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 /* Default = Worker */
28 #define DLB2_SW_CREDIT_P_QUANTA_DEFAULT 256 /* Producer */
29 #define DLB2_SW_CREDIT_C_QUANTA_DEFAULT 256 /* Consumer */
30 #define DLB2_DEPTH_THRESH_DEFAULT 256
32 /* command line arg strings */
33 #define NUMA_NODE_ARG "numa_node"
34 #define DLB2_MAX_NUM_EVENTS "max_num_events"
35 #define DLB2_NUM_DIR_CREDITS "num_dir_credits"
36 #define DEV_ID_ARG "dev_id"
37 #define DLB2_QID_DEPTH_THRESH_ARG "qid_depth_thresh"
38 #define DLB2_COS_ARG "cos"
39 #define DLB2_POLL_INTERVAL_ARG "poll_interval"
40 #define DLB2_SW_CREDIT_QUANTA_ARG "sw_credit_quanta"
41 #define DLB2_HW_CREDIT_QUANTA_ARG "hw_credit_quanta"
42 #define DLB2_DEPTH_THRESH_ARG "default_depth_thresh"
43 #define DLB2_VECTOR_OPTS_ENAB_ARG "vector_opts_enable"
45 /* Begin HW related defines and structs */
48 #define DLB2_HW_V2_5 1
49 #define DLB2_MAX_NUM_DOMAINS 32
50 #define DLB2_MAX_NUM_VFS 16
51 #define DLB2_MAX_NUM_LDB_QUEUES 32
52 #define DLB2_MAX_NUM_LDB_PORTS 64
53 #define DLB2_MAX_NUM_DIR_PORTS_V2 DLB2_MAX_NUM_DIR_QUEUES_V2
54 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
55 #define DLB2_MAX_NUM_DIR_PORTS(ver) (ver == DLB2_HW_V2 ? \
56 DLB2_MAX_NUM_DIR_PORTS_V2 : \
57 DLB2_MAX_NUM_DIR_PORTS_V2_5)
58 #define DLB2_MAX_NUM_DIR_QUEUES_V2 64 /* DIR == directed */
59 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
60 /* When needed for array sizing, the DLB 2.5 macro is used */
61 #define DLB2_MAX_NUM_DIR_QUEUES(ver) (ver == DLB2_HW_V2 ? \
62 DLB2_MAX_NUM_DIR_QUEUES_V2 : \
63 DLB2_MAX_NUM_DIR_QUEUES_V2_5)
64 #define DLB2_MAX_NUM_FLOWS (64 * 1024)
65 #define DLB2_MAX_NUM_LDB_CREDITS (8 * 1024)
66 #define DLB2_MAX_NUM_DIR_CREDITS(ver) (ver == DLB2_HW_V2 ? 4096 : 0)
67 #define DLB2_MAX_NUM_CREDITS(ver) (ver == DLB2_HW_V2 ? \
68 0 : DLB2_MAX_NUM_LDB_CREDITS)
69 #define DLB2_MAX_NUM_LDB_CREDIT_POOLS 64
70 #define DLB2_MAX_NUM_DIR_CREDIT_POOLS 64
71 #define DLB2_MAX_NUM_HIST_LIST_ENTRIES 2048
72 #define DLB2_MAX_NUM_QIDS_PER_LDB_CQ 8
73 #define DLB2_QID_PRIORITIES 8
74 #define DLB2_MAX_DEVICE_PATH 32
75 #define DLB2_MIN_DEQUEUE_TIMEOUT_NS 1
76 /* Note: "- 1" here to support the timeout range check in eventdev_autotest */
77 #define DLB2_MAX_DEQUEUE_TIMEOUT_NS (UINT32_MAX - 1)
78 #define DLB2_SW_CREDIT_BATCH_SZ 32 /* Default - Worker */
79 #define DLB2_SW_CREDIT_P_BATCH_SZ 256 /* Producer */
80 #define DLB2_SW_CREDIT_C_BATCH_SZ 256 /* Consumer */
81 #define DLB2_NUM_SN_GROUPS 2
82 #define DLB2_MAX_LDB_SN_ALLOC 1024
83 #define DLB2_MAX_QUEUE_DEPTH_THRESHOLD 8191
85 /* 2048 total hist list entries and 64 total ldb ports, which
86 * makes for 2048/64 == 32 hist list entries per port. However, CQ
87 * depth must be a power of 2 and must also be >= HIST LIST entries.
88 * As a result we just limit the maximum dequeue depth to 32.
90 #define DLB2_MIN_CQ_DEPTH 1
91 #define DLB2_MAX_CQ_DEPTH 32
92 #define DLB2_MIN_HARDWARE_CQ_DEPTH 8
93 #define DLB2_NUM_HIST_LIST_ENTRIES_PER_LDB_PORT \
96 #define DLB2_HW_DEVICE_FROM_PCI_ID(_pdev) \
97 (((_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_PF) || \
98 (_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_VF)) ? \
99 DLB2_HW_V2_5 : DLB2_HW_V2)
102 * Static per queue/port provisioning values
104 #define DLB2_NUM_ATOMIC_INFLIGHTS_PER_QUEUE 64
106 #define CQ_BASE(is_dir) ((is_dir) ? DLB2_DIR_CQ_BASE : DLB2_LDB_CQ_BASE)
107 #define CQ_SIZE(is_dir) ((is_dir) ? DLB2_DIR_CQ_MAX_SIZE : \
108 DLB2_LDB_CQ_MAX_SIZE)
109 #define PP_BASE(is_dir) ((is_dir) ? DLB2_DIR_PP_BASE : DLB2_LDB_PP_BASE)
111 #define DLB2_NUM_QES_PER_CACHE_LINE 4
113 #define DLB2_MAX_ENQUEUE_DEPTH 64
114 #define DLB2_MIN_ENQUEUE_DEPTH 4
116 #define DLB2_NAME_SIZE 64
119 #define DLB2_2K (2 * DLB2_1K)
120 #define DLB2_4K (4 * DLB2_1K)
121 #define DLB2_16K (16 * DLB2_1K)
122 #define DLB2_32K (32 * DLB2_1K)
123 #define DLB2_1MB (DLB2_1K * DLB2_1K)
124 #define DLB2_16MB (16 * DLB2_1MB)
126 /* Use the upper 3 bits of the event priority to select the DLB2 priority */
127 #define EV_TO_DLB2_PRIO(x) ((x) >> 5)
128 #define DLB2_TO_EV_PRIO(x) ((x) << 5)
135 enum dlb2_hw_port_types {
138 DLB2_NUM_PORT_TYPES /* Must be last */
141 enum dlb2_hw_queue_types {
144 DLB2_NUM_QUEUE_TYPES /* Must be last */
147 #define DLB2_COMBINED_POOL DLB2_LDB_QUEUE
149 #define PORT_TYPE(p) ((p)->is_directed ? DLB2_DIR_PORT : DLB2_LDB_PORT)
151 /* Do not change - must match hardware! */
152 enum dlb2_hw_sched_type {
153 DLB2_SCHED_ATOMIC = 0,
154 DLB2_SCHED_UNORDERED,
157 /* DLB2_NUM_HW_SCHED_TYPES must be last */
158 DLB2_NUM_HW_SCHED_TYPES
161 struct dlb2_hw_rsrcs {
162 int32_t nb_events_limit;
163 uint32_t num_queues; /* Total queues (lb + dir) */
164 uint32_t num_ldb_queues; /* Number of available ldb queues */
165 uint32_t num_ldb_ports; /* Number of load balanced ports */
166 uint32_t num_dir_ports; /* Number of directed ports */
169 uint32_t num_ldb_credits; /* Number of ldb credits */
170 uint32_t num_dir_credits; /* Number of dir credits */
173 uint32_t num_credits; /* Number of combined credits */
176 uint32_t reorder_window_size; /* Size of reorder window */
179 struct dlb2_hw_resource_info {
180 /**> Max resources that can be provided */
181 struct dlb2_hw_rsrcs hw_rsrc_max;
182 int num_sched_domains;
186 enum dlb2_enqueue_type {
188 * New : Used to inject a new packet into the QM.
192 * Forward : Enqueues a packet, and
193 * - if atomic: release any lock it holds in the QM
194 * - if ordered: release the packet for egress re-ordering
198 * Enqueue Drop : Release an inflight packet. Must be called with
199 * event == NULL. Used to drop a packet.
201 * Note that all packets dequeued from a load-balanced port must be
202 * released, either with DLB2_ENQ_DROP or DLB2_ENQ_FWD.
206 /* marker for array sizing etc. */
210 /* hw-specific format - do not change */
212 struct dlb2_event_type {
218 union dlb2_opaque_data {
219 uint16_t opaque_data;
220 struct dlb2_event_type event_type;
223 struct dlb2_msg_info {
225 uint8_t sched_type:2;
230 #define DLB2_NEW_CMD_BYTE 0x08
231 #define DLB2_FWD_CMD_BYTE 0x0A
232 #define DLB2_COMP_CMD_BYTE 0x02
233 #define DLB2_POP_CMD_BYTE 0x01
234 #define DLB2_NOOP_CMD_BYTE 0x00
236 /* hw-specific format - do not change */
237 struct dlb2_enqueue_qe {
240 union dlb2_opaque_data u;
242 uint8_t sched_type:2;
265 /* hw-specific format - do not change */
266 struct dlb2_cq_pop_qe {
268 union dlb2_opaque_data u;
270 uint8_t sched_type:2;
293 /* hw-specific format - do not change */
294 struct dlb2_dequeue_qe {
296 union dlb2_opaque_data u;
298 uint8_t sched_type:2;
301 uint16_t flow_id:16; /* was pp_id in v1 */
304 uint8_t qid_depth:2; /* 2 bits in v2 */
310 union dlb2_port_config {
311 struct dlb2_create_ldb_port_args ldb;
312 struct dlb2_create_dir_port_args dir;
315 enum dlb2_port_state {
321 enum dlb2_configuration_state {
322 /* The resource has not been configured */
324 /* The resource was configured, but the device was stopped */
325 DLB2_PREV_CONFIGURED,
326 /* The resource is currently configured */
334 uint16_t dir_credits;
335 uint32_t dequeue_depth;
336 enum dlb2_token_pop_mode token_pop_mode;
337 union dlb2_port_config cfg;
338 uint32_t *credit_pool[DLB2_NUM_QUEUE_TYPES]; /* use __atomic builtins */
341 uint16_t cached_ldb_credits;
342 uint16_t ldb_credits;
343 uint16_t cached_dir_credits;
346 uint16_t cached_credits;
351 uint16_t owed_tokens;
352 int16_t issued_releases;
353 int16_t token_pop_thresh;
356 uint16_t cq_idx_unmasked;
357 uint16_t cq_depth_mask;
358 uint16_t gen_bit_shift;
359 uint64_t cq_rolling_mask; /*
360 * rotate to always have right expected
363 uint64_t cq_rolling_mask_2;
364 void *cq_addr_cached; /* avoid multiple refs */
365 enum dlb2_port_state state;
366 enum dlb2_configuration_state config_state;
368 uint8_t *qid_mappings;
369 struct dlb2_enqueue_qe *qe4; /* Cache line's worth of QEs (4) */
370 struct dlb2_enqueue_qe *int_arm_qe;
371 struct dlb2_cq_pop_qe *consume_qe;
372 struct dlb2_eventdev *dlb2; /* back ptr */
373 struct dlb2_eventdev_port *ev_port; /* back ptr */
374 bool use_scalar; /* force usage of scalar code */
375 uint16_t hw_credit_quanta;
378 /* Per-process per-port mmio and memory pointers */
379 struct process_local_port_data {
381 struct dlb2_dequeue_qe *cq_base;
382 const struct rte_memzone *mz;
386 struct dlb2_eventdev;
388 struct dlb2_port_low_level_io_functions {
389 void (*pp_enqueue_four)(void *qe4, void *pp_addr);
397 uint32_t num_ldb_credits;
398 uint32_t num_dir_credits;
401 uint32_t num_credits;
404 struct dlb2_create_sched_domain_args resources;
408 DLB2_COS_DEFAULT = -1,
416 struct dlb2_config cfg;
417 struct dlb2_hw_resource_info info;
418 void *pf_dev; /* opaque pointer to PF PMD dev (struct dlb2_dev) */
420 enum dlb2_cos cos_id;
421 rte_spinlock_t resource_lock; /* for MP support */
422 } __rte_cache_aligned;
424 /* End HW related defines and structs */
426 /* Begin DLB2 PMD Eventdev related defines and structs */
428 #define DLB2_MAX_NUM_QUEUES(ver) \
429 (DLB2_MAX_NUM_DIR_QUEUES(ver) + DLB2_MAX_NUM_LDB_QUEUES)
431 #define DLB2_MAX_NUM_PORTS(ver) \
432 (DLB2_MAX_NUM_DIR_PORTS(ver) + DLB2_MAX_NUM_LDB_PORTS)
434 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
435 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
436 #define DLB2_MAX_NUM_QUEUES_ALL \
437 (DLB2_MAX_NUM_DIR_QUEUES_V2_5 + DLB2_MAX_NUM_LDB_QUEUES)
438 #define DLB2_MAX_NUM_PORTS_ALL \
439 (DLB2_MAX_NUM_DIR_PORTS_V2_5 + DLB2_MAX_NUM_LDB_PORTS)
440 #define DLB2_MAX_INPUT_QUEUE_DEPTH 256
442 /** Structure to hold the queue to port link establishment attributes */
444 struct dlb2_event_queue_link {
451 struct dlb2_traffic_stats {
454 uint64_t rx_interrupt_wait;
455 uint64_t rx_umonitor_umwait;
457 uint64_t total_polls;
461 uint64_t tx_nospc_ldb_hw_credits;
462 uint64_t tx_nospc_dir_hw_credits;
465 uint64_t tx_nospc_hw_credits;
468 uint64_t tx_nospc_inflight_max;
469 uint64_t tx_nospc_new_event_limit;
470 uint64_t tx_nospc_inflight_credits;
473 /* DLB2 HW sets the 2bit qid_depth in rx QEs based on the programmable depth
474 * threshold. The global default value in config/common_base (or rte_config.h)
475 * can be overridden on a per-qid basis using a vdev command line parameter.
476 * 3: depth > threshold
477 * 2: threshold >= depth > 3/4 threshold
478 * 1: 3/4 threshold >= depth > 1/2 threshold
479 * 0: depth <= 1/2 threshold.
481 #define DLB2_QID_DEPTH_LE50 0
482 #define DLB2_QID_DEPTH_GT50_LE75 1
483 #define DLB2_QID_DEPTH_GT75_LE100 2
484 #define DLB2_QID_DEPTH_GT100 3
485 #define DLB2_NUM_QID_DEPTH_STAT_VALS 4 /* 2 bits */
487 struct dlb2_queue_stats {
489 uint64_t qid_depth[DLB2_NUM_QID_DEPTH_STAT_VALS];
492 struct dlb2_port_stats {
493 struct dlb2_traffic_stats traffic;
494 uint64_t tx_op_cnt[4]; /* indexed by rte_event.op */
495 uint64_t tx_implicit_rel;
496 uint64_t tx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
498 uint64_t rx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
499 uint64_t rx_sched_invalid;
500 struct dlb2_queue_stats queue[DLB2_MAX_NUM_QUEUES_ALL];
503 struct dlb2_eventdev_port {
504 struct dlb2_port qm_port; /* hw specific data structure */
505 struct rte_event_port_conf conf; /* user-supplied configuration */
506 uint16_t inflight_credits; /* num credits this port has right now */
507 uint16_t credit_update_quanta;
508 struct dlb2_eventdev *dlb2; /* backlink optimization */
509 struct dlb2_port_stats stats __rte_cache_aligned;
510 struct dlb2_event_queue_link link[DLB2_MAX_NUM_QIDS_PER_LDB_CQ];
512 uint32_t id; /* port id */
513 /* num releases yet to be completed on this port.
514 * Only applies to load-balanced ports.
516 uint16_t outstanding_releases;
517 uint16_t inflight_max; /* app requested max inflights for this port */
518 /* setup_done is set when the event port is setup */
520 /* enq_configured is set when the qm port is created */
522 uint8_t implicit_release; /* release events before dequeueing */
523 } __rte_cache_aligned;
526 uint32_t num_qid_inflights; /* User config */
527 uint32_t num_atm_inflights; /* User config */
528 enum dlb2_configuration_state config_state;
529 int sched_type; /* LB queue only */
534 struct dlb2_eventdev_queue {
535 struct dlb2_queue qm_queue;
536 struct rte_event_queue_conf conf; /* User config */
537 int depth_threshold; /* use default if 0 */
543 enum dlb2_run_state {
544 DLB2_RUN_STATE_STOPPED = 0,
545 DLB2_RUN_STATE_STOPPING,
546 DLB2_RUN_STATE_STARTING,
547 DLB2_RUN_STATE_STARTED
550 struct dlb2_eventdev {
551 struct dlb2_eventdev_port ev_ports[DLB2_MAX_NUM_PORTS_ALL];
552 struct dlb2_eventdev_queue ev_queues[DLB2_MAX_NUM_QUEUES_ALL];
553 uint8_t qm_ldb_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
554 uint8_t qm_dir_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
555 /* store num stats and offset of the stats for each queue */
556 uint16_t xstats_count_per_qid[DLB2_MAX_NUM_QUEUES_ALL];
557 uint16_t xstats_offset_for_qid[DLB2_MAX_NUM_QUEUES_ALL];
558 /* store num stats and offset of the stats for each port */
559 uint16_t xstats_count_per_port[DLB2_MAX_NUM_PORTS_ALL];
560 uint16_t xstats_offset_for_port[DLB2_MAX_NUM_PORTS_ALL];
561 struct dlb2_get_num_resources_args hw_rsrc_query_results;
562 uint32_t xstats_count_mode_queue;
563 struct dlb2_hw_dev qm_instance; /* strictly hw related */
564 uint64_t global_dequeue_wait_ticks;
565 struct dlb2_xstats_entry *xstats;
566 struct rte_eventdev *event_dev; /* backlink to dev */
567 uint32_t xstats_count_mode_dev;
568 uint32_t xstats_count_mode_port;
569 uint32_t xstats_count;
570 uint32_t inflights; /* use __atomic builtins */
571 uint32_t new_event_limit;
572 int max_num_events_override;
573 int num_dir_credits_override;
574 bool vector_opts_enabled;
575 volatile enum dlb2_run_state run_state;
576 uint16_t num_dir_queues; /* total num of evdev dir queues requested */
579 uint16_t num_dir_credits;
580 uint16_t num_ldb_credits;
583 uint16_t num_credits;
586 uint16_t num_queues; /* total queues */
587 uint16_t num_ldb_queues; /* total num of evdev ldb queues requested */
588 uint16_t num_ports; /* total num of evdev ports requested */
589 uint16_t num_ldb_ports; /* total num of ldb ports requested */
590 uint16_t num_dir_ports; /* total num of dir ports requested */
592 bool global_dequeue_wait; /* Not using per dequeue wait if true */
593 enum dlb2_cq_poll_modes poll_mode;
595 int sw_credit_quanta;
596 int hw_credit_quanta;
597 int default_depth_thresh;
603 uint16_t max_ldb_credits;
604 uint16_t max_dir_credits;
605 /* use __atomic builtins */ /* shared hw cred */
606 uint32_t ldb_credit_pool __rte_cache_aligned;
607 /* use __atomic builtins */ /* shared hw cred */
608 uint32_t dir_credit_pool __rte_cache_aligned;
611 uint16_t max_credits;
612 /* use __atomic builtins */ /* shared hw cred */
613 uint32_t credit_pool __rte_cache_aligned;
618 /* used for collecting and passing around the dev args */
619 struct dlb2_qid_depth_thresholds {
620 int val[DLB2_MAX_NUM_QUEUES_ALL];
623 struct dlb2_devargs {
626 int num_dir_credits_override;
628 struct dlb2_qid_depth_thresholds qid_depth_thresholds;
629 enum dlb2_cos cos_id;
631 int sw_credit_quanta;
632 int hw_credit_quanta;
633 int default_depth_thresh;
634 bool vector_opts_enabled;
637 /* End Eventdev related defines and structs */
639 /* Forwards for non-inlined functions */
641 void dlb2_eventdev_dump(struct rte_eventdev *dev, FILE *f);
643 int dlb2_xstats_init(struct dlb2_eventdev *dlb2);
645 void dlb2_xstats_uninit(struct dlb2_eventdev *dlb2);
647 int dlb2_eventdev_xstats_get(const struct rte_eventdev *dev,
648 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
649 const unsigned int ids[], uint64_t values[], unsigned int n);
651 int dlb2_eventdev_xstats_get_names(const struct rte_eventdev *dev,
652 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
653 struct rte_event_dev_xstats_name *xstat_names,
654 unsigned int *ids, unsigned int size);
656 uint64_t dlb2_eventdev_xstats_get_by_name(const struct rte_eventdev *dev,
657 const char *name, unsigned int *id);
659 int dlb2_eventdev_xstats_reset(struct rte_eventdev *dev,
660 enum rte_event_dev_xstats_mode mode,
661 int16_t queue_port_id,
662 const uint32_t ids[],
665 int test_dlb2_eventdev(void);
667 int dlb2_primary_eventdev_probe(struct rte_eventdev *dev,
669 struct dlb2_devargs *dlb2_args);
671 int dlb2_secondary_eventdev_probe(struct rte_eventdev *dev,
674 uint32_t dlb2_get_queue_depth(struct dlb2_eventdev *dlb2,
675 struct dlb2_eventdev_queue *queue);
677 int dlb2_parse_params(const char *params,
679 struct dlb2_devargs *dlb2_args,
683 extern struct process_local_port_data dlb2_port[][DLB2_NUM_PORT_TYPES];
685 #endif /* _DLB2_PRIV_H_ */