OpenCloudOS-Kernel/crypto/cryptd.c

1131 lines
29 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Software async crypto daemon.
*
* Copyright (c) 2006 Herbert Xu <herbert@gondor.apana.org.au>
*
* Added AEAD support to cryptd.
* Authors: Tadeusz Struk (tadeusz.struk@intel.com)
* Adrian Hoban <adrian.hoban@intel.com>
* Gabriele Paoloni <gabriele.paoloni@intel.com>
* Aidan O'Mahony (aidan.o.mahony@intel.com)
* Copyright (c) 2010, Intel Corporation.
*/
#include <crypto/internal/hash.h>
#include <crypto/internal/aead.h>
#include <crypto/internal/skcipher.h>
#include <crypto/cryptd.h>
#include <linux/refcount.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/workqueue.h>
static unsigned int cryptd_max_cpu_qlen = 1000;
module_param(cryptd_max_cpu_qlen, uint, 0);
MODULE_PARM_DESC(cryptd_max_cpu_qlen, "Set cryptd Max queue depth");
static struct workqueue_struct *cryptd_wq;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
struct cryptd_cpu_queue {
struct crypto_queue queue;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
struct work_struct work;
};
struct cryptd_queue {
/*
* Protected by disabling BH to allow enqueueing from softinterrupt and
* dequeuing from kworker (cryptd_queue_worker()).
*/
struct cryptd_cpu_queue __percpu *cpu_queue;
};
struct cryptd_instance_ctx {
struct crypto_spawn spawn;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
struct cryptd_queue *queue;
};
struct skcipherd_instance_ctx {
struct crypto_skcipher_spawn spawn;
struct cryptd_queue *queue;
};
struct hashd_instance_ctx {
struct crypto_shash_spawn spawn;
struct cryptd_queue *queue;
};
struct aead_instance_ctx {
struct crypto_aead_spawn aead_spawn;
struct cryptd_queue *queue;
};
struct cryptd_skcipher_ctx {
refcount_t refcnt;
struct crypto_skcipher *child;
};
struct cryptd_skcipher_request_ctx {
struct skcipher_request req;
};
struct cryptd_hash_ctx {
refcount_t refcnt;
struct crypto_shash *child;
};
struct cryptd_hash_request_ctx {
crypto_completion_t complete;
void *data;
struct shash_desc desc;
};
struct cryptd_aead_ctx {
refcount_t refcnt;
struct crypto_aead *child;
};
struct cryptd_aead_request_ctx {
struct aead_request req;
};
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
static void cryptd_queue_worker(struct work_struct *work);
static int cryptd_init_queue(struct cryptd_queue *queue,
unsigned int max_cpu_qlen)
{
int cpu;
struct cryptd_cpu_queue *cpu_queue;
queue->cpu_queue = alloc_percpu(struct cryptd_cpu_queue);
if (!queue->cpu_queue)
return -ENOMEM;
for_each_possible_cpu(cpu) {
cpu_queue = per_cpu_ptr(queue->cpu_queue, cpu);
crypto_init_queue(&cpu_queue->queue, max_cpu_qlen);
INIT_WORK(&cpu_queue->work, cryptd_queue_worker);
}
pr_info("cryptd: max_cpu_qlen set to %d\n", max_cpu_qlen);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
return 0;
}
static void cryptd_fini_queue(struct cryptd_queue *queue)
{
int cpu;
struct cryptd_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 cryptd_enqueue_request(struct cryptd_queue *queue,
struct crypto_async_request *request)
{
int err;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
struct cryptd_cpu_queue *cpu_queue;
refcount_t *refcnt;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
local_bh_disable();
cpu_queue = this_cpu_ptr(queue->cpu_queue);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
err = crypto_enqueue_request(&cpu_queue->queue, request);
refcnt = crypto_tfm_ctx(request->tfm);
if (err == -ENOSPC)
goto out;
queue_work_on(smp_processor_id(), cryptd_wq, &cpu_queue->work);
if (!refcount_read(refcnt))
goto out;
refcount_inc(refcnt);
out:
local_bh_enable();
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
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 cryptd_queue_worker(struct work_struct *work)
{
struct cryptd_cpu_queue *cpu_queue;
struct crypto_async_request *req, *backlog;
cpu_queue = container_of(work, struct cryptd_cpu_queue, work);
/*
* Only handle one request at a time to avoid hogging crypto workqueue.
*/
local_bh_disable();
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
backlog = crypto_get_backlog(&cpu_queue->queue);
req = crypto_dequeue_request(&cpu_queue->queue);
local_bh_enable();
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
if (!req)
return;
if (backlog)
crypto_request_complete(backlog, -EINPROGRESS);
crypto_request_complete(req, 0);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
if (cpu_queue->queue.qlen)
queue_work(cryptd_wq, &cpu_queue->work);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
}
static inline struct cryptd_queue *cryptd_get_queue(struct crypto_tfm *tfm)
{
struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
struct cryptd_instance_ctx *ictx = crypto_instance_ctx(inst);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
return ictx->queue;
}
static void cryptd_type_and_mask(struct crypto_attr_type *algt,
u32 *type, u32 *mask)
{
/*
* cryptd is allowed to wrap internal algorithms, but in that case the
* resulting cryptd instance will be marked as internal as well.
*/
*type = algt->type & CRYPTO_ALG_INTERNAL;
*mask = algt->mask & CRYPTO_ALG_INTERNAL;
/* No point in cryptd wrapping an algorithm that's already async. */
*mask |= CRYPTO_ALG_ASYNC;
*mask |= crypto_algt_inherited_mask(algt);
}
static int cryptd_init_instance(struct crypto_instance *inst,
struct crypto_alg *alg)
{
if (snprintf(inst->alg.cra_driver_name, CRYPTO_MAX_ALG_NAME,
"cryptd(%s)",
alg->cra_driver_name) >= CRYPTO_MAX_ALG_NAME)
return -ENAMETOOLONG;
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;
return 0;
}
static int cryptd_skcipher_setkey(struct crypto_skcipher *parent,
const u8 *key, unsigned int keylen)
{
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(parent);
struct crypto_skcipher *child = ctx->child;
crypto_skcipher_clear_flags(child, CRYPTO_TFM_REQ_MASK);
crypto_skcipher_set_flags(child,
crypto_skcipher_get_flags(parent) &
CRYPTO_TFM_REQ_MASK);
return crypto_skcipher_setkey(child, key, keylen);
}
static struct skcipher_request *cryptd_skcipher_prepare(
struct skcipher_request *req, int err)
{
struct cryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req);
struct skcipher_request *subreq = &rctx->req;
struct cryptd_skcipher_ctx *ctx;
struct crypto_skcipher *child;
req->base.complete = subreq->base.complete;
req->base.data = subreq->base.data;
if (unlikely(err == -EINPROGRESS))
return NULL;
ctx = crypto_skcipher_ctx(crypto_skcipher_reqtfm(req));
child = ctx->child;
skcipher_request_set_tfm(subreq, child);
skcipher_request_set_callback(subreq, CRYPTO_TFM_REQ_MAY_SLEEP,
NULL, NULL);
skcipher_request_set_crypt(subreq, req->src, req->dst, req->cryptlen,
req->iv);
return subreq;
}
static void cryptd_skcipher_complete(struct skcipher_request *req, int err,
crypto_completion_t complete)
{
struct cryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req);
struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm);
struct skcipher_request *subreq = &rctx->req;
int refcnt = refcount_read(&ctx->refcnt);
local_bh_disable();
skcipher_request_complete(req, err);
local_bh_enable();
if (unlikely(err == -EINPROGRESS)) {
subreq->base.complete = req->base.complete;
subreq->base.data = req->base.data;
req->base.complete = complete;
req->base.data = req;
} else if (refcnt && refcount_dec_and_test(&ctx->refcnt))
crypto_free_skcipher(tfm);
}
static void cryptd_skcipher_encrypt(void *data, int err)
{
struct skcipher_request *req = data;
struct skcipher_request *subreq;
subreq = cryptd_skcipher_prepare(req, err);
if (likely(subreq))
err = crypto_skcipher_encrypt(subreq);
cryptd_skcipher_complete(req, err, cryptd_skcipher_encrypt);
}
static void cryptd_skcipher_decrypt(void *data, int err)
{
struct skcipher_request *req = data;
struct skcipher_request *subreq;
subreq = cryptd_skcipher_prepare(req, err);
if (likely(subreq))
err = crypto_skcipher_decrypt(subreq);
cryptd_skcipher_complete(req, err, cryptd_skcipher_decrypt);
}
static int cryptd_skcipher_enqueue(struct skcipher_request *req,
crypto_completion_t compl)
{
struct cryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req);
struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
struct skcipher_request *subreq = &rctx->req;
struct cryptd_queue *queue;
queue = cryptd_get_queue(crypto_skcipher_tfm(tfm));
subreq->base.complete = req->base.complete;
subreq->base.data = req->base.data;
req->base.complete = compl;
req->base.data = req;
return cryptd_enqueue_request(queue, &req->base);
}
static int cryptd_skcipher_encrypt_enqueue(struct skcipher_request *req)
{
return cryptd_skcipher_enqueue(req, cryptd_skcipher_encrypt);
}
static int cryptd_skcipher_decrypt_enqueue(struct skcipher_request *req)
{
return cryptd_skcipher_enqueue(req, cryptd_skcipher_decrypt);
}
static int cryptd_skcipher_init_tfm(struct crypto_skcipher *tfm)
{
struct skcipher_instance *inst = skcipher_alg_instance(tfm);
struct skcipherd_instance_ctx *ictx = skcipher_instance_ctx(inst);
struct crypto_skcipher_spawn *spawn = &ictx->spawn;
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm);
struct crypto_skcipher *cipher;
cipher = crypto_spawn_skcipher(spawn);
if (IS_ERR(cipher))
return PTR_ERR(cipher);
ctx->child = cipher;
crypto_skcipher_set_reqsize(
tfm, sizeof(struct cryptd_skcipher_request_ctx) +
crypto_skcipher_reqsize(cipher));
return 0;
}
static void cryptd_skcipher_exit_tfm(struct crypto_skcipher *tfm)
{
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm);
crypto_free_skcipher(ctx->child);
}
static void cryptd_skcipher_free(struct skcipher_instance *inst)
{
struct skcipherd_instance_ctx *ctx = skcipher_instance_ctx(inst);
crypto_drop_skcipher(&ctx->spawn);
kfree(inst);
}
static int cryptd_create_skcipher(struct crypto_template *tmpl,
struct rtattr **tb,
struct crypto_attr_type *algt,
struct cryptd_queue *queue)
{
struct skcipherd_instance_ctx *ctx;
struct skcipher_instance *inst;
struct skcipher_alg *alg;
u32 type;
u32 mask;
int err;
cryptd_type_and_mask(algt, &type, &mask);
inst = kzalloc(sizeof(*inst) + sizeof(*ctx), GFP_KERNEL);
if (!inst)
return -ENOMEM;
ctx = skcipher_instance_ctx(inst);
ctx->queue = queue;
err = crypto_grab_skcipher(&ctx->spawn, skcipher_crypto_instance(inst),
crypto_attr_alg_name(tb[1]), type, mask);
if (err)
goto err_free_inst;
alg = crypto_spawn_skcipher_alg(&ctx->spawn);
err = cryptd_init_instance(skcipher_crypto_instance(inst), &alg->base);
if (err)
goto err_free_inst;
inst->alg.base.cra_flags |= CRYPTO_ALG_ASYNC |
(alg->base.cra_flags & CRYPTO_ALG_INTERNAL);
inst->alg.ivsize = crypto_skcipher_alg_ivsize(alg);
inst->alg.chunksize = crypto_skcipher_alg_chunksize(alg);
inst->alg.min_keysize = crypto_skcipher_alg_min_keysize(alg);
inst->alg.max_keysize = crypto_skcipher_alg_max_keysize(alg);
inst->alg.base.cra_ctxsize = sizeof(struct cryptd_skcipher_ctx);
inst->alg.init = cryptd_skcipher_init_tfm;
inst->alg.exit = cryptd_skcipher_exit_tfm;
inst->alg.setkey = cryptd_skcipher_setkey;
inst->alg.encrypt = cryptd_skcipher_encrypt_enqueue;
inst->alg.decrypt = cryptd_skcipher_decrypt_enqueue;
inst->free = cryptd_skcipher_free;
err = skcipher_register_instance(tmpl, inst);
if (err) {
err_free_inst:
cryptd_skcipher_free(inst);
}
return err;
}
static int cryptd_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 cryptd_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 cryptd_hash_request_ctx) +
crypto_shash_descsize(hash));
return 0;
}
static void cryptd_hash_exit_tfm(struct crypto_tfm *tfm)
{
struct cryptd_hash_ctx *ctx = crypto_tfm_ctx(tfm);
crypto_free_shash(ctx->child);
}
static int cryptd_hash_setkey(struct crypto_ahash *parent,
const u8 *key, unsigned int keylen)
{
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(parent);
struct crypto_shash *child = ctx->child;
crypto_shash_clear_flags(child, CRYPTO_TFM_REQ_MASK);
crypto_shash_set_flags(child, crypto_ahash_get_flags(parent) &
CRYPTO_TFM_REQ_MASK);
return crypto_shash_setkey(child, key, keylen);
}
static int cryptd_hash_enqueue(struct ahash_request *req,
crypto_completion_t compl)
{
struct cryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
struct cryptd_queue *queue =
cryptd_get_queue(crypto_ahash_tfm(tfm));
rctx->complete = req->base.complete;
rctx->data = req->base.data;
req->base.complete = compl;
req->base.data = req;
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
return cryptd_enqueue_request(queue, &req->base);
}
static struct shash_desc *cryptd_hash_prepare(struct ahash_request *req,
int err)
{
struct cryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
req->base.complete = rctx->complete;
req->base.data = rctx->data;
if (unlikely(err == -EINPROGRESS))
return NULL;
return &rctx->desc;
}
static void cryptd_hash_complete(struct ahash_request *req, int err,
crypto_completion_t complete)
{
struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(tfm);
int refcnt = refcount_read(&ctx->refcnt);
local_bh_disable();
ahash_request_complete(req, err);
local_bh_enable();
if (err == -EINPROGRESS) {
req->base.complete = complete;
req->base.data = req;
} else if (refcnt && refcount_dec_and_test(&ctx->refcnt))
crypto_free_ahash(tfm);
}
static void cryptd_hash_init(void *data, int err)
{
struct ahash_request *req = data;
struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(tfm);
struct crypto_shash *child = ctx->child;
struct shash_desc *desc;
desc = cryptd_hash_prepare(req, err);
if (unlikely(!desc))
goto out;
desc->tfm = child;
err = crypto_shash_init(desc);
out:
cryptd_hash_complete(req, err, cryptd_hash_init);
}
static int cryptd_hash_init_enqueue(struct ahash_request *req)
{
return cryptd_hash_enqueue(req, cryptd_hash_init);
}
static void cryptd_hash_update(void *data, int err)
{
struct ahash_request *req = data;
struct shash_desc *desc;
desc = cryptd_hash_prepare(req, err);
if (likely(desc))
err = shash_ahash_update(req, desc);
cryptd_hash_complete(req, err, cryptd_hash_update);
}
static int cryptd_hash_update_enqueue(struct ahash_request *req)
{
return cryptd_hash_enqueue(req, cryptd_hash_update);
}
static void cryptd_hash_final(void *data, int err)
{
struct ahash_request *req = data;
struct shash_desc *desc;
desc = cryptd_hash_prepare(req, err);
if (likely(desc))
err = crypto_shash_final(desc, req->result);
cryptd_hash_complete(req, err, cryptd_hash_final);
}
static int cryptd_hash_final_enqueue(struct ahash_request *req)
{
return cryptd_hash_enqueue(req, cryptd_hash_final);
}
static void cryptd_hash_finup(void *data, int err)
{
struct ahash_request *req = data;
struct shash_desc *desc;
desc = cryptd_hash_prepare(req, err);
if (likely(desc))
err = shash_ahash_finup(req, desc);
cryptd_hash_complete(req, err, cryptd_hash_finup);
}
static int cryptd_hash_finup_enqueue(struct ahash_request *req)
{
return cryptd_hash_enqueue(req, cryptd_hash_finup);
}
static void cryptd_hash_digest(void *data, int err)
{
struct ahash_request *req = data;
struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(tfm);
struct crypto_shash *child = ctx->child;
struct shash_desc *desc;
desc = cryptd_hash_prepare(req, err);
if (unlikely(!desc))
goto out;
desc->tfm = child;
err = shash_ahash_digest(req, desc);
out:
cryptd_hash_complete(req, err, cryptd_hash_digest);
}
static int cryptd_hash_digest_enqueue(struct ahash_request *req)
{
return cryptd_hash_enqueue(req, cryptd_hash_digest);
}
static int cryptd_hash_export(struct ahash_request *req, void *out)
{
struct cryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
return crypto_shash_export(&rctx->desc, out);
}
static int cryptd_hash_import(struct ahash_request *req, const void *in)
{
struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(tfm);
struct shash_desc *desc = cryptd_shash_desc(req);
desc->tfm = ctx->child;
return crypto_shash_import(desc, in);
}
static void cryptd_hash_free(struct ahash_instance *inst)
{
struct hashd_instance_ctx *ctx = ahash_instance_ctx(inst);
crypto_drop_shash(&ctx->spawn);
kfree(inst);
}
static int cryptd_create_hash(struct crypto_template *tmpl, struct rtattr **tb,
struct crypto_attr_type *algt,
struct cryptd_queue *queue)
{
struct hashd_instance_ctx *ctx;
struct ahash_instance *inst;
struct shash_alg *alg;
u32 type;
u32 mask;
int err;
cryptd_type_and_mask(algt, &type, &mask);
inst = kzalloc(sizeof(*inst) + sizeof(*ctx), GFP_KERNEL);
if (!inst)
return -ENOMEM;
ctx = ahash_instance_ctx(inst);
ctx->queue = queue;
err = crypto_grab_shash(&ctx->spawn, ahash_crypto_instance(inst),
crypto_attr_alg_name(tb[1]), type, mask);
if (err)
goto err_free_inst;
alg = crypto_spawn_shash_alg(&ctx->spawn);
err = cryptd_init_instance(ahash_crypto_instance(inst), &alg->base);
if (err)
goto err_free_inst;
inst->alg.halg.base.cra_flags |= CRYPTO_ALG_ASYNC |
(alg->base.cra_flags & (CRYPTO_ALG_INTERNAL|
CRYPTO_ALG_OPTIONAL_KEY));
inst->alg.halg.digestsize = alg->digestsize;
inst->alg.halg.statesize = alg->statesize;
inst->alg.halg.base.cra_ctxsize = sizeof(struct cryptd_hash_ctx);
inst->alg.halg.base.cra_init = cryptd_hash_init_tfm;
inst->alg.halg.base.cra_exit = cryptd_hash_exit_tfm;
inst->alg.init = cryptd_hash_init_enqueue;
inst->alg.update = cryptd_hash_update_enqueue;
inst->alg.final = cryptd_hash_final_enqueue;
inst->alg.finup = cryptd_hash_finup_enqueue;
inst->alg.export = cryptd_hash_export;
inst->alg.import = cryptd_hash_import;
if (crypto_shash_alg_has_setkey(alg))
inst->alg.setkey = cryptd_hash_setkey;
inst->alg.digest = cryptd_hash_digest_enqueue;
inst->free = cryptd_hash_free;
err = ahash_register_instance(tmpl, inst);
if (err) {
err_free_inst:
cryptd_hash_free(inst);
}
return err;
}
static int cryptd_aead_setkey(struct crypto_aead *parent,
const u8 *key, unsigned int keylen)
{
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(parent);
struct crypto_aead *child = ctx->child;
return crypto_aead_setkey(child, key, keylen);
}
static int cryptd_aead_setauthsize(struct crypto_aead *parent,
unsigned int authsize)
{
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(parent);
struct crypto_aead *child = ctx->child;
return crypto_aead_setauthsize(child, authsize);
}
static void cryptd_aead_crypt(struct aead_request *req,
struct crypto_aead *child, int err,
int (*crypt)(struct aead_request *req),
crypto_completion_t compl)
{
struct cryptd_aead_request_ctx *rctx;
struct aead_request *subreq;
struct cryptd_aead_ctx *ctx;
struct crypto_aead *tfm;
int refcnt;
rctx = aead_request_ctx(req);
subreq = &rctx->req;
req->base.complete = subreq->base.complete;
req->base.data = subreq->base.data;
tfm = crypto_aead_reqtfm(req);
if (unlikely(err == -EINPROGRESS))
goto out;
aead_request_set_tfm(subreq, child);
aead_request_set_callback(subreq, CRYPTO_TFM_REQ_MAY_SLEEP,
NULL, NULL);
aead_request_set_crypt(subreq, req->src, req->dst, req->cryptlen,
req->iv);
aead_request_set_ad(subreq, req->assoclen);
err = crypt(subreq);
out:
ctx = crypto_aead_ctx(tfm);
refcnt = refcount_read(&ctx->refcnt);
local_bh_disable();
aead_request_complete(req, err);
local_bh_enable();
if (err == -EINPROGRESS) {
subreq->base.complete = req->base.complete;
subreq->base.data = req->base.data;
req->base.complete = compl;
req->base.data = req;
} else if (refcnt && refcount_dec_and_test(&ctx->refcnt))
crypto_free_aead(tfm);
}
static void cryptd_aead_encrypt(void *data, int err)
{
struct aead_request *req = data;
struct cryptd_aead_ctx *ctx;
struct crypto_aead *child;
ctx = crypto_aead_ctx(crypto_aead_reqtfm(req));
child = ctx->child;
cryptd_aead_crypt(req, child, err, crypto_aead_alg(child)->encrypt,
cryptd_aead_encrypt);
}
static void cryptd_aead_decrypt(void *data, int err)
{
struct aead_request *req = data;
struct cryptd_aead_ctx *ctx;
struct crypto_aead *child;
ctx = crypto_aead_ctx(crypto_aead_reqtfm(req));
child = ctx->child;
cryptd_aead_crypt(req, child, err, crypto_aead_alg(child)->decrypt,
cryptd_aead_decrypt);
}
static int cryptd_aead_enqueue(struct aead_request *req,
crypto_completion_t compl)
{
struct cryptd_aead_request_ctx *rctx = aead_request_ctx(req);
struct crypto_aead *tfm = crypto_aead_reqtfm(req);
struct cryptd_queue *queue = cryptd_get_queue(crypto_aead_tfm(tfm));
struct aead_request *subreq = &rctx->req;
subreq->base.complete = req->base.complete;
subreq->base.data = req->base.data;
req->base.complete = compl;
req->base.data = req;
return cryptd_enqueue_request(queue, &req->base);
}
static int cryptd_aead_encrypt_enqueue(struct aead_request *req)
{
return cryptd_aead_enqueue(req, cryptd_aead_encrypt );
}
static int cryptd_aead_decrypt_enqueue(struct aead_request *req)
{
return cryptd_aead_enqueue(req, cryptd_aead_decrypt );
}
static int cryptd_aead_init_tfm(struct crypto_aead *tfm)
{
struct aead_instance *inst = aead_alg_instance(tfm);
struct aead_instance_ctx *ictx = aead_instance_ctx(inst);
struct crypto_aead_spawn *spawn = &ictx->aead_spawn;
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(tfm);
struct crypto_aead *cipher;
cipher = crypto_spawn_aead(spawn);
if (IS_ERR(cipher))
return PTR_ERR(cipher);
ctx->child = cipher;
crypto_aead_set_reqsize(
tfm, sizeof(struct cryptd_aead_request_ctx) +
crypto_aead_reqsize(cipher));
return 0;
}
static void cryptd_aead_exit_tfm(struct crypto_aead *tfm)
{
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(tfm);
crypto_free_aead(ctx->child);
}
static void cryptd_aead_free(struct aead_instance *inst)
{
struct aead_instance_ctx *ctx = aead_instance_ctx(inst);
crypto_drop_aead(&ctx->aead_spawn);
kfree(inst);
}
static int cryptd_create_aead(struct crypto_template *tmpl,
struct rtattr **tb,
struct crypto_attr_type *algt,
struct cryptd_queue *queue)
{
struct aead_instance_ctx *ctx;
struct aead_instance *inst;
struct aead_alg *alg;
u32 type;
u32 mask;
int err;
cryptd_type_and_mask(algt, &type, &mask);
inst = kzalloc(sizeof(*inst) + sizeof(*ctx), GFP_KERNEL);
if (!inst)
return -ENOMEM;
ctx = aead_instance_ctx(inst);
ctx->queue = queue;
err = crypto_grab_aead(&ctx->aead_spawn, aead_crypto_instance(inst),
crypto_attr_alg_name(tb[1]), type, mask);
if (err)
goto err_free_inst;
alg = crypto_spawn_aead_alg(&ctx->aead_spawn);
err = cryptd_init_instance(aead_crypto_instance(inst), &alg->base);
if (err)
goto err_free_inst;
inst->alg.base.cra_flags |= CRYPTO_ALG_ASYNC |
(alg->base.cra_flags & CRYPTO_ALG_INTERNAL);
inst->alg.base.cra_ctxsize = sizeof(struct cryptd_aead_ctx);
inst->alg.ivsize = crypto_aead_alg_ivsize(alg);
inst->alg.maxauthsize = crypto_aead_alg_maxauthsize(alg);
inst->alg.init = cryptd_aead_init_tfm;
inst->alg.exit = cryptd_aead_exit_tfm;
inst->alg.setkey = cryptd_aead_setkey;
inst->alg.setauthsize = cryptd_aead_setauthsize;
inst->alg.encrypt = cryptd_aead_encrypt_enqueue;
inst->alg.decrypt = cryptd_aead_decrypt_enqueue;
inst->free = cryptd_aead_free;
err = aead_register_instance(tmpl, inst);
if (err) {
err_free_inst:
cryptd_aead_free(inst);
}
return err;
}
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
static struct cryptd_queue queue;
static int cryptd_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_SKCIPHER:
return cryptd_create_skcipher(tmpl, tb, algt, &queue);
case CRYPTO_ALG_TYPE_HASH:
return cryptd_create_hash(tmpl, tb, algt, &queue);
case CRYPTO_ALG_TYPE_AEAD:
return cryptd_create_aead(tmpl, tb, algt, &queue);
}
return -EINVAL;
}
static struct crypto_template cryptd_tmpl = {
.name = "cryptd",
.create = cryptd_create,
.module = THIS_MODULE,
};
struct cryptd_skcipher *cryptd_alloc_skcipher(const char *alg_name,
u32 type, u32 mask)
{
char cryptd_alg_name[CRYPTO_MAX_ALG_NAME];
struct cryptd_skcipher_ctx *ctx;
struct crypto_skcipher *tfm;
if (snprintf(cryptd_alg_name, CRYPTO_MAX_ALG_NAME,
"cryptd(%s)", alg_name) >= CRYPTO_MAX_ALG_NAME)
return ERR_PTR(-EINVAL);
tfm = crypto_alloc_skcipher(cryptd_alg_name, type, mask);
if (IS_ERR(tfm))
return ERR_CAST(tfm);
if (tfm->base.__crt_alg->cra_module != THIS_MODULE) {
crypto_free_skcipher(tfm);
return ERR_PTR(-EINVAL);
}
ctx = crypto_skcipher_ctx(tfm);
refcount_set(&ctx->refcnt, 1);
return container_of(tfm, struct cryptd_skcipher, base);
}
EXPORT_SYMBOL_GPL(cryptd_alloc_skcipher);
struct crypto_skcipher *cryptd_skcipher_child(struct cryptd_skcipher *tfm)
{
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(&tfm->base);
return ctx->child;
}
EXPORT_SYMBOL_GPL(cryptd_skcipher_child);
bool cryptd_skcipher_queued(struct cryptd_skcipher *tfm)
{
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(&tfm->base);
return refcount_read(&ctx->refcnt) - 1;
}
EXPORT_SYMBOL_GPL(cryptd_skcipher_queued);
void cryptd_free_skcipher(struct cryptd_skcipher *tfm)
{
struct cryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(&tfm->base);
if (refcount_dec_and_test(&ctx->refcnt))
crypto_free_skcipher(&tfm->base);
}
EXPORT_SYMBOL_GPL(cryptd_free_skcipher);
struct cryptd_ahash *cryptd_alloc_ahash(const char *alg_name,
u32 type, u32 mask)
{
char cryptd_alg_name[CRYPTO_MAX_ALG_NAME];
struct cryptd_hash_ctx *ctx;
struct crypto_ahash *tfm;
if (snprintf(cryptd_alg_name, CRYPTO_MAX_ALG_NAME,
"cryptd(%s)", alg_name) >= CRYPTO_MAX_ALG_NAME)
return ERR_PTR(-EINVAL);
tfm = crypto_alloc_ahash(cryptd_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);
}
ctx = crypto_ahash_ctx(tfm);
refcount_set(&ctx->refcnt, 1);
return __cryptd_ahash_cast(tfm);
}
EXPORT_SYMBOL_GPL(cryptd_alloc_ahash);
struct crypto_shash *cryptd_ahash_child(struct cryptd_ahash *tfm)
{
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(&tfm->base);
return ctx->child;
}
EXPORT_SYMBOL_GPL(cryptd_ahash_child);
struct shash_desc *cryptd_shash_desc(struct ahash_request *req)
{
struct cryptd_hash_request_ctx *rctx = ahash_request_ctx(req);
return &rctx->desc;
}
EXPORT_SYMBOL_GPL(cryptd_shash_desc);
bool cryptd_ahash_queued(struct cryptd_ahash *tfm)
{
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(&tfm->base);
return refcount_read(&ctx->refcnt) - 1;
}
EXPORT_SYMBOL_GPL(cryptd_ahash_queued);
void cryptd_free_ahash(struct cryptd_ahash *tfm)
{
struct cryptd_hash_ctx *ctx = crypto_ahash_ctx(&tfm->base);
if (refcount_dec_and_test(&ctx->refcnt))
crypto_free_ahash(&tfm->base);
}
EXPORT_SYMBOL_GPL(cryptd_free_ahash);
struct cryptd_aead *cryptd_alloc_aead(const char *alg_name,
u32 type, u32 mask)
{
char cryptd_alg_name[CRYPTO_MAX_ALG_NAME];
struct cryptd_aead_ctx *ctx;
struct crypto_aead *tfm;
if (snprintf(cryptd_alg_name, CRYPTO_MAX_ALG_NAME,
"cryptd(%s)", alg_name) >= CRYPTO_MAX_ALG_NAME)
return ERR_PTR(-EINVAL);
tfm = crypto_alloc_aead(cryptd_alg_name, type, mask);
if (IS_ERR(tfm))
return ERR_CAST(tfm);
if (tfm->base.__crt_alg->cra_module != THIS_MODULE) {
crypto_free_aead(tfm);
return ERR_PTR(-EINVAL);
}
ctx = crypto_aead_ctx(tfm);
refcount_set(&ctx->refcnt, 1);
return __cryptd_aead_cast(tfm);
}
EXPORT_SYMBOL_GPL(cryptd_alloc_aead);
struct crypto_aead *cryptd_aead_child(struct cryptd_aead *tfm)
{
struct cryptd_aead_ctx *ctx;
ctx = crypto_aead_ctx(&tfm->base);
return ctx->child;
}
EXPORT_SYMBOL_GPL(cryptd_aead_child);
bool cryptd_aead_queued(struct cryptd_aead *tfm)
{
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(&tfm->base);
return refcount_read(&ctx->refcnt) - 1;
}
EXPORT_SYMBOL_GPL(cryptd_aead_queued);
void cryptd_free_aead(struct cryptd_aead *tfm)
{
struct cryptd_aead_ctx *ctx = crypto_aead_ctx(&tfm->base);
if (refcount_dec_and_test(&ctx->refcnt))
crypto_free_aead(&tfm->base);
}
EXPORT_SYMBOL_GPL(cryptd_free_aead);
static int __init cryptd_init(void)
{
int err;
cryptd_wq = alloc_workqueue("cryptd", WQ_MEM_RECLAIM | WQ_CPU_INTENSIVE,
1);
if (!cryptd_wq)
return -ENOMEM;
err = cryptd_init_queue(&queue, cryptd_max_cpu_qlen);
if (err)
goto err_destroy_wq;
err = crypto_register_template(&cryptd_tmpl);
if (err)
goto err_fini_queue;
return 0;
err_fini_queue:
cryptd_fini_queue(&queue);
err_destroy_wq:
destroy_workqueue(cryptd_wq);
return err;
}
static void __exit cryptd_exit(void)
{
destroy_workqueue(cryptd_wq);
crypto: cryptd - Per-CPU thread implementation based on kcrypto_wq Original cryptd thread implementation has scalability issue, this patch solve the issue with a per-CPU thread implementation. struct cryptd_queue is defined to be a per-CPU queue, which holds one struct cryptd_cpu_queue for each CPU. In struct cryptd_cpu_queue, a struct crypto_queue holds all requests for the CPU, a struct work_struct is used to run all requests for the CPU. Testing based on dm-crypt on an Intel Core 2 E6400 (two cores) machine shows 19.2% performance gain. The testing script is as follow: -------------------- script begin --------------------------- #!/bin/sh dmc_create() { # Create a crypt device using dmsetup dmsetup create $2 --table "0 `blockdev --getsize $1` crypt cbc(aes-asm)?cryptd?plain:plain babebabebabebabebabebabebabebabe 0 $1 0" } dmsetup remove crypt0 dmsetup remove crypt1 dd if=/dev/zero of=/dev/ram0 bs=1M count=4 >& /dev/null dd if=/dev/zero of=/dev/ram1 bs=1M count=4 >& /dev/null dmc_create /dev/ram0 crypt0 dmc_create /dev/ram1 crypt1 cat >tr.sh <<EOF #!/bin/sh for n in \$(seq 10); do dd if=/dev/dm-0 of=/dev/null >& /dev/null & dd if=/dev/dm-1 of=/dev/null >& /dev/null & done wait EOF for n in $(seq 10); do /usr/bin/time sh tr.sh done rm tr.sh -------------------- script end --------------------------- The separator of dm-crypt parameter is changed from "-" to "?", because "-" is used in some cipher driver name too, and cryptds need to specify cipher driver name instead of cipher name. The test result on an Intel Core2 E6400 (two cores) is as follow: without patch: -----------------wo begin -------------------------- 0.04user 0.38system 0:00.39elapsed 107%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6566minor)pagefaults 0swaps 0.07user 0.35system 0:00.35elapsed 121%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6567minor)pagefaults 0swaps 0.06user 0.34system 0:00.30elapsed 135%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.36elapsed 119%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6607minor)pagefaults 0swaps 0.06user 0.36system 0:00.35elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.37system 0:00.31elapsed 136%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.34system 0:00.30elapsed 126%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6597minor)pagefaults 0swaps 0.06user 0.32system 0:00.31elapsed 125%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6571minor)pagefaults 0swaps 0.06user 0.34system 0:00.31elapsed 134%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6581minor)pagefaults 0swaps 0.05user 0.38system 0:00.31elapsed 138%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6600minor)pagefaults 0swaps -----------------wo end -------------------------- with patch: ------------------w begin -------------------------- 0.02user 0.31system 0:00.24elapsed 141%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6554minor)pagefaults 0swaps 0.05user 0.34system 0:00.31elapsed 127%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6606minor)pagefaults 0swaps 0.07user 0.33system 0:00.26elapsed 155%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6559minor)pagefaults 0swaps 0.07user 0.32system 0:00.26elapsed 151%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.05user 0.34system 0:00.26elapsed 150%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6603minor)pagefaults 0swaps 0.03user 0.36system 0:00.31elapsed 124%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 147%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6586minor)pagefaults 0swaps 0.03user 0.37system 0:00.27elapsed 146%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6562minor)pagefaults 0swaps 0.04user 0.36system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6594minor)pagefaults 0swaps 0.04user 0.35system 0:00.26elapsed 154%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+6557minor)pagefaults 0swaps ------------------w end -------------------------- The middle value of elapsed time is: wo cryptwq: 0.31 w cryptwq: 0.26 The performance gain is about (0.31-0.26)/0.26 = 0.192. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-02-19 14:42:19 +08:00
cryptd_fini_queue(&queue);
crypto_unregister_template(&cryptd_tmpl);
}
subsys_initcall(cryptd_init);
module_exit(cryptd_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Software async crypto daemon");
MODULE_ALIAS_CRYPTO("cryptd");