linux/tools/perf/bench/epoll-ctl.c
Athira Rajeev 299687e18a perf bench: Fix epoll bench to correct usage of affinity for machines with #CPUs > 1K
The 'perf bench epoll' testcase fails on systems with more than 1K CPUs.

Testcase: perf bench epoll all

Result snippet:
<<>>
Run summary [PID 106497]: 1399 threads monitoring on 64 file-descriptors for 8 secs.

perf: pthread_create: No such file or directory
<<>>

In epoll benchmarks (ctl, wait) pthread_create is invoked in do_threads
from respective bench_epoll_*  function. Though the logs shows direct
failure from pthread_create, the actual failure is from
"sched_setaffinity" returning EINVAL (invalid argument).

This happens because the default mask size in glibc is 1024. To overcome
this 1024 CPUs mask size limitation of cpu_set_t, change the mask size
using the CPU_*_S macros.

Patch addresses this by fixing all the epoll benchmarks to use CPU_ALLOC
to allocate cpumask, CPU_ALLOC_SIZE for size, and CPU_SET_S to set the
mask.

Reported-by: Disha Goel <disgoel@linux.vnet.ibm.com>
Signed-off-by: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Tested-by: Disha Goel <disgoel@linux.vnet.ibm.com>
Acked-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kajol Jain <kjain@linux.ibm.com>
Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nageswara R Sastry <rnsastry@linux.ibm.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: linuxppc-dev@lists.ozlabs.org
Link: https://lore.kernel.org/r/20220406175113.87881-3-atrajeev@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2022-04-09 12:34:29 -03:00

428 lines
9.9 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (C) 2018 Davidlohr Bueso.
*
* Benchmark the various operations allowed for epoll_ctl(2).
* The idea is to concurrently stress a single epoll instance
*/
#ifdef HAVE_EVENTFD_SUPPORT
/* For the CLR_() macros */
#include <string.h>
#include <pthread.h>
#include <errno.h>
#include <inttypes.h>
#include <signal.h>
#include <stdlib.h>
#include <unistd.h>
#include <linux/compiler.h>
#include <linux/kernel.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/epoll.h>
#include <sys/eventfd.h>
#include <perf/cpumap.h>
#include "../util/stat.h"
#include <subcmd/parse-options.h>
#include "bench.h"
#include <err.h>
#define printinfo(fmt, arg...) \
do { if (__verbose) printf(fmt, ## arg); } while (0)
static unsigned int nthreads = 0;
static unsigned int nsecs = 8;
static bool done, __verbose, randomize;
/*
* epoll related shared variables.
*/
/* Maximum number of nesting allowed inside epoll sets */
#define EPOLL_MAXNESTS 4
enum {
OP_EPOLL_ADD,
OP_EPOLL_MOD,
OP_EPOLL_DEL,
EPOLL_NR_OPS,
};
static int epollfd;
static int *epollfdp;
static bool noaffinity;
static unsigned int nested = 0;
/* amount of fds to monitor, per thread */
static unsigned int nfds = 64;
static pthread_mutex_t thread_lock;
static unsigned int threads_starting;
static struct stats all_stats[EPOLL_NR_OPS];
static pthread_cond_t thread_parent, thread_worker;
struct worker {
int tid;
pthread_t thread;
unsigned long ops[EPOLL_NR_OPS];
int *fdmap;
};
static const struct option options[] = {
OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
OPT_UINTEGER('r', "runtime", &nsecs, "Specify runtime (in seconds)"),
OPT_UINTEGER('f', "nfds", &nfds, "Specify amount of file descriptors to monitor for each thread"),
OPT_BOOLEAN( 'n', "noaffinity", &noaffinity, "Disables CPU affinity"),
OPT_UINTEGER( 'N', "nested", &nested, "Nesting level epoll hierarchy (default is 0, no nesting)"),
OPT_BOOLEAN( 'R', "randomize", &randomize, "Perform random operations on random fds"),
OPT_BOOLEAN( 'v', "verbose", &__verbose, "Verbose mode"),
OPT_END()
};
static const char * const bench_epoll_ctl_usage[] = {
"perf bench epoll ctl <options>",
NULL
};
static void toggle_done(int sig __maybe_unused,
siginfo_t *info __maybe_unused,
void *uc __maybe_unused)
{
/* inform all threads that we're done for the day */
done = true;
gettimeofday(&bench__end, NULL);
timersub(&bench__end, &bench__start, &bench__runtime);
}
static void nest_epollfd(void)
{
unsigned int i;
struct epoll_event ev;
if (nested > EPOLL_MAXNESTS)
nested = EPOLL_MAXNESTS;
printinfo("Nesting level(s): %d\n", nested);
epollfdp = calloc(nested, sizeof(int));
if (!epollfdp)
err(EXIT_FAILURE, "calloc");
for (i = 0; i < nested; i++) {
epollfdp[i] = epoll_create(1);
if (epollfd < 0)
err(EXIT_FAILURE, "epoll_create");
}
ev.events = EPOLLHUP; /* anything */
ev.data.u64 = i; /* any number */
for (i = nested - 1; i; i--) {
if (epoll_ctl(epollfdp[i - 1], EPOLL_CTL_ADD,
epollfdp[i], &ev) < 0)
err(EXIT_FAILURE, "epoll_ctl");
}
if (epoll_ctl(epollfd, EPOLL_CTL_ADD, *epollfdp, &ev) < 0)
err(EXIT_FAILURE, "epoll_ctl");
}
static inline void do_epoll_op(struct worker *w, int op, int fd)
{
int error;
struct epoll_event ev;
ev.events = EPOLLIN;
ev.data.u64 = fd;
switch (op) {
case OP_EPOLL_ADD:
error = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev);
break;
case OP_EPOLL_MOD:
ev.events = EPOLLOUT;
error = epoll_ctl(epollfd, EPOLL_CTL_MOD, fd, &ev);
break;
case OP_EPOLL_DEL:
error = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL);
break;
default:
error = 1;
break;
}
if (!error)
w->ops[op]++;
}
static inline void do_random_epoll_op(struct worker *w)
{
unsigned long rnd1 = random(), rnd2 = random();
int op, fd;
fd = w->fdmap[rnd1 % nfds];
op = rnd2 % EPOLL_NR_OPS;
do_epoll_op(w, op, fd);
}
static void *workerfn(void *arg)
{
unsigned int i;
struct worker *w = (struct worker *) arg;
struct timespec ts = { .tv_sec = 0,
.tv_nsec = 250 };
pthread_mutex_lock(&thread_lock);
threads_starting--;
if (!threads_starting)
pthread_cond_signal(&thread_parent);
pthread_cond_wait(&thread_worker, &thread_lock);
pthread_mutex_unlock(&thread_lock);
/* Let 'em loose */
do {
/* random */
if (randomize) {
do_random_epoll_op(w);
} else {
for (i = 0; i < nfds; i++) {
do_epoll_op(w, OP_EPOLL_ADD, w->fdmap[i]);
do_epoll_op(w, OP_EPOLL_MOD, w->fdmap[i]);
do_epoll_op(w, OP_EPOLL_DEL, w->fdmap[i]);
}
}
nanosleep(&ts, NULL);
} while (!done);
return NULL;
}
static void init_fdmaps(struct worker *w, int pct)
{
unsigned int i;
int inc;
struct epoll_event ev;
if (!pct)
return;
inc = 100/pct;
for (i = 0; i < nfds; i+=inc) {
ev.data.fd = w->fdmap[i];
ev.events = EPOLLIN;
if (epoll_ctl(epollfd, EPOLL_CTL_ADD, w->fdmap[i], &ev) < 0)
err(EXIT_FAILURE, "epoll_ct");
}
}
static int do_threads(struct worker *worker, struct perf_cpu_map *cpu)
{
pthread_attr_t thread_attr, *attrp = NULL;
cpu_set_t *cpuset;
unsigned int i, j;
int ret = 0;
int nrcpus;
size_t size;
if (!noaffinity)
pthread_attr_init(&thread_attr);
nrcpus = perf_cpu_map__nr(cpu);
cpuset = CPU_ALLOC(nrcpus);
BUG_ON(!cpuset);
size = CPU_ALLOC_SIZE(nrcpus);
for (i = 0; i < nthreads; i++) {
struct worker *w = &worker[i];
w->tid = i;
w->fdmap = calloc(nfds, sizeof(int));
if (!w->fdmap)
return 1;
for (j = 0; j < nfds; j++) {
w->fdmap[j] = eventfd(0, EFD_NONBLOCK);
if (w->fdmap[j] < 0)
err(EXIT_FAILURE, "eventfd");
}
/*
* Lets add 50% of the fdmap to the epoll instance, and
* do it before any threads are started; otherwise there is
* an initial bias of the call failing (mod and del ops).
*/
if (randomize)
init_fdmaps(w, 50);
if (!noaffinity) {
CPU_ZERO_S(size, cpuset);
CPU_SET_S(perf_cpu_map__cpu(cpu, i % perf_cpu_map__nr(cpu)).cpu,
size, cpuset);
ret = pthread_attr_setaffinity_np(&thread_attr, size, cpuset);
if (ret) {
CPU_FREE(cpuset);
err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
}
attrp = &thread_attr;
}
ret = pthread_create(&w->thread, attrp, workerfn,
(void *)(struct worker *) w);
if (ret) {
CPU_FREE(cpuset);
err(EXIT_FAILURE, "pthread_create");
}
}
CPU_FREE(cpuset);
if (!noaffinity)
pthread_attr_destroy(&thread_attr);
return ret;
}
static void print_summary(void)
{
int i;
unsigned long avg[EPOLL_NR_OPS];
double stddev[EPOLL_NR_OPS];
for (i = 0; i < EPOLL_NR_OPS; i++) {
avg[i] = avg_stats(&all_stats[i]);
stddev[i] = stddev_stats(&all_stats[i]);
}
printf("\nAveraged %ld ADD operations (+- %.2f%%)\n",
avg[OP_EPOLL_ADD], rel_stddev_stats(stddev[OP_EPOLL_ADD],
avg[OP_EPOLL_ADD]));
printf("Averaged %ld MOD operations (+- %.2f%%)\n",
avg[OP_EPOLL_MOD], rel_stddev_stats(stddev[OP_EPOLL_MOD],
avg[OP_EPOLL_MOD]));
printf("Averaged %ld DEL operations (+- %.2f%%)\n",
avg[OP_EPOLL_DEL], rel_stddev_stats(stddev[OP_EPOLL_DEL],
avg[OP_EPOLL_DEL]));
}
int bench_epoll_ctl(int argc, const char **argv)
{
int j, ret = 0;
struct sigaction act;
struct worker *worker = NULL;
struct perf_cpu_map *cpu;
struct rlimit rl, prevrl;
unsigned int i;
argc = parse_options(argc, argv, options, bench_epoll_ctl_usage, 0);
if (argc) {
usage_with_options(bench_epoll_ctl_usage, options);
exit(EXIT_FAILURE);
}
memset(&act, 0, sizeof(act));
sigfillset(&act.sa_mask);
act.sa_sigaction = toggle_done;
sigaction(SIGINT, &act, NULL);
cpu = perf_cpu_map__new(NULL);
if (!cpu)
goto errmem;
/* a single, main epoll instance */
epollfd = epoll_create(1);
if (epollfd < 0)
err(EXIT_FAILURE, "epoll_create");
/*
* Deal with nested epolls, if any.
*/
if (nested)
nest_epollfd();
/* default to the number of CPUs */
if (!nthreads)
nthreads = perf_cpu_map__nr(cpu);
worker = calloc(nthreads, sizeof(*worker));
if (!worker)
goto errmem;
if (getrlimit(RLIMIT_NOFILE, &prevrl))
err(EXIT_FAILURE, "getrlimit");
rl.rlim_cur = rl.rlim_max = nfds * nthreads * 2 + 50;
printinfo("Setting RLIMIT_NOFILE rlimit from %" PRIu64 " to: %" PRIu64 "\n",
(uint64_t)prevrl.rlim_max, (uint64_t)rl.rlim_max);
if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
err(EXIT_FAILURE, "setrlimit");
printf("Run summary [PID %d]: %d threads doing epoll_ctl ops "
"%d file-descriptors for %d secs.\n\n",
getpid(), nthreads, nfds, nsecs);
for (i = 0; i < EPOLL_NR_OPS; i++)
init_stats(&all_stats[i]);
pthread_mutex_init(&thread_lock, NULL);
pthread_cond_init(&thread_parent, NULL);
pthread_cond_init(&thread_worker, NULL);
threads_starting = nthreads;
gettimeofday(&bench__start, NULL);
do_threads(worker, cpu);
pthread_mutex_lock(&thread_lock);
while (threads_starting)
pthread_cond_wait(&thread_parent, &thread_lock);
pthread_cond_broadcast(&thread_worker);
pthread_mutex_unlock(&thread_lock);
sleep(nsecs);
toggle_done(0, NULL, NULL);
printinfo("main thread: toggling done\n");
for (i = 0; i < nthreads; i++) {
ret = pthread_join(worker[i].thread, NULL);
if (ret)
err(EXIT_FAILURE, "pthread_join");
}
/* cleanup & report results */
pthread_cond_destroy(&thread_parent);
pthread_cond_destroy(&thread_worker);
pthread_mutex_destroy(&thread_lock);
for (i = 0; i < nthreads; i++) {
unsigned long t[EPOLL_NR_OPS];
for (j = 0; j < EPOLL_NR_OPS; j++) {
t[j] = worker[i].ops[j];
update_stats(&all_stats[j], t[j]);
}
if (nfds == 1)
printf("[thread %2d] fdmap: %p [ add: %04ld; mod: %04ld; del: %04lds ops ]\n",
worker[i].tid, &worker[i].fdmap[0],
t[OP_EPOLL_ADD], t[OP_EPOLL_MOD], t[OP_EPOLL_DEL]);
else
printf("[thread %2d] fdmap: %p ... %p [ add: %04ld ops; mod: %04ld ops; del: %04ld ops ]\n",
worker[i].tid, &worker[i].fdmap[0],
&worker[i].fdmap[nfds-1],
t[OP_EPOLL_ADD], t[OP_EPOLL_MOD], t[OP_EPOLL_DEL]);
}
print_summary();
close(epollfd);
return ret;
errmem:
err(EXIT_FAILURE, "calloc");
}
#endif // HAVE_EVENTFD_SUPPORT