1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2020 Intel Corporation
11 #include <rte_eventdev.h>
12 #include <rte_config.h>
13 #include "dlb2_user.h"
15 #include "rte_pmd_dlb2.h"
17 #ifndef RTE_LIBRTE_PMD_DLB2_QUELL_STATS
18 #define DLB2_INC_STAT(_stat, _incr_val) ((_stat) += _incr_val)
20 #define DLB2_INC_STAT(_stat, _incr_val)
23 #define EVDEV_DLB2_NAME_PMD dlb2_event
25 /* Default values for command line devargs */
26 #define DLB2_POLL_INTERVAL_DEFAULT 1000
27 #define DLB2_SW_CREDIT_QUANTA_DEFAULT 32
28 #define DLB2_DEPTH_THRESH_DEFAULT 256
30 /* command line arg strings */
31 #define NUMA_NODE_ARG "numa_node"
32 #define DLB2_MAX_NUM_EVENTS "max_num_events"
33 #define DLB2_NUM_DIR_CREDITS "num_dir_credits"
34 #define DEV_ID_ARG "dev_id"
35 #define DLB2_DEFER_SCHED_ARG "defer_sched"
36 #define DLB2_QID_DEPTH_THRESH_ARG "qid_depth_thresh"
37 #define DLB2_COS_ARG "cos"
38 #define DLB2_POLL_INTERVAL_ARG "poll_interval"
39 #define DLB2_SW_CREDIT_QUANTA_ARG "sw_credit_quanta"
40 #define DLB2_DEPTH_THRESH_ARG "default_depth_thresh"
41 #define DLB2_VECTOR_OPTS_DISAB_ARG "vector_opts_disable"
43 /* Begin HW related defines and structs */
46 #define DLB2_HW_V2_5 1
47 #define DLB2_MAX_NUM_DOMAINS 32
48 #define DLB2_MAX_NUM_VFS 16
49 #define DLB2_MAX_NUM_LDB_QUEUES 32
50 #define DLB2_MAX_NUM_LDB_PORTS 64
51 #define DLB2_MAX_NUM_DIR_PORTS_V2 DLB2_MAX_NUM_DIR_QUEUES_V2
52 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
53 #define DLB2_MAX_NUM_DIR_PORTS(ver) (ver == DLB2_HW_V2 ? \
54 DLB2_MAX_NUM_DIR_PORTS_V2 : \
55 DLB2_MAX_NUM_DIR_PORTS_V2_5)
56 #define DLB2_MAX_NUM_DIR_QUEUES_V2 64 /* DIR == directed */
57 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
58 /* When needed for array sizing, the DLB 2.5 macro is used */
59 #define DLB2_MAX_NUM_DIR_QUEUES(ver) (ver == DLB2_HW_V2 ? \
60 DLB2_MAX_NUM_DIR_QUEUES_V2 : \
61 DLB2_MAX_NUM_DIR_QUEUES_V2_5)
62 #define DLB2_MAX_NUM_FLOWS (64 * 1024)
63 #define DLB2_MAX_NUM_LDB_CREDITS (8 * 1024)
64 #define DLB2_MAX_NUM_DIR_CREDITS(ver) (ver == DLB2_HW_V2 ? 4096 : 0)
65 #define DLB2_MAX_NUM_CREDITS(ver) (ver == DLB2_HW_V2 ? \
66 0 : DLB2_MAX_NUM_LDB_CREDITS)
67 #define DLB2_MAX_NUM_LDB_CREDIT_POOLS 64
68 #define DLB2_MAX_NUM_DIR_CREDIT_POOLS 64
69 #define DLB2_MAX_NUM_HIST_LIST_ENTRIES 2048
70 #define DLB2_MAX_NUM_QIDS_PER_LDB_CQ 8
71 #define DLB2_QID_PRIORITIES 8
72 #define DLB2_MAX_DEVICE_PATH 32
73 #define DLB2_MIN_DEQUEUE_TIMEOUT_NS 1
74 /* Note: "- 1" here to support the timeout range check in eventdev_autotest */
75 #define DLB2_MAX_DEQUEUE_TIMEOUT_NS (UINT32_MAX - 1)
76 #define DLB2_SW_CREDIT_BATCH_SZ 32
77 #define DLB2_NUM_SN_GROUPS 2
78 #define DLB2_MAX_LDB_SN_ALLOC 1024
79 #define DLB2_MAX_QUEUE_DEPTH_THRESHOLD 8191
81 /* 2048 total hist list entries and 64 total ldb ports, which
82 * makes for 2048/64 == 32 hist list entries per port. However, CQ
83 * depth must be a power of 2 and must also be >= HIST LIST entries.
84 * As a result we just limit the maximum dequeue depth to 32.
86 #define DLB2_MIN_CQ_DEPTH 1
87 #define DLB2_MAX_CQ_DEPTH 32
88 #define DLB2_MIN_HARDWARE_CQ_DEPTH 8
89 #define DLB2_NUM_HIST_LIST_ENTRIES_PER_LDB_PORT \
92 #define DLB2_HW_DEVICE_FROM_PCI_ID(_pdev) \
93 (((_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_PF) || \
94 (_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_VF)) ? \
95 DLB2_HW_V2_5 : DLB2_HW_V2)
98 * Static per queue/port provisioning values
100 #define DLB2_NUM_ATOMIC_INFLIGHTS_PER_QUEUE 64
102 #define CQ_BASE(is_dir) ((is_dir) ? DLB2_DIR_CQ_BASE : DLB2_LDB_CQ_BASE)
103 #define CQ_SIZE(is_dir) ((is_dir) ? DLB2_DIR_CQ_MAX_SIZE : \
104 DLB2_LDB_CQ_MAX_SIZE)
105 #define PP_BASE(is_dir) ((is_dir) ? DLB2_DIR_PP_BASE : DLB2_LDB_PP_BASE)
107 #define DLB2_NUM_QES_PER_CACHE_LINE 4
109 #define DLB2_MAX_ENQUEUE_DEPTH 64
110 #define DLB2_MIN_ENQUEUE_DEPTH 4
112 #define DLB2_NAME_SIZE 64
115 #define DLB2_2K (2 * DLB2_1K)
116 #define DLB2_4K (4 * DLB2_1K)
117 #define DLB2_16K (16 * DLB2_1K)
118 #define DLB2_32K (32 * DLB2_1K)
119 #define DLB2_1MB (DLB2_1K * DLB2_1K)
120 #define DLB2_16MB (16 * DLB2_1MB)
122 /* Use the upper 3 bits of the event priority to select the DLB2 priority */
123 #define EV_TO_DLB2_PRIO(x) ((x) >> 5)
124 #define DLB2_TO_EV_PRIO(x) ((x) << 5)
131 enum dlb2_hw_port_types {
134 DLB2_NUM_PORT_TYPES /* Must be last */
137 enum dlb2_hw_queue_types {
140 DLB2_NUM_QUEUE_TYPES /* Must be last */
143 #define DLB2_COMBINED_POOL DLB2_LDB_QUEUE
145 #define PORT_TYPE(p) ((p)->is_directed ? DLB2_DIR_PORT : DLB2_LDB_PORT)
147 /* Do not change - must match hardware! */
148 enum dlb2_hw_sched_type {
149 DLB2_SCHED_ATOMIC = 0,
150 DLB2_SCHED_UNORDERED,
153 /* DLB2_NUM_HW_SCHED_TYPES must be last */
154 DLB2_NUM_HW_SCHED_TYPES
157 struct dlb2_hw_rsrcs {
158 int32_t nb_events_limit;
159 uint32_t num_queues; /* Total queues (lb + dir) */
160 uint32_t num_ldb_queues; /* Number of available ldb queues */
161 uint32_t num_ldb_ports; /* Number of load balanced ports */
162 uint32_t num_dir_ports; /* Number of directed ports */
165 uint32_t num_ldb_credits; /* Number of ldb credits */
166 uint32_t num_dir_credits; /* Number of dir credits */
169 uint32_t num_credits; /* Number of combined credits */
172 uint32_t reorder_window_size; /* Size of reorder window */
175 struct dlb2_hw_resource_info {
176 /**> Max resources that can be provided */
177 struct dlb2_hw_rsrcs hw_rsrc_max;
178 int num_sched_domains;
182 enum dlb2_enqueue_type {
184 * New : Used to inject a new packet into the QM.
188 * Forward : Enqueues a packet, and
189 * - if atomic: release any lock it holds in the QM
190 * - if ordered: release the packet for egress re-ordering
194 * Enqueue Drop : Release an inflight packet. Must be called with
195 * event == NULL. Used to drop a packet.
197 * Note that all packets dequeued from a load-balanced port must be
198 * released, either with DLB2_ENQ_DROP or DLB2_ENQ_FWD.
202 /* marker for array sizing etc. */
206 /* hw-specific format - do not change */
208 struct dlb2_event_type {
214 union dlb2_opaque_data {
215 uint16_t opaque_data;
216 struct dlb2_event_type event_type;
219 struct dlb2_msg_info {
221 uint8_t sched_type:2;
226 #define DLB2_NEW_CMD_BYTE 0x08
227 #define DLB2_FWD_CMD_BYTE 0x0A
228 #define DLB2_COMP_CMD_BYTE 0x02
229 #define DLB2_POP_CMD_BYTE 0x01
230 #define DLB2_NOOP_CMD_BYTE 0x00
232 /* hw-specific format - do not change */
233 struct dlb2_enqueue_qe {
236 union dlb2_opaque_data u;
238 uint8_t sched_type:2;
261 /* hw-specific format - do not change */
262 struct dlb2_cq_pop_qe {
264 union dlb2_opaque_data u;
266 uint8_t sched_type:2;
289 /* hw-specific format - do not change */
290 struct dlb2_dequeue_qe {
292 union dlb2_opaque_data u;
294 uint8_t sched_type:2;
297 uint16_t flow_id:16; /* was pp_id in v1 */
300 uint8_t qid_depth:2; /* 2 bits in v2 */
306 union dlb2_port_config {
307 struct dlb2_create_ldb_port_args ldb;
308 struct dlb2_create_dir_port_args dir;
311 enum dlb2_port_state {
317 enum dlb2_configuration_state {
318 /* The resource has not been configured */
320 /* The resource was configured, but the device was stopped */
321 DLB2_PREV_CONFIGURED,
322 /* The resource is currently configured */
330 uint16_t dir_credits;
331 uint32_t dequeue_depth;
332 enum dlb2_token_pop_mode token_pop_mode;
333 union dlb2_port_config cfg;
334 uint32_t *credit_pool[DLB2_NUM_QUEUE_TYPES]; /* use __atomic builtins */
337 uint16_t cached_ldb_credits;
338 uint16_t ldb_credits;
339 uint16_t cached_dir_credits;
342 uint16_t cached_credits;
347 uint16_t owed_tokens;
348 int16_t issued_releases;
349 int16_t token_pop_thresh;
352 uint16_t cq_idx_unmasked;
353 uint16_t cq_depth_mask;
354 uint16_t gen_bit_shift;
355 uint64_t cq_rolling_mask; /*
356 * rotate to always have right expected
359 uint64_t cq_rolling_mask_2;
360 void *cq_addr_cached; /* avoid multiple refs */
361 enum dlb2_port_state state;
362 enum dlb2_configuration_state config_state;
364 uint8_t *qid_mappings;
365 struct dlb2_enqueue_qe *qe4; /* Cache line's worth of QEs (4) */
366 struct dlb2_enqueue_qe *int_arm_qe;
367 struct dlb2_cq_pop_qe *consume_qe;
368 struct dlb2_eventdev *dlb2; /* back ptr */
369 struct dlb2_eventdev_port *ev_port; /* back ptr */
370 bool use_scalar; /* force usage of scalar code */
373 /* Per-process per-port mmio and memory pointers */
374 struct process_local_port_data {
376 struct dlb2_dequeue_qe *cq_base;
377 const struct rte_memzone *mz;
381 struct dlb2_eventdev;
383 struct dlb2_port_low_level_io_functions {
384 void (*pp_enqueue_four)(void *qe4, void *pp_addr);
392 uint32_t num_ldb_credits;
393 uint32_t num_dir_credits;
396 uint32_t num_credits;
399 struct dlb2_create_sched_domain_args resources;
403 DLB2_COS_DEFAULT = -1,
411 struct dlb2_config cfg;
412 struct dlb2_hw_resource_info info;
413 void *pf_dev; /* opaque pointer to PF PMD dev (struct dlb2_dev) */
415 enum dlb2_cos cos_id;
416 rte_spinlock_t resource_lock; /* for MP support */
417 } __rte_cache_aligned;
419 /* End HW related defines and structs */
421 /* Begin DLB2 PMD Eventdev related defines and structs */
423 #define DLB2_MAX_NUM_QUEUES(ver) \
424 (DLB2_MAX_NUM_DIR_QUEUES(ver) + DLB2_MAX_NUM_LDB_QUEUES)
426 #define DLB2_MAX_NUM_PORTS(ver) \
427 (DLB2_MAX_NUM_DIR_PORTS(ver) + DLB2_MAX_NUM_LDB_PORTS)
429 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
430 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
431 #define DLB2_MAX_NUM_QUEUES_ALL \
432 (DLB2_MAX_NUM_DIR_QUEUES_V2_5 + DLB2_MAX_NUM_LDB_QUEUES)
433 #define DLB2_MAX_NUM_PORTS_ALL \
434 (DLB2_MAX_NUM_DIR_PORTS_V2_5 + DLB2_MAX_NUM_LDB_PORTS)
435 #define DLB2_MAX_INPUT_QUEUE_DEPTH 256
437 /** Structure to hold the queue to port link establishment attributes */
439 struct dlb2_event_queue_link {
446 struct dlb2_traffic_stats {
449 uint64_t rx_interrupt_wait;
450 uint64_t rx_umonitor_umwait;
452 uint64_t total_polls;
456 uint64_t tx_nospc_ldb_hw_credits;
457 uint64_t tx_nospc_dir_hw_credits;
460 uint64_t tx_nospc_hw_credits;
463 uint64_t tx_nospc_inflight_max;
464 uint64_t tx_nospc_new_event_limit;
465 uint64_t tx_nospc_inflight_credits;
468 /* DLB2 HW sets the 2bit qid_depth in rx QEs based on the programmable depth
469 * threshold. The global default value in config/common_base (or rte_config.h)
470 * can be overridden on a per-qid basis using a vdev command line parameter.
471 * 3: depth > threshold
472 * 2: threshold >= depth > 3/4 threshold
473 * 1: 3/4 threshold >= depth > 1/2 threshold
474 * 0: depth <= 1/2 threshold.
476 #define DLB2_QID_DEPTH_LE50 0
477 #define DLB2_QID_DEPTH_GT50_LE75 1
478 #define DLB2_QID_DEPTH_GT75_LE100 2
479 #define DLB2_QID_DEPTH_GT100 3
480 #define DLB2_NUM_QID_DEPTH_STAT_VALS 4 /* 2 bits */
482 struct dlb2_queue_stats {
484 uint64_t qid_depth[DLB2_NUM_QID_DEPTH_STAT_VALS];
487 struct dlb2_port_stats {
488 struct dlb2_traffic_stats traffic;
489 uint64_t tx_op_cnt[4]; /* indexed by rte_event.op */
490 uint64_t tx_implicit_rel;
491 uint64_t tx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
493 uint64_t rx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
494 uint64_t rx_sched_invalid;
495 struct dlb2_queue_stats queue[DLB2_MAX_NUM_QUEUES_ALL];
498 struct dlb2_eventdev_port {
499 struct dlb2_port qm_port; /* hw specific data structure */
500 struct rte_event_port_conf conf; /* user-supplied configuration */
501 uint16_t inflight_credits; /* num credits this port has right now */
502 uint16_t credit_update_quanta;
503 struct dlb2_eventdev *dlb2; /* backlink optimization */
504 struct dlb2_port_stats stats __rte_cache_aligned;
505 struct dlb2_event_queue_link link[DLB2_MAX_NUM_QIDS_PER_LDB_CQ];
507 uint32_t id; /* port id */
508 /* num releases yet to be completed on this port.
509 * Only applies to load-balanced ports.
511 uint16_t outstanding_releases;
512 uint16_t inflight_max; /* app requested max inflights for this port */
513 /* setup_done is set when the event port is setup */
515 /* enq_configured is set when the qm port is created */
517 uint8_t implicit_release; /* release events before dequeueing */
518 } __rte_cache_aligned;
521 uint32_t num_qid_inflights; /* User config */
522 uint32_t num_atm_inflights; /* User config */
523 enum dlb2_configuration_state config_state;
524 int sched_type; /* LB queue only */
529 struct dlb2_eventdev_queue {
530 struct dlb2_queue qm_queue;
531 struct rte_event_queue_conf conf; /* User config */
532 int depth_threshold; /* use default if 0 */
538 enum dlb2_run_state {
539 DLB2_RUN_STATE_STOPPED = 0,
540 DLB2_RUN_STATE_STOPPING,
541 DLB2_RUN_STATE_STARTING,
542 DLB2_RUN_STATE_STARTED
545 struct dlb2_eventdev {
546 struct dlb2_eventdev_port ev_ports[DLB2_MAX_NUM_PORTS_ALL];
547 struct dlb2_eventdev_queue ev_queues[DLB2_MAX_NUM_QUEUES_ALL];
548 uint8_t qm_ldb_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
549 uint8_t qm_dir_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
550 /* store num stats and offset of the stats for each queue */
551 uint16_t xstats_count_per_qid[DLB2_MAX_NUM_QUEUES_ALL];
552 uint16_t xstats_offset_for_qid[DLB2_MAX_NUM_QUEUES_ALL];
553 /* store num stats and offset of the stats for each port */
554 uint16_t xstats_count_per_port[DLB2_MAX_NUM_PORTS_ALL];
555 uint16_t xstats_offset_for_port[DLB2_MAX_NUM_PORTS_ALL];
556 struct dlb2_get_num_resources_args hw_rsrc_query_results;
557 uint32_t xstats_count_mode_queue;
558 struct dlb2_hw_dev qm_instance; /* strictly hw related */
559 uint64_t global_dequeue_wait_ticks;
560 struct dlb2_xstats_entry *xstats;
561 struct rte_eventdev *event_dev; /* backlink to dev */
562 uint32_t xstats_count_mode_dev;
563 uint32_t xstats_count_mode_port;
564 uint32_t xstats_count;
565 uint32_t inflights; /* use __atomic builtins */
566 uint32_t new_event_limit;
567 int max_num_events_override;
568 int num_dir_credits_override;
569 bool vector_opts_disabled;
570 volatile enum dlb2_run_state run_state;
571 uint16_t num_dir_queues; /* total num of evdev dir queues requested */
574 uint16_t num_dir_credits;
575 uint16_t num_ldb_credits;
578 uint16_t num_credits;
581 uint16_t num_queues; /* total queues */
582 uint16_t num_ldb_queues; /* total num of evdev ldb queues requested */
583 uint16_t num_ports; /* total num of evdev ports requested */
584 uint16_t num_ldb_ports; /* total num of ldb ports requested */
585 uint16_t num_dir_ports; /* total num of dir ports requested */
587 bool global_dequeue_wait; /* Not using per dequeue wait if true */
589 enum dlb2_cq_poll_modes poll_mode;
591 int sw_credit_quanta;
592 int default_depth_thresh;
598 uint16_t max_ldb_credits;
599 uint16_t max_dir_credits;
600 /* use __atomic builtins */ /* shared hw cred */
601 uint32_t ldb_credit_pool __rte_cache_aligned;
602 /* use __atomic builtins */ /* shared hw cred */
603 uint32_t dir_credit_pool __rte_cache_aligned;
606 uint16_t max_credits;
607 /* use __atomic builtins */ /* shared hw cred */
608 uint32_t credit_pool __rte_cache_aligned;
613 /* used for collecting and passing around the dev args */
614 struct dlb2_qid_depth_thresholds {
615 int val[DLB2_MAX_NUM_QUEUES_ALL];
618 struct dlb2_devargs {
621 int num_dir_credits_override;
624 struct dlb2_qid_depth_thresholds qid_depth_thresholds;
625 enum dlb2_cos cos_id;
627 int sw_credit_quanta;
628 int default_depth_thresh;
629 bool vector_opts_disabled;
632 /* End Eventdev related defines and structs */
634 /* Forwards for non-inlined functions */
636 void dlb2_eventdev_dump(struct rte_eventdev *dev, FILE *f);
638 int dlb2_xstats_init(struct dlb2_eventdev *dlb2);
640 void dlb2_xstats_uninit(struct dlb2_eventdev *dlb2);
642 int dlb2_eventdev_xstats_get(const struct rte_eventdev *dev,
643 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
644 const unsigned int ids[], uint64_t values[], unsigned int n);
646 int dlb2_eventdev_xstats_get_names(const struct rte_eventdev *dev,
647 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
648 struct rte_event_dev_xstats_name *xstat_names,
649 unsigned int *ids, unsigned int size);
651 uint64_t dlb2_eventdev_xstats_get_by_name(const struct rte_eventdev *dev,
652 const char *name, unsigned int *id);
654 int dlb2_eventdev_xstats_reset(struct rte_eventdev *dev,
655 enum rte_event_dev_xstats_mode mode,
656 int16_t queue_port_id,
657 const uint32_t ids[],
660 int test_dlb2_eventdev(void);
662 int dlb2_primary_eventdev_probe(struct rte_eventdev *dev,
664 struct dlb2_devargs *dlb2_args);
666 int dlb2_secondary_eventdev_probe(struct rte_eventdev *dev,
669 uint32_t dlb2_get_queue_depth(struct dlb2_eventdev *dlb2,
670 struct dlb2_eventdev_queue *queue);
672 int dlb2_parse_params(const char *params,
674 struct dlb2_devargs *dlb2_args,
678 extern struct process_local_port_data dlb2_port[][DLB2_NUM_PORT_TYPES];
680 #endif /* _DLB2_PRIV_H_ */