mirror_zfs/module/zfs/zfs_chksum.c
Tino Reichardt 68aa3379ec
Skip checksum benchmarks on systems with slow cpu
The checksum benchmarking on module load may take a really long time
on embedded systems with a slow cpu. Avoid all benchmarks >= 1MiB on
systems, where EdonR is slower then 300 MiB/s.

This limit is currently hardcoded via the define LIMIT_PERF_MBS.

This is the new benchmark output of a slow Intel Atom:

```
 implementation    1k    4k   16k   64k  256k    1m    4m   16m
 edonr-generic    209   257   268   259   262     0     0     0
 skein-generic    129   150   151   150   150     0     0     0
 sha256-generic    50    55    56    56    56     0     0     0
 sha512-generic    76    86    88    89    88     0     0     0
 blake3-generic    63    62    62    62    61     0     0     0
 blake3-sse2      114   292   301   307   309     0     0     0
```

Reviewed-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tino Reichardt <milky-zfs@mcmilk.de>
Closes #13695
2022-08-01 09:51:45 -07:00

356 lines
9.3 KiB
C

/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or https://opensource.org/licenses/CDDL-1.0.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2021-2022 Tino Reichardt <milky-zfs@mcmilk.de>
*/
#include <sys/types.h>
#include <sys/spa.h>
#include <sys/zio_checksum.h>
#include <sys/zfs_context.h>
#include <sys/zfs_chksum.h>
#include <sys/blake3.h>
/* limit benchmarking to max 256KiB, when EdonR is slower then this: */
#define LIMIT_PERF_MBS 300
typedef struct {
const char *name;
const char *impl;
uint64_t bs1k;
uint64_t bs4k;
uint64_t bs16k;
uint64_t bs64k;
uint64_t bs256k;
uint64_t bs1m;
uint64_t bs4m;
uint64_t bs16m;
zio_cksum_salt_t salt;
zio_checksum_t *(func);
zio_checksum_tmpl_init_t *(init);
zio_checksum_tmpl_free_t *(free);
} chksum_stat_t;
static chksum_stat_t *chksum_stat_data = 0;
static int chksum_stat_cnt = 0;
static kstat_t *chksum_kstat = NULL;
/*
* i3-1005G1 test output:
*
* implementation 1k 4k 16k 64k 256k 1m 4m
* fletcher-4 5421 15001 26468 32555 34720 32801 18847
* edonr-generic 1196 1602 1761 1749 1762 1759 1751
* skein-generic 546 591 608 615 619 612 616
* sha256-generic 246 270 274 274 277 275 276
* sha256-avx 262 296 304 307 307 307 306
* sha256-sha-ni 769 1072 1172 1220 1219 1232 1228
* sha256-openssl 240 300 316 314 304 285 276
* sha512-generic 333 374 385 392 391 393 392
* sha512-openssl 353 441 467 476 472 467 426
* sha512-avx 362 444 473 475 479 476 478
* sha512-avx2 394 500 530 538 543 545 542
* blake3-generic 308 313 313 313 312 313 312
* blake3-sse2 402 1289 1423 1446 1432 1458 1413
* blake3-sse41 427 1470 1625 1704 1679 1607 1629
* blake3-avx2 428 1920 3095 3343 3356 3318 3204
* blake3-avx512 473 2687 4905 5836 5844 5643 5374
*/
static int
chksum_kstat_headers(char *buf, size_t size)
{
ssize_t off = 0;
off += snprintf(buf + off, size, "%-23s", "implementation");
off += snprintf(buf + off, size - off, "%8s", "1k");
off += snprintf(buf + off, size - off, "%8s", "4k");
off += snprintf(buf + off, size - off, "%8s", "16k");
off += snprintf(buf + off, size - off, "%8s", "64k");
off += snprintf(buf + off, size - off, "%8s", "256k");
off += snprintf(buf + off, size - off, "%8s", "1m");
off += snprintf(buf + off, size - off, "%8s", "4m");
(void) snprintf(buf + off, size - off, "%8s\n", "16m");
return (0);
}
static int
chksum_kstat_data(char *buf, size_t size, void *data)
{
chksum_stat_t *cs;
ssize_t off = 0;
char b[24];
cs = (chksum_stat_t *)data;
snprintf(b, 23, "%s-%s", cs->name, cs->impl);
off += snprintf(buf + off, size - off, "%-23s", b);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs1k);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs4k);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs16k);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs64k);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs256k);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs1m);
off += snprintf(buf + off, size - off, "%8llu",
(u_longlong_t)cs->bs4m);
(void) snprintf(buf + off, size - off, "%8llu\n",
(u_longlong_t)cs->bs16m);
return (0);
}
static void *
chksum_kstat_addr(kstat_t *ksp, loff_t n)
{
if (n < chksum_stat_cnt)
ksp->ks_private = (void *)(chksum_stat_data + n);
else
ksp->ks_private = NULL;
return (ksp->ks_private);
}
static void
chksum_run(chksum_stat_t *cs, abd_t *abd, void *ctx, int round,
uint64_t *result)
{
hrtime_t start;
uint64_t run_bw, run_time_ns, run_count = 0, size = 0;
uint32_t l, loops = 0;
zio_cksum_t zcp;
switch (round) {
case 1: /* 1k */
size = 1<<10; loops = 128; break;
case 2: /* 2k */
size = 1<<12; loops = 64; break;
case 3: /* 4k */
size = 1<<14; loops = 32; break;
case 4: /* 16k */
size = 1<<16; loops = 16; break;
case 5: /* 256k */
size = 1<<18; loops = 8; break;
case 6: /* 1m */
size = 1<<20; loops = 4; break;
case 7: /* 4m */
size = 1<<22; loops = 1; break;
case 8: /* 16m */
size = 1<<24; loops = 1; break;
}
kpreempt_disable();
start = gethrtime();
do {
for (l = 0; l < loops; l++, run_count++)
cs->func(abd, size, ctx, &zcp);
run_time_ns = gethrtime() - start;
} while (run_time_ns < MSEC2NSEC(1));
kpreempt_enable();
run_bw = size * run_count * NANOSEC;
run_bw /= run_time_ns; /* B/s */
*result = run_bw/1024/1024; /* MiB/s */
}
#define LIMIT_INIT 0
#define LIMIT_NEEDED 1
#define LIMIT_NOLIMIT 2
static void
chksum_benchit(chksum_stat_t *cs)
{
abd_t *abd;
void *ctx = 0;
void *salt = &cs->salt.zcs_bytes;
static int chksum_stat_limit = LIMIT_INIT;
memset(salt, 0, sizeof (cs->salt.zcs_bytes));
if (cs->init)
ctx = cs->init(&cs->salt);
/* allocate test memory via abd linear interface */
abd = abd_alloc_linear(1<<20, B_FALSE);
chksum_run(cs, abd, ctx, 1, &cs->bs1k);
chksum_run(cs, abd, ctx, 2, &cs->bs4k);
chksum_run(cs, abd, ctx, 3, &cs->bs16k);
chksum_run(cs, abd, ctx, 4, &cs->bs64k);
chksum_run(cs, abd, ctx, 5, &cs->bs256k);
/* check if we ran on a slow cpu */
if (chksum_stat_limit == LIMIT_INIT) {
if (cs->bs1k < LIMIT_PERF_MBS) {
chksum_stat_limit = LIMIT_NEEDED;
} else {
chksum_stat_limit = LIMIT_NOLIMIT;
}
}
/* skip benchmarks >= 1MiB when the CPU is to slow */
if (chksum_stat_limit == LIMIT_NEEDED)
goto abort;
chksum_run(cs, abd, ctx, 6, &cs->bs1m);
abd_free(abd);
/* allocate test memory via abd non linear interface */
abd = abd_alloc(1<<24, B_FALSE);
chksum_run(cs, abd, ctx, 7, &cs->bs4m);
chksum_run(cs, abd, ctx, 8, &cs->bs16m);
abort:
abd_free(abd);
/* free up temp memory */
if (cs->free)
cs->free(ctx);
}
/*
* Initialize and benchmark all supported implementations.
*/
static void
chksum_benchmark(void)
{
#ifndef _KERNEL
/* we need the benchmark only for the kernel module */
return;
#endif
chksum_stat_t *cs;
int cbid = 0, id;
uint64_t max = 0;
/* space for the benchmark times */
chksum_stat_cnt = 4;
chksum_stat_cnt += blake3_get_impl_count();
chksum_stat_data = (chksum_stat_t *)kmem_zalloc(
sizeof (chksum_stat_t) * chksum_stat_cnt, KM_SLEEP);
/* edonr - needs to be the first one here (slow CPU check) */
cs = &chksum_stat_data[cbid++];
cs->init = abd_checksum_edonr_tmpl_init;
cs->func = abd_checksum_edonr_native;
cs->free = abd_checksum_edonr_tmpl_free;
cs->name = "edonr";
cs->impl = "generic";
chksum_benchit(cs);
/* skein */
cs = &chksum_stat_data[cbid++];
cs->init = abd_checksum_skein_tmpl_init;
cs->func = abd_checksum_skein_native;
cs->free = abd_checksum_skein_tmpl_free;
cs->name = "skein";
cs->impl = "generic";
chksum_benchit(cs);
/* sha256 */
cs = &chksum_stat_data[cbid++];
cs->init = 0;
cs->func = abd_checksum_SHA256;
cs->free = 0;
cs->name = "sha256";
cs->impl = "generic";
chksum_benchit(cs);
/* sha512 */
cs = &chksum_stat_data[cbid++];
cs->init = 0;
cs->func = abd_checksum_SHA512_native;
cs->free = 0;
cs->name = "sha512";
cs->impl = "generic";
chksum_benchit(cs);
/* blake3 */
for (id = 0; id < blake3_get_impl_count(); id++) {
blake3_set_impl_id(id);
cs = &chksum_stat_data[cbid++];
cs->init = abd_checksum_blake3_tmpl_init;
cs->func = abd_checksum_blake3_native;
cs->free = abd_checksum_blake3_tmpl_free;
cs->name = "blake3";
cs->impl = blake3_get_impl_name();
chksum_benchit(cs);
if (cs->bs256k > max) {
max = cs->bs256k;
blake3_set_impl_fastest(id);
}
}
}
void
chksum_init(void)
{
#ifdef _KERNEL
blake3_per_cpu_ctx_init();
#endif
/* Benchmark supported implementations */
chksum_benchmark();
/* Install kstats for all implementations */
chksum_kstat = kstat_create("zfs", 0, "chksum_bench", "misc",
KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VIRTUAL);
if (chksum_kstat != NULL) {
chksum_kstat->ks_data = NULL;
chksum_kstat->ks_ndata = UINT32_MAX;
kstat_set_raw_ops(chksum_kstat,
chksum_kstat_headers,
chksum_kstat_data,
chksum_kstat_addr);
kstat_install(chksum_kstat);
}
/* setup implementations */
blake3_setup_impl();
}
void
chksum_fini(void)
{
if (chksum_kstat != NULL) {
kstat_delete(chksum_kstat);
chksum_kstat = NULL;
}
if (chksum_stat_cnt) {
kmem_free(chksum_stat_data,
sizeof (chksum_stat_t) * chksum_stat_cnt);
chksum_stat_cnt = 0;
chksum_stat_data = 0;
}
#ifdef _KERNEL
blake3_per_cpu_ctx_fini();
#endif
}