git-lng-odp May 2018

git-lng-odp@lists.linaro.org

1 participants
23 discussions

[COMMIT] branch master updated. v1.19.0.0_tigermoth-31-gd189d704

by git＠git-us.linaro.org

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "". The branch, master has been updated via d189d704a29f490583057ccb5c762432a4e4b7a7 (commit) from 122b74c0074329f1758b2ed5d241e0e9a790087b (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit d189d704a29f490583057ccb5c762432a4e4b7a7 Author: Maxim Uvarov <maxim.uvarov(a)linaro.org> Date: Fri May 4 12:33:26 2018 +0300 linux-gen: pktio: ipc free ring memory on error ipc pktio can fail on open if second process is not yet ready to connect. But on this error free of ring memory is missing which leads to taking all available shm descriptors. https://bugs.linaro.org/show_bug.cgi?id=3788 Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> diff --git a/platform/linux-generic/pktio/ipc.c b/platform/linux-generic/pktio/ipc.c index 08e3e4bc..d1e30ab8 100644 --- a/platform/linux-generic/pktio/ipc.c +++ b/platform/linux-generic/pktio/ipc.c @@ -347,6 +347,8 @@ static int ipc_pktio_open(odp_pktio_t id ODP_UNUSED, pktio_entry->s.ipc.rx.cache = _ring_create("ipc_rx_cache", PKTIO_IPC_ENTRIES, _RING_NO_LIST); + if (!pktio_entry->s.ipc.rx.cache) + return -1; /* Shared info about remote pktio */ if (sscanf(dev, "ipc:%d:%s", &pid, tail) == 2) { @@ -355,12 +357,15 @@ static int ipc_pktio_open(odp_pktio_t id ODP_UNUSED, snprintf(name, sizeof(name), "ipc:%s_info", tail); IPC_ODP_DBG("lookup for name %s for pid %d\n", name, pid); shm = odp_shm_import(name, pid, name); - if (ODP_SHM_INVALID == shm) + if (ODP_SHM_INVALID == shm) { + _ring_destroy("ipc_rx_cache"); return -1; + } pinfo = odp_shm_addr(shm); if (!pinfo->master.init_done) { odp_shm_free(shm); + _ring_destroy("ipc_rx_cache"); return -1; } pktio_entry->s.ipc.pinfo = pinfo; @@ -374,6 +379,7 @@ static int ipc_pktio_open(odp_pktio_t id ODP_UNUSED, ODP_CACHE_LINE_SIZE, _ODP_ISHM_EXPORT | _ODP_ISHM_LOCK); if (ODP_SHM_INVALID == shm) { + _ring_destroy("ipc_rx_cache"); ODP_ERR("can not create shm %s\n", name); return -1; } ----------------------------------------------------------------------- Summary of changes: platform/linux-generic/pktio/ipc.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) hooks/post-receive --

6 years, 9 months

[COMMIT] branch master updated. v1.19.0.0_tigermoth-30-g122b74c0

by git＠git-us.linaro.org

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "". The branch, master has been updated via 122b74c0074329f1758b2ed5d241e0e9a790087b (commit) from bd37df82b49faefef078558cdb2a54477bec6c5a (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 122b74c0074329f1758b2ed5d241e0e9a790087b Author: Dmitry Eremin-Solenikov <dmitry.ereminsolenikov(a)linaro.org> Date: Mon Apr 23 15:59:38 2018 +0300 build: odp_dpdk: add dependencies for crypto PMDs Signed-off-by: Dmitry Eremin-Solenikov <dmitry.ereminsolenikov(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Reviewed-by: Matias Elo <matias.elo(a)nokia.com> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/m4/odp_dpdk.m4 b/m4/odp_dpdk.m4 index ec248e76..2ef5253c 100644 --- a/m4/odp_dpdk.m4 +++ b/m4/odp_dpdk.m4 @@ -12,6 +12,12 @@ AS_CASE([$cur_driver], [rte_pmd_mlx4], [AS_VAR_APPEND([DPDK_LIBS], [" -lmlx4 -libverbs"])], [rte_pmd_mlx5], [AS_VAR_APPEND([DPDK_LIBS], [" -lmlx5 -libverbs"])], [rte_pmd_pcap], [AS_VAR_APPEND([DPDK_LIBS], [" -lpcap"])], + [rte_pmd_aesni_gcm], [AS_VAR_APPEND([DPDK_LIBS], [" -lIPSec_MB"])], + [rte_pmd_aesni_mb], [AS_VAR_APPEND([DPDK_LIBS], [" -lIPSec_MB"])], + [rte_pmd_kasumi], [AS_VAR_APPEND([DPDK_LIBS], [" -lsso_kasumi"])], + [rte_pmd_snow3g], [AS_VAR_APPEND([DPDK_LIBS], [" -lsso_snow3g"])], + [rte_pmd_zuc], [AS_VAR_APPEND([DPDK_LIBS], [" -lsso_zuc"])], + [rte_pmd_qat], [AS_VAR_APPEND([DPDK_LIBS], [" -lcrypto"])], [rte_pmd_openssl], [AS_VAR_APPEND([DPDK_LIBS], [" -lcrypto"])]) done AS_VAR_APPEND([DPDK_PMDS], [--no-whole-archive]) ----------------------------------------------------------------------- Summary of changes: m4/odp_dpdk.m4 | 6 ++++++ 1 file changed, 6 insertions(+) hooks/post-receive --

6 years, 9 months

[COMMIT] branch master updated. v1.19.0.0_tigermoth-29-gbd37df82

by git＠git-us.linaro.org

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "". The branch, master has been updated via bd37df82b49faefef078558cdb2a54477bec6c5a (commit) via dea03d0a24117764ea9b8763e640a4836031e82b (commit) via 5c745f55a160b0a72d8e493604917a0d63dd71a5 (commit) via 15a30694560ba62160fa3a8ffddeb51b800e351a (commit) via 27a84cdfc0dfb82e3b86067e473658c091377892 (commit) from 3600304c30c4a345538f6c06ca50183b14cd12b1 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit bd37df82b49faefef078558cdb2a54477bec6c5a Author: Petri Savolainen <petri.savolainen(a)linaro.org> Date: Fri Apr 27 13:33:58 2018 +0300 test: sched_pktio: add inactivity timer Add a timer per packet input queue to test timer performance with packet IO. The timer is reset every time packets are received from a queue, or when timer expires. Timer testing is enabled by setting the timeout period in usec with -t option. Signed-off-by: Petri Savolainen <petri.savolainen(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/test/performance/odp_sched_pktio.c b/test/performance/odp_sched_pktio.c index 840fe8da..46a08c92 100644 --- a/test/performance/odp_sched_pktio.c +++ b/test/performance/odp_sched_pktio.c @@ -25,8 +25,10 @@ #define BURST_SIZE 32 #define CHECK_PERIOD 10000 #define TEST_PASSED_LIMIT 5000 +#define TIMEOUT_OFFSET_NS 1000000 typedef struct test_options_t { + long int timeout_us; int num_worker; int num_pktio; int num_pktio_queue; @@ -80,6 +82,14 @@ typedef struct { } pktio[MAX_PKTIOS]; + struct { + odp_timer_pool_t timer_pool; + odp_pool_t timeout_pool; + uint64_t timeout_tick; + odp_timer_t timer[MAX_PKTIOS][MAX_PKTIO_QUEUES]; + + } timer; + worker_arg_t worker_arg[MAX_WORKERS]; worker_stat_t worker_stat[MAX_WORKERS]; @@ -186,6 +196,121 @@ static int worker_thread(void *arg) return 0; } +static int worker_thread_timers(void *arg) +{ + odp_event_t ev[BURST_SIZE]; + int num, sent, drop, out, tmos, i, src_pktio, src_queue; + odp_pktout_queue_t pktout; + odp_queue_t queue; + queue_context_t *queue_context; + odp_timer_t timer; + odp_timer_set_t ret; + worker_arg_t *worker_arg = arg; + test_global_t *test_global = worker_arg->test_global_ptr; + int worker_id = worker_arg->worker_id; + uint32_t polls = 0; + uint64_t tick = test_global->timer.timeout_tick; + + printf("Worker (timers) %i started\n", worker_id); + + /* Wait for other workers to start */ + odp_barrier_wait(&test_global->worker_start); + + while (1) { + odp_packet_t pkt[BURST_SIZE]; + + num = odp_schedule_multi(&queue, ODP_SCHED_NO_WAIT, + ev, BURST_SIZE); + + polls++; + + if (polls == CHECK_PERIOD) { + polls = 0; + if (test_global->stop_workers) + break; + } + + if (num <= 0) + continue; + + tmos = 0; + queue_context = odp_queue_context(queue); + src_pktio = queue_context->src_pktio; + src_queue = queue_context->src_queue; + timer = test_global->timer.timer[src_pktio][src_queue]; + + for (i = 0; i < num; i++) { + if (odp_unlikely(odp_event_type(ev[i]) == + ODP_EVENT_TIMEOUT)) { + tmos++; + ret = odp_timer_set_rel(timer, tick, &ev[i]); + + if (odp_unlikely(ret != ODP_TIMER_SUCCESS)) { + /* Should never happen */ + printf("Expired timer reset failed\n"); + odp_event_free(ev[i]); + } + + if (odp_unlikely(tmos > 1)) { + /* Should never happen */ + printf("Too many timeouts\n"); + } + } else { + pkt[i - tmos] = odp_packet_from_event(ev[i]); + } + } + + if (tmos == 0) { + /* Reset timer with existing timeout event */ + ret = odp_timer_set_rel(timer, tick, NULL); + + if (odp_unlikely(ret != ODP_TIMER_SUCCESS && + ret != ODP_TIMER_TOOEARLY)) { + /* Should never happen. Reset should either + * succeed or be too close to timer expiration + * in which case timeout event will be received + * soon. */ + printf("Timer reset failed %i\n", ret); + } + } + + num = num - tmos; + + if (DEBUG_PRINT) + printf("worker %i: [%i/%i] -> [%i/%i], %i packets " + "%i timeouts\n", + worker_id, + queue_context->src_pktio, + queue_context->src_queue, + queue_context->dst_pktio, + queue_context->dst_queue, num, tmos); + + pktout = queue_context->dst_pktout; + out = queue_context->dst_pktio; + + fill_eth_addr(pkt, num, test_global, out); + + sent = odp_pktout_send(pktout, pkt, num); + + if (odp_unlikely(sent < 0)) + sent = 0; + + drop = num - sent; + + if (odp_unlikely(drop)) + odp_packet_free_multi(&pkt[sent], drop); + + if (odp_unlikely(test_global->opt.collect_stat)) { + test_global->worker_stat[worker_id].rx_pkt += num; + test_global->worker_stat[worker_id].tx_pkt += sent; + } + } + + printf("Worker %i stopped\n", worker_id); + + return 0; +} + static void sig_handler(int signo) { (void)signo; @@ -210,6 +335,7 @@ static void print_usage(const char *progname) " -i, --interface <name> Packet IO interfaces (comma-separated, no spaces)\n" " -c, --num_cpu <number> Worker thread count. Default: 1\n" " -q, --num_queue <number> Number of pktio queues. Default: Worker thread count\n" + " -t, --timeout <number> Flow inactivity timeout (in usec) per packet. Default: 0 (don't use timers)\n" " -s, --stat Collect statistics.\n" " -h, --help Display help and exit.\n\n", NO_PATH(progname)); @@ -224,11 +350,12 @@ static int parse_options(int argc, char *argv[], test_options_t *test_options) {"interface", required_argument, NULL, 'i'}, {"num_cpu", required_argument, NULL, 'c'}, {"num_queue", required_argument, NULL, 'q'}, + {"timeout", required_argument, NULL, 't'}, {"stat", no_argument, NULL, 's'}, {"help", no_argument, NULL, 'h'}, {NULL, 0, NULL, 0} }; - const char *shortopts = "+i:c:q:sh"; + const char *shortopts = "+i:c:q:t:sh"; int ret = 0; memset(test_options, 0, sizeof(test_options_t)); @@ -283,6 +410,9 @@ static int parse_options(int argc, char *argv[], test_options_t *test_options) case 'q': test_options->num_pktio_queue = atoi(optarg); break; + case 't': + test_options->timeout_us = atol(optarg); + break; case 's': test_options->collect_stat = 1; break; @@ -382,6 +512,7 @@ static void print_config(test_global_t *test_global) test_global->opt.num_pktio_queue); printf(" collect statistics: %u\n", test_global->opt.collect_stat); + printf(" timeout usec: %li\n", test_global->opt.timeout_us); printf("\n"); } @@ -691,6 +822,178 @@ static int close_pktios(test_global_t *test_global) return ret; } +static int create_timers(test_global_t *test_global) +{ + int num_timer, num_pktio, num_queue, i, j; + odp_pool_t pool; + odp_pool_param_t pool_param; + odp_timer_pool_t timer_pool; + odp_timer_pool_param_t timer_param; + odp_timer_capability_t timer_capa; + odp_timer_t timer; + odp_queue_t queue; + uint64_t res_ns, tick; + uint64_t timeout_ns = 1000 * test_global->opt.timeout_us; + + num_pktio = test_global->opt.num_pktio; + num_queue = test_global->opt.num_pktio_queue; + num_timer = num_pktio * num_queue; + + /* Always init globals for destroy calls */ + test_global->timer.timer_pool = ODP_TIMER_POOL_INVALID; + test_global->timer.timeout_pool = ODP_POOL_INVALID; + + for (i = 0; i < num_pktio; i++) + for (j = 0; j < num_queue; j++) + test_global->timer.timer[i][j] = ODP_TIMER_INVALID; + + /* Timers not used */ + if (test_global->opt.timeout_us == 0) + return 0; + + if (odp_timer_capability(ODP_CLOCK_CPU, &timer_capa)) { + printf("Timer capa failed\n"); + return -1; + } + + res_ns = timeout_ns / 10; + + if (timer_capa.highest_res_ns > res_ns) { + printf("Timeout too short. Min timeout %" PRIu64 " usec\n", + timer_capa.highest_res_ns / 100); + return -1; + } + + memset(&timer_param, 0, sizeof(odp_timer_pool_param_t)); + + timer_param.res_ns = res_ns; + timer_param.min_tmo = timeout_ns; + timer_param.max_tmo = timeout_ns; + timer_param.num_timers = num_timer; + timer_param.clk_src = ODP_CLOCK_CPU; + + timer_pool = odp_timer_pool_create("sched_pktio_timer", &timer_param); + + if (timer_pool == ODP_TIMER_POOL_INVALID) { + printf("Timer pool create failed\n"); + return -1; + } + + test_global->timer.timer_pool = timer_pool; + tick = odp_timer_ns_to_tick(timer_pool, timeout_ns); + test_global->timer.timeout_tick = tick; + + odp_timer_pool_start(); + + for (i = 0; i < num_pktio; i++) { + for (j = 0; j < num_queue; j++) { + queue = test_global->pktio[i].input_queue[j]; + timer = odp_timer_alloc(timer_pool, queue, NULL); + + if (timer == ODP_TIMER_INVALID) { + printf("Timer alloc failed.\n"); + return -1; + } + + test_global->timer.timer[i][j] = timer; + } + } + + odp_pool_param_init(&pool_param); + pool_param.type = ODP_POOL_TIMEOUT; + pool_param.tmo.num = num_timer; + + pool = odp_pool_create("timeout pool", &pool_param); + + if (pool == ODP_POOL_INVALID) { + printf("Timeout pool create failed.\n"); + return -1; + } + + test_global->timer.timeout_pool = pool; + + return 0; +} + +static int start_timers(test_global_t *test_global) +{ + int i, j; + odp_event_t event; + odp_timeout_t timeout; + odp_timer_t timer; + odp_timer_set_t ret; + uint64_t timeout_tick = test_global->timer.timeout_tick; + int num_pktio = test_global->opt.num_pktio; + int num_queue = test_global->opt.num_pktio_queue; + odp_pool_t pool = test_global->timer.timeout_pool; + + /* Timers not used */ + if (test_global->opt.timeout_us == 0) + return 0; + + /* Delay the first timeout so that workers have time to startup */ + timeout_tick += odp_timer_ns_to_tick(test_global->timer.timer_pool, + TIMEOUT_OFFSET_NS); + + for (i = 0; i < num_pktio; i++) { + for (j = 0; j < num_queue; j++) { + timer = test_global->timer.timer[i][j]; + + timeout = odp_timeout_alloc(pool); + if (timeout == ODP_TIMEOUT_INVALID) { + printf("Timeout alloc failed\n"); + return -1; + } + + event = odp_timeout_to_event(timeout); + + ret = odp_timer_set_rel(timer, timeout_tick, &event); + + if (ret != ODP_TIMER_SUCCESS) { + printf("Timer set failed\n"); + return -1; + } + } + } + + return 0; +} + +static void destroy_timers(test_global_t *test_global) +{ + int i, j; + odp_event_t event; + odp_timer_t timer; + int num_pktio = test_global->opt.num_pktio; + int num_queue = test_global->opt.num_pktio_queue; + odp_timer_pool_t timer_pool = test_global->timer.timer_pool; + odp_pool_t pool = test_global->timer.timeout_pool; + + if (timer_pool == ODP_TIMER_POOL_INVALID) + return; + + for (i = 0; i < num_pktio; i++) { + for (j = 0; j < num_queue; j++) { + timer = test_global->timer.timer[i][j]; + + if (timer == ODP_TIMER_INVALID) + break; + + event = odp_timer_free(timer); + + if (event != ODP_EVENT_INVALID) + odp_event_free(event); + } + } + + if (pool != ODP_POOL_INVALID) { + if (odp_pool_destroy(pool)) + printf("Timeout pool destroy failed\n"); + } + + odp_timer_pool_destroy(timer_pool); +} + static void start_workers(odph_odpthread_t thread[], test_global_t *test_global) { @@ -700,7 +1003,12 @@ static void start_workers(odph_odpthread_t thread[], int num = test_global->opt.num_worker; memset(&param, 0, sizeof(odph_odpthread_params_t)); - param.start = worker_thread; + + if (test_global->opt.timeout_us) + param.start = worker_thread_timers; + else + param.start = worker_thread; + param.thr_type = ODP_THREAD_WORKER; param.instance = test_global->instance; @@ -745,8 +1053,11 @@ int main(int argc, char *argv[]) init.not_used.feat.cls = 1; init.not_used.feat.crypto = 1; init.not_used.feat.ipsec = 1; - init.not_used.feat.timer = 1; init.not_used.feat.tm = 1; + init.not_used.feat.timer = 1; + + if (test_options.timeout_us) + init.not_used.feat.timer = 0; /* Init ODP before calling anything else */ if (odp_init_global(&instance, &init, NULL)) { @@ -789,6 +1100,12 @@ int main(int argc, char *argv[]) link_pktios(test_global); + if (create_timers(test_global)) + goto quit; + + if (start_timers(test_global)) + goto quit; + odp_barrier_init(&test_global->worker_start, test_global->opt.num_worker + 1); @@ -813,6 +1130,7 @@ quit: stop_pktios(test_global); empty_queues(); close_pktios(test_global); + destroy_timers(test_global); if (test_global->opt.collect_stat) { print_stat(test_global, odp_time_diff_ns(t2, t1)); commit dea03d0a24117764ea9b8763e640a4836031e82b Author: Petri Savolainen <petri.savolainen(a)linaro.org> Date: Thu Apr 26 17:13:40 2018 +0300 test: sched_pktio: parse options before odp global init Options can be used to set parameters for odp_init_global(). Signed-off-by: Petri Savolainen <petri.savolainen(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/test/performance/odp_sched_pktio.c b/test/performance/odp_sched_pktio.c index f03c2911..840fe8da 100644 --- a/test/performance/odp_sched_pktio.c +++ b/test/performance/odp_sched_pktio.c @@ -26,6 +26,15 @@ #define CHECK_PERIOD 10000 #define TEST_PASSED_LIMIT 5000 +typedef struct test_options_t { + int num_worker; + int num_pktio; + int num_pktio_queue; + uint8_t collect_stat; + char pktio_name[MAX_PKTIOS][MAX_PKTIO_NAME + 1]; + +} test_options_t; + typedef struct { int worker_id; void *test_global_ptr; @@ -48,12 +57,7 @@ typedef struct { volatile int stop_workers; odp_barrier_t worker_start; - struct { - int num_worker; - int num_pktio; - int num_pktio_queue; - uint8_t collect_stat; - } opt; + test_options_t opt; int max_workers; odp_cpumask_t cpumask; @@ -66,7 +70,6 @@ typedef struct { uint32_t pkt_num; struct { - char name[MAX_PKTIO_NAME + 1]; odp_pktio_t pktio; int pktio_index; int started; @@ -212,7 +215,7 @@ static void print_usage(const char *progname) NO_PATH(progname)); } -static int parse_options(int argc, char *argv[], test_global_t *test_global) +static int parse_options(int argc, char *argv[], test_options_t *test_options) { int i, opt, long_index; char *name, *str; @@ -228,8 +231,10 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) const char *shortopts = "+i:c:q:sh"; int ret = 0; - test_global->opt.num_worker = 1; - test_global->opt.num_pktio_queue = 0; + memset(test_options, 0, sizeof(test_options_t)); + + test_options->num_worker = 1; + test_options->num_pktio_queue = 0; /* let helper collect its own arguments (e.g. --odph_proc) */ odph_parse_options(argc, argv, shortopts, longopts); @@ -263,23 +268,23 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) break; } - name = test_global->pktio[i].name; + name = test_options->pktio_name[i]; memcpy(name, str, len); str += len + 1; i++; } - test_global->opt.num_pktio = i; + test_options->num_pktio = i; break; case 'c': - test_global->opt.num_worker = atoi(optarg); + test_options->num_worker = atoi(optarg); break; case 'q': - test_global->opt.num_pktio_queue = atoi(optarg); + test_options->num_pktio_queue = atoi(optarg); break; case 's': - test_global->opt.collect_stat = 1; + test_options->collect_stat = 1; break; case 'h': print_usage(argv[0]); @@ -291,8 +296,8 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) } } - if (test_global->opt.num_pktio_queue == 0) - test_global->opt.num_pktio_queue = test_global->opt.num_worker; + if (test_options->num_pktio_queue == 0) + test_options->num_pktio_queue = test_options->num_worker; return ret; } @@ -370,7 +375,7 @@ static void print_config(test_global_t *test_global) " interface names: ", test_global->opt.num_pktio); for (i = 0; i < test_global->opt.num_pktio; i++) - printf(" %s", test_global->pktio[i].name); + printf(" %s", test_global->opt.pktio_name[i]); printf("\n" " queues per interface: %i\n", @@ -462,7 +467,7 @@ static int open_pktios(test_global_t *test_global) /* Open and configure interfaces */ for (i = 0; i < num_pktio; i++) { - name = test_global->pktio[i].name; + name = test_global->opt.pktio_name[i]; pktio = odp_pktio_open(name, pool, &pktio_param); if (pktio == ODP_PKTIO_INVALID) { @@ -605,7 +610,7 @@ static int start_pktios(test_global_t *test_global) for (i = 0; i < test_global->opt.num_pktio; i++) { if (odp_pktio_start(test_global->pktio[i].pktio)) { printf("Error (%s): Pktio start failed.\n", - test_global->pktio[i].name); + test_global->opt.pktio_name[i]); return -1; } @@ -630,7 +635,7 @@ static int stop_pktios(test_global_t *test_global) if (odp_pktio_stop(pktio)) { printf("Error (%s): Pktio stop failed.\n", - test_global->pktio[i].name); + test_global->opt.pktio_name[i]); ret = -1; } } @@ -668,7 +673,7 @@ static int close_pktios(test_global_t *test_global) if (odp_pktio_close(pktio)) { printf("Error (%s): Pktio close failed.\n", - test_global->pktio[i].name); + test_global->opt.pktio_name[i]); ret = -1; } } @@ -727,10 +732,14 @@ int main(int argc, char *argv[]) odp_shm_t shm; odp_time_t t1, t2; odph_odpthread_t thread[MAX_WORKERS]; + test_options_t test_options; int ret = 0; signal(SIGINT, sig_handler); + if (parse_options(argc, argv, &test_options)) + return -1; + /* List features not to be used (may optimize performance) */ odp_init_param_init(&init); init.not_used.feat.cls = 1; @@ -766,8 +775,7 @@ int main(int argc, char *argv[]) test_global->instance = instance; test_global->pool = ODP_POOL_INVALID; - if (parse_options(argc, argv, test_global)) - goto quit; + memcpy(&test_global->opt, &test_options, sizeof(test_options_t)); odp_sys_info_print(); commit 5c745f55a160b0a72d8e493604917a0d63dd71a5 Author: Petri Savolainen <petri.savolainen(a)linaro.org> Date: Thu Apr 26 14:18:57 2018 +0300 test: sched_pktio: ensure packet order per flow Select pktout queue based on input queue, instead of worker id. Packet order is maintained from input to output as each flow is routed always through the same queues. Signed-off-by: Petri Savolainen <petri.savolainen(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/test/performance/odp_sched_pktio.c b/test/performance/odp_sched_pktio.c index 1a16bb50..f03c2911 100644 --- a/test/performance/odp_sched_pktio.c +++ b/test/performance/odp_sched_pktio.c @@ -14,6 +14,7 @@ #include <odp_api.h> #include <odp/helper/odph_api.h> +#define DEBUG_PRINT 0 #define MAX_WORKERS 64 #define MAX_PKTIOS 32 #define MAX_PKTIO_NAME 31 @@ -23,7 +24,6 @@ #define MIN_PKT_SEG_LEN 64 #define BURST_SIZE 32 #define CHECK_PERIOD 10000 -#define MAX_PKTIO_INDEXES 256 #define TEST_PASSED_LIMIT 5000 typedef struct { @@ -36,6 +36,14 @@ typedef struct ODP_ALIGNED_CACHE { uint64_t tx_pkt; } worker_stat_t; +typedef struct queue_context_t { + odp_pktout_queue_t dst_pktout; + uint8_t dst_pktio; + uint8_t dst_queue; + uint8_t src_pktio; + uint8_t src_queue; +} queue_context_t; + typedef struct { volatile int stop_workers; odp_barrier_t worker_start; @@ -65,14 +73,12 @@ typedef struct { odph_ethaddr_t my_addr; odp_queue_t input_queue[MAX_PKTIO_QUEUES]; odp_pktout_queue_t pktout[MAX_PKTIO_QUEUES]; + queue_context_t queue_context[MAX_PKTIO_QUEUES]; } pktio[MAX_PKTIOS]; worker_arg_t worker_arg[MAX_WORKERS]; - /* Maps pktio input index to pktio[] index for output */ - uint8_t pktio_map[MAX_PKTIO_INDEXES]; - worker_stat_t worker_stat[MAX_WORKERS]; uint64_t rx_pkt_sum; uint64_t tx_pkt_sum; @@ -108,8 +114,10 @@ static inline void fill_eth_addr(odp_packet_t pkt[], int num, static int worker_thread(void *arg) { odp_event_t ev[BURST_SIZE]; - int num, sent, drop, in, out; + int num, sent, drop, out; odp_pktout_queue_t pktout; + odp_queue_t queue; + queue_context_t *queue_context; worker_arg_t *worker_arg = arg; test_global_t *test_global = worker_arg->test_global_ptr; int worker_id = worker_arg->worker_id; @@ -123,7 +131,7 @@ static int worker_thread(void *arg) while (1) { odp_packet_t pkt[BURST_SIZE]; - num = odp_schedule_multi(NULL, ODP_SCHED_NO_WAIT, + num = odp_schedule_multi(&queue, ODP_SCHED_NO_WAIT, ev, BURST_SIZE); polls++; @@ -137,11 +145,20 @@ static int worker_thread(void *arg) if (num <= 0) continue; + queue_context = odp_queue_context(queue); + + if (DEBUG_PRINT) + printf("worker %i: [%i/%i] -> [%i/%i], %i packets\n", + worker_id, + queue_context->src_pktio, + queue_context->src_queue, + queue_context->dst_pktio, + queue_context->dst_queue, num); + odp_packet_from_event_multi(pkt, ev, num); - in = odp_packet_input_index(pkt[0]); - out = test_global->pktio_map[in]; - pktout = test_global->pktio[out].pktout[worker_id]; + pktout = queue_context->dst_pktout; + out = queue_context->dst_pktio; fill_eth_addr(pkt, num, test_global, out); @@ -307,12 +324,6 @@ static int config_setup(test_global_t *test_global) return -1; } - if (MAX_PKTIO_INDEXES <= odp_pktio_max_index()) { - printf("Error: Larger pktio_map[] table needed: %u\n", - odp_pktio_max_index()); - return -1; - } - if (odp_pool_capability(&pool_capa)) { printf("Error: Pool capability failed.\n"); return -1; @@ -420,6 +431,7 @@ static int open_pktios(test_global_t *test_global) unsigned int num_queue; char *name; int i, num_pktio, ret; + unsigned int j; num_pktio = test_global->opt.num_pktio; num_queue = test_global->opt.num_pktio_queue; @@ -518,6 +530,21 @@ static int open_pktios(test_global_t *test_global) return -1; } + for (j = 0; j < num_queue; j++) { + odp_queue_t queue; + void *ctx; + uint32_t len = sizeof(queue_context_t); + + queue = test_global->pktio[i].input_queue[j]; + ctx = &test_global->pktio[i].queue_context[j]; + + if (odp_queue_context_set(queue, ctx, len)) { + printf("Error (%s): Queue ctx set failed.\n", + name); + return -1; + } + } + odp_pktout_queue_param_init(&pktout_param); pktout_param.num_queues = num_queue; pktout_param.op_mode = ODP_PKTIO_OP_MT_UNSAFE; @@ -542,20 +569,30 @@ static int open_pktios(test_global_t *test_global) static void link_pktios(test_global_t *test_global) { int i, num_pktio, input, output; + int num_queue; + odp_pktout_queue_t pktout; + queue_context_t *ctx; num_pktio = test_global->opt.num_pktio; + num_queue = test_global->opt.num_pktio_queue; printf("Forwarding table (pktio indexes)\n"); /* If single interface loopback, otherwise forward to the next * interface. */ - for (i = 0; i < num_pktio; i++) { - input = test_global->pktio[i].pktio_index; - output = (i + 1) % num_pktio; - test_global->pktio_map[input] = output; - printf(" input %i, output %i\n", - input, - test_global->pktio[output].pktio_index); + for (input = 0; input < num_pktio; input++) { + output = (input + 1) % num_pktio; + printf(" input %i, output %i\n", input, output); + + for (i = 0; i < num_queue; i++) { + ctx = &test_global->pktio[input].queue_context[i]; + pktout = test_global->pktio[output].pktout[i]; + ctx->dst_pktout = pktout; + ctx->dst_pktio = output; + ctx->dst_queue = i; + ctx->src_pktio = input; + ctx->src_queue = i; + } } printf("\n"); commit 15a30694560ba62160fa3a8ffddeb51b800e351a Author: Petri Savolainen <petri.savolainen(a)linaro.org> Date: Thu Apr 26 14:50:39 2018 +0300 test: sched_pktio: add pktio queue count option Added option for selecting number of input / output queues per pktio interface. For simplicity, the same number of queues is configured for all interfaces and both (input / output) directions. Signed-off-by: Petri Savolainen <petri.savolainen(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/test/performance/odp_sched_pktio.c b/test/performance/odp_sched_pktio.c index 16d14b19..1a16bb50 100644 --- a/test/performance/odp_sched_pktio.c +++ b/test/performance/odp_sched_pktio.c @@ -43,6 +43,7 @@ typedef struct { struct { int num_worker; int num_pktio; + int num_pktio_queue; uint8_t collect_stat; } opt; @@ -56,9 +57,6 @@ typedef struct { uint32_t pkt_len; uint32_t pkt_num; - unsigned int num_input_queues; - unsigned int num_output_queues; - struct { char name[MAX_PKTIO_NAME + 1]; odp_pktio_t pktio; @@ -189,10 +187,11 @@ static void print_usage(const char *progname) "Usage: %s [options]\n" "\n" "OPTIONS:\n" - " -i, --interface <name> Packet IO interfaces (comma-separated, no spaces)\n" - " -c, --count <number> Worker thread count. Default: 1\n" - " -s, --stat Collect statistics.\n" - " -h, --help Display help and exit.\n\n", + " -i, --interface <name> Packet IO interfaces (comma-separated, no spaces)\n" + " -c, --num_cpu <number> Worker thread count. Default: 1\n" + " -q, --num_queue <number> Number of pktio queues. Default: Worker thread count\n" + " -s, --stat Collect statistics.\n" + " -h, --help Display help and exit.\n\n", NO_PATH(progname)); } @@ -203,15 +202,17 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) int len, str_len; const struct option longopts[] = { {"interface", required_argument, NULL, 'i'}, - {"count", required_argument, NULL, 'c'}, + {"num_cpu", required_argument, NULL, 'c'}, + {"num_queue", required_argument, NULL, 'q'}, {"stat", no_argument, NULL, 's'}, {"help", no_argument, NULL, 'h'}, {NULL, 0, NULL, 0} }; - const char *shortopts = "+i:c:sh"; + const char *shortopts = "+i:c:q:sh"; int ret = 0; test_global->opt.num_worker = 1; + test_global->opt.num_pktio_queue = 0; /* let helper collect its own arguments (e.g. --odph_proc) */ odph_parse_options(argc, argv, shortopts, longopts); @@ -257,6 +258,9 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) case 'c': test_global->opt.num_worker = atoi(optarg); break; + case 'q': + test_global->opt.num_pktio_queue = atoi(optarg); + break; case 's': test_global->opt.collect_stat = 1; break; @@ -270,6 +274,9 @@ static int parse_options(int argc, char *argv[], test_global_t *test_global) } } + if (test_global->opt.num_pktio_queue == 0) + test_global->opt.num_pktio_queue = test_global->opt.num_worker; + return ret; } @@ -323,12 +330,6 @@ static int config_setup(test_global_t *test_global) test_global->pkt_len = pkt_len; test_global->pkt_num = pkt_num; - if (test_global->num_input_queues == 0) - test_global->num_input_queues = test_global->opt.num_worker; - - if (test_global->num_output_queues == 0) - test_global->num_output_queues = test_global->opt.num_worker; - return 0; } @@ -361,9 +362,8 @@ static void print_config(test_global_t *test_global) printf(" %s", test_global->pktio[i].name); printf("\n" - " num input queues: %i\n" - " num output queues: %i\n", - test_global->num_input_queues, test_global->num_output_queues); + " queues per interface: %i\n", + test_global->opt.num_pktio_queue); printf(" collect statistics: %u\n", test_global->opt.collect_stat); @@ -417,13 +417,12 @@ static int open_pktios(test_global_t *test_global) odp_pktin_queue_param_t pktin_param; odp_pktout_queue_param_t pktout_param; odp_schedule_sync_t sched_sync; - unsigned int num_input, num_output; + unsigned int num_queue; char *name; int i, num_pktio, ret; - num_pktio = test_global->opt.num_pktio; - num_input = test_global->num_input_queues; - num_output = test_global->num_output_queues; + num_pktio = test_global->opt.num_pktio; + num_queue = test_global->opt.num_pktio_queue; odp_pool_param_init(&pool_param); pool_param.pkt.seg_len = MIN_PKT_SEG_LEN; @@ -477,15 +476,15 @@ static int open_pktios(test_global_t *test_global) return -1; } - if (num_input > pktio_capa.max_input_queues) { + if (num_queue > pktio_capa.max_input_queues) { printf("Error (%s): Too many input queues: %u\n", - name, num_input); + name, num_queue); return -1; } - if (num_output > pktio_capa.max_output_queues) { + if (num_queue > pktio_capa.max_output_queues) { printf("Error (%s): Too many output queues: %u\n", - name, num_output); + name, num_queue); return -1; } @@ -500,12 +499,12 @@ static int open_pktios(test_global_t *test_global) pktin_param.queue_param.sched.sync = sched_sync; pktin_param.queue_param.sched.group = ODP_SCHED_GROUP_ALL; - if (num_input > 1) { + if (num_queue > 1) { pktin_param.hash_enable = 1; pktin_param.hash_proto.proto.ipv4_udp = 1; } - pktin_param.num_queues = num_input; + pktin_param.num_queues = num_queue; if (odp_pktin_queue_config(pktio, &pktin_param)) { printf("Error (%s): Pktin config failed.\n", name); @@ -514,13 +513,13 @@ static int open_pktios(test_global_t *test_global) if (odp_pktin_event_queue(pktio, test_global->pktio[i].input_queue, - num_input) != (int)num_input) { + num_queue) != (int)num_queue) { printf("Error (%s): Input queue query failed.\n", name); return -1; } odp_pktout_queue_param_init(&pktout_param); - pktout_param.num_queues = num_output; + pktout_param.num_queues = num_queue; pktout_param.op_mode = ODP_PKTIO_OP_MT_UNSAFE; if (odp_pktout_queue_config(pktio, &pktout_param)) { @@ -530,7 +529,7 @@ static int open_pktios(test_global_t *test_global) if (odp_pktout_queue(pktio, test_global->pktio[i].pktout, - num_output) != (int)num_output) { + num_queue) != (int)num_queue) { printf("Error (%s): Output queue query failed.\n", name); return -1; commit 27a84cdfc0dfb82e3b86067e473658c091377892 Author: Petri Savolainen <petri.savolainen(a)linaro.org> Date: Fri Apr 27 12:50:52 2018 +0300 linux-gen: timer: set and check highest resolution When using inline_timers, highest resolution was left to 0 ns and was checked to be non-zero. In practice, few nanosecond resolution is not needed or practical. Limit highest supported resolution to 500 ns. Signed-off-by: Petri Savolainen <petri.savolainen(a)linaro.org> Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org> Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org> diff --git a/platform/linux-generic/odp_timer.c b/platform/linux-generic/odp_timer.c index a445a57b..37d0304c 100644 --- a/platform/linux-generic/odp_timer.c +++ b/platform/linux-generic/odp_timer.c @@ -76,7 +76,7 @@ static _odp_atomic_flag_t locks[NUM_LOCKS]; /* Multiple locks per cache line! */ #endif /* Max timer resolution in nanoseconds */ -static uint64_t highest_res_ns; +static uint64_t highest_res_ns = 500; static uint64_t min_res_ns = INT64_MAX; /****************************************************************************** @@ -1077,13 +1077,10 @@ int odp_timer_capability(odp_timer_clk_src_t clk_src, return ret; } -odp_timer_pool_t -odp_timer_pool_create(const char *name, - const odp_timer_pool_param_t *param) +odp_timer_pool_t odp_timer_pool_create(const char *name, + const odp_timer_pool_param_t *param) { - /* Verify that buffer pool can be used for timeouts */ - /* Verify that we have a valid (non-zero) timer resolution */ - if (param->res_ns == 0) { + if (param->res_ns < highest_res_ns) { __odp_errno = EINVAL; return ODP_TIMER_POOL_INVALID; } ----------------------------------------------------------------------- Summary of changes: platform/linux-generic/odp_timer.c | 11 +- test/performance/odp_sched_pktio.c | 510 +++++++++++++++++++++++++++++++------ 2 files changed, 440 insertions(+), 81 deletions(-) hooks/post-receive --

6 years, 9 months

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

git-lng-odp May 2018