Before: 1Gping-throughput.sh: Seconds: 20 1Gping-throughput.sh: Packets: 1000001,1000004 1Gping-throughput.sh: Interrupts: 966239,1 1Gping-throughput.sh: Notifications: 1089(14275),963192(36813) Net IRQs triggered: 966443(12192),1(962475) 1Mping.sh: Seconds: 35 1Mping.sh: Packets: 1000001,1000004 1Mping.sh: Interrupts: 1000001,1 1Mping.sh: Notifications: 15385(0),1000004(1) Net IRQs triggered: 1000001(0),1(1000004) After: 1Gping-throughput.sh: Seconds: 23 1Gping-throughput.sh: Packets: 1000001,1000004 1Gping-throughput.sh: Interrupts: 999906,1 1Gping-throughput.sh: Notifications: 15385(0),999966(39) Net IRQs triggered: 999932(16),1(999918) 1Mping.sh: Seconds: 24 1Mping.sh: Packets: 1000001,1000004 1Mping.sh: Interrupts: 1000001,1 1Mping.sh: Notifications: 15385(0),1000003(2) Net IRQs triggered: 1000001(5),1(1000003) --- Documentation/lguest/lguest.c | 154 +++++++++++++++++++----------------------- 1 file changed, 71 insertions(+), 83 deletions(-) diff --git a/Documentation/lguest/lguest.c b/Documentation/lguest/lguest.c --- a/Documentation/lguest/lguest.c +++ b/Documentation/lguest/lguest.c @@ -799,8 +799,6 @@ static void console_output(struct virtqu * kernel network stack. */ struct net_info { - /* We write packets to the pipe. */ - int p[2]; /* Our hardware address */ unsigned char ethaddr[ETH_ALEN]; unsigned char ipaddr[4]; @@ -887,45 +885,69 @@ static void csum_replace(__u16 *sum, u32 *sum = csum_fold(csum_partial(diff, sizeof(diff), *sum ^ 0xFFFF)); } -static unsigned int copy_iov(char *buffer, - const struct iovec *iov, unsigned iovnum, - unsigned buflen) +static unsigned int copy_iov(struct iovec *dstiov, unsigned dstnum, + const struct iovec *srciov, unsigned srcnum) { - unsigned int len = 0; + unsigned int dstoff = 0, srcoff = 0; + unsigned int totlen = 0; - while (iovnum) { - if (len + iov->iov_len > buflen) - err(1, "iovec too long"); - memcpy(buffer + len, iov->iov_base, iov->iov_len); - len += iov->iov_len; - iovnum--; - iov++; + while (srcnum) { + unsigned int len; + + if (dstiov->iov_len - dstoff > srciov->iov_len - srcoff) + len = srciov->iov_len - srcoff; + else + len = dstiov->iov_len - dstoff; + + memcpy(dstiov->iov_base + dstoff, srciov->iov_base + srcoff, + len); + dstoff += len; + srcoff += len; + totlen += len; + if (dstoff == dstiov->iov_len) { + dstiov++; + dstnum--; + if (dstnum == 0) + errx(1, "dst buf too short (%u)", totlen); + dstoff = 0; + } + if (srcoff == srciov->iov_len) { + srciov++; + srcnum--; + srcoff = 0; + } } - - return len; + return totlen; } -static void net_output(struct virtqueue *vq) +static void do_net_output(struct net_info *net_info, + struct virtqueue *outvq, struct virtqueue *invq) { - struct net_info *net_info = vq->dev->priv; - unsigned int head, out, in; - struct iovec iov[vq->vring.num]; + unsigned int out_head, out, in, in_head, dummy; + struct iovec outiov[outvq->vring.num], iniov[invq->vring.num]; struct ether_header *ehdr; - char pkt[2 + sizeof(struct virtio_net_hdr) + 1514]; unsigned int len; - vq->irq = &xmit_irq; - vq->irq_suppressed = &xmit_irq_suppressed; + in_head = wait_for_vq_desc(invq, iniov, &dummy, &in); + if (dummy) + errx(1, "Output buffers in net input queue?"); - head = wait_for_vq_desc(vq, iov, &out, &in); - if (in) + /* If we're going to block, make sure we trigger receive first. */ + if (lg_last_avail(outvq) == outvq->vring.avail->idx) + trigger_irq(invq); + + out_head = wait_for_vq_desc(outvq, outiov, &out, &dummy); + if (dummy) errx(1, "Input buffers in net output queue?"); - len = copy_iov(pkt+2, iov, out, sizeof(pkt)); + /* We transfer and operate on packet inside guest. This only works + * so well because guest currently uses v. simple buffer layout. */ + len = copy_iov(iniov, in, outiov, out); + if (len < sizeof(struct virtio_net_hdr) + 14) errx(1, "Short packet %u", len); - ehdr = (struct ether_header *)(pkt + 2 + sizeof(struct virtio_net_hdr)); + ehdr = iniov[1].iov_base; len -= sizeof(struct virtio_net_hdr); if (ehdr->ether_type == htons(ETHERTYPE_ARP)) { struct arphdr *arp = (struct arphdr *)(ehdr + 1); @@ -973,67 +995,36 @@ static void net_output(struct virtqueue /* Copy our ethaddr to source ethaddr. */ memcpy(ehdr->ether_shost, net_info->ethaddr, ETH_ALEN); - /* We put the lenght first, to packetize. */ - *(u16 *)pkt = len + sizeof(struct virtio_net_hdr); - if (write(net_info->p[1], pkt, 2 + sizeof(struct virtio_net_hdr) + len) - != 2 + sizeof(struct virtio_net_hdr) + len) - err(1, "Writing to pipe"); + add_used(invq, in_head, len+sizeof(struct virtio_net_hdr)); consume: - add_used(vq, head, len+sizeof(struct virtio_net_hdr)); + add_used(outvq, out_head, len+sizeof(struct virtio_net_hdr)); } -/* Will reading from this file descriptor block? */ -static bool will_block(int fd) +static void net_output(struct virtqueue *outvq) { - fd_set fdset; - struct timeval zero = { 0, 0 }; - FD_ZERO(&fdset); - FD_SET(fd, &fdset); - return select(fd+1, &fdset, NULL, NULL, &zero) != 1; -} + struct net_info *net_info = outvq->dev->priv; + struct virtqueue *invq = outvq->dev->vq; + unsigned long args[] = { LHREQ_EVENTFD, + invq->config.pfn*getpagesize(), 0 }; -static void trim_iov(struct iovec *iov, unsigned num_iov, unsigned len) -{ - unsigned int i; + /* Create a zero-initialized eventfd. */ + invq->eventfd = eventfd(0, 0); + if (invq->eventfd < 0) + err(1, "Creating eventfd"); + args[2] = invq->eventfd; - for (i = 0; i < num_iov; i++) { - if (iov[i].iov_len > len) - iov[i].iov_len = len; - len -= iov[i].iov_len; - } -} + /* Attach an eventfd for input virtqueue. */ + if (write(lguest_fd, &args, sizeof(args)) != 0) + err(1, "Attaching eventfd"); -/* This is where we handle packets coming in from the tun device to our - * Guest. */ -static void net_input(struct virtqueue *vq) -{ - int len; - unsigned int head, out, in; - struct iovec iov[vq->vring.num]; - u16 pkt_len; - struct net_info *net_info = vq->dev->priv; + /* Set up stats. */ + invq->irq = &recv_irq; + invq->irq_suppressed = &recv_irq_suppressed; + outvq->irq = &xmit_irq; + outvq->irq_suppressed = &xmit_irq_suppressed; - vq->irq = &recv_irq; - vq->irq_suppressed = &recv_irq_suppressed; - - head = wait_for_vq_desc(vq, iov, &out, &in); - if (out) - errx(1, "Output buffers in net input queue?"); - - /* Deliver interrupt now, since we're about to sleep. */ - if (vq->pending_used && will_block(net_info->p[0])) - trigger_irq(vq); - - if (read(net_info->p[0], &pkt_len, 2) != 2) - err(1, "Failed to read length from pipe."); - - /* Don't read in more than the one packet! */ - trim_iov(iov, in, pkt_len); - - len = readv(net_info->p[0], iov, in); - if (len != pkt_len) - err(1, "Failed to read from pipe: %u vs %u", len, pkt_len); - add_used(vq, head, len); + for (;;) + do_net_output(net_info, outvq, invq); } /* This is the helper to create threads. */ @@ -1430,16 +1421,13 @@ static void setup_dummy_net(char *arg) struct virtio_net_config conf; u32 ip; - if (pipe(net_info->p) != 0) - err(1, "Creating net pipe"); - /* First we create a new network device. */ dev = new_device("net", VIRTIO_ID_NET); dev->priv = net_info; /* Network devices need a receive and a send queue, just like * console. */ - add_virtqueue(dev, VIRTQUEUE_NUM, net_input); + add_virtqueue(dev, VIRTQUEUE_NUM, NULL); add_virtqueue(dev, VIRTQUEUE_NUM, net_output); /* Set IP address. */