From patchwork Thu Jun 21 14:23:22 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Eads, Gage" X-Patchwork-Id: 41370 X-Patchwork-Delegate: jerinj@marvell.com Return-Path: X-Original-To: patchwork@dpdk.org Delivered-To: patchwork@dpdk.org Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 373891BB65; Thu, 21 Jun 2018 16:23:35 +0200 (CEST) Received: from mga09.intel.com (mga09.intel.com [134.134.136.24]) by dpdk.org (Postfix) with ESMTP id 88CC41B5F1 for ; Thu, 21 Jun 2018 16:23:31 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga004.fm.intel.com ([10.253.24.48]) by orsmga102.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 21 Jun 2018 07:23:31 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.51,252,1526367600"; d="scan'208";a="65140508" Received: from txasoft-yocto.an.intel.com ([10.123.72.192]) by fmsmga004.fm.intel.com with ESMTP; 21 Jun 2018 07:23:30 -0700 From: Gage Eads To: dev@dpdk.org Cc: jerin.jacob@caviumnetworks.com, harry.van.haaren@intel.com, bruce.richardson@intel.com, nikhil.rao@intel.com, erik.g.carrillo@intel.com, abhinandan.gujjar@intel.com, thomas@monjalon.net Date: Thu, 21 Jun 2018 09:23:22 -0500 Message-Id: <20180621142323.17598-2-gage.eads@intel.com> X-Mailer: git-send-email 2.13.6 In-Reply-To: <20180621142323.17598-1-gage.eads@intel.com> References: <20180614135130.30849-1-gage.eads@intel.com> <20180621142323.17598-1-gage.eads@intel.com> Subject: [dpdk-dev] [PATCH v3 1/2] service: add mechanism for quiescing a service 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" Existing service functions allow us to stop a service, but doing so doesn't guarantee that the service has finished running on a service core. This commit introduces rte_service_may_be_active(), which returns whether the service may be executing on one or more lcores currently, or definitely is not. The service core layer supports this function by setting a flag when a service core is going to execute a service, and unsetting the flag when the core is no longer able to run the service (its runstate becomes stopped or the lcore is no longer mapped). With this new function, applications can set a service's runstate to stopped, then poll rte_service_may_be_active() until it returns false. At that point, the service is quiesced. Signed-off-by: Gage Eads Acked-by: Harry van Haaren --- lib/librte_eal/common/include/rte_service.h | 20 ++++++++++++++ lib/librte_eal/common/rte_service.c | 32 ++++++++++++++++++--- lib/librte_eal/rte_eal_version.map | 1 + test/test/test_service_cores.c | 43 +++++++++++++++++++++++++++++ 4 files changed, 92 insertions(+), 4 deletions(-) diff --git a/lib/librte_eal/common/include/rte_service.h b/lib/librte_eal/common/include/rte_service.h index aea4d91b9..20f713b19 100644 --- a/lib/librte_eal/common/include/rte_service.h +++ b/lib/librte_eal/common/include/rte_service.h @@ -162,6 +162,26 @@ int32_t rte_service_runstate_set(uint32_t id, uint32_t runstate); int32_t rte_service_runstate_get(uint32_t id); /** + * @warning + * @b EXPERIMENTAL: this API may change, or be removed, without prior notice + * + * This function returns whether the service may be currently executing on + * at least one lcore, or definitely is not. This function can be used to + * determine if, after setting the service runstate to stopped, the service + * is still executing a service lcore. + * + * Care must be taken if calling this function when the service runstate is + * running, since the result of this function may be incorrect by the time the + * function returns due to service cores running in parallel. + * + * @retval 1 Service may be running on one or more lcores + * @retval 0 Service is not running on any lcore + * @retval -EINVAL Invalid service id + */ +int32_t __rte_experimental +rte_service_may_be_active(uint32_t id); + +/** * Enable or disable the check for a service-core being mapped to the service. * An application can disable the check when takes the responsibility to run a * service itself using *rte_service_run_iter_on_app_lcore*. diff --git a/lib/librte_eal/common/rte_service.c b/lib/librte_eal/common/rte_service.c index 73507aacb..af9a660d4 100644 --- a/lib/librte_eal/common/rte_service.c +++ b/lib/librte_eal/common/rte_service.c @@ -52,6 +52,7 @@ struct rte_service_spec_impl { rte_atomic32_t num_mapped_cores; uint64_t calls; uint64_t cycles_spent; + uint8_t active_on_lcore[RTE_MAX_LCORE]; } __rte_cache_aligned; /* the internal values of a service core */ @@ -347,15 +348,19 @@ rte_service_runner_do_callback(struct rte_service_spec_impl *s, static inline int32_t -service_run(uint32_t i, struct core_state *cs, uint64_t service_mask) +service_run(uint32_t i, int lcore, struct core_state *cs, uint64_t service_mask) { if (!service_valid(i)) return -EINVAL; struct rte_service_spec_impl *s = &rte_services[i]; if (s->comp_runstate != RUNSTATE_RUNNING || s->app_runstate != RUNSTATE_RUNNING || - !(service_mask & (UINT64_C(1) << i))) + !(service_mask & (UINT64_C(1) << i))) { + s->active_on_lcore[lcore] = 0; return -ENOEXEC; + } + + s->active_on_lcore[lcore] = 1; /* check do we need cmpset, if MT safe or <= 1 core * mapped, atomic ops are not required. @@ -374,6 +379,25 @@ service_run(uint32_t i, struct core_state *cs, uint64_t service_mask) return 0; } +int32_t __rte_experimental +rte_service_may_be_active(uint32_t id) +{ + uint32_t ids[RTE_MAX_LCORE] = {0}; + struct rte_service_spec_impl *s = &rte_services[id]; + int32_t lcore_count = rte_service_lcore_list(ids, RTE_MAX_LCORE); + int i; + + if (!service_valid(id)) + return -EINVAL; + + for (i = 0; i < lcore_count; i++) { + if (s->active_on_lcore[ids[i]]) + return 1; + } + + return 0; +} + int32_t rte_service_run_iter_on_app_lcore(uint32_t id, uint32_t serialize_mt_unsafe) { @@ -398,7 +422,7 @@ int32_t rte_service_run_iter_on_app_lcore(uint32_t id, return -EBUSY; } - int ret = service_run(id, cs, UINT64_MAX); + int ret = service_run(id, rte_lcore_id(), cs, UINT64_MAX); if (serialize_mt_unsafe) rte_atomic32_dec(&s->num_mapped_cores); @@ -419,7 +443,7 @@ rte_service_runner_func(void *arg) for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) { /* return value ignored as no change to code flow */ - service_run(i, cs, service_mask); + service_run(i, lcore, cs, service_mask); } rte_smp_rmb(); diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map index f7dd0e7bc..7a9ccaa0a 100644 --- a/lib/librte_eal/rte_eal_version.map +++ b/lib/librte_eal/rte_eal_version.map @@ -294,6 +294,7 @@ EXPERIMENTAL { rte_mp_request_sync; rte_mp_request_async; rte_mp_sendmsg; + rte_service_may_be_active; rte_socket_count; rte_socket_id_by_idx; rte_vfio_dma_map; diff --git a/test/test/test_service_cores.c b/test/test/test_service_cores.c index 86b4073d2..e6973d642 100644 --- a/test/test/test_service_cores.c +++ b/test/test/test_service_cores.c @@ -771,6 +771,48 @@ service_lcore_start_stop(void) return unregister_all(); } +/* stop a service and wait for it to become inactive */ +static int +service_may_be_active(void) +{ + const uint32_t sid = 0; + int i; + + /* expected failure cases */ + TEST_ASSERT_EQUAL(-EINVAL, rte_service_may_be_active(10000), + "Invalid service may be active check did not fail"); + + /* start the service */ + TEST_ASSERT_EQUAL(0, rte_service_runstate_set(sid, 1), + "Starting valid service failed"); + TEST_ASSERT_EQUAL(0, rte_service_lcore_add(slcore_id), + "Add service core failed when not in use before"); + TEST_ASSERT_EQUAL(0, rte_service_map_lcore_set(sid, slcore_id, 1), + "Enabling valid service on valid core failed"); + TEST_ASSERT_EQUAL(0, rte_service_lcore_start(slcore_id), + "Service core start after add failed"); + + /* ensures core really is running the service function */ + TEST_ASSERT_EQUAL(1, service_lcore_running_check(), + "Service core expected to poll service but it didn't"); + + /* stop the service */ + TEST_ASSERT_EQUAL(0, rte_service_runstate_set(sid, 0), + "Error: Service stop returned non-zero"); + + /* give the service 100ms to stop running */ + for (i = 0; i < 100; i++) { + if (!rte_service_may_be_active(sid)) + break; + rte_delay_ms(SERVICE_DELAY); + } + + TEST_ASSERT_EQUAL(0, rte_service_may_be_active(sid), + "Error: Service not stopped after 100ms"); + + return unregister_all(); +} + static struct unit_test_suite service_tests = { .suite_name = "service core test suite", .setup = testsuite_setup, @@ -790,6 +832,7 @@ static struct unit_test_suite service_tests = { TEST_CASE_ST(dummy_register, NULL, service_mt_safe_poll), TEST_CASE_ST(dummy_register, NULL, service_app_lcore_mt_safe), TEST_CASE_ST(dummy_register, NULL, service_app_lcore_mt_unsafe), + TEST_CASE_ST(dummy_register, NULL, service_may_be_active), TEST_CASES_END() /**< NULL terminate unit test array */ } }; From patchwork Thu Jun 21 14:23:23 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Eads, Gage" X-Patchwork-Id: 41371 X-Patchwork-Delegate: jerinj@marvell.com Return-Path: X-Original-To: patchwork@dpdk.org Delivered-To: patchwork@dpdk.org Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 9D6EE1BB83; Thu, 21 Jun 2018 16:23:37 +0200 (CEST) Received: from mga09.intel.com (mga09.intel.com [134.134.136.24]) by dpdk.org (Postfix) with ESMTP id 404651BB4E for ; Thu, 21 Jun 2018 16:23:33 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga004.fm.intel.com ([10.253.24.48]) by orsmga102.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 21 Jun 2018 07:23:32 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.51,252,1526367600"; d="scan'208";a="65140523" Received: from txasoft-yocto.an.intel.com ([10.123.72.192]) by fmsmga004.fm.intel.com with ESMTP; 21 Jun 2018 07:23:31 -0700 From: Gage Eads To: dev@dpdk.org Cc: jerin.jacob@caviumnetworks.com, harry.van.haaren@intel.com, bruce.richardson@intel.com, nikhil.rao@intel.com, erik.g.carrillo@intel.com, abhinandan.gujjar@intel.com, thomas@monjalon.net Date: Thu, 21 Jun 2018 09:23:23 -0500 Message-Id: <20180621142323.17598-3-gage.eads@intel.com> X-Mailer: git-send-email 2.13.6 In-Reply-To: <20180621142323.17598-1-gage.eads@intel.com> References: <20180614135130.30849-1-gage.eads@intel.com> <20180621142323.17598-1-gage.eads@intel.com> Subject: [dpdk-dev] [PATCH v3 2/2] event/sw: support device stop flush callback 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" This commit also adds a flush callback test to the sw eventdev's selftest suite. Signed-off-by: Gage Eads Acked-by: Harry van Haaren --- drivers/event/sw/sw_evdev.c | 114 ++++++++++++++++++++++++++++++++++- drivers/event/sw/sw_evdev_selftest.c | 81 ++++++++++++++++++++++++- 2 files changed, 192 insertions(+), 3 deletions(-) diff --git a/drivers/event/sw/sw_evdev.c b/drivers/event/sw/sw_evdev.c index 10f0e1ad4..331c518ff 100644 --- a/drivers/event/sw/sw_evdev.c +++ b/drivers/event/sw/sw_evdev.c @@ -361,9 +361,99 @@ sw_init_qid_iqs(struct sw_evdev *sw) } } +static int +sw_qids_empty(struct sw_evdev *sw) +{ + unsigned int i, j; + + for (i = 0; i < sw->qid_count; i++) { + for (j = 0; j < SW_IQS_MAX; j++) { + if (iq_count(&sw->qids[i].iq[j])) + return 0; + } + } + + return 1; +} + +static int +sw_ports_empty(struct sw_evdev *sw) +{ + unsigned int i; + + for (i = 0; i < sw->port_count; i++) { + if ((rte_event_ring_count(sw->ports[i].rx_worker_ring)) || + rte_event_ring_count(sw->ports[i].cq_worker_ring)) + return 0; + } + + return 1; +} + +static void +sw_drain_ports(struct rte_eventdev *dev) +{ + struct sw_evdev *sw = sw_pmd_priv(dev); + eventdev_stop_flush_t flush; + unsigned int i; + uint8_t dev_id; + void *arg; + + flush = dev->dev_ops->dev_stop_flush; + dev_id = dev->data->dev_id; + arg = dev->data->dev_stop_flush_arg; + + for (i = 0; i < sw->port_count; i++) { + struct rte_event ev; + + while (rte_event_dequeue_burst(dev_id, i, &ev, 1, 0)) { + if (flush) + flush(dev_id, ev, arg); + + ev.op = RTE_EVENT_OP_RELEASE; + rte_event_enqueue_burst(dev_id, i, &ev, 1); + } + } +} + +static void +sw_drain_queue(struct rte_eventdev *dev, struct sw_iq *iq) +{ + struct sw_evdev *sw = sw_pmd_priv(dev); + eventdev_stop_flush_t flush; + uint8_t dev_id; + void *arg; + + flush = dev->dev_ops->dev_stop_flush; + dev_id = dev->data->dev_id; + arg = dev->data->dev_stop_flush_arg; + + while (iq_count(iq) > 0) { + struct rte_event ev; + + iq_dequeue_burst(sw, iq, &ev, 1); + + if (flush) + flush(dev_id, ev, arg); + } +} + +static void +sw_drain_queues(struct rte_eventdev *dev) +{ + struct sw_evdev *sw = sw_pmd_priv(dev); + unsigned int i, j; + + for (i = 0; i < sw->qid_count; i++) { + for (j = 0; j < SW_IQS_MAX; j++) + sw_drain_queue(dev, &sw->qids[i].iq[j]); + } +} + static void -sw_clean_qid_iqs(struct sw_evdev *sw) +sw_clean_qid_iqs(struct rte_eventdev *dev) { + struct sw_evdev *sw = sw_pmd_priv(dev); int i, j; /* Release the IQ memory of all configured qids */ @@ -729,10 +819,30 @@ static void sw_stop(struct rte_eventdev *dev) { struct sw_evdev *sw = sw_pmd_priv(dev); - sw_clean_qid_iqs(sw); + int32_t runstate; + + /* Stop the scheduler if it's running */ + runstate = rte_service_runstate_get(sw->service_id); + if (runstate == 1) + rte_service_runstate_set(sw->service_id, 0); + + while (rte_service_may_be_active(sw->service_id)) + rte_pause(); + + /* Flush all events out of the device */ + while (!(sw_qids_empty(sw) && sw_ports_empty(sw))) { + sw_event_schedule(dev); + sw_drain_ports(dev); + sw_drain_queues(dev); + } + + sw_clean_qid_iqs(dev); sw_xstats_uninit(sw); sw->started = 0; rte_smp_wmb(); + + if (runstate == 1) + rte_service_runstate_set(sw->service_id, 1); } static int diff --git a/drivers/event/sw/sw_evdev_selftest.c b/drivers/event/sw/sw_evdev_selftest.c index 78d30e07a..c40912db5 100644 --- a/drivers/event/sw/sw_evdev_selftest.c +++ b/drivers/event/sw/sw_evdev_selftest.c @@ -28,6 +28,7 @@ #define MAX_PORTS 16 #define MAX_QIDS 16 #define NUM_PACKETS (1<<18) +#define DEQUEUE_DEPTH 128 static int evdev; @@ -147,7 +148,7 @@ init(struct test *t, int nb_queues, int nb_ports) .nb_event_ports = nb_ports, .nb_event_queue_flows = 1024, .nb_events_limit = 4096, - .nb_event_port_dequeue_depth = 128, + .nb_event_port_dequeue_depth = DEQUEUE_DEPTH, .nb_event_port_enqueue_depth = 128, }; int ret; @@ -2807,6 +2808,78 @@ holb(struct test *t) /* test to check we avoid basic head-of-line blocking */ return -1; } +static void +flush(uint8_t dev_id __rte_unused, struct rte_event event, void *arg) +{ + *((uint8_t *) arg) += (event.u64 == 0xCA11BACC) ? 1 : 0; +} + +static int +dev_stop_flush(struct test *t) /* test to check we can properly flush events */ +{ + const struct rte_event new_ev = { + .op = RTE_EVENT_OP_NEW, + .u64 = 0xCA11BACC, + .queue_id = 0 + }; + struct rte_event ev = new_ev; + uint8_t count = 0; + int i; + + if (init(t, 1, 1) < 0 || + create_ports(t, 1) < 0 || + create_atomic_qids(t, 1) < 0) { + printf("%d: Error initializing device\n", __LINE__); + return -1; + } + + /* Link the queue so *_start() doesn't error out */ + if (rte_event_port_link(evdev, t->port[0], NULL, NULL, 0) != 1) { + printf("%d: Error linking queue to port\n", __LINE__); + goto err; + } + + if (rte_event_dev_start(evdev) < 0) { + printf("%d: Error with start call\n", __LINE__); + goto err; + } + + for (i = 0; i < DEQUEUE_DEPTH + 1; i++) { + if (rte_event_enqueue_burst(evdev, t->port[0], &ev, 1) != 1) { + printf("%d: Error enqueuing events\n", __LINE__); + goto err; + } + } + + /* Schedule the events from the port to the IQ. At least one event + * should be remaining in the queue. + */ + rte_service_run_iter_on_app_lcore(t->service_id, 1); + + if (rte_event_dev_stop_flush_callback_register(evdev, flush, &count)) { + printf("%d: Error installing the flush callback\n", __LINE__); + goto err; + } + + cleanup(t); + + if (count == 0) { + printf("%d: Error executing the flush callback\n", __LINE__); + goto err; + } + + if (rte_event_dev_stop_flush_callback_register(evdev, NULL, NULL)) { + printf("%d: Error uninstalling the flush callback\n", __LINE__); + goto err; + } + + return 0; +err: + rte_event_dev_dump(evdev, stdout); + cleanup(t); + return -1; +} + static int worker_loopback_worker_fn(void *arg) { @@ -3211,6 +3284,12 @@ test_sw_eventdev(void) printf("ERROR - Head-of-line-blocking test FAILED.\n"); goto test_fail; } + printf("*** Running Stop Flush test...\n"); + ret = dev_stop_flush(t); + if (ret != 0) { + printf("ERROR - Stop Flush test FAILED.\n"); + goto test_fail; + } if (rte_lcore_count() >= 3) { printf("*** Running Worker loopback test...\n"); ret = worker_loopback(t, 0);