Spdk/test/unit/lib/thread/thread.c/thread_ut.c

2565 lines
69 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright (C) 2016 Intel Corporation.
* All rights reserved.
* Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
*/
#include "spdk/stdinc.h"
#include "spdk_cunit.h"
#include "thread/thread_internal.h"
#include "thread/thread.c"
#include "common/lib/ut_multithread.c"
static int g_sched_rc = 0;
static int
_thread_schedule(struct spdk_thread *thread)
{
return g_sched_rc;
}
static bool
_thread_op_supported(enum spdk_thread_op op)
{
switch (op) {
case SPDK_THREAD_OP_NEW:
return true;
default:
return false;
}
}
static int
_thread_op(struct spdk_thread *thread, enum spdk_thread_op op)
{
switch (op) {
case SPDK_THREAD_OP_NEW:
return _thread_schedule(thread);
default:
return -ENOTSUP;
}
}
static void
thread_alloc(void)
{
struct spdk_thread *thread;
/* No schedule callback */
spdk_thread_lib_init(NULL, 0);
thread = spdk_thread_create(NULL, NULL);
SPDK_CU_ASSERT_FATAL(thread != NULL);
spdk_set_thread(thread);
spdk_thread_exit(thread);
while (!spdk_thread_is_exited(thread)) {
spdk_thread_poll(thread, 0, 0);
}
spdk_thread_destroy(thread);
spdk_thread_lib_fini();
/* Schedule callback exists */
spdk_thread_lib_init(_thread_schedule, 0);
/* Scheduling succeeds */
g_sched_rc = 0;
thread = spdk_thread_create(NULL, NULL);
SPDK_CU_ASSERT_FATAL(thread != NULL);
spdk_set_thread(thread);
spdk_thread_exit(thread);
while (!spdk_thread_is_exited(thread)) {
spdk_thread_poll(thread, 0, 0);
}
spdk_thread_destroy(thread);
/* Scheduling fails */
g_sched_rc = -1;
thread = spdk_thread_create(NULL, NULL);
SPDK_CU_ASSERT_FATAL(thread == NULL);
spdk_thread_lib_fini();
/* Scheduling callback exists with extended thread library initialization. */
spdk_thread_lib_init_ext(_thread_op, _thread_op_supported, 0,
SPDK_DEFAULT_MSG_MEMPOOL_SIZE);
/* Scheduling succeeds */
g_sched_rc = 0;
thread = spdk_thread_create(NULL, NULL);
SPDK_CU_ASSERT_FATAL(thread != NULL);
spdk_set_thread(thread);
spdk_thread_exit(thread);
while (!spdk_thread_is_exited(thread)) {
spdk_thread_poll(thread, 0, 0);
}
spdk_thread_destroy(thread);
/* Scheduling fails */
g_sched_rc = -1;
thread = spdk_thread_create(NULL, NULL);
SPDK_CU_ASSERT_FATAL(thread == NULL);
spdk_thread_lib_fini();
}
static void
send_msg_cb(void *ctx)
{
bool *done = ctx;
*done = true;
}
static void
thread_send_msg(void)
{
struct spdk_thread *thread0;
bool done = false;
allocate_threads(2);
set_thread(0);
thread0 = spdk_get_thread();
set_thread(1);
/* Simulate thread 1 sending a message to thread 0. */
spdk_thread_send_msg(thread0, send_msg_cb, &done);
/* We have not polled thread 0 yet, so done should be false. */
CU_ASSERT(!done);
/*
* Poll thread 1. The message was sent to thread 0, so this should be
* a nop and done should still be false.
*/
poll_thread(1);
CU_ASSERT(!done);
/*
* Poll thread 0. This should execute the message and done should then
* be true.
*/
poll_thread(0);
CU_ASSERT(done);
free_threads();
}
static int
poller_run_done(void *ctx)
{
bool *poller_run = ctx;
*poller_run = true;
return -1;
}
static void
thread_poller(void)
{
struct spdk_poller *poller = NULL;
bool poller_run = false;
allocate_threads(1);
set_thread(0);
MOCK_SET(spdk_get_ticks, 0);
/* Register a poller with no-wait time and test execution */
poller = spdk_poller_register(poller_run_done, &poller_run, 0);
CU_ASSERT(poller != NULL);
poll_threads();
CU_ASSERT(poller_run == true);
spdk_poller_unregister(&poller);
CU_ASSERT(poller == NULL);
/* Register a poller with 1000us wait time and test single execution */
poller_run = false;
poller = spdk_poller_register(poller_run_done, &poller_run, 1000);
CU_ASSERT(poller != NULL);
poll_threads();
CU_ASSERT(poller_run == false);
spdk_delay_us(1000);
poll_threads();
CU_ASSERT(poller_run == true);
poller_run = false;
poll_threads();
CU_ASSERT(poller_run == false);
spdk_delay_us(1000);
poll_threads();
CU_ASSERT(poller_run == true);
spdk_poller_unregister(&poller);
CU_ASSERT(poller == NULL);
free_threads();
}
struct poller_ctx {
struct spdk_poller *poller;
bool run;
};
static int
poller_run_pause(void *ctx)
{
struct poller_ctx *poller_ctx = ctx;
poller_ctx->run = true;
spdk_poller_pause(poller_ctx->poller);
return 0;
}
/* Verify the same poller can be switched multiple times between
* pause and resume while it runs.
*/
static int
poller_run_pause_resume_pause(void *ctx)
{
struct poller_ctx *poller_ctx = ctx;
poller_ctx->run = true;
spdk_poller_pause(poller_ctx->poller);
spdk_poller_resume(poller_ctx->poller);
spdk_poller_pause(poller_ctx->poller);
return 0;
}
static void
poller_msg_pause_cb(void *ctx)
{
struct spdk_poller *poller = ctx;
spdk_poller_pause(poller);
}
static void
poller_msg_resume_cb(void *ctx)
{
struct spdk_poller *poller = ctx;
spdk_poller_resume(poller);
}
static void
poller_pause(void)
{
struct poller_ctx poller_ctx = {};
unsigned int delay[] = { 0, 1000 };
unsigned int i;
allocate_threads(1);
set_thread(0);
/* Register a poller that pauses itself */
poller_ctx.poller = spdk_poller_register(poller_run_pause, &poller_ctx, 0);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
/* Register a poller that switches between pause and resume itself */
poller_ctx.poller = spdk_poller_register(poller_run_pause_resume_pause, &poller_ctx, 0);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
/* Verify that resuming an unpaused poller doesn't do anything */
poller_ctx.poller = spdk_poller_register(poller_run_done, &poller_ctx.run, 0);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
spdk_poller_resume(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
/* Verify that pausing the same poller twice works too */
spdk_poller_pause(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_pause(poller_ctx.poller);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_resume(poller_ctx.poller);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
/* Verify that a poller is run when it's resumed immediately after pausing */
poller_ctx.run = false;
spdk_poller_pause(poller_ctx.poller);
spdk_poller_resume(poller_ctx.poller);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
/* Poll the thread to make sure the previous poller gets unregistered */
poll_threads();
CU_ASSERT_EQUAL(spdk_thread_has_pollers(spdk_get_thread()), false);
/* Verify that it's possible to unregister a paused poller */
poller_ctx.poller = spdk_poller_register(poller_run_done, &poller_ctx.run, 0);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_pause(poller_ctx.poller);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_unregister(&poller_ctx.poller);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
CU_ASSERT_EQUAL(spdk_thread_has_pollers(spdk_get_thread()), false);
/* Register pollers with 0 and 1000us wait time and pause/resume them */
for (i = 0; i < SPDK_COUNTOF(delay); ++i) {
poller_ctx.poller = spdk_poller_register(poller_run_done, &poller_ctx.run, delay[i]);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
spdk_delay_us(delay[i]);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_pause(poller_ctx.poller);
spdk_delay_us(delay[i]);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_resume(poller_ctx.poller);
spdk_delay_us(delay[i]);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
/* Verify that the poller can be paused/resumed from spdk_thread_send_msg */
spdk_thread_send_msg(spdk_get_thread(), poller_msg_pause_cb, poller_ctx.poller);
spdk_delay_us(delay[i]);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_thread_send_msg(spdk_get_thread(), poller_msg_resume_cb, poller_ctx.poller);
poll_threads();
if (delay[i] > 0) {
spdk_delay_us(delay[i]);
poll_threads();
}
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
/* Register a timed poller that pauses itself */
poller_ctx.poller = spdk_poller_register(poller_run_pause, &poller_ctx, delay[i]);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
spdk_delay_us(delay[i]);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
poller_ctx.run = false;
spdk_delay_us(delay[i]);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_resume(poller_ctx.poller);
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_delay_us(delay[i]);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
/* Register a timed poller that switches between pause and resume itself */
poller_ctx.poller = spdk_poller_register(poller_run_pause_resume_pause,
&poller_ctx, delay[i]);
CU_ASSERT_PTR_NOT_NULL(poller_ctx.poller);
spdk_delay_us(delay[i]);
poller_ctx.run = false;
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
poller_ctx.run = false;
spdk_delay_us(delay[i]);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_poller_resume(poller_ctx.poller);
CU_ASSERT_EQUAL(poller_ctx.run, false);
spdk_delay_us(delay[i]);
poll_threads();
CU_ASSERT_EQUAL(poller_ctx.run, true);
spdk_poller_unregister(&poller_ctx.poller);
CU_ASSERT_PTR_NULL(poller_ctx.poller);
}
free_threads();
}
static void
for_each_cb(void *ctx)
{
int *count = ctx;
(*count)++;
}
static void
thread_for_each(void)
{
int count = 0;
int i;
allocate_threads(3);
set_thread(0);
spdk_for_each_thread(for_each_cb, &count, for_each_cb);
/* We have not polled thread 0 yet, so count should be 0 */
CU_ASSERT(count == 0);
/* Poll each thread to verify the message is passed to each */
for (i = 0; i < 3; i++) {
poll_thread(i);
CU_ASSERT(count == (i + 1));
}
/*
* After each thread is called, the completion calls it
* one more time.
*/
poll_thread(0);
CU_ASSERT(count == 4);
free_threads();
}
static int
channel_create(void *io_device, void *ctx_buf)
{
int *ch_count = io_device;
(*ch_count)++;
return 0;
}
static void
channel_destroy(void *io_device, void *ctx_buf)
{
int *ch_count = io_device;
(*ch_count)--;
}
static void
channel_msg(struct spdk_io_channel_iter *i)
{
int *msg_count = spdk_io_channel_iter_get_ctx(i);
(*msg_count)++;
spdk_for_each_channel_continue(i, 0);
}
static void
channel_cpl(struct spdk_io_channel_iter *i, int status)
{
int *msg_count = spdk_io_channel_iter_get_ctx(i);
(*msg_count)++;
}
static void
for_each_channel_remove(void)
{
struct spdk_io_channel *ch0, *ch1, *ch2;
int ch_count = 0;
int msg_count = 0;
allocate_threads(3);
set_thread(0);
spdk_io_device_register(&ch_count, channel_create, channel_destroy, sizeof(int), NULL);
ch0 = spdk_get_io_channel(&ch_count);
set_thread(1);
ch1 = spdk_get_io_channel(&ch_count);
set_thread(2);
ch2 = spdk_get_io_channel(&ch_count);
CU_ASSERT(ch_count == 3);
/*
* Test that io_channel handles the case where we start to iterate through
* the channels, and during the iteration, one of the channels is deleted.
* This is done in some different and sometimes non-intuitive orders, because
* some operations are deferred and won't execute until their threads are
* polled.
*
* Case #1: Put the I/O channel before spdk_for_each_channel.
*/
set_thread(0);
spdk_put_io_channel(ch0);
CU_ASSERT(ch_count == 3);
poll_threads();
CU_ASSERT(ch_count == 2);
spdk_for_each_channel(&ch_count, channel_msg, &msg_count, channel_cpl);
CU_ASSERT(msg_count == 0);
poll_threads();
CU_ASSERT(msg_count == 3);
msg_count = 0;
/*
* Case #2: Put the I/O channel after spdk_for_each_channel, but before
* thread 0 is polled.
*/
ch0 = spdk_get_io_channel(&ch_count);
CU_ASSERT(ch_count == 3);
spdk_for_each_channel(&ch_count, channel_msg, &msg_count, channel_cpl);
spdk_put_io_channel(ch0);
CU_ASSERT(ch_count == 3);
poll_threads();
CU_ASSERT(ch_count == 2);
CU_ASSERT(msg_count == 4);
set_thread(1);
spdk_put_io_channel(ch1);
CU_ASSERT(ch_count == 2);
set_thread(2);
spdk_put_io_channel(ch2);
CU_ASSERT(ch_count == 2);
poll_threads();
CU_ASSERT(ch_count == 0);
spdk_io_device_unregister(&ch_count, NULL);
poll_threads();
free_threads();
}
struct unreg_ctx {
bool ch_done;
bool foreach_done;
};
static void
unreg_ch_done(struct spdk_io_channel_iter *i)
{
struct unreg_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
ctx->ch_done = true;
SPDK_CU_ASSERT_FATAL(i->cur_thread != NULL);
spdk_for_each_channel_continue(i, 0);
}
static void
unreg_foreach_done(struct spdk_io_channel_iter *i, int status)
{
struct unreg_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
ctx->foreach_done = true;
}
static void
for_each_channel_unreg(void)
{
struct spdk_io_channel *ch0;
struct io_device *dev;
struct unreg_ctx ctx = {};
int io_target = 0;
allocate_threads(1);
set_thread(0);
CU_ASSERT(RB_EMPTY(&g_io_devices));
spdk_io_device_register(&io_target, channel_create, channel_destroy, sizeof(int), NULL);
CU_ASSERT(!RB_EMPTY(&g_io_devices));
dev = RB_MIN(io_device_tree, &g_io_devices);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(RB_NEXT(io_device_tree, &g_io_devices, dev) == NULL);
ch0 = spdk_get_io_channel(&io_target);
spdk_io_device_register(&io_target, channel_create, channel_destroy, sizeof(int), NULL);
/*
* There is already a device registered at &io_target, so a new io_device should not
* have been added to g_io_devices.
*/
CU_ASSERT(dev == RB_MIN(io_device_tree, &g_io_devices));
CU_ASSERT(RB_NEXT(io_device_tree, &g_io_devices, dev) == NULL);
spdk_for_each_channel(&io_target, unreg_ch_done, &ctx, unreg_foreach_done);
spdk_io_device_unregister(&io_target, NULL);
/*
* There is an outstanding foreach call on the io_device, so the unregister should not
* have immediately removed the device.
*/
CU_ASSERT(dev == RB_MIN(io_device_tree, &g_io_devices));
poll_thread(0);
CU_ASSERT(ctx.ch_done == true);
CU_ASSERT(ctx.foreach_done == true);
/*
* There are no more foreach operations outstanding, so the device should be
* unregistered.
*/
CU_ASSERT(RB_EMPTY(&g_io_devices));
set_thread(0);
spdk_put_io_channel(ch0);
poll_threads();
free_threads();
}
static void
thread_name(void)
{
struct spdk_thread *thread;
const char *name;
spdk_thread_lib_init(NULL, 0);
/* Create thread with no name, which automatically generates one */
thread = spdk_thread_create(NULL, NULL);
spdk_set_thread(thread);
thread = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread != NULL);
name = spdk_thread_get_name(thread);
CU_ASSERT(name != NULL);
spdk_thread_exit(thread);
while (!spdk_thread_is_exited(thread)) {
spdk_thread_poll(thread, 0, 0);
}
spdk_thread_destroy(thread);
/* Create thread named "test_thread" */
thread = spdk_thread_create("test_thread", NULL);
spdk_set_thread(thread);
thread = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread != NULL);
name = spdk_thread_get_name(thread);
SPDK_CU_ASSERT_FATAL(name != NULL);
CU_ASSERT(strcmp(name, "test_thread") == 0);
spdk_thread_exit(thread);
while (!spdk_thread_is_exited(thread)) {
spdk_thread_poll(thread, 0, 0);
}
spdk_thread_destroy(thread);
spdk_thread_lib_fini();
}
static uint64_t g_device1;
static uint64_t g_device2;
static uint64_t g_device3;
static uint64_t g_ctx1 = 0x1111;
static uint64_t g_ctx2 = 0x2222;
static int g_create_cb_calls = 0;
static int g_destroy_cb_calls = 0;
static int
create_cb_1(void *io_device, void *ctx_buf)
{
CU_ASSERT(io_device == &g_device1);
*(uint64_t *)ctx_buf = g_ctx1;
g_create_cb_calls++;
return 0;
}
static void
destroy_cb_1(void *io_device, void *ctx_buf)
{
CU_ASSERT(io_device == &g_device1);
CU_ASSERT(*(uint64_t *)ctx_buf == g_ctx1);
g_destroy_cb_calls++;
}
static int
create_cb_2(void *io_device, void *ctx_buf)
{
CU_ASSERT(io_device == &g_device2);
*(uint64_t *)ctx_buf = g_ctx2;
g_create_cb_calls++;
return 0;
}
static void
destroy_cb_2(void *io_device, void *ctx_buf)
{
CU_ASSERT(io_device == &g_device2);
CU_ASSERT(*(uint64_t *)ctx_buf == g_ctx2);
g_destroy_cb_calls++;
}
static void
channel(void)
{
struct spdk_io_channel *ch1, *ch2;
void *ctx;
allocate_threads(1);
set_thread(0);
spdk_io_device_register(&g_device1, create_cb_1, destroy_cb_1, sizeof(g_ctx1), NULL);
spdk_io_device_register(&g_device2, create_cb_2, destroy_cb_2, sizeof(g_ctx2), NULL);
g_create_cb_calls = 0;
ch1 = spdk_get_io_channel(&g_device1);
CU_ASSERT(g_create_cb_calls == 1);
SPDK_CU_ASSERT_FATAL(ch1 != NULL);
CU_ASSERT(spdk_io_channel_get_io_device(ch1) == &g_device1);
g_create_cb_calls = 0;
ch2 = spdk_get_io_channel(&g_device1);
CU_ASSERT(g_create_cb_calls == 0);
CU_ASSERT(ch1 == ch2);
SPDK_CU_ASSERT_FATAL(ch2 != NULL);
CU_ASSERT(spdk_io_channel_get_io_device(ch2) == &g_device1);
g_destroy_cb_calls = 0;
spdk_put_io_channel(ch2);
poll_threads();
CU_ASSERT(g_destroy_cb_calls == 0);
g_create_cb_calls = 0;
ch2 = spdk_get_io_channel(&g_device2);
CU_ASSERT(g_create_cb_calls == 1);
CU_ASSERT(ch1 != ch2);
SPDK_CU_ASSERT_FATAL(ch2 != NULL);
CU_ASSERT(spdk_io_channel_get_io_device(ch2) == &g_device2);
ctx = spdk_io_channel_get_ctx(ch2);
CU_ASSERT(*(uint64_t *)ctx == g_ctx2);
g_destroy_cb_calls = 0;
spdk_put_io_channel(ch1);
poll_threads();
CU_ASSERT(g_destroy_cb_calls == 1);
g_destroy_cb_calls = 0;
spdk_put_io_channel(ch2);
poll_threads();
CU_ASSERT(g_destroy_cb_calls == 1);
ch1 = spdk_get_io_channel(&g_device3);
CU_ASSERT(ch1 == NULL);
spdk_io_device_unregister(&g_device1, NULL);
poll_threads();
spdk_io_device_unregister(&g_device2, NULL);
poll_threads();
CU_ASSERT(RB_EMPTY(&g_io_devices));
free_threads();
CU_ASSERT(TAILQ_EMPTY(&g_threads));
}
static int
create_cb(void *io_device, void *ctx_buf)
{
uint64_t *refcnt = (uint64_t *)ctx_buf;
CU_ASSERT(*refcnt == 0);
*refcnt = 1;
return 0;
}
static void
destroy_cb(void *io_device, void *ctx_buf)
{
uint64_t *refcnt = (uint64_t *)ctx_buf;
CU_ASSERT(*refcnt == 1);
*refcnt = 0;
}
/**
* This test is checking that a sequence of get, put, get, put without allowing
* the deferred put operation to complete doesn't result in releasing the memory
* for the channel twice.
*/
static void
channel_destroy_races(void)
{
uint64_t device;
struct spdk_io_channel *ch;
allocate_threads(1);
set_thread(0);
spdk_io_device_register(&device, create_cb, destroy_cb, sizeof(uint64_t), NULL);
ch = spdk_get_io_channel(&device);
SPDK_CU_ASSERT_FATAL(ch != NULL);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel(&device);
SPDK_CU_ASSERT_FATAL(ch != NULL);
spdk_put_io_channel(ch);
poll_threads();
spdk_io_device_unregister(&device, NULL);
poll_threads();
CU_ASSERT(RB_EMPTY(&g_io_devices));
free_threads();
CU_ASSERT(TAILQ_EMPTY(&g_threads));
}
static void
thread_exit_test(void)
{
struct spdk_thread *thread;
struct spdk_io_channel *ch;
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
struct spdk_poller *poller1, *poller2;
void *ctx;
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
bool done1 = false, done2 = false, poller1_run = false, poller2_run = false;
int rc __attribute__((unused));
MOCK_SET(spdk_get_ticks, 10);
MOCK_SET(spdk_get_ticks_hz, 1);
allocate_threads(4);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Test if all pending messages are reaped for the exiting thread, and the
* thread moves to the exited state.
*/
set_thread(0);
thread = spdk_get_thread();
/* Sending message to thread 0 will be accepted. */
rc = spdk_thread_send_msg(thread, send_msg_cb, &done1);
CU_ASSERT(rc == 0);
CU_ASSERT(!done1);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Move thread 0 to the exiting state. */
spdk_thread_exit(thread);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(spdk_thread_is_exited(thread) == false);
/* Sending message to thread 0 will be still accepted. */
rc = spdk_thread_send_msg(thread, send_msg_cb, &done2);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(rc == 0);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Thread 0 will reap pending messages. */
poll_thread(0);
CU_ASSERT(done1 == true);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(done2 == true);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Thread 0 will move to the exited state. */
CU_ASSERT(spdk_thread_is_exited(thread) == true);
/* Test releasing I/O channel is reaped even after the thread moves to
* the exiting state
*/
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
set_thread(1);
spdk_io_device_register(&g_device1, create_cb_1, destroy_cb_1, sizeof(g_ctx1), NULL);
g_create_cb_calls = 0;
ch = spdk_get_io_channel(&g_device1);
CU_ASSERT(g_create_cb_calls == 1);
SPDK_CU_ASSERT_FATAL(ch != NULL);
ctx = spdk_io_channel_get_ctx(ch);
CU_ASSERT(*(uint64_t *)ctx == g_ctx1);
g_destroy_cb_calls = 0;
spdk_put_io_channel(ch);
thread = spdk_get_thread();
spdk_thread_exit(thread);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Thread 1 will not move to the exited state yet because I/O channel release
* does not complete yet.
*/
CU_ASSERT(spdk_thread_is_exited(thread) == false);
/* Thread 1 will be able to get the another reference of I/O channel
* even after the thread moves to the exiting state.
*/
g_create_cb_calls = 0;
ch = spdk_get_io_channel(&g_device1);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(g_create_cb_calls == 0);
SPDK_CU_ASSERT_FATAL(ch != NULL);
ctx = spdk_io_channel_get_ctx(ch);
CU_ASSERT(*(uint64_t *)ctx == g_ctx1);
spdk_put_io_channel(ch);
poll_threads();
CU_ASSERT(g_destroy_cb_calls == 1);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Thread 1 will move to the exited state after I/O channel is released.
* are released.
*/
CU_ASSERT(spdk_thread_is_exited(thread) == true);
spdk_io_device_unregister(&g_device1, NULL);
poll_threads();
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
/* Test if unregistering poller is reaped for the exiting thread, and the
* thread moves to the exited thread.
*/
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
set_thread(2);
thread = spdk_get_thread();
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
poller1 = spdk_poller_register(poller_run_done, &poller1_run, 0);
CU_ASSERT(poller1 != NULL);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
spdk_poller_unregister(&poller1);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
spdk_thread_exit(thread);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
poller2 = spdk_poller_register(poller_run_done, &poller2_run, 0);
poll_threads();
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(poller1_run == false);
CU_ASSERT(poller2_run == true);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(spdk_thread_is_exited(thread) == false);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
spdk_poller_unregister(&poller2);
poll_threads();
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
CU_ASSERT(spdk_thread_is_exited(thread) == true);
/* Test if the exiting thread is exited forcefully after timeout. */
set_thread(3);
thread = spdk_get_thread();
poller1 = spdk_poller_register(poller_run_done, &poller1_run, 0);
CU_ASSERT(poller1 != NULL);
spdk_thread_exit(thread);
CU_ASSERT(spdk_thread_is_exited(thread) == false);
MOCK_SET(spdk_get_ticks, 11);
poll_threads();
CU_ASSERT(spdk_thread_is_exited(thread) == false);
/* Cause timeout forcefully. */
MOCK_SET(spdk_get_ticks, 15);
poll_threads();
CU_ASSERT(spdk_thread_is_exited(thread) == true);
spdk_poller_unregister(&poller1);
poll_threads();
MOCK_CLEAR(spdk_get_ticks);
MOCK_CLEAR(spdk_get_ticks_hz);
free_threads();
}
lib/thread: spdk_thread_poll() measures run time per call by reading end TSC Update spdk_thread_poll() to count SPDK thread stats correctly on multiple SPDK threads per reactor configuration. spdk_thread_poll() gets start time and reads TSC at end as end time, and then gets delta between them as run time. Run time is added to idle time or busy time according to the result of polling. Reactor overhead is included into the next thread which calls spdk_thread_poll() now. spdk_thread_poll() saves the end time to the current thread to use it as the start time of the next thread. Unit test framework for this patch and the next patch need to access thread->tsc_last. In the next patch, reactor will use the end time of the current thread to the start time of the next thread in reactor_run() to realize the idea. Hence add an new API spdk_thread_get_last_tsc(). The corresponding variable is named as tsc_last and it is good and is aligned with DPDK (DPDK has used tsc_start and tsc_end as variable name). But last_tsc will be better as API name because the last TSC value is easier to understand. Then add necessary unit test and update the unit test framework. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I5e465e9283c032acb427576d0c90f9e1414f2271 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1048 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Changpeng Liu <changpeng.liu@intel.com> Reviewed-by: Paul Luse <paul.e.luse@intel.com>
2020-03-13 02:53:00 +00:00
static int
poller_run_idle(void *ctx)
{
uint64_t delay_us = (uint64_t)ctx;
spdk_delay_us(delay_us);
return 0;
}
static int
poller_run_busy(void *ctx)
{
uint64_t delay_us = (uint64_t)ctx;
spdk_delay_us(delay_us);
return 1;
}
static void
thread_update_stats_test(void)
lib/thread: spdk_thread_poll() measures run time per call by reading end TSC Update spdk_thread_poll() to count SPDK thread stats correctly on multiple SPDK threads per reactor configuration. spdk_thread_poll() gets start time and reads TSC at end as end time, and then gets delta between them as run time. Run time is added to idle time or busy time according to the result of polling. Reactor overhead is included into the next thread which calls spdk_thread_poll() now. spdk_thread_poll() saves the end time to the current thread to use it as the start time of the next thread. Unit test framework for this patch and the next patch need to access thread->tsc_last. In the next patch, reactor will use the end time of the current thread to the start time of the next thread in reactor_run() to realize the idea. Hence add an new API spdk_thread_get_last_tsc(). The corresponding variable is named as tsc_last and it is good and is aligned with DPDK (DPDK has used tsc_start and tsc_end as variable name). But last_tsc will be better as API name because the last TSC value is easier to understand. Then add necessary unit test and update the unit test framework. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I5e465e9283c032acb427576d0c90f9e1414f2271 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1048 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Changpeng Liu <changpeng.liu@intel.com> Reviewed-by: Paul Luse <paul.e.luse@intel.com>
2020-03-13 02:53:00 +00:00
{
struct spdk_poller *poller;
struct spdk_thread *thread;
MOCK_SET(spdk_get_ticks, 10);
allocate_threads(1);
set_thread(0);
thread = spdk_get_thread();
CU_ASSERT(thread->tsc_last == 10);
CU_ASSERT(thread->stats.idle_tsc == 0);
CU_ASSERT(thread->stats.busy_tsc == 0);
/* Test if idle_tsc is updated expectedly. */
poller = spdk_poller_register(poller_run_idle, (void *)1000, 0);
CU_ASSERT(poller != NULL);
spdk_delay_us(100);
poll_thread_times(0, 1);
CU_ASSERT(thread->tsc_last == 1110);
CU_ASSERT(thread->stats.idle_tsc == 1000);
CU_ASSERT(thread->stats.busy_tsc == 0);
spdk_delay_us(100);
poll_thread_times(0, 1);
CU_ASSERT(thread->tsc_last == 2210);
CU_ASSERT(thread->stats.idle_tsc == 2000);
CU_ASSERT(thread->stats.busy_tsc == 0);
spdk_poller_unregister(&poller);
/* Test if busy_tsc is updated expectedly. */
poller = spdk_poller_register(poller_run_busy, (void *)100000, 0);
CU_ASSERT(poller != NULL);
spdk_delay_us(10000);
poll_thread_times(0, 1);
CU_ASSERT(thread->tsc_last == 112210);
CU_ASSERT(thread->stats.idle_tsc == 2000);
CU_ASSERT(thread->stats.busy_tsc == 100000);
spdk_delay_us(10000);
poll_thread_times(0, 1);
CU_ASSERT(thread->tsc_last == 222210);
CU_ASSERT(thread->stats.idle_tsc == 2000);
CU_ASSERT(thread->stats.busy_tsc == 200000);
spdk_poller_unregister(&poller);
lib/thread: thread_poll() polls until the exiting thread is exited Extract _spdk_thread_exit() from spdk_thread_exit() and _spdk_thread_poll() calls _spdk_thread_exit() if the thread is in the exiting state. spdk_thread_exit() changes to move the state to the exiting state. The spdk_thread_poll() loop will end after the thread moves to the exited state because the caller of spdk_thread_poll() will check if the thread is in the exited state, and break the loop if true. If the user does not call spdk_thread_exit() explicitly, the reactor has to terminate all existing threads at its shutdown. In this case, multiple threads may have some dependency to release I/O channels or unregister pollers. So the reactor has the large two loops, the first loop calls spdk_thread_exit() on all threads, the second loop calls spdk_thread_destroy() if exited or spdk_thread_poll() otherwise for each thread until all threads are destroyed. Besides, change the return value of spdk_thread_exit() to return always 0. Keep it for ABI compatibility. Change ERRLOG to INFOLOG for _spdk_thread_exit() because it is called repeatedly now. Remove the check of I/O reference count from _spdk_thread_exit() because _free_thread() cannot free I/O channel. Refine the unit test accordingly. Fixes issue #1288. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Iee5fb984a96bfac53110fe991dd994ded31dffa4 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1423 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-04-04 23:19:43 +00:00
MOCK_CLEAR(spdk_get_ticks);
lib/thread: spdk_thread_poll() measures run time per call by reading end TSC Update spdk_thread_poll() to count SPDK thread stats correctly on multiple SPDK threads per reactor configuration. spdk_thread_poll() gets start time and reads TSC at end as end time, and then gets delta between them as run time. Run time is added to idle time or busy time according to the result of polling. Reactor overhead is included into the next thread which calls spdk_thread_poll() now. spdk_thread_poll() saves the end time to the current thread to use it as the start time of the next thread. Unit test framework for this patch and the next patch need to access thread->tsc_last. In the next patch, reactor will use the end time of the current thread to the start time of the next thread in reactor_run() to realize the idea. Hence add an new API spdk_thread_get_last_tsc(). The corresponding variable is named as tsc_last and it is good and is aligned with DPDK (DPDK has used tsc_start and tsc_end as variable name). But last_tsc will be better as API name because the last TSC value is easier to understand. Then add necessary unit test and update the unit test framework. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I5e465e9283c032acb427576d0c90f9e1414f2271 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1048 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Changpeng Liu <changpeng.liu@intel.com> Reviewed-by: Paul Luse <paul.e.luse@intel.com>
2020-03-13 02:53:00 +00:00
free_threads();
}
struct ut_nested_ch {
struct spdk_io_channel *child;
struct spdk_poller *poller;
};
struct ut_nested_dev {
struct ut_nested_dev *child;
};
static int
ut_null_poll(void *ctx)
{
return -1;
}
static int
ut_nested_ch_create_cb(void *io_device, void *ctx_buf)
{
struct ut_nested_ch *_ch = ctx_buf;
struct ut_nested_dev *_dev = io_device;
struct ut_nested_dev *_child;
_child = _dev->child;
if (_child != NULL) {
_ch->child = spdk_get_io_channel(_child);
SPDK_CU_ASSERT_FATAL(_ch->child != NULL);
} else {
_ch->child = NULL;
}
_ch->poller = spdk_poller_register(ut_null_poll, NULL, 0);
SPDK_CU_ASSERT_FATAL(_ch->poller != NULL);
return 0;
}
static void
ut_nested_ch_destroy_cb(void *io_device, void *ctx_buf)
{
struct ut_nested_ch *_ch = ctx_buf;
struct spdk_io_channel *child;
child = _ch->child;
if (child != NULL) {
spdk_put_io_channel(child);
}
spdk_poller_unregister(&_ch->poller);
}
static void
ut_check_nested_ch_create(struct spdk_io_channel *ch, struct io_device *dev)
{
CU_ASSERT(ch->ref == 1);
CU_ASSERT(ch->dev == dev);
CU_ASSERT(dev->refcnt == 1);
}
static void
ut_check_nested_ch_destroy_pre(struct spdk_io_channel *ch, struct io_device *dev)
{
CU_ASSERT(ch->ref == 0);
CU_ASSERT(ch->destroy_ref == 1);
CU_ASSERT(dev->refcnt == 1);
}
static void
ut_check_nested_ch_destroy_post(struct io_device *dev)
{
CU_ASSERT(dev->refcnt == 0);
}
static void
ut_check_nested_poller_register(struct spdk_poller *poller)
{
SPDK_CU_ASSERT_FATAL(poller != NULL);
}
static void
nested_channel(void)
{
struct ut_nested_dev _dev1, _dev2, _dev3;
struct ut_nested_ch *_ch1, *_ch2, *_ch3;
struct io_device *dev1, *dev2, *dev3;
struct spdk_io_channel *ch1, *ch2, *ch3;
struct spdk_poller *poller;
struct spdk_thread *thread;
allocate_threads(1);
set_thread(0);
thread = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread != NULL);
_dev1.child = &_dev2;
_dev2.child = &_dev3;
_dev3.child = NULL;
spdk_io_device_register(&_dev1, ut_nested_ch_create_cb, ut_nested_ch_destroy_cb,
sizeof(struct ut_nested_ch), "dev1");
spdk_io_device_register(&_dev2, ut_nested_ch_create_cb, ut_nested_ch_destroy_cb,
sizeof(struct ut_nested_ch), "dev2");
spdk_io_device_register(&_dev3, ut_nested_ch_create_cb, ut_nested_ch_destroy_cb,
sizeof(struct ut_nested_ch), "dev3");
dev1 = io_device_get(&_dev1);
SPDK_CU_ASSERT_FATAL(dev1 != NULL);
dev2 = io_device_get(&_dev2);
SPDK_CU_ASSERT_FATAL(dev2 != NULL);
dev3 = io_device_get(&_dev3);
SPDK_CU_ASSERT_FATAL(dev3 != NULL);
/* A single call spdk_get_io_channel() to dev1 will also create channels
* to dev2 and dev3 continuously. Pollers will be registered together.
*/
ch1 = spdk_get_io_channel(&_dev1);
SPDK_CU_ASSERT_FATAL(ch1 != NULL);
_ch1 = spdk_io_channel_get_ctx(ch1);
ch2 = _ch1->child;
SPDK_CU_ASSERT_FATAL(ch2 != NULL);
_ch2 = spdk_io_channel_get_ctx(ch2);
ch3 = _ch2->child;
SPDK_CU_ASSERT_FATAL(ch3 != NULL);
_ch3 = spdk_io_channel_get_ctx(ch3);
CU_ASSERT(_ch3->child == NULL);
ut_check_nested_ch_create(ch1, dev1);
ut_check_nested_ch_create(ch2, dev2);
ut_check_nested_ch_create(ch3, dev3);
poller = spdk_poller_register(ut_null_poll, NULL, 0);
ut_check_nested_poller_register(poller);
ut_check_nested_poller_register(_ch1->poller);
ut_check_nested_poller_register(_ch2->poller);
ut_check_nested_poller_register(_ch3->poller);
spdk_poller_unregister(&poller);
poll_thread_times(0, 1);
/* A single call spdk_put_io_channel() to dev1 will also destroy channels
* to dev2 and dev3 continuously. Pollers will be unregistered together.
*/
spdk_put_io_channel(ch1);
/* Start exiting the current thread after unregistering the non-nested
* I/O channel.
*/
spdk_thread_exit(thread);
ut_check_nested_ch_destroy_pre(ch1, dev1);
poll_thread_times(0, 1);
ut_check_nested_ch_destroy_post(dev1);
CU_ASSERT(spdk_thread_is_exited(thread) == false);
ut_check_nested_ch_destroy_pre(ch2, dev2);
poll_thread_times(0, 1);
ut_check_nested_ch_destroy_post(dev2);
CU_ASSERT(spdk_thread_is_exited(thread) == false);
ut_check_nested_ch_destroy_pre(ch3, dev3);
poll_thread_times(0, 1);
ut_check_nested_ch_destroy_post(dev3);
CU_ASSERT(spdk_thread_is_exited(thread) == true);
spdk_io_device_unregister(&_dev1, NULL);
spdk_io_device_unregister(&_dev2, NULL);
spdk_io_device_unregister(&_dev3, NULL);
CU_ASSERT(RB_EMPTY(&g_io_devices));
free_threads();
CU_ASSERT(TAILQ_EMPTY(&g_threads));
}
static int
create_cb2(void *io_device, void *ctx_buf)
{
uint64_t *devcnt = (uint64_t *)io_device;
*devcnt += 1;
return 0;
}
static void
destroy_cb2(void *io_device, void *ctx_buf)
{
uint64_t *devcnt = (uint64_t *)io_device;
CU_ASSERT(*devcnt > 0);
*devcnt -= 1;
}
static void
unregister_cb2(void *io_device)
{
uint64_t *devcnt = (uint64_t *)io_device;
CU_ASSERT(*devcnt == 0);
}
static void
device_unregister_and_thread_exit_race(void)
{
uint64_t device = 0;
struct spdk_io_channel *ch1, *ch2;
struct spdk_thread *thread1, *thread2;
/* Create two threads and each thread gets a channel from the same device. */
allocate_threads(2);
set_thread(0);
thread1 = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread1 != NULL);
spdk_io_device_register(&device, create_cb2, destroy_cb2, sizeof(uint64_t), NULL);
ch1 = spdk_get_io_channel(&device);
SPDK_CU_ASSERT_FATAL(ch1 != NULL);
set_thread(1);
thread2 = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread2 != NULL);
ch2 = spdk_get_io_channel(&device);
SPDK_CU_ASSERT_FATAL(ch2 != NULL);
set_thread(0);
/* Move thread 0 to the exiting state, but it should keep exiting until two channels
* and a device are released.
*/
spdk_thread_exit(thread1);
poll_thread(0);
spdk_put_io_channel(ch1);
spdk_io_device_unregister(&device, unregister_cb2);
poll_thread(0);
CU_ASSERT(spdk_thread_is_exited(thread1) == false);
set_thread(1);
/* Move thread 1 to the exiting state, but it should keep exiting until its channel
* is released.
*/
spdk_thread_exit(thread2);
poll_thread(1);
CU_ASSERT(spdk_thread_is_exited(thread2) == false);
spdk_put_io_channel(ch2);
poll_thread(1);
CU_ASSERT(spdk_thread_is_exited(thread1) == false);
CU_ASSERT(spdk_thread_is_exited(thread2) == true);
poll_thread(0);
CU_ASSERT(spdk_thread_is_exited(thread1) == true);
free_threads();
}
static int
dummy_poller(void *arg)
{
return SPDK_POLLER_IDLE;
}
static void
cache_closest_timed_poller(void)
{
struct spdk_thread *thread;
struct spdk_poller *poller1, *poller2, *poller3, *tmp;
allocate_threads(1);
set_thread(0);
thread = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread != NULL);
poller1 = spdk_poller_register(dummy_poller, NULL, 1000);
SPDK_CU_ASSERT_FATAL(poller1 != NULL);
poller2 = spdk_poller_register(dummy_poller, NULL, 1500);
SPDK_CU_ASSERT_FATAL(poller2 != NULL);
poller3 = spdk_poller_register(dummy_poller, NULL, 1800);
SPDK_CU_ASSERT_FATAL(poller3 != NULL);
poll_threads();
/* When multiple timed pollers are inserted, the cache should
* have the closest timed poller.
*/
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == poller1);
spdk_delay_us(1000);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller2);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == poller2);
/* If we unregister a timed poller by spdk_poller_unregister()
* when it is waiting, it is marked as being unregistered and
* is actually unregistered when it is expired.
*
* Hence if we unregister the closest timed poller when it is waiting,
* the cache is not updated to the next timed poller until it is expired.
*/
tmp = poller2;
spdk_poller_unregister(&poller2);
CU_ASSERT(poller2 == NULL);
spdk_delay_us(499);
poll_threads();
CU_ASSERT(thread->first_timed_poller == tmp);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == tmp);
spdk_delay_us(1);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller3);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == poller3);
/* If we pause a timed poller by spdk_poller_pause() when it is waiting,
* it is marked as being paused and is actually paused when it is expired.
*
* Hence if we pause the closest timed poller when it is waiting, the cache
* is not updated to the next timed poller until it is expired.
*/
spdk_poller_pause(poller3);
spdk_delay_us(299);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller3);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == poller3);
spdk_delay_us(1);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(RB_MIN(timed_pollers_tree, &thread->timed_pollers) == poller1);
/* After unregistering all timed pollers, the cache should
* be NULL.
*/
spdk_poller_unregister(&poller1);
spdk_poller_unregister(&poller3);
spdk_delay_us(200);
poll_threads();
CU_ASSERT(thread->first_timed_poller == NULL);
CU_ASSERT(RB_EMPTY(&thread->timed_pollers));
free_threads();
}
static void
multi_timed_pollers_have_same_expiration(void)
{
struct spdk_thread *thread;
struct spdk_poller *poller1, *poller2, *poller3, *poller4, *tmp;
uint64_t start_ticks;
allocate_threads(1);
set_thread(0);
thread = spdk_get_thread();
SPDK_CU_ASSERT_FATAL(thread != NULL);
/*
* case 1: multiple timed pollers have the same next_run_tick.
*/
start_ticks = spdk_get_ticks();
poller1 = spdk_poller_register(dummy_poller, NULL, 500);
SPDK_CU_ASSERT_FATAL(poller1 != NULL);
poller2 = spdk_poller_register(dummy_poller, NULL, 500);
SPDK_CU_ASSERT_FATAL(poller2 != NULL);
poller3 = spdk_poller_register(dummy_poller, NULL, 1000);
SPDK_CU_ASSERT_FATAL(poller3 != NULL);
poller4 = spdk_poller_register(dummy_poller, NULL, 1500);
SPDK_CU_ASSERT_FATAL(poller4 != NULL);
/* poller1 and poller2 have the same next_run_tick but cache has poller1
* because poller1 is registered earlier than poller2.
*/
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(poller1->next_run_tick == start_ticks + 500);
CU_ASSERT(poller2->next_run_tick == start_ticks + 500);
CU_ASSERT(poller3->next_run_tick == start_ticks + 1000);
CU_ASSERT(poller4->next_run_tick == start_ticks + 1500);
/* after 500 usec, poller1 and poller2 are expired. */
spdk_delay_us(500);
CU_ASSERT(spdk_get_ticks() == start_ticks + 500);
poll_threads();
/* poller1, poller2, and poller3 have the same next_run_tick but cache
* has poller3 because poller3 is not expired yet.
*/
CU_ASSERT(thread->first_timed_poller == poller3);
CU_ASSERT(poller1->next_run_tick == start_ticks + 1000);
CU_ASSERT(poller2->next_run_tick == start_ticks + 1000);
CU_ASSERT(poller3->next_run_tick == start_ticks + 1000);
CU_ASSERT(poller4->next_run_tick == start_ticks + 1500);
/* after 500 usec, poller1, poller2, and poller3 are expired. */
spdk_delay_us(500);
CU_ASSERT(spdk_get_ticks() == start_ticks + 1000);
poll_threads();
/* poller1, poller2, and poller4 have the same next_run_tick but cache
* has poller4 because poller4 is not expired yet.
*/
CU_ASSERT(thread->first_timed_poller == poller4);
CU_ASSERT(poller1->next_run_tick == start_ticks + 1500);
CU_ASSERT(poller2->next_run_tick == start_ticks + 1500);
CU_ASSERT(poller3->next_run_tick == start_ticks + 2000);
CU_ASSERT(poller4->next_run_tick == start_ticks + 1500);
/* after 500 usec, poller1, poller2, and poller4 are expired. */
spdk_delay_us(500);
CU_ASSERT(spdk_get_ticks() == start_ticks + 1500);
poll_threads();
/* poller1, poller2, and poller3 have the same next_run_tick but cache
* has poller3 because poller3 is updated earlier than poller1 and poller2.
*/
CU_ASSERT(thread->first_timed_poller == poller3);
CU_ASSERT(poller1->next_run_tick == start_ticks + 2000);
CU_ASSERT(poller2->next_run_tick == start_ticks + 2000);
CU_ASSERT(poller3->next_run_tick == start_ticks + 2000);
CU_ASSERT(poller4->next_run_tick == start_ticks + 3000);
spdk_poller_unregister(&poller1);
spdk_poller_unregister(&poller2);
spdk_poller_unregister(&poller3);
spdk_poller_unregister(&poller4);
spdk_delay_us(1500);
CU_ASSERT(spdk_get_ticks() == start_ticks + 3000);
poll_threads();
CU_ASSERT(thread->first_timed_poller == NULL);
CU_ASSERT(RB_EMPTY(&thread->timed_pollers));
/*
* case 2: unregister timed pollers while multiple timed pollers are registered.
*/
start_ticks = spdk_get_ticks();
poller1 = spdk_poller_register(dummy_poller, NULL, 500);
SPDK_CU_ASSERT_FATAL(poller1 != NULL);
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(poller1->next_run_tick == start_ticks + 500);
/* after 250 usec, register poller2 and poller3. */
spdk_delay_us(250);
CU_ASSERT(spdk_get_ticks() == start_ticks + 250);
poller2 = spdk_poller_register(dummy_poller, NULL, 500);
SPDK_CU_ASSERT_FATAL(poller2 != NULL);
poller3 = spdk_poller_register(dummy_poller, NULL, 750);
SPDK_CU_ASSERT_FATAL(poller3 != NULL);
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(poller1->next_run_tick == start_ticks + 500);
CU_ASSERT(poller2->next_run_tick == start_ticks + 750);
CU_ASSERT(poller3->next_run_tick == start_ticks + 1000);
/* unregister poller2 which is not the closest. */
tmp = poller2;
spdk_poller_unregister(&poller2);
/* after 250 usec, poller1 is expired. */
spdk_delay_us(250);
CU_ASSERT(spdk_get_ticks() == start_ticks + 500);
poll_threads();
/* poller2 is not unregistered yet because it is not expired. */
CU_ASSERT(thread->first_timed_poller == tmp);
CU_ASSERT(poller1->next_run_tick == start_ticks + 1000);
CU_ASSERT(tmp->next_run_tick == start_ticks + 750);
CU_ASSERT(poller3->next_run_tick == start_ticks + 1000);
spdk_delay_us(250);
CU_ASSERT(spdk_get_ticks() == start_ticks + 750);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller3);
CU_ASSERT(poller1->next_run_tick == start_ticks + 1000);
CU_ASSERT(poller3->next_run_tick == start_ticks + 1000);
spdk_poller_unregister(&poller3);
spdk_delay_us(250);
CU_ASSERT(spdk_get_ticks() == start_ticks + 1000);
poll_threads();
CU_ASSERT(thread->first_timed_poller == poller1);
CU_ASSERT(poller1->next_run_tick == start_ticks + 1500);
spdk_poller_unregister(&poller1);
spdk_delay_us(500);
CU_ASSERT(spdk_get_ticks() == start_ticks + 1500);
poll_threads();
CU_ASSERT(thread->first_timed_poller == NULL);
CU_ASSERT(RB_EMPTY(&thread->timed_pollers));
free_threads();
}
static int
dummy_create_cb(void *io_device, void *ctx_buf)
{
return 0;
}
static void
dummy_destroy_cb(void *io_device, void *ctx_buf)
{
}
/* We had a bug that the compare function for the io_device tree
* did not work as expected because subtraction caused overflow
* when the difference between two keys was more than 32 bits.
* This test case verifies the fix for the bug.
*/
static void
io_device_lookup(void)
{
struct io_device dev1, dev2, *dev;
struct spdk_io_channel *ch;
/* The compare function io_device_cmp() had a overflow bug.
* Verify the fix first.
*/
dev1.io_device = (void *)0x7FFFFFFF;
dev2.io_device = NULL;
CU_ASSERT(io_device_cmp(&dev1, &dev2) > 0);
CU_ASSERT(io_device_cmp(&dev2, &dev1) < 0);
/* Check if overflow due to 32 bits does not occur. */
dev1.io_device = (void *)0x80000000;
CU_ASSERT(io_device_cmp(&dev1, &dev2) > 0);
CU_ASSERT(io_device_cmp(&dev2, &dev1) < 0);
dev1.io_device = (void *)0x100000000;
CU_ASSERT(io_device_cmp(&dev1, &dev2) > 0);
CU_ASSERT(io_device_cmp(&dev2, &dev1) < 0);
dev1.io_device = (void *)0x8000000000000000;
CU_ASSERT(io_device_cmp(&dev1, &dev2) > 0);
CU_ASSERT(io_device_cmp(&dev2, &dev1) < 0);
allocate_threads(1);
set_thread(0);
spdk_io_device_register((void *)0x1, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)0x7FFFFFFF, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)0x80000000, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)0x100000000, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)0x8000000000000000, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)0x8000000100000000, dummy_create_cb, dummy_destroy_cb, 0, NULL);
spdk_io_device_register((void *)UINT64_MAX, dummy_create_cb, dummy_destroy_cb, 0, NULL);
/* RB_MIN and RB_NEXT should return devs in ascending order by addresses.
* RB_FOREACH uses RB_MIN and RB_NEXT internally.
*/
dev = RB_MIN(io_device_tree, &g_io_devices);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x1);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x7FFFFFFF);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x80000000);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x100000000);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x8000000000000000);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)0x8000000100000000);
dev = RB_NEXT(io_device_tree, &g_io_devices, dev);
SPDK_CU_ASSERT_FATAL(dev != NULL);
CU_ASSERT(dev->io_device == (void *)UINT64_MAX);
/* Verify spdk_get_io_channel() creates io_channels associated with the
* correct io_devices.
*/
ch = spdk_get_io_channel((void *)0x1);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x1);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)0x7FFFFFFF);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x7FFFFFFF);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)0x80000000);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x80000000);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)0x100000000);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x100000000);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)0x8000000000000000);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x8000000000000000);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)0x8000000100000000);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)0x8000000100000000);
spdk_put_io_channel(ch);
ch = spdk_get_io_channel((void *)UINT64_MAX);
SPDK_CU_ASSERT_FATAL(ch != NULL);
CU_ASSERT(ch->dev->io_device == (void *)UINT64_MAX);
spdk_put_io_channel(ch);
poll_threads();
spdk_io_device_unregister((void *)0x1, NULL);
spdk_io_device_unregister((void *)0x7FFFFFFF, NULL);
spdk_io_device_unregister((void *)0x80000000, NULL);
spdk_io_device_unregister((void *)0x100000000, NULL);
spdk_io_device_unregister((void *)0x8000000000000000, NULL);
spdk_io_device_unregister((void *)0x8000000100000000, NULL);
spdk_io_device_unregister((void *)UINT64_MAX, NULL);
poll_threads();
CU_ASSERT(RB_EMPTY(&g_io_devices));
free_threads();
}
static enum spin_error g_spin_err;
static uint32_t g_spin_err_count = 0;
static void
ut_track_abort(enum spin_error err)
{
g_spin_err = err;
g_spin_err_count++;
}
static void
spdk_spin(void)
{
struct spdk_spinlock lock;
g_spin_abort_fn = ut_track_abort;
/* Do not need to be on an SPDK thread to initialize an spdk_spinlock */
g_spin_err_count = 0;
spdk_spin_init(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Trying to take a lock while not on an SPDK thread is an error */
g_spin_err_count = 0;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_NOT_SPDK_THREAD);
/* Trying to check if a lock is held while not on an SPDK thread is an error */
g_spin_err_count = 0;
spdk_spin_held(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_NOT_SPDK_THREAD);
/* Do not need to be on an SPDK thread to destroy an spdk_spinlock */
g_spin_err_count = 0;
spdk_spin_destroy(&lock);
CU_ASSERT(g_spin_err_count == 0);
allocate_threads(2);
set_thread(0);
/* Can initialize an spdk_spinlock on an SPDK thread */
g_spin_err_count = 0;
spdk_spin_init(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Can take spinlock */
g_spin_err_count = 0;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Can release spinlock */
g_spin_err_count = 0;
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Deadlock detected */
g_spin_err_count = 0;
g_spin_err = SPIN_ERR_NONE;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 0);
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_DEADLOCK);
/* Cannot unlock from wrong thread */
set_thread(1);
g_spin_err_count = 0;
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_WRONG_THREAD);
/* Get back to a known good state */
set_thread(0);
g_spin_err_count = 0;
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Cannot release the same lock twice */
g_spin_err_count = 0;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 0);
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 0);
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_WRONG_THREAD);
/* A lock that is not held is properly recognized */
g_spin_err_count = 0;
CU_ASSERT(!spdk_spin_held(&lock));
CU_ASSERT(g_spin_err_count == 0);
/* A lock that is held is recognized as held by only the thread that holds it. */
set_thread(1);
g_spin_err_count = 0;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 0);
CU_ASSERT(spdk_spin_held(&lock));
CU_ASSERT(g_spin_err_count == 0);
set_thread(0);
CU_ASSERT(!spdk_spin_held(&lock));
CU_ASSERT(g_spin_err_count == 0);
/* After releasing, no one thinks it is held */
set_thread(1);
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 0);
CU_ASSERT(!spdk_spin_held(&lock));
CU_ASSERT(g_spin_err_count == 0);
set_thread(0);
CU_ASSERT(!spdk_spin_held(&lock));
CU_ASSERT(g_spin_err_count == 0);
/* Destroying a lock that is held is an error. */
set_thread(0);
g_spin_err_count = 0;
spdk_spin_lock(&lock);
CU_ASSERT(g_spin_err_count == 0);
spdk_spin_destroy(&lock);
CU_ASSERT(g_spin_err_count == 1);
CU_ASSERT(g_spin_err == SPIN_ERR_LOCK_HELD);
g_spin_err_count = 0;
spdk_spin_unlock(&lock);
CU_ASSERT(g_spin_err_count == 0);
/* Clean up */
g_spin_err_count = 0;
spdk_spin_destroy(&lock);
CU_ASSERT(g_spin_err_count == 0);
free_threads();
g_spin_abort_fn = __posix_abort;
}
thread: introduce iobuf buffer pools The idea behind "iobuf" is to have a single place for allocating data buffers across different libraries. That way, each library won't need to allocate its own mempools, therefore decreasing the memory footprint of the whole application. There are two reasons for putting these kind of functions in the thread library. Firstly, the code is pretty small, so it doesn't make sense to create a new library. Secondly, it relies on the IO channel abstraction, so users will need to pull in the thread library anyway. It's very much inspired by the way bdev layer handles data buffers (much of the code was directly copied over). There are two global mempools, one for small and one for large buffers, and per-thread queues that hold requests waiting for a buffer. The main difference is that we also need to track which module requested a buffer in order to allow users to iterate over its pending requests. The usage is fairly simple: ``` /* Embed spdk_iobuf_channel into an existing IO channel */ struct foo_channel { ... struct spdk_iobuf_channel iobuf; }; /* Embed spdk_iobuf_entry into objects that will request buffers */ struct foo_object { ... struct spdk_iobuf_entry entry; }; /* Register the module as iobuf user */ spdk_iobuf_register_module("foo"); /* Initialize iobuf channel in foo_channel's create cb */ spdk_iobuf_channel_init(&foo_channel->iobuf, "foo", 0, 0); /* Finally, request a buffer... */ buf = spdk_iobuf_get(&foo_channel->iobuf, length, &foo_objet.entry, buf_get_cb); ... /* ...and release it */ spdk_iobuf_put(&foo_channel->iobuf, buf, length); ``` Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com> Change-Id: Ifaa6934c03ed6587ddba972198e606921bd85008 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/15326 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Community-CI: Mellanox Build Bot Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com> Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2022-10-26 15:21:03 +00:00
struct ut_iobuf_entry {
struct spdk_iobuf_channel *ioch;
struct spdk_iobuf_entry iobuf;
void *buf;
uint32_t thread_id;
const char *module;
};
static void
ut_iobuf_finish_cb(void *ctx)
{
*(int *)ctx = 1;
}
static void
ut_iobuf_get_buf_cb(struct spdk_iobuf_entry *entry, void *buf)
{
struct ut_iobuf_entry *ut_entry = SPDK_CONTAINEROF(entry, struct ut_iobuf_entry, iobuf);
ut_entry->buf = buf;
}
static int
ut_iobuf_foreach_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_arg)
{
struct ut_iobuf_entry *ut_entry = SPDK_CONTAINEROF(entry, struct ut_iobuf_entry, iobuf);
ut_entry->buf = cb_arg;
return 0;
}
static void
iobuf(void)
{
struct spdk_iobuf_opts opts = {
.small_pool_count = 2,
.large_pool_count = 2,
.small_bufsize = 2,
.large_bufsize = 4,
};
struct ut_iobuf_entry *entry;
struct spdk_iobuf_channel mod0_ch[2], mod1_ch[2];
struct ut_iobuf_entry mod0_entries[] = {
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 1, .module = "ut_module0", },
{ .thread_id = 1, .module = "ut_module0", },
{ .thread_id = 1, .module = "ut_module0", },
{ .thread_id = 1, .module = "ut_module0", },
};
struct ut_iobuf_entry mod1_entries[] = {
{ .thread_id = 0, .module = "ut_module1", },
{ .thread_id = 0, .module = "ut_module1", },
{ .thread_id = 0, .module = "ut_module1", },
{ .thread_id = 0, .module = "ut_module1", },
{ .thread_id = 1, .module = "ut_module1", },
{ .thread_id = 1, .module = "ut_module1", },
{ .thread_id = 1, .module = "ut_module1", },
{ .thread_id = 1, .module = "ut_module1", },
};
int rc, finish = 0;
uint32_t i;
allocate_cores(2);
allocate_threads(2);
set_thread(0);
/* We cannot use spdk_iobuf_set_opts(), as it won't allow us to use such small pools */
g_iobuf.opts = opts;
rc = spdk_iobuf_initialize();
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_register_module("ut_module0");
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_register_module("ut_module1");
CU_ASSERT_EQUAL(rc, 0);
set_thread(0);
rc = spdk_iobuf_channel_init(&mod0_ch[0], "ut_module0", 0, 0);
CU_ASSERT_EQUAL(rc, 0);
set_thread(1);
rc = spdk_iobuf_channel_init(&mod0_ch[1], "ut_module0", 0, 0);
CU_ASSERT_EQUAL(rc, 0);
for (i = 0; i < SPDK_COUNTOF(mod0_entries); ++i) {
mod0_entries[i].ioch = &mod0_ch[mod0_entries[i].thread_id];
}
set_thread(0);
rc = spdk_iobuf_channel_init(&mod1_ch[0], "ut_module1", 0, 0);
CU_ASSERT_EQUAL(rc, 0);
set_thread(1);
rc = spdk_iobuf_channel_init(&mod1_ch[1], "ut_module1", 0, 0);
CU_ASSERT_EQUAL(rc, 0);
for (i = 0; i < SPDK_COUNTOF(mod1_entries); ++i) {
mod1_entries[i].ioch = &mod1_ch[mod1_entries[i].thread_id];
}
/* First check that it's possible to retrieve the whole pools from a single module */
set_thread(0);
entry = &mod0_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* The next two should be put onto the large buf wait queue */
entry = &mod0_entries[2];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Pick the two next buffers from the small pool */
set_thread(1);
entry = &mod0_entries[4];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[5];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* The next two should be put onto the small buf wait queue */
entry = &mod0_entries[6];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[7];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Now return one of the large buffers to the pool and verify that the first request's
* (entry 2) callback was executed and it was removed from the wait queue.
*/
set_thread(0);
entry = &mod0_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[2];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[3];
CU_ASSERT_PTR_NULL(entry->buf);
/* Return the second buffer and check that the other request is satisfied */
entry = &mod0_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[3];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Return the remaining two buffers */
entry = &mod0_entries[2];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[3];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
/* Check that it didn't change the requests waiting for the small buffers */
entry = &mod0_entries[6];
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[7];
CU_ASSERT_PTR_NULL(entry->buf);
/* Do the same test as above, this time using the small pool */
set_thread(1);
entry = &mod0_entries[4];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod0_entries[6];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[7];
CU_ASSERT_PTR_NULL(entry->buf);
/* Return the second buffer and check that the other request is satisfied */
entry = &mod0_entries[5];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod0_entries[7];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Return the remaining two buffers */
entry = &mod0_entries[6];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod0_entries[7];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
/* Now check requesting buffers from different modules - first request all of them from one
* module, starting from the large pool
*/
set_thread(0);
entry = &mod0_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Request all of them from the small one */
set_thread(1);
entry = &mod0_entries[4];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[5];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Request one buffer per module from each pool */
set_thread(0);
entry = &mod1_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Change the order from the small pool and request a buffer from mod0 first */
set_thread(1);
entry = &mod0_entries[6];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[4];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Now return one buffer to the large pool */
set_thread(0);
entry = &mod0_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
/* Make sure the request from mod1 got the buffer, as it was the first to request it */
entry = &mod1_entries[0];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[3];
CU_ASSERT_PTR_NULL(entry->buf);
/* Return second buffer to the large pool and check the outstanding mod0 request */
entry = &mod0_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[3];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Return the remaining two buffers */
entry = &mod1_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[3];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
/* Check the same for the small pool, but this time the order of the request is reversed
* (mod0 before mod1)
*/
set_thread(1);
entry = &mod0_entries[4];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod0_entries[6];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* mod1 request was second in this case, so it still needs to wait */
entry = &mod1_entries[4];
CU_ASSERT_PTR_NULL(entry->buf);
/* Return the second requested buffer */
entry = &mod0_entries[5];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod1_entries[4];
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Return the remaining two buffers */
entry = &mod0_entries[6];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod1_entries[4];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
/* Request buffers to make the pools empty */
set_thread(0);
entry = &mod0_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod1_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod1_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Queue more requests from both modules */
entry = &mod0_entries[2];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[2];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Check that abort correctly remove an entry from the queue */
entry = &mod0_entries[2];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 4);
entry = &mod1_entries[3];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 2);
entry = &mod0_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
CU_ASSERT_PTR_NOT_NULL(mod1_entries[2].buf);
entry = &mod0_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
CU_ASSERT_PTR_NOT_NULL(mod0_entries[3].buf);
/* Clean up */
entry = &mod1_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod1_entries[2];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod1_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod0_entries[3];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
/* Request buffers to make the pools empty */
set_thread(0);
entry = &mod0_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod1_entries[0];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod0_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
entry = &mod1_entries[1];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
/* Request a buffer from each queue and each module on thread 0 */
set_thread(0);
entry = &mod0_entries[2];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[2];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[3];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Do the same on thread 1 */
set_thread(1);
entry = &mod0_entries[6];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[6];
entry->buf = spdk_iobuf_get(entry->ioch, 4, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod0_entries[7];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
entry = &mod1_entries[7];
entry->buf = spdk_iobuf_get(entry->ioch, 2, &entry->iobuf, ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(entry->buf);
/* Now do the foreach and check that correct entries are iterated over by assigning their
* ->buf pointers to different values.
*/
set_thread(0);
rc = spdk_iobuf_for_each_entry(&mod0_ch[0], &mod0_ch[0].large,
ut_iobuf_foreach_cb, (void *)0xdeadbeef);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod0_ch[0], &mod0_ch[0].small,
ut_iobuf_foreach_cb, (void *)0xbeefdead);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod1_ch[0], &mod1_ch[0].large,
ut_iobuf_foreach_cb, (void *)0xfeedbeef);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod1_ch[0], &mod1_ch[0].small,
ut_iobuf_foreach_cb, (void *)0xbeeffeed);
CU_ASSERT_EQUAL(rc, 0);
set_thread(1);
rc = spdk_iobuf_for_each_entry(&mod0_ch[1], &mod0_ch[1].large,
ut_iobuf_foreach_cb, (void *)0xcafebabe);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod0_ch[1], &mod0_ch[1].small,
ut_iobuf_foreach_cb, (void *)0xbabecafe);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod1_ch[1], &mod1_ch[1].large,
ut_iobuf_foreach_cb, (void *)0xbeefcafe);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_for_each_entry(&mod1_ch[1], &mod1_ch[1].small,
ut_iobuf_foreach_cb, (void *)0xcafebeef);
CU_ASSERT_EQUAL(rc, 0);
/* thread 0 */
CU_ASSERT_PTR_EQUAL(mod0_entries[2].buf, (void *)0xdeadbeef);
CU_ASSERT_PTR_EQUAL(mod0_entries[3].buf, (void *)0xbeefdead);
CU_ASSERT_PTR_EQUAL(mod1_entries[2].buf, (void *)0xfeedbeef);
CU_ASSERT_PTR_EQUAL(mod1_entries[3].buf, (void *)0xbeeffeed);
/* thread 1 */
CU_ASSERT_PTR_EQUAL(mod0_entries[6].buf, (void *)0xcafebabe);
CU_ASSERT_PTR_EQUAL(mod0_entries[7].buf, (void *)0xbabecafe);
CU_ASSERT_PTR_EQUAL(mod1_entries[6].buf, (void *)0xbeefcafe);
CU_ASSERT_PTR_EQUAL(mod1_entries[7].buf, (void *)0xcafebeef);
/* Clean everything up */
set_thread(0);
entry = &mod0_entries[2];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 4);
entry = &mod0_entries[3];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 2);
entry = &mod1_entries[2];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 4);
entry = &mod1_entries[3];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 2);
entry = &mod0_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod1_entries[0];
spdk_iobuf_put(entry->ioch, entry->buf, 4);
entry = &mod0_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
entry = &mod1_entries[1];
spdk_iobuf_put(entry->ioch, entry->buf, 2);
set_thread(1);
entry = &mod0_entries[6];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 4);
entry = &mod0_entries[7];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 2);
entry = &mod1_entries[6];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 4);
entry = &mod1_entries[7];
spdk_iobuf_entry_abort(entry->ioch, &entry->iobuf, 2);
set_thread(0);
spdk_iobuf_channel_fini(&mod0_ch[0]);
poll_threads();
spdk_iobuf_channel_fini(&mod1_ch[0]);
poll_threads();
set_thread(1);
spdk_iobuf_channel_fini(&mod0_ch[1]);
poll_threads();
spdk_iobuf_channel_fini(&mod1_ch[1]);
poll_threads();
spdk_iobuf_finish(ut_iobuf_finish_cb, &finish);
poll_threads();
CU_ASSERT_EQUAL(finish, 1);
free_threads();
free_cores();
}
static void
iobuf_cache(void)
{
struct spdk_iobuf_opts opts = {
.small_pool_count = 4,
.large_pool_count = 4,
.small_bufsize = 2,
.large_bufsize = 4,
};
struct spdk_iobuf_channel iobuf_ch[2];
struct ut_iobuf_entry *entry;
struct ut_iobuf_entry mod0_entries[] = {
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
{ .thread_id = 0, .module = "ut_module0", },
};
struct ut_iobuf_entry mod1_entries[] = {
{ .thread_id = 0, .module = "ut_module1", },
{ .thread_id = 0, .module = "ut_module1", },
};
int rc, finish = 0;
uint32_t i, j, bufsize;
allocate_cores(1);
allocate_threads(1);
set_thread(0);
/* We cannot use spdk_iobuf_set_opts(), as it won't allow us to use such small pools */
g_iobuf.opts = opts;
rc = spdk_iobuf_initialize();
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_register_module("ut_module0");
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_register_module("ut_module1");
CU_ASSERT_EQUAL(rc, 0);
/* First check that channel initialization fails when it's not possible to fill in the cache
* from the pool.
*/
rc = spdk_iobuf_channel_init(&iobuf_ch[0], "ut_module0", 5, 1);
CU_ASSERT_EQUAL(rc, -ENOMEM);
rc = spdk_iobuf_channel_init(&iobuf_ch[0], "ut_module0", 1, 5);
CU_ASSERT_EQUAL(rc, -ENOMEM);
rc = spdk_iobuf_channel_init(&iobuf_ch[0], "ut_module0", 4, 4);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 4, 4);
CU_ASSERT_EQUAL(rc, -ENOMEM);
spdk_iobuf_channel_fini(&iobuf_ch[0]);
poll_threads();
/* Initialize one channel with cache, acquire buffers, and check that a second one can be
* created once the buffers acquired from the first one are returned to the pool
*/
rc = spdk_iobuf_channel_init(&iobuf_ch[0], "ut_module0", 2, 2);
CU_ASSERT_EQUAL(rc, 0);
for (i = 0; i < 3; ++i) {
mod0_entries[i].buf = spdk_iobuf_get(&iobuf_ch[0], 4, &mod0_entries[i].iobuf,
ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(mod0_entries[i].buf);
}
/* It should be able to create a channel with a single entry in the cache */
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 2, 1);
CU_ASSERT_EQUAL(rc, 0);
spdk_iobuf_channel_fini(&iobuf_ch[1]);
poll_threads();
/* But not with two entries */
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 2, 2);
CU_ASSERT_EQUAL(rc, -ENOMEM);
for (i = 0; i < 2; ++i) {
spdk_iobuf_put(&iobuf_ch[0], mod0_entries[i].buf, 4);
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 2, 2);
CU_ASSERT_EQUAL(rc, -ENOMEM);
}
spdk_iobuf_put(&iobuf_ch[0], mod0_entries[2].buf, 4);
/* The last buffer should be released back to the pool, so we should be able to create a new
* channel
*/
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 2, 2);
CU_ASSERT_EQUAL(rc, 0);
spdk_iobuf_channel_fini(&iobuf_ch[0]);
spdk_iobuf_channel_fini(&iobuf_ch[1]);
poll_threads();
/* Check that the pool is only used when the cache is empty and that the cache guarantees a
* certain set of buffers
*/
rc = spdk_iobuf_channel_init(&iobuf_ch[0], "ut_module0", 2, 2);
CU_ASSERT_EQUAL(rc, 0);
rc = spdk_iobuf_channel_init(&iobuf_ch[1], "ut_module1", 1, 1);
CU_ASSERT_EQUAL(rc, 0);
uint32_t buffer_sizes[] = { 2, 4 };
for (i = 0; i < SPDK_COUNTOF(buffer_sizes); ++i) {
bufsize = buffer_sizes[i];
for (j = 0; j < 3; ++j) {
entry = &mod0_entries[j];
entry->buf = spdk_iobuf_get(&iobuf_ch[0], bufsize, &entry->iobuf,
ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(entry->buf);
}
mod1_entries[0].buf = spdk_iobuf_get(&iobuf_ch[1], bufsize, &mod1_entries[0].iobuf,
ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NOT_NULL(mod1_entries[0].buf);
/* The whole pool is exhausted now */
mod1_entries[1].buf = spdk_iobuf_get(&iobuf_ch[1], bufsize, &mod1_entries[1].iobuf,
ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(mod1_entries[1].buf);
mod0_entries[3].buf = spdk_iobuf_get(&iobuf_ch[0], bufsize, &mod0_entries[3].iobuf,
ut_iobuf_get_buf_cb);
CU_ASSERT_PTR_NULL(mod0_entries[3].buf);
/* If there are outstanding requests waiting for a buffer, they should have priority
* over filling in the cache, even if they're from different modules.
*/
spdk_iobuf_put(&iobuf_ch[0], mod0_entries[2].buf, bufsize);
/* Also make sure the queue is FIFO and doesn't care about which module requested
* and which module released the buffer.
*/
CU_ASSERT_PTR_NOT_NULL(mod1_entries[1].buf);
CU_ASSERT_PTR_NULL(mod0_entries[3].buf);
/* Return the buffers back */
spdk_iobuf_entry_abort(&iobuf_ch[0], &mod0_entries[3].iobuf, bufsize);
for (j = 0; j < 2; ++j) {
spdk_iobuf_put(&iobuf_ch[0], mod0_entries[j].buf, bufsize);
spdk_iobuf_put(&iobuf_ch[1], mod1_entries[j].buf, bufsize);
}
}
spdk_iobuf_channel_fini(&iobuf_ch[0]);
spdk_iobuf_channel_fini(&iobuf_ch[1]);
poll_threads();
spdk_iobuf_finish(ut_iobuf_finish_cb, &finish);
poll_threads();
CU_ASSERT_EQUAL(finish, 1);
free_threads();
free_cores();
}
int
main(int argc, char **argv)
{
CU_pSuite suite = NULL;
unsigned int num_failures;
CU_set_error_action(CUEA_ABORT);
CU_initialize_registry();
suite = CU_add_suite("io_channel", NULL, NULL);
CU_ADD_TEST(suite, thread_alloc);
CU_ADD_TEST(suite, thread_send_msg);
CU_ADD_TEST(suite, thread_poller);
CU_ADD_TEST(suite, poller_pause);
CU_ADD_TEST(suite, thread_for_each);
CU_ADD_TEST(suite, for_each_channel_remove);
CU_ADD_TEST(suite, for_each_channel_unreg);
CU_ADD_TEST(suite, thread_name);
CU_ADD_TEST(suite, channel);
CU_ADD_TEST(suite, channel_destroy_races);
CU_ADD_TEST(suite, thread_exit_test);
CU_ADD_TEST(suite, thread_update_stats_test);
CU_ADD_TEST(suite, nested_channel);
CU_ADD_TEST(suite, device_unregister_and_thread_exit_race);
CU_ADD_TEST(suite, cache_closest_timed_poller);
CU_ADD_TEST(suite, multi_timed_pollers_have_same_expiration);
CU_ADD_TEST(suite, io_device_lookup);
CU_ADD_TEST(suite, spdk_spin);
thread: introduce iobuf buffer pools The idea behind "iobuf" is to have a single place for allocating data buffers across different libraries. That way, each library won't need to allocate its own mempools, therefore decreasing the memory footprint of the whole application. There are two reasons for putting these kind of functions in the thread library. Firstly, the code is pretty small, so it doesn't make sense to create a new library. Secondly, it relies on the IO channel abstraction, so users will need to pull in the thread library anyway. It's very much inspired by the way bdev layer handles data buffers (much of the code was directly copied over). There are two global mempools, one for small and one for large buffers, and per-thread queues that hold requests waiting for a buffer. The main difference is that we also need to track which module requested a buffer in order to allow users to iterate over its pending requests. The usage is fairly simple: ``` /* Embed spdk_iobuf_channel into an existing IO channel */ struct foo_channel { ... struct spdk_iobuf_channel iobuf; }; /* Embed spdk_iobuf_entry into objects that will request buffers */ struct foo_object { ... struct spdk_iobuf_entry entry; }; /* Register the module as iobuf user */ spdk_iobuf_register_module("foo"); /* Initialize iobuf channel in foo_channel's create cb */ spdk_iobuf_channel_init(&foo_channel->iobuf, "foo", 0, 0); /* Finally, request a buffer... */ buf = spdk_iobuf_get(&foo_channel->iobuf, length, &foo_objet.entry, buf_get_cb); ... /* ...and release it */ spdk_iobuf_put(&foo_channel->iobuf, buf, length); ``` Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com> Change-Id: Ifaa6934c03ed6587ddba972198e606921bd85008 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/15326 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Community-CI: Mellanox Build Bot Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com> Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2022-10-26 15:21:03 +00:00
CU_ADD_TEST(suite, iobuf);
CU_ADD_TEST(suite, iobuf_cache);
CU_basic_set_mode(CU_BRM_VERBOSE);
CU_basic_run_tests();
num_failures = CU_get_number_of_failures();
CU_cleanup_registry();
return num_failures;
}