numam-spdk/lib/event/reactor.c
Darek Stojaczyk f50d3d7759 reactor: pin spdk threads to their cores
We're about to drop legacy event messages from SPDK libs
and for that we'll replace various lcore numbers with
spdk_thread objects. For now SPDK libs can't spawn their
own threads, so they must use spdk_for_each_thread() and
spdk_get_thread() to retrieve different thread pointers.

The vhost library offers API to construct a vhost device
to be polled on specified cores, and in order to keep that
functionality we'll need to know which core each thread
is polling on. We would like to achieve that with
spdk_thread_get_cpumask(), but right now it always returns
all cores, so this patch changes it to return just
a single cpu on which the thread is actually pinned. It's
only a stop gap, eventually the vhost library will spawn
its own threads with custom cpumasks.

Change-Id: I15947727123c51b23f63727d52079770bfb2e07b
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/452204
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-05-02 17:06:34 +00:00

493 lines
12 KiB
C

/*-
* BSD LICENSE
*
* Copyright (c) Intel Corporation.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
* * Neither the name of Intel Corporation nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "spdk/stdinc.h"
#include "spdk/likely.h"
#include "spdk_internal/event.h"
#include "spdk_internal/log.h"
#include "spdk_internal/thread.h"
#include "spdk/log.h"
#include "spdk/thread.h"
#include "spdk/env.h"
#include "spdk/util.h"
#ifdef __linux__
#include <sys/prctl.h>
#endif
#ifdef __FreeBSD__
#include <pthread_np.h>
#endif
#define SPDK_EVENT_BATCH_SIZE 8
enum spdk_reactor_state {
SPDK_REACTOR_STATE_INVALID = 0,
SPDK_REACTOR_STATE_INITIALIZED = 1,
SPDK_REACTOR_STATE_RUNNING = 2,
SPDK_REACTOR_STATE_EXITING = 3,
SPDK_REACTOR_STATE_SHUTDOWN = 4,
};
struct spdk_lw_thread {
TAILQ_ENTRY(spdk_lw_thread) link;
};
struct spdk_reactor {
/* Logical core number for this reactor. */
uint32_t lcore;
/* Lightweight threads running on this reactor */
TAILQ_HEAD(, spdk_lw_thread) threads;
/* Poller for get the rusage for the reactor. */
struct spdk_poller *rusage_poller;
/* The last known rusage values */
struct rusage rusage;
struct spdk_ring *events;
} __attribute__((aligned(64)));
static struct spdk_reactor *g_reactors;
static enum spdk_reactor_state g_reactor_state = SPDK_REACTOR_STATE_INVALID;
static bool g_context_switch_monitor_enabled = true;
static void spdk_reactor_construct(struct spdk_reactor *w, uint32_t lcore);
static struct spdk_mempool *g_spdk_event_mempool = NULL;
static struct spdk_cpuset *g_spdk_app_core_mask;
static struct spdk_reactor *
spdk_reactor_get(uint32_t lcore)
{
struct spdk_reactor *reactor;
reactor = spdk_likely(g_reactors) ? &g_reactors[lcore] : NULL;
return reactor;
}
struct spdk_event *
spdk_event_allocate(uint32_t lcore, spdk_event_fn fn, void *arg1, void *arg2)
{
struct spdk_event *event = NULL;
struct spdk_reactor *reactor = spdk_reactor_get(lcore);
if (!reactor) {
assert(false);
return NULL;
}
event = spdk_mempool_get(g_spdk_event_mempool);
if (event == NULL) {
assert(false);
return NULL;
}
event->lcore = lcore;
event->fn = fn;
event->arg1 = arg1;
event->arg2 = arg2;
return event;
}
void
spdk_event_call(struct spdk_event *event)
{
int rc;
struct spdk_reactor *reactor;
reactor = spdk_reactor_get(event->lcore);
assert(reactor->events != NULL);
rc = spdk_ring_enqueue(reactor->events, (void **)&event, 1);
if (rc != 1) {
assert(false);
}
}
static inline uint32_t
_spdk_event_queue_run_batch(struct spdk_reactor *reactor)
{
unsigned count, i;
void *events[SPDK_EVENT_BATCH_SIZE];
struct spdk_thread *thread;
struct spdk_lw_thread *lw_thread;
#ifdef DEBUG
/*
* spdk_ring_dequeue() fills events and returns how many entries it wrote,
* so we will never actually read uninitialized data from events, but just to be sure
* (and to silence a static analyzer false positive), initialize the array to NULL pointers.
*/
memset(events, 0, sizeof(events));
#endif
count = spdk_ring_dequeue(reactor->events, events, SPDK_EVENT_BATCH_SIZE);
if (count == 0) {
return 0;
}
/* Execute the events. There are still some remaining events
* that must occur on an SPDK thread. To accomodate those, try to
* run them on the first thread in the list, if it exists. */
lw_thread = TAILQ_FIRST(&reactor->threads);
if (lw_thread) {
thread = spdk_thread_get_from_ctx(lw_thread);
} else {
thread = NULL;
}
spdk_set_thread(thread);
for (i = 0; i < count; i++) {
struct spdk_event *event = events[i];
assert(event != NULL);
event->fn(event->arg1, event->arg2);
}
spdk_set_thread(NULL);
spdk_mempool_put_bulk(g_spdk_event_mempool, events, count);
return count;
}
#define CONTEXT_SWITCH_MONITOR_PERIOD 1000000
static int
get_rusage(struct spdk_reactor *reactor)
{
struct rusage rusage;
if (getrusage(RUSAGE_THREAD, &rusage) != 0) {
return -1;
}
if (rusage.ru_nvcsw != reactor->rusage.ru_nvcsw || rusage.ru_nivcsw != reactor->rusage.ru_nivcsw) {
SPDK_INFOLOG(SPDK_LOG_REACTOR,
"Reactor %d: %ld voluntary context switches and %ld involuntary context switches in the last second.\n",
reactor->lcore, rusage.ru_nvcsw - reactor->rusage.ru_nvcsw,
rusage.ru_nivcsw - reactor->rusage.ru_nivcsw);
}
reactor->rusage = rusage;
return -1;
}
void
spdk_reactor_enable_context_switch_monitor(bool enable)
{
/* This global is being read by multiple threads, so this isn't
* strictly thread safe. However, we're toggling between true and
* false here, and if a thread sees the value update later than it
* should, it's no big deal. */
g_context_switch_monitor_enabled = enable;
}
bool
spdk_reactor_context_switch_monitor_enabled(void)
{
return g_context_switch_monitor_enabled;
}
static void
_set_thread_name(const char *thread_name)
{
#if defined(__linux__)
prctl(PR_SET_NAME, thread_name, 0, 0, 0);
#elif defined(__FreeBSD__)
pthread_set_name_np(pthread_self(), thread_name);
#else
#error missing platform support for thread name
#endif
}
static int
_spdk_reactor_run(void *arg)
{
struct spdk_reactor *reactor = arg;
struct spdk_thread *thread;
uint64_t last_rusage = 0;
struct spdk_lw_thread *lw_thread, *tmp;
char thread_name[32];
SPDK_NOTICELOG("Reactor started on core %u\n", reactor->lcore);
/* Rename the POSIX thread because the reactor is tied to the POSIX
* thread in the SPDK event library.
*/
snprintf(thread_name, sizeof(thread_name), "reactor_%u", reactor->lcore);
_set_thread_name(thread_name);
while (1) {
uint64_t now;
/* For each loop through the reactor, capture the time. This time
* is used for all threads. */
now = spdk_get_ticks();
_spdk_event_queue_run_batch(reactor);
TAILQ_FOREACH_SAFE(lw_thread, &reactor->threads, link, tmp) {
thread = spdk_thread_get_from_ctx(lw_thread);
spdk_thread_poll(thread, 0, now);
}
if (g_reactor_state != SPDK_REACTOR_STATE_RUNNING) {
break;
}
if (g_context_switch_monitor_enabled) {
if ((last_rusage + CONTEXT_SWITCH_MONITOR_PERIOD) < now) {
get_rusage(reactor);
last_rusage = now;
}
}
}
TAILQ_FOREACH_SAFE(lw_thread, &reactor->threads, link, tmp) {
thread = spdk_thread_get_from_ctx(lw_thread);
TAILQ_REMOVE(&reactor->threads, lw_thread, link);
spdk_thread_exit(thread);
}
return 0;
}
static void
spdk_reactor_construct(struct spdk_reactor *reactor, uint32_t lcore)
{
reactor->lcore = lcore;
TAILQ_INIT(&reactor->threads);
reactor->events = spdk_ring_create(SPDK_RING_TYPE_MP_SC, 65536, SPDK_ENV_SOCKET_ID_ANY);
assert(reactor->events != NULL);
}
int
spdk_app_parse_core_mask(const char *mask, struct spdk_cpuset *cpumask)
{
int ret;
struct spdk_cpuset *validmask;
ret = spdk_cpuset_parse(cpumask, mask);
if (ret < 0) {
return ret;
}
validmask = spdk_app_get_core_mask();
spdk_cpuset_and(cpumask, validmask);
return 0;
}
struct spdk_cpuset *
spdk_app_get_core_mask(void)
{
return g_spdk_app_core_mask;
}
void
spdk_reactors_start(void)
{
struct spdk_reactor *reactor;
struct spdk_cpuset *tmp_cpumask;
uint32_t i, current_core;
int rc;
char thread_name[32];
tmp_cpumask = spdk_cpuset_alloc();
if (tmp_cpumask == NULL) {
SPDK_ERRLOG("spdk_cpuset_alloc() failed\n");
assert(false);
return;
}
g_reactor_state = SPDK_REACTOR_STATE_RUNNING;
g_spdk_app_core_mask = spdk_cpuset_alloc();
current_core = spdk_env_get_current_core();
SPDK_ENV_FOREACH_CORE(i) {
if (i != current_core) {
reactor = spdk_reactor_get(i);
rc = spdk_env_thread_launch_pinned(reactor->lcore, _spdk_reactor_run, reactor);
if (rc < 0) {
SPDK_ERRLOG("Unable to start reactor thread on core %u\n", reactor->lcore);
assert(false);
return;
}
/* For now, for each reactor spawn one thread. */
snprintf(thread_name, sizeof(thread_name), "reactor_%u", reactor->lcore);
spdk_cpuset_zero(tmp_cpumask);
spdk_cpuset_set_cpu(tmp_cpumask, i, true);
spdk_thread_create(thread_name, tmp_cpumask);
}
spdk_cpuset_set_cpu(g_spdk_app_core_mask, i, true);
}
spdk_cpuset_free(tmp_cpumask);
/* Start the master reactor */
reactor = spdk_reactor_get(current_core);
_spdk_reactor_run(reactor);
spdk_env_thread_wait_all();
g_reactor_state = SPDK_REACTOR_STATE_SHUTDOWN;
spdk_cpuset_free(g_spdk_app_core_mask);
g_spdk_app_core_mask = NULL;
}
void
spdk_reactors_stop(void *arg1)
{
g_reactor_state = SPDK_REACTOR_STATE_EXITING;
}
static pthread_mutex_t g_scheduler_mtx = PTHREAD_MUTEX_INITIALIZER;
static uint32_t g_next_core = UINT32_MAX;
static void
_schedule_thread(void *arg1, void *arg2)
{
struct spdk_lw_thread *lw_thread = arg1;
struct spdk_reactor *reactor;
reactor = spdk_reactor_get(spdk_env_get_current_core());
TAILQ_INSERT_TAIL(&reactor->threads, lw_thread, link);
}
static void
spdk_reactor_schedule_thread(struct spdk_thread *thread)
{
uint32_t core;
struct spdk_lw_thread *lw_thread;
struct spdk_event *evt;
lw_thread = spdk_thread_get_ctx(thread);
assert(lw_thread != NULL);
memset(lw_thread, 0, sizeof(*lw_thread));
pthread_mutex_lock(&g_scheduler_mtx);
if (g_next_core > spdk_env_get_last_core()) {
g_next_core = spdk_env_get_first_core();
}
core = g_next_core;
g_next_core = spdk_env_get_next_core(g_next_core);
pthread_mutex_unlock(&g_scheduler_mtx);
evt = spdk_event_allocate(core, _schedule_thread, lw_thread, NULL);
spdk_event_call(evt);
}
int
spdk_reactors_init(void)
{
int rc;
uint32_t i, last_core;
struct spdk_reactor *reactor;
char mempool_name[32];
snprintf(mempool_name, sizeof(mempool_name), "evtpool_%d", getpid());
g_spdk_event_mempool = spdk_mempool_create(mempool_name,
262144 - 1, /* Power of 2 minus 1 is optimal for memory consumption */
sizeof(struct spdk_event),
SPDK_MEMPOOL_DEFAULT_CACHE_SIZE,
SPDK_ENV_SOCKET_ID_ANY);
if (g_spdk_event_mempool == NULL) {
SPDK_ERRLOG("spdk_event_mempool creation failed\n");
return -1;
}
/* struct spdk_reactor must be aligned on 64 byte boundary */
last_core = spdk_env_get_last_core();
rc = posix_memalign((void **)&g_reactors, 64,
(last_core + 1) * sizeof(struct spdk_reactor));
if (rc != 0) {
SPDK_ERRLOG("Could not allocate array size=%u for g_reactors\n",
last_core + 1);
spdk_mempool_free(g_spdk_event_mempool);
return -1;
}
memset(g_reactors, 0, (last_core + 1) * sizeof(struct spdk_reactor));
spdk_thread_lib_init(spdk_reactor_schedule_thread, sizeof(struct spdk_lw_thread));
SPDK_ENV_FOREACH_CORE(i) {
reactor = spdk_reactor_get(i);
spdk_reactor_construct(reactor, i);
}
g_reactor_state = SPDK_REACTOR_STATE_INITIALIZED;
return 0;
}
void
spdk_reactors_fini(void)
{
uint32_t i;
struct spdk_reactor *reactor;
spdk_thread_lib_fini();
SPDK_ENV_FOREACH_CORE(i) {
reactor = spdk_reactor_get(i);
if (spdk_likely(reactor != NULL) && reactor->events != NULL) {
spdk_ring_free(reactor->events);
}
}
spdk_mempool_free(g_spdk_event_mempool);
free(g_reactors);
g_reactors = NULL;
}
SPDK_LOG_REGISTER_COMPONENT("reactor", SPDK_LOG_REACTOR)