static inline void
run_gen_bench_impl(const char *impl)
{
int fn, ncols;
uint64_t ds, iter_cnt, iter, disksize;
struct rusage start, stop;
double elapsed, d_bw;
/* Benchmark generate functions */
for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
/* create suitable raidz_map */
ncols = rto_opts.rto_dcols + fn + 1;
zio_bench.io_size = 1ULL << ds;
rm_bench = vdev_raidz_map_alloc(&zio_bench,
BENCH_ASHIFT, ncols, fn+1);
/* estimate iteration count */
iter_cnt = GEN_BENCH_MEMORY;
iter_cnt /= zio_bench.io_size;
getrusage(RUSAGE_THREAD, &start);
for (iter = 0; iter < iter_cnt; iter++)
vdev_raidz_generate_parity(rm_bench);
getrusage(RUSAGE_THREAD, &stop);
elapsed = get_time_diff(&start, &stop);
disksize = (1ULL << ds) / rto_opts.rto_dcols;
d_bw = (double)iter_cnt * (double)disksize;
d_bw /= (1024.0 * 1024.0 * elapsed);
LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
impl,
raidz_gen_name[fn],
rto_opts.rto_dcols,
(1ULL<<ds),
d_bw,
d_bw * (double)(ncols),
(unsigned) iter_cnt);
vdev_raidz_map_free(rm_bench);
}
}
}
void
vdev_raidz_math_init(void)
{
raidz_impl_ops_t *curr_impl;
zio_t *bench_zio = NULL;
raidz_map_t *bench_rm = NULL;
uint64_t bench_parity;
int i, c, fn;
/* move supported impl into raidz_supp_impl */
for (i = 0, c = 0; i < ARRAY_SIZE(raidz_all_maths); i++) {
curr_impl = (raidz_impl_ops_t *)raidz_all_maths[i];
/* initialize impl */
if (curr_impl->init)
curr_impl->init();
if (curr_impl->is_supported())
raidz_supp_impl[c++] = (raidz_impl_ops_t *)curr_impl;
}
membar_producer(); /* complete raidz_supp_impl[] init */
raidz_supp_impl_cnt = c; /* number of supported impl */
#if !defined(_KERNEL)
/* Skip benchmarking and use last implementation as fastest */
memcpy(&vdev_raidz_fastest_impl, raidz_supp_impl[raidz_supp_impl_cnt-1],
sizeof (vdev_raidz_fastest_impl));
strcpy(vdev_raidz_fastest_impl.name, "fastest");
raidz_math_initialized = B_TRUE;
/* Use 'cycle' math selection method for userspace */
VERIFY0(vdev_raidz_impl_set("cycle"));
return;
#endif
/* Fake an zio and run the benchmark on a warmed up buffer */
bench_zio = kmem_zalloc(sizeof (zio_t), KM_SLEEP);
bench_zio->io_offset = 0;
bench_zio->io_size = BENCH_ZIO_SIZE; /* only data columns */
bench_zio->io_abd = abd_alloc_linear(BENCH_ZIO_SIZE, B_TRUE);
memset(abd_to_buf(bench_zio->io_abd), 0xAA, BENCH_ZIO_SIZE);
/* Benchmark parity generation methods */
for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
bench_parity = fn + 1;
/* New raidz_map is needed for each generate_p/q/r */
bench_rm = vdev_raidz_map_alloc(bench_zio, SPA_MINBLOCKSHIFT,
BENCH_D_COLS + bench_parity, bench_parity);
benchmark_raidz_impl(bench_rm, fn, benchmark_gen_impl);
vdev_raidz_map_free(bench_rm);
}
/* Benchmark data reconstruction methods */
bench_rm = vdev_raidz_map_alloc(bench_zio, SPA_MINBLOCKSHIFT,
BENCH_COLS, PARITY_PQR);
for (fn = 0; fn < RAIDZ_REC_NUM; fn++)
benchmark_raidz_impl(bench_rm, fn, benchmark_rec_impl);
vdev_raidz_map_free(bench_rm);
/* cleanup the bench zio */
abd_free(bench_zio->io_abd);
kmem_free(bench_zio, sizeof (zio_t));
/* install kstats for all impl */
raidz_math_kstat = kstat_create("zfs", 0, "vdev_raidz_bench", "misc",
KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VIRTUAL);
if (raidz_math_kstat != NULL) {
raidz_math_kstat->ks_data = NULL;
raidz_math_kstat->ks_ndata = UINT32_MAX;
kstat_set_raw_ops(raidz_math_kstat,
raidz_math_kstat_headers,
raidz_math_kstat_data,
raidz_math_kstat_addr);
kstat_install(raidz_math_kstat);
}
/* Finish initialization */
atomic_swap_32(&zfs_vdev_raidz_impl, user_sel_impl);
raidz_math_initialized = B_TRUE;
}
static void
run_rec_bench_impl(const char *impl)
{
int fn, ncols, nbad;
uint64_t ds, iter_cnt, iter, disksize;
hrtime_t start;
double elapsed, d_bw;
static const int tgt[7][3] = {
{1, 2, 3}, /* rec_p: bad QR & D[0] */
{0, 2, 3}, /* rec_q: bad PR & D[0] */
{0, 1, 3}, /* rec_r: bad PQ & D[0] */
{2, 3, 4}, /* rec_pq: bad R & D[0][1] */
{1, 3, 4}, /* rec_pr: bad Q & D[0][1] */
{0, 3, 4}, /* rec_qr: bad P & D[0][1] */
{3, 4, 5} /* rec_pqr: bad & D[0][1][2] */
};
for (fn = 0; fn < RAIDZ_REC_NUM; fn++) {
for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
/* create suitable raidz_map */
ncols = rto_opts.rto_dcols + PARITY_PQR;
zio_bench.io_size = 1ULL << ds;
/*
* raidz block is too short to test
* the requested method
*/
if (zio_bench.io_size / rto_opts.rto_dcols <
(1ULL << BENCH_ASHIFT))
continue;
rm_bench = vdev_raidz_map_alloc(&zio_bench,
BENCH_ASHIFT, ncols, PARITY_PQR);
/* estimate iteration count */
iter_cnt = (REC_BENCH_MEMORY);
iter_cnt /= zio_bench.io_size;
/* calculate how many bad columns there are */
nbad = MIN(3, raidz_ncols(rm_bench) -
raidz_parity(rm_bench));
start = gethrtime();
for (iter = 0; iter < iter_cnt; iter++)
vdev_raidz_reconstruct(rm_bench, tgt[fn], nbad);
elapsed = NSEC2SEC((double) (gethrtime() - start));
disksize = (1ULL << ds) / rto_opts.rto_dcols;
d_bw = (double)iter_cnt * (double)(disksize);
d_bw /= (1024.0 * 1024.0 * elapsed);
LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
impl,
raidz_rec_name[fn],
rto_opts.rto_dcols,
(1ULL<<ds),
d_bw,
d_bw * (double)ncols,
(unsigned) iter_cnt);
vdev_raidz_map_free(rm_bench);
}
}
}
void
vdev_raidz_math_init(void)
{
raidz_impl_ops_t *curr_impl;
zio_t *bench_zio = NULL;
raidz_map_t *bench_rm = NULL;
uint64_t bench_parity;
int i, c, fn;
/* init & vdev_raidz_impl_lock */
rw_init(&vdev_raidz_impl_lock, NULL, RW_DEFAULT, NULL);
/* move supported impl into raidz_supp_impl */
for (i = 0, c = 0; i < ARRAY_SIZE(raidz_all_maths); i++) {
curr_impl = (raidz_impl_ops_t *) raidz_all_maths[i];
/* initialize impl */
if (curr_impl->init)
curr_impl->init();
if (curr_impl->is_supported()) {
/* init kstat */
init_raidz_kstat(&raidz_impl_kstats[c],
curr_impl->name);
raidz_supp_impl[c++] = (raidz_impl_ops_t *) curr_impl;
}
}
raidz_supp_impl_cnt = c; /* number of supported impl */
raidz_supp_impl[c] = NULL; /* sentinel */
/* init kstat for original routines */
init_raidz_kstat(&(raidz_impl_kstats[raidz_supp_impl_cnt]), "original");
#if !defined(_KERNEL)
/*
* Skip benchmarking and use last implementation as fastest
*/
memcpy(&vdev_raidz_fastest_impl, raidz_supp_impl[raidz_supp_impl_cnt-1],
sizeof (vdev_raidz_fastest_impl));
vdev_raidz_fastest_impl.name = "fastest";
raidz_math_initialized = B_TRUE;
/* Use 'cycle' math selection method for userspace */
VERIFY0(vdev_raidz_impl_set("cycle"));
return;
#endif
/* Fake an zio and run the benchmark on it */
bench_zio = kmem_zalloc(sizeof (zio_t), KM_SLEEP);
bench_zio->io_offset = 0;
bench_zio->io_size = BENCH_ZIO_SIZE; /* only data columns */
bench_zio->io_data = zio_data_buf_alloc(BENCH_ZIO_SIZE);
VERIFY(bench_zio->io_data);
/* Benchmark parity generation methods */
for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
bench_parity = fn + 1;
/* New raidz_map is needed for each generate_p/q/r */
bench_rm = vdev_raidz_map_alloc(bench_zio, 9,
BENCH_D_COLS + bench_parity, bench_parity);
benchmark_raidz_impl(bench_rm, fn, benchmark_gen_impl);
vdev_raidz_map_free(bench_rm);
}
/* Benchmark data reconstruction methods */
bench_rm = vdev_raidz_map_alloc(bench_zio, 9, BENCH_COLS, PARITY_PQR);
for (fn = 0; fn < RAIDZ_REC_NUM; fn++)
benchmark_raidz_impl(bench_rm, fn, benchmark_rec_impl);
vdev_raidz_map_free(bench_rm);
/* cleanup the bench zio */
zio_data_buf_free(bench_zio->io_data, BENCH_ZIO_SIZE);
kmem_free(bench_zio, sizeof (zio_t));
/* install kstats for all impl */
raidz_math_kstat = kstat_create("zfs", 0, "vdev_raidz_bench",
"misc", KSTAT_TYPE_NAMED,
sizeof (raidz_impl_kstat_t) / sizeof (kstat_named_t) *
(raidz_supp_impl_cnt + 1), KSTAT_FLAG_VIRTUAL);
if (raidz_math_kstat != NULL) {
raidz_math_kstat->ks_data = raidz_impl_kstats;
kstat_install(raidz_math_kstat);
}
/* Finish initialization */
raidz_math_initialized = B_TRUE;
if (!vdev_raidz_impl_user_set)
VERIFY0(vdev_raidz_impl_set("fastest"));
}
