X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fevent%2Fdsw%2Fdsw_evdev.h;h=08889a09909299ecf22acd16624664312c4a61d5;hb=300b796262a1;hp=a5399dda5e88ed687b6d937bd6bde73172b61349;hpb=2b7bc6a5b2b4c48990d9ef3e46d8645f94ee60b1;p=dpdk.git diff --git a/drivers/event/dsw/dsw_evdev.h b/drivers/event/dsw/dsw_evdev.h index a5399dda5e..08889a0990 100644 --- a/drivers/event/dsw/dsw_evdev.h +++ b/drivers/event/dsw/dsw_evdev.h @@ -10,7 +10,6 @@ #define DSW_PMD_NAME RTE_STR(event_dsw) -/* Code changes are required to allow more ports. */ #define DSW_MAX_PORTS (64) #define DSW_MAX_PORT_DEQUEUE_DEPTH (128) #define DSW_MAX_PORT_ENQUEUE_DEPTH (128) @@ -20,8 +19,20 @@ #define DSW_MAX_EVENTS (16384) -/* Code changes are required to allow more flows than 32k. */ -#define DSW_MAX_FLOWS_BITS (15) +/* Multiple 24-bit flow ids will map to the same DSW-level flow. The + * number of DSW flows should be high enough make it unlikely that + * flow ids of several large flows hash to the same DSW-level flow. + * Such collisions will limit parallism and thus the number of cores + * that may be utilized. However, configuring a large number of DSW + * flows might potentially, depending on traffic and actual + * application flow id value range, result in each such DSW-level flow + * being very small. The effect of migrating such flows will be small, + * in terms amount of processing load redistributed. This will in turn + * reduce the load balancing speed, since flow migration rate has an + * upper limit. Code changes are required to allow > 32k DSW-level + * flows. + */ +#define DSW_MAX_FLOWS_BITS (13) #define DSW_MAX_FLOWS (1<<(DSW_MAX_FLOWS_BITS)) #define DSW_MAX_FLOWS_MASK (DSW_MAX_FLOWS-1) @@ -73,7 +84,51 @@ #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_REBALANCE_THRESHOLD (DSW_LOAD_FROM_PERCENT(3)) + +#define DSW_MAX_EVENTS_RECORDED (128) + +#define DSW_MAX_FLOWS_PER_MIGRATION (8) + +/* Only one outstanding migration per port is allowed */ +#define DSW_MAX_PAUSED_FLOWS (DSW_MAX_PORTS*DSW_MAX_FLOWS_PER_MIGRATION) + +/* 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; @@ -98,6 +153,7 @@ struct dsw_port { uint16_t ops_since_bg_task; + /* most recent 'background' processing */ uint64_t last_bg; /* For port load measurement. */ @@ -108,13 +164,64 @@ struct dsw_port { uint64_t busy_cycles; uint64_t total_busy_cycles; + /* For the ctl interface and flow migration mechanism. */ + uint64_t next_emigration; + uint64_t migration_interval; + enum dsw_migration_state migration_state; + + uint64_t emigration_start; + uint64_t emigrations; + uint64_t emigration_latency; + + uint8_t emigration_target_port_ids[DSW_MAX_FLOWS_PER_MIGRATION]; + struct dsw_queue_flow + emigration_target_qfs[DSW_MAX_FLOWS_PER_MIGRATION]; + uint8_t emigration_targets_len; + uint8_t cfm_cnt; + + uint64_t immigrations; + + 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 enqueue_calls; + uint64_t new_enqueued; + uint64_t forward_enqueued; + uint64_t release_enqueued; + uint64_t queue_enqueued[DSW_MAX_QUEUES]; + + uint64_t dequeue_calls; + 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; + int16_t load __rte_cache_aligned; + /* Estimate of flows currently migrating to this port. */ + int32_t immigration_load __rte_cache_aligned; } __rte_cache_aligned; struct dsw_queue { @@ -134,9 +241,20 @@ struct dsw_evdev { uint8_t num_queues; int32_t max_inflight; - rte_atomic32_t credits_on_loan __rte_cache_aligned; + int32_t credits_on_loan __rte_cache_aligned; }; +#define DSW_CTL_PAUS_REQ (0) +#define DSW_CTL_UNPAUS_REQ (1) +#define DSW_CTL_CFM (2) + +struct dsw_ctl_msg { + uint8_t type; + uint8_t originating_port_id; + uint8_t qfs_len; + struct dsw_queue_flow qfs[DSW_MAX_FLOWS_PER_MIGRATION]; +} __rte_aligned(4); + 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[], @@ -152,6 +270,17 @@ 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) {