6
+ add IDX, inp7
+ mov inp0, _data_ptr (arg1)
+ mov inp1, _data_ptr + 1*8(arg1)
+ mov inp2, _data_ptr + 2*8(arg1)
+ mov inp3, _data_ptr + 3*8(arg1)
+ mov inp4, _data_ptr + 4*8(arg1)
+ mov inp5, _data_ptr + 5*8(arg1)
+ mov inp6, _data_ptr + 6*8(arg1)
+ mov inp7, _data_ptr + 7*8(arg1)
+
+ ################
+ ## Postamble
+
+ mov RSP_SAVE, %rsp
+ pop RSP_SAVE
+
+ ret
+ENDPROC(sha1_x8_avx2)
+
+
+.data
+
+.align 32
+K00_19:
+.octa 0x5A8279995A8279995A8279995A827999
+.octa 0x5A8279995A8279995A8279995A827999
+K20_39:
+.octa 0x6ED9EBA16ED9EBA16ED9EBA16ED9EBA1
+.octa 0x6ED9EBA16ED9EBA16ED9EBA16ED9EBA1
+K40_59:
+.octa 0x8F1BBCDC8F1BBCDC8F1BBCDC8F1BBCDC
+.octa 0x8F1BBCDC8F1BBCDC8F1BBCDC8F1BBCDC
+K60_79:
+.octa 0xCA62C1D6CA62C1D6CA62C1D6CA62C1D6
+.octa 0xCA62C1D6CA62C1D6CA62C1D6CA62C1D6
+PSHUFFLE_BYTE_FLIP_MASK:
+.octa 0x0c0d0e0f08090a0b0405060700010203
+.octa 0x0c0d0e0f08090a0b0405060700010203
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 5/7] crypto: SHA1 multibuffer scheduler
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
` (4 preceding siblings ...)
2014-07-11 20:32 ` [PATCH v4 4/7] crypto: SHA1 multibuffer crypto computation (x8 AVX2) Tim Chen
@ 2014-07-11 20:33 ` Tim Chen
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
2014-07-11 20:33 ` [PATCH v4 7/7] crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle Tim Chen
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:33 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch introduces the multi-buffer scheduler 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>
---
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha-mb/Makefile | 11 +
arch/x86/crypto/sha-mb/sha1_mb.c | 943 +++++++++++++++++++++++++++++++++++++++
3 files changed, 956 insertions(+)
create mode 100644 arch/x86/crypto/sha-mb/Makefile
create mode 100644 arch/x86/crypto/sha-mb/sha1_mb.c
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index 61d6e28..1189267 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -25,6 +25,7 @@ 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
@@ -88,3 +89,4 @@ crc32-pclmul-y := crc32-pclmul_asm.o crc32-pclmul_glue.o
sha256-ssse3-y := sha256-ssse3-asm.o sha256-avx-asm.o sha256-avx2-asm.o sha256_ssse3_glue.o
sha512-ssse3-y := sha512-ssse3-asm.o sha512-avx-asm.o sha512-avx2-asm.o sha512_ssse3_glue.o
crct10dif-pclmul-y := crct10dif-pcl-asm_64.o crct10dif-pclmul_glue.o
+
diff --git a/arch/x86/crypto/sha-mb/Makefile b/arch/x86/crypto/sha-mb/Makefile
new file mode 100644
index 0000000..2f87563
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/Makefile
@@ -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
index 0000000..9c5feae
--- /dev/null
+++ b/arch/x86/crypto/sha-mb/sha1_mb.c
@@ -0,0 +1,943 @@
+/*
+ * 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;
+
+ /* initialize tag */
+ rctx->tag.arrival = jiffies; /* tag the arrival time */
+ rctx->tag.seq_num = cstate->next_seq_num++;
+ next_flush = rctx->tag.arrival + usecs_to_jiffies(FLUSH_INTERVAL);
+ rctx->tag.expire = next_flush;
+
+ spin_lock(&cstate->work_lock);
+ list_add_tail(&rctx->waiter, &cstate->work_list);
+ spin_unlock(&cstate->work_lock);
+
+ if (!cstate->flusher_engaged) {
+ queue_delayed_work_on(smp_processor_id(), kcrypto_wq,
+ &cstate->flush, usecs_to_jiffies(FLUSH_INTERVAL));
+ cstate->flusher_engaged = true;
+ cstate->next_flush = next_flush;
+ }
+}
+
+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;
+ 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;
+ queue_delayed_work_on(smp_processor_id(), kcrypto_wq,
+ &cstate->flush, get_delay(next_flush));
+ } else {
+ next_flush = 0;
+ cstate->flusher_engaged = false;
+ }
+ 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");
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
` (5 preceding siblings ...)
2014-07-11 20:33 ` [PATCH v4 5/7] crypto: SHA1 multibuffer scheduler Tim Chen
@ 2014-07-11 20:33 ` Tim Chen
2014-07-12 9:25 ` Kirill Tkhai
` (2 more replies)
2014-07-11 20:33 ` [PATCH v4 7/7] crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle Tim Chen
7 siblings, 3 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:33 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This function will help a thread decide if it wants to to do work
that can be delayed, to accumulate more tasks for more efficient
batch processing later.
However, if no other tasks are running on the cpu, it can take
advantgae of the available cpu cycles to complete the tasks
for immediate processing to minimize delay, otherwise it will yield.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
include/linux/sched.h | 1 +
kernel/sched/core.c | 8 ++++++++
2 files changed, 9 insertions(+)
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 7cb07fd..0884250 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -168,6 +168,7 @@ extern int nr_threads;
DECLARE_PER_CPU(unsigned long, process_counts);
extern int nr_processes(void);
extern unsigned long nr_running(void);
+extern unsigned long nr_running_cpu(int cpu);
extern unsigned long nr_iowait(void);
extern unsigned long nr_iowait_cpu(int cpu);
extern unsigned long this_cpu_load(void);
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 9cae286..d5bb8e6 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2283,6 +2283,14 @@ unsigned long nr_running(void)
return sum;
}
+unsigned long nr_running_cpu(int cpu)
+{
+ if (cpumask_test_cpu(cpu, cpu_online_mask))
+ return cpu_rq(cpu)->nr_running;
+ else
+ return 0;
+}
+
unsigned long long nr_context_switches(void)
{
int i;
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* [PATCH v4 7/7] crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
` (6 preceding siblings ...)
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
@ 2014-07-11 20:33 ` Tim Chen
7 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-11 20:33 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Tim Chen, Jussi Kivilinna, linux-crypto,
linux-kernel
This patch adds a notifier to the SHA1 multi-buffer algorithm
when CPU is giong idle, so it can take advantage of the available
CPU power to flush out any partially completed jobs. This
will eliminate possible extended latency in the multi-buffer
algorithm.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha-mb/sha1_mb.c | 61 ++++++++++++++++++++++++++++++++++++++++
include/crypto/mcryptd.h | 1 +
2 files changed, 62 insertions(+)
diff --git a/arch/x86/crypto/sha-mb/sha1_mb.c b/arch/x86/crypto/sha-mb/sha1_mb.c
index 9c5feae..fd9b219 100644
--- a/arch/x86/crypto/sha-mb/sha1_mb.c
+++ b/arch/x86/crypto/sha-mb/sha1_mb.c
@@ -69,6 +69,7 @@
#include <asm/xcr.h>
#include <asm/xsave.h>
#include <linux/hardirq.h>
+#include <linux/sched.h>
#include <asm/fpu-internal.h>
#include "sha_mb_ctx.h"
@@ -820,6 +821,60 @@ static struct ahash_alg sha1_mb_async_alg = {
},
};
+void sha1_mb_force_flush(struct mcryptd_alg_cstate *cstate)
+{
+ struct mcryptd_hash_request_ctx *rctx;
+ struct sha1_hash_ctx *sha_ctx;
+
+ /* force flush uncompleted jobs in all data lanes before cpu becomes idle */
+ while (!list_empty(&cstate->work_list)) {
+ /* turn off flusher as we are flushing here */
+ if (cstate->flusher_engaged)
+ cstate->flusher_engaged = false;
+
+ 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);
+ }
+
+ return;
+}
+
+void sha1_mb_earlyflush(struct work_struct *__work)
+{
+ struct mcryptd_alg_cstate *alg_cpu_state;
+
+ /* do not do early flush if other tasks are running */
+ if (nr_running_cpu(smp_processor_id()) > 1)
+ return;
+
+ alg_cpu_state = container_of(__work, struct mcryptd_alg_cstate, early_flush);
+ sha1_mb_force_flush(alg_cpu_state);
+}
+
+static int sha1_mb_idle_notifier(struct notifier_block *nb, unsigned long val,
+ void *data)
+{
+ struct mcryptd_alg_cstate *cstate =
+ this_cpu_ptr(sha1_mb_alg_state.alg_cstate);
+
+ if (val == IDLE_START && cstate->flusher_engaged)
+ queue_work_on(smp_processor_id(), kcrypto_wq, &cstate->early_flush);
+
+ return 0;
+}
+
+static struct notifier_block sha1_mb_idle_nb = {
+ .notifier_call = sha1_mb_idle_notifier,
+};
+
unsigned long sha1_mb_flusher(struct mcryptd_alg_cstate *cstate)
{
struct mcryptd_hash_request_ctx *rctx;
@@ -830,6 +885,9 @@ unsigned long sha1_mb_flusher(struct mcryptd_alg_cstate *cstate)
cur_time = jiffies;
+ if (!cstate->flusher_engaged)
+ return 0;
+
while (!list_empty(&cstate->work_list)) {
rctx = list_entry(cstate->work_list.next,
struct mcryptd_hash_request_ctx, waiter);
@@ -889,6 +947,7 @@ static int __init sha1_mb_mod_init(void)
cpu_state->next_seq_num = 0;
cpu_state->flusher_engaged = false;
INIT_DELAYED_WORK(&cpu_state->flush, mcryptd_flusher);
+ INIT_WORK(&cpu_state->early_flush, sha1_mb_earlyflush);
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);
@@ -907,6 +966,7 @@ static int __init sha1_mb_mod_init(void)
if (err)
goto err1;
+ idle_notifier_register(&sha1_mb_idle_nb);
return 0;
err1:
@@ -925,6 +985,7 @@ static void __exit sha1_mb_mod_fini(void)
int cpu;
struct mcryptd_alg_cstate *cpu_state;
+ idle_notifier_unregister(&sha1_mb_idle_nb);
crypto_unregister_ahash(&sha1_mb_async_alg);
crypto_unregister_shash(&sha1_mb_shash_alg);
for_each_possible_cpu(cpu) {
diff --git a/include/crypto/mcryptd.h b/include/crypto/mcryptd.h
index b2b9055..2ef1824 100644
--- a/include/crypto/mcryptd.h
+++ b/include/crypto/mcryptd.h
@@ -82,6 +82,7 @@ struct mcryptd_alg_cstate {
unsigned next_seq_num;
bool flusher_engaged;
struct delayed_work flush;
+ struct work_struct early_flush;
int cpu;
struct mcryptd_alg_state *alg_state;
void *mgr;
--
1.7.11.7
^ permalink raw reply related [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
@ 2014-07-12 9:25 ` Kirill Tkhai
2014-07-14 17:51 ` Tim Chen
2014-07-12 14:21 ` Tadeusz Struk
2014-07-14 10:16 ` Peter Zijlstra
2 siblings, 1 reply; 35+ messages in thread
From: Kirill Tkhai @ 2014-07-12 9:25 UTC (permalink / raw)
To: Tim Chen, Herbert Xu, H. Peter Anvin, David S.Miller,
Peter Zijlstra, Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On 12.07.2014 00:33, Tim Chen wrote:
> This function will help a thread decide if it wants to to do work
> that can be delayed, to accumulate more tasks for more efficient
> batch processing later.
>
> However, if no other tasks are running on the cpu, it can take
> advantgae of the available cpu cycles to complete the tasks
> for immediate processing to minimize delay, otherwise it will yield.
>
> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
> ---
> include/linux/sched.h | 1 +
> kernel/sched/core.c | 8 ++++++++
> 2 files changed, 9 insertions(+)
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 7cb07fd..0884250 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -168,6 +168,7 @@ extern int nr_threads;
> DECLARE_PER_CPU(unsigned long, process_counts);
> extern int nr_processes(void);
> extern unsigned long nr_running(void);
> +extern unsigned long nr_running_cpu(int cpu);
> extern unsigned long nr_iowait(void);
> extern unsigned long nr_iowait_cpu(int cpu);
> extern unsigned long this_cpu_load(void);
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 9cae286..d5bb8e6 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -2283,6 +2283,14 @@ unsigned long nr_running(void)
> return sum;
> }
>
> +unsigned long nr_running_cpu(int cpu)
> +{
> + if (cpumask_test_cpu(cpu, cpu_online_mask))
> + return cpu_rq(cpu)->nr_running;
> + else
> + return 0;
> +}
> +
Offline cpu should have nr_running equal to 0. We park last
enqueued thread (migration_thread) at the end of take_cpu_down().
So, it's enough to return cpu_rq(cpu)->nr_running.
> unsigned long long nr_context_switches(void)
> {
> int i;
>
Thanks,
Kirill
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
2014-07-12 9:25 ` Kirill Tkhai
@ 2014-07-12 14:21 ` Tadeusz Struk
2014-07-14 23:51 ` Tim Chen
2014-07-14 10:16 ` Peter Zijlstra
2 siblings, 1 reply; 35+ messages in thread
From: Tadeusz Struk @ 2014-07-12 14:21 UTC (permalink / raw)
To: Tim Chen, Herbert Xu, H. Peter Anvin, David S.Miller,
Peter Zijlstra, Ingo Molnar
Cc: Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On 07/11/2014 01:33 PM, Tim Chen wrote:
> +unsigned long nr_running_cpu(int cpu)
> +{
> + if (cpumask_test_cpu(cpu, cpu_online_mask))
> + return cpu_rq(cpu)->nr_running;
> + else
> + return 0;
> +}
> +
EXPORT_SYMBOL?
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
2014-07-12 9:25 ` Kirill Tkhai
2014-07-12 14:21 ` Tadeusz Struk
@ 2014-07-14 10:16 ` Peter Zijlstra
2014-07-14 16:10 ` Tim Chen
2 siblings, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-14 10:16 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 901 bytes --]
On Fri, Jul 11, 2014 at 01:33:04PM -0700, Tim Chen wrote:
> This function will help a thread decide if it wants to to do work
> that can be delayed, to accumulate more tasks for more efficient
> batch processing later.
>
> However, if no other tasks are running on the cpu, it can take
> advantgae of the available cpu cycles to complete the tasks
> for immediate processing to minimize delay, otherwise it will yield.
Ugh.. and ignore topology and everything else.
Yet another scheduler on top of the scheduler.
We have the padata muck, also only ever used by crypto.
We have the workqueue nonsense, used all over the place
And we have btrfs doing their own padata like muck.
And I'm sure there's at least one more out there, just because.
Why do we want yet another thing?
I'm inclined to go NAK and get people to reduce the amount of async
queueing and processing crap.
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 10:16 ` Peter Zijlstra
@ 2014-07-14 16:10 ` Tim Chen
2014-07-14 16:14 ` Peter Zijlstra
0 siblings, 1 reply; 35+ messages in thread
From: Tim Chen @ 2014-07-14 16:10 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Mon, 2014-07-14 at 12:16 +0200, Peter Zijlstra wrote:
> On Fri, Jul 11, 2014 at 01:33:04PM -0700, Tim Chen wrote:
> > This function will help a thread decide if it wants to to do work
> > that can be delayed, to accumulate more tasks for more efficient
> > batch processing later.
> >
> > However, if no other tasks are running on the cpu, it can take
> > advantgae of the available cpu cycles to complete the tasks
> > for immediate processing to minimize delay, otherwise it will yield.
>
> Ugh.. and ignore topology and everything else.
>
> Yet another scheduler on top of the scheduler.
>
> We have the padata muck, also only ever used by crypto.
> We have the workqueue nonsense, used all over the place
> And we have btrfs doing their own padata like muck.
> And I'm sure there's at least one more out there, just because.
>
> Why do we want yet another thing?
>
> I'm inclined to go NAK and get people to reduce the amount of async
> queueing and processing crap.
The mult-buffer class of crypto algorithms is by nature
asynchronous. The algorithm gathers several crypto jobs, and
put the buffer from each job in a data lane of the SIMD register.
This allows for parallel processing and increases throughput.
The gathering of the crypto jobs is an async process and
queuing is necessary for this class of algorithm.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 16:10 ` Tim Chen
@ 2014-07-14 16:14 ` Peter Zijlstra
2014-07-14 17:05 ` Tim Chen
0 siblings, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-14 16:14 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 1561 bytes --]
On Mon, Jul 14, 2014 at 09:10:14AM -0700, Tim Chen wrote:
> On Mon, 2014-07-14 at 12:16 +0200, Peter Zijlstra wrote:
> > On Fri, Jul 11, 2014 at 01:33:04PM -0700, Tim Chen wrote:
> > > This function will help a thread decide if it wants to to do work
> > > that can be delayed, to accumulate more tasks for more efficient
> > > batch processing later.
> > >
> > > However, if no other tasks are running on the cpu, it can take
> > > advantgae of the available cpu cycles to complete the tasks
> > > for immediate processing to minimize delay, otherwise it will yield.
> >
> > Ugh.. and ignore topology and everything else.
> >
> > Yet another scheduler on top of the scheduler.
> >
> > We have the padata muck, also only ever used by crypto.
> > We have the workqueue nonsense, used all over the place
> > And we have btrfs doing their own padata like muck.
> > And I'm sure there's at least one more out there, just because.
> >
> > Why do we want yet another thing?
> >
> > I'm inclined to go NAK and get people to reduce the amount of async
> > queueing and processing crap.
>
> The mult-buffer class of crypto algorithms is by nature
> asynchronous. The algorithm gathers several crypto jobs, and
> put the buffer from each job in a data lane of the SIMD register.
> This allows for parallel processing and increases throughput.
> The gathering of the crypto jobs is an async process and
> queuing is necessary for this class of algorithm.
How is that related to me saying we've got too much of this crap
already?
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 16:14 ` Peter Zijlstra
@ 2014-07-14 17:05 ` Tim Chen
2014-07-14 18:17 ` Peter Zijlstra
0 siblings, 1 reply; 35+ messages in thread
From: Tim Chen @ 2014-07-14 17:05 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Mon, 2014-07-14 at 18:14 +0200, Peter Zijlstra wrote:
> On Mon, Jul 14, 2014 at 09:10:14AM -0700, Tim Chen wrote:
> > On Mon, 2014-07-14 at 12:16 +0200, Peter Zijlstra wrote:
> > > On Fri, Jul 11, 2014 at 01:33:04PM -0700, Tim Chen wrote:
> > > > This function will help a thread decide if it wants to to do work
> > > > that can be delayed, to accumulate more tasks for more efficient
> > > > batch processing later.
> > > >
> > > > However, if no other tasks are running on the cpu, it can take
> > > > advantgae of the available cpu cycles to complete the tasks
> > > > for immediate processing to minimize delay, otherwise it will yield.
> > >
> > > Ugh.. and ignore topology and everything else.
> > >
> > > Yet another scheduler on top of the scheduler.
> > >
> > > We have the padata muck, also only ever used by crypto.
> > > We have the workqueue nonsense, used all over the place
> > > And we have btrfs doing their own padata like muck.
> > > And I'm sure there's at least one more out there, just because.
> > >
> > > Why do we want yet another thing?
> > >
> > > I'm inclined to go NAK and get people to reduce the amount of async
> > > queueing and processing crap.
> >
> > The mult-buffer class of crypto algorithms is by nature
> > asynchronous. The algorithm gathers several crypto jobs, and
> > put the buffer from each job in a data lane of the SIMD register.
> > This allows for parallel processing and increases throughput.
> > The gathering of the crypto jobs is an async process and
> > queuing is necessary for this class of algorithm.
>
> How is that related to me saying we've got too much of this crap
> already?
I was trying to explain why the algorithm is implemented this way
because of its batching nature.
There is a whole class of async algorithm that can provide
substantial speedup by doing batch processing and uses workqueue.
The multi-buffer sha1 version has 2.2x speedup over existing
AVX2 version, and can have even more speedup when AVX3
comes round. Workqueue is a natural way to implement
this. I don't think a throughput speedup of 2.2x is "crap".
We are not inventing anything new, but ask for a
very simple helper function to know if there's something else
running on our cpu to help us make a better decision
of whether we should flush the batched jobs immediately.
And also asynchronous crypto interface is already used substantially
in crypto and has a well established infrastructure.
Thanks.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-12 9:25 ` Kirill Tkhai
@ 2014-07-14 17:51 ` Tim Chen
0 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-14 17:51 UTC (permalink / raw)
To: tkhai
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Sat, 2014-07-12 at 13:25 +0400, Kirill Tkhai wrote:
> >
> > +unsigned long nr_running_cpu(int cpu)
> > +{
> > + if (cpumask_test_cpu(cpu, cpu_online_mask))
> > + return cpu_rq(cpu)->nr_running;
> > + else
> > + return 0;
> > +}
> > +
>
> Offline cpu should have nr_running equal to 0. We park last
> enqueued thread (migration_thread) at the end of take_cpu_down().
>
> So, it's enough to return cpu_rq(cpu)->nr_running.
Thanks. This seems reasonable.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 17:05 ` Tim Chen
@ 2014-07-14 18:17 ` Peter Zijlstra
2014-07-14 19:08 ` Tim Chen
0 siblings, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-14 18:17 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 1215 bytes --]
On Mon, Jul 14, 2014 at 10:05:34AM -0700, Tim Chen wrote:
> I was trying to explain why the algorithm is implemented this way
> because of its batching nature.
>
> There is a whole class of async algorithm that can provide
> substantial speedup by doing batch processing and uses workqueue.
> The multi-buffer sha1 version has 2.2x speedup over existing
> AVX2 version, and can have even more speedup when AVX3
> comes round. Workqueue is a natural way to implement
> this. I don't think a throughput speedup of 2.2x is "crap".
>
> We are not inventing anything new, but ask for a
> very simple helper function to know if there's something else
> running on our cpu to help us make a better decision
> of whether we should flush the batched jobs immediately.
>
> And also asynchronous crypto interface is already used substantially
> in crypto and has a well established infrastructure.
The crap I was talking about is that there's a metric ton of 'async'
interfaces all different.
Your multi-buffer thing isn't generic either, it seems lmiited to sha1.
It does not reuse padata, it does not extend workqueues, it does not
remove the btrfs nonsense, it adds yet anotehr thing.
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 18:17 ` Peter Zijlstra
@ 2014-07-14 19:08 ` Tim Chen
2014-07-14 19:15 ` Peter Zijlstra
0 siblings, 1 reply; 35+ messages in thread
From: Tim Chen @ 2014-07-14 19:08 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Mon, 2014-07-14 at 20:17 +0200, Peter Zijlstra wrote:
> On Mon, Jul 14, 2014 at 10:05:34AM -0700, Tim Chen wrote:
> > I was trying to explain why the algorithm is implemented this way
> > because of its batching nature.
> >
> > There is a whole class of async algorithm that can provide
> > substantial speedup by doing batch processing and uses workqueue.
> > The multi-buffer sha1 version has 2.2x speedup over existing
> > AVX2 version, and can have even more speedup when AVX3
> > comes round. Workqueue is a natural way to implement
> > this. I don't think a throughput speedup of 2.2x is "crap".
> >
> > We are not inventing anything new, but ask for a
> > very simple helper function to know if there's something else
> > running on our cpu to help us make a better decision
> > of whether we should flush the batched jobs immediately.
> >
> > And also asynchronous crypto interface is already used substantially
> > in crypto and has a well established infrastructure.
>
> The crap I was talking about is that there's a metric ton of 'async'
> interfaces all different.
Async interfaces when used appropriately, actually speed things up
substantially for crypto. We actually have a case with
ecyrptfs not using the async crypto interface, causing cpu to stall
and slowing things down substantially with AES-NI. And async interface
with workqueue speed things up (30% to 35% on encryption with SSD).
http://marc.info/?l=ecryptfs-users&m=136520541407248
http://www.spinics.net/lists/ecryptfs/msg00228.html
>
> Your multi-buffer thing isn't generic either, it seems lmiited to sha1.
We actually have many other multi-buffer crypto algorithms already
published for encryption and other IPSec usages. So
multi-buffer algorithm is not just limited to SHA1.
We hope to port those to the kernel crypto library eventually.
http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-multi-buffer-ipsec-implementations-ia-processors-paper.pdf
> It does not reuse padata,
padata tries to speed things up by parallelizing jobs to *multiple*
cpus. Whereas multi-buffer tries to speed things up by speeding things
up by using multiple data lanes in SIMD register in a *single* cpu.
These two usages are complementary but not the same.
> it does not extend workqueues,
Why do I need to extend workqueues if the existing ones already
meet my needs?
> it does not
> remove the btrfs nonsense,
Not much I can do about btrfs as I don't understand the issues there.
> it adds yet anotehr thing.
Thanks.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 19:08 ` Tim Chen
@ 2014-07-14 19:15 ` Peter Zijlstra
2014-07-14 19:50 ` Tim Chen
2014-07-15 13:36 ` Peter Zijlstra
0 siblings, 2 replies; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-14 19:15 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 1302 bytes --]
On Mon, Jul 14, 2014 at 12:08:28PM -0700, Tim Chen wrote:
> On Mon, 2014-07-14 at 20:17 +0200, Peter Zijlstra wrote:
> > Your multi-buffer thing isn't generic either, it seems lmiited to sha1.
>
> We actually have many other multi-buffer crypto algorithms already
> published for encryption and other IPSec usages. So
> multi-buffer algorithm is not just limited to SHA1.
> We hope to port those to the kernel crypto library eventually.
> http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-multi-buffer-ipsec-implementations-ia-processors-paper.pdf
That's all nice and such; but the code as I've seen in these patches is
very much sha1 specific. The mb part isn't separated out.
> > It does not reuse padata,
> padata tries to speed things up by parallelizing jobs to *multiple*
> cpus. Whereas multi-buffer tries to speed things up by speeding things
> up by using multiple data lanes in SIMD register in a *single* cpu.
> These two usages are complementary but not the same.
And if its single cpu, wth do you need that nr_running thing for another
cpu for?
Also, this difference wasn't clear to me.
I still loathe all the async work, because it makes a mockery of
accounting etc.. but that's a story for another day I suppose :-(
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 19:15 ` Peter Zijlstra
@ 2014-07-14 19:50 ` Tim Chen
2014-07-15 9:50 ` Peter Zijlstra
2014-07-15 13:36 ` Peter Zijlstra
1 sibling, 1 reply; 35+ messages in thread
From: Tim Chen @ 2014-07-14 19:50 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Mon, 2014-07-14 at 21:15 +0200, Peter Zijlstra wrote:
> On Mon, Jul 14, 2014 at 12:08:28PM -0700, Tim Chen wrote:
> > On Mon, 2014-07-14 at 20:17 +0200, Peter Zijlstra wrote:
>
> > > Your multi-buffer thing isn't generic either, it seems lmiited to sha1.
> >
> > We actually have many other multi-buffer crypto algorithms already
> > published for encryption and other IPSec usages. So
> > multi-buffer algorithm is not just limited to SHA1.
> > We hope to port those to the kernel crypto library eventually.
> > http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-multi-buffer-ipsec-implementations-ia-processors-paper.pdf
>
> That's all nice and such; but the code as I've seen in these patches is
> very much sha1 specific. The mb part isn't separated out.
There is a generic multi-buffer infrastructure portion that manages
pulling and queuing jobs on the crypto workqueue, and it is separated out
in patch 1 of the patchset. The other portions are algorithm specific that defines
algorithm specific data structure and does the crypto computation
for a particular algorithm, mostly in
assemblies and C glue code. The infrastructure code is
meant to be reused for other similar
multi-buffer algorithms.
>
> > > It does not reuse padata,
> > padata tries to speed things up by parallelizing jobs to *multiple*
> > cpus. Whereas multi-buffer tries to speed things up by speeding things
> > up by using multiple data lanes in SIMD register in a *single* cpu.
> > These two usages are complementary but not the same.
>
> And if its single cpu, wth do you need that nr_running thing for another
> cpu for?
We use nr_running_cpu to check whether there are other tasks running on
the *current* cpu, (not for another cpu), to decide if we should flush
and compute crypto jobs accumulated. If there's nobody else running,
we can take advantage of available cpu cycles on the cpu we are running
on to do computation on the existing jobs in a SIMD mannner.
Waiting a bit longer may accumulate more jobs to process in parallel
in a single SIMD instruction, but will have more delay.
>
> Also, this difference wasn't clear to me.
>
> I still loathe all the async work, because it makes a mockery of
> accounting etc.. but that's a story for another day I suppose :-(
>
>
Thanks.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-12 14:21 ` Tadeusz Struk
@ 2014-07-14 23:51 ` Tim Chen
0 siblings, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-14 23:51 UTC (permalink / raw)
To: Tadeusz Struk
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Peter Zijlstra,
Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Sat, 2014-07-12 at 07:21 -0700, Tadeusz Struk wrote:
> On 07/11/2014 01:33 PM, Tim Chen wrote:
> > +unsigned long nr_running_cpu(int cpu)
> > +{
> > + if (cpumask_test_cpu(cpu, cpu_online_mask))
> > + return cpu_rq(cpu)->nr_running;
> > + else
> > + return 0;
> > +}
> > +
> EXPORT_SYMBOL?
Yes, thanks.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 19:50 ` Tim Chen
@ 2014-07-15 9:50 ` Peter Zijlstra
2014-07-15 12:07 ` Peter Zijlstra
2014-07-15 18:40 ` Tim Chen
0 siblings, 2 replies; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 9:50 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 2106 bytes --]
On Mon, Jul 14, 2014 at 12:50:50PM -0700, Tim Chen wrote:
> There is a generic multi-buffer infrastructure portion that manages
> pulling and queuing jobs on the crypto workqueue, and it is separated out
> in patch 1 of the patchset.
There's one very weird multi-line comment in that patch.
> The other portions are algorithm specific that defines
> algorithm specific data structure and does the crypto computation
> for a particular algorithm, mostly in
> assemblies and C glue code. The infrastructure code is
> meant to be reused for other similar
> multi-buffer algorithms.
The flushing part that uses the sched thing is sha1 specific, even
though it strikes me as not being so. Flushing buffers on idle seems
like a 'generic' thing.
> We use nr_running_cpu to check whether there are other tasks running on
> the *current* cpu, (not for another cpu),
And yet, the function allows you do to exactly that..
> to decide if we should flush
> and compute crypto jobs accumulated. If there's nobody else running,
> we can take advantage of available cpu cycles on the cpu we are running
> on to do computation on the existing jobs in a SIMD mannner.
> Waiting a bit longer may accumulate more jobs to process in parallel
> in a single SIMD instruction, but will have more delay.
So you already have an idle notifier (which is x86 only, we should fix
that I suppose), and you then double check there really isn't anything
else running.
How much, if anything, does that second check buy you? There's just not
a single word on that.
Also, there is not a word on the latency vs throughput tradeoff you
make. I can imagine that for very short idle durations you loose, not
win with this thing.
So for now I still see no reason for doing this.
Also, I wonder about SMT, the point of this is to make best use of the
SIMD pipelines, does it still make sense to use siblings at the same
time even though you're running hand crafted ASM to stuff the pipelines
to the brim? Should this thing be SMT aware and not gather queues for
both siblings?
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 9:50 ` Peter Zijlstra
@ 2014-07-15 12:07 ` Peter Zijlstra
2014-07-15 12:59 ` Thomas Gleixner
2014-07-15 18:40 ` Tim Chen
2014-07-15 18:40 ` Tim Chen
1 sibling, 2 replies; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 12:07 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 504 bytes --]
On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> So you already have an idle notifier (which is x86 only, we should fix
> that I suppose), and you then double check there really isn't anything
> else running.
Note that we've already done a large part of the expense of going idle
by the time we call that idle notifier -- in specific, we've
reprogrammed the clock to stop the tick.
Its really wasteful to then generate work again, which means we have to
again reprogram the clock etc.
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 12:07 ` Peter Zijlstra
@ 2014-07-15 12:59 ` Thomas Gleixner
2014-07-15 14:45 ` Mike Galbraith
2014-07-15 18:41 ` Tim Chen
2014-07-15 18:40 ` Tim Chen
1 sibling, 2 replies; 35+ messages in thread
From: Thomas Gleixner @ 2014-07-15 12:59 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Tim Chen, Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, 15 Jul 2014, Peter Zijlstra wrote:
> On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> > So you already have an idle notifier (which is x86 only, we should fix
> > that I suppose), and you then double check there really isn't anything
> > else running.
>
> Note that we've already done a large part of the expense of going idle
> by the time we call that idle notifier -- in specific, we've
> reprogrammed the clock to stop the tick.
>
> Its really wasteful to then generate work again, which means we have to
> again reprogram the clock etc.
Doing anything which is not related to idle itself in the idle
notifier is just plain wrong.
If that stuff wants to utilize idle slots, we really need to come up
with a generic and general solution. Otherwise we'll grow those warts
all over the architecture space, with slightly different ways of
wreckaging the world an some more.
This whole attidute of people thinking that they need their own
specialized scheduling around the real scheduler is a PITA. All this
stuff is just damanging any sensible approach of power saving, load
balancing, etc.
What we really want is infrastructure, which allows the scheduler to
actively query the async work situation and based on the results
actively decide when to process it and where.
Thanks,
tglx
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-14 19:15 ` Peter Zijlstra
2014-07-14 19:50 ` Tim Chen
@ 2014-07-15 13:36 ` Peter Zijlstra
2014-07-15 15:21 ` Tejun Heo
1 sibling, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 13:36 UTC (permalink / raw)
To: Tim Chen
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel, tj
[-- Attachment #1: Type: text/plain, Size: 645 bytes --]
On Mon, Jul 14, 2014 at 09:15:04PM +0200, Peter Zijlstra wrote:
> I still loathe all the async work, because it makes a mockery of
> accounting etc.. but that's a story for another day I suppose :-(
So, just to expand on this, we're already getting 'bug' reports because
worker threads are not cgroup aware. If work gets generated inside some
cgroup, the worker doesn't care and runs the worker thread wherever
(typically the root cgroup).
This means that the 'work' escapes the cgroup confines and creates
resource inversion etc. The same is of course true for nice and RT
priorities.
TJ, are you aware of this and/or given it any throught?
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 12:59 ` Thomas Gleixner
@ 2014-07-15 14:45 ` Mike Galbraith
2014-07-15 14:53 ` Peter Zijlstra
2014-07-15 18:41 ` Tim Chen
1 sibling, 1 reply; 35+ messages in thread
From: Mike Galbraith @ 2014-07-15 14:45 UTC (permalink / raw)
To: Thomas Gleixner
Cc: Peter Zijlstra, Tim Chen, Herbert Xu, H. Peter Anvin,
David S.Miller, Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal,
James Guilford, Wajdi Feghali, Jussi Kivilinna, linux-crypto,
linux-kernel
On Tue, 2014-07-15 at 14:59 +0200, Thomas Gleixner wrote:
> On Tue, 15 Jul 2014, Peter Zijlstra wrote:
>
> > On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> > > So you already have an idle notifier (which is x86 only, we should fix
> > > that I suppose), and you then double check there really isn't anything
> > > else running.
> >
> > Note that we've already done a large part of the expense of going idle
> > by the time we call that idle notifier -- in specific, we've
> > reprogrammed the clock to stop the tick.
> >
> > Its really wasteful to then generate work again, which means we have to
> > again reprogram the clock etc.
>
> Doing anything which is not related to idle itself in the idle
> notifier is just plain wrong.
>
> If that stuff wants to utilize idle slots, we really need to come up
> with a generic and general solution. Otherwise we'll grow those warts
> all over the architecture space, with slightly different ways of
> wreckaging the world an some more.
>
> This whole attidute of people thinking that they need their own
> specialized scheduling around the real scheduler is a PITA. All this
> stuff is just damanging any sensible approach of power saving, load
> balancing, etc.
Not to mention that we're already too rotund...
pipe-test scheduling cross core, ie ~0 work, ~pure full fastpath. All
kernels with same (obese) distro config, with drivers reduced to what my
boxen need. Squint a little, there is some jitter. These kernels are
all adjusted to eliminate various regressions that would otherwise skew
results up to and including _very_ badly. See "virgin", the numbers are
much more useful without that particular skew methinks :)
3.0.101-default 3.753363 usecs/loop -- avg 3.770737 530.4 KHz 1.000
3.1.10-default 3.723843 usecs/loop -- avg 3.716058 538.2 KHz 1.014
3.2.51-default 3.728060 usecs/loop -- avg 3.710372 539.0 KHz 1.016
3.3.8-default 3.906174 usecs/loop -- avg 3.900399 512.8 KHz .966
3.4.97-default 3.864158 usecs/loop -- avg 3.865281 517.4 KHz .975
3.5.7-default 3.967481 usecs/loop -- avg 3.962757 504.7 KHz .951
3.6.11-default 3.851186 usecs/loop -- avg 3.845321 520.1 KHz .980
3.7.10-default 3.777869 usecs/loop -- avg 3.776913 529.5 KHz .998
3.8.13-default 4.049927 usecs/loop -- avg 4.041905 494.8 KHz .932
3.9.11-default 3.973046 usecs/loop -- avg 3.974208 503.2 KHz .948
3.10.27-default 4.189598 usecs/loop -- avg 4.189298 477.4 KHz .900
3.11.10-default 4.293870 usecs/loop -- avg 4.297979 465.3 KHz .877
3.12.24-default 4.321570 usecs/loop -- avg 4.321961 462.8 KHz .872
3.13.11-default 4.137845 usecs/loop -- avg 4.134863 483.7 KHz .911
3.14.10-default 4.145348 usecs/loop -- avg 4.139987 483.1 KHz .910 1.000
3.15.4-default 4.355594 usecs/loop -- avg 4.351961 459.6 KHz .866 .951 1.000
3.16.0-default 4.537279 usecs/loop -- avg 4.543532 440.2 KHz .829 .911 .957
3.16.0-virgin 6.377331 usecs/loop -- avg 6.352794 314.8 KHz 0.sob
my local config, group sched, namespaces etc disabled
3.0.101-smp 3.692377 usecs/loop -- avg 3.690774 541.9 KHz 1.000
3.1.10-smp 3.573832 usecs/loop -- avg 3.563269 561.3 KHz 1.035
3.2.51-smp 3.632690 usecs/loop -- avg 3.628220 551.2 KHz 1.017
3.3.8-smp 3.801838 usecs/loop -- avg 3.803441 525.8 KHz .970
3.4.97-smp 3.836087 usecs/loop -- avg 3.843501 520.4 KHz .960
3.5.7-smp 3.646927 usecs/loop -- avg 3.646288 548.5 KHz 1.012
3.6.11-smp 3.674402 usecs/loop -- avg 3.680929 543.3 KHz 1.002
3.7.10-smp 3.644274 usecs/loop -- avg 3.644566 548.8 KHz 1.012
3.8.13-smp 3.678164 usecs/loop -- avg 3.675524 544.1 KHz 1.004
3.9.11-smp 3.834943 usecs/loop -- avg 3.845852 520.0 KHz .959
3.10.27-smp 3.651881 usecs/loop -- avg 3.634515 550.3 KHz 1.015
3.11.10-smp 3.716159 usecs/loop -- avg 3.720603 537.5 KHz .991
3.12.24-smp 3.862634 usecs/loop -- avg 3.872252 516.5 KHz .953
3.13.11-smp 3.803254 usecs/loop -- avg 3.802553 526.0 KHz .970
3.14.10-smp 4.010009 usecs/loop -- avg 4.009019 498.9 KHz .920
3.15.4-smp 3.882398 usecs/loop -- avg 3.884095 514.9 KHz .950
3.16.0-master 4.061003 usecs/loop -- avg 4.058244 492.8 KHz .909
echo 0 > sched_wakeup_granularity_ns, taskset -c 3 pipe-test 1 (shortest path)
3.0.101-default 3.352267 usecs/loop -- avg 3.352434 596.6 KHz 1.000
3.16.0-default 3.596559 usecs/loop -- avg 3.594023 556.5 KHz .932
3.0.101-smp 3.089251 usecs/loop -- avg 3.089556 647.3 KHz 1.000
3.16.0-master 3.254721 usecs/loop -- avg 3.251534 615.1 KHz .950
sched+idle is becoming more of a not-so-fastpath. Pure sched is not as
bad, but still, we're getting fat.
netperf TCP_RR trans/sec (unbound)
3.0.101-default 91360.56 1.000
3.16.0-default 72523.30 .793
3.0.101-smp 92166.23 1.000
3.16.0-master 81235.30 .881
echo 0 > sched_wakeup_granularity_ns, bound to cpu3
3.0.101-smp 94289.95 1.000
3.16.0-master 81219.02 .861
Leanest meanest kernel ever to run on this box (2.6.22 + cfs-2.6.25 etc)
did that bound TCP_RR at ~114k IIRC. My userspace became too new to
boot that kernel without a squabble, but I think I recall correctly.
-Mike
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 14:45 ` Mike Galbraith
@ 2014-07-15 14:53 ` Peter Zijlstra
2014-07-15 18:06 ` Mike Galbraith
0 siblings, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 14:53 UTC (permalink / raw)
To: Mike Galbraith
Cc: Thomas Gleixner, Tim Chen, Herbert Xu, H. Peter Anvin,
David S.Miller, Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal,
James Guilford, Wajdi Feghali, Jussi Kivilinna, linux-crypto,
linux-kernel
[-- Attachment #1: Type: text/plain, Size: 797 bytes --]
On Tue, Jul 15, 2014 at 04:45:25PM +0200, Mike Galbraith wrote:
>
> 3.0.101-default 3.753363 usecs/loop -- avg 3.770737 530.4 KHz 1.000
> 3.14.10-default 4.145348 usecs/loop -- avg 4.139987 483.1 KHz .910 1.000
> 3.15.4-default 4.355594 usecs/loop -- avg 4.351961 459.6 KHz .866 .951 1.000
> 3.16.0-default 4.537279 usecs/loop -- avg 4.543532 440.2 KHz .829 .911 .957
>
> 3.0.101-smp 3.692377 usecs/loop -- avg 3.690774 541.9 KHz 1.000
> 3.14.10-smp 4.010009 usecs/loop -- avg 4.009019 498.9 KHz .920
> 3.15.4-smp 3.882398 usecs/loop -- avg 3.884095 514.9 KHz .950
> 3.16.0-master 4.061003 usecs/loop -- avg 4.058244 492.8 KHz .909
Urgh,.. I need to go fix that :/
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 13:36 ` Peter Zijlstra
@ 2014-07-15 15:21 ` Tejun Heo
2014-07-15 16:37 ` Peter Zijlstra
0 siblings, 1 reply; 35+ messages in thread
From: Tejun Heo @ 2014-07-15 15:21 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Tim Chen, Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, Jul 15, 2014 at 03:36:27PM +0200, Peter Zijlstra wrote:
> So, just to expand on this, we're already getting 'bug' reports because
> worker threads are not cgroup aware. If work gets generated inside some
> cgroup, the worker doesn't care and runs the worker thread wherever
> (typically the root cgroup).
>
> This means that the 'work' escapes the cgroup confines and creates
> resource inversion etc. The same is of course true for nice and RT
> priorities.
>
> TJ, are you aware of this and/or given it any throught?
Yeap, I'm aware of the issue but haven't read any actual bug reports
yet. Can you point me to the reports?
Given that worker pool management is dynamic, spawning separate pools
for individual cgroups on-demand should be doable. Haven't been able
to decide how much we should be willing to pay in terms of complexity
yet.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 15:21 ` Tejun Heo
@ 2014-07-15 16:37 ` Peter Zijlstra
0 siblings, 0 replies; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 16:37 UTC (permalink / raw)
To: Tejun Heo
Cc: Tim Chen, Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 2505 bytes --]
On Tue, Jul 15, 2014 at 11:21:49AM -0400, Tejun Heo wrote:
> On Tue, Jul 15, 2014 at 03:36:27PM +0200, Peter Zijlstra wrote:
> > So, just to expand on this, we're already getting 'bug' reports because
> > worker threads are not cgroup aware. If work gets generated inside some
> > cgroup, the worker doesn't care and runs the worker thread wherever
> > (typically the root cgroup).
> >
> > This means that the 'work' escapes the cgroup confines and creates
> > resource inversion etc. The same is of course true for nice and RT
> > priorities.
> >
> > TJ, are you aware of this and/or given it any throught?
>
> Yeap, I'm aware of the issue but haven't read any actual bug reports
> yet. Can you point me to the reports?
lkml.kernel.org/r/53A8EC1E.1060504@linux.vnet.ibm.com
The root level workqueue thingies disturb the cgroup level scheduling to
'some' extend.
That whole thread is somewhat confusing and I think there's more than
just this going on, but they're really seeing this as a pain point.
> Given that worker pool management is dynamic, spawning separate pools
> for individual cgroups on-demand should be doable. Haven't been able
> to decide how much we should be willing to pay in terms of complexity
> yet.
Yah, I figured. Back before you ripped up the workqueue I had a
worklet-PI patch in -rt, which basically sorted and ran works in a
RR/FIFO priority order, including boosting the current work when a
higher prio one was pending etc.
I never really figured out a way to make the new concurrent stuff do
something like that, and this 'problem' here is harder still, because
they're not static prios etc.
Ideally we'd run the works _in_ the same task-context (from a scheduler
POV) as the task creating the work. There's some very obvious problems
of implementation there, and some less obvious others, so bleh.
Also, there's the whole softirq trainwreck, which has many of the same
problems. Much of the network stack isn't necessarily aware for whom
they're doing work, so no way to propagate.
Point in case for the crypto stuff I suppose, that's a combination of
the two, god only knows who we should be accounting it to and in what
context things should run.
Ideally a socket has a 'single' (ha! if only) owner, and we'd know
throughout the entirely rx/tx paths, but I doubt we actually have that.
(Note that there's people really suffering because of this..)
Same for the 'shiny' block-mq stuff I suppose :-(
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 14:53 ` Peter Zijlstra
@ 2014-07-15 18:06 ` Mike Galbraith
2014-07-15 19:03 ` Peter Zijlstra
0 siblings, 1 reply; 35+ messages in thread
From: Mike Galbraith @ 2014-07-15 18:06 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Thomas Gleixner, Tim Chen, Herbert Xu, H. Peter Anvin,
David S.Miller, Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal,
James Guilford, Wajdi Feghali, Jussi Kivilinna, linux-crypto,
linux-kernel
On Tue, 2014-07-15 at 16:53 +0200, Peter Zijlstra wrote:
> On Tue, Jul 15, 2014 at 04:45:25PM +0200, Mike Galbraith wrote:
> >
> > 3.0.101-default 3.753363 usecs/loop -- avg 3.770737 530.4 KHz 1.000
> > 3.14.10-default 4.145348 usecs/loop -- avg 4.139987 483.1 KHz .910 1.000
> > 3.15.4-default 4.355594 usecs/loop -- avg 4.351961 459.6 KHz .866 .951 1.000
> > 3.16.0-default 4.537279 usecs/loop -- avg 4.543532 440.2 KHz .829 .911 .957
> >
> > 3.0.101-smp 3.692377 usecs/loop -- avg 3.690774 541.9 KHz 1.000
> > 3.14.10-smp 4.010009 usecs/loop -- avg 4.009019 498.9 KHz .920
> > 3.15.4-smp 3.882398 usecs/loop -- avg 3.884095 514.9 KHz .950
> > 3.16.0-master 4.061003 usecs/loop -- avg 4.058244 492.8 KHz .909
>
> Urgh,.. I need to go fix that :/
I'm poking about. It's not just one thing 'course, just lots of change
adding up to less than wonderful. Idle changes are costing some, for
obese config, avg goop. The select_next_task() reorganization appears
to be costing, but eyeballing, I can see no excuse for that at all.
-Mike
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 9:50 ` Peter Zijlstra
2014-07-15 12:07 ` Peter Zijlstra
@ 2014-07-15 18:40 ` Tim Chen
1 sibling, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-15 18:40 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, 2014-07-15 at 11:50 +0200, Peter Zijlstra wrote:
> On Mon, Jul 14, 2014 at 12:50:50PM -0700, Tim Chen wrote:
>
> > There is a generic multi-buffer infrastructure portion that manages
> > pulling and queuing jobs on the crypto workqueue, and it is separated out
> > in patch 1 of the patchset.
>
> There's one very weird multi-line comment in that patch.
>
> > The other portions are algorithm specific that defines
> > algorithm specific data structure and does the crypto computation
> > for a particular algorithm, mostly in
> > assemblies and C glue code. The infrastructure code is
> > meant to be reused for other similar
> > multi-buffer algorithms.
>
> The flushing part that uses the sched thing is sha1 specific, even
> though it strikes me as not being so. Flushing buffers on idle seems
> like a 'generic' thing.
>
> > We use nr_running_cpu to check whether there are other tasks running on
> > the *current* cpu, (not for another cpu),
>
> And yet, the function allows you do to exactly that..
How about a function "single_task_running()"? Something like
bool single_task_running()
{
if (cpu_rq(smp_processor_id())->nr_running == 1)
return true;
else
return false;
}
>
> > to decide if we should flush
> > and compute crypto jobs accumulated. If there's nobody else running,
> > we can take advantage of available cpu cycles on the cpu we are running
> > on to do computation on the existing jobs in a SIMD mannner.
> > Waiting a bit longer may accumulate more jobs to process in parallel
> > in a single SIMD instruction, but will have more delay.
>
> So you already have an idle notifier (which is x86 only, we should fix
> that I suppose), and you then double check there really isn't anything
> else running.
>
> How much, if anything, does that second check buy you? There's just not
> a single word on that.
>
> Also, there is not a word on the latency vs throughput tradeoff you
> make. I can imagine that for very short idle durations you loose, not
> win with this thing.
I am not crazy about the idle_notifier implementation either. Thinking
a bit more, I can probably achieve most of what I need in the code path
that pull jobs from the work queue. I'll flush the remaining
partially processed jobs if I don't have any more jobs to process and no
other tasks are running (i.e. single_task_running() is true).
I'll try to move this logic into the generic multi-buffer job handler
path and out of the idle_notifier.
>
> So for now I still see no reason for doing this.
>
> Also, I wonder about SMT, the point of this is to make best use of the
> SIMD pipelines, does it still make sense to use siblings at the same
> time even though you're running hand crafted ASM to stuff the pipelines
> to the brim? Should this thing be SMT aware and not gather queues for
> both siblings?
I am processing the crypto job on the same cpu that the crypto
jobs originate from to get best cache utilization.
It will be the scheduler's decision to move the task that's
generating a lot crypto jobs to another cpu, if the sibling
is also very busy.
Thanks for your comments and reviews.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 12:07 ` Peter Zijlstra
2014-07-15 12:59 ` Thomas Gleixner
@ 2014-07-15 18:40 ` Tim Chen
1 sibling, 0 replies; 35+ messages in thread
From: Tim Chen @ 2014-07-15 18:40 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Herbert Xu, H. Peter Anvin, David S.Miller, Ingo Molnar,
Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, 2014-07-15 at 14:07 +0200, Peter Zijlstra wrote:
> On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> > So you already have an idle notifier (which is x86 only, we should fix
> > that I suppose), and you then double check there really isn't anything
> > else running.
>
> Note that we've already done a large part of the expense of going idle
> by the time we call that idle notifier -- in specific, we've
> reprogrammed the clock to stop the tick.
>
> Its really wasteful to then generate work again, which means we have to
> again reprogram the clock etc.
Will try another version of the patch without using the idle notifier.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 12:59 ` Thomas Gleixner
2014-07-15 14:45 ` Mike Galbraith
@ 2014-07-15 18:41 ` Tim Chen
2014-07-15 20:46 ` Thomas Gleixner
1 sibling, 1 reply; 35+ messages in thread
From: Tim Chen @ 2014-07-15 18:41 UTC (permalink / raw)
To: Thomas Gleixner
Cc: Peter Zijlstra, Herbert Xu, H. Peter Anvin, David S.Miller,
Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, 2014-07-15 at 14:59 +0200, Thomas Gleixner wrote:
> On Tue, 15 Jul 2014, Peter Zijlstra wrote:
>
> > On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> > > So you already have an idle notifier (which is x86 only, we should fix
> > > that I suppose), and you then double check there really isn't anything
> > > else running.
> >
> > Note that we've already done a large part of the expense of going idle
> > by the time we call that idle notifier -- in specific, we've
> > reprogrammed the clock to stop the tick.
> >
> > Its really wasteful to then generate work again, which means we have to
> > again reprogram the clock etc.
>
> Doing anything which is not related to idle itself in the idle
> notifier is just plain wrong.
I don't like the kicking the multi-buffer job flush using idle_notifier
path either. I'll try another version of the patch by doing this in the
multi-buffer job handler path.
>
> If that stuff wants to utilize idle slots, we really need to come up
> with a generic and general solution. Otherwise we'll grow those warts
> all over the architecture space, with slightly different ways of
> wreckaging the world an some more.
>
> This whole attidute of people thinking that they need their own
> specialized scheduling around the real scheduler is a PITA. All this
> stuff is just damanging any sensible approach of power saving, load
> balancing, etc.
>
> What we really want is infrastructure, which allows the scheduler to
> actively query the async work situation and based on the results
> actively decide when to process it and where.
I agree with you. It will be great if we have such infrastructure.
Thanks.
Tim
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 18:06 ` Mike Galbraith
@ 2014-07-15 19:03 ` Peter Zijlstra
2014-07-15 19:24 ` Mike Galbraith
0 siblings, 1 reply; 35+ messages in thread
From: Peter Zijlstra @ 2014-07-15 19:03 UTC (permalink / raw)
To: Mike Galbraith
Cc: Thomas Gleixner, Tim Chen, Herbert Xu, H. Peter Anvin,
David S.Miller, Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal,
James Guilford, Wajdi Feghali, Jussi Kivilinna, linux-crypto,
linux-kernel
[-- Attachment #1: Type: text/plain, Size: 1437 bytes --]
On Tue, Jul 15, 2014 at 08:06:55PM +0200, Mike Galbraith wrote:
> On Tue, 2014-07-15 at 16:53 +0200, Peter Zijlstra wrote:
> > On Tue, Jul 15, 2014 at 04:45:25PM +0200, Mike Galbraith wrote:
> > >
> > > 3.0.101-default 3.753363 usecs/loop -- avg 3.770737 530.4 KHz 1.000
> > > 3.14.10-default 4.145348 usecs/loop -- avg 4.139987 483.1 KHz .910 1.000
> > > 3.15.4-default 4.355594 usecs/loop -- avg 4.351961 459.6 KHz .866 .951 1.000
> > > 3.16.0-default 4.537279 usecs/loop -- avg 4.543532 440.2 KHz .829 .911 .957
> > >
> > > 3.0.101-smp 3.692377 usecs/loop -- avg 3.690774 541.9 KHz 1.000
> > > 3.14.10-smp 4.010009 usecs/loop -- avg 4.009019 498.9 KHz .920
> > > 3.15.4-smp 3.882398 usecs/loop -- avg 3.884095 514.9 KHz .950
> > > 3.16.0-master 4.061003 usecs/loop -- avg 4.058244 492.8 KHz .909
> >
> > Urgh,.. I need to go fix that :/
>
> I'm poking about. It's not just one thing 'course, just lots of change
> adding up to less than wonderful. Idle changes are costing some, for
> obese config, avg goop. The select_next_task() reorganization appears
> to be costing, but eyeballing, I can see no excuse for that at all.
How is the idle stuff costing, cpu-affine pipe-test should pretty much
peg a cpu at 100%, right? Or did I mis-understand and are you running a
loose pipe-test?
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 19:03 ` Peter Zijlstra
@ 2014-07-15 19:24 ` Mike Galbraith
0 siblings, 0 replies; 35+ messages in thread
From: Mike Galbraith @ 2014-07-15 19:24 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Thomas Gleixner, Tim Chen, Herbert Xu, H. Peter Anvin,
David S.Miller, Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal,
James Guilford, Wajdi Feghali, Jussi Kivilinna, linux-crypto,
linux-kernel
On Tue, 2014-07-15 at 21:03 +0200, Peter Zijlstra wrote:
> On Tue, Jul 15, 2014 at 08:06:55PM +0200, Mike Galbraith wrote:
> > On Tue, 2014-07-15 at 16:53 +0200, Peter Zijlstra wrote:
> > > On Tue, Jul 15, 2014 at 04:45:25PM +0200, Mike Galbraith wrote:
> > > >
> > > > 3.0.101-default 3.753363 usecs/loop -- avg 3.770737 530.4 KHz 1.000
> > > > 3.14.10-default 4.145348 usecs/loop -- avg 4.139987 483.1 KHz .910 1.000
> > > > 3.15.4-default 4.355594 usecs/loop -- avg 4.351961 459.6 KHz .866 .951 1.000
> > > > 3.16.0-default 4.537279 usecs/loop -- avg 4.543532 440.2 KHz .829 .911 .957
> > > >
> > > > 3.0.101-smp 3.692377 usecs/loop -- avg 3.690774 541.9 KHz 1.000
> > > > 3.14.10-smp 4.010009 usecs/loop -- avg 4.009019 498.9 KHz .920
> > > > 3.15.4-smp 3.882398 usecs/loop -- avg 3.884095 514.9 KHz .950
> > > > 3.16.0-master 4.061003 usecs/loop -- avg 4.058244 492.8 KHz .909
> > >
> > > Urgh,.. I need to go fix that :/
> >
> > I'm poking about. It's not just one thing 'course, just lots of change
> > adding up to less than wonderful. Idle changes are costing some, for
> > obese config, avg goop. The select_next_task() reorganization appears
> > to be costing, but eyeballing, I can see no excuse for that at all.
>
> How is the idle stuff costing, cpu-affine pipe-test should pretty much
> peg a cpu at 100%, right? Or did I mis-understand and are you running a
> loose pipe-test?
Exactly, I'm measuring the cost of popping in and out of idle while
trying to reclaim overlap (which doesn't exist in pipe-test case).
-Mike
^ permalink raw reply [flat|nested] 35+ messages in thread
* Re: [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu
2014-07-15 18:41 ` Tim Chen
@ 2014-07-15 20:46 ` Thomas Gleixner
0 siblings, 0 replies; 35+ messages in thread
From: Thomas Gleixner @ 2014-07-15 20:46 UTC (permalink / raw)
To: Tim Chen
Cc: Peter Zijlstra, Herbert Xu, H. Peter Anvin, David S.Miller,
Ingo Molnar, Chandramouli Narayanan, Vinodh Gopal, James Guilford,
Wajdi Feghali, Jussi Kivilinna, linux-crypto, linux-kernel
On Tue, 15 Jul 2014, Tim Chen wrote:
> On Tue, 2014-07-15 at 14:59 +0200, Thomas Gleixner wrote:
> > On Tue, 15 Jul 2014, Peter Zijlstra wrote:
> >
> > > On Tue, Jul 15, 2014 at 11:50:45AM +0200, Peter Zijlstra wrote:
> > > > So you already have an idle notifier (which is x86 only, we should fix
> > > > that I suppose), and you then double check there really isn't anything
> > > > else running.
> > >
> > > Note that we've already done a large part of the expense of going idle
> > > by the time we call that idle notifier -- in specific, we've
> > > reprogrammed the clock to stop the tick.
> > >
> > > Its really wasteful to then generate work again, which means we have to
> > > again reprogram the clock etc.
> >
> > Doing anything which is not related to idle itself in the idle
> > notifier is just plain wrong.
>
> I don't like the kicking the multi-buffer job flush using idle_notifier
> path either. I'll try another version of the patch by doing this in the
> multi-buffer job handler path.
>
> >
> > If that stuff wants to utilize idle slots, we really need to come up
> > with a generic and general solution. Otherwise we'll grow those warts
> > all over the architecture space, with slightly different ways of
> > wreckaging the world an some more.
> >
> > This whole attidute of people thinking that they need their own
> > specialized scheduling around the real scheduler is a PITA. All this
> > stuff is just damanging any sensible approach of power saving, load
> > balancing, etc.
> >
> > What we really want is infrastructure, which allows the scheduler to
> > actively query the async work situation and based on the results
> > actively decide when to process it and where.
>
> I agree with you. It will be great if we have such infrastructure.
You are heartly invited to come up with that. :)
^ permalink raw reply [flat|nested] 35+ messages in thread
end of thread, other threads:[~2014-07-15 20:47 UTC | newest]
Thread overview: 35+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <cover.1405074379.git.tim.c.chen@linux.intel.com>
2014-07-11 20:32 ` [PATCH v4 0/7] crypto: SHA1 multibuffer implementation Tim Chen
2014-07-11 20:32 ` [PATCH v4 1/7] crypto: SHA1 multibuffer crypto hash infrastructure Tim Chen
2014-07-11 20:32 ` [PATCH v4 2/7] crypto: SHA1 multibuffer algorithm data structures Tim Chen
2014-07-11 20:32 ` [PATCH v4 3/7] crypto: SHA1 multibuffer submit and flush routines for AVX2 Tim Chen
2014-07-11 20:32 ` [PATCH v4 4/7] crypto: SHA1 multibuffer crypto computation (x8 AVX2) Tim Chen
2014-07-11 20:33 ` [PATCH v4 5/7] crypto: SHA1 multibuffer scheduler Tim Chen
2014-07-11 20:33 ` [PATCH v4 6/7] sched: add function nr_running_cpu to expose number of tasks running on cpu Tim Chen
2014-07-12 9:25 ` Kirill Tkhai
2014-07-14 17:51 ` Tim Chen
2014-07-12 14:21 ` Tadeusz Struk
2014-07-14 23:51 ` Tim Chen
2014-07-14 10:16 ` Peter Zijlstra
2014-07-14 16:10 ` Tim Chen
2014-07-14 16:14 ` Peter Zijlstra
2014-07-14 17:05 ` Tim Chen
2014-07-14 18:17 ` Peter Zijlstra
2014-07-14 19:08 ` Tim Chen
2014-07-14 19:15 ` Peter Zijlstra
2014-07-14 19:50 ` Tim Chen
2014-07-15 9:50 ` Peter Zijlstra
2014-07-15 12:07 ` Peter Zijlstra
2014-07-15 12:59 ` Thomas Gleixner
2014-07-15 14:45 ` Mike Galbraith
2014-07-15 14:53 ` Peter Zijlstra
2014-07-15 18:06 ` Mike Galbraith
2014-07-15 19:03 ` Peter Zijlstra
2014-07-15 19:24 ` Mike Galbraith
2014-07-15 18:41 ` Tim Chen
2014-07-15 20:46 ` Thomas Gleixner
2014-07-15 18:40 ` Tim Chen
2014-07-15 18:40 ` Tim Chen
2014-07-15 13:36 ` Peter Zijlstra
2014-07-15 15:21 ` Tejun Heo
2014-07-15 16:37 ` Peter Zijlstra
2014-07-11 20:33 ` [PATCH v4 7/7] crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle Tim Chen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).