perf-eh_elf/bench/futex-wake-parallel.c
Linus Torvalds 16c00db4bb Merge tag 'afs-fixes-20180514' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
Pull AFS fixes from David Howells:
 "Here's a set of patches that fix a number of bugs in the in-kernel AFS
  client, including:

   - Fix directory locking to not use individual page locks for
     directory reading/scanning but rather to use a semaphore on the
     afs_vnode struct as the directory contents must be read in a single
     blob and data from different reads must not be mixed as the entire
     contents may be shuffled about between reads.

   - Fix address list parsing to handle port specifiers correctly.

   - Only give up callback records on a server if we actually talked to
     that server (we might not be able to access a server).

   - Fix some callback handling bugs, including refcounting,
     whole-volume callbacks and when callbacks actually get broken in
     response to a CB.CallBack op.

   - Fix some server/address rotation bugs, including giving up if we
     can't probe a server; giving up if a server says it doesn't have a
     volume, but there are more servers to try.

   - Fix the decoding of fetched statuses to be OpenAFS compatible.

   - Fix the handling of server lookups in Cache Manager ops (such as
     CB.InitCallBackState3) to use a UUID if possible and to handle no
     server being found.

   - Fix a bug in server lookup where not all addresses are compared.

   - Fix the non-encryption of calls that prevents some servers from
     being accessed (this also requires an AF_RXRPC patch that has
     already gone in through the net tree).

  There's also a patch that adds tracepoints to log Cache Manager ops
  that don't find a matching server, either by UUID or by address"

* tag 'afs-fixes-20180514' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
  afs: Fix the non-encryption of calls
  afs: Fix CB.CallBack handling
  afs: Fix whole-volume callback handling
  afs: Fix afs_find_server search loop
  afs: Fix the handling of an unfound server in CM operations
  afs: Add a tracepoint to record callbacks from unlisted servers
  afs: Fix the handling of CB.InitCallBackState3 to find the server by UUID
  afs: Fix VNOVOL handling in address rotation
  afs: Fix AFSFetchStatus decoder to provide OpenAFS compatibility
  afs: Fix server rotation's handling of fileserver probe failure
  afs: Fix refcounting in callback registration
  afs: Fix giving up callbacks on server destruction
  afs: Fix address list parsing
  afs: Fix directory page locking
2018-05-15 10:48:36 -07:00

323 lines
8.6 KiB
C

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (C) 2015 Davidlohr Bueso.
*
* Block a bunch of threads and let parallel waker threads wakeup an
* equal amount of them. The program output reflects the avg latency
* for each individual thread to service its share of work. Ultimately
* it can be used to measure futex_wake() changes.
*/
#include "bench.h"
#include <linux/compiler.h>
#include "../util/debug.h"
#ifndef HAVE_PTHREAD_BARRIER
int bench_futex_wake_parallel(int argc __maybe_unused, const char **argv __maybe_unused)
{
pr_err("%s: pthread_barrier_t unavailable, disabling this test...\n", __func__);
return 0;
}
#else /* HAVE_PTHREAD_BARRIER */
/* For the CLR_() macros */
#include <string.h>
#include <pthread.h>
#include <signal.h>
#include "../util/stat.h"
#include <subcmd/parse-options.h>
#include <linux/kernel.h>
#include <linux/time64.h>
#include <errno.h>
#include "futex.h"
#include "cpumap.h"
#include <err.h>
#include <stdlib.h>
#include <sys/time.h>
struct thread_data {
pthread_t worker;
unsigned int nwoken;
struct timeval runtime;
};
static unsigned int nwakes = 1;
/* all threads will block on the same futex -- hash bucket chaos ;) */
static u_int32_t futex = 0;
static pthread_t *blocked_worker;
static bool done = false, silent = false, fshared = false;
static unsigned int nblocked_threads = 0, nwaking_threads = 0;
static pthread_mutex_t thread_lock;
static pthread_cond_t thread_parent, thread_worker;
static pthread_barrier_t barrier;
static struct stats waketime_stats, wakeup_stats;
static unsigned int threads_starting;
static int futex_flag = 0;
static const struct option options[] = {
OPT_UINTEGER('t', "threads", &nblocked_threads, "Specify amount of threads"),
OPT_UINTEGER('w', "nwakers", &nwaking_threads, "Specify amount of waking threads"),
OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"),
OPT_BOOLEAN( 'S', "shared", &fshared, "Use shared futexes instead of private ones"),
OPT_END()
};
static const char * const bench_futex_wake_parallel_usage[] = {
"perf bench futex wake-parallel <options>",
NULL
};
static void *waking_workerfn(void *arg)
{
struct thread_data *waker = (struct thread_data *) arg;
struct timeval start, end;
pthread_barrier_wait(&barrier);
gettimeofday(&start, NULL);
waker->nwoken = futex_wake(&futex, nwakes, futex_flag);
if (waker->nwoken != nwakes)
warnx("couldn't wakeup all tasks (%d/%d)",
waker->nwoken, nwakes);
gettimeofday(&end, NULL);
timersub(&end, &start, &waker->runtime);
pthread_exit(NULL);
return NULL;
}
static void wakeup_threads(struct thread_data *td, pthread_attr_t thread_attr)
{
unsigned int i;
pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_JOINABLE);
pthread_barrier_init(&barrier, NULL, nwaking_threads + 1);
/* create and block all threads */
for (i = 0; i < nwaking_threads; i++) {
/*
* Thread creation order will impact per-thread latency
* as it will affect the order to acquire the hb spinlock.
* For now let the scheduler decide.
*/
if (pthread_create(&td[i].worker, &thread_attr,
waking_workerfn, (void *)&td[i]))
err(EXIT_FAILURE, "pthread_create");
}
pthread_barrier_wait(&barrier);
for (i = 0; i < nwaking_threads; i++)
if (pthread_join(td[i].worker, NULL))
err(EXIT_FAILURE, "pthread_join");
pthread_barrier_destroy(&barrier);
}
static void *blocked_workerfn(void *arg __maybe_unused)
{
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);
while (1) { /* handle spurious wakeups */
if (futex_wait(&futex, 0, NULL, futex_flag) != EINTR)
break;
}
pthread_exit(NULL);
return NULL;
}
static void block_threads(pthread_t *w, pthread_attr_t thread_attr,
struct cpu_map *cpu)
{
cpu_set_t cpuset;
unsigned int i;
threads_starting = nblocked_threads;
/* create and block all threads */
for (i = 0; i < nblocked_threads; i++) {
CPU_ZERO(&cpuset);
CPU_SET(cpu->map[i % cpu->nr], &cpuset);
if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpuset))
err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
if (pthread_create(&w[i], &thread_attr, blocked_workerfn, NULL))
err(EXIT_FAILURE, "pthread_create");
}
}
static void print_run(struct thread_data *waking_worker, unsigned int run_num)
{
unsigned int i, wakeup_avg;
double waketime_avg, waketime_stddev;
struct stats __waketime_stats, __wakeup_stats;
init_stats(&__wakeup_stats);
init_stats(&__waketime_stats);
for (i = 0; i < nwaking_threads; i++) {
update_stats(&__waketime_stats, waking_worker[i].runtime.tv_usec);
update_stats(&__wakeup_stats, waking_worker[i].nwoken);
}
waketime_avg = avg_stats(&__waketime_stats);
waketime_stddev = stddev_stats(&__waketime_stats);
wakeup_avg = avg_stats(&__wakeup_stats);
printf("[Run %d]: Avg per-thread latency (waking %d/%d threads) "
"in %.4f ms (+-%.2f%%)\n", run_num + 1, wakeup_avg,
nblocked_threads, waketime_avg / USEC_PER_MSEC,
rel_stddev_stats(waketime_stddev, waketime_avg));
}
static void print_summary(void)
{
unsigned int wakeup_avg;
double waketime_avg, waketime_stddev;
waketime_avg = avg_stats(&waketime_stats);
waketime_stddev = stddev_stats(&waketime_stats);
wakeup_avg = avg_stats(&wakeup_stats);
printf("Avg per-thread latency (waking %d/%d threads) in %.4f ms (+-%.2f%%)\n",
wakeup_avg,
nblocked_threads,
waketime_avg / USEC_PER_MSEC,
rel_stddev_stats(waketime_stddev, waketime_avg));
}
static void do_run_stats(struct thread_data *waking_worker)
{
unsigned int i;
for (i = 0; i < nwaking_threads; i++) {
update_stats(&waketime_stats, waking_worker[i].runtime.tv_usec);
update_stats(&wakeup_stats, waking_worker[i].nwoken);
}
}
static void toggle_done(int sig __maybe_unused,
siginfo_t *info __maybe_unused,
void *uc __maybe_unused)
{
done = true;
}
int bench_futex_wake_parallel(int argc, const char **argv)
{
int ret = 0;
unsigned int i, j;
struct sigaction act;
pthread_attr_t thread_attr;
struct thread_data *waking_worker;
struct cpu_map *cpu;
argc = parse_options(argc, argv, options,
bench_futex_wake_parallel_usage, 0);
if (argc) {
usage_with_options(bench_futex_wake_parallel_usage, options);
exit(EXIT_FAILURE);
}
sigfillset(&act.sa_mask);
act.sa_sigaction = toggle_done;
sigaction(SIGINT, &act, NULL);
cpu = cpu_map__new(NULL);
if (!cpu)
err(EXIT_FAILURE, "calloc");
if (!nblocked_threads)
nblocked_threads = cpu->nr;
/* some sanity checks */
if (nwaking_threads > nblocked_threads || !nwaking_threads)
nwaking_threads = nblocked_threads;
if (nblocked_threads % nwaking_threads)
errx(EXIT_FAILURE, "Must be perfectly divisible");
/*
* Each thread will wakeup nwakes tasks in
* a single futex_wait call.
*/
nwakes = nblocked_threads/nwaking_threads;
blocked_worker = calloc(nblocked_threads, sizeof(*blocked_worker));
if (!blocked_worker)
err(EXIT_FAILURE, "calloc");
if (!fshared)
futex_flag = FUTEX_PRIVATE_FLAG;
printf("Run summary [PID %d]: blocking on %d threads (at [%s] "
"futex %p), %d threads waking up %d at a time.\n\n",
getpid(), nblocked_threads, fshared ? "shared":"private",
&futex, nwaking_threads, nwakes);
init_stats(&wakeup_stats);
init_stats(&waketime_stats);
pthread_attr_init(&thread_attr);
pthread_mutex_init(&thread_lock, NULL);
pthread_cond_init(&thread_parent, NULL);
pthread_cond_init(&thread_worker, NULL);
for (j = 0; j < bench_repeat && !done; j++) {
waking_worker = calloc(nwaking_threads, sizeof(*waking_worker));
if (!waking_worker)
err(EXIT_FAILURE, "calloc");
/* create, launch & block all threads */
block_threads(blocked_worker, thread_attr, cpu);
/* make sure all threads are already blocked */
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);
usleep(100000);
/* Ok, all threads are patiently blocked, start waking folks up */
wakeup_threads(waking_worker, thread_attr);
for (i = 0; i < nblocked_threads; i++) {
ret = pthread_join(blocked_worker[i], NULL);
if (ret)
err(EXIT_FAILURE, "pthread_join");
}
do_run_stats(waking_worker);
if (!silent)
print_run(waking_worker, j);
free(waking_worker);
}
/* cleanup & report results */
pthread_cond_destroy(&thread_parent);
pthread_cond_destroy(&thread_worker);
pthread_mutex_destroy(&thread_lock);
pthread_attr_destroy(&thread_attr);
print_summary();
free(blocked_worker);
return ret;
}
#endif /* HAVE_PTHREAD_BARRIER */