From patchwork Tue Sep 15 08:38:36 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Moessbauer, Felix" X-Patchwork-Id: 77702 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 62D5FA04C7; Tue, 15 Sep 2020 10:39:18 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 944A01C0D1; Tue, 15 Sep 2020 10:39:08 +0200 (CEST) Received: from gecko.sbs.de (gecko.sbs.de [194.138.37.40]) by dpdk.org (Postfix) with ESMTP id 464BB1C0D0 for ; Tue, 15 Sep 2020 10:39:07 +0200 (CEST) Received: from mail1.sbs.de (mail1.sbs.de [192.129.41.35]) by gecko.sbs.de (8.15.2/8.15.2) with ESMTPS id 08F8d3Xv010469 (version=TLSv1.2 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK); Tue, 15 Sep 2020 10:39:03 +0200 Received: from derbion.ppmd.siemens.net (derbion.ppmd.siemens.net [139.25.68.102]) by mail1.sbs.de (8.15.2/8.15.2) with ESMTP id 08F8cuB9007442; Tue, 15 Sep 2020 10:39:03 +0200 From: Felix Moessbauer To: thomas@monjalon.net Cc: dev@dpdk.org, ktraynor@redhat.com, ali.rizvi@emumba.com, muhammad.ahamd@emumba.com, david.hunt@intel.com, reshma.pattan@intel.com, konstantin.ananyev@intel.com, bruce.richardson@intel.com, ferruh.yigit@intel.com, jerinj@marvell.com, honnappa.nagarahalli@arm.com, olivier.matz@6wind.com, lijuan.tu@intel.com, ohilyard@iol.unh.edu, mb@smartsharesystems.com, viacheslavo@nvidia.com, stephen@networkplumber.org, anatoly.burakov@intel.com, Felix Moessbauer , Henning Schild Date: Tue, 15 Sep 2020 10:38:36 +0200 Message-Id: <20200915083836.18805-3-felix.moessbauer@siemens.com> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20200715155409.9835-3-felix.moessbauer@siemens.com> References: <20200715155409.9835-3-felix.moessbauer@siemens.com> MIME-Version: 1.0 Subject: [dpdk-dev] [PATCH 2/2] Add l2reflect measurement application X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" The l2reflect application implements a ping-pong benchmark to measure the latency between two instances. For communication, we use raw ethernet and send one packet at a time. The timing data is collected locally and min/max/avg values are displayed in a TUI. Finally, a histogram of the latencies is printed which can be further processed with the jitterdebugger visualization scripts. To debug latency spikes, a max threshold can be defined. If it is hit, a trace point is created on both instances. Signed-off-by: Felix Moessbauer Signed-off-by: Henning Schild --- app/l2reflect/colors.c | 34 ++ app/l2reflect/colors.h | 19 + app/l2reflect/l2reflect.h | 48 ++ app/l2reflect/main.c | 907 ++++++++++++++++++++++++++++++++++++++ app/l2reflect/meson.build | 25 ++ app/l2reflect/stats.c | 178 ++++++++ app/l2reflect/stats.h | 61 +++ app/meson.build | 1 + 8 files changed, 1273 insertions(+) create mode 100644 app/l2reflect/colors.c create mode 100644 app/l2reflect/colors.h create mode 100644 app/l2reflect/l2reflect.h create mode 100644 app/l2reflect/main.c create mode 100644 app/l2reflect/meson.build create mode 100644 app/l2reflect/stats.c create mode 100644 app/l2reflect/stats.h diff --git a/app/l2reflect/colors.c b/app/l2reflect/colors.c new file mode 100644 index 000000000..af881d878 --- /dev/null +++ b/app/l2reflect/colors.c @@ -0,0 +1,34 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + */ + +#include "colors.h" + +const struct color_palette *colors; + +static const struct color_palette color_palette_default = { + .red = "\x1b[01;31m", + .green = "\x1b[01;32m", + .yellow = "\x1b[01;33m", + .blue = "\x1b[01;34m", + .magenta = "\x1b[01;35m", + .cyan = "\x1b[01;36m", + .reset = "\x1b[0m" +}; + +static const struct color_palette color_palette_bw = { .red = "", + .green = "", + .yellow = "", + .blue = "", + .magenta = "", + .cyan = "", + .reset = "" }; + +void +enable_colors(int enable) +{ + if (enable) + colors = &color_palette_default; + else + colors = &color_palette_bw; +} diff --git a/app/l2reflect/colors.h b/app/l2reflect/colors.h new file mode 100644 index 000000000..82d33a8b8 --- /dev/null +++ b/app/l2reflect/colors.h @@ -0,0 +1,19 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + */ +#ifndef _COLORS_H_ +#define _COLORS_H_ + +/* posix terminal colors */ +struct color_palette { + const char *red, *green, *yellow, *blue, *magenta, *cyan, *reset; +}; + +/* ptr to the current tui color palette */ +extern const struct color_palette *colors; + +/* disable colored output */ +void +enable_colors(int enable); + +#endif /* _COLORS_H_ */ diff --git a/app/l2reflect/l2reflect.h b/app/l2reflect/l2reflect.h new file mode 100644 index 000000000..e7b19d537 --- /dev/null +++ b/app/l2reflect/l2reflect.h @@ -0,0 +1,48 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + */ +#ifndef _L2REFLECT_H_ +#define _L2REFLECT_H_ +#define MAGIC_TRACE_PAYLOAD 0xd00faffeaffed00full +/* IEEE Std 802 - Local Experimental Ethertype */ +#define ETHER_TYPE_L2REFLECT 0x88B5 +/* Used to compare MAC addresses. */ +#define MAC_ADDR_CMP 0xFFFFFFFFFFFFull + +enum { + TRACE_TYPE_DATA, + TRACE_TYPE_HELO, + TRACE_TYPE_EHLO, + TRACE_TYPE_RSET, + TRACE_TYPE_QUIT, +}; + +struct my_magic_packet { + struct rte_ether_hdr eth; + uint8_t type; + uint64_t magic; + uint64_t breakval; +}; + +enum STATE { + /* elect the initial sender */ + S_ELECT_LEADER = 1, + /* reset the counters */ + S_RESET_TRX = 2, + /* measurement S_RUNNING */ + S_RUNNING = 4, + /* terminated by local event */ + S_LOCAL_TERM = 8, + /* terminated by remote event */ + S_REMOTE_TERM = 16 +}; + +extern int l2reflect_hist; +extern atomic_int l2reflect_output_hist; +extern uint64_t l2reflect_sleep_msec; +extern uint16_t l2reflect_port_number; +extern atomic_int l2reflect_state; +extern struct rte_ether_addr l2reflect_port_eth_addr; +extern struct rte_ether_addr l2reflect_remote_eth_addr; + +#endif /* _L2REFLECT_H_ */ diff --git a/app/l2reflect/main.c b/app/l2reflect/main.c new file mode 100644 index 000000000..23ace81a5 --- /dev/null +++ b/app/l2reflect/main.c @@ -0,0 +1,907 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + * + * The l2reflect application implements a ping-pong benchmark to + * measure the latency between two instances. For communication, + * we use raw ethernet and send one packet at a time. The timing data + * is collected locally and min/max/avg values are displayed in a TUI. + * Finally, a histogram of the latencies is printed which can be + * further processed with the jitterdebugger visualization scripts. + * To debug latency spikes, a max threshold can be defined. + * If it is hit, a trace point is created on both instances. + * + * Examples: + * launch (non-rt kernel): l2reflect --lcores 0@0,1@6 -n 1 + * launch (rt kernel): l2reflect --lcores 0@0,1@6 -n 1 -- -P 50 -r -l + * + * For histogram data, launch with -H -F , e.g. + * -H 10 -F histogram.json for a histogram with 10 usec buckets which + * is written to a histogram.json file. This file can then be visualized + * using the jitterdebugger plotting scripts: + * jitterplot hist histogram.json + * + * While the application is running, it can be controlled by sending + * signals to one of the processes: + * - SIGUSR1: reset the min/max/avg on both instances + * - SIGUSR2: output / write the current histogram + * - SIGHUB/SIGINT: gracefully terminate both instances + * + * Note on wiring: + * The l2reflect application sends the packets via a physical ethernet + * interface. When running both instances on a single system, at least + * two dedicated physical ports and a (physical) loopback between them + * is required. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#ifdef HAS_SYS_IO +#include +#endif +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "l2reflect.h" +#include "colors.h" +#include "stats.h" + +#define RTE_LOGTYPE_L2REFLECT RTE_LOGTYPE_USER1 + +#define NSEC_PER_SEC 1000000000 + +#define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) +#define NB_MBUF 2047 + +#define MAX_PKT_BURST 32 +/* warmup a few round before starting the measurement */ +#define WARMUP_ROUNDS 42 + +#define MAX(x, y) (((x) > (y)) ? (x) : (y)) +#define MIN(x, y) (((x) < (y)) ? (x) : (y)) + +int l2reflect_hist; +atomic_int l2reflect_output_hist; +uint64_t l2reflect_sleep_msec; +uint16_t l2reflect_port_number; +atomic_int l2reflect_state; +struct rte_ether_addr l2reflect_port_eth_addr; +struct rte_ether_addr l2reflect_remote_eth_addr; + + +static struct timespec last_sent, last_recv; +static int quiet, disable_int, priority, policy, l2reflect_mlock, + l2reflect_interrupt, trace_fd; + +static atomic_int sleep_start; + +static uint64_t rounds; +static int quiet, disable_int, priority, policy, l2reflect_mlock, + l2reflect_interrupt, trace_fd; + +/* Configurable number of RX/TX ring descriptors */ +#define RTE_TEST_RX_DESC_DEFAULT 128 +#define RTE_TEST_TX_DESC_DEFAULT 512 +static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; +static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; + +static struct rte_eth_conf port_conf = { + .rxmode = { + .split_hdr_size = 0, + }, + .txmode = { + .mq_mode = ETH_MQ_TX_NONE, + }, +}; + +static uint32_t l2reflect_q; +static uint64_t l2reflect_break_usec = UINT64_MAX; +static uint64_t l2reflect_break_usec; + +static struct rte_ether_addr ether_bcast_addr = { + .addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff } +}; + +struct rte_mempool *l2reflect_pktmbuf_pool; + +static void +l2reflect_usage(const char *prgname) +{ + printf("%s [EAL options] -- [-p PORT] -P [PRIO] [-b USEC] [-r] [-f] [-l]" + "[-s] [-q] [-d] [-S] [-i MSEC] [-c]\n" + " -p PORT: port to configure\n" + " -P PRIO: scheduling priority to use\n" + " -b USEC: break when latency > USEC\n" + " -r: scheduling policy SCHED_RR\n" + " -f: scheduling policy SCHED_FIFO\n" + " -l: lock memory (mlockall)\n" + " -s: send one packet on startup\n" + " -q: quiet, do not print stats\n" +#ifdef HAS_SYS_IO + " -d: die die die cli\n" +#endif + " -H USEC: create histogram of latencies with USEC time slices\n" + " -F FILE: write histogram to file\n" + " -S: start processing threads in sleep, wake with SIGUSR2\n" + " -i MSEC: use interrupts instead of polling (cont. on interrupt or after MSEC)\n" + " -c: disable colored output\n", + prgname); +} + +/* Parse the argument given in the command line of the application */ +static int +l2reflect_parse_args(int argc, char **argv) +{ + int opt, ret; + char **argvopt; + int option_index; + int opt_colors = 1; + char *prgname = argv[0]; + static struct option lgopts[] = { { NULL, 0, 0, 0 } }; + + argvopt = argv; + policy = SCHED_OTHER; + hist_filename = NULL; + l2reflect_output_hist = 0; + + while ((opt = getopt_long(argc, argvopt, "p:P:b:H:F:i:sqdrflSc", lgopts, + &option_index)) != EOF) { + switch (opt) { + /* port */ + case 'p': + l2reflect_port_number = + (uint16_t)strtoul(optarg, NULL, 10); + break; + case 'P': + priority = strtoul(optarg, NULL, 10); + if (priority > 0) { + if (policy == SCHED_OTHER) + policy = SCHED_RR; + l2reflect_mlock = 1; + } + break; + case 'b': + l2reflect_break_usec = strtoul(optarg, NULL, 10); + break; + case 'S': + sleep_start = 1; + break; + case 'q': + quiet = 1; + break; + case 'd': + disable_int = 1; + break; + case 'r': + policy = SCHED_RR; + break; + case 'f': + policy = SCHED_FIFO; + break; + case 'l': + l2reflect_mlock = 1; + break; + case 'H': + l2reflect_hist = 1; + hist_bucket_usec = strtoul(optarg, NULL, 10); +#ifndef RTE_HAS_JANSSON + printf("not compiled with cjson support\n"); + return -1; +#endif + break; + case 'F': + hist_filename = strndup(optarg, 128); + break; + case 'i': + l2reflect_interrupt = 1; + l2reflect_sleep_msec = strtoul(optarg, NULL, 10); + break; + case 'c': + opt_colors = 0; + break; + default: + l2reflect_usage(prgname); + return -1; + } + } + + if (optind >= 0) + argv[optind - 1] = prgname; + + if (hist_filename && !l2reflect_hist) { + printf("-F switch requires -H switch as well\n"); + return -1; + } + + /* output is redirected, disable coloring */ + if (!isatty(fileno(stdout))) + opt_colors = 0; + + enable_colors(opt_colors); + + ret = optind - 1; + optind = 0; /* reset getopt lib */ + return ret; +} + +__rte_format_printf(1, 0) +static void +trace_write(const char *fmt, ...) +{ + va_list ap; + char buf[256]; + int n, err; + + if (trace_fd == 0) + trace_fd = open("/sys/kernel/debug/tracing/trace_marker", + O_WRONLY); + if (trace_fd < 0) + return; + + va_start(ap, fmt); + n = vsnprintf(buf, 256, fmt, ap); + va_end(ap); + + err = write(trace_fd, buf, n); + assert(err >= 1); +} + +/* Send a burst of one packet */ +static int +l2reflect_send_packet(struct rte_mbuf **m, uint16_t port) +{ + unsigned int ret; + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + uint16_t type; + + eth = rte_pktmbuf_mtod(*m, struct rte_ether_hdr *); + pkt = (struct my_magic_packet *)eth; + type = pkt->type; + + ret = rte_eth_tx_burst(port, l2reflect_q, m, 1); + if (unlikely(ret < 1)) { + rte_pktmbuf_free(*m); + } else { + if (type == TRACE_TYPE_DATA) { + clock_gettime(CLOCK_MONOTONIC, &last_sent); + l2reflect_state = S_RUNNING; + } + } + return 0; +} + +static void +l2reflect_simple_forward(struct rte_mbuf *m) +{ + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + + eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); + pkt = (struct my_magic_packet *)eth; + + /* dst addr */ + rte_ether_addr_copy(ð->s_addr, ð->d_addr); + /* src addr */ + rte_ether_addr_copy(&l2reflect_port_eth_addr, ð->s_addr); + + if ((eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_L2REFLECT)) && + (pkt->magic == MAGIC_TRACE_PAYLOAD)) { + /* and the native one */ + trace_write("sending traced packet\n"); + } + + l2reflect_send_packet(&m, l2reflect_port_number); +} + +static struct rte_mbuf * +l2reflect_new_pkt(unsigned int type) +{ + struct rte_mbuf *m; + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + + m = rte_pktmbuf_alloc(l2reflect_pktmbuf_pool); + if (m == NULL) + rte_exit(EXIT_FAILURE, "rte_pktmbuf_alloc failed\n"); + + eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); + + if (l2reflect_state != S_ELECT_LEADER) + rte_ether_addr_copy(&l2reflect_remote_eth_addr, ð->d_addr); + else + rte_ether_addr_copy(ðer_bcast_addr, ð->d_addr); + + /* src addr */ + rte_ether_addr_copy(&l2reflect_port_eth_addr, ð->s_addr); + eth->ether_type = rte_cpu_to_be_16(ETHER_TYPE_L2REFLECT); + + m->data_len = 64; + m->nb_segs = 1; + m->pkt_len = 64; + m->l2_len = sizeof(struct rte_ether_hdr); + + pkt = (struct my_magic_packet *)eth; + pkt->type = type; + pkt->breakval = l2reflect_break_usec; + + return m; +} + +static void +l2reflect_send_reset(void) +{ + struct rte_mbuf *m; + m = l2reflect_new_pkt(TRACE_TYPE_RSET); + l2reflect_send_packet(&m, l2reflect_port_number); +} + +static void +l2reflect_send_quit(void) +{ + struct rte_mbuf *m; + m = l2reflect_new_pkt(TRACE_TYPE_QUIT); + l2reflect_send_packet(&m, l2reflect_port_number); +} + +static void +l2reflect_new_ball(void) +{ + struct rte_mbuf *pnewball; + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + char mac_src_str[32]; + char mac_dst_str[32]; + + printf("Should create a packet to play with ...\n"); + pnewball = l2reflect_new_pkt(TRACE_TYPE_DATA); + + eth = rte_pktmbuf_mtod(pnewball, struct rte_ether_hdr *); + + rte_ether_format_addr(mac_src_str, sizeof(mac_src_str), ð->s_addr); + rte_ether_format_addr(mac_dst_str, sizeof(mac_dst_str), ð->d_addr); + printf("from MAC address: %s to %s\n\n", mac_src_str, mac_dst_str); + + pkt = (struct my_magic_packet *)eth; + + /* we are tracing lets tell the others */ + if (l2reflect_break_usec) + pkt->magic = MAGIC_TRACE_PAYLOAD; + + l2reflect_send_packet(&pnewball, l2reflect_port_number); +} + +static inline int64_t +calcdiff_ns(struct timespec t1, struct timespec t2) +{ + int64_t diff; + diff = NSEC_PER_SEC * (int64_t)((int)t1.tv_sec - (int)t2.tv_sec); + diff += ((int)t1.tv_nsec - (int)t2.tv_nsec); + return diff; +} + +/* filter the received packets for actual l2reflect messages */ +static inline unsigned int +l2reflect_rx_filter( + struct rte_mbuf **bufs, + unsigned int nb_rx, + unsigned int data_only) +{ + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + unsigned int i, ret; + + ret = 0; + for (i = 0; i < nb_rx; i++) { + eth = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *); + /* + * if the packet is not from our partner + * (and we already have a partner), drop it + */ + if (l2reflect_state != S_ELECT_LEADER && + !rte_is_same_ether_addr(ð->s_addr, + &l2reflect_remote_eth_addr)) + goto drop; + + /* check for the l2reflect ether type */ + if (eth->ether_type != rte_cpu_to_be_16(ETHER_TYPE_L2REFLECT)) + goto drop; + + /* check if the message is a control message */ + pkt = (struct my_magic_packet *)eth; + if (data_only && (pkt->type != TRACE_TYPE_DATA && + pkt->type != TRACE_TYPE_RSET && + pkt->type != TRACE_TYPE_QUIT)) + goto drop; + + bufs[ret++] = bufs[i]; + continue; +drop: + rte_pktmbuf_free(bufs[i]); + } + + return ret; +} + +/* + * automatically elect the leader of the benchmark by + * sending out HELO packets and waiting for responses. + * On response, the mac addresses are compared and the + * numerically larger one becomes the leader. + */ +static int +elect_leader(uint16_t portid) +{ + struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; + struct rte_mbuf *m; + struct rte_ether_hdr *eth; + struct my_magic_packet *pkt; + uint64_t my_mac, other_mac, breakval; + unsigned int i, nb_rx; + int i_win; + int searching = 1; + + other_mac = 0ULL; + my_mac = 0ULL; + breakval = 0; + + while (l2reflect_state == S_ELECT_LEADER) { + nb_rx = rte_eth_rx_burst(portid, l2reflect_q, pkts_burst, + MAX_PKT_BURST); + nb_rx = l2reflect_rx_filter(pkts_burst, nb_rx, 0); + for (i = 0; i < nb_rx && searching; i++) { + eth = rte_pktmbuf_mtod(pkts_burst[i], + struct rte_ether_hdr *); + pkt = (struct my_magic_packet *)eth; + if (pkt->type == TRACE_TYPE_HELO || + pkt->type == TRACE_TYPE_EHLO) { + l2reflect_state = S_RUNNING; + rte_ether_addr_copy(ð->s_addr, + &l2reflect_remote_eth_addr); + rte_ether_addr_copy( + ð->s_addr, + (struct rte_ether_addr *)&other_mac); + rte_ether_addr_copy( + &l2reflect_port_eth_addr, + (struct rte_ether_addr *)&my_mac); + breakval = pkt->breakval; + /* break, but cleanup */ + searching = 0; + } + /* we got a HELO packet, respond with EHLO */ + if (pkt->type == TRACE_TYPE_HELO) { + m = l2reflect_new_pkt(TRACE_TYPE_EHLO); + RTE_LOG(INFO, L2REFLECT, "found one EHLO\n"); + l2reflect_send_packet(&m, + l2reflect_port_number); + } + rte_pktmbuf_free(pkts_burst[i]); + } + m = l2reflect_new_pkt(TRACE_TYPE_HELO); + RTE_LOG(INFO, L2REFLECT, "looking for player HELO\n"); + l2reflect_send_packet(&m, l2reflect_port_number); + usleep(500000); + } + /* leave election logic */ + if (l2reflect_state != S_RUNNING) + return 0; + + if (my_mac == other_mac) + rte_exit(EXIT_FAILURE, "talking to myself ... confused\n"); + + /* the one with the bigger MAC is the leader */ + i_win = (my_mac > other_mac); + + RTE_LOG(INFO, L2REFLECT, "i am the \"%s\"\n", i_win ? "rick" : "morty"); + + /* looser takes tracing break value from winner */ + if (!i_win) + l2reflect_break_usec = breakval; + + return i_win; +} + +/* + * add the measured time diff to the statistics. + * return false if threshold is hit + */ +static inline bool +add_to_record(const uint64_t diff) +{ + record.rounds++; + /* do not count the first rounds, diff would be too high */ + if (record.rounds > WARMUP_ROUNDS) { + if (l2reflect_hist) { + const uint64_t bucket = + MIN(diff / (hist_bucket_usec * 1000), + HIST_CAP_BUCKET); + record.hist[bucket]++; + } + + record.avg_round += (double)diff; + if (diff < record.min_round) + record.min_round = diff; + if (diff > record.max_round) { + record.max_round = diff; + if (l2reflect_break_usec && + (record.max_round > + l2reflect_break_usec * + 1000)) + return false; + } + } + return true; +} + +/* + * process a single packet. + * return false if latency threshold is hit + */ +static inline bool +process_packet( + struct my_magic_packet *pkt, + uint64_t *diff) +{ + if (pkt->type == TRACE_TYPE_DATA) { + clock_gettime(CLOCK_MONOTONIC, &last_recv); + *diff = calcdiff_ns(last_recv, last_sent); + if (!add_to_record(*diff)) + return false; + } + if (pkt->magic == MAGIC_TRACE_PAYLOAD) { + /* and the native one */ + trace_write("received traced packet\n"); + } + return true; +} + +/* main processing loop */ +static void +l2reflect_main_loop(void) +{ + struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; + struct rte_mbuf *m; + unsigned int lcore_id; + unsigned int j, nb_rx, nb_evt; + uint16_t portid; + uint64_t diff = 0; + int sender; + struct my_magic_packet *pkt; + struct rte_ether_hdr *eth; + struct rte_epoll_event event; + + lcore_id = rte_lcore_id(); + + RTE_LOG(INFO, L2REFLECT, "entering main loop on lcore %u\n", lcore_id); + + portid = l2reflect_port_number; + RTE_LOG(INFO, L2REFLECT, " -- lcoreid=%u portid=%u\n", lcore_id, + portid); + +restart: + init_record(); + l2reflect_state = S_ELECT_LEADER; + sender = elect_leader(portid); + + /* we are the sender so we bring one ball into the game */ + if (sender) + l2reflect_new_ball(); + + while (l2reflect_state == S_RUNNING) { + if (l2reflect_interrupt) { + rte_eth_dev_rx_intr_enable(portid, l2reflect_q); + /* wait for interrupt or timeout */ + nb_evt = rte_epoll_wait(RTE_EPOLL_PER_THREAD, &event, 1, + l2reflect_sleep_msec); + rte_eth_dev_rx_intr_disable(portid, l2reflect_q); + if (nb_evt == 0 && rounds > WARMUP_ROUNDS) + ++record.timeouts; + } + + nb_rx = rte_eth_rx_burst(portid, l2reflect_q, pkts_burst, + MAX_PKT_BURST); + + /* TODO use drivers/hw to filter mac */ + nb_rx = l2reflect_rx_filter(pkts_burst, nb_rx, 1); + + /* remote is telling us to reset or stop */ + if (nb_rx) { + eth = rte_pktmbuf_mtod(pkts_burst[0], + struct rte_ether_hdr *); + pkt = (struct my_magic_packet *)eth; + if (eth->ether_type == + rte_cpu_to_be_16(ETHER_TYPE_L2REFLECT)) { + if (pkt->type == TRACE_TYPE_RSET) { + rte_pktmbuf_free(pkts_burst[0]); + goto restart; + } + if (pkt->type == TRACE_TYPE_QUIT) { + l2reflect_state = S_REMOTE_TERM; + break; + } + } + } + + if (l2reflect_state == S_RUNNING && nb_rx) { + eth = rte_pktmbuf_mtod(pkts_burst[0], + struct rte_ether_hdr *); + pkt = (struct my_magic_packet *)eth; + if (eth->ether_type == + rte_cpu_to_be_16(ETHER_TYPE_L2REFLECT)) { + if (!process_packet(pkt, &diff)) + break; + } + } + for (j = 0; j < nb_rx; j++) { + m = pkts_burst[j]; + rte_prefetch0(rte_pktmbuf_mtod(m, void *)); + l2reflect_simple_forward(m); + } + } + const int state_cpy = l2reflect_state; + switch (state_cpy) { + case S_RESET_TRX: + l2reflect_send_reset(); + l2reflect_state = S_ELECT_LEADER; + case S_ELECT_LEADER: + goto restart; + } + + if (state_cpy == S_LOCAL_TERM) { + l2reflect_send_quit(); + if (record.max_round > l2reflect_break_usec) { + printf("hit latency threshold (%" PRIu64 + ".%03u > %" PRIu64 ")\n", + diff / 1000, (unsigned int)(diff % 1000), + l2reflect_break_usec); + } + } else if (state_cpy == S_REMOTE_TERM) { + printf("received message that remote hit threshold (or is cancelled)\n"); + } +} + +static int +l2reflect_launch_one_lcore(__rte_unused void *dummy) +{ + struct sched_param param; + int err; + + if (sleep_start) { + printf("Sleeping and waiting for SIGCONT\n"); + while (sleep_start) + usleep(10000); + + printf("Got SIGCONT, continuing"); + } + if (l2reflect_mlock) { + err = mlockall(MCL_CURRENT | MCL_FUTURE); + if (err) + rte_exit(EXIT_FAILURE, "mlockall failed: %s\n", + strerror(errno)); + } + if (priority > 0 || policy != SCHED_OTHER) { + memset(¶m, 0, sizeof(param)); + param.sched_priority = priority; + err = sched_setscheduler(0, policy, ¶m); + if (err) + rte_exit(EXIT_FAILURE, + "sched_setscheduler failed: %s\n", + strerror(errno)); + } + if (l2reflect_interrupt) { + err = rte_eth_dev_rx_intr_ctl_q(l2reflect_port_number, + l2reflect_q, + RTE_EPOLL_PER_THREAD, + RTE_INTR_EVENT_ADD, NULL); + if (err) + rte_exit(EXIT_FAILURE, + "could not register I/O interrupt\n"); + } + l2reflect_main_loop(); + return 0; +} + +static void +sig_handler(int signum) +{ + switch (signum) { + case SIGUSR1: + if (l2reflect_state == S_RUNNING) + l2reflect_state = S_RESET_TRX; + break; + case SIGUSR2: + l2reflect_output_hist = 1; + break; + case SIGCONT: + sleep_start = 0; + break; + case SIGHUP: + case SIGINT: + l2reflect_state = S_LOCAL_TERM; + break; + } +} + +int +main(int argc, char **argv) +{ + struct rte_eth_dev_info dev_info; + int ret; + uint32_t i; + uint16_t nb_ports; + unsigned int lcore_id; + struct sigaction action; + bzero(&action, sizeof(action)); + char mempool_name[128]; + char mac_str[32]; + + action.sa_handler = sig_handler; + if (sigaction(SIGHUP, &action, NULL) < 0 || + sigaction(SIGUSR1, &action, NULL) < 0 || + sigaction(SIGUSR2, &action, NULL) < 0 || + sigaction(SIGCONT, &action, NULL) < 0 || + sigaction(SIGINT, &action, NULL) < 0) { + rte_exit(EXIT_FAILURE, "Could not register signal handler\n"); + } + + lcore_id = rte_lcore_id(); + /* init EAL */ + ret = rte_eal_init(argc, argv); + if (ret < 0) + rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n"); + argc -= ret; + argv += ret; + + /* parse application arguments (after the EAL ones) */ + ret = l2reflect_parse_args(argc, argv); + if (ret < 0) + rte_exit(EXIT_FAILURE, "Invalid L2REFLECT arguments\n"); + + snprintf(mempool_name, sizeof(mempool_name), "mbuf_pool_%d", getpid()); + printf("About to create mempool \"%s\"\n", mempool_name); + /* create the mbuf pool */ + l2reflect_pktmbuf_pool = + rte_mempool_create(mempool_name, NB_MBUF, MBUF_SIZE, + MAX_PKT_BURST, sizeof(struct rte_pktmbuf_pool_private), + rte_pktmbuf_pool_init, NULL, + rte_pktmbuf_init, NULL, rte_socket_id(), 0); + + if (l2reflect_pktmbuf_pool == NULL) + rte_exit(EXIT_FAILURE, + "Cannot init/find mbuf pool name %s\nError: %d %s\n", + mempool_name, rte_errno, rte_strerror(rte_errno)); + + nb_ports = rte_eth_dev_count_avail(); + if (nb_ports == 0) + rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n"); + if (l2reflect_port_number + 1 > nb_ports) + rte_exit(EXIT_FAILURE, "Chosen port %d does not exist - bye\n", + l2reflect_port_number); + printf("We have %d ports and will use port %d\n", nb_ports, + l2reflect_port_number); + + rte_eth_dev_info_get(l2reflect_port_number, &dev_info); + printf("Initializing port %u... ", l2reflect_port_number); + fflush(stdout); + + if (l2reflect_interrupt) + port_conf.intr_conf.rxq = 1; + ret = rte_eth_dev_configure(l2reflect_port_number, 1, 1, &port_conf); + if (ret < 0) + rte_exit(EXIT_FAILURE, + "Cannot configure device: err=%s, port=%u\n", + strerror(-ret), l2reflect_port_number); + + rte_eth_macaddr_get(l2reflect_port_number, &l2reflect_port_eth_addr); + + /* init RX queues */ + fflush(stdout); + for (i = 0; i <= l2reflect_q; i++) { + ret = rte_eth_rx_queue_setup( + l2reflect_port_number, i, nb_rxd, + rte_eth_dev_socket_id(l2reflect_port_number), NULL, + l2reflect_pktmbuf_pool); + if (ret < 0) + rte_exit( + EXIT_FAILURE, + "rte_eth_rx_queue_setup:err=%s, port=%u q=%u\n", + strerror(-ret), l2reflect_port_number, i); + } + + /* init one TX queue on each port */ + fflush(stdout); + ret = rte_eth_tx_queue_setup( + l2reflect_port_number, 0, nb_txd, + rte_eth_dev_socket_id(l2reflect_port_number), NULL); + if (ret < 0) + rte_exit(EXIT_FAILURE, + "rte_eth_tx_queue_setup:err=%s, port=%u\n", + strerror(-ret), (unsigned int)l2reflect_port_number); + + /* Start device */ + ret = rte_eth_dev_start(l2reflect_port_number); + if (ret < 0) + rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%s, port=%u\n", + strerror(-ret), (unsigned int)l2reflect_port_number); + + rte_eth_promiscuous_enable(l2reflect_port_number); + + rte_ether_format_addr(mac_str, sizeof(mac_str), + &l2reflect_port_eth_addr); + printf("Port %u, MAC address: %s\n\n", + (unsigned int)l2reflect_port_number, mac_str); + + /* + * in quiet mode the master executes the main packet loop + * otherwise the one slave does it and the master prints stats + */ + if (quiet) { + assert(rte_lcore_count() == 1); +#ifdef HAS_SYS_IO + if (disable_int) { + iopl(3); + asm("cli"); + } +#endif + l2reflect_launch_one_lcore(NULL); + } else { + assert(rte_lcore_count() == 2); + /* the slave reflects the packets */ + RTE_LCORE_FOREACH_SLAVE(lcore_id) + { + rte_eal_remote_launch(l2reflect_launch_one_lcore, NULL, + lcore_id); + } + + /* the master prints the stats */ + init_record(); + l2reflect_stats_loop(); + rte_eal_mp_wait_lcore(); + } + rte_eal_cleanup(); + + if (l2reflect_hist) + output_histogram_snapshot(); + + cleanup_record(); + + if (trace_fd) + close(trace_fd); + + return 0; +} diff --git a/app/l2reflect/meson.build b/app/l2reflect/meson.build new file mode 100644 index 000000000..d5cfa906f --- /dev/null +++ b/app/l2reflect/meson.build @@ -0,0 +1,25 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2020 Siemens AG + +# meson file, for building this example as part of a main DPDK build. +cc = meson.get_compiler('c') + +jansson = dependency('libjansson', required: false) + +if not jansson.found() + jansson = cc.find_library('jansson', required: false) +endif + +if not is_linux + build = false +endif +if cc.has_header('sys/io.h') + cflags += '-DHAS_SYS_IO' +endif + +sources = files('main.c', 'stats.c', 'colors.c') +deps += ['ethdev'] +if jansson.found() + ext_deps += jansson + cflags += '-DRTE_HAS_JANSSON' +endif diff --git a/app/l2reflect/stats.c b/app/l2reflect/stats.c new file mode 100644 index 000000000..3f2ef9b1e --- /dev/null +++ b/app/l2reflect/stats.c @@ -0,0 +1,178 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + */ +#include +#include +#include +#include +#ifdef RTE_HAS_JANSSON +#include +#endif +#include "colors.h" +#include "stats.h" + +unsigned int hist_bucket_usec; +struct stats record; +char *hist_filename; + +void +init_record(void) +{ + record.max_round = 0; + record.min_round = MIN_INITIAL; + record.rounds = 0; + record.timeouts = 0; + record.avg_round = 0; + if (l2reflect_hist) { + if (!record.hist_size) { + record.hist = + calloc(HIST_NUM_BUCKETS, sizeof(uint64_t)); + record.hist_size = HIST_NUM_BUCKETS; + } else { + memset(record.hist, 0, + record.hist_size * sizeof(uint64_t)); + } + } +} + +void +cleanup_record(void) +{ + if (l2reflect_hist) { + free(record.hist); + record.hist = NULL; + record.hist_size = 0; + } +} + +void +output_histogram_snapshot(void) +{ + char *json = serialize_histogram(&record); + FILE *fd = stderr; + if (hist_filename) + fd = fopen(hist_filename, "w"); + fputs(json, fd); + fputs("\n", fd); + free(json); + if (hist_filename) + fclose(fd); +} + +void +print_stats(void) +{ + const char clr[] = { 27, '[', '2', 'J', '\0' }; + const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' }; + const uint64_t bytes_in_gib = 0x40000000; + struct rte_eth_stats stats; + char mac_str_me[32]; + char mac_str_remote[32]; + + rte_eth_stats_get(l2reflect_port_number, &stats); + rte_ether_format_addr(mac_str_me, sizeof(mac_str_me), + &l2reflect_port_eth_addr); + rte_ether_format_addr(mac_str_remote, sizeof(mac_str_remote), + &l2reflect_remote_eth_addr); + + /* Clear screen and move to top left */ + printf("%s%s", clr, topLeft); + + printf("%sNetworking Roundtrip Test%s\n", colors->green, colors->reset); + printf("\n%sPort statistics ====================================%s", + colors->magenta, colors->reset); + + printf("\nMe: %s <--> Remote: %s", mac_str_me, mac_str_remote); + printf("\nStatistics for port %d PID %d on lcore %d ---------" + "\nPackets tx: %22" PRIu64 "\nPackets rx: %22" PRIu64 + "\nBytes tx: %24" PRIu64 " (%8.2f GiB)" + "\nBytes rx: %24" PRIu64 " (%8.2f GiB)" + "\nErrors tx: %23" PRIu64 "\nErrors rx: %23" PRIu64 + "\nTimeouts rx: %21" PRIu64 " (>%9" PRIu64 "ms)", + l2reflect_port_number, getpid(), rte_lcore_id(), + stats.opackets, stats.ipackets, stats.obytes, + (double)stats.obytes / bytes_in_gib, stats.ibytes, + (double)stats.ibytes / bytes_in_gib, stats.oerrors, + stats.ierrors, record.timeouts, l2reflect_sleep_msec); + printf("\n%sPort timing statistics =============================%s", + colors->magenta, colors->reset); + if (l2reflect_state == S_ELECT_LEADER || + record.min_round == MIN_INITIAL) { + printf("\n\nBenchmark not started yet...\n"); + } else { + printf("\n%sMax%s roundtrip: %19" PRIu64 " us", colors->red, + colors->reset, record.max_round / 1000); + printf("\n%sAvg%s roundtrip: %19" PRIu64 " us", colors->yellow, + colors->reset, + record.rounds ? (uint64_t)(record.avg_round / + record.rounds / 1000) : + 0); + printf("\n%sMin%s roundtrip: %19" PRIu64 " us", colors->green, + colors->reset, record.min_round / 1000); + } + printf("\n%s====================================================%s\n", + colors->magenta, colors->reset); +} + +void +l2reflect_stats_loop(void) +{ + while (!(l2reflect_state & (S_LOCAL_TERM | S_REMOTE_TERM))) { + print_stats(); + if (l2reflect_hist && l2reflect_output_hist) { + output_histogram_snapshot(); + l2reflect_output_hist = 0; + } + sleep(1); + } +} + +char * +serialize_histogram(__rte_unused const struct stats *record) +{ +#ifndef RTE_HAS_JANSSON + return strdup("to print histogram, build with jansson support"); +#else + char *str = NULL; + char key[8]; + unsigned int i; + json_t *hist0, *cpu0, *all_cpus, *output; + + output = json_object(); + /* version: 1 */ + json_object_set_new(output, "version", json_integer(1)); + + /* cpu 0 histogram */ + hist0 = json_object(); + for (i = 0; i < record->hist_size; ++i) { + /* only log positive numbers to meet jitterplot format */ + if (record->hist[i] != 0) { + snprintf(key, 8, "%u", i * hist_bucket_usec); + json_object_set(hist0, key, + json_integer(record->hist[i])); + } + } + + /* cpu 0 stats */ + cpu0 = json_object(); + json_object_set_new(cpu0, "histogram", hist0); + json_object_set_new(cpu0, "count", json_integer(record->rounds)); + json_object_set_new(cpu0, "min", json_integer(record->min_round)); + json_object_set_new(cpu0, "max", json_integer(record->max_round)); + json_object_set_new( + cpu0, "avg", + json_integer((record->avg_round / record->rounds / 1000))); + + /* combine objects */ + all_cpus = json_object(); + json_object_set_new(all_cpus, "0", cpu0); + json_object_set_new(output, "cpu", all_cpus); + + str = json_dumps(output, JSON_ENSURE_ASCII | JSON_INDENT(2)); + + /* cleanup */ + json_decref(output); + + return str; +#endif +} diff --git a/app/l2reflect/stats.h b/app/l2reflect/stats.h new file mode 100644 index 000000000..d13f81a54 --- /dev/null +++ b/app/l2reflect/stats.h @@ -0,0 +1,61 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Siemens AG + */ +#ifndef _STATS_H_ +#define _STATS_H_ +#include +#include +#include + +#include + +#include "l2reflect.h" + +#define MIN_INITIAL ULONG_MAX +#define HIST_NUM_BUCKETS 128 +#define HIST_CAP_BUCKET (HIST_NUM_BUCKETS - 1) + +/* runtime statistics */ +struct stats { + uint64_t max_round; + uint64_t min_round; + uint64_t rounds; + uint64_t timeouts; + double avg_round; + unsigned int hist_size; + /* each slot is 10us */ + uint64_t *hist; +}; + +/* size of each histogram bucket in usec */ +extern unsigned int hist_bucket_usec; +extern struct stats record; +extern char *hist_filename; + +void +init_record(void); +void +cleanup_record(void); + +void +l2reflect_stats_loop(void); + +/* + * Write the histogram to file / stdio without any locking. + * When called during the measurement, values are approximations + * (racy reads). + */ +void +output_histogram_snapshot(void); + +/* Print out statistics on packets dropped */ +void +print_stats(void); + +/* + * get a JSON representation of the record + */ +char * +serialize_histogram(const struct stats *record); + +#endif /* _STATS_H_ */ diff --git a/app/meson.build b/app/meson.build index ab43a8403..52d119f7d 100644 --- a/app/meson.build +++ b/app/meson.build @@ -6,6 +6,7 @@ if is_windows endif apps = [ + 'l2reflect', 'pdump', 'proc-info', 'test-acl',