Date:   Wed, 17 Jan 2018 15:06:11 -0800
From:   Jesus Sanchez-Palencia <jesus.sanchez-palencia@...el.com>
To:     netdev@...r.kernel.org
Cc:     jhs@...atatu.com, xiyou.wangcong@...il.com, jiri@...nulli.us,
        vinicius.gomes@...el.com, richardcochran@...il.com,
        intel-wired-lan@...ts.osuosl.org, anna-maria@...utronix.de,
        henrik@...tad.us, tglx@...utronix.de, john.stultz@...aro.org,
        andre.guedes@...el.com, ivan.briano@...el.com,
        levi.pearson@...man.com
Subject: [RFC v2 net-next 00/10] Time based packet transmission

This series is the v2 of the Time based packet transmission RFC, which was
originally proposed by Richard Cochran: https://lwn.net/Articles/733962/ .

It introduces a new socket option (SO_TXTIME), a new qdisc (tbs) and implements
support for hw offloading on the igb driver for the Intel i210 NIC. The tbs
qdisc also supports SW best effort that can be used as a fallback.

The main changes since v1 are:
 - the tstamp field from sk_buffs is now used;
 - ktime_t is the type used for the field added to struct sockcm_cookie instead
   of u64;
 - the tbs qdisc is introduced with SW best effort and hw offloading;
 - the igb implementation for HW offloading was re-written, allowing both tbs
   and cbs qdiscs to co-exist with proper driver support.

The tbs qdisc is designed so it buffers packets until a configurable time before
their deadline (tx times). It uses a rbtree internally, thus the buffered
packets are always 'ordered' by the earliest deadline.

The other configurable parameter from the tbs qdisc is the clockid to be used.
In order to provide that, this series adds a new API to pkt_sched.h (i.e.
qdisc_watchdog_init_clockid()).

As an usage example:

    $ tc qdisc replace dev enp2s0 parent root handle 100 mqprio num_tc 3 \
               map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0

    $ tc qdisc add dev enp2s0 parent 100:1 tbs delta 60000 clockid 11 \
               offload 1

In this example first the mqprio qdisc is setup, then the tbs qdisc is
configured onto the first hw Tx queue and tries to enable HW offloading (i.e.
offload 1). Also, it is configured so the timestamps on each packet are in
reference to the clockid '11' (CLOCK_TAI) and so packets are dequeued from
the qdisc 60000 nanoseconds before their transmission time.

The tbs qdisc will drop any packets with a transmission time in the past or
when a deadline is missed. Queueing packets in advance plus configuring the
delta parameter for the system correctly makes all the difference in reducing
the number of drops. Moreover, note that the delta parameter ends up defining
the Tx time when SW best effort is used given that the timestamps won't be used
by the NIC on this case.

For testing, we've followed a similar approach from the v1 testing:
  1. Prepared a PC and the Device Under Test (DUT) each with an Intel
     i210 card connected.
  2. The DUT was a Intel(R) Core(TM) i5-7600 CPU @ 3.50GHz running on top of
     kernel 4.15.0-rc8+ with about 50 usec maximum latency under cyclictest.
  3. Synchronized the DUT's PHC to the PC's PHC using ptp4l.
  4. Synchronized the DUT's system clock to its PHC using phc2sys.
  5. Measured the arrival time of the packets at the PC's PHC using
     hardware time stamping.

First, a baseline test was ran for 10 minutes with the plain kernel only:

|                 | plain kernel @ 1ms |
|-----------------+--------------------+
| min (ns):       |    +4.820000e+02   |
| max (ns):       |    +9.999300e+05   |
| pk-pk:          |    +9.994480e+05   |
| mean (ns):      |    +3.464421e+04   |
| stddev:         |    +1.305947e+05   |
| count:          |           600000   |


Tests were then ran for 10 minutes with a period of 1 millisecond using both
SW best effort and HW offloading. For last, we repeated the HW offloading test
with a 250 microsecond period. The measured offset from the expected period is
shown below, plus the tbs delta parameter that was used in each case.

|                 |    tbs SW @ 1ms   |  tbs HW @ 1ms  | tbs HW @ 250 us |
|-----------------+-------------------+----------------+-----------------|
| min (ns):       |    +1.510000e+02  |  +4.420000e+02 |   +4.260000e+02 |
| max (ns):       |    +9.977030e+05  |  +5.060000e+02 |   +5.060000e+02 |
| pk-pk:          |    +9.975520e+05  |  +6.400000e+01 |   +8.000000e+01 |
| mean (ns):      |    +1.416511e+04  |  +4.687228e+02 |   +4.600596e+02 |
| stddev:         |    +5.750639e+04  |  +9.868569e+00 |   +1.287626e+01 |
| count:          |           600000  |         600000 |         2400000 |
| dropped:        |                3  |              0 |               0 |
| tbs delta (ns): |           130000  |         130000 |          130000 |


The code used for testing is appended below. The wake_tx parameter (-d) used
for all tests was 600000 ns and the priority parameter was 90 (-p). The
baseline test (plain kernel) used a wake_tx parameter of 130000 ns.

Our main questions at this stage are related to the qdisc:
 - does the proposed design attend all use cases?
 - should the qdisc really drop packets that expired after being queued even
   for the SW best effort mode?
 - once one expired packet is found and dropped during a dequeue, should we
   traverse the rbtree and drop other expired packets if any, or should we
   keep deferring that to the next dequeue call?

For last, most of the To Dos we still have before a final patchset are related
to further testing the igb support:
 - testing with L2 only talkers + AF_PACKET sockets;
 - testing tbs in conjunction with cbs;

Thanks,
Jesus


Jesus Sanchez-Palencia (4):
  igb: Refactor igb_configure_cbs()
  igb: Only change Tx arbitration when CBS is on
  igb: Refactor igb_offload_cbs()
  igb: Add support for TBS offload

Richard Cochran (4):
  net: Add a new socket option for a future transmit time.
  net: ipv4: raw: Hook into time based transmission.
  net: ipv4: udp: Hook into time based transmission.
  net: packet: Hook into time based transmission.

Vinicius Costa Gomes (2):
  net/sched: Allow creating a Qdisc watchdog with other clocks
  net/sched: Introduce the TBS Qdisc

 arch/alpha/include/uapi/asm/socket.h           |   3 +
 arch/frv/include/uapi/asm/socket.h             |   3 +
 arch/ia64/include/uapi/asm/socket.h            |   3 +
 arch/m32r/include/uapi/asm/socket.h            |   3 +
 arch/mips/include/uapi/asm/socket.h            |   3 +
 arch/mn10300/include/uapi/asm/socket.h         |   3 +
 arch/parisc/include/uapi/asm/socket.h          |   3 +
 arch/s390/include/uapi/asm/socket.h            |   3 +
 arch/sparc/include/uapi/asm/socket.h           |   3 +
 arch/xtensa/include/uapi/asm/socket.h          |   3 +
 drivers/net/ethernet/intel/igb/e1000_defines.h |  16 +
 drivers/net/ethernet/intel/igb/igb.h           |   1 +
 drivers/net/ethernet/intel/igb/igb_main.c      | 239 +++++++++++----
 include/linux/netdevice.h                      |   1 +
 include/net/pkt_sched.h                        |   7 +
 include/net/sock.h                             |   2 +
 include/uapi/asm-generic/socket.h              |   3 +
 include/uapi/linux/pkt_sched.h                 |  17 ++
 net/core/sock.c                                |  16 +
 net/ipv4/raw.c                                 |   2 +
 net/ipv4/udp.c                                 |   5 +-
 net/packet/af_packet.c                         |   6 +
 net/sched/Kconfig                              |  11 +
 net/sched/Makefile                             |   1 +
 net/sched/sch_api.c                            |  11 +-
 net/sched/sch_tbs.c                            | 392 +++++++++++++++++++++++++
 26 files changed, 699 insertions(+), 61 deletions(-)
 create mode 100644 net/sched/sch_tbs.c

-- 
2.15.1

---8<---
/*
 * This program demonstrates transmission of UDP packets using the
 * system TAI timer.
 *
 * Copyright (C) 2017 linutronix GmbH
 *
 * Large portions taken from the linuxptp stack.
 * Copyright (C) 2011, 2012 Richard Cochran <richardcochran@...il.com>
 *
 * Some portions taken from the sgd test program.
 * Copyright (C) 2015 linutronix GmbH
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along
 * with this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
 */
#define _GNU_SOURCE /*for CPU_SET*/
#include <arpa/inet.h>
#include <errno.h>
#include <fcntl.h>
#include <ifaddrs.h>
#include <linux/ethtool.h>
#include <linux/net_tstamp.h>
#include <linux/sockios.h>
#include <net/if.h>
#include <netinet/in.h>
#include <poll.h>
#include <pthread.h>
#include <sched.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>

#define DEFAULT_PERIOD	1000000
#define DEFAULT_DELAY	500000
#define MCAST_IPADDR	"239.1.1.1"
#define UDP_PORT	7788

#ifndef SO_TXTIME
#define SO_TXTIME	61
#endif

#define pr_err(s)	fprintf(stderr, s "\n")
#define pr_info(s)	fprintf(stdout, s "\n")

static int running = 1, use_so_txtime = 1;
static int period_nsec = DEFAULT_PERIOD;
static int waketx_delay = DEFAULT_DELAY;
static struct in_addr mcast_addr;

static int mcast_bind(int fd, int index)
{
	int err;
	struct ip_mreqn req;
	memset(&req, 0, sizeof(req));
	req.imr_ifindex = index;
	err = setsockopt(fd, IPPROTO_IP, IP_MULTICAST_IF, &req, sizeof(req));
	if (err) {
		pr_err("setsockopt IP_MULTICAST_IF failed: %m");
		return -1;
	}
	return 0;
}

static int mcast_join(int fd, int index, const struct sockaddr *grp,
		      socklen_t grplen)
{
	int err, off = 0;
	struct ip_mreqn req;
	struct sockaddr_in *sa = (struct sockaddr_in *) grp;

	memset(&req, 0, sizeof(req));
	memcpy(&req.imr_multiaddr, &sa->sin_addr, sizeof(struct in_addr));
	req.imr_ifindex = index;
	err = setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &req, sizeof(req));
	if (err) {
		pr_err("setsockopt IP_ADD_MEMBERSHIP failed: %m");
		return -1;
	}
	err = setsockopt(fd, IPPROTO_IP, IP_MULTICAST_LOOP, &off, sizeof(off));
	if (err) {
		pr_err("setsockopt IP_MULTICAST_LOOP failed: %m");
		return -1;
	}
	return 0;
}

static void normalize(struct timespec *ts)
{
	while (ts->tv_nsec > 999999999) {
		ts->tv_sec += 1;
		ts->tv_nsec -= 1000000000;
	}
}

static int sk_interface_index(int fd, const char *name)
{
	struct ifreq ifreq;
	int err;

	memset(&ifreq, 0, sizeof(ifreq));
	strncpy(ifreq.ifr_name, name, sizeof(ifreq.ifr_name) - 1);
	err = ioctl(fd, SIOCGIFINDEX, &ifreq);
	if (err < 0) {
		pr_err("ioctl SIOCGIFINDEX failed: %m");
		return err;
	}
	return ifreq.ifr_ifindex;
}

static int open_socket(const char *name, struct in_addr mc_addr, short port)
{
	struct sockaddr_in addr;
	int fd, index, on = 1;
	int priority = 3;

	memset(&addr, 0, sizeof(addr));
	addr.sin_family = AF_INET;
	addr.sin_addr.s_addr = htonl(INADDR_ANY);
	addr.sin_port = htons(port);

	fd = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);
	if (fd < 0) {
		pr_err("socket failed: %m");
		goto no_socket;
	}
	index = sk_interface_index(fd, name);
	if (index < 0)
		goto no_option;

	if (setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &priority, sizeof(priority))) {
		pr_err("Couldn't set priority");
		goto no_option;
	}
	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on))) {
		pr_err("setsockopt SO_REUSEADDR failed: %m");
		goto no_option;
	}
	if (bind(fd, (struct sockaddr *) &addr, sizeof(addr))) {
		pr_err("bind failed: %m");
		goto no_option;
	}
	if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, name, strlen(name))) {
		pr_err("setsockopt SO_BINDTODEVICE failed: %m");
		goto no_option;
	}
	addr.sin_addr = mc_addr;
	if (mcast_join(fd, index, (struct sockaddr *) &addr, sizeof(addr))) {
		pr_err("mcast_join failed");
		goto no_option;
	}
	if (mcast_bind(fd, index)) {
		goto no_option;
	}
	if (use_so_txtime && setsockopt(fd, SOL_SOCKET, SO_TXTIME, &on, sizeof(on))) {
		pr_err("setsockopt SO_TXTIME failed: %m");
		goto no_option;
	}

	return fd;
no_option:
	close(fd);
no_socket:
	return -1;
}

static int udp_open(const char *name)
{
	int fd;

	if (!inet_aton(MCAST_IPADDR, &mcast_addr))
		return -1;

	fd = open_socket(name, mcast_addr, UDP_PORT);

	return fd;
}

static int udp_send(int fd, void *buf, int len, __u64 txtime)
{
	union {
		char buf[CMSG_SPACE(sizeof(__u64))];
		struct cmsghdr align;
	} u;
	struct sockaddr_in sin;
	struct cmsghdr *cmsg;
	struct msghdr msg;
	struct iovec iov;
	ssize_t cnt;

	memset(&sin, 0, sizeof(sin));
	sin.sin_family = AF_INET;
	sin.sin_addr = mcast_addr;
	sin.sin_port = htons(UDP_PORT);

	iov.iov_base = buf;
	iov.iov_len = len;

	memset(&msg, 0, sizeof(msg));
	msg.msg_name = &sin;
	msg.msg_namelen = sizeof(sin);
	msg.msg_iov = &iov;
	msg.msg_iovlen = 1;

	/*
	 * We specify the transmission time in the CMSG.
	 */
	if (use_so_txtime) {
		msg.msg_control = u.buf;
		msg.msg_controllen = sizeof(u.buf);
		cmsg = CMSG_FIRSTHDR(&msg);
		cmsg->cmsg_level = SOL_SOCKET;
		cmsg->cmsg_type = SO_TXTIME;
		cmsg->cmsg_len = CMSG_LEN(sizeof(__u64));
		*((__u64 *) CMSG_DATA(cmsg)) = txtime;
	}
	cnt = sendmsg(fd, &msg, 0);
	if (cnt < 1) {
		pr_err("sendmsg failed: %m");
		return cnt;
	}
	return cnt;
}

static unsigned char tx_buffer[256];
static int marker;

static int run_nanosleep(clockid_t clkid, int fd)
{
	struct timespec ts;
	int cnt, err;
	__u64 txtime;

	clock_gettime(clkid, &ts);

	/* Start one to two seconds in the future. */
	ts.tv_sec += 1;
	ts.tv_nsec = 1000000000 - waketx_delay;
	normalize(&ts);

	txtime = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
	txtime += waketx_delay;

	while (running) {
		err = clock_nanosleep(clkid, TIMER_ABSTIME, &ts, NULL);
		switch (err) {
		case 0:
			cnt = udp_send(fd, tx_buffer, sizeof(tx_buffer), txtime);
			if (cnt != sizeof(tx_buffer)) {
				pr_err("udp_send failed");
			}
			memset(tx_buffer, marker++, sizeof(tx_buffer));
			ts.tv_nsec += period_nsec;
			normalize(&ts);
			txtime += period_nsec;
			break;
		case EINTR:
			continue;
		default:
			fprintf(stderr, "clock_nanosleep returned %d: %s",
				err, strerror(err));
			return err;
		}
	}

	return 0;
}

static int set_realtime(pthread_t thread, int priority, int cpu)
{
	cpu_set_t cpuset;
	struct sched_param sp;
	int err, policy;

	int min = sched_get_priority_min(SCHED_FIFO);
	int max = sched_get_priority_max(SCHED_FIFO);

	fprintf(stderr, "min %d max %d\n", min, max);

	if (priority < 0) {
		return 0;
	}

	err = pthread_getschedparam(thread, &policy, &sp);
	if (err) {
		fprintf(stderr, "pthread_getschedparam: %s\n", strerror(err));
		return -1;
	}

	sp.sched_priority = priority;

	err = pthread_setschedparam(thread, SCHED_FIFO, &sp);
	if (err) {
		fprintf(stderr, "pthread_setschedparam: %s\n", strerror(err));
		return -1;
	}

	if (cpu < 0) {
		return 0;
	}
	CPU_ZERO(&cpuset);
	CPU_SET(cpu, &cpuset);
	err = pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
	if (err) {
		fprintf(stderr, "pthread_setaffinity_np: %s\n", strerror(err));
		return -1;
	}

	return 0;
}

static void usage(char *progname)
{
	fprintf(stderr,
		"\n"
		"usage: %s [options]\n"
		"\n"
		" -c [num]   run on CPU 'num'\n"
		" -d [num]   delay from wake up to transmission in nanoseconds (default %d)\n"
		" -h         prints this message and exits\n"
		" -i [name]  use network interface 'name'\n"
		" -p [num]   run with RT priorty 'num'\n"
		" -P [num]   period in nanoseconds (default %d)\n"
		" -u         do not use SO_TXTIME\n"
		"\n",
		progname, DEFAULT_DELAY, DEFAULT_PERIOD);
}

int main(int argc, char *argv[])
{
	int c, cpu = -1, err, fd, priority = -1;
	clockid_t clkid = CLOCK_TAI;
	char *iface = NULL, *progname;

	/* Process the command line arguments. */
	progname = strrchr(argv[0], '/');
	progname = progname ? 1 + progname : argv[0];
	while (EOF != (c = getopt(argc, argv, "c:d:hi:p:P:u"))) {
		switch (c) {
		case 'c':
			cpu = atoi(optarg);
			break;
		case 'd':
			waketx_delay = atoi(optarg);
			break;
		case 'h':
			usage(progname);
			return 0;
		case 'i':
			iface = optarg;
			break;
		case 'p':
			priority = atoi(optarg);
			break;
		case 'P':
			period_nsec = atoi(optarg);
			break;
		case 'u':
			use_so_txtime = 0;
			break;
		case '?':
			usage(progname);
			return -1;
		}
	}

	if (waketx_delay > 999999999 || waketx_delay < 0) {
		pr_err("Bad wake up to transmission delay.");
		usage(progname);
		return -1;
	}

	if (period_nsec < 1000) {
		pr_err("Bad period.");
		usage(progname);
		return -1;
	}

	if (!iface) {
		pr_err("Need a network interface.");
		usage(progname);
		return -1;
	}

	if (set_realtime(pthread_self(), priority, cpu)) {
		return -1;
	}

	fd = udp_open(iface);
	if (fd < 0) {
		return -1;
	}

	err = run_nanosleep(clkid, fd);

	close(fd);
	return err;
}

Powered by blists - more mailing lists