hostap/src/utils/eloop.c

1149 lines
27 KiB
C
Raw Normal View History

/*
* Event loop based on select() loop
* Copyright (c) 2002-2009, Jouni Malinen <j@w1.fi>
*
* This software may be distributed under the terms of the BSD license.
* See README for more details.
*/
#include "includes.h"
#include <assert.h>
#include "common.h"
#include "trace.h"
2009-12-19 19:26:22 +01:00
#include "list.h"
#include "eloop.h"
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#if defined(CONFIG_ELOOP_POLL) && defined(CONFIG_ELOOP_EPOLL)
#error Do not define both of poll and epoll
#endif
#if !defined(CONFIG_ELOOP_POLL) && !defined(CONFIG_ELOOP_EPOLL)
#define CONFIG_ELOOP_SELECT
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#endif
#ifdef CONFIG_ELOOP_POLL
#include <poll.h>
#endif /* CONFIG_ELOOP_POLL */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
#include <sys/epoll.h>
#endif /* CONFIG_ELOOP_EPOLL */
struct eloop_sock {
int sock;
void *eloop_data;
void *user_data;
eloop_sock_handler handler;
WPA_TRACE_REF(eloop);
WPA_TRACE_REF(user);
WPA_TRACE_INFO
};
struct eloop_timeout {
2009-12-19 19:26:22 +01:00
struct dl_list list;
struct os_reltime time;
void *eloop_data;
void *user_data;
eloop_timeout_handler handler;
WPA_TRACE_REF(eloop);
WPA_TRACE_REF(user);
WPA_TRACE_INFO
};
struct eloop_signal {
int sig;
void *user_data;
eloop_signal_handler handler;
int signaled;
};
struct eloop_sock_table {
int count;
struct eloop_sock *table;
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
eloop_event_type type;
int changed;
};
struct eloop_data {
int max_sock;
int count; /* sum of all table counts */
#ifdef CONFIG_ELOOP_POLL
int max_pollfd_map; /* number of pollfds_map currently allocated */
int max_poll_fds; /* number of pollfds currently allocated */
struct pollfd *pollfds;
struct pollfd **pollfds_map;
#endif /* CONFIG_ELOOP_POLL */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
int epollfd;
int epoll_max_event_num;
int epoll_max_fd;
struct eloop_sock *epoll_table;
struct epoll_event *epoll_events;
#endif /* CONFIG_ELOOP_EPOLL */
struct eloop_sock_table readers;
struct eloop_sock_table writers;
struct eloop_sock_table exceptions;
2009-12-19 19:26:22 +01:00
struct dl_list timeout;
int signal_count;
struct eloop_signal *signals;
int signaled;
int pending_terminate;
int terminate;
};
static struct eloop_data eloop;
#ifdef WPA_TRACE
static void eloop_sigsegv_handler(int sig)
{
wpa_trace_show("eloop SIGSEGV");
abort();
}
static void eloop_trace_sock_add_ref(struct eloop_sock_table *table)
{
int i;
if (table == NULL || table->table == NULL)
return;
for (i = 0; i < table->count; i++) {
wpa_trace_add_ref(&table->table[i], eloop,
table->table[i].eloop_data);
wpa_trace_add_ref(&table->table[i], user,
table->table[i].user_data);
}
}
static void eloop_trace_sock_remove_ref(struct eloop_sock_table *table)
{
int i;
if (table == NULL || table->table == NULL)
return;
for (i = 0; i < table->count; i++) {
wpa_trace_remove_ref(&table->table[i], eloop,
table->table[i].eloop_data);
wpa_trace_remove_ref(&table->table[i], user,
table->table[i].user_data);
}
}
#else /* WPA_TRACE */
#define eloop_trace_sock_add_ref(table) do { } while (0)
#define eloop_trace_sock_remove_ref(table) do { } while (0)
#endif /* WPA_TRACE */
int eloop_init(void)
{
os_memset(&eloop, 0, sizeof(eloop));
2009-12-19 19:26:22 +01:00
dl_list_init(&eloop.timeout);
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
eloop.epollfd = epoll_create1(0);
if (eloop.epollfd < 0) {
wpa_printf(MSG_ERROR, "%s: epoll_create1 failed. %s\n",
__func__, strerror(errno));
return -1;
}
eloop.readers.type = EVENT_TYPE_READ;
eloop.writers.type = EVENT_TYPE_WRITE;
eloop.exceptions.type = EVENT_TYPE_EXCEPTION;
#endif /* CONFIG_ELOOP_EPOLL */
#ifdef WPA_TRACE
signal(SIGSEGV, eloop_sigsegv_handler);
#endif /* WPA_TRACE */
return 0;
}
static int eloop_sock_table_add_sock(struct eloop_sock_table *table,
int sock, eloop_sock_handler handler,
void *eloop_data, void *user_data)
{
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
struct eloop_sock *temp_table;
struct epoll_event ev, *temp_events;
int next;
#endif /* CONFIG_ELOOP_EPOLL */
struct eloop_sock *tmp;
int new_max_sock;
if (sock > eloop.max_sock)
new_max_sock = sock;
else
new_max_sock = eloop.max_sock;
if (table == NULL)
return -1;
#ifdef CONFIG_ELOOP_POLL
if (new_max_sock >= eloop.max_pollfd_map) {
struct pollfd **nmap;
nmap = os_realloc_array(eloop.pollfds_map, new_max_sock + 50,
sizeof(struct pollfd *));
if (nmap == NULL)
return -1;
eloop.max_pollfd_map = new_max_sock + 50;
eloop.pollfds_map = nmap;
}
if (eloop.count + 1 > eloop.max_poll_fds) {
struct pollfd *n;
int nmax = eloop.count + 1 + 50;
n = os_realloc_array(eloop.pollfds, nmax,
sizeof(struct pollfd));
if (n == NULL)
return -1;
eloop.max_poll_fds = nmax;
eloop.pollfds = n;
}
#endif /* CONFIG_ELOOP_POLL */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
if (new_max_sock >= eloop.epoll_max_fd) {
next = eloop.epoll_max_fd == 0 ? 16 : eloop.epoll_max_fd * 2;
temp_table = os_realloc_array(eloop.epoll_table, next,
sizeof(struct eloop_sock));
if (temp_table == NULL)
return -1;
eloop.epoll_max_fd = next;
eloop.epoll_table = temp_table;
}
if (eloop.count + 1 > eloop.epoll_max_event_num) {
next = eloop.epoll_max_event_num == 0 ? 8 :
eloop.epoll_max_event_num * 2;
temp_events = os_realloc_array(eloop.epoll_events, next,
sizeof(struct epoll_event));
if (temp_events == NULL) {
wpa_printf(MSG_ERROR, "%s: malloc for epoll failed. "
"%s\n", __func__, strerror(errno));
return -1;
}
eloop.epoll_max_event_num = next;
eloop.epoll_events = temp_events;
}
#endif /* CONFIG_ELOOP_EPOLL */
eloop_trace_sock_remove_ref(table);
tmp = os_realloc_array(table->table, table->count + 1,
sizeof(struct eloop_sock));
if (tmp == NULL) {
eloop_trace_sock_add_ref(table);
return -1;
}
tmp[table->count].sock = sock;
tmp[table->count].eloop_data = eloop_data;
tmp[table->count].user_data = user_data;
tmp[table->count].handler = handler;
wpa_trace_record(&tmp[table->count]);
table->count++;
table->table = tmp;
eloop.max_sock = new_max_sock;
eloop.count++;
table->changed = 1;
eloop_trace_sock_add_ref(table);
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
os_memset(&ev, 0, sizeof(ev));
switch (table->type) {
case EVENT_TYPE_READ:
ev.events = EPOLLIN;
break;
case EVENT_TYPE_WRITE:
ev.events = EPOLLOUT;
break;
/*
* Exceptions are always checked when using epoll, but I suppose it's
* possible that someone registered a socket *only* for exception
* handling.
*/
case EVENT_TYPE_EXCEPTION:
ev.events = EPOLLERR | EPOLLHUP;
break;
}
ev.data.fd = sock;
if (epoll_ctl(eloop.epollfd, EPOLL_CTL_ADD, sock, &ev) < 0) {
wpa_printf(MSG_ERROR, "%s: epoll_ctl(ADD) for fd=%d "
"failed. %s\n", __func__, sock, strerror(errno));
return -1;
}
os_memcpy(&eloop.epoll_table[sock], &table->table[table->count - 1],
sizeof(struct eloop_sock));
#endif /* CONFIG_ELOOP_EPOLL */
return 0;
}
static void eloop_sock_table_remove_sock(struct eloop_sock_table *table,
int sock)
{
int i;
if (table == NULL || table->table == NULL || table->count == 0)
return;
for (i = 0; i < table->count; i++) {
if (table->table[i].sock == sock)
break;
}
if (i == table->count)
return;
eloop_trace_sock_remove_ref(table);
if (i != table->count - 1) {
os_memmove(&table->table[i], &table->table[i + 1],
(table->count - i - 1) *
sizeof(struct eloop_sock));
}
table->count--;
eloop.count--;
table->changed = 1;
eloop_trace_sock_add_ref(table);
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
if (epoll_ctl(eloop.epollfd, EPOLL_CTL_DEL, sock, NULL) < 0) {
wpa_printf(MSG_ERROR, "%s: epoll_ctl(DEL) for fd=%d "
"failed. %s\n", __func__, sock, strerror(errno));
return;
}
os_memset(&eloop.epoll_table[sock], 0, sizeof(struct eloop_sock));
#endif /* CONFIG_ELOOP_EPOLL */
}
#ifdef CONFIG_ELOOP_POLL
static struct pollfd * find_pollfd(struct pollfd **pollfds_map, int fd, int mx)
{
if (fd < mx && fd >= 0)
return pollfds_map[fd];
return NULL;
}
static int eloop_sock_table_set_fds(struct eloop_sock_table *readers,
struct eloop_sock_table *writers,
struct eloop_sock_table *exceptions,
struct pollfd *pollfds,
struct pollfd **pollfds_map,
int max_pollfd_map)
{
int i;
int nxt = 0;
int fd;
struct pollfd *pfd;
/* Clear pollfd lookup map. It will be re-populated below. */
os_memset(pollfds_map, 0, sizeof(struct pollfd *) * max_pollfd_map);
if (readers && readers->table) {
for (i = 0; i < readers->count; i++) {
fd = readers->table[i].sock;
assert(fd >= 0 && fd < max_pollfd_map);
pollfds[nxt].fd = fd;
pollfds[nxt].events = POLLIN;
pollfds[nxt].revents = 0;
pollfds_map[fd] = &(pollfds[nxt]);
nxt++;
}
}
if (writers && writers->table) {
for (i = 0; i < writers->count; i++) {
/*
* See if we already added this descriptor, update it
* if so.
*/
fd = writers->table[i].sock;
assert(fd >= 0 && fd < max_pollfd_map);
pfd = pollfds_map[fd];
if (!pfd) {
pfd = &(pollfds[nxt]);
pfd->events = 0;
pfd->fd = fd;
pollfds[i].revents = 0;
pollfds_map[fd] = pfd;
nxt++;
}
pfd->events |= POLLOUT;
}
}
/*
* Exceptions are always checked when using poll, but I suppose it's
* possible that someone registered a socket *only* for exception
* handling. Set the POLLIN bit in this case.
*/
if (exceptions && exceptions->table) {
for (i = 0; i < exceptions->count; i++) {
/*
* See if we already added this descriptor, just use it
* if so.
*/
fd = exceptions->table[i].sock;
assert(fd >= 0 && fd < max_pollfd_map);
pfd = pollfds_map[fd];
if (!pfd) {
pfd = &(pollfds[nxt]);
pfd->events = POLLIN;
pfd->fd = fd;
pollfds[i].revents = 0;
pollfds_map[fd] = pfd;
nxt++;
}
}
}
return nxt;
}
static int eloop_sock_table_dispatch_table(struct eloop_sock_table *table,
struct pollfd **pollfds_map,
int max_pollfd_map,
short int revents)
{
int i;
struct pollfd *pfd;
if (!table || !table->table)
return 0;
table->changed = 0;
for (i = 0; i < table->count; i++) {
pfd = find_pollfd(pollfds_map, table->table[i].sock,
max_pollfd_map);
if (!pfd)
continue;
if (!(pfd->revents & revents))
continue;
table->table[i].handler(table->table[i].sock,
table->table[i].eloop_data,
table->table[i].user_data);
if (table->changed)
return 1;
}
return 0;
}
static void eloop_sock_table_dispatch(struct eloop_sock_table *readers,
struct eloop_sock_table *writers,
struct eloop_sock_table *exceptions,
struct pollfd **pollfds_map,
int max_pollfd_map)
{
if (eloop_sock_table_dispatch_table(readers, pollfds_map,
max_pollfd_map, POLLIN | POLLERR |
POLLHUP))
return; /* pollfds may be invalid at this point */
if (eloop_sock_table_dispatch_table(writers, pollfds_map,
max_pollfd_map, POLLOUT))
return; /* pollfds may be invalid at this point */
eloop_sock_table_dispatch_table(exceptions, pollfds_map,
max_pollfd_map, POLLERR | POLLHUP);
}
#endif /* CONFIG_ELOOP_POLL */
#ifdef CONFIG_ELOOP_SELECT
static void eloop_sock_table_set_fds(struct eloop_sock_table *table,
fd_set *fds)
{
int i;
FD_ZERO(fds);
if (table->table == NULL)
return;
for (i = 0; i < table->count; i++) {
assert(table->table[i].sock >= 0);
FD_SET(table->table[i].sock, fds);
}
}
static void eloop_sock_table_dispatch(struct eloop_sock_table *table,
fd_set *fds)
{
int i;
if (table == NULL || table->table == NULL)
return;
table->changed = 0;
for (i = 0; i < table->count; i++) {
if (FD_ISSET(table->table[i].sock, fds)) {
table->table[i].handler(table->table[i].sock,
table->table[i].eloop_data,
table->table[i].user_data);
if (table->changed)
break;
}
}
}
#endif /* CONFIG_ELOOP_SELECT */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
static void eloop_sock_table_dispatch(struct epoll_event *events, int nfds)
{
struct eloop_sock *table;
int i;
for (i = 0; i < nfds; i++) {
table = &eloop.epoll_table[events[i].data.fd];
if (table->handler == NULL)
continue;
table->handler(table->sock, table->eloop_data,
table->user_data);
if (eloop.readers.changed ||
eloop.writers.changed ||
eloop.exceptions.changed)
break;
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
}
}
#endif /* CONFIG_ELOOP_EPOLL */
int eloop_sock_requeue(void)
{
return 0;
}
static void eloop_sock_table_destroy(struct eloop_sock_table *table)
{
if (table) {
int i;
for (i = 0; i < table->count && table->table; i++) {
wpa_printf(MSG_INFO, "ELOOP: remaining socket: "
"sock=%d eloop_data=%p user_data=%p "
"handler=%p",
table->table[i].sock,
table->table[i].eloop_data,
table->table[i].user_data,
table->table[i].handler);
wpa_trace_dump_funcname("eloop unregistered socket "
"handler",
table->table[i].handler);
wpa_trace_dump("eloop sock", &table->table[i]);
}
os_free(table->table);
}
}
int eloop_register_read_sock(int sock, eloop_sock_handler handler,
void *eloop_data, void *user_data)
{
return eloop_register_sock(sock, EVENT_TYPE_READ, handler,
eloop_data, user_data);
}
void eloop_unregister_read_sock(int sock)
{
eloop_unregister_sock(sock, EVENT_TYPE_READ);
}
static struct eloop_sock_table *eloop_get_sock_table(eloop_event_type type)
{
switch (type) {
case EVENT_TYPE_READ:
return &eloop.readers;
case EVENT_TYPE_WRITE:
return &eloop.writers;
case EVENT_TYPE_EXCEPTION:
return &eloop.exceptions;
}
return NULL;
}
int eloop_register_sock(int sock, eloop_event_type type,
eloop_sock_handler handler,
void *eloop_data, void *user_data)
{
struct eloop_sock_table *table;
assert(sock >= 0);
table = eloop_get_sock_table(type);
return eloop_sock_table_add_sock(table, sock, handler,
eloop_data, user_data);
}
void eloop_unregister_sock(int sock, eloop_event_type type)
{
struct eloop_sock_table *table;
table = eloop_get_sock_table(type);
eloop_sock_table_remove_sock(table, sock);
}
int eloop_register_timeout(unsigned int secs, unsigned int usecs,
eloop_timeout_handler handler,
void *eloop_data, void *user_data)
{
2009-12-19 19:26:22 +01:00
struct eloop_timeout *timeout, *tmp;
os_time_t now_sec;
timeout = os_zalloc(sizeof(*timeout));
if (timeout == NULL)
return -1;
if (os_get_reltime(&timeout->time) < 0) {
os_free(timeout);
return -1;
}
now_sec = timeout->time.sec;
timeout->time.sec += secs;
if (timeout->time.sec < now_sec) {
/*
* Integer overflow - assume long enough timeout to be assumed
* to be infinite, i.e., the timeout would never happen.
*/
wpa_printf(MSG_DEBUG, "ELOOP: Too long timeout (secs=%u) to "
"ever happen - ignore it", secs);
os_free(timeout);
return 0;
}
timeout->time.usec += usecs;
while (timeout->time.usec >= 1000000) {
timeout->time.sec++;
timeout->time.usec -= 1000000;
}
timeout->eloop_data = eloop_data;
timeout->user_data = user_data;
timeout->handler = handler;
wpa_trace_add_ref(timeout, eloop, eloop_data);
wpa_trace_add_ref(timeout, user, user_data);
wpa_trace_record(timeout);
2009-12-19 19:26:22 +01:00
/* Maintain timeouts in order of increasing time */
dl_list_for_each(tmp, &eloop.timeout, struct eloop_timeout, list) {
if (os_reltime_before(&timeout->time, &tmp->time)) {
2009-12-19 19:26:22 +01:00
dl_list_add(tmp->list.prev, &timeout->list);
return 0;
}
}
2009-12-19 19:26:22 +01:00
dl_list_add_tail(&eloop.timeout, &timeout->list);
return 0;
}
static void eloop_remove_timeout(struct eloop_timeout *timeout)
{
dl_list_del(&timeout->list);
wpa_trace_remove_ref(timeout, eloop, timeout->eloop_data);
wpa_trace_remove_ref(timeout, user, timeout->user_data);
os_free(timeout);
}
int eloop_cancel_timeout(eloop_timeout_handler handler,
void *eloop_data, void *user_data)
{
2009-12-19 19:26:22 +01:00
struct eloop_timeout *timeout, *prev;
int removed = 0;
2009-12-19 19:26:22 +01:00
dl_list_for_each_safe(timeout, prev, &eloop.timeout,
struct eloop_timeout, list) {
if (timeout->handler == handler &&
(timeout->eloop_data == eloop_data ||
eloop_data == ELOOP_ALL_CTX) &&
(timeout->user_data == user_data ||
user_data == ELOOP_ALL_CTX)) {
eloop_remove_timeout(timeout);
removed++;
2009-12-19 19:26:22 +01:00
}
}
return removed;
}
int eloop_cancel_timeout_one(eloop_timeout_handler handler,
void *eloop_data, void *user_data,
struct os_reltime *remaining)
{
struct eloop_timeout *timeout, *prev;
int removed = 0;
struct os_reltime now;
os_get_reltime(&now);
remaining->sec = remaining->usec = 0;
dl_list_for_each_safe(timeout, prev, &eloop.timeout,
struct eloop_timeout, list) {
if (timeout->handler == handler &&
(timeout->eloop_data == eloop_data) &&
(timeout->user_data == user_data)) {
removed = 1;
if (os_reltime_before(&now, &timeout->time))
os_reltime_sub(&timeout->time, &now, remaining);
eloop_remove_timeout(timeout);
break;
}
}
return removed;
}
int eloop_is_timeout_registered(eloop_timeout_handler handler,
void *eloop_data, void *user_data)
{
struct eloop_timeout *tmp;
2009-12-19 19:26:22 +01:00
dl_list_for_each(tmp, &eloop.timeout, struct eloop_timeout, list) {
if (tmp->handler == handler &&
tmp->eloop_data == eloop_data &&
tmp->user_data == user_data)
return 1;
}
return 0;
}
int eloop_deplete_timeout(unsigned int req_secs, unsigned int req_usecs,
eloop_timeout_handler handler, void *eloop_data,
void *user_data)
{
struct os_reltime now, requested, remaining;
struct eloop_timeout *tmp;
dl_list_for_each(tmp, &eloop.timeout, struct eloop_timeout, list) {
if (tmp->handler == handler &&
tmp->eloop_data == eloop_data &&
tmp->user_data == user_data) {
requested.sec = req_secs;
requested.usec = req_usecs;
os_get_reltime(&now);
os_reltime_sub(&tmp->time, &now, &remaining);
if (os_reltime_before(&requested, &remaining)) {
eloop_cancel_timeout(handler, eloop_data,
user_data);
eloop_register_timeout(requested.sec,
requested.usec,
handler, eloop_data,
user_data);
return 1;
}
return 0;
}
}
return -1;
}
int eloop_replenish_timeout(unsigned int req_secs, unsigned int req_usecs,
eloop_timeout_handler handler, void *eloop_data,
void *user_data)
{
struct os_reltime now, requested, remaining;
struct eloop_timeout *tmp;
dl_list_for_each(tmp, &eloop.timeout, struct eloop_timeout, list) {
if (tmp->handler == handler &&
tmp->eloop_data == eloop_data &&
tmp->user_data == user_data) {
requested.sec = req_secs;
requested.usec = req_usecs;
os_get_reltime(&now);
os_reltime_sub(&tmp->time, &now, &remaining);
if (os_reltime_before(&remaining, &requested)) {
eloop_cancel_timeout(handler, eloop_data,
user_data);
eloop_register_timeout(requested.sec,
requested.usec,
handler, eloop_data,
user_data);
return 1;
}
return 0;
}
}
return -1;
}
#ifndef CONFIG_NATIVE_WINDOWS
static void eloop_handle_alarm(int sig)
{
wpa_printf(MSG_ERROR, "eloop: could not process SIGINT or SIGTERM in "
"two seconds. Looks like there\n"
"is a bug that ends up in a busy loop that "
"prevents clean shutdown.\n"
"Killing program forcefully.\n");
exit(1);
}
#endif /* CONFIG_NATIVE_WINDOWS */
static void eloop_handle_signal(int sig)
{
int i;
#ifndef CONFIG_NATIVE_WINDOWS
if ((sig == SIGINT || sig == SIGTERM) && !eloop.pending_terminate) {
/* Use SIGALRM to break out from potential busy loops that
* would not allow the program to be killed. */
eloop.pending_terminate = 1;
signal(SIGALRM, eloop_handle_alarm);
alarm(2);
}
#endif /* CONFIG_NATIVE_WINDOWS */
eloop.signaled++;
for (i = 0; i < eloop.signal_count; i++) {
if (eloop.signals[i].sig == sig) {
eloop.signals[i].signaled++;
break;
}
}
}
static void eloop_process_pending_signals(void)
{
int i;
if (eloop.signaled == 0)
return;
eloop.signaled = 0;
if (eloop.pending_terminate) {
#ifndef CONFIG_NATIVE_WINDOWS
alarm(0);
#endif /* CONFIG_NATIVE_WINDOWS */
eloop.pending_terminate = 0;
}
for (i = 0; i < eloop.signal_count; i++) {
if (eloop.signals[i].signaled) {
eloop.signals[i].signaled = 0;
eloop.signals[i].handler(eloop.signals[i].sig,
eloop.signals[i].user_data);
}
}
}
int eloop_register_signal(int sig, eloop_signal_handler handler,
void *user_data)
{
struct eloop_signal *tmp;
tmp = os_realloc_array(eloop.signals, eloop.signal_count + 1,
sizeof(struct eloop_signal));
if (tmp == NULL)
return -1;
tmp[eloop.signal_count].sig = sig;
tmp[eloop.signal_count].user_data = user_data;
tmp[eloop.signal_count].handler = handler;
tmp[eloop.signal_count].signaled = 0;
eloop.signal_count++;
eloop.signals = tmp;
signal(sig, eloop_handle_signal);
return 0;
}
int eloop_register_signal_terminate(eloop_signal_handler handler,
void *user_data)
{
int ret = eloop_register_signal(SIGINT, handler, user_data);
if (ret == 0)
ret = eloop_register_signal(SIGTERM, handler, user_data);
return ret;
}
int eloop_register_signal_reconfig(eloop_signal_handler handler,
void *user_data)
{
#ifdef CONFIG_NATIVE_WINDOWS
return 0;
#else /* CONFIG_NATIVE_WINDOWS */
return eloop_register_signal(SIGHUP, handler, user_data);
#endif /* CONFIG_NATIVE_WINDOWS */
}
void eloop_run(void)
{
#ifdef CONFIG_ELOOP_POLL
int num_poll_fds;
int timeout_ms = 0;
#endif /* CONFIG_ELOOP_POLL */
#ifdef CONFIG_ELOOP_SELECT
fd_set *rfds, *wfds, *efds;
struct timeval _tv;
#endif /* CONFIG_ELOOP_SELECT */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
int timeout_ms = -1;
#endif /* CONFIG_ELOOP_EPOLL */
int res;
struct os_reltime tv, now;
#ifdef CONFIG_ELOOP_SELECT
rfds = os_malloc(sizeof(*rfds));
wfds = os_malloc(sizeof(*wfds));
efds = os_malloc(sizeof(*efds));
if (rfds == NULL || wfds == NULL || efds == NULL)
goto out;
#endif /* CONFIG_ELOOP_SELECT */
while (!eloop.terminate &&
2009-12-19 19:26:22 +01:00
(!dl_list_empty(&eloop.timeout) || eloop.readers.count > 0 ||
eloop.writers.count > 0 || eloop.exceptions.count > 0)) {
2009-12-19 19:26:22 +01:00
struct eloop_timeout *timeout;
if (eloop.pending_terminate) {
/*
* This may happen in some corner cases where a signal
* is received during a blocking operation. We need to
* process the pending signals and exit if requested to
* avoid hitting the SIGALRM limit if the blocking
* operation took more than two seconds.
*/
eloop_process_pending_signals();
if (eloop.terminate)
break;
}
2009-12-19 19:26:22 +01:00
timeout = dl_list_first(&eloop.timeout, struct eloop_timeout,
list);
if (timeout) {
os_get_reltime(&now);
if (os_reltime_before(&now, &timeout->time))
os_reltime_sub(&timeout->time, &now, &tv);
else
tv.sec = tv.usec = 0;
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#if defined(CONFIG_ELOOP_POLL) || defined(CONFIG_ELOOP_EPOLL)
timeout_ms = tv.sec * 1000 + tv.usec / 1000;
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#endif /* defined(CONFIG_ELOOP_POLL) || defined(CONFIG_ELOOP_EPOLL) */
#ifdef CONFIG_ELOOP_SELECT
_tv.tv_sec = tv.sec;
_tv.tv_usec = tv.usec;
#endif /* CONFIG_ELOOP_SELECT */
}
#ifdef CONFIG_ELOOP_POLL
num_poll_fds = eloop_sock_table_set_fds(
&eloop.readers, &eloop.writers, &eloop.exceptions,
eloop.pollfds, eloop.pollfds_map,
eloop.max_pollfd_map);
res = poll(eloop.pollfds, num_poll_fds,
timeout ? timeout_ms : -1);
#endif /* CONFIG_ELOOP_POLL */
#ifdef CONFIG_ELOOP_SELECT
eloop_sock_table_set_fds(&eloop.readers, rfds);
eloop_sock_table_set_fds(&eloop.writers, wfds);
eloop_sock_table_set_fds(&eloop.exceptions, efds);
res = select(eloop.max_sock + 1, rfds, wfds, efds,
2009-12-19 19:26:22 +01:00
timeout ? &_tv : NULL);
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#endif /* CONFIG_ELOOP_SELECT */
#ifdef CONFIG_ELOOP_EPOLL
if (eloop.count == 0) {
res = 0;
} else {
res = epoll_wait(eloop.epollfd, eloop.epoll_events,
eloop.count, timeout_ms);
}
#endif /* CONFIG_ELOOP_EPOLL */
if (res < 0 && errno != EINTR && errno != 0) {
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
wpa_printf(MSG_ERROR, "eloop: %s: %s",
#ifdef CONFIG_ELOOP_POLL
"poll"
#endif /* CONFIG_ELOOP_POLL */
#ifdef CONFIG_ELOOP_SELECT
"select"
#endif /* CONFIG_ELOOP_SELECT */
#ifdef CONFIG_ELOOP_EPOLL
"epoll"
#endif /* CONFIG_ELOOP_EPOLL */
, strerror(errno));
goto out;
}
eloop.readers.changed = 0;
eloop.writers.changed = 0;
eloop.exceptions.changed = 0;
eloop_process_pending_signals();
/* check if some registered timeouts have occurred */
timeout = dl_list_first(&eloop.timeout, struct eloop_timeout,
list);
2009-12-19 19:26:22 +01:00
if (timeout) {
os_get_reltime(&now);
if (!os_reltime_before(&now, &timeout->time)) {
void *eloop_data = timeout->eloop_data;
void *user_data = timeout->user_data;
eloop_timeout_handler handler =
timeout->handler;
eloop_remove_timeout(timeout);
handler(eloop_data, user_data);
}
}
if (res <= 0)
continue;
if (eloop.readers.changed ||
eloop.writers.changed ||
eloop.exceptions.changed) {
/*
* Sockets may have been closed and reopened with the
* same FD in the signal or timeout handlers, so we
* must skip the previous results and check again
* whether any of the currently registered sockets have
* events.
*/
continue;
}
#ifdef CONFIG_ELOOP_POLL
eloop_sock_table_dispatch(&eloop.readers, &eloop.writers,
&eloop.exceptions, eloop.pollfds_map,
eloop.max_pollfd_map);
#endif /* CONFIG_ELOOP_POLL */
#ifdef CONFIG_ELOOP_SELECT
eloop_sock_table_dispatch(&eloop.readers, rfds);
eloop_sock_table_dispatch(&eloop.writers, wfds);
eloop_sock_table_dispatch(&eloop.exceptions, efds);
#endif /* CONFIG_ELOOP_SELECT */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
eloop_sock_table_dispatch(eloop.epoll_events, res);
#endif /* CONFIG_ELOOP_EPOLL */
}
eloop.terminate = 0;
out:
#ifdef CONFIG_ELOOP_SELECT
os_free(rfds);
os_free(wfds);
os_free(efds);
#endif /* CONFIG_ELOOP_SELECT */
return;
}
void eloop_terminate(void)
{
eloop.terminate = 1;
}
void eloop_destroy(void)
{
struct eloop_timeout *timeout, *prev;
struct os_reltime now;
os_get_reltime(&now);
2009-12-19 19:26:22 +01:00
dl_list_for_each_safe(timeout, prev, &eloop.timeout,
struct eloop_timeout, list) {
int sec, usec;
2009-12-19 19:26:22 +01:00
sec = timeout->time.sec - now.sec;
usec = timeout->time.usec - now.usec;
if (timeout->time.usec < now.usec) {
sec--;
usec += 1000000;
}
2009-12-19 19:26:22 +01:00
wpa_printf(MSG_INFO, "ELOOP: remaining timeout: %d.%06d "
"eloop_data=%p user_data=%p handler=%p",
sec, usec, timeout->eloop_data, timeout->user_data,
timeout->handler);
wpa_trace_dump_funcname("eloop unregistered timeout handler",
timeout->handler);
2009-12-19 19:26:22 +01:00
wpa_trace_dump("eloop timeout", timeout);
eloop_remove_timeout(timeout);
}
eloop_sock_table_destroy(&eloop.readers);
eloop_sock_table_destroy(&eloop.writers);
eloop_sock_table_destroy(&eloop.exceptions);
os_free(eloop.signals);
#ifdef CONFIG_ELOOP_POLL
os_free(eloop.pollfds);
os_free(eloop.pollfds_map);
#endif /* CONFIG_ELOOP_POLL */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#ifdef CONFIG_ELOOP_EPOLL
os_free(eloop.epoll_table);
os_free(eloop.epoll_events);
close(eloop.epollfd);
#endif /* CONFIG_ELOOP_EPOLL */
}
int eloop_terminated(void)
{
return eloop.terminate || eloop.pending_terminate;
}
void eloop_wait_for_read_sock(int sock)
{
#ifdef CONFIG_ELOOP_POLL
struct pollfd pfd;
if (sock < 0)
return;
os_memset(&pfd, 0, sizeof(pfd));
pfd.fd = sock;
pfd.events = POLLIN;
poll(&pfd, 1, -1);
#endif /* CONFIG_ELOOP_POLL */
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#if defined(CONFIG_ELOOP_SELECT) || defined(CONFIG_ELOOP_EPOLL)
/*
* We can use epoll() here. But epoll() requres 4 system calls.
* epoll_create1(), epoll_ctl() for ADD, epoll_wait, and close() for
* epoll fd. So select() is better for performance here.
*/
fd_set rfds;
if (sock < 0)
return;
FD_ZERO(&rfds);
FD_SET(sock, &rfds);
select(sock + 1, &rfds, NULL, NULL, NULL);
eloop: Add epoll option for better performance This patch adds epoll option for the eloop implementation. This can be selected with the CONFIG_ELOOP_EPOLL=y build option. [merit] See Table1. Table1. comparison table +--------+--------+-----------+------------+-------------+ | | add fd | remove fd | prepare fd | dispatch fd | +--------+--------+-----------+------------+-------------+ | select | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | poll | O(1) | O(1) | O(N) | O(N) | +--------+--------+-----------+------------+-------------+ | epoll | O(1) | O(1) | 0 | O(M) | +--------+--------+-----------+------------+-------------+ "add fd" is addition of fd by eloop_sock_table_add_sock(). "remove fd" is removal of fd by eloop_sock_table_remove_sock(). "prepare fd" is preparation of fds before wait in eloop_run(). "dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch(). "N" is all watching fds. "M" is fds which could be dispatched after waiting. As shown in Table1, epoll option has better performance on "prepare fd" column. Because select/poll option requires setting fds before every select()/poll(). But epoll_wait() doesn't need it. And epoll option has also better performance on "dispatch fd" column. Because select/poll option needs to check all registered fds to find out dispatchable fds. But epoll option doesn't require checking all registered fds. Because epoll_wait() returns dispatchable fd set. So epoll option is effective for GO/AP functionality. [demerit] The epoll option requires additional heap memory. In case of P2P GO, it is about 8K bytes. Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
2014-05-13 02:35:48 +02:00
#endif /* defined(CONFIG_ELOOP_SELECT) || defined(CONFIG_ELOOP_EPOLL) */
}
#ifdef CONFIG_ELOOP_SELECT
#undef CONFIG_ELOOP_SELECT
#endif /* CONFIG_ELOOP_SELECT */