epoll optimization, add ff_kevent_do_each

example/Makefile

@@ -17,6 +17,7 @@ LIBS+= -Wl,--no-whole-archive -lrt -lm -ldl -lcrypto -pthread
 TARGET="helloworld"
 all:
 	cc -O -gdwarf-2  -I../lib -o ${TARGET} main.c ${LIBS}
+	cc -O -gdwarf-2  -I../lib -o ${TARGET}_epoll main_epoll.c ${LIBS}
 
 .PHONY: clean
 clean:

lib/Makefile

@@ -128,8 +128,7 @@ FF_SRCS+=			\
 	ff_syscall_wrapper.c \
 	ff_subr_prf.c    \
 	ff_vfs_ops.c  \
-	ff_veth.c  \
-	ff_epoll.c
+	ff_veth.c
 
 FF_HOST_SRCS+= \
 	ff_host_interface.c \
@@ -138,6 +137,7 @@ FF_HOST_SRCS+= \
 	ff_dpdk_if.c \
 	ff_dpdk_kni.c \
 	ff_dpdk_pcap.c \
+	ff_epoll.c \
 	ff_init.c
 
 ifdef FF_IPSEC

lib/ff_api.h

@@ -101,6 +101,9 @@ int ff_poll(struct pollfd fds[], nfds_t nfds, int timeout);
 int ff_kqueue(void);
 int ff_kevent(int kq, const struct kevent *changelist, int nchanges, 
     struct kevent *eventlist, int nevents, const struct timespec *timeout);
+int ff_kevent_do_each(int kq, const struct kevent *changelist, int nchanges, 
+    void *eventlist, int nevents, const struct timespec *timeout, 
+    void (*do_each)(void **, struct kevent *));
 
 #ifdef __cplusplus
 }

lib/ff_api.symlist

@@ -29,10 +29,7 @@ ff_shutdown
 ff_sysctl 
 ff_kqueue
 ff_kevent
-ff_epoll_create
-ff_epoll_ctl
-ff_epoll_wait
-ff_epoll_close
+ff_kevent_do_each
 ff_veth_attach
 ff_veth_detach
 ff_veth_process_packet
@@ -40,4 +37,4 @@ ff_mbuf_gethdr
 ff_mbuf_get
 ff_mbuf_free
 ff_mbuf_copydata
-ff_mbuf_tx_offload
+ff_mbuf_tx_offload

lib/ff_epoll.c

@@ -1,34 +1,24 @@
-#include <sys/param.h>
-#include <sys/limits.h>
-#include <sys/uio.h>
-#include <sys/proc.h>
-#include <sys/syscallsubr.h>
-#include <sys/module.h>
-#include <sys/param.h>
-#include <sys/malloc.h>
-#include <sys/socketvar.h>
-#include <sys/event.h>
-#include <sys/kernel.h>
-#include <sys/refcount.h>
-#include <sys/sysctl.h>
-#include <sys/pcpu.h>
-#include <sys/select.h>
-#include <sys/poll.h>
-#include <sys/event.h>
-#include <sys/file.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <sched.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <assert.h>
+#include <unistd.h>
 #include <netinet/in.h>
-#include <netinet/tcp.h>
-#include <sys/ttycom.h>
-#include <sys/filio.h>
-#include <sys/sysproto.h>
-#include <sys/fcntl.h>
-#include <machine/stdarg.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/time.h>
+#include <sys/select.h>
+#include <sys/syscall.h>
+#include <arpa/inet.h>
+#include <sys/epoll.h>
 
 #include "ff_api.h"
-#include "ff_epoll.h"
 #include "ff_errno.h"
-#include "ff_host_interface.h"
-
 
 
 int
@@ -37,16 +27,16 @@ ff_epoll_create(int size __attribute__((__unused__)))
 	return ff_kqueue();
 }
 
-
 int 
 ff_epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)
 {
+	struct kevent kev[3];
+
 	if (!event && op != EPOLL_CTL_DEL) {
-        ff_os_errno(ff_EINVAL);
+        errno = EINVAL;
 		return -1;
 	}
 
-	struct kevent kev[3];
 	if (op == EPOLL_CTL_ADD){
 		EV_SET(&kev[0], fd, EVFILT_READ,
 			EV_ADD | (event->events & EPOLLIN ? 0 : EV_DISABLE), 0, 0, NULL);
@@ -66,62 +56,48 @@ ff_epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)
 		EV_SET(&kev[2], fd, EVFILT_USER, 0,
 		    NOTE_FFCOPY | (event->events & EPOLLRDHUP ? 1 : 0), 0, NULL);		
 	} else {
-		ff_os_errno(ff_EINVAL);
+		errno = EINVAL;
 		return -1;
 	}
 
 	return ff_kevent(epfd, kev, 3, NULL, 0, NULL);
 }
 
+static void 
+ff_event_to_epoll(void **ev, struct kevent *kev)
+{
+    unsigned int event_one = 0;
+    struct epoll_event **ppev = (struct epoll_event **)ev;
+
+    if (kev->filter & EVFILT_READ) {
+        event_one |= EPOLLIN;
+    } 
+    if (kev->filter & EVFILT_WRITE) {
+        event_one |= EPOLLOUT;
+    }
+
+    if (kev->flags & EV_ERROR) {
+        event_one |= EPOLLERR;
+    }
+
+    if (kev->flags & EV_EOF) {
+        event_one |= EPOLLIN;		
+    }
+
+    (*ppev)->events   = event_one;
+    (*ppev)->data.fd  = kev->ident;
+    (*ppev)++;
+}
+
 int 
 ff_epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout)
 {
+    int i, ret;
 	if (!events || maxevents < 1) {
-		ff_os_errno(ff_EINVAL);
+		errno = EINVAL;
 		return -1;
 	}
-	
-	struct kevent *evlist = malloc(sizeof(struct kevent)*maxevents, M_DEVBUF, M_ZERO|M_NOWAIT);
-	if(NULL == evlist){
-		ff_os_errno(ff_EINVAL);
-		return -1;		
-	}
-	memset(evlist, 0, sizeof(struct kevent)*maxevents);
-	
-	int ret = ff_kevent(epfd, NULL, 0, evlist, maxevents, NULL);
-	if (ret == -1) {
-		free(evlist, M_DEVBUF);
-		return ret;
-	}
 
-	unsigned int event_one = 0;
-	for (int i = 0; i < ret; ++i) {
-		event_one = 0;
-		if (evlist[i].filter & EVFILT_READ) {
-			event_one |= EPOLLIN;
-		} 
-		if (evlist[i].filter & EVFILT_WRITE) {
-			event_one |= EPOLLOUT;
-		}
-
-		if (evlist[i].flags & EV_ERROR) {
-			event_one |= EPOLLERR;
-		}
-
-		if (evlist[i].flags & EV_EOF) {
-			event_one |= EPOLLIN;		
-		}
-		events[i].events   = event_one;
-		events[i].data.fd  = evlist[i].ident;
-	}
-	
-	free(evlist, M_DEVBUF);
-	return ret;
-}
-
-int 
-ff_epoll_close(int epfd)
-{
-	return ff_close(epfd);
+	return ff_kevent_do_each(epfd, NULL, 0, events, maxevents, NULL, ff_event_to_epoll);
 }
 

lib/ff_epoll.h

@@ -1,73 +1,11 @@
 #ifndef _FF_EPOLL_H
 #define _FF_EPOLL_H
 
-//#include <sys/stdint.h>
-//#include <sys/queue.h>
+#include <sys/epoll.h>
 
-#ifndef _KERNEL
-#include <stdint.h>
-#else
-#include <sys/stdint.h>
-#include <sys/types.h>
-#endif
-
-enum EPOLL_EVENTS
-  {
-    EPOLLIN = 0x001,
-#define EPOLLIN EPOLLIN
-    EPOLLPRI = 0x002,
-#define EPOLLPRI EPOLLPRI
-    EPOLLOUT = 0x004,
-#define EPOLLOUT EPOLLOUT
-    EPOLLRDNORM = 0x040,
-#define EPOLLRDNORM EPOLLRDNORM
-    EPOLLRDBAND = 0x080,
-#define EPOLLRDBAND EPOLLRDBAND
-    EPOLLWRNORM = 0x100,
-#define EPOLLWRNORM EPOLLWRNORM
-    EPOLLWRBAND = 0x200,
-#define EPOLLWRBAND EPOLLWRBAND
-    EPOLLMSG = 0x400,
-#define EPOLLMSG EPOLLMSG
-    EPOLLERR = 0x008,
-#define EPOLLERR EPOLLERR
-    EPOLLHUP = 0x010,
-#define EPOLLHUP EPOLLHUP
-    EPOLLRDHUP = 0x2000,
-#define EPOLLRDHUP EPOLLRDHUP
-    EPOLLWAKEUP = 1u << 29,
-#define EPOLLWAKEUP EPOLLWAKEUP
-    EPOLLONESHOT = 1u << 30,
-#define EPOLLONESHOT EPOLLONESHOT
-    EPOLLET = 1u << 31
-#define EPOLLET EPOLLET
-  };
-
-
-/* Valid opcodes ( "op" parameter ) to issue to epoll_ctl().  */
-#define EPOLL_CTL_ADD 1	/* Add a file descriptor to the interface.  */
-#define EPOLL_CTL_DEL 2	/* Remove a file descriptor from the interface.  */
-#define EPOLL_CTL_MOD 3	/* Change file descriptor epoll_event structure.  */
-
-
-typedef union epoll_data
-{
-  void *ptr;
-  int fd;
-  uint32_t u32;
-  uint64_t u64;
-} epoll_data_t;
-
-struct epoll_event
-{
-  uint32_t events;	/* Epoll events */
-  epoll_data_t data;	/* User data variable */
-};
-
-/*warpper epoll api.*/
 int ff_epoll_create(int size);
 int ff_epoll_ctl(int epfd, int op, int fd, struct epoll_event *event);
-int ff_epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout); 
-int ff_epoll_close(int epfd);
+int ff_epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout);
 
 #endif
+

lib/ff_syscall_wrapper.c

@@ -936,20 +936,31 @@ struct sys_kevent_args {
     int fd;
     const struct kevent *changelist;
     int nchanges;
-    struct kevent *eventlist;
+    void *eventlist;
     int nevents;
     const struct timespec *timeout;
+    void (*do_each)(void **, struct kevent *);
 };
 
 static int
 kevent_copyout(void *arg, struct kevent *kevp, int count)
 {
+    int i;
+    struct kevent *ke, **el;
     struct sys_kevent_args *uap;
 
     uap = (struct sys_kevent_args *)arg;
-    bcopy(kevp, uap->eventlist, count * sizeof *kevp);
 
-    uap->eventlist += count;
+    for (ke = kevp, i = 0; i < count; i++, ke++) {
+        if (uap->do_each) {
+            uap->do_each(&(uap->eventlist), ke);
+
+        } else {
+            el = (struct kevent **)(&uap->eventlist);
+            **el = *ke;
+            (*el)++;
+        }
+    }
 
     return (0);
 }
@@ -971,21 +982,25 @@ kevent_copyin(void *arg, struct kevent *kevp, int count)
 }
 
 int
-ff_kevent(int kq, const struct kevent *changelist, int nchanges, 
-    struct kevent *eventlist, int nevents, const struct timespec *timeout)
+ff_kevent_do_each(int kq, const struct kevent *changelist, int nchanges, 
+    void *eventlist, int nevents, const struct timespec *timeout, 
+    void (*do_each)(void **, struct kevent *))
 {
     int rc;
     struct timespec ts;
     ts.tv_sec = 0;
     ts.tv_nsec = 0;
+
     struct sys_kevent_args ska = {
         kq,
         changelist,
         nchanges,
         eventlist,
         nevents,
-        &ts
+        &ts,
+        do_each
     };
+
     struct kevent_copyops k_ops = {
         &ska,
         kevent_copyout,
@@ -1002,3 +1017,11 @@ kern_fail:
     ff_os_errno(rc);
     return (-1);
 }
+
+int
+ff_kevent(int kq, const struct kevent *changelist, int nchanges, 
+    struct kevent *eventlist, int nevents, const struct timespec *timeout)
+{
+    return ff_kevent_do_each(kq, changelist, nchanges, eventlist, nevents, timeout, NULL);
+}
+