X-Git-Url: http://rtime.felk.cvut.cz/gitweb/can-benchmark.git/blobdiff_plain/1323549a8a6d9b15f4415f7fe3b5baaa91292630..0050e301dae44fac57f8b30a22f07ec870618c2d:/latester/latester.c

diff --git a/latester/latester.c b/latester/latester.c
index 8d38540..ff627dc 100644
--- a/latester/latester.c
+++ b/latester/latester.c
@@ -1,17 +1,21 @@
-/**************************************************************************/
-/* CAN latency tester                                                     */
-/* Copyright (C) 2010 Michal Sojka, DCE FEE CTU Prague                    */
-/* License: GPLv2     	     	    	    				  */
-/**************************************************************************/
+/*******************************************************************************/
+/* CAN latency tester                                                          */
+/* Copyright (C) 2010, 2011, 2012, 2013, 2014 Michal Sojka, DCE FEE CTU Prague */
+/* License: GPLv2     	     	    	    				       */
+/*******************************************************************************/
 
+#define _GNU_SOURCE
 #include <ctype.h>
 #include <errno.h>
 #include <error.h>
 #include <fcntl.h>
 #include <math.h>
 #include <net/if.h>
+#include <poll.h>
+#include <popt.h>
 #include <pthread.h>
 #include <semaphore.h>
+#include <sched.h>
 #include <signal.h>
 #include <stdbool.h>
 #include <stdint.h>
@@ -19,21 +23,22 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
 #include <sys/time.h>
 #include <sys/types.h>
-//#include <ul_list.h>
+#include <talloc.h>
 #include <unistd.h>
-#include <sched.h>
-#include <sys/mman.h>
-#include <sys/socket.h>
+
+#include "canframelen.h"
 #include <linux/can.h>
 #include <linux/can/raw.h>
-#include <poll.h>
-#include <talloc.h>
-#include <popt.h>
 
 #include "histogram.h"
 
+//#define FTRACE
+
 #ifndef DEBUG
 #define dbg(level, fmt, arg...) do {} while (0)
 #else
@@ -56,16 +61,38 @@ struct options {
 	unsigned timeout_ms;
 	unsigned count;
 	unsigned oneattime;
-	char *file;
-	char *histogram_fn;
-} opt = {
+	char *name;
+	int length;
+	int userhist;
+	int quiet;
+
+	/* Temporary variables */
+	FILE *f_msgs;
+	FILE *f_hist;
+	FILE *f_hist_gw;
+	FILE *f_stat;
+};
+
+struct options opt = {
 	.id = 10,
 	.period_us = 0,
 	.timeout_ms = 1000,
-	
+	.length = 2,
 };
 
+struct {
+	unsigned enobufs;
+	unsigned overrun;
+	unsigned lost;
+	struct timespec tic, tac;
+	unsigned timeouts;
+	unsigned invalid_frame;
+} stats;
+
 int num_interfaces = 0;
+int count = 0;			/* Number of sent messages */
+unsigned msg_in_progress = 0;
+int completion_pipe[2];
 
 struct msg_info {
 	canid_t id;
@@ -73,24 +100,60 @@ struct msg_info {
 	struct timespec ts_sent, ts_sent_kern;
 	struct timespec ts_rx_onwire, ts_rx_onwire_kern;
 	struct timespec ts_rx_final, ts_rx_final_kern;
+	struct can_frame sent, received;
+	unsigned lat_measured_us, tx_time_us;
 };
 
 #define MAX_INFOS 10000
 struct msg_info msg_infos[MAX_INFOS];
-uint16_t curr_msg = 0;
 
-struct histogram histogram;
+struct histogram histogram, histogram_gw;
 
-static inline struct msg_info *frame2info(struct can_frame *frame)
+void sprint_canframe(char *buf , struct can_frame *cf, int sep) {
+	/* documentation see lib.h */
+
+	int i,offset;
+	int dlc = (cf->can_dlc > 8)? 8 : cf->can_dlc;
+
+	if (cf->can_id & CAN_ERR_FLAG) {
+		sprintf(buf, "%08X#", cf->can_id & (CAN_ERR_MASK|CAN_ERR_FLAG));
+		offset = 9;
+	} else if (cf->can_id & CAN_EFF_FLAG) {
+		sprintf(buf, "%08X#", cf->can_id & CAN_EFF_MASK);
+		offset = 9;
+	} else {
+		sprintf(buf, "%03X#", cf->can_id & CAN_SFF_MASK);
+		offset = 4;
+	}
+
+	if (cf->can_id & CAN_RTR_FLAG) /* there are no ERR frames with RTR */
+		sprintf(buf+offset, "R");
+	else
+		for (i = 0; i < dlc; i++) {
+			sprintf(buf+offset, "%02X", cf->data[i]);
+			offset += 2;
+			if (sep && (i+1 < dlc))
+				sprintf(buf+offset++, ".");
+		}
+}
+
+static inline uint16_t frame_index(struct can_frame *frame)
 {
 	uint16_t idx;
-	if (frame->can_dlc == 2) {
+	if (frame->can_dlc >= 2) {
 		memcpy(&idx, frame->data, sizeof(idx));
 		if (idx >= MAX_INFOS)
 			error(1, 0, "%s idx too high", __FUNCTION__);
-	} else
+	} else {
+
 		error(1, 0, "%s error", __FUNCTION__);
-	return &msg_infos[idx];
+	}
+	return idx;
+}
+
+static inline struct msg_info *frame2info(struct can_frame *frame)
+{
+	return &msg_infos[frame_index(frame)];
 }
 
 static inline char *tstamp_str(const void *ctx, struct timespec *tstamp)
@@ -99,37 +162,59 @@ static inline char *tstamp_str(const void *ctx, struct timespec *tstamp)
 			       tstamp->tv_sec, tstamp->tv_nsec/1000);
 }
 
-void print_msg_info(struct msg_info *mi)
+int timespec_subtract (struct timespec *result, struct timespec *x, struct timespec *yy);
+
+void msg_info_print(FILE *f, struct msg_info *mi)
 {
-	struct timespec diff;
+	struct timespec diff, routing_delay, tx_time;
 	void *local = talloc_new (NULL);
-	
+	static long num = 0;
+	char sent[64], received[64];
+
+	if (!f)
+		return;
+
+	sprint_canframe(sent, &mi->sent, true);
+	sprint_canframe(received, &mi->received, true);
+
 #define S(ts) tstamp_str(local, &ts)
 #define DIFF(a, b) (timespec_subtract(&diff, &b, &a), S(diff))
-	
-	if (num_interfaces == 2) 
-		printf("%s -> %s (%s) = %s (%s)\n",
-		       S(mi->ts_sent), S(mi->ts_rx_final_kern), S(mi->ts_rx_final),
-		       DIFF(mi->ts_sent, mi->ts_rx_onwire_kern),
-		       DIFF(mi->ts_sent, mi->ts_rx_onwire));
-	else
-		printf("%s -> %s (%s) -> %s (%s) = %s (%s), %s (%s)\n",
-		       S(mi->ts_sent),
-		       S(mi->ts_rx_onwire_kern), S(mi->ts_rx_onwire),
-		       S(mi->ts_rx_final_kern), S(mi->ts_rx_final),
-		       DIFF(mi->ts_sent, mi->ts_rx_onwire_kern),
-		       DIFF(mi->ts_sent, mi->ts_rx_onwire),
-		       DIFF(mi->ts_rx_onwire_kern, mi->ts_rx_final_kern),
-		       DIFF(mi->ts_rx_onwire, mi->ts_rx_final));
+
+	switch (num_interfaces) {
+	case 2:
+		fprintf(f, "%ld: %s %s -> %s (%s) %s = %s (%s) %d\n",
+			num, S(mi->ts_sent), sent, S(mi->ts_rx_final_kern), S(mi->ts_rx_final), received,
+			DIFF(mi->ts_sent, mi->ts_rx_final_kern),
+			DIFF(mi->ts_sent, mi->ts_rx_final),
+			mi->tx_time_us);
+		break;
+	case 3:
+		tx_time.tv_sec = 0;
+		tx_time.tv_nsec = mi->tx_time_us*1000;
+		timespec_subtract(&routing_delay, &mi->ts_rx_final_kern, &mi->ts_rx_onwire_kern);
+		timespec_subtract(&routing_delay, &routing_delay, &tx_time);
+
+		fprintf(f, "%ld: %s %s -> %s (%s) -> %s (%s) %s = %s (%s), %s (%s) %d %s\n",
+			num, S(mi->ts_sent), sent,
+			S(mi->ts_rx_onwire_kern), S(mi->ts_rx_onwire),
+			S(mi->ts_rx_final_kern), S(mi->ts_rx_final), received,
+			DIFF(mi->ts_sent, mi->ts_rx_onwire_kern),
+			DIFF(mi->ts_sent, mi->ts_rx_onwire),
+			DIFF(mi->ts_rx_onwire_kern, mi->ts_rx_final_kern),
+			DIFF(mi->ts_rx_onwire, mi->ts_rx_final),
+			mi->tx_time_us,
+			S(routing_delay));
+		break;
+	}
 #undef S
 #undef DIFF
+	num++;
 	talloc_free (local);
 }
 
-
 /* Subtract the `struct timespec' values X and Y, storing the result in
    RESULT.  Return 1 if the difference is negative, otherwise 0.  */
-     
+
 int timespec_subtract (struct timespec *result, struct timespec *x, struct timespec *yy)
 {
 	struct timespec ylocal = *yy, *y = &ylocal;
@@ -144,12 +229,12 @@ int timespec_subtract (struct timespec *result, struct timespec *x, struct times
 		y->tv_nsec += 1000000000 * nsec;
 		y->tv_sec -= nsec;
 	}
-     
+
 	/* Compute the time remaining to wait.
 	   `tv_nsec' is certainly positive. */
 	result->tv_sec = x->tv_sec - y->tv_sec;
 	result->tv_nsec = x->tv_nsec - y->tv_nsec;
-     
+
 	/* Return 1 if result is negative. */
 	return x->tv_sec < y->tv_sec;
 }
@@ -160,11 +245,27 @@ void dbg_print_timespec(char *msg, struct timespec *tv)
 	printf("%s sec=%ld nsec=%ld\n", msg, tv->tv_sec, tv->tv_nsec);
 }
 
-static inline unsigned get_msg_latency_us(struct msg_info *mi)
+static inline void calc_msg_latencies(struct msg_info *mi)
 {
 	struct timespec diff;
-	timespec_subtract(&diff, &mi->ts_rx_final_kern, &mi->ts_rx_onwire_kern);
-	return diff.tv_sec * 1000000 + diff.tv_nsec/1000;
+	switch (num_interfaces) {
+	case 3:
+		if (opt.userhist)
+			timespec_subtract(&diff, &mi->ts_rx_final, &mi->ts_rx_onwire);
+		else
+			timespec_subtract(&diff, &mi->ts_rx_final_kern, &mi->ts_rx_onwire_kern);
+		break;
+	case 2:
+		if (opt.userhist)
+			timespec_subtract(&diff, &mi->ts_rx_final, &mi->ts_sent);
+		else
+			timespec_subtract(&diff, &mi->ts_rx_final_kern, &mi->ts_sent);
+		break;
+	default:
+		return;
+	}
+	mi->lat_measured_us = diff.tv_sec * 1000000 + diff.tv_nsec/1000;
+	mi->tx_time_us = calc_frame_length(&mi->received);
 }
 
 void set_sched_policy_and_prio(int policy, int rtprio)
@@ -192,51 +293,168 @@ static inline int sock_get_if_index(int s, const char *if_name)
 {
 	struct ifreq ifr;
 	MEMSET_ZERO(ifr);
-	
+
 	strcpy(ifr.ifr_name, if_name);
 	if (ioctl(s, SIOCGIFINDEX, &ifr) < 0)
 		error(1, errno, "SIOCGIFINDEX '%s'", if_name);
 	return ifr.ifr_ifindex;
 }
 
-static inline get_tstamp(struct timespec *ts)
+static inline void get_tstamp(struct timespec *ts)
 {
 	clock_gettime(CLOCK_REALTIME, ts);
 }
 
+
+int trace_fd = -1;
+int marker_fd = -1;
+
+void init_ftrace()
+{
+#ifdef FTRACE
+	char *debugfs;
+	char path[256];
+	FILE *f;
+
+	debugfs = "/sys/kernel/debug";
+	if (debugfs) {
+		strcpy(path, debugfs);
+		strcat(path,"/tracing/tracing_on");
+		trace_fd = open(path, O_WRONLY);
+		if (trace_fd >= 0)
+			write(trace_fd, "1", 1);
+
+		strcpy(path, debugfs);
+		strcat(path,"/tracing/trace_marker");
+		marker_fd = open(path, O_WRONLY);
+
+		strcpy(path, debugfs);
+		strcat(path,"/tracing/set_ftrace_pid");
+		f = fopen(path, "w");
+		fprintf(f, "%d\n", getpid());
+		fclose(f);
+		system("echo function_graph > /sys/kernel/debug/tracing/current_tracer");
+		system("echo can_send > /sys/kernel/debug/tracing/set_graph_function");
+		system("echo > /sys/kernel/debug/tracing/trace");
+		system("echo 1 > /sys/kernel/debug/tracing/tracing_enabled");
+	}
+#endif	/* FTRACE */
+}
+
+static inline void trace_on()
+{
+	if (trace_fd >= 0)
+		write(trace_fd, "1", 1);
+}
+
+static inline void trace_off(int ret)
+{
+	if (marker_fd >= 0) {
+		char marker[100];
+		sprintf(marker, "write returned %d\n", ret);
+		write(marker_fd, marker, strlen(marker));
+	}
+	if (trace_fd >= 0)
+		write(trace_fd, "0", 1);
+}
+
+static inline void msg_info_free(struct msg_info *mi)
+{
+	mi->id = -1;
+}
+
+static inline bool msg_info_used(struct msg_info *mi)
+{
+	return mi->id != -1;
+}
+
 int send_frame(int socket)
 {
 	struct can_frame frame;
 	struct msg_info *mi;
 	int ret;
+	static int curr_msg = -1;
+	int i;
+	uint16_t idx;
+
+	MEMSET_ZERO(frame);
+	i = curr_msg+1;
+	while (msg_info_used(&msg_infos[i]) && i != curr_msg) {
+		i++;
+		if (i >= MAX_INFOS)
+			i = 0;
+	}
+	if (i == curr_msg)
+		error(1, 0, "Msg info table is full! Probably, many packets were lost.");
+	else
+		curr_msg = i;
 
 	frame.can_id = opt.id;
-	frame.can_dlc = 2;
-	memcpy(frame.data, &curr_msg, sizeof(curr_msg));
+	if (opt.length < 2)
+		error(1, 0, "Length < 2 is not yet supported");
+	frame.can_dlc = opt.length;
+	idx = curr_msg;
+	memcpy(frame.data, &idx, sizeof(idx));
 	mi = frame2info(&frame);
 
+	mi->id = frame.can_id;
+	mi->length = frame.can_dlc;
 	get_tstamp(&mi->ts_sent);
+	mi->sent = frame;
+
+	trace_on();
 	ret = write(socket, &frame, sizeof(frame));
+	trace_off(ret);
+
+	if (ret == -1 || num_interfaces == 1)
+		msg_info_free(mi);
 	return ret;
 }
 
-static inline void get_next_timeout(struct timespec *timeout)
+static inline void send_and_check(int s)
+{
+	int ret;
+	ret = send_frame(s);
+	if (ret != sizeof(struct can_frame)) {
+/* 		if (ret == -1 && errno == ENOBUFS && opt.period_us == 0 && !opt.oneattime) { */
+/* 			stats.enobufs++; */
+/* 			/\* Ignore this error - pfifo_fast qeuue is full *\/ */
+/* 		} else */
+			error(1, errno, "send_frame (line %d)", __LINE__);
+	} else {
+		count++;
+		msg_in_progress++;
+	}
+}
+
+static inline void get_next_timeout(struct timespec *timeout, bool advance)
 {
 	struct timespec now;
 	static struct timespec last = {-1, 0 };
 
 	clock_gettime(CLOCK_MONOTONIC, &now);
-	
-	if (last.tv_sec == -1)
+
+	if (last.tv_sec == -1) {
 		last = now;
+		last.tv_nsec = last.tv_nsec/1000000*1000000;
+	}
 	if (opt.period_us != 0) {
-		last.tv_sec += opt.period_us/1000000;
-		last.tv_nsec += (opt.period_us%1000000)*1000;
-		while (last.tv_nsec >= 1000000000) {
-			last.tv_nsec -= 1000000000;
-			last.tv_sec++;
-		}		
-		timespec_subtract(timeout, &last, &now);
+		if (advance) {
+			last.tv_sec += opt.period_us/1000000;
+			last.tv_nsec += (opt.period_us%1000000)*1000;
+			while (last.tv_nsec >= 1000000000) {
+				last.tv_nsec -= 1000000000;
+				last.tv_sec++;
+			}
+		}
+		if (timespec_subtract(timeout, &last, &now) /* is negative */) {
+			stats.overrun++;
+			memset(timeout, 0, sizeof(*timeout));
+		}
+/* 		printf("next %ld.%06ld  now %ld.%06ld  --> timeout %ld.%06ld\n", */
+/* 		       last.tv_sec, last.tv_nsec/1000, */
+/* 		       now.tv_sec, now.tv_nsec/1000, */
+/* 		       timeout->tv_sec, timeout->tv_nsec/1000); */
 	} else if (opt.timeout_ms != 0) {
 		timeout->tv_sec = opt.timeout_ms/1000;
 		timeout->tv_nsec = (opt.timeout_ms%1000)*1000000;
@@ -263,7 +481,7 @@ void receive(int s, struct can_frame *frame, struct timespec *ts_kern, struct ti
 	/* these settings may be modified by recvmsg() */
 	iov.iov_len = sizeof(*frame);
 	msg.msg_namelen = sizeof(addr);
-	msg.msg_controllen = sizeof(ctrlmsg);  
+	msg.msg_controllen = sizeof(ctrlmsg);
 	msg.msg_flags = 0;
 
 	nbytes = recvmsg(s, &msg, 0);
@@ -279,9 +497,12 @@ void receive(int s, struct can_frame *frame, struct timespec *ts_kern, struct ti
 	     cmsg && (cmsg->cmsg_level == SOL_SOCKET);
 	     cmsg = CMSG_NXTHDR(&msg,cmsg)) {
 		if (cmsg->cmsg_type == SO_TIMESTAMPNS)
-			*ts_kern = *(struct timespec *)CMSG_DATA(cmsg);
-		else if (cmsg->cmsg_type == SO_RXQ_OVFL)
-			dropcnt += *(__u32 *)CMSG_DATA(cmsg);
+			memcpy(ts_kern, CMSG_DATA(cmsg), sizeof(struct timespec));
+		else if (cmsg->cmsg_type == SO_RXQ_OVFL) {
+			uint32_t ovfl;
+			memcpy(&ovfl, CMSG_DATA(cmsg), sizeof(ovfl));
+			dropcnt += ovfl;
+		}
 	}
 
 }
@@ -293,31 +514,40 @@ void process_tx(int s)
 
 void process_on_wire_rx(int s)
 {
-	struct timespec ts_kern, ts_user, ts_diff;
+	struct timespec ts_kern, ts_user;
 	struct can_frame frame;
 	struct msg_info *mi;
 	receive(s, &frame, &ts_kern, &ts_user);
 	mi = frame2info(&frame);
-	mi->ts_rx_onwire_kern = ts_kern;
-	mi->ts_rx_onwire = ts_user;
+	if (msg_info_used(mi)) {
+		mi->ts_rx_onwire_kern = ts_kern;
+		mi->ts_rx_onwire = ts_user;
+	} else
+		stats.invalid_frame++;
 }
 
 
 void process_final_rx(int s)
 {
-	struct timespec ts_kern, ts_user, ts_diff;
+	struct timespec ts_kern, ts_user;
 	struct can_frame frame;
 	struct msg_info *mi;
+	int ret;
+
 	receive(s, &frame, &ts_kern, &ts_user);
 	mi = frame2info(&frame);
 	mi->ts_rx_final_kern = ts_kern;
 	mi->ts_rx_final = ts_user;
+	mi->received = frame;
+
+	calc_msg_latencies(mi);
 
-	if (opt.histogram_fn)
-		histogram_add(&histogram, get_msg_latency_us(mi));
+	histogram_add(&histogram, mi->lat_measured_us);
+	histogram_add(&histogram_gw, mi->lat_measured_us - mi->tx_time_us);
 
-	if (!opt.file)
-		print_msg_info(mi);
+	ret = write(completion_pipe[1], &mi, sizeof(mi));
+	if (ret == -1)
+		error(1, errno, "completion_pipe write");
 }
 
 void *measure_thread(void *arg)
@@ -326,12 +556,10 @@ void *measure_thread(void *arg)
 	struct pollfd pfd[3];
 	struct timespec timeout;
 	struct sockaddr_can addr;
-	sigset_t set;
-	int count = 0;
-	unsigned msg_in_progress = 0;
+	int consecutive_timeouts = 0;
 
 	MEMSET_ZERO(pfd);
-	
+
 	for (i=0; i<num_interfaces; i++) {
 		if ((s = socket(PF_CAN, SOCK_RAW, CAN_RAW)) < 0)
 			error(1, errno, "socket");
@@ -356,10 +584,10 @@ void *measure_thread(void *arg)
 			       &timestamp_on, sizeof(timestamp_on)) < 0)
 			error(1, errno, "setsockopt SO_TIMESTAMP");
 
-//		const int dropmonitor_on = 1;
-//		if (setsockopt(s, SOL_SOCKET, SO_RXQ_OVFL,
-//			       &dropmonitor_on, sizeof(dropmonitor_on)) < 0)
-//			error(1, errno, "setsockopt SO_RXQ_OVFL not supported by your Linux Kernel");
+		const int dropmonitor_on = 1;
+		if (setsockopt(s, SOL_SOCKET, SO_RXQ_OVFL,
+			       &dropmonitor_on, sizeof(dropmonitor_on)) < 0)
+			error(1, errno, "setsockopt SO_RXQ_OVFL not supported by your Linux Kernel");
 
 		pfd[i].fd = s;
 		if (i == 0)
@@ -368,30 +596,23 @@ void *measure_thread(void *arg)
 			pfd[i].events = POLLIN;
 	}
 
-	set_sched_policy_and_prio(SCHED_FIFO, 99);
+	set_sched_policy_and_prio(SCHED_FIFO, 40);
 
-#define SEND()						\
-	do {	printf("send\n");			\
-		ret = send_frame(pfd[0].fd);		\
-		if (ret != sizeof(struct can_frame))	\
-			error(1, errno, "send_frame %d", __LINE__);	\
-		count++;				\
-		msg_in_progress++;			\
-	} while (0)
+#define SEND() send_and_check(pfd[0].fd)
 
-	if (opt.oneattime) {
+	if (opt.oneattime)
 		SEND();
-		count = 1;
-	}
 
+	get_tstamp(&stats.tic);
+	ret = 0;
 	while (!finish_flag &&
 	       (opt.count == 0 || count < opt.count || msg_in_progress != 0)) {
 
-		get_next_timeout(&timeout);
-		printf("ppoll");
-		fflush(stdout);
+		get_next_timeout(&timeout, ret == 0);
+		//printf("timeout %ld.%06ld\n", timeout.tv_sec, timeout.tv_nsec/1000);
+		//printf("ppoll"); fflush(stdout);
 		ret = ppoll(pfd, num_interfaces, &timeout, NULL);
-		printf("=%d\n", ret);
+		//printf("=%d\n", ret);
 		switch (ret) {
 		case -1: // Error
 			if (!INTERRUPTED_SYSCALL(errno))
@@ -401,9 +622,15 @@ void *measure_thread(void *arg)
 			if (opt.period_us) {
 				if (opt.count == 0 || count < opt.count) {
 					SEND();
-				} else {
-					error(1, 0, "poll timeout");
 				}
+			} else {
+				/* Lost message - send a new one */
+				stats.timeouts++;
+				consecutive_timeouts++;
+				if (consecutive_timeouts < 10)
+					SEND();
+				else /* Something is really broken */
+					finish_flag = 1;
 			}
 			break;
 		default: // Event
@@ -416,13 +643,16 @@ void *measure_thread(void *arg)
 			}
 			pfd[0].revents = 0;
 
-			if (num_interfaces == 3 && pfd[1].revents != 0) {
+			if (num_interfaces == 3 && pfd[1].revents & POLLIN) {
 				process_on_wire_rx(pfd[1].fd);
 				pfd[1].revents = 0;
 			}
+			if (num_interfaces == 3 && pfd[1].revents & ~POLLIN)
+				error(1, 0, "Unexpected pfd[1].revents: 0x%04x\n", pfd[1].revents);
 
 			i = (num_interfaces == 2) ? 1 : 2;
-			if (pfd[i].revents != 0) {
+			if (pfd[i].revents & POLLIN) {
+				consecutive_timeouts = 0;
 				process_final_rx(pfd[i].fd);
 				msg_in_progress--;
 				pfd[i].revents = 0;
@@ -431,9 +661,13 @@ void *measure_thread(void *arg)
 					SEND();
 				}
 			}
+			if (pfd[i].revents & ~POLLIN)
+				error(1, 0, "Unexpected pfd[%d].revents: 0x%04x\n", i, pfd[i].revents);
 		}
 	}
 
+	get_tstamp(&stats.tac);
+
 	for (i=0; i<num_interfaces; i++)
 		close(pfd[i].fd);
 
@@ -448,8 +682,10 @@ struct poptOption optionsTable[] = {
 	{ "timeout",'t', POPT_ARG_INT|POPT_ARGFLAG_SHOW_DEFAULT,  &opt.timeout_ms,0, "Timeout when period is zero", "ms"},
 	{ "oneattime",'o', POPT_ARG_NONE,			  &opt.oneattime,0,  "Send the next message only when the previous was finally received"},
 	{ "verbose",'v', POPT_ARG_NONE,			  	  NULL, 'v',   	     "Send the next message only when the previous was finally received"},
-	{ "file",   'f', POPT_ARG_STRING, 			  &opt.file,    0,   "File where to store results", "filename"},
-	{ "histogram", 'h', POPT_ARG_STRING, 			  &opt.histogram_fn, 0, "Store histogram in file", "filename"},
+	{ "name",   'n', POPT_ARG_STRING,			  &opt.name, 0,	     "Prefix of the generated files"},
+	{ "length", 'l', POPT_ARG_INT|POPT_ARGFLAG_SHOW_DEFAULT,  &opt.length, 0,    "The length of generated messages", "bytes"},
+	{ "userhist", 'u', POPT_ARG_NONE,  			  &opt.userhist, 0,  "Generate histogram from userspace timestamps"},
+	{ "quiet",  'q', POPT_ARG_NONE,  			  &opt.quiet, 0,     "Do not print progress and statistics"},
 	POPT_AUTOHELP
 	{ NULL, 0, 0, NULL, 0 }
 };
@@ -458,7 +694,8 @@ int parse_options(int argc, const char *argv[])
 {
 	int c;
 	poptContext optCon;   /* context for parsing command-line options */
-	
+	void *local = talloc_new (NULL);
+
 	optCon = poptGetContext(NULL, argc, argv, optionsTable, 0);
 	//poptSetOtherOptionHelp(optCon, "[OPTIONS]* <port>");
 
@@ -474,53 +711,194 @@ int parse_options(int argc, const char *argv[])
 		error(1, 0, "%s: %s\n",
 		      poptBadOption(optCon, POPT_BADOPTION_NOALIAS),
 		      poptStrerror(c));
-	
-	if (num_interfaces < 2 || num_interfaces > 3)
-		error(1, 0, "-d option must be given exactly 2 or 3 times");
+
+	if (num_interfaces < 1 || num_interfaces > 3)
+		error(1, 0, "-d option must only be given one, two or three times");
 
 	if (opt.oneattime && opt.period_us)
 		error(1, 0, "oneattime and period cannot be specified at the same time");
 
-	poptFreeContext(optCon);
+	if (opt.name) {
+		char *f = talloc_asprintf(local, "%s-msgs.txt", opt.name);
+		opt.f_msgs = fopen(f, "w");
+		if (!opt.f_msgs)
+			error(1, errno, "fopen: %s", f);
+	}
+
+	if (opt.name) {
+		char *f = talloc_asprintf(local, "%s-hist-raw.txt", opt.name);
+		opt.f_hist = fopen(f, "w");
+		if (!opt.f_hist)
+			error(1, errno, "fopen: %s", f);
+	}
+
+	if (opt.name) {
+		char *f = talloc_asprintf(local, "%s-hist.txt", opt.name);
+		opt.f_hist_gw = fopen(f, "w");
+		if (!opt.f_hist_gw)
+			error(1, errno, "fopen: %s", f);
+	}
+
+	if (opt.name) {
+		char *f = talloc_asprintf(local, "%s-stat.txt", opt.name);
+		opt.f_stat = fopen(f, "w");
+		if (!opt.f_stat)
+			error(1, errno, "fopen: %s", f);
+	}
 
+	poptFreeContext(optCon);
+	talloc_free(local);
 	return 0;
 }
 
+void print_progress()
+{
+	if (! opt.quiet) {
+		if (num_interfaces > 1)
+			printf("\rSent %5d, in progress %5d", count, msg_in_progress);
+		else
+			printf("\rSent %5d", count);
+		fflush(stdout);
+	}
+}
 
 int main(int argc, const char *argv[])
 {
 	pthread_t thread;
-	sigset_t set;
-	int ret;
-	FILE *fhist;
-	
+	int ret, i;
+
 	parse_options(argc, argv);
-			   
+
 	mlockall(MCL_CURRENT | MCL_FUTURE);
 
 	signal(SIGINT, term_handler);
 	signal(SIGTERM, term_handler);
 
-	if (opt.histogram_fn) {
-		histogram_init(&histogram, 1000000, 5);
-		fhist = fopen(opt.histogram_fn, "w");
-		if (fhist == NULL)
-			error(1, errno, "open %s", opt.histogram_fn);
+	for (i=0; i<MAX_INFOS; i++)
+		msg_infos[i].id = -1;
+
+	histogram_init(&histogram, 5000000, 1);
+	histogram_init(&histogram_gw, 5000000, 1);
+
+	ret = pipe(completion_pipe);
+	if (ret == -1)
+		error(1, errno, "pipe");
+	ret = fcntl(completion_pipe[1], F_SETFL, O_NONBLOCK);
+	if (ret == -1)
+		error(1, errno, "pipe fcntl");
+
+	init_ftrace();
+	if (getenv("LATESTER_CONTROL_HACKBENCH")) {
+		char cmd[1000];
+		sprintf(cmd, "ssh -x -a -S $HOME/.ssh/cangw-connection root@192.168.2.3 'kill -CONT -%s'",
+			getenv("LATESTER_CONTROL_HACKBENCH"));
+		printf("Running: %s\n", cmd);
+		system(cmd);
 	}
 
-
 	pthread_create(&thread, 0, measure_thread, NULL);
 
-/* 	while (!finish_flag) { */
-/* 		sleep(1); */
-/* 	} */
+	struct timespec next, now, diff, allsent = {0,0};
+	clock_gettime(CLOCK_MONOTONIC, &next);
+	int completed = 0;
+	while (!finish_flag && (opt.count == 0 || completed < opt.count)) {
+		struct pollfd pfd[1];
+		pfd[0].fd = completion_pipe[0];
+		pfd[0].events = POLLIN;
+		ret = poll(pfd, 1, 100);
+		if (ret == -1 && !INTERRUPTED_SYSCALL(errno))
+			error(1, errno, "poll main");
+		if (ret > 0 && (pfd[0].revents & POLLIN)) {
+			struct msg_info *mi;
+			int ret;
+			ret = read(completion_pipe[0], &mi, sizeof(mi));
+			if (ret < sizeof(mi))
+				error(1, errno, "read completion returned %d", ret);
+			msg_info_print(opt.f_msgs, mi);
+			msg_info_free(mi);
+			completed++;
+		}
+
+		clock_gettime(CLOCK_MONOTONIC, &now);
+		if (timespec_subtract(&diff, &next, &now)) {
+			print_progress();
+			next.tv_nsec += 100000000;
+			while (next.tv_nsec >= 1000000000) {
+				next.tv_nsec -= 1000000000;
+				next.tv_sec++;
+			}
+		}
+		if (opt.count != 0 && count >= opt.count) {
+			if (allsent.tv_sec == 0)
+				allsent = now;
+			timespec_subtract(&diff, &now, &allsent);
+			if (diff.tv_sec >= 1)
+				finish_flag = 1;
+		}
+	}
+	print_progress();
+	if (!opt.quiet)
+		printf("\n");
+
+	stats.lost = msg_in_progress;
 
 	pthread_join(thread, NULL);
 
-	if (opt.histogram_fn) {
-		histogram_fprint(&histogram, fhist);
-		fclose(fhist);
+	if (getenv("LATESTER_CONTROL_HACKBENCH")) {
+		char cmd[1000];
+		sprintf(cmd, "ssh -x -a -S $HOME/.ssh/cangw-connection root@192.168.2.3 'kill -STOP -%s'",
+			getenv("LATESTER_CONTROL_HACKBENCH"));
+		printf("Running: %s\n", cmd);
+		system(cmd);
 	}
-	
+
+	close(completion_pipe[0]);
+	close(completion_pipe[1]);
+
+	histogram_fprint(&histogram, opt.f_hist);
+	histogram_fprint(&histogram_gw, opt.f_hist_gw);
+	if (opt.f_hist)
+		fclose(opt.f_hist);
+	if (opt.f_hist_gw)
+		fclose(opt.f_hist_gw);
+	if (opt.f_msgs)
+		fclose(opt.f_msgs);
+
+	if (opt.f_stat) {
+		struct histogram_stats hs;
+		fprintf(opt.f_stat, "cmdline='");
+		for (i=0; i<argc; i++)
+			fprintf(opt.f_stat, "%s%s", argv[i], i < argc-1 ? " " : "");
+		fprintf(opt.f_stat, "'\n");
+
+		timespec_subtract(&diff, &stats.tac, &stats.tic);
+		fprintf(opt.f_stat, "duration=%s # seconds\n", tstamp_str(NULL, &diff));
+
+		fprintf(opt.f_stat, "sent=%d\n", count);
+		fprintf(opt.f_stat, "overrun=%d\n", stats.overrun);
+		if (stats.overrun && !opt.quiet)
+			printf("overrun=%d\n", stats.overrun);
+		fprintf(opt.f_stat, "enobufs=%d\n", stats.enobufs);
+		if (stats.enobufs && !opt.quiet)
+			printf("enobufs=%d\n", stats.enobufs);
+		fprintf(opt.f_stat, "lost=%d\n", stats.lost);
+		if (stats.lost && !opt.quiet)
+			printf("lost=%d\n", stats.lost);
+		fprintf(opt.f_stat, "timeouts=%d\n", stats.timeouts);
+		if (stats.timeouts && !opt.quiet)
+			printf("timeouts=%d\n", stats.timeouts);
+		fprintf(opt.f_stat, "invalid_frame=%d\n", stats.timeouts);
+		if (stats.timeouts && !opt.quiet)
+			printf("invalid_frame=%d\n", stats.timeouts);
+
+		histogram_stats(&histogram_gw, &hs);
+		double avg = hs.count ? (double)hs.sum/hs.count : 0;
+		fprintf(opt.f_stat, "avg=%g\n", avg);
+		for (i = 0; i <= 20; i++)
+			fprintf(opt.f_stat, "percentile%d=%d\n", i*5, hs.percentile[i*5]);
+
+		fclose(opt.f_stat);
+	}
+
 	return 0;
 }