performance improvements of select handler by Nick Mathewson; I added

better recovery when memory allocation fails; something that needs to be
done for the poll improvements, too.


svn:r166

select.c

@@ -44,6 +44,9 @@
 #include <string.h>
 #include <unistd.h>
 #include <errno.h>
+#ifdef CHECK_INVARIANTS
+#include <assert.h>
+#endif
 
 #include "event.h"
 #include "event-internal.h"
@@ -59,8 +62,12 @@ extern volatile sig_atomic_t evsignal_caught;
 struct selectop {
 	int event_fds;		/* Highest fd in fd set */
 	int event_fdsz;
-	fd_set *event_readset;
+	fd_set *event_readset_in;
-	fd_set *event_writeset;
+	fd_set *event_writeset_in;
+	fd_set *event_readset_out;
+	fd_set *event_writeset_out;
+	struct event **event_r_by_fd;
+	struct event **event_w_by_fd;
 	sigset_t evsigmask;
 };
 
@@ -79,6 +86,8 @@ const struct eventop selectops = {
 	select_dispatch
 };
 
+static int select_resize(struct selectop *sop, int fdsz);
+
 void *
 select_init(void)
 {
@@ -91,11 +100,40 @@ select_init(void)
 	if (!(sop = calloc(1, sizeof(struct selectop))))
 		return (NULL);
 
+	select_resize(sop, howmany(32 + 1, NFDBITS)*sizeof(fd_mask));
+
 	evsignal_init(&sop->evsigmask);
 
 	return (sop);
 }
 
+#ifdef CHECK_INVARIANTS
+static void
+check_selectop(struct selectop *sop)
+{
+	int i;
+	for (i=0;i<=sop->event_fds;++i) {
+		if (FD_ISSET(i, sop->event_readset_in)) {
+			assert(sop->event_r_by_fd[i]);
+			assert(sop->event_r_by_fd[i]->ev_events & EV_READ);
+			assert(sop->event_r_by_fd[i]->ev_fd == i);
+		} else {
+			assert(! sop->event_r_by_fd[i]);
+		}
+		if (FD_ISSET(i, sop->event_writeset_in)) {
+			assert(sop->event_w_by_fd[i]);
+			assert(sop->event_w_by_fd[i]->ev_events & EV_WRITE);
+			assert(sop->event_w_by_fd[i]->ev_fd == i);
+		} else {
+			assert(! sop->event_w_by_fd[i]);
+		}
+	}
+
+}
+#else
+#define check_selectop(sop)
+#endif
+
 /*
  * Called with the highest fd that we know about.  If it is 0, completely
  * recalculate everything.
@@ -105,41 +143,8 @@ int
 select_recalc(struct event_base *base, void *arg, int max)
 {
 	struct selectop *sop = arg;
-	fd_set *readset, *writeset;
-	struct event *ev;
-	int fdsz;
 
-	if (sop->event_fds < max)
+	check_selectop(sop);
-		sop->event_fds = max;
-
-	if (!sop->event_fds) {
-		TAILQ_FOREACH(ev, &base->eventqueue, ev_next)
-			if (ev->ev_fd > sop->event_fds)
-				sop->event_fds = ev->ev_fd;
-	}
-
-	fdsz = howmany(sop->event_fds + 1, NFDBITS) * sizeof(fd_mask);
-	if (fdsz > sop->event_fdsz) {
-		if ((readset = realloc(sop->event_readset, fdsz)) == NULL) {
-			event_warn("malloc");
-			return (-1);
-		}
-
-		if ((writeset = realloc(sop->event_writeset, fdsz)) == NULL) {
-			event_warn("malloc");
-			free(readset);
-			return (-1);
-		}
-
-		memset((char *)readset + sop->event_fdsz, 0,
-		    fdsz - sop->event_fdsz);
-		memset((char *)writeset + sop->event_fdsz, 0,
-		    fdsz - sop->event_fdsz);
-
-		sop->event_readset = readset;
-		sop->event_writeset = writeset;
-		sop->event_fdsz = fdsz;
-	}
 
 	return (evsignal_recalc(&sop->evsigmask));
 }
@@ -147,26 +152,23 @@ select_recalc(struct event_base *base, void *arg, int max)
 int
 select_dispatch(struct event_base *base, void *arg, struct timeval *tv)
 {
-	int maxfd, res;
+	int res, i;
-	struct event *ev, *next;
 	struct selectop *sop = arg;
 
-	memset(sop->event_readset, 0, sop->event_fdsz);
+	check_selectop(sop);
-	memset(sop->event_writeset, 0, sop->event_fdsz);
 
-	TAILQ_FOREACH(ev, &base->eventqueue, ev_next) {
+	memcpy(sop->event_readset_out, sop->event_readset_in,
-		if (ev->ev_events & EV_WRITE)
+	       sop->event_fdsz);
-			FD_SET(ev->ev_fd, sop->event_writeset);
+	memcpy(sop->event_writeset_out, sop->event_writeset_in,
-		if (ev->ev_events & EV_READ)
+	       sop->event_fdsz);
-			FD_SET(ev->ev_fd, sop->event_readset);
-	}
 
 	if (evsignal_deliver(&sop->evsigmask) == -1)
 		return (-1);
 
-	res = select(sop->event_fds + 1, sop->event_readset, 
+	res = select(sop->event_fds + 1, sop->event_readset_out,
-	    sop->event_writeset, NULL, tv);
+	    sop->event_writeset_out, NULL, tv);
 
+	check_selectop(sop);
 	if (evsignal_recalc(&sop->evsigmask) == -1)
 		return (-1);
 
@@ -183,32 +185,94 @@ select_dispatch(struct event_base *base, void *arg, struct timeval *tv)
 
 	event_debug(("%s: select reports %d", __func__, res));
 
-	maxfd = 0;
+	check_selectop(sop);
-	for (ev = TAILQ_FIRST(&base->eventqueue); ev != NULL; ev = next) {
+	for (i = 0; i <= sop->event_fds; ++i) {
-		next = TAILQ_NEXT(ev, ev_next);
+		struct event *r_ev = NULL, *w_ev = NULL;
-
 		res = 0;
-		if (FD_ISSET(ev->ev_fd, sop->event_readset))
+		if (FD_ISSET(i, sop->event_readset_out)) {
+			r_ev = sop->event_r_by_fd[i];
 			res |= EV_READ;
-		if (FD_ISSET(ev->ev_fd, sop->event_writeset))
+		}
+		if (FD_ISSET(i, sop->event_writeset_out)) {
+			w_ev = sop->event_w_by_fd[i];
 			res |= EV_WRITE;
-		res &= ev->ev_events;
+		}
-
+		if (r_ev && (res & r_ev->ev_events)) {
-		if (res) {
+			if (!(r_ev->ev_events & EV_PERSIST))
-			if (!(ev->ev_events & EV_PERSIST))
+				event_del(r_ev);
-				event_del(ev);
+			event_active(r_ev, res & r_ev->ev_events, 1);
-			else if (ev->ev_fd > maxfd)
+		}
-				maxfd = ev->ev_fd;
+		if (w_ev && w_ev != r_ev && (res & w_ev->ev_events)) {
-			event_active(ev, res, 1);
+			if (!(w_ev->ev_events & EV_PERSIST))
-		} else if (ev->ev_fd > maxfd)
+				event_del(w_ev);
-			maxfd = ev->ev_fd;
+			event_active(w_ev, res & w_ev->ev_events, 1);
+		}
 	}
-
+	check_selectop(sop);
-	sop->event_fds = maxfd;
 
 	return (0);
 }
 
+
+static int
+select_resize(struct selectop *sop, int fdsz)
+{
+	int n_events, n_events_old;
+
+	fd_set *readset_in = NULL;
+	fd_set *writeset_in = NULL;
+	fd_set *readset_out = NULL;
+	fd_set *writeset_out = NULL;
+	struct event **r_by_fd = NULL;
+	struct event **w_by_fd = NULL;
+
+	n_events = (fdsz/sizeof(fd_mask)) * NFDBITS;
+	n_events_old = (sop->event_fdsz/sizeof(fd_mask)) * NFDBITS;
+
+	if (sop->event_readset_in)
+		check_selectop(sop);
+
+	if ((readset_in = realloc(sop->event_readset_in, fdsz)) == NULL)
+		goto error;
+	sop->event_readset_in = readset_in;
+	if ((readset_out = realloc(sop->event_readset_out, fdsz)) == NULL)
+		goto error;
+	sop->event_readset_out = readset_out;
+	if ((writeset_in = realloc(sop->event_writeset_in, fdsz)) == NULL)
+		goto error;
+	sop->event_writeset_in = writeset_in;
+	if ((writeset_out = realloc(sop->event_writeset_out, fdsz)) == NULL)
+		goto error;
+	sop->event_writeset_out = writeset_out;
+	if ((r_by_fd = realloc(sop->event_r_by_fd,
+		 n_events*sizeof(struct event*))) == NULL)
+		goto error;
+	sop->event_r_by_fd = r_by_fd;
+	if ((w_by_fd = realloc(sop->event_w_by_fd,
+		 n_events * sizeof(struct event*))) == NULL)
+		goto error;
+	sop->event_w_by_fd = w_by_fd;
+
+	memset((char *)sop->event_readset_in + sop->event_fdsz, 0,
+	    fdsz - sop->event_fdsz);
+	memset((char *)sop->event_writeset_in + sop->event_fdsz, 0,
+	    fdsz - sop->event_fdsz);
+	memset(sop->event_r_by_fd + n_events_old, 0,
+	    (n_events-n_events_old) * sizeof(struct event*));
+	memset(sop->event_w_by_fd + n_events_old, 0,
+	    (n_events-n_events_old) * sizeof(struct event*));
+
+	sop->event_fdsz = fdsz;
+	check_selectop(sop);
+
+	return (0);
+
+ error:
+	event_warn("malloc");
+	return (-1);
+}
+
+
 int
 select_add(void *arg, struct event *ev)
 {
@@ -217,12 +281,40 @@ select_add(void *arg, struct event *ev)
 	if (ev->ev_events & EV_SIGNAL)
 		return (evsignal_add(&sop->evsigmask, ev));
 
-	/* 
+	check_selectop(sop);
+	/*
 	 * Keep track of the highest fd, so that we can calculate the size
 	 * of the fd_sets for select(2)
 	 */
-	if (sop->event_fds < ev->ev_fd)
+	if (sop->event_fds < ev->ev_fd) {
+		int fdsz = sop->event_fdsz;
+
+		if (fdsz < sizeof(fd_mask))
+			fdsz = sizeof(fd_mask);
+
+		while (fdsz <
+		    (howmany(ev->ev_fd + 1, NFDBITS) * sizeof(fd_mask)))
+			fdsz *= 2;
+
+		if (fdsz != sop->event_fdsz) {
+			if (select_resize(sop, fdsz)) {
+				check_selectop(sop);
+				return (-1);
+			}
+		}
+
 		sop->event_fds = ev->ev_fd;
+	}
+
+	if (ev->ev_events & EV_READ) {
+		FD_SET(ev->ev_fd, sop->event_readset_in);
+		sop->event_r_by_fd[ev->ev_fd] = ev;
+	}
+	if (ev->ev_events & EV_WRITE) {
+		FD_SET(ev->ev_fd, sop->event_writeset_in);
+		sop->event_w_by_fd[ev->ev_fd] = ev;
+	}
+	check_selectop(sop);
 
 	return (0);
 }
@@ -236,8 +328,25 @@ select_del(void *arg, struct event *ev)
 {
 	struct selectop *sop = arg;
 
-	if (!(ev->ev_events & EV_SIGNAL))
+	check_selectop(sop);
-		return (0);
+	if (ev->ev_events & EV_SIGNAL)
+		return (evsignal_del(&sop->evsigmask, ev));
 
-	return (evsignal_del(&sop->evsigmask, ev));
+	if (sop->event_fds < ev->ev_fd) {
+		check_selectop(sop);
+		return (0);
+	}
+
+	if (ev->ev_events & EV_READ) {
+		FD_CLR(ev->ev_fd, sop->event_readset_in);
+		sop->event_r_by_fd[ev->ev_fd] = NULL;
+	}
+
+	if (ev->ev_events & EV_WRITE) {
+		FD_CLR(ev->ev_fd, sop->event_writeset_in);
+		sop->event_w_by_fd[ev->ev_fd] = NULL;
+	}
+
+	check_selectop(sop);
+	return (0);
 }