// SPDX-License-Identifier: GPL-2.0 /* Toeplitz test * * 1. Read packets and their rx_hash using PF_PACKET/TPACKET_V3 * 2. Compute the rx_hash in software based on the packet contents * 3. Compare the two * * Optionally, either '-C $rx_irq_cpu_list' or '-r $rps_bitmap' may be given. * * If '-C $rx_irq_cpu_list' is given, also * * 4. Identify the cpu on which the packet arrived with PACKET_FANOUT_CPU * 5. Compute the rxqueue that RSS would select based on this rx_hash * 6. Using the $rx_irq_cpu_list map, identify the arriving cpu based on rxq irq * 7. Compare the cpus from 4 and 6 * * Else if '-r $rps_bitmap' is given, also * * 4. Identify the cpu on which the packet arrived with PACKET_FANOUT_CPU * 5. Compute the cpu that RPS should select based on rx_hash and $rps_bitmap * 6. Compare the cpus from 4 and 5 */ #define _GNU_SOURCE #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 #define TOEPLITZ_KEY_MIN_LEN 40 #define TOEPLITZ_KEY_MAX_LEN 60 #define TOEPLITZ_STR_LEN(K) (((K) * 3) - 1) /* hex encoded: AA:BB:CC:...:ZZ */ #define TOEPLITZ_STR_MIN_LEN TOEPLITZ_STR_LEN(TOEPLITZ_KEY_MIN_LEN) #define TOEPLITZ_STR_MAX_LEN TOEPLITZ_STR_LEN(TOEPLITZ_KEY_MAX_LEN) #define FOUR_TUPLE_MAX_LEN ((sizeof(struct in6_addr) * 2) + (sizeof(uint16_t) * 2)) #define RSS_MAX_CPUS (1 << 16) /* real constraint is PACKET_FANOUT_MAX */ #define RPS_MAX_CPUS 16UL /* must be a power of 2 */ /* configuration options (cmdline arguments) */ static uint16_t cfg_dport = 8000; static int cfg_family = AF_INET6; static char *cfg_ifname = "eth0"; static int cfg_num_queues; static int cfg_num_rps_cpus; static bool cfg_sink; static int cfg_type = SOCK_STREAM; static int cfg_timeout_msec = 1000; static bool cfg_verbose; /* global vars */ static int num_cpus; static int ring_block_nr; static int ring_block_sz; /* stats */ static int frames_received; static int frames_nohash; static int frames_error; #define log_verbose(args...) do { if (cfg_verbose) fprintf(stderr, args); } while (0) /* tpacket ring */ struct ring_state { int fd; char *mmap; int idx; int cpu; }; static unsigned int rx_irq_cpus[RSS_MAX_CPUS]; /* map from rxq to cpu */ static int rps_silo_to_cpu[RPS_MAX_CPUS]; static unsigned char toeplitz_key[TOEPLITZ_KEY_MAX_LEN]; static struct ring_state rings[RSS_MAX_CPUS]; static inline uint32_t toeplitz(const unsigned char *four_tuple, const unsigned char *key) { int i, bit, ret = 0; uint32_t key32; key32 = ntohl(*((uint32_t *)key)); key += 4; for (i = 0; i < FOUR_TUPLE_MAX_LEN; i++) { for (bit = 7; bit >= 0; bit--) { if (four_tuple[i] & (1 << bit)) ret ^= key32; key32 <<= 1; key32 |= !!(key[0] & (1 << bit)); } key++; } return ret; } /* Compare computed cpu with arrival cpu from packet_fanout_cpu */ static void verify_rss(uint32_t rx_hash, int cpu) { int queue = rx_hash % cfg_num_queues; log_verbose(" rxq %d (cpu %d)", queue, rx_irq_cpus[queue]); if (rx_irq_cpus[queue] != cpu) { log_verbose(". error: rss cpu mismatch (%d)", cpu); frames_error++; } } static void verify_rps(uint64_t rx_hash, int cpu) { int silo = (rx_hash * cfg_num_rps_cpus) >> 32; log_verbose(" silo %d (cpu %d)", silo, rps_silo_to_cpu[silo]); if (rps_silo_to_cpu[silo] != cpu) { log_verbose(". error: rps cpu mismatch (%d)", cpu); frames_error++; } } static void log_rxhash(int cpu, uint32_t rx_hash, const char *addrs, int addr_len) { char saddr[INET6_ADDRSTRLEN], daddr[INET6_ADDRSTRLEN]; uint16_t *ports; if (!inet_ntop(cfg_family, addrs, saddr, sizeof(saddr)) || !inet_ntop(cfg_family, addrs + addr_len, daddr, sizeof(daddr))) error(1, 0, "address parse error"); ports = (void *)addrs + (addr_len * 2); log_verbose("cpu %d: rx_hash 0x%08x [saddr %s daddr %s sport %02hu dport %02hu]", cpu, rx_hash, saddr, daddr, ntohs(ports[0]), ntohs(ports[1])); } /* Compare computed rxhash with rxhash received from tpacket_v3 */ static void verify_rxhash(const char *pkt, uint32_t rx_hash, int cpu) { unsigned char four_tuple[FOUR_TUPLE_MAX_LEN] = {0}; uint32_t rx_hash_sw; const char *addrs; int addr_len; if (cfg_family == AF_INET) { addr_len = sizeof(struct in_addr); addrs = pkt + offsetof(struct iphdr, saddr); } else { addr_len = sizeof(struct in6_addr); addrs = pkt + offsetof(struct ip6_hdr, ip6_src); } memcpy(four_tuple, addrs, (addr_len * 2) + (sizeof(uint16_t) * 2)); rx_hash_sw = toeplitz(four_tuple, toeplitz_key); if (cfg_verbose) log_rxhash(cpu, rx_hash, addrs, addr_len); if (rx_hash != rx_hash_sw) { log_verbose(" != expected 0x%x\n", rx_hash_sw); frames_error++; return; } log_verbose(" OK"); if (cfg_num_queues) verify_rss(rx_hash, cpu); else if (cfg_num_rps_cpus) verify_rps(rx_hash, cpu); log_verbose("\n"); } static char *recv_frame(const struct ring_state *ring, char *frame) { struct tpacket3_hdr *hdr = (void *)frame; if (hdr->hv1.tp_rxhash) verify_rxhash(frame + hdr->tp_net, hdr->hv1.tp_rxhash, ring->cpu); else frames_nohash++; return frame + hdr->tp_next_offset; } /* A single TPACKET_V3 block can hold multiple frames */ static void recv_block(struct ring_state *ring) { struct tpacket_block_desc *block; char *frame; int i; block = (void *)(ring->mmap + ring->idx * ring_block_sz); if (!(block->hdr.bh1.block_status & TP_STATUS_USER)) return; frame = (char *)block; frame += block->hdr.bh1.offset_to_first_pkt; for (i = 0; i < block->hdr.bh1.num_pkts; i++) { frame = recv_frame(ring, frame); frames_received++; } block->hdr.bh1.block_status = TP_STATUS_KERNEL; ring->idx = (ring->idx + 1) % ring_block_nr; } /* simple test: sleep once unconditionally and then process all rings */ static void process_rings(void) { int i; usleep(1000 * cfg_timeout_msec); for (i = 0; i < num_cpus; i++) recv_block(&rings[i]); fprintf(stderr, "count: pass=%u nohash=%u fail=%u\n", frames_received - frames_nohash - frames_error, frames_nohash, frames_error); } static char *setup_ring(int fd) { struct tpacket_req3 req3 = {0}; void *ring; req3.tp_retire_blk_tov = cfg_timeout_msec; req3.tp_feature_req_word = TP_FT_REQ_FILL_RXHASH; req3.tp_frame_size = 2048; req3.tp_frame_nr = 1 << 10; req3.tp_block_nr = 2; req3.tp_block_size = req3.tp_frame_size * req3.tp_frame_nr; req3.tp_block_size /= req3.tp_block_nr; if (setsockopt(fd, SOL_PACKET, PACKET_RX_RING, &req3, sizeof(req3))) error(1, errno, "setsockopt PACKET_RX_RING"); ring_block_sz = req3.tp_block_size; ring_block_nr = req3.tp_block_nr; ring = mmap(0, req3.tp_block_size * req3.tp_block_nr, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED | MAP_POPULATE, fd, 0); if (ring == MAP_FAILED) error(1, 0, "mmap failed"); return ring; } static void __set_filter(int fd, int off_proto, uint8_t proto, int off_dport) { struct sock_filter filter[] = { BPF_STMT(BPF_LD + BPF_B + BPF_ABS, SKF_AD_OFF + SKF_AD_PKTTYPE), BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, PACKET_HOST, 0, 4), BPF_STMT(BPF_LD + BPF_B + BPF_ABS, off_proto), BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, proto, 0, 2), BPF_STMT(BPF_LD + BPF_H + BPF_ABS, off_dport), BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, cfg_dport, 1, 0), BPF_STMT(BPF_RET + BPF_K, 0), BPF_STMT(BPF_RET + BPF_K, 0xFFFF), }; struct sock_fprog prog = {}; prog.filter = filter; prog.len = sizeof(filter) / sizeof(struct sock_filter); if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &prog, sizeof(prog))) error(1, errno, "setsockopt filter"); } /* filter on transport protocol and destination port */ static void set_filter(int fd) { const int off_dport = offsetof(struct tcphdr, dest); /* same for udp */ uint8_t proto; proto = cfg_type == SOCK_STREAM ? IPPROTO_TCP : IPPROTO_UDP; if (cfg_family == AF_INET) __set_filter(fd, offsetof(struct iphdr, protocol), proto, sizeof(struct iphdr) + off_dport); else __set_filter(fd, offsetof(struct ip6_hdr, ip6_nxt), proto, sizeof(struct ip6_hdr) + off_dport); } /* drop everything: used temporarily during setup */ static void set_filter_null(int fd) { struct sock_filter filter[] = { BPF_STMT(BPF_RET + BPF_K, 0), }; struct sock_fprog prog = {}; prog.filter = filter; prog.len = sizeof(filter) / sizeof(struct sock_filter); if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &prog, sizeof(prog))) error(1, errno, "setsockopt filter"); } static int create_ring(char **ring) { struct fanout_args args = { .id = 1, .type_flags = PACKET_FANOUT_CPU, .max_num_members = RSS_MAX_CPUS }; struct sockaddr_ll ll = { 0 }; int fd, val; fd = socket(PF_PACKET, SOCK_DGRAM, 0); if (fd == -1) error(1, errno, "socket creation failed"); val = TPACKET_V3; if (setsockopt(fd, SOL_PACKET, PACKET_VERSION, &val, sizeof(val))) error(1, errno, "setsockopt PACKET_VERSION"); *ring = setup_ring(fd); /* block packets until all rings are added to the fanout group: * else packets can arrive during setup and get misclassified */ set_filter_null(fd); ll.sll_family = AF_PACKET; ll.sll_ifindex = if_nametoindex(cfg_ifname); ll.sll_protocol = cfg_family == AF_INET ? htons(ETH_P_IP) : htons(ETH_P_IPV6); if (bind(fd, (void *)&ll, sizeof(ll))) error(1, errno, "bind"); /* must come after bind: verifies all programs in group match */ if (setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &args, sizeof(args))) { /* on failure, retry using old API if that is sufficient: * it has a hard limit of 256 sockets, so only try if * (a) only testing rxhash, not RSS or (b) <= 256 cpus. * in this API, the third argument is left implicit. */ if (cfg_num_queues || num_cpus > 256 || setsockopt(fd, SOL_PACKET, PACKET_FANOUT, &args, sizeof(uint32_t))) error(1, errno, "setsockopt PACKET_FANOUT cpu"); } return fd; } /* setup inet(6) socket to blackhole the test traffic, if arg '-s' */ static int setup_sink(void) { int fd, val; fd = socket(cfg_family, cfg_type, 0); if (fd == -1) error(1, errno, "socket %d.%d", cfg_family, cfg_type); val = 1 << 20; if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &val, sizeof(val))) error(1, errno, "setsockopt rcvbuf"); return fd; } static void setup_rings(void) { int i; for (i = 0; i < num_cpus; i++) { rings[i].cpu = i; rings[i].fd = create_ring(&rings[i].mmap); } /* accept packets once all rings in the fanout group are up */ for (i = 0; i < num_cpus; i++) set_filter(rings[i].fd); } static void cleanup_rings(void) { int i; for (i = 0; i < num_cpus; i++) { if (munmap(rings[i].mmap, ring_block_nr * ring_block_sz)) error(1, errno, "munmap"); if (close(rings[i].fd)) error(1, errno, "close"); } } static void parse_cpulist(const char *arg) { do { rx_irq_cpus[cfg_num_queues++] = strtol(arg, NULL, 10); arg = strchr(arg, ','); if (!arg) break; arg++; // skip ',' } while (1); } static void show_cpulist(void) { int i; for (i = 0; i < cfg_num_queues; i++) fprintf(stderr, "rxq %d: cpu %d\n", i, rx_irq_cpus[i]); } static void show_silos(void) { int i; for (i = 0; i < cfg_num_rps_cpus; i++) fprintf(stderr, "silo %d: cpu %d\n", i, rps_silo_to_cpu[i]); } static void parse_toeplitz_key(const char *str, int slen, unsigned char *key) { int i, ret, off; if (slen < TOEPLITZ_STR_MIN_LEN || slen > TOEPLITZ_STR_MAX_LEN + 1) error(1, 0, "invalid toeplitz key"); for (i = 0, off = 0; off < slen; i++, off += 3) { ret = sscanf(str + off, "%hhx", &key[i]); if (ret != 1) error(1, 0, "key parse error at %d off %d len %d", i, off, slen); } } static void parse_rps_bitmap(const char *arg) { unsigned long bitmap; int i; bitmap = strtoul(arg, NULL, 0); if (bitmap & ~(RPS_MAX_CPUS - 1)) error(1, 0, "rps bitmap 0x%lx out of bounds 0..%lu", bitmap, RPS_MAX_CPUS - 1); for (i = 0; i < RPS_MAX_CPUS; i++) if (bitmap & 1UL << i) rps_silo_to_cpu[cfg_num_rps_cpus++] = i; } static void parse_opts(int argc, char **argv) { static struct option long_options[] = { {"dport", required_argument, 0, 'd'}, {"cpus", required_argument, 0, 'C'}, {"key", required_argument, 0, 'k'}, {"iface", required_argument, 0, 'i'}, {"ipv4", no_argument, 0, '4'}, {"ipv6", no_argument, 0, '6'}, {"sink", no_argument, 0, 's'}, {"tcp", no_argument, 0, 't'}, {"timeout", required_argument, 0, 'T'}, {"udp", no_argument, 0, 'u'}, {"verbose", no_argument, 0, 'v'}, {"rps", required_argument, 0, 'r'}, {0, 0, 0, 0} }; bool have_toeplitz = false; int index, c; while ((c = getopt_long(argc, argv, "46C:d:i:k:r:stT:u:v", long_options, &index)) != -1) { switch (c) { case '4': cfg_family = AF_INET; break; case '6': cfg_family = AF_INET6; break; case 'C': parse_cpulist(optarg); break; case 'd': cfg_dport = strtol(optarg, NULL, 0); break; case 'i': cfg_ifname = optarg; break; case 'k': parse_toeplitz_key(optarg, strlen(optarg), toeplitz_key); have_toeplitz = true; break; case 'r': parse_rps_bitmap(optarg); break; case 's': cfg_sink = true; break; case 't': cfg_type = SOCK_STREAM; break; case 'T': cfg_timeout_msec = strtol(optarg, NULL, 0); break; case 'u': cfg_type = SOCK_DGRAM; break; case 'v': cfg_verbose = true; break; default: error(1, 0, "unknown option %c", optopt); break; } } if (!have_toeplitz) error(1, 0, "Must supply rss key ('-k')"); num_cpus = get_nprocs(); if (num_cpus > RSS_MAX_CPUS) error(1, 0, "increase RSS_MAX_CPUS"); if (cfg_num_queues && cfg_num_rps_cpus) error(1, 0, "Can't supply both RSS cpus ('-C') and RPS map ('-r')"); if (cfg_verbose) { show_cpulist(); show_silos(); } } int main(int argc, char **argv) { const int min_tests = 10; int fd_sink = -1; parse_opts(argc, argv); if (cfg_sink) fd_sink = setup_sink(); setup_rings(); process_rings(); cleanup_rings(); if (cfg_sink && close(fd_sink)) error(1, errno, "close sink"); if (frames_received - frames_nohash < min_tests) error(1, 0, "too few frames for verification"); return frames_error; }