From 2741a0493c04067d7acb0e44035aa27618b7d204 Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:41 +0800 Subject: [PATCH 1/7] samples/bpf: xdpsock: Add VLAN support for Tx-only operation In multi-queue environment testing, the support for VLAN-tag based steering is useful. So, this patch adds the capability to add VLAN tag (VLAN ID and Priority) to the generated Tx frame. To set the VLAN ID=10 and Priority=2 for Tx only through TxQ=3: $ xdpsock -i eth0 -t -N -z -q 3 -V -J 10 -K 2 If VLAN ID (-J) and Priority (-K) is set, it default to VLAN ID = 1 VLAN Priority = 0. For example, VLAN-tagged Tx only, xdp copy mode through TxQ=1: $ xdpsock -i eth0 -t -N -c -q 1 -V Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Acked-by: Song Liu Link: https://lore.kernel.org/bpf/20211230035447.523177-2-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 88 ++++++++++++++++++++++++++++++++------ 1 file changed, 74 insertions(+), 14 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index 616d663d55aa..d5e298ccf491 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -56,6 +56,12 @@ #define DEBUG_HEXDUMP 0 +#define VLAN_PRIO_MASK 0xe000 /* Priority Code Point */ +#define VLAN_PRIO_SHIFT 13 +#define VLAN_VID_MASK 0x0fff /* VLAN Identifier */ +#define VLAN_VID__DEFAULT 1 +#define VLAN_PRI__DEFAULT 0 + typedef __u64 u64; typedef __u32 u32; typedef __u16 u16; @@ -81,6 +87,9 @@ static u32 opt_batch_size = 64; static int opt_pkt_count; static u16 opt_pkt_size = MIN_PKT_SIZE; static u32 opt_pkt_fill_pattern = 0x12345678; +static bool opt_vlan_tag; +static u16 opt_pkt_vlan_id = VLAN_VID__DEFAULT; +static u16 opt_pkt_vlan_pri = VLAN_PRI__DEFAULT; static bool opt_extra_stats; static bool opt_quiet; static bool opt_app_stats; @@ -101,6 +110,14 @@ static u32 prog_id; static bool opt_busy_poll; static bool opt_reduced_cap; +struct vlan_ethhdr { + unsigned char h_dest[6]; + unsigned char h_source[6]; + __be16 h_vlan_proto; + __be16 h_vlan_TCI; + __be16 h_vlan_encapsulated_proto; +}; + struct xsk_ring_stats { unsigned long rx_npkts; unsigned long tx_npkts; @@ -740,11 +757,13 @@ static inline u16 udp_csum(u32 saddr, u32 daddr, u32 len, #define ETH_FCS_SIZE 4 -#define PKT_HDR_SIZE (sizeof(struct ethhdr) + sizeof(struct iphdr) + \ +#define ETH_HDR_SIZE (opt_vlan_tag ? sizeof(struct vlan_ethhdr) : \ + sizeof(struct ethhdr)) +#define PKT_HDR_SIZE (ETH_HDR_SIZE + sizeof(struct iphdr) + \ sizeof(struct udphdr)) #define PKT_SIZE (opt_pkt_size - ETH_FCS_SIZE) -#define IP_PKT_SIZE (PKT_SIZE - sizeof(struct ethhdr)) +#define IP_PKT_SIZE (PKT_SIZE - ETH_HDR_SIZE) #define UDP_PKT_SIZE (IP_PKT_SIZE - sizeof(struct iphdr)) #define UDP_PKT_DATA_SIZE (UDP_PKT_SIZE - sizeof(struct udphdr)) @@ -752,17 +771,42 @@ static u8 pkt_data[XSK_UMEM__DEFAULT_FRAME_SIZE]; static void gen_eth_hdr_data(void) { - struct udphdr *udp_hdr = (struct udphdr *)(pkt_data + - sizeof(struct ethhdr) + - sizeof(struct iphdr)); - struct iphdr *ip_hdr = (struct iphdr *)(pkt_data + - sizeof(struct ethhdr)); - struct ethhdr *eth_hdr = (struct ethhdr *)pkt_data; + struct udphdr *udp_hdr; + struct iphdr *ip_hdr; + + if (opt_vlan_tag) { + struct vlan_ethhdr *veth_hdr = (struct vlan_ethhdr *)pkt_data; + u16 vlan_tci = 0; + + udp_hdr = (struct udphdr *)(pkt_data + + sizeof(struct vlan_ethhdr) + + sizeof(struct iphdr)); + ip_hdr = (struct iphdr *)(pkt_data + + sizeof(struct vlan_ethhdr)); + + /* ethernet & VLAN header */ + memcpy(veth_hdr->h_dest, "\x3c\xfd\xfe\x9e\x7f\x71", ETH_ALEN); + memcpy(veth_hdr->h_source, "\xec\xb1\xd7\x98\x3a\xc0", ETH_ALEN); + veth_hdr->h_vlan_proto = htons(ETH_P_8021Q); + vlan_tci = opt_pkt_vlan_id & VLAN_VID_MASK; + vlan_tci |= (opt_pkt_vlan_pri << VLAN_PRIO_SHIFT) & VLAN_PRIO_MASK; + veth_hdr->h_vlan_TCI = htons(vlan_tci); + veth_hdr->h_vlan_encapsulated_proto = htons(ETH_P_IP); + } else { + struct ethhdr *eth_hdr = (struct ethhdr *)pkt_data; + + udp_hdr = (struct udphdr *)(pkt_data + + sizeof(struct ethhdr) + + sizeof(struct iphdr)); + ip_hdr = (struct iphdr *)(pkt_data + + sizeof(struct ethhdr)); + + /* ethernet header */ + memcpy(eth_hdr->h_dest, "\x3c\xfd\xfe\x9e\x7f\x71", ETH_ALEN); + memcpy(eth_hdr->h_source, "\xec\xb1\xd7\x98\x3a\xc0", ETH_ALEN); + eth_hdr->h_proto = htons(ETH_P_IP); + } - /* ethernet header */ - memcpy(eth_hdr->h_dest, "\x3c\xfd\xfe\x9e\x7f\x71", ETH_ALEN); - memcpy(eth_hdr->h_source, "\xec\xb1\xd7\x98\x3a\xc0", ETH_ALEN); - eth_hdr->h_proto = htons(ETH_P_IP); /* IP header */ ip_hdr->version = IPVERSION; @@ -920,6 +964,9 @@ static struct option long_options[] = { {"tx-pkt-count", required_argument, 0, 'C'}, {"tx-pkt-size", required_argument, 0, 's'}, {"tx-pkt-pattern", required_argument, 0, 'P'}, + {"tx-vlan", no_argument, 0, 'V'}, + {"tx-vlan-id", required_argument, 0, 'J'}, + {"tx-vlan-pri", required_argument, 0, 'K'}, {"extra-stats", no_argument, 0, 'x'}, {"quiet", no_argument, 0, 'Q'}, {"app-stats", no_argument, 0, 'a'}, @@ -960,6 +1007,9 @@ static void usage(const char *prog) " (Default: %d bytes)\n" " Min size: %d, Max size %d.\n" " -P, --tx-pkt-pattern=nPacket fill pattern. Default: 0x%x\n" + " -V, --tx-vlan Send VLAN tagged packets (For -t|--txonly)\n" + " -J, --tx-vlan-id=n Tx VLAN ID [1-4095]. Default: %d (For -V|--tx-vlan)\n" + " -K, --tx-vlan-pri=n Tx VLAN Priority [0-7]. Default: %d (For -V|--tx-vlan)\n" " -x, --extra-stats Display extra statistics.\n" " -Q, --quiet Do not display any stats.\n" " -a, --app-stats Display application (syscall) statistics.\n" @@ -969,7 +1019,8 @@ static void usage(const char *prog) "\n"; fprintf(stderr, str, prog, XSK_UMEM__DEFAULT_FRAME_SIZE, opt_batch_size, MIN_PKT_SIZE, MIN_PKT_SIZE, - XSK_UMEM__DEFAULT_FRAME_SIZE, opt_pkt_fill_pattern); + XSK_UMEM__DEFAULT_FRAME_SIZE, opt_pkt_fill_pattern, + VLAN_VID__DEFAULT, VLAN_PRI__DEFAULT); exit(EXIT_FAILURE); } @@ -981,7 +1032,7 @@ static void parse_command_line(int argc, char **argv) opterr = 0; for (;;) { - c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:xQaI:BR", + c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:VJ:K:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1062,6 +1113,15 @@ static void parse_command_line(int argc, char **argv) case 'P': opt_pkt_fill_pattern = strtol(optarg, NULL, 16); break; + case 'V': + opt_vlan_tag = true; + break; + case 'J': + opt_pkt_vlan_id = atoi(optarg); + break; + case 'K': + opt_pkt_vlan_pri = atoi(optarg); + break; case 'x': opt_extra_stats = 1; break; From 6440a6c23f6c72c57dbdf7928d92d3fc1aef6edc Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:42 +0800 Subject: [PATCH 2/7] samples/bpf: xdpsock: Add Dest and Src MAC setting for Tx-only operation To set Dest MAC address (-G|--tx-dmac) only: $ xdpsock -i eth0 -t -N -z -G aa:bb:cc:dd:ee:ff To set Source MAC address (-H|--tx-smac) only: $ xdpsock -i eth0 -t -N -z -H 11:22:33:44:55:66 To set both Dest and Source MAC address: $ xdpsock -i eth0 -t -N -z -G aa:bb:cc:dd:ee:ff \ -H 11:22:33:44:55:66 The default Dest and Source MAC address remain the same as before. Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Acked-by: Song Liu Acked-by: Jesper Dangaard Brouer Link: https://lore.kernel.org/bpf/20211230035447.523177-3-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 35 ++++++++++++++++++++++++++++++----- 1 file changed, 30 insertions(+), 5 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index d5e298ccf491..c9a8748a460f 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -14,6 +14,7 @@ #include #include #include +#include #include #include #include @@ -90,6 +91,10 @@ static u32 opt_pkt_fill_pattern = 0x12345678; static bool opt_vlan_tag; static u16 opt_pkt_vlan_id = VLAN_VID__DEFAULT; static u16 opt_pkt_vlan_pri = VLAN_PRI__DEFAULT; +static struct ether_addr opt_txdmac = {{ 0x3c, 0xfd, 0xfe, + 0x9e, 0x7f, 0x71 }}; +static struct ether_addr opt_txsmac = {{ 0xec, 0xb1, 0xd7, + 0x98, 0x3a, 0xc0 }}; static bool opt_extra_stats; static bool opt_quiet; static bool opt_app_stats; @@ -785,8 +790,8 @@ static void gen_eth_hdr_data(void) sizeof(struct vlan_ethhdr)); /* ethernet & VLAN header */ - memcpy(veth_hdr->h_dest, "\x3c\xfd\xfe\x9e\x7f\x71", ETH_ALEN); - memcpy(veth_hdr->h_source, "\xec\xb1\xd7\x98\x3a\xc0", ETH_ALEN); + memcpy(veth_hdr->h_dest, &opt_txdmac, ETH_ALEN); + memcpy(veth_hdr->h_source, &opt_txsmac, ETH_ALEN); veth_hdr->h_vlan_proto = htons(ETH_P_8021Q); vlan_tci = opt_pkt_vlan_id & VLAN_VID_MASK; vlan_tci |= (opt_pkt_vlan_pri << VLAN_PRIO_SHIFT) & VLAN_PRIO_MASK; @@ -802,8 +807,8 @@ static void gen_eth_hdr_data(void) sizeof(struct ethhdr)); /* ethernet header */ - memcpy(eth_hdr->h_dest, "\x3c\xfd\xfe\x9e\x7f\x71", ETH_ALEN); - memcpy(eth_hdr->h_source, "\xec\xb1\xd7\x98\x3a\xc0", ETH_ALEN); + memcpy(eth_hdr->h_dest, &opt_txdmac, ETH_ALEN); + memcpy(eth_hdr->h_source, &opt_txsmac, ETH_ALEN); eth_hdr->h_proto = htons(ETH_P_IP); } @@ -967,6 +972,8 @@ static struct option long_options[] = { {"tx-vlan", no_argument, 0, 'V'}, {"tx-vlan-id", required_argument, 0, 'J'}, {"tx-vlan-pri", required_argument, 0, 'K'}, + {"tx-dmac", required_argument, 0, 'G'}, + {"tx-smac", required_argument, 0, 'H'}, {"extra-stats", no_argument, 0, 'x'}, {"quiet", no_argument, 0, 'Q'}, {"app-stats", no_argument, 0, 'a'}, @@ -1010,6 +1017,8 @@ static void usage(const char *prog) " -V, --tx-vlan Send VLAN tagged packets (For -t|--txonly)\n" " -J, --tx-vlan-id=n Tx VLAN ID [1-4095]. Default: %d (For -V|--tx-vlan)\n" " -K, --tx-vlan-pri=n Tx VLAN Priority [0-7]. Default: %d (For -V|--tx-vlan)\n" + " -G, --tx-dmac= Dest MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" + " -H, --tx-smac= Src MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -x, --extra-stats Display extra statistics.\n" " -Q, --quiet Do not display any stats.\n" " -a, --app-stats Display application (syscall) statistics.\n" @@ -1032,7 +1041,7 @@ static void parse_command_line(int argc, char **argv) opterr = 0; for (;;) { - c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:VJ:K:xQaI:BR", + c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:VJ:K:G:H:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1122,6 +1131,22 @@ static void parse_command_line(int argc, char **argv) case 'K': opt_pkt_vlan_pri = atoi(optarg); break; + case 'G': + if (!ether_aton_r(optarg, + (struct ether_addr *)&opt_txdmac)) { + fprintf(stderr, "Invalid dmac address:%s\n", + optarg); + usage(basename(argv[0])); + } + break; + case 'H': + if (!ether_aton_r(optarg, + (struct ether_addr *)&opt_txsmac)) { + fprintf(stderr, "Invalid smac address:%s\n", + optarg); + usage(basename(argv[0])); + } + break; case 'x': opt_extra_stats = 1; break; From 5a3882542acda1ac5f0a22dddf7f7f8533d3a8cc Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:43 +0800 Subject: [PATCH 3/7] samples/bpf: xdpsock: Add clockid selection support User specifies the clock selection by using -w CLOCK or --clock=CLOCK where CLOCK=[REALTIME, TAI, BOOTTIME, MONOTONIC]. The default CLOCK selection is MONOTONIC. The implementation of clock selection parsing is borrowed from iproute2/tc/q_taprio.c Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211230035447.523177-4-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 40 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 38 insertions(+), 2 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index c9a8748a460f..e6e9a20375cb 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -114,6 +114,7 @@ static u32 opt_num_xsks = 1; static u32 prog_id; static bool opt_busy_poll; static bool opt_reduced_cap; +static clockid_t opt_clock = CLOCK_MONOTONIC; struct vlan_ethhdr { unsigned char h_dest[6]; @@ -178,15 +179,40 @@ struct xsk_socket_info { u32 outstanding_tx; }; +static const struct clockid_map { + const char *name; + clockid_t clockid; +} clockids_map[] = { + { "REALTIME", CLOCK_REALTIME }, + { "TAI", CLOCK_TAI }, + { "BOOTTIME", CLOCK_BOOTTIME }, + { "MONOTONIC", CLOCK_MONOTONIC }, + { NULL } +}; + static int num_socks; struct xsk_socket_info *xsks[MAX_SOCKS]; int sock; +static int get_clockid(clockid_t *id, const char *name) +{ + const struct clockid_map *clk; + + for (clk = clockids_map; clk->name; clk++) { + if (strcasecmp(clk->name, name) == 0) { + *id = clk->clockid; + return 0; + } + } + + return -1; +} + static unsigned long get_nsecs(void) { struct timespec ts; - clock_gettime(CLOCK_MONOTONIC, &ts); + clock_gettime(opt_clock, &ts); return ts.tv_sec * 1000000000UL + ts.tv_nsec; } @@ -965,6 +991,7 @@ static struct option long_options[] = { {"shared-umem", no_argument, 0, 'M'}, {"force", no_argument, 0, 'F'}, {"duration", required_argument, 0, 'd'}, + {"clock", required_argument, 0, 'w'}, {"batch-size", required_argument, 0, 'b'}, {"tx-pkt-count", required_argument, 0, 'C'}, {"tx-pkt-size", required_argument, 0, 's'}, @@ -1006,6 +1033,7 @@ static void usage(const char *prog) " -F, --force Force loading the XDP prog\n" " -d, --duration=n Duration in secs to run command.\n" " Default: forever.\n" + " -w, --clock=CLOCK Clock NAME (default MONOTONIC).\n" " -b, --batch-size=n Batch size for sending or receiving\n" " packets. Default: %d\n" " -C, --tx-pkt-count=n Number of packets to send.\n" @@ -1041,7 +1069,7 @@ static void parse_command_line(int argc, char **argv) opterr = 0; for (;;) { - c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:VJ:K:G:H:xQaI:BR", + c = getopt_long(argc, argv, "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1075,6 +1103,14 @@ static void parse_command_line(int argc, char **argv) case 'n': opt_interval = atoi(optarg); break; + case 'w': + if (get_clockid(&opt_clock, optarg)) { + fprintf(stderr, + "ERROR: Invalid clock %s. Default to CLOCK_MONOTONIC.\n", + optarg); + opt_clock = CLOCK_MONOTONIC; + } + break; case 'z': opt_xdp_bind_flags |= XDP_ZEROCOPY; break; From fa0d27a1d5a8c1f07b0229348b0d178233694fbc Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:44 +0800 Subject: [PATCH 4/7] samples/bpf: xdpsock: Add cyclic TX operation capability Tx cycle time is in micro-seconds unit. By combining the batch size (-b M) and Tx cycle time (-T|--tx-cycle N), xdpsock now can transmit batch-size of packets every N-us periodically. Cyclic TX operation is not applicable if --poll mode is used. To transmit 16 packets every 1ms cycle time for total of 100000 packets silently: $ xdpsock -i eth0 -T -N -z -T 1000 -b 16 -C 100000 To print cyclic TX schedule variance stats, use --app-stats|-a: $ xdpsock -i eth0 -T -N -z -T 1000 -b 16 -C 100000 -a sock0@eth0:0 txonly xdp-drv pps pkts 0.00 rx 0 0 tx 0 100000 calls/s count rx empty polls 0 0 fill fail polls 0 0 copy tx sendtos 0 0 tx wakeup sendtos 0 6254 opt polls 0 0 period min ave max cycle Cyclic TX 1000000 53507 75334 712642 6250 Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211230035447.523177-5-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 87 +++++++++++++++++++++++++++++++++++--- 1 file changed, 81 insertions(+), 6 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index e6e9a20375cb..a2a42ec4b0e9 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -63,12 +63,19 @@ #define VLAN_VID__DEFAULT 1 #define VLAN_PRI__DEFAULT 0 +#define NSEC_PER_SEC 1000000000UL +#define NSEC_PER_USEC 1000 + typedef __u64 u64; typedef __u32 u32; typedef __u16 u16; typedef __u8 u8; static unsigned long prev_time; +static long tx_cycle_diff_min; +static long tx_cycle_diff_max; +static double tx_cycle_diff_ave; +static long tx_cycle_cnt; enum benchmark_type { BENCH_RXDROP = 0, @@ -115,6 +122,7 @@ static u32 prog_id; static bool opt_busy_poll; static bool opt_reduced_cap; static clockid_t opt_clock = CLOCK_MONOTONIC; +static unsigned long opt_tx_cycle_ns; struct vlan_ethhdr { unsigned char h_dest[6]; @@ -305,6 +313,15 @@ static void dump_app_stats(long dt) xsks[i]->app_stats.prev_tx_wakeup_sendtos = xsks[i]->app_stats.tx_wakeup_sendtos; xsks[i]->app_stats.prev_opt_polls = xsks[i]->app_stats.opt_polls; } + + if (opt_tx_cycle_ns) { + printf("\n%-18s %-10s %-10s %-10s %-10s %-10s\n", + "", "period", "min", "ave", "max", "cycle"); + printf("%-18s %-10lu %-10lu %-10lu %-10lu %-10lu\n", + "Cyclic TX", opt_tx_cycle_ns, tx_cycle_diff_min, + (long)(tx_cycle_diff_ave / tx_cycle_cnt), + tx_cycle_diff_max, tx_cycle_cnt); + } } static bool get_interrupt_number(void) @@ -1001,6 +1018,7 @@ static struct option long_options[] = { {"tx-vlan-pri", required_argument, 0, 'K'}, {"tx-dmac", required_argument, 0, 'G'}, {"tx-smac", required_argument, 0, 'H'}, + {"tx-cycle", required_argument, 0, 'T'}, {"extra-stats", no_argument, 0, 'x'}, {"quiet", no_argument, 0, 'Q'}, {"app-stats", no_argument, 0, 'a'}, @@ -1047,6 +1065,7 @@ static void usage(const char *prog) " -K, --tx-vlan-pri=n Tx VLAN Priority [0-7]. Default: %d (For -V|--tx-vlan)\n" " -G, --tx-dmac= Dest MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -H, --tx-smac= Src MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" + " -T, --tx-cycle=n Tx cycle time in micro-seconds (For -t|--txonly).\n" " -x, --extra-stats Display extra statistics.\n" " -Q, --quiet Do not display any stats.\n" " -a, --app-stats Display application (syscall) statistics.\n" @@ -1069,7 +1088,7 @@ static void parse_command_line(int argc, char **argv) opterr = 0; for (;;) { - c = getopt_long(argc, argv, "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:xQaI:BR", + c = getopt_long(argc, argv, "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:T:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1183,6 +1202,10 @@ static void parse_command_line(int argc, char **argv) usage(basename(argv[0])); } break; + case 'T': + opt_tx_cycle_ns = atoi(optarg); + opt_tx_cycle_ns *= NSEC_PER_USEC; + break; case 'x': opt_extra_stats = 1; break; @@ -1388,7 +1411,7 @@ static void rx_drop_all(void) } } -static void tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) +static int tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) { u32 idx; unsigned int i; @@ -1397,7 +1420,7 @@ static void tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) batch_size) { complete_tx_only(xsk, batch_size); if (benchmark_done) - return; + return 0; } for (i = 0; i < batch_size; i++) { @@ -1413,6 +1436,8 @@ static void tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) *frame_nb += batch_size; *frame_nb %= NUM_FRAMES; complete_tx_only(xsk, batch_size); + + return batch_size; } static inline int get_batch_size(int pkt_cnt) @@ -1446,16 +1471,39 @@ static void tx_only_all(void) { struct pollfd fds[MAX_SOCKS] = {}; u32 frame_nb[MAX_SOCKS] = {}; + unsigned long next_tx_ns = 0; int pkt_cnt = 0; int i, ret; + if (opt_poll && opt_tx_cycle_ns) { + fprintf(stderr, + "Error: --poll and --tx-cycles are both set\n"); + return; + } + for (i = 0; i < num_socks; i++) { fds[0].fd = xsk_socket__fd(xsks[i]->xsk); fds[0].events = POLLOUT; } + if (opt_tx_cycle_ns) { + /* Align Tx time to micro-second boundary */ + next_tx_ns = (get_nsecs() / NSEC_PER_USEC + 1) * + NSEC_PER_USEC; + next_tx_ns += opt_tx_cycle_ns; + + /* Initialize periodic Tx scheduling variance */ + tx_cycle_diff_min = 1000000000; + tx_cycle_diff_max = 0; + tx_cycle_diff_ave = 0.0; + } + while ((opt_pkt_count && pkt_cnt < opt_pkt_count) || !opt_pkt_count) { int batch_size = get_batch_size(pkt_cnt); + struct timespec next; + int tx_cnt = 0; + long diff; + int err; if (opt_poll) { for (i = 0; i < num_socks; i++) @@ -1468,13 +1516,40 @@ static void tx_only_all(void) continue; } - for (i = 0; i < num_socks; i++) - tx_only(xsks[i], &frame_nb[i], batch_size); + if (opt_tx_cycle_ns) { + next.tv_sec = next_tx_ns / NSEC_PER_SEC; + next.tv_nsec = next_tx_ns % NSEC_PER_SEC; + err = clock_nanosleep(opt_clock, TIMER_ABSTIME, &next, NULL); + if (err) { + if (err != EINTR) + fprintf(stderr, + "clock_nanosleep failed. Err:%d errno:%d\n", + err, errno); + break; + } - pkt_cnt += batch_size; + /* Measure periodic Tx scheduling variance */ + diff = get_nsecs() - next_tx_ns; + if (diff < tx_cycle_diff_min) + tx_cycle_diff_min = diff; + + if (diff > tx_cycle_diff_max) + tx_cycle_diff_max = diff; + + tx_cycle_diff_ave += (double)diff; + tx_cycle_cnt++; + } + + for (i = 0; i < num_socks; i++) + tx_cnt += tx_only(xsks[i], &frame_nb[i], batch_size); + + pkt_cnt += tx_cnt; if (benchmark_done) break; + + if (opt_tx_cycle_ns) + next_tx_ns += opt_tx_cycle_ns; } if (opt_pkt_count) From fa24d0b1d57825d1a5b802339728d4d8ac20b6d6 Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:45 +0800 Subject: [PATCH 5/7] samples/bpf: xdpsock: Add sched policy and priority support By default, TX schedule policy is SCHED_OTHER (round-robin time-sharing). To improve TX cyclic scheduling, we add SCHED_FIFO policy and its priority by using -W FIFO or --policy=FIFO and -U or --schpri=. A) From xdpsock --app-stats, for SCHED_OTHER policy: $ xdpsock -i eth0 -t -N -z -T 1000 -b 16 -C 100000 -a period min ave max cycle Cyclic TX 1000000 53507 75334 712642 6250 B) For SCHED_FIFO policy and schpri=50: $ xdpsock -i eth0 -t -N -z -T 1000 -b 16 -C 100000 -a -W FIFO -U 50 period min ave max cycle Cyclic TX 1000000 3699 24859 54397 6250 Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211230035447.523177-6-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 61 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 59 insertions(+), 2 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index a2a42ec4b0e9..b7d0f536f974 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -31,6 +31,7 @@ #include #include #include +#include #include #include @@ -66,6 +67,8 @@ #define NSEC_PER_SEC 1000000000UL #define NSEC_PER_USEC 1000 +#define SCHED_PRI__DEFAULT 0 + typedef __u64 u64; typedef __u32 u32; typedef __u16 u16; @@ -123,6 +126,8 @@ static bool opt_busy_poll; static bool opt_reduced_cap; static clockid_t opt_clock = CLOCK_MONOTONIC; static unsigned long opt_tx_cycle_ns; +static int opt_schpolicy = SCHED_OTHER; +static int opt_schprio = SCHED_PRI__DEFAULT; struct vlan_ethhdr { unsigned char h_dest[6]; @@ -198,6 +203,15 @@ static const struct clockid_map { { NULL } }; +static const struct sched_map { + const char *name; + int policy; +} schmap[] = { + { "OTHER", SCHED_OTHER }, + { "FIFO", SCHED_FIFO }, + { NULL } +}; + static int num_socks; struct xsk_socket_info *xsks[MAX_SOCKS]; int sock; @@ -216,6 +230,20 @@ static int get_clockid(clockid_t *id, const char *name) return -1; } +static int get_schpolicy(int *policy, const char *name) +{ + const struct sched_map *sch; + + for (sch = schmap; sch->name; sch++) { + if (strcasecmp(sch->name, name) == 0) { + *policy = sch->policy; + return 0; + } + } + + return -1; +} + static unsigned long get_nsecs(void) { struct timespec ts; @@ -1019,6 +1047,8 @@ static struct option long_options[] = { {"tx-dmac", required_argument, 0, 'G'}, {"tx-smac", required_argument, 0, 'H'}, {"tx-cycle", required_argument, 0, 'T'}, + {"policy", required_argument, 0, 'W'}, + {"schpri", required_argument, 0, 'U'}, {"extra-stats", no_argument, 0, 'x'}, {"quiet", no_argument, 0, 'Q'}, {"app-stats", no_argument, 0, 'a'}, @@ -1066,6 +1096,8 @@ static void usage(const char *prog) " -G, --tx-dmac= Dest MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -H, --tx-smac= Src MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -T, --tx-cycle=n Tx cycle time in micro-seconds (For -t|--txonly).\n" + " -W, --policy=POLICY Schedule policy. Default: SCHED_OTHER\n" + " -U, --schpri=n Schedule priority. Default: %d\n" " -x, --extra-stats Display extra statistics.\n" " -Q, --quiet Do not display any stats.\n" " -a, --app-stats Display application (syscall) statistics.\n" @@ -1076,7 +1108,8 @@ static void usage(const char *prog) fprintf(stderr, str, prog, XSK_UMEM__DEFAULT_FRAME_SIZE, opt_batch_size, MIN_PKT_SIZE, MIN_PKT_SIZE, XSK_UMEM__DEFAULT_FRAME_SIZE, opt_pkt_fill_pattern, - VLAN_VID__DEFAULT, VLAN_PRI__DEFAULT); + VLAN_VID__DEFAULT, VLAN_PRI__DEFAULT, + SCHED_PRI__DEFAULT); exit(EXIT_FAILURE); } @@ -1088,7 +1121,8 @@ static void parse_command_line(int argc, char **argv) opterr = 0; for (;;) { - c = getopt_long(argc, argv, "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:T:xQaI:BR", + c = getopt_long(argc, argv, + "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:T:W:U:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1206,6 +1240,17 @@ static void parse_command_line(int argc, char **argv) opt_tx_cycle_ns = atoi(optarg); opt_tx_cycle_ns *= NSEC_PER_USEC; break; + case 'W': + if (get_schpolicy(&opt_schpolicy, optarg)) { + fprintf(stderr, + "ERROR: Invalid policy %s. Default to SCHED_OTHER.\n", + optarg); + opt_schpolicy = SCHED_OTHER; + } + break; + case 'U': + opt_schprio = atoi(optarg); + break; case 'x': opt_extra_stats = 1; break; @@ -1780,6 +1825,7 @@ int main(int argc, char **argv) struct __user_cap_data_struct data[2] = { { 0 } }; struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY}; bool rx = false, tx = false; + struct sched_param schparam; struct xsk_umem_info *umem; struct bpf_object *obj; int xsks_map_fd = 0; @@ -1881,6 +1927,16 @@ int main(int argc, char **argv) prev_time = get_nsecs(); start_time = prev_time; + /* Configure sched priority for better wake-up accuracy */ + memset(&schparam, 0, sizeof(schparam)); + schparam.sched_priority = opt_schprio; + ret = sched_setscheduler(0, opt_schpolicy, &schparam); + if (ret) { + fprintf(stderr, "Error(%d) in setting priority(%d): %s\n", + errno, opt_schprio, strerror(errno)); + goto out; + } + if (opt_bench == BENCH_RXDROP) rx_drop_all(); else if (opt_bench == BENCH_TXONLY) @@ -1888,6 +1944,7 @@ int main(int argc, char **argv) else l2fwd_all(); +out: benchmark_done = true; if (!opt_quiet) From 8121e78932018df48758985e00651e16ff34ae5f Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:46 +0800 Subject: [PATCH 6/7] samples/bpf: xdpsock: Add time-out for cleaning Tx When user sets tx-pkt-count and in case where there are invalid Tx frame, the complete_tx_only_all() process polls indefinitely. So, this patch adds a time-out mechanism into the process so that the application can terminate automatically after it retries 3*polling interval duration. v1->v2: Thanks to Jesper's and Song Liu's suggestion. - clean-up git message to remove polling log - make the Tx time-out retries configurable with 1s granularity Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211230035447.523177-7-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index b7d0f536f974..319cb3cdb226 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -113,6 +113,7 @@ static u32 irq_no; static int irqs_at_init = -1; static int opt_poll; static int opt_interval = 1; +static int opt_retries = 3; static u32 opt_xdp_bind_flags = XDP_USE_NEED_WAKEUP; static u32 opt_umem_flags; static int opt_unaligned_chunks; @@ -1028,6 +1029,7 @@ static struct option long_options[] = { {"xdp-skb", no_argument, 0, 'S'}, {"xdp-native", no_argument, 0, 'N'}, {"interval", required_argument, 0, 'n'}, + {"retries", required_argument, 0, 'O'}, {"zero-copy", no_argument, 0, 'z'}, {"copy", no_argument, 0, 'c'}, {"frame-size", required_argument, 0, 'f'}, @@ -1072,6 +1074,7 @@ static void usage(const char *prog) " -S, --xdp-skb=n Use XDP skb-mod\n" " -N, --xdp-native=n Enforce XDP native mode\n" " -n, --interval=n Specify statistics update interval (default 1 sec).\n" + " -O, --retries=n Specify time-out retries (1s interval) attempt (default 3).\n" " -z, --zero-copy Force zero-copy mode.\n" " -c, --copy Force copy mode.\n" " -m, --no-need-wakeup Turn off use of driver need wakeup flag.\n" @@ -1122,7 +1125,7 @@ static void parse_command_line(int argc, char **argv) for (;;) { c = getopt_long(argc, argv, - "Frtli:q:pSNn:w:czf:muMd:b:C:s:P:VJ:K:G:H:T:W:U:xQaI:BR", + "Frtli:q:pSNn:w:O:czf:muMd:b:C:s:P:VJ:K:G:H:T:W:U:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1164,6 +1167,9 @@ static void parse_command_line(int argc, char **argv) opt_clock = CLOCK_MONOTONIC; } break; + case 'O': + opt_retries = atoi(optarg); + break; case 'z': opt_xdp_bind_flags |= XDP_ZEROCOPY; break; @@ -1509,7 +1515,8 @@ static void complete_tx_only_all(void) pending = !!xsks[i]->outstanding_tx; } } - } while (pending); + sleep(1); + } while (pending && opt_retries-- > 0); } static void tx_only_all(void) From eb68db45b747756c351ea84e9af55a69468d0549 Mon Sep 17 00:00:00 2001 From: Ong Boon Leong Date: Thu, 30 Dec 2021 11:54:47 +0800 Subject: [PATCH 7/7] samples/bpf: xdpsock: Add timestamp for Tx-only operation It may be useful to add timestamp for Tx packets for continuous or cyclic transmit operation. The timestamp and sequence ID of a Tx packet are stored according to pktgen header format. To enable per-packet timestamp, use -y|--tstamp option. If timestamp is off, pktgen header is not included in the UDP payload. This means receiving side can use the magic number for pktgen for differentiation. The implementation supports both VLAN tagged and untagged option. By default, the minimum packet size is set at 64B. However, if VLAN tagged is on (-V), the minimum packet size is increased to 66B just so to fit the pktgen_hdr size. Added hex_dump() into the code path just for future cross-checking. As before, simply change to "#define DEBUG_HEXDUMP 1" to inspect the accuracy of TX packet. Signed-off-by: Ong Boon Leong Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211230035447.523177-8-boon.leong.ong@intel.com --- samples/bpf/xdpsock_user.c | 77 +++++++++++++++++++++++++++++++++----- 1 file changed, 68 insertions(+), 9 deletions(-) diff --git a/samples/bpf/xdpsock_user.c b/samples/bpf/xdpsock_user.c index 319cb3cdb226..aa50864e4415 100644 --- a/samples/bpf/xdpsock_user.c +++ b/samples/bpf/xdpsock_user.c @@ -111,6 +111,7 @@ static bool opt_app_stats; static const char *opt_irq_str = ""; static u32 irq_no; static int irqs_at_init = -1; +static u32 sequence; static int opt_poll; static int opt_interval = 1; static int opt_retries = 3; @@ -129,6 +130,7 @@ static clockid_t opt_clock = CLOCK_MONOTONIC; static unsigned long opt_tx_cycle_ns; static int opt_schpolicy = SCHED_OTHER; static int opt_schprio = SCHED_PRI__DEFAULT; +static bool opt_tstamp; struct vlan_ethhdr { unsigned char h_dest[6]; @@ -138,6 +140,14 @@ struct vlan_ethhdr { __be16 h_vlan_encapsulated_proto; }; +#define PKTGEN_MAGIC 0xbe9be955 +struct pktgen_hdr { + __be32 pgh_magic; + __be32 seq_num; + __be32 tv_sec; + __be32 tv_usec; +}; + struct xsk_ring_stats { unsigned long rx_npkts; unsigned long tx_npkts; @@ -836,18 +846,25 @@ static inline u16 udp_csum(u32 saddr, u32 daddr, u32 len, #define ETH_HDR_SIZE (opt_vlan_tag ? sizeof(struct vlan_ethhdr) : \ sizeof(struct ethhdr)) +#define PKTGEN_HDR_SIZE (opt_tstamp ? sizeof(struct pktgen_hdr) : 0) #define PKT_HDR_SIZE (ETH_HDR_SIZE + sizeof(struct iphdr) + \ - sizeof(struct udphdr)) + sizeof(struct udphdr) + PKTGEN_HDR_SIZE) +#define PKTGEN_HDR_OFFSET (ETH_HDR_SIZE + sizeof(struct iphdr) + \ + sizeof(struct udphdr)) +#define PKTGEN_SIZE_MIN (PKTGEN_HDR_OFFSET + sizeof(struct pktgen_hdr) + \ + ETH_FCS_SIZE) #define PKT_SIZE (opt_pkt_size - ETH_FCS_SIZE) #define IP_PKT_SIZE (PKT_SIZE - ETH_HDR_SIZE) #define UDP_PKT_SIZE (IP_PKT_SIZE - sizeof(struct iphdr)) -#define UDP_PKT_DATA_SIZE (UDP_PKT_SIZE - sizeof(struct udphdr)) +#define UDP_PKT_DATA_SIZE (UDP_PKT_SIZE - \ + (sizeof(struct udphdr) + PKTGEN_HDR_SIZE)) static u8 pkt_data[XSK_UMEM__DEFAULT_FRAME_SIZE]; static void gen_eth_hdr_data(void) { + struct pktgen_hdr *pktgen_hdr; struct udphdr *udp_hdr; struct iphdr *ip_hdr; @@ -860,7 +877,10 @@ static void gen_eth_hdr_data(void) sizeof(struct iphdr)); ip_hdr = (struct iphdr *)(pkt_data + sizeof(struct vlan_ethhdr)); - + pktgen_hdr = (struct pktgen_hdr *)(pkt_data + + sizeof(struct vlan_ethhdr) + + sizeof(struct iphdr) + + sizeof(struct udphdr)); /* ethernet & VLAN header */ memcpy(veth_hdr->h_dest, &opt_txdmac, ETH_ALEN); memcpy(veth_hdr->h_source, &opt_txsmac, ETH_ALEN); @@ -877,7 +897,10 @@ static void gen_eth_hdr_data(void) sizeof(struct iphdr)); ip_hdr = (struct iphdr *)(pkt_data + sizeof(struct ethhdr)); - + pktgen_hdr = (struct pktgen_hdr *)(pkt_data + + sizeof(struct ethhdr) + + sizeof(struct iphdr) + + sizeof(struct udphdr)); /* ethernet header */ memcpy(eth_hdr->h_dest, &opt_txdmac, ETH_ALEN); memcpy(eth_hdr->h_source, &opt_txsmac, ETH_ALEN); @@ -906,6 +929,9 @@ static void gen_eth_hdr_data(void) udp_hdr->dest = htons(0x1000); udp_hdr->len = htons(UDP_PKT_SIZE); + if (opt_tstamp) + pktgen_hdr->pgh_magic = htonl(PKTGEN_MAGIC); + /* UDP data */ memset32_htonl(pkt_data + PKT_HDR_SIZE, opt_pkt_fill_pattern, UDP_PKT_DATA_SIZE); @@ -1049,6 +1075,7 @@ static struct option long_options[] = { {"tx-dmac", required_argument, 0, 'G'}, {"tx-smac", required_argument, 0, 'H'}, {"tx-cycle", required_argument, 0, 'T'}, + {"tstamp", no_argument, 0, 'y'}, {"policy", required_argument, 0, 'W'}, {"schpri", required_argument, 0, 'U'}, {"extra-stats", no_argument, 0, 'x'}, @@ -1099,6 +1126,7 @@ static void usage(const char *prog) " -G, --tx-dmac= Dest MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -H, --tx-smac= Src MAC addr of TX frame in aa:bb:cc:dd:ee:ff format (For -V|--tx-vlan)\n" " -T, --tx-cycle=n Tx cycle time in micro-seconds (For -t|--txonly).\n" + " -y, --tstamp Add time-stamp to packet (For -t|--txonly).\n" " -W, --policy=POLICY Schedule policy. Default: SCHED_OTHER\n" " -U, --schpri=n Schedule priority. Default: %d\n" " -x, --extra-stats Display extra statistics.\n" @@ -1125,7 +1153,7 @@ static void parse_command_line(int argc, char **argv) for (;;) { c = getopt_long(argc, argv, - "Frtli:q:pSNn:w:O:czf:muMd:b:C:s:P:VJ:K:G:H:T:W:U:xQaI:BR", + "Frtli:q:pSNn:w:O:czf:muMd:b:C:s:P:VJ:K:G:H:T:yW:U:xQaI:BR", long_options, &option_index); if (c == -1) break; @@ -1246,6 +1274,9 @@ static void parse_command_line(int argc, char **argv) opt_tx_cycle_ns = atoi(optarg); opt_tx_cycle_ns *= NSEC_PER_USEC; break; + case 'y': + opt_tstamp = 1; + break; case 'W': if (get_schpolicy(&opt_schpolicy, optarg)) { fprintf(stderr, @@ -1462,9 +1493,10 @@ static void rx_drop_all(void) } } -static int tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) +static int tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, + int batch_size, unsigned long tx_ns) { - u32 idx; + u32 idx, tv_sec, tv_usec; unsigned int i; while (xsk_ring_prod__reserve(&xsk->tx, batch_size, &idx) < @@ -1474,11 +1506,31 @@ static int tx_only(struct xsk_socket_info *xsk, u32 *frame_nb, int batch_size) return 0; } + if (opt_tstamp) { + tv_sec = (u32)(tx_ns / NSEC_PER_SEC); + tv_usec = (u32)((tx_ns % NSEC_PER_SEC) / 1000); + } + for (i = 0; i < batch_size; i++) { struct xdp_desc *tx_desc = xsk_ring_prod__tx_desc(&xsk->tx, idx + i); tx_desc->addr = (*frame_nb + i) * opt_xsk_frame_size; tx_desc->len = PKT_SIZE; + + if (opt_tstamp) { + struct pktgen_hdr *pktgen_hdr; + u64 addr = tx_desc->addr; + char *pkt; + + pkt = xsk_umem__get_data(xsk->umem->buffer, addr); + pktgen_hdr = (struct pktgen_hdr *)(pkt + PKTGEN_HDR_OFFSET); + + pktgen_hdr->seq_num = htonl(sequence++); + pktgen_hdr->tv_sec = htonl(tv_sec); + pktgen_hdr->tv_usec = htonl(tv_usec); + + hex_dump(pkt, PKT_SIZE, addr); + } } xsk_ring_prod__submit(&xsk->tx, batch_size); @@ -1552,6 +1604,7 @@ static void tx_only_all(void) while ((opt_pkt_count && pkt_cnt < opt_pkt_count) || !opt_pkt_count) { int batch_size = get_batch_size(pkt_cnt); + unsigned long tx_ns = 0; struct timespec next; int tx_cnt = 0; long diff; @@ -1581,7 +1634,8 @@ static void tx_only_all(void) } /* Measure periodic Tx scheduling variance */ - diff = get_nsecs() - next_tx_ns; + tx_ns = get_nsecs(); + diff = tx_ns - next_tx_ns; if (diff < tx_cycle_diff_min) tx_cycle_diff_min = diff; @@ -1590,10 +1644,12 @@ static void tx_only_all(void) tx_cycle_diff_ave += (double)diff; tx_cycle_cnt++; + } else if (opt_tstamp) { + tx_ns = get_nsecs(); } for (i = 0; i < num_socks; i++) - tx_cnt += tx_only(xsks[i], &frame_nb[i], batch_size); + tx_cnt += tx_only(xsks[i], &frame_nb[i], batch_size, tx_ns); pkt_cnt += tx_cnt; @@ -1895,6 +1951,9 @@ int main(int argc, char **argv) apply_setsockopt(xsks[i]); if (opt_bench == BENCH_TXONLY) { + if (opt_tstamp && opt_pkt_size < PKTGEN_SIZE_MIN) + opt_pkt_size = PKTGEN_SIZE_MIN; + gen_eth_hdr_data(); for (i = 0; i < NUM_FRAMES; i++)