X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fevent%2Fdsw%2Fdsw_evdev.h;h=dc28ab125d8ed1d7c6142a3ac95e91c78c5f2bc1;hb=fb69492c042b96625fa8f5c4e49a326db22cab7c;hp=f8e94e4a4e67bd574c159c96aba9987ef4e5277c;hpb=1c8e3caa3bfba26f3b22ec52baf33224d08d690c;p=dpdk.git diff --git a/drivers/event/dsw/dsw_evdev.h b/drivers/event/dsw/dsw_evdev.h index f8e94e4a4e..dc28ab125d 100644 --- a/drivers/event/dsw/dsw_evdev.h +++ b/drivers/event/dsw/dsw_evdev.h @@ -36,6 +36,15 @@ */ #define DSW_PARALLEL_FLOWS (1024) +/* 'Background tasks' are polling the control rings for * + * migration-related messages, or flush the output buffer (so + * buffered events doesn't linger too long). Shouldn't be too low, + * since the system won't benefit from the 'batching' effects from + * the output buffer, and shouldn't be too high, since it will make + * buffered events linger too long in case the port goes idle. + */ +#define DSW_MAX_PORT_OPS_PER_BG_TASK (128) + /* Avoid making small 'loans' from the central in-flight event credit * pool, to improve efficiency. */ @@ -50,6 +59,63 @@ */ #define DSW_IN_RING_SIZE (DSW_MAX_EVENTS) +#define DSW_MAX_LOAD (INT16_MAX) +#define DSW_LOAD_FROM_PERCENT(x) ((int16_t)(((x)*DSW_MAX_LOAD)/100)) +#define DSW_LOAD_TO_PERCENT(x) ((100*x)/DSW_MAX_LOAD) + +/* The thought behind keeping the load update interval shorter than + * the migration interval is that the load from newly migrated flows + * should 'show up' on the load measurement before new migrations are + * considered. This is to avoid having too many flows, from too many + * source ports, to be migrated too quickly to a lightly loaded port - + * in particular since this might cause the system to oscillate. + */ +#define DSW_LOAD_UPDATE_INTERVAL (DSW_MIGRATION_INTERVAL/4) +#define DSW_OLD_LOAD_WEIGHT (1) + +/* The minimum time (in us) between two flow migrations. What puts an + * upper limit on the actual migration rate is primarily the pace in + * which the ports send and receive control messages, which in turn is + * largely a function of how much cycles are spent the processing of + * an event burst. + */ +#define DSW_MIGRATION_INTERVAL (1000) +#define DSW_MIN_SOURCE_LOAD_FOR_MIGRATION (DSW_LOAD_FROM_PERCENT(70)) +#define DSW_MAX_TARGET_LOAD_FOR_MIGRATION (DSW_LOAD_FROM_PERCENT(95)) + +#define DSW_MAX_EVENTS_RECORDED (128) + +/* Only one outstanding migration per port is allowed */ +#define DSW_MAX_PAUSED_FLOWS (DSW_MAX_PORTS) + +/* Enough room for paus request/confirm and unpaus request/confirm for + * all possible senders. + */ +#define DSW_CTL_IN_RING_SIZE ((DSW_MAX_PORTS-1)*4) + +/* With DSW_SORT_DEQUEUED enabled, the scheduler will, at the point of + * dequeue(), arrange events so that events with the same flow id on + * the same queue forms a back-to-back "burst", and also so that such + * bursts of different flow ids, but on the same queue, also come + * consecutively. All this in an attempt to improve data and + * instruction cache usage for the application, at the cost of a + * scheduler overhead increase. + */ + +/* #define DSW_SORT_DEQUEUED */ + +struct dsw_queue_flow { + uint8_t queue_id; + uint16_t flow_hash; +}; + +enum dsw_migration_state { + DSW_MIGRATION_STATE_IDLE, + DSW_MIGRATION_STATE_PAUSING, + DSW_MIGRATION_STATE_FORWARDING, + DSW_MIGRATION_STATE_UNPAUSING +}; + struct dsw_port { uint16_t id; @@ -71,10 +137,69 @@ struct dsw_port { uint16_t next_parallel_flow_id; + uint16_t ops_since_bg_task; + + /* most recent 'background' processing */ + uint64_t last_bg; + + /* For port load measurement. */ + uint64_t next_load_update; + uint64_t load_update_interval; + uint64_t measurement_start; + uint64_t busy_start; + uint64_t busy_cycles; + uint64_t total_busy_cycles; + + /* For the ctl interface and flow migration mechanism. */ + uint64_t next_migration; + uint64_t migration_interval; + enum dsw_migration_state migration_state; + + uint64_t migration_start; + uint64_t migrations; + uint64_t migration_latency; + + uint8_t migration_target_port_id; + struct dsw_queue_flow migration_target_qf; + uint8_t cfm_cnt; + + uint16_t paused_flows_len; + struct dsw_queue_flow paused_flows[DSW_MAX_PAUSED_FLOWS]; + + /* In a very contrived worst case all inflight events can be + * laying around paused here. + */ + uint16_t paused_events_len; + struct rte_event paused_events[DSW_MAX_EVENTS]; + + uint16_t seen_events_len; + uint16_t seen_events_idx; + struct dsw_queue_flow seen_events[DSW_MAX_EVENTS_RECORDED]; + + uint64_t new_enqueued; + uint64_t forward_enqueued; + uint64_t release_enqueued; + uint64_t queue_enqueued[DSW_MAX_QUEUES]; + + uint64_t dequeued; + uint64_t queue_dequeued[DSW_MAX_QUEUES]; + uint16_t out_buffer_len[DSW_MAX_PORTS]; struct rte_event out_buffer[DSW_MAX_PORTS][DSW_MAX_PORT_OUT_BUFFER]; + uint16_t in_buffer_len; + uint16_t in_buffer_start; + /* This buffer may contain events that were read up from the + * in_ring during the flow migration process. + */ + struct rte_event in_buffer[DSW_MAX_EVENTS]; + struct rte_event_ring *in_ring __rte_cache_aligned; + + struct rte_ring *ctl_in_ring __rte_cache_aligned; + + /* Estimate of current port load. */ + rte_atomic16_t load __rte_cache_aligned; } __rte_cache_aligned; struct dsw_queue { @@ -97,6 +222,20 @@ struct dsw_evdev { rte_atomic32_t credits_on_loan __rte_cache_aligned; }; +#define DSW_CTL_PAUS_REQ (0) +#define DSW_CTL_UNPAUS_REQ (1) +#define DSW_CTL_CFM (2) + +/* sizeof(struct dsw_ctl_msg) must be equal or less than + * sizeof(void *), to fit on the control ring. + */ +struct dsw_ctl_msg { + uint8_t type:2; + uint8_t originating_port_id:6; + uint8_t queue_id; + uint16_t flow_hash; +} __rte_packed; + uint16_t dsw_event_enqueue(void *port, const struct rte_event *event); uint16_t dsw_event_enqueue_burst(void *port, const struct rte_event events[], @@ -112,6 +251,17 @@ uint16_t dsw_event_dequeue(void *port, struct rte_event *ev, uint64_t wait); uint16_t dsw_event_dequeue_burst(void *port, struct rte_event *events, uint16_t num, uint64_t wait); +int dsw_xstats_get_names(const struct rte_eventdev *dev, + enum rte_event_dev_xstats_mode mode, + uint8_t queue_port_id, + struct rte_event_dev_xstats_name *xstats_names, + unsigned int *ids, unsigned int size); +int dsw_xstats_get(const struct rte_eventdev *dev, + enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id, + const unsigned int ids[], uint64_t values[], unsigned int n); +uint64_t dsw_xstats_get_by_name(const struct rte_eventdev *dev, + const char *name, unsigned int *id); + static inline struct dsw_evdev * dsw_pmd_priv(const struct rte_eventdev *eventdev) {