int main(int argc, const char *argv[])
{
if (nice(2*MAX_PRIO) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (nice(-2*MAX_PRIO) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setpriority(-1,0) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (getpriority(-1) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setpriority(getpid(),2*MAX_PRIO) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setpriority(getpid(),-1) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setaffinity(-1, 0) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setaffinity(getpid(), -1) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (setaffinity(getpid(), 3) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
if (getaffinity(-1) != -1)
handle_error("Error: error checking for syscalls should be implemented.\n");
printf(stdout, "hw4-test-err succeeded\n");
exit();
}
static void* slave_main(void* arg) {
slave_conf_t* my_wc = (slave_conf_t*)arg;
pthread_mutex_t *my_mut = mut + my_wc->thread_id;
pthread_cond_t *my_cond = cond + my_wc->thread_id;
BENCHDECLARE
#if BENCHMARKING
unsigned long* t_[BS];
int a1, a2;
a1 = my_wc->thread_id;
a2 = my_wc->thread_id + nrn_nthread;
t_[a1] = t1_[a1];
t_[a2] = t1_[a2];
#endif
setaffinity(my_wc->thread_id);
for(;;) {
if (busywait_) {
while(my_wc->flag == 0) {;}
if (my_wc->flag == 1) {
BENCHBEGIN(a1)
(*my_wc->job)(nrn_threads + my_wc->thread_id);
BENCHADD(a2)
}else{
return (void*)0;
}
my_wc->flag = 0;
pthread_cond_signal(my_cond);
}else{
int
main(void)
{
struct queue *q = qinit();
int i;
pthread_t kids[POP_CNT + PUSH_CNT];
setaffinity(0);
msgs = calloc(MSG_CNT, sizeof(struct msg));
for (i = 0; i < POP_CNT; i++)
start_thread(i, pop_task, q, &kids[i]);
for (; i < POP_CNT + PUSH_CNT; i++)
start_thread(i, push_task, q, &kids[i]);
for (i = 0; i < POP_CNT + PUSH_CNT; i++)
pthread_join(kids[i], NULL);
quit = 0;
if (pop_total < MSG_CNT) {
printf("flushing: %d\n", MSG_CNT - pop_total);
for (i = 0; i < POP_CNT; i++)
start_thread(i, pop_flush_task, q, &kids[i]);
for (i = 0; i < POP_CNT; i++)
pthread_join(kids[i], NULL);
}
printf("pop total: %d\n", pop_total);
printf("pop cycles/msg: %lu\n", pop_cycles / pop_total);
printf("push cycles/msg: %lu\n", push_cycles / MSG_CNT);
return 0;
}
/* set the affinity for the current task to the given CPU */
int on_cpu(unsigned cpu){
cpu_set_t mask;
CPU_ZERO(&mask);
CPU_SET(cpu,&mask);
if (setaffinity(mask) < 0){
perror("sched_setaffinity");
return -1;
}
return 0;
}
int TestMain () {
#if _WIN32||_WIN64 || __linux__ || __FreeBSD_version >= 701000
#if _WIN32||_WIN64
SYSTEM_INFO si;
GetSystemInfo(&si);
if ( si.dwNumberOfProcessors < 2 )
return Harness::Skipped;
int availableProcs = (int)si.dwNumberOfProcessors / 2;
DWORD_PTR mask = 1;
for ( int i = 1; i < availableProcs; ++i )
mask |= mask << 1;
bool err = !SetProcessAffinityMask( GetCurrentProcess(), mask );
#else /* !WIN */
#if __linux__
int maxProcs = get_nprocs();
typedef cpu_set_t mask_t;
#if __TBB_MAIN_THREAD_AFFINITY_BROKEN
#define setaffinity(mask) sched_setaffinity(0 /*get the mask of the calling thread*/, sizeof(mask_t), &mask)
#else
#define setaffinity(mask) sched_setaffinity(getpid(), sizeof(mask_t), &mask)
#endif
#else /* __FreeBSD__ */
int maxProcs = sysconf(_SC_NPROCESSORS_ONLN);
typedef cpuset_t mask_t;
#if __TBB_MAIN_THREAD_AFFINITY_BROKEN
#define setaffinity(mask) cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(mask_t), &mask)
#else
#define setaffinity(mask) cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_PID, -1, sizeof(mask_t), &mask)
#endif
#endif /* __FreeBSD__ */
if ( maxProcs < 2 )
return Harness::Skipped;
mask_t newMask;
CPU_ZERO(&newMask);
int availableProcs = min(maxProcs, (int)sizeof(mask_t) * CHAR_BIT) / 2;
for ( int i = 0; i < availableProcs; ++i )
CPU_SET( i, &newMask );
int err = setaffinity( newMask );
#endif /* !WIN */
ASSERT( !err, "Setting process affinity failed" );
ASSERT( tbb::task_scheduler_init::default_num_threads() == availableProcs, NULL );
ASSERT( (int)tbb::tbb_thread::hardware_concurrency() == availableProcs, NULL );
return Harness::Done;
#else /* !(WIN || LIN || BSD) */
return Harness::Skipped;
#endif /* !(WIN || LIN || BSD) */
}
int
usnet_setup(int argc, char *argv[])
{
//struct nm_desc *nmd;
char *p;
int ret;
(void)argc;
(void)argv;
(void)p;
//(void)nmd;
setaffinity(0);
ret = usnet_get_options(argc, argv);
if ( ret < 0 ) {
show_help();
exit(0);
}
g_nmd = usnet_init(g_nmd, (char*)g_interface, 0);
if (1) {
struct netmap_if *nifp = g_nmd->nifp;
struct nmreq *req = &g_nmd->req;
int i;
D("fisrt_tx_ring=%d, last_tx_ring=%d", g_nmd->first_tx_ring, g_nmd->last_tx_ring);
D("nifp at offset %d, %d tx %d rx region %d",
req->nr_offset, req->nr_tx_rings, req->nr_rx_rings,
req->nr_arg2);
for (i = 0; i <= req->nr_tx_rings; i++) {
struct netmap_ring *ring = NETMAP_TXRING(nifp, i);
D(" TX%d at 0x%p slots %d", i,
(void *)((char *)ring - (char *)nifp), ring->num_slots);
}
for (i = 0; i <= req->nr_rx_rings; i++) {
struct netmap_ring *ring = NETMAP_RXRING(nifp, i);
D(" RX%d at 0x%p slots %d", i,
(void *)((char *)ring - (char *)nifp), ring->num_slots);
}
}
return 0;
}
void *run(void* args)
{
u_int burst = 1024, wait_link = 4;
struct thread_args_t *targ = (struct thread_args_t *)args;
struct nm_desc *desc = (struct nm_desc*)targ->desc;
if (setaffinity(targ->thread, targ->affinity))
{
D("setaffinity failed.");
}
int efd = epoll_create(1);
if (efd < 0 )
{
D("epoll create failed.");
exit(1);
}
struct epoll_event event, events;
memset(&event.data, 0, sizeof(event.data));
event.data.fd = desc->fd;
event.events = EPOLLIN;
epoll_ctl(efd, EPOLL_CTL_ADD, desc->fd, &event);
D("Wait %d secs for link to come up...", wait_link);
sleep(wait_link);
D("Ready to go ....%d", desc->fd);
/* main loop */
signal(SIGINT, sigint_h);
while (!do_abort) {
if (epoll_wait(efd, &events, 1, 200) > 0 )
{
rx(targ, desc, burst);
}
}
return NULL;
}
static void *
pinger_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd pfd = { .fd = targ->fd, .events = POLLIN };
struct netmap_if *nifp = targ->nmd->nifp;
int i, rx = 0, n = targ->g->npackets;
void *frame;
int size;
uint32_t sent = 0;
struct timespec ts, now, last_print;
uint32_t count = 0, min = 1000000000, av = 0;
frame = &targ->pkt;
frame += sizeof(targ->pkt.vh) - targ->g->virt_header;
size = targ->g->pkt_size + targ->g->virt_header;
if (targ->g->nthreads > 1) {
D("can only ping with 1 thread");
return NULL;
}
clock_gettime(CLOCK_REALTIME_PRECISE, &last_print);
now = last_print;
while (n == 0 || (int)sent < n) {
struct netmap_ring *ring = NETMAP_TXRING(nifp, 0);
struct netmap_slot *slot;
char *p;
for (i = 0; i < 1; i++) { /* XXX why the loop for 1 pkt ? */
slot = &ring->slot[ring->cur];
slot->len = size;
p = NETMAP_BUF(ring, slot->buf_idx);
if (nm_ring_empty(ring)) {
D("-- ouch, cannot send");
} else {
struct tstamp *tp;
nm_pkt_copy(frame, p, size);
clock_gettime(CLOCK_REALTIME_PRECISE, &ts);
bcopy(&sent, p+42, sizeof(sent));
tp = (struct tstamp *)(p+46);
tp->sec = (uint32_t)ts.tv_sec;
tp->nsec = (uint32_t)ts.tv_nsec;
sent++;
ring->head = ring->cur = nm_ring_next(ring, ring->cur);
}
}
/* should use a parameter to decide how often to send */
if (poll(&pfd, 1, 3000) <= 0) {
D("poll error/timeout on queue %d: %s", targ->me,
strerror(errno));
continue;
}
/* see what we got back */
for (i = targ->nmd->first_tx_ring;
i <= targ->nmd->last_tx_ring; i++) {
ring = NETMAP_RXRING(nifp, i);
while (!nm_ring_empty(ring)) {
uint32_t seq;
struct tstamp *tp;
slot = &ring->slot[ring->cur];
p = NETMAP_BUF(ring, slot->buf_idx);
clock_gettime(CLOCK_REALTIME_PRECISE, &now);
bcopy(p+42, &seq, sizeof(seq));
tp = (struct tstamp *)(p+46);
ts.tv_sec = (time_t)tp->sec;
ts.tv_nsec = (long)tp->nsec;
ts.tv_sec = now.tv_sec - ts.tv_sec;
ts.tv_nsec = now.tv_nsec - ts.tv_nsec;
if (ts.tv_nsec < 0) {
ts.tv_nsec += 1000000000;
ts.tv_sec--;
}
if (1) D("seq %d/%d delta %d.%09d", seq, sent,
(int)ts.tv_sec, (int)ts.tv_nsec);
if (ts.tv_nsec < (int)min)
min = ts.tv_nsec;
count ++;
av += ts.tv_nsec;
ring->head = ring->cur = nm_ring_next(ring, ring->cur);
rx++;
}
}
//D("tx %d rx %d", sent, rx);
//usleep(100000);
ts.tv_sec = now.tv_sec - last_print.tv_sec;
ts.tv_nsec = now.tv_nsec - last_print.tv_nsec;
if (ts.tv_nsec < 0) {
ts.tv_nsec += 1000000000;
ts.tv_sec--;
}
if (ts.tv_sec >= 1) {
D("count %d min %d av %d",
count, min, av/count);
count = 0;
av = 0;
min = 100000000;
last_print = now;
}
}
return NULL;
}
/*
* reply to ping requests
*/
static void *
ponger_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd pfd = { .fd = targ->fd, .events = POLLIN };
struct netmap_if *nifp = targ->nmd->nifp;
struct netmap_ring *txring, *rxring;
int i, rx = 0, sent = 0, n = targ->g->npackets;
if (targ->g->nthreads > 1) {
D("can only reply ping with 1 thread");
return NULL;
}
D("understood ponger %d but don't know how to do it", n);
while (n == 0 || sent < n) {
uint32_t txcur, txavail;
//#define BUSYWAIT
#ifdef BUSYWAIT
ioctl(pfd.fd, NIOCRXSYNC, NULL);
#else
if (poll(&pfd, 1, 1000) <= 0) {
D("poll error/timeout on queue %d: %s", targ->me,
strerror(errno));
continue;
}
#endif
txring = NETMAP_TXRING(nifp, 0);
txcur = txring->cur;
txavail = nm_ring_space(txring);
/* see what we got back */
for (i = targ->nmd->first_rx_ring; i <= targ->nmd->last_rx_ring; i++) {
rxring = NETMAP_RXRING(nifp, i);
while (!nm_ring_empty(rxring)) {
uint16_t *spkt, *dpkt;
uint32_t cur = rxring->cur;
struct netmap_slot *slot = &rxring->slot[cur];
char *src, *dst;
src = NETMAP_BUF(rxring, slot->buf_idx);
//D("got pkt %p of size %d", src, slot->len);
rxring->head = rxring->cur = nm_ring_next(rxring, cur);
rx++;
if (txavail == 0)
continue;
dst = NETMAP_BUF(txring,
txring->slot[txcur].buf_idx);
/* copy... */
dpkt = (uint16_t *)dst;
spkt = (uint16_t *)src;
nm_pkt_copy(src, dst, slot->len);
dpkt[0] = spkt[3];
dpkt[1] = spkt[4];
dpkt[2] = spkt[5];
dpkt[3] = spkt[0];
dpkt[4] = spkt[1];
dpkt[5] = spkt[2];
txring->slot[txcur].len = slot->len;
/* XXX swap src dst mac */
txcur = nm_ring_next(txring, txcur);
txavail--;
sent++;
}
}
txring->head = txring->cur = txcur;
targ->count = sent;
#ifdef BUSYWAIT
ioctl(pfd.fd, NIOCTXSYNC, NULL);
#endif
//D("tx %d rx %d", sent, rx);
}
return NULL;
}
static __inline int
timespec_ge(const struct timespec *a, const struct timespec *b)
{
if (a->tv_sec > b->tv_sec)
return (1);
if (a->tv_sec < b->tv_sec)
return (0);
if (a->tv_nsec >= b->tv_nsec)
return (1);
return (0);
}
static __inline struct timespec
timeval2spec(const struct timeval *a)
{
struct timespec ts = {
.tv_sec = a->tv_sec,
.tv_nsec = a->tv_usec * 1000
};
return ts;
}
static __inline struct timeval
timespec2val(const struct timespec *a)
{
struct timeval tv = {
.tv_sec = a->tv_sec,
.tv_usec = a->tv_nsec / 1000
};
return tv;
}
static __inline struct timespec
timespec_add(struct timespec a, struct timespec b)
{
struct timespec ret = { a.tv_sec + b.tv_sec, a.tv_nsec + b.tv_nsec };
if (ret.tv_nsec >= 1000000000) {
ret.tv_sec++;
ret.tv_nsec -= 1000000000;
}
return ret;
}
static __inline struct timespec
timespec_sub(struct timespec a, struct timespec b)
{
struct timespec ret = { a.tv_sec - b.tv_sec, a.tv_nsec - b.tv_nsec };
if (ret.tv_nsec < 0) {
ret.tv_sec--;
ret.tv_nsec += 1000000000;
}
return ret;
}
/*
* wait until ts, either busy or sleeping if more than 1ms.
* Return wakeup time.
*/
static struct timespec
wait_time(struct timespec ts)
{
for (;;) {
struct timespec w, cur;
clock_gettime(CLOCK_REALTIME_PRECISE, &cur);
w = timespec_sub(ts, cur);
if (w.tv_sec < 0)
return cur;
else if (w.tv_sec > 0 || w.tv_nsec > 1000000)
poll(NULL, 0, 1);
}
}
static void *
sender_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd pfd = { .fd = targ->fd, .events = POLLOUT };
struct netmap_if *nifp;
struct netmap_ring *txring;
int i, n = targ->g->npackets / targ->g->nthreads;
int64_t sent = 0;
int options = targ->g->options | OPT_COPY;
struct timespec nexttime = { 0, 0}; // XXX silence compiler
int rate_limit = targ->g->tx_rate;
struct pkt *pkt = &targ->pkt;
void *frame;
int size;
if (targ->frame == NULL) {
frame = pkt;
frame += sizeof(pkt->vh) - targ->g->virt_header;
size = targ->g->pkt_size + targ->g->virt_header;
} else {
frame = targ->frame;
size = targ->g->pkt_size;
}
D("start, fd %d main_fd %d", targ->fd, targ->g->main_fd);
if (setaffinity(targ->thread, targ->affinity))
goto quit;
/* main loop.*/
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->tic);
if (rate_limit) {
targ->tic = timespec_add(targ->tic, (struct timespec){2,0});
targ->tic.tv_nsec = 0;
wait_time(targ->tic);
nexttime = targ->tic;
}
if (targ->g->dev_type == DEV_TAP) {
D("writing to file desc %d", targ->g->main_fd);
for (i = 0; !targ->cancel && (n == 0 || sent < n); i++) {
if (write(targ->g->main_fd, frame, size) != -1)
sent++;
update_addresses(pkt, targ->g);
if (i > 10000) {
targ->count = sent;
i = 0;
}
}
#ifndef NO_PCAP
} else if (targ->g->dev_type == DEV_PCAP) {
pcap_t *p = targ->g->p;
for (i = 0; !targ->cancel && (n == 0 || sent < n); i++) {
if (pcap_inject(p, frame, size) != -1)
sent++;
update_addresses(pkt, targ->g);
if (i > 10000) {
targ->count = sent;
i = 0;
}
}
#endif /* NO_PCAP */
} else {
int tosend = 0;
int frags = targ->g->frags;
nifp = targ->nmd->nifp;
while (!targ->cancel && (n == 0 || sent < n)) {
if (rate_limit && tosend <= 0) {
tosend = targ->g->burst;
nexttime = timespec_add(nexttime, targ->g->tx_period);
wait_time(nexttime);
}
/*
* wait for available room in the send queue(s)
*/
if (poll(&pfd, 1, 2000) <= 0) {
if (targ->cancel)
break;
D("poll error/timeout on queue %d: %s", targ->me,
strerror(errno));
// goto quit;
}
if (pfd.revents & POLLERR) {
D("poll error");
goto quit;
}
/*
* scan our queues and send on those with room
*/
if (options & OPT_COPY && sent > 100000 && !(targ->g->options & OPT_COPY) ) {
D("drop copy");
options &= ~OPT_COPY;
}
for (i = targ->nmd->first_tx_ring; i <= targ->nmd->last_tx_ring; i++) {
int m, limit = rate_limit ? tosend : targ->g->burst;
if (n > 0 && n - sent < limit)
limit = n - sent;
txring = NETMAP_TXRING(nifp, i);
if (nm_ring_empty(txring))
continue;
if (frags > 1)
limit = ((limit + frags - 1) / frags) * frags;
m = send_packets(txring, pkt, frame, size, targ->g,
limit, options, frags);
ND("limit %d tail %d frags %d m %d",
limit, txring->tail, frags, m);
sent += m;
targ->count = sent;
if (rate_limit) {
tosend -= m;
if (tosend <= 0)
break;
}
}
}
/* flush any remaining packets */
D("flush tail %d head %d on thread %p",
txring->tail, txring->head,
pthread_self());
ioctl(pfd.fd, NIOCTXSYNC, NULL);
/* final part: wait all the TX queues to be empty. */
for (i = targ->nmd->first_tx_ring; i <= targ->nmd->last_tx_ring; i++) {
txring = NETMAP_TXRING(nifp, i);
while (nm_tx_pending(txring)) {
RD(5, "pending tx tail %d head %d on ring %d",
txring->tail, txring->head, i);
ioctl(pfd.fd, NIOCTXSYNC, NULL);
usleep(1); /* wait 1 tick */
}
}
} /* end DEV_NETMAP */
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->toc);
targ->completed = 1;
targ->count = sent;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
#ifndef NO_PCAP
static void
receive_pcap(u_char *user, const struct pcap_pkthdr * h,
const u_char * bytes)
{
int *count = (int *)user;
(void)h; /* UNUSED */
(void)bytes; /* UNUSED */
(*count)++;
}
#endif /* !NO_PCAP */
static int
receive_packets(struct netmap_ring *ring, u_int limit, int dump)
{
u_int cur, rx, n;
cur = ring->cur;
n = nm_ring_space(ring);
if (n < limit)
limit = n;
for (rx = 0; rx < limit; rx++) {
struct netmap_slot *slot = &ring->slot[cur];
char *p = NETMAP_BUF(ring, slot->buf_idx);
if (dump)
dump_payload(p, slot->len, ring, cur);
cur = nm_ring_next(ring, cur);
}
ring->head = ring->cur = cur;
return (rx);
}
static void *
receiver_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd pfd = { .fd = targ->fd, .events = POLLIN };
struct netmap_if *nifp;
struct netmap_ring *rxring;
int i;
uint64_t received = 0;
if (setaffinity(targ->thread, targ->affinity))
goto quit;
D("reading from %s fd %d main_fd %d",
targ->g->ifname, targ->fd, targ->g->main_fd);
/* unbounded wait for the first packet. */
for (;!targ->cancel;) {
i = poll(&pfd, 1, 1000);
if (i > 0 && !(pfd.revents & POLLERR))
break;
RD(1, "waiting for initial packets, poll returns %d %d",
i, pfd.revents);
}
/* main loop, exit after 1s silence */
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->tic);
if (targ->g->dev_type == DEV_TAP) {
while (!targ->cancel) {
char buf[MAX_BODYSIZE];
/* XXX should we poll ? */
if (read(targ->g->main_fd, buf, sizeof(buf)) > 0)
targ->count++;
}
#ifndef NO_PCAP
} else if (targ->g->dev_type == DEV_PCAP) {
while (!targ->cancel) {
/* XXX should we poll ? */
pcap_dispatch(targ->g->p, targ->g->burst, receive_pcap,
(u_char *)&targ->count);
}
#endif /* !NO_PCAP */
} else {
int dump = targ->g->options & OPT_DUMP;
nifp = targ->nmd->nifp;
while (!targ->cancel) {
/* Once we started to receive packets, wait at most 1 seconds
before quitting. */
if (poll(&pfd, 1, 1 * 1000) <= 0 && !targ->g->forever) {
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->toc);
targ->toc.tv_sec -= 1; /* Subtract timeout time. */
goto out;
}
if (pfd.revents & POLLERR) {
D("poll err");
goto quit;
}
for (i = targ->nmd->first_rx_ring; i <= targ->nmd->last_rx_ring; i++) {
int m;
rxring = NETMAP_RXRING(nifp, i);
if (nm_ring_empty(rxring))
continue;
m = receive_packets(rxring, targ->g->burst, dump);
received += m;
}
targ->count = received;
}
}
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->toc);
out:
targ->completed = 1;
targ->count = received;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
/* very crude code to print a number in normalized form.
* Caller has to make sure that the buffer is large enough.
*/
static const char *
norm(char *buf, double val)
{
char *units[] = { "", "K", "M", "G", "T" };
u_int i;
for (i = 0; val >=1000 && i < sizeof(units)/sizeof(char *) - 1; i++)
val /= 1000;
sprintf(buf, "%.2f %s", val, units[i]);
return buf;
}
static void
tx_output(uint64_t sent, int size, double delta)
{
double bw, raw_bw, pps;
char b1[40], b2[80], b3[80];
printf("Sent %llu packets, %d bytes each, in %.2f seconds.\n",
(unsigned long long)sent, size, delta);
if (delta == 0)
delta = 1e-6;
if (size < 60) /* correct for min packet size */
size = 60;
pps = sent / delta;
bw = (8.0 * size * sent) / delta;
/* raw packets have4 bytes crc + 20 bytes framing */
raw_bw = (8.0 * (size + 24) * sent) / delta;
printf("Speed: %spps Bandwidth: %sbps (raw %sbps)\n",
norm(b1, pps), norm(b2, bw), norm(b3, raw_bw) );
}
static void
rx_output(uint64_t received, double delta)
{
double pps;
char b1[40];
printf("Received %llu packets, in %.2f seconds.\n",
(unsigned long long) received, delta);
if (delta == 0)
delta = 1e-6;
pps = received / delta;
printf("Speed: %spps\n", norm(b1, pps));
}
int main (int argc, char *argv[])
{
char *threads_spec;
pthread_t ebbr, egll, zzzz;
struct spec b, l, p, m;
char *some_mem __attribute__((unused)) = malloc(100);
if (argc > 5 && atoi(argv[5])) setaffinity();
setup_sigusr_handler();
if (argc > 1)
loops = atoi(argv[1]);
if (argc > 2)
sleepms = atoi(argv[2]);
if (argc > 3)
burn = atoll(argv[3]);
if (argc > 4)
threads_spec = argv[4];
else
threads_spec = "BSBSBSBS";
fprintf(stderr, "loops/sleep_ms/burn/threads_spec/affinity: %d %d %d %s %d\n",
loops, sleepms, burn, threads_spec, argc > 5 && atoi(argv[5]));
fflush(stderr);
b.name = "Brussels";
b.burn = *threads_spec++ == 'B';
b.sleep = *threads_spec++ == 'S';
b.t = -1;
if (b.burn || b.sleep) {
b.t = 1;
pthread_create(&ebbr, NULL, sleeper_or_burner, &b);
wait_ready();
}
l.name = "London";
l.burn = *threads_spec++ == 'B';
l.sleep = *threads_spec++ == 'S';
l.t = -1;
if (l.burn || l.sleep) {
l.t = 2;
pthread_create(&egll, NULL, sleeper_or_burner, &l);
wait_ready();
}
p.name = "Petaouchnok";
p.burn = *threads_spec++ == 'B';
p.sleep = *threads_spec++ == 'S';
p.t = -1;
if (p.burn || p.sleep) {
p.t = 3;
pthread_create(&zzzz, NULL, sleeper_or_burner, &p);
wait_ready();
}
m.name = "main";
m.burn = *threads_spec++ == 'B';
m.sleep = *threads_spec++ == 'S';
m.t = 0;
sleeper_or_burner(&m);
if (b.t != -1) pthread_join(ebbr, NULL);
if (l.t != -1) pthread_join(egll, NULL);
if (p.t != -1) pthread_join(zzzz, NULL);
return 0;
}
void setaffinity(uint64_t const proc)
{
setaffinity(std::vector<uint64_t>(1,proc));
}
static __inline struct timespec
timeval2spec(const struct timeval *a)
{
struct timespec ts = {
.tv_sec = a->tv_sec,
.tv_nsec = a->tv_usec * 1000
};
return ts;
}
static __inline struct timeval
timespec2val(const struct timespec *a)
{
struct timeval tv = {
.tv_sec = a->tv_sec,
.tv_usec = a->tv_nsec / 1000
};
return tv;
}
static __inline struct timespec
timespec_add(struct timespec a, struct timespec b)
{
struct timespec ret = { a.tv_sec + b.tv_sec, a.tv_nsec + b.tv_nsec };
if (ret.tv_nsec >= 1000000000) {
ret.tv_sec++;
ret.tv_nsec -= 1000000000;
}
return ret;
}
static __inline struct timespec
timespec_sub(struct timespec a, struct timespec b)
{
struct timespec ret = { a.tv_sec - b.tv_sec, a.tv_nsec - b.tv_nsec };
if (ret.tv_nsec < 0) {
ret.tv_sec--;
ret.tv_nsec += 1000000000;
}
return ret;
}
/*
* wait until ts, either busy or sleeping if more than 1ms.
* Return wakeup time.
*/
static struct timespec
wait_time(struct timespec ts)
{
for (;;) {
struct timespec w, cur;
clock_gettime(CLOCK_REALTIME_PRECISE, &cur);
w = timespec_sub(ts, cur);
if (w.tv_sec < 0)
return cur;
else if (w.tv_sec > 0 || w.tv_nsec > 1000000)
poll(NULL, 0, 1);
}
}
static void *
sender_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd fds[1];
struct netmap_if *nifp = targ->nifp;
struct netmap_ring *txring;
int i, n = targ->g->npackets / targ->g->nthreads, sent = 0;
int options = targ->g->options | OPT_COPY;
struct timespec nexttime = { 0, 0}; // XXX silence compiler
int rate_limit = targ->g->tx_rate;
struct pkt *pkt = &targ->pkt;
void *frame;
int size;
frame = pkt;
frame += sizeof(pkt->vh) - targ->g->virt_header;
size = targ->g->pkt_size + targ->g->virt_header;
D("start");
if (setaffinity(targ->thread, targ->affinity))
goto quit;
/* setup poll(2) mechanism. */
memset(fds, 0, sizeof(fds));
fds[0].fd = targ->fd;
fds[0].events = (POLLOUT);
/* main loop.*/
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->tic);
if (rate_limit) {
targ->tic = timespec_add(targ->tic, (struct timespec){2,0});
targ->tic.tv_nsec = 0;
wait_time(targ->tic);
nexttime = targ->tic;
}
if (targ->g->dev_type == DEV_PCAP) {
pcap_t *p = targ->g->p;
for (i = 0; !targ->cancel && (n == 0 || sent < n); i++) {
if (pcap_inject(p, frame, size) != -1)
sent++;
update_addresses(pkt, targ->g);
if (i > 10000) {
targ->count = sent;
i = 0;
}
}
} else if (targ->g->dev_type == DEV_TAP) { /* tap */
D("writing to file desc %d", targ->g->main_fd);
for (i = 0; !targ->cancel && (n == 0 || sent < n); i++) {
if (write(targ->g->main_fd, frame, size) != -1)
sent++;
update_addresses(pkt, targ->g);
if (i > 10000) {
targ->count = sent;
i = 0;
}
}
} else {
int tosend = 0;
int frags = targ->g->frags;
while (!targ->cancel && (n == 0 || sent < n)) {
if (rate_limit && tosend <= 0) {
tosend = targ->g->burst;
nexttime = timespec_add(nexttime, targ->g->tx_period);
wait_time(nexttime);
}
/*
* wait for available room in the send queue(s)
*/
if (poll(fds, 1, 2000) <= 0) {
if (targ->cancel)
break;
D("poll error/timeout on queue %d: %s", targ->me,
strerror(errno));
goto quit;
}
if (fds[0].revents & POLLERR) {
D("poll error");
goto quit;
}
/*
* scan our queues and send on those with room
*/
if (options & OPT_COPY && sent > 100000 && !(targ->g->options & OPT_COPY) ) {
D("drop copy");
options &= ~OPT_COPY;
}
for (i = targ->qfirst; i < targ->qlast; i++) {
int m, limit = rate_limit ? tosend : targ->g->burst;
if (n > 0 && n - sent < limit)
limit = n - sent;
txring = NETMAP_TXRING(nifp, i);
if (nm_ring_empty(txring))
continue;
if (frags > 1)
limit = ((limit + frags - 1) / frags) * frags;
m = send_packets(txring, pkt, frame, size, targ->g,
limit, options, frags);
ND("limit %d avail %d frags %d m %d",
limit, txring->avail, frags, m);
sent += m;
targ->count = sent;
if (rate_limit) {
tosend -= m;
if (tosend <= 0)
break;
}
}
}
/* flush any remaining packets */
ioctl(fds[0].fd, NIOCTXSYNC, NULL);
/* final part: wait all the TX queues to be empty. */
for (i = targ->qfirst; i < targ->qlast; i++) {
txring = NETMAP_TXRING(nifp, i);
while (nm_tx_pending(txring)) {
ioctl(fds[0].fd, NIOCTXSYNC, NULL);
usleep(1); /* wait 1 tick */
}
}
}
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->toc);
targ->completed = 1;
targ->count = sent;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
static void
receive_pcap(u_char *user, const struct pcap_pkthdr * h,
const u_char * bytes)
{
int *count = (int *)user;
(void)h; /* UNUSED */
(void)bytes; /* UNUSED */
(*count)++;
}
static void *
receiver_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd fds[1];
struct netmap_if *nifp = targ->nifp;
struct netmap_ring *rxring;
int i;
uint64_t received = 0;
if (setaffinity(targ->thread, targ->affinity))
goto quit;
/* setup poll(2) mechanism. */
memset(fds, 0, sizeof(fds));
fds[0].fd = targ->fd;
fds[0].events = (POLLIN);
/* unbounded wait for the first packet. */
for (;;) {
i = poll(fds, 1, 1000);
if (i > 0 && !(fds[0].revents & POLLERR))
break;
RD(1, "waiting for initial packets, poll returns %d %d", i, fds[0].revents);
}
/* main loop, exit after 1s silence */
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->tic);
if (targ->g->dev_type == DEV_PCAP) {
while (!targ->cancel) {
/* XXX should we poll ? */
pcap_dispatch(targ->g->p, targ->g->burst, receive_pcap, NULL);
}
} else if (targ->g->dev_type == DEV_TAP) {
D("reading from %s fd %d", targ->g->ifname, targ->g->main_fd);
while (!targ->cancel) {
char buf[2048];
/* XXX should we poll ? */
if (read(targ->g->main_fd, buf, sizeof(buf)) > 0)
targ->count++;
}
} else {
int dump = targ->g->options & OPT_DUMP;
while (!targ->cancel) {
/* Once we started to receive packets, wait at most 1 seconds
before quitting. */
if (poll(fds, 1, 1 * 1000) <= 0 && !targ->g->forever) {
clock_gettime(CLOCK_REALTIME_PRECISE, &targ->toc);
targ->toc.tv_sec -= 1; /* Subtract timeout time. */
break;
}
if (fds[0].revents & POLLERR) {
D("poll err");
goto quit;
}
for (i = targ->qfirst; i < targ->qlast; i++) {
int m;
rxring = NETMAP_RXRING(nifp, i);
if (nm_ring_empty(rxring))
continue;
m = receive_packets(rxring, targ->g->burst, dump);
received += m;
}
targ->count = received;
// tell the card we have read the data
//ioctl(fds[0].fd, NIOCRXSYNC, NULL);
}
}
targ->completed = 1;
targ->count = received;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
static void *
receiver_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd fds[1];
struct netmap_if *nifp = targ->nifp;
struct netmap_ring *rxring;
int i, received = 0;
if (setaffinity(targ->thread, targ->affinity))
goto quit;
/* setup poll(2) mechanism. */
memset(fds, 0, sizeof(fds));
fds[0].fd = targ->fd;
fds[0].events = (POLLIN);
/* unbounded wait for the first packet. */
for (;;) {
i = poll(fds, 1, 1000);
if (i > 0 && !(fds[0].revents & POLLERR))
break;
D("waiting for initial packets, poll returns %d %d", i, fds[0].revents);
}
/* main loop, exit after 1s silence */
gettimeofday(&targ->tic, NULL);
if (targ->g->use_pcap) {
for (;;) {
/* XXX should we poll ? */
pcap_dispatch(targ->g->p, targ->g->burst, receive_pcap, NULL);
}
} else {
while (1) {
/* Once we started to receive packets, wait at most 1 seconds
before quitting. */
if (poll(fds, 1, 1 * 1000) <= 0) {
gettimeofday(&targ->toc, NULL);
targ->toc.tv_sec -= 1; /* Subtract timeout time. */
break;
}
for (i = targ->qfirst; i < targ->qlast; i++) {
int m;
rxring = NETMAP_RXRING(nifp, i);
if (rxring->avail == 0)
continue;
m = receive_packets(rxring, targ->g->burst,
SKIP_PAYLOAD);
received += m;
targ->count = received;
}
// tell the card we have read the data
//ioctl(fds[0].fd, NIOCRXSYNC, NULL);
}
}
targ->completed = 1;
targ->count = received;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
static void *
sender_body(void *data)
{
struct targ *targ = (struct targ *) data;
struct pollfd fds[1];
struct netmap_if *nifp = targ->nifp;
struct netmap_ring *txring;
int i, n = targ->g->npackets / targ->g->nthreads, sent = 0;
int options = targ->g->options | OPT_COPY;
D("start");
if (setaffinity(targ->thread, targ->affinity))
goto quit;
/* setup poll(2) mechanism. */
memset(fds, 0, sizeof(fds));
fds[0].fd = targ->fd;
fds[0].events = (POLLOUT);
/* main loop.*/
gettimeofday(&targ->tic, NULL);
if (targ->g->use_pcap) {
int size = targ->g->pkt_size;
void *pkt = &targ->pkt;
pcap_t *p = targ->g->p;
for (i = 0; n == 0 || sent < n; i++) {
if (pcap_inject(p, pkt, size) != -1)
sent++;
if (i > 10000) {
targ->count = sent;
i = 0;
}
}
} else {
while (n == 0 || sent < n) {
/*
* wait for available room in the send queue(s)
*/
if (poll(fds, 1, 2000) <= 0) {
D("poll error/timeout on queue %d", targ->me);
goto quit;
}
/*
* scan our queues and send on those with room
*/
if (options & OPT_COPY && sent > 100000 && !(targ->g->options & OPT_COPY) ) {
D("drop copy");
options &= ~OPT_COPY;
}
for (i = targ->qfirst; i < targ->qlast; i++) {
int m, limit = targ->g->burst;
if (n > 0 && n - sent < limit)
limit = n - sent;
txring = NETMAP_TXRING(nifp, i);
if (txring->avail == 0)
continue;
m = send_packets(txring, &targ->pkt, targ->g->pkt_size,
limit, options);
sent += m;
targ->count = sent;
}
}
/* flush any remaining packets */
ioctl(fds[0].fd, NIOCTXSYNC, NULL);
/* final part: wait all the TX queues to be empty. */
for (i = targ->qfirst; i < targ->qlast; i++) {
txring = NETMAP_TXRING(nifp, i);
while (!NETMAP_TX_RING_EMPTY(txring)) {
ioctl(fds[0].fd, NIOCTXSYNC, NULL);
usleep(1); /* wait 1 tick */
}
}
}
gettimeofday(&targ->toc, NULL);
targ->completed = 1;
targ->count = sent;
quit:
/* reset the ``used`` flag. */
targ->used = 0;
return (NULL);
}
