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
31 #define DLB2_MIN_CQ_DEPTH_OVERRIDE 32
32 #define DLB2_MAX_CQ_DEPTH_OVERRIDE 1024
34 /* command line arg strings */
35 #define NUMA_NODE_ARG "numa_node"
36 #define DLB2_MAX_NUM_EVENTS "max_num_events"
37 #define DLB2_NUM_DIR_CREDITS "num_dir_credits"
38 #define DEV_ID_ARG "dev_id"
39 #define DLB2_QID_DEPTH_THRESH_ARG "qid_depth_thresh"
40 #define DLB2_COS_ARG "cos"
41 #define DLB2_POLL_INTERVAL_ARG "poll_interval"
42 #define DLB2_SW_CREDIT_QUANTA_ARG "sw_credit_quanta"
43 #define DLB2_HW_CREDIT_QUANTA_ARG "hw_credit_quanta"
44 #define DLB2_DEPTH_THRESH_ARG "default_depth_thresh"
45 #define DLB2_VECTOR_OPTS_ENAB_ARG "vector_opts_enable"
46 #define DLB2_MAX_CQ_DEPTH "max_cq_depth"
48 /* Begin HW related defines and structs */
51 #define DLB2_HW_V2_5 1
52 #define DLB2_MAX_NUM_DOMAINS 32
53 #define DLB2_MAX_NUM_VFS 16
54 #define DLB2_MAX_NUM_LDB_QUEUES 32
55 #define DLB2_MAX_NUM_LDB_PORTS 64
56 #define DLB2_MAX_NUM_DIR_PORTS_V2 DLB2_MAX_NUM_DIR_QUEUES_V2
57 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
58 #define DLB2_MAX_NUM_DIR_PORTS(ver) (ver == DLB2_HW_V2 ? \
59 DLB2_MAX_NUM_DIR_PORTS_V2 : \
60 DLB2_MAX_NUM_DIR_PORTS_V2_5)
61 #define DLB2_MAX_NUM_DIR_QUEUES_V2 64 /* DIR == directed */
62 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
63 /* When needed for array sizing, the DLB 2.5 macro is used */
64 #define DLB2_MAX_NUM_DIR_QUEUES(ver) (ver == DLB2_HW_V2 ? \
65 DLB2_MAX_NUM_DIR_QUEUES_V2 : \
66 DLB2_MAX_NUM_DIR_QUEUES_V2_5)
67 #define DLB2_MAX_NUM_FLOWS (64 * 1024)
68 #define DLB2_MAX_NUM_LDB_CREDITS (8 * 1024)
69 #define DLB2_MAX_NUM_DIR_CREDITS(ver) (ver == DLB2_HW_V2 ? 4096 : 0)
70 #define DLB2_MAX_NUM_CREDITS(ver) (ver == DLB2_HW_V2 ? \
71 0 : DLB2_MAX_NUM_LDB_CREDITS)
72 #define DLB2_MAX_NUM_LDB_CREDIT_POOLS 64
73 #define DLB2_MAX_NUM_DIR_CREDIT_POOLS 64
74 #define DLB2_MAX_NUM_HIST_LIST_ENTRIES 2048
75 #define DLB2_MAX_NUM_QIDS_PER_LDB_CQ 8
76 #define DLB2_QID_PRIORITIES 8
77 #define DLB2_MAX_DEVICE_PATH 32
78 #define DLB2_MIN_DEQUEUE_TIMEOUT_NS 1
79 /* Note: "- 1" here to support the timeout range check in eventdev_autotest */
80 #define DLB2_MAX_DEQUEUE_TIMEOUT_NS (UINT32_MAX - 1)
81 #define DLB2_SW_CREDIT_BATCH_SZ 32 /* Default - Worker */
82 #define DLB2_SW_CREDIT_P_BATCH_SZ 256 /* Producer */
83 #define DLB2_SW_CREDIT_C_BATCH_SZ 256 /* Consumer */
84 #define DLB2_NUM_SN_GROUPS 2
85 #define DLB2_MAX_LDB_SN_ALLOC 1024
86 #define DLB2_MAX_QUEUE_DEPTH_THRESHOLD 8191
88 /* 2048 total hist list entries and 64 total ldb ports, which
89 * makes for 2048/64 == 32 hist list entries per port. However, CQ
90 * depth must be a power of 2 and must also be >= HIST LIST entries.
91 * As a result we just limit the maximum dequeue depth to 32.
93 #define DLB2_MAX_HL_ENTRIES 2048
94 #define DLB2_MIN_CQ_DEPTH 1
95 #define DLB2_DEFAULT_CQ_DEPTH 32
96 #define DLB2_MIN_HARDWARE_CQ_DEPTH 8
97 #define DLB2_NUM_HIST_LIST_ENTRIES_PER_LDB_PORT \
100 #define DLB2_HW_DEVICE_FROM_PCI_ID(_pdev) \
101 (((_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_PF) || \
102 (_pdev->id.device_id == PCI_DEVICE_ID_INTEL_DLB2_5_VF)) ? \
103 DLB2_HW_V2_5 : DLB2_HW_V2)
106 * Static per queue/port provisioning values
108 #define DLB2_NUM_ATOMIC_INFLIGHTS_PER_QUEUE 64
110 #define CQ_BASE(is_dir) ((is_dir) ? DLB2_DIR_CQ_BASE : DLB2_LDB_CQ_BASE)
111 #define CQ_SIZE(is_dir) ((is_dir) ? DLB2_DIR_CQ_MAX_SIZE : \
112 DLB2_LDB_CQ_MAX_SIZE)
113 #define PP_BASE(is_dir) ((is_dir) ? DLB2_DIR_PP_BASE : DLB2_LDB_PP_BASE)
115 #define DLB2_NUM_QES_PER_CACHE_LINE 4
117 #define DLB2_MAX_ENQUEUE_DEPTH 64
118 #define DLB2_MIN_ENQUEUE_DEPTH 4
120 #define DLB2_NAME_SIZE 64
123 #define DLB2_2K (2 * DLB2_1K)
124 #define DLB2_4K (4 * DLB2_1K)
125 #define DLB2_16K (16 * DLB2_1K)
126 #define DLB2_32K (32 * DLB2_1K)
127 #define DLB2_1MB (DLB2_1K * DLB2_1K)
128 #define DLB2_16MB (16 * DLB2_1MB)
130 /* Use the upper 3 bits of the event priority to select the DLB2 priority */
131 #define EV_TO_DLB2_PRIO(x) ((x) >> 5)
132 #define DLB2_TO_EV_PRIO(x) ((x) << 5)
139 enum dlb2_hw_port_types {
142 DLB2_NUM_PORT_TYPES /* Must be last */
145 enum dlb2_hw_queue_types {
148 DLB2_NUM_QUEUE_TYPES /* Must be last */
151 #define DLB2_COMBINED_POOL DLB2_LDB_QUEUE
153 #define PORT_TYPE(p) ((p)->is_directed ? DLB2_DIR_PORT : DLB2_LDB_PORT)
155 /* Do not change - must match hardware! */
156 enum dlb2_hw_sched_type {
157 DLB2_SCHED_ATOMIC = 0,
158 DLB2_SCHED_UNORDERED,
161 /* DLB2_NUM_HW_SCHED_TYPES must be last */
162 DLB2_NUM_HW_SCHED_TYPES
165 struct dlb2_hw_rsrcs {
166 int32_t nb_events_limit;
167 uint32_t num_queues; /* Total queues (lb + dir) */
168 uint32_t num_ldb_queues; /* Number of available ldb queues */
169 uint32_t num_ldb_ports; /* Number of load balanced ports */
170 uint32_t num_dir_ports; /* Number of directed ports */
173 uint32_t num_ldb_credits; /* Number of ldb credits */
174 uint32_t num_dir_credits; /* Number of dir credits */
177 uint32_t num_credits; /* Number of combined credits */
180 uint32_t reorder_window_size; /* Size of reorder window */
183 struct dlb2_hw_resource_info {
184 /**> Max resources that can be provided */
185 struct dlb2_hw_rsrcs hw_rsrc_max;
186 int num_sched_domains;
190 enum dlb2_enqueue_type {
192 * New : Used to inject a new packet into the QM.
196 * Forward : Enqueues a packet, and
197 * - if atomic: release any lock it holds in the QM
198 * - if ordered: release the packet for egress re-ordering
202 * Enqueue Drop : Release an inflight packet. Must be called with
203 * event == NULL. Used to drop a packet.
205 * Note that all packets dequeued from a load-balanced port must be
206 * released, either with DLB2_ENQ_DROP or DLB2_ENQ_FWD.
210 /* marker for array sizing etc. */
214 /* hw-specific format - do not change */
216 struct dlb2_event_type {
222 union dlb2_opaque_data {
223 uint16_t opaque_data;
224 struct dlb2_event_type event_type;
227 struct dlb2_msg_info {
229 uint8_t sched_type:2;
234 #define DLB2_NEW_CMD_BYTE 0x08
235 #define DLB2_FWD_CMD_BYTE 0x0A
236 #define DLB2_COMP_CMD_BYTE 0x02
237 #define DLB2_POP_CMD_BYTE 0x01
238 #define DLB2_NOOP_CMD_BYTE 0x00
240 /* hw-specific format - do not change */
241 struct dlb2_enqueue_qe {
244 union dlb2_opaque_data u;
246 uint8_t sched_type:2;
269 /* hw-specific format - do not change */
270 struct dlb2_cq_pop_qe {
272 union dlb2_opaque_data u;
274 uint8_t sched_type:2;
297 /* hw-specific format - do not change */
298 struct dlb2_dequeue_qe {
300 union dlb2_opaque_data u;
302 uint8_t sched_type:2;
305 uint16_t flow_id:16; /* was pp_id in v1 */
308 uint8_t qid_depth:2; /* 2 bits in v2 */
314 union dlb2_port_config {
315 struct dlb2_create_ldb_port_args ldb;
316 struct dlb2_create_dir_port_args dir;
319 enum dlb2_port_state {
325 enum dlb2_configuration_state {
326 /* The resource has not been configured */
328 /* The resource was configured, but the device was stopped */
329 DLB2_PREV_CONFIGURED,
330 /* The resource is currently configured */
338 uint16_t dir_credits;
339 uint32_t dequeue_depth;
340 enum dlb2_token_pop_mode token_pop_mode;
341 union dlb2_port_config cfg;
342 uint32_t *credit_pool[DLB2_NUM_QUEUE_TYPES]; /* use __atomic builtins */
345 uint16_t cached_ldb_credits;
346 uint16_t ldb_credits;
347 uint16_t cached_dir_credits;
350 uint16_t cached_credits;
355 uint16_t owed_tokens;
356 int16_t issued_releases;
357 int16_t token_pop_thresh;
360 uint16_t cq_idx_unmasked;
361 uint16_t cq_depth_mask;
362 uint16_t gen_bit_shift;
363 uint64_t cq_rolling_mask; /*
364 * rotate to always have right expected
367 uint64_t cq_rolling_mask_2;
368 void *cq_addr_cached; /* avoid multiple refs */
369 enum dlb2_port_state state;
370 enum dlb2_configuration_state config_state;
372 uint8_t *qid_mappings;
373 struct dlb2_enqueue_qe *qe4; /* Cache line's worth of QEs (4) */
374 struct dlb2_enqueue_qe *int_arm_qe;
375 struct dlb2_cq_pop_qe *consume_qe;
376 struct dlb2_eventdev *dlb2; /* back ptr */
377 struct dlb2_eventdev_port *ev_port; /* back ptr */
378 bool use_scalar; /* force usage of scalar code */
379 uint16_t hw_credit_quanta;
382 /* Per-process per-port mmio and memory pointers */
383 struct process_local_port_data {
385 struct dlb2_dequeue_qe *cq_base;
386 const struct rte_memzone *mz;
390 struct dlb2_eventdev;
392 struct dlb2_port_low_level_io_functions {
393 void (*pp_enqueue_four)(void *qe4, void *pp_addr);
401 uint32_t num_ldb_credits;
402 uint32_t num_dir_credits;
405 uint32_t num_credits;
408 struct dlb2_create_sched_domain_args resources;
412 DLB2_COS_DEFAULT = -1,
420 struct dlb2_config cfg;
421 struct dlb2_hw_resource_info info;
422 void *pf_dev; /* opaque pointer to PF PMD dev (struct dlb2_dev) */
424 enum dlb2_cos cos_id;
425 rte_spinlock_t resource_lock; /* for MP support */
426 } __rte_cache_aligned;
428 /* End HW related defines and structs */
430 /* Begin DLB2 PMD Eventdev related defines and structs */
432 #define DLB2_MAX_NUM_QUEUES(ver) \
433 (DLB2_MAX_NUM_DIR_QUEUES(ver) + DLB2_MAX_NUM_LDB_QUEUES)
435 #define DLB2_MAX_NUM_PORTS(ver) \
436 (DLB2_MAX_NUM_DIR_PORTS(ver) + DLB2_MAX_NUM_LDB_PORTS)
438 #define DLB2_MAX_NUM_DIR_QUEUES_V2_5 96
439 #define DLB2_MAX_NUM_DIR_PORTS_V2_5 DLB2_MAX_NUM_DIR_QUEUES_V2_5
440 #define DLB2_MAX_NUM_QUEUES_ALL \
441 (DLB2_MAX_NUM_DIR_QUEUES_V2_5 + DLB2_MAX_NUM_LDB_QUEUES)
442 #define DLB2_MAX_NUM_PORTS_ALL \
443 (DLB2_MAX_NUM_DIR_PORTS_V2_5 + DLB2_MAX_NUM_LDB_PORTS)
444 #define DLB2_MAX_INPUT_QUEUE_DEPTH 256
446 /** Structure to hold the queue to port link establishment attributes */
448 struct dlb2_event_queue_link {
455 struct dlb2_traffic_stats {
458 uint64_t rx_interrupt_wait;
459 uint64_t rx_umonitor_umwait;
461 uint64_t total_polls;
465 uint64_t tx_nospc_ldb_hw_credits;
466 uint64_t tx_nospc_dir_hw_credits;
469 uint64_t tx_nospc_hw_credits;
472 uint64_t tx_nospc_inflight_max;
473 uint64_t tx_nospc_new_event_limit;
474 uint64_t tx_nospc_inflight_credits;
477 /* DLB2 HW sets the 2bit qid_depth in rx QEs based on the programmable depth
478 * threshold. The global default value in config/common_base (or rte_config.h)
479 * can be overridden on a per-qid basis using a vdev command line parameter.
480 * 3: depth > threshold
481 * 2: threshold >= depth > 3/4 threshold
482 * 1: 3/4 threshold >= depth > 1/2 threshold
483 * 0: depth <= 1/2 threshold.
485 #define DLB2_QID_DEPTH_LE50 0
486 #define DLB2_QID_DEPTH_GT50_LE75 1
487 #define DLB2_QID_DEPTH_GT75_LE100 2
488 #define DLB2_QID_DEPTH_GT100 3
489 #define DLB2_NUM_QID_DEPTH_STAT_VALS 4 /* 2 bits */
491 struct dlb2_queue_stats {
493 uint64_t qid_depth[DLB2_NUM_QID_DEPTH_STAT_VALS];
496 struct dlb2_port_stats {
497 struct dlb2_traffic_stats traffic;
498 uint64_t tx_op_cnt[4]; /* indexed by rte_event.op */
499 uint64_t tx_implicit_rel;
500 uint64_t tx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
502 uint64_t rx_sched_cnt[DLB2_NUM_HW_SCHED_TYPES];
503 uint64_t rx_sched_invalid;
504 struct dlb2_queue_stats queue[DLB2_MAX_NUM_QUEUES_ALL];
507 struct dlb2_eventdev_port {
508 struct dlb2_port qm_port; /* hw specific data structure */
509 struct rte_event_port_conf conf; /* user-supplied configuration */
510 uint16_t inflight_credits; /* num credits this port has right now */
511 uint16_t credit_update_quanta;
512 struct dlb2_eventdev *dlb2; /* backlink optimization */
513 struct dlb2_port_stats stats __rte_cache_aligned;
514 struct dlb2_event_queue_link link[DLB2_MAX_NUM_QIDS_PER_LDB_CQ];
516 uint32_t id; /* port id */
517 /* num releases yet to be completed on this port.
518 * Only applies to load-balanced ports.
520 uint16_t outstanding_releases;
521 uint16_t inflight_max; /* app requested max inflights for this port */
522 /* setup_done is set when the event port is setup */
524 /* enq_configured is set when the qm port is created */
526 uint8_t implicit_release; /* release events before dequeuing */
527 } __rte_cache_aligned;
530 uint32_t num_qid_inflights; /* User config */
531 uint32_t num_atm_inflights; /* User config */
532 enum dlb2_configuration_state config_state;
533 int sched_type; /* LB queue only */
538 struct dlb2_eventdev_queue {
539 struct dlb2_queue qm_queue;
540 struct rte_event_queue_conf conf; /* User config */
541 int depth_threshold; /* use default if 0 */
547 enum dlb2_run_state {
548 DLB2_RUN_STATE_STOPPED = 0,
549 DLB2_RUN_STATE_STOPPING,
550 DLB2_RUN_STATE_STARTING,
551 DLB2_RUN_STATE_STARTED
554 struct dlb2_eventdev {
555 struct dlb2_eventdev_port ev_ports[DLB2_MAX_NUM_PORTS_ALL];
556 struct dlb2_eventdev_queue ev_queues[DLB2_MAX_NUM_QUEUES_ALL];
557 uint8_t qm_ldb_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
558 uint8_t qm_dir_to_ev_queue_id[DLB2_MAX_NUM_QUEUES_ALL];
559 /* store num stats and offset of the stats for each queue */
560 uint16_t xstats_count_per_qid[DLB2_MAX_NUM_QUEUES_ALL];
561 uint16_t xstats_offset_for_qid[DLB2_MAX_NUM_QUEUES_ALL];
562 /* store num stats and offset of the stats for each port */
563 uint16_t xstats_count_per_port[DLB2_MAX_NUM_PORTS_ALL];
564 uint16_t xstats_offset_for_port[DLB2_MAX_NUM_PORTS_ALL];
565 struct dlb2_get_num_resources_args hw_rsrc_query_results;
566 uint32_t xstats_count_mode_queue;
567 struct dlb2_hw_dev qm_instance; /* strictly hw related */
568 uint64_t global_dequeue_wait_ticks;
569 struct dlb2_xstats_entry *xstats;
570 struct rte_eventdev *event_dev; /* backlink to dev */
571 uint32_t xstats_count_mode_dev;
572 uint32_t xstats_count_mode_port;
573 uint32_t xstats_count;
574 uint32_t inflights; /* use __atomic builtins */
575 uint32_t new_event_limit;
576 int max_num_events_override;
577 int num_dir_credits_override;
578 bool vector_opts_enabled;
580 volatile enum dlb2_run_state run_state;
581 uint16_t num_dir_queues; /* total num of evdev dir queues requested */
584 uint16_t num_dir_credits;
585 uint16_t num_ldb_credits;
588 uint16_t num_credits;
591 uint16_t num_queues; /* total queues */
592 uint16_t num_ldb_queues; /* total num of evdev ldb queues requested */
593 uint16_t num_ports; /* total num of evdev ports requested */
594 uint16_t num_ldb_ports; /* total num of ldb ports requested */
595 uint16_t num_dir_ports; /* total num of dir ports requested */
597 bool global_dequeue_wait; /* Not using per dequeue wait if true */
598 enum dlb2_cq_poll_modes poll_mode;
600 int sw_credit_quanta;
601 int hw_credit_quanta;
602 int default_depth_thresh;
608 uint16_t max_ldb_credits;
609 uint16_t max_dir_credits;
610 /* use __atomic builtins */ /* shared hw cred */
611 uint32_t ldb_credit_pool __rte_cache_aligned;
612 /* use __atomic builtins */ /* shared hw cred */
613 uint32_t dir_credit_pool __rte_cache_aligned;
616 uint16_t max_credits;
617 /* use __atomic builtins */ /* shared hw cred */
618 uint32_t credit_pool __rte_cache_aligned;
623 /* used for collecting and passing around the dev args */
624 struct dlb2_qid_depth_thresholds {
625 int val[DLB2_MAX_NUM_QUEUES_ALL];
628 struct dlb2_devargs {
631 int num_dir_credits_override;
633 struct dlb2_qid_depth_thresholds qid_depth_thresholds;
634 enum dlb2_cos cos_id;
636 int sw_credit_quanta;
637 int hw_credit_quanta;
638 int default_depth_thresh;
639 bool vector_opts_enabled;
643 /* End Eventdev related defines and structs */
645 /* Forwards for non-inlined functions */
647 void dlb2_eventdev_dump(struct rte_eventdev *dev, FILE *f);
649 int dlb2_xstats_init(struct dlb2_eventdev *dlb2);
651 void dlb2_xstats_uninit(struct dlb2_eventdev *dlb2);
653 int dlb2_eventdev_xstats_get(const struct rte_eventdev *dev,
654 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
655 const unsigned int ids[], uint64_t values[], unsigned int n);
657 int dlb2_eventdev_xstats_get_names(const struct rte_eventdev *dev,
658 enum rte_event_dev_xstats_mode mode, uint8_t queue_port_id,
659 struct rte_event_dev_xstats_name *xstat_names,
660 unsigned int *ids, unsigned int size);
662 uint64_t dlb2_eventdev_xstats_get_by_name(const struct rte_eventdev *dev,
663 const char *name, unsigned int *id);
665 int dlb2_eventdev_xstats_reset(struct rte_eventdev *dev,
666 enum rte_event_dev_xstats_mode mode,
667 int16_t queue_port_id,
668 const uint32_t ids[],
671 int test_dlb2_eventdev(void);
673 int dlb2_primary_eventdev_probe(struct rte_eventdev *dev,
675 struct dlb2_devargs *dlb2_args);
677 int dlb2_secondary_eventdev_probe(struct rte_eventdev *dev,
680 uint32_t dlb2_get_queue_depth(struct dlb2_eventdev *dlb2,
681 struct dlb2_eventdev_queue *queue);
683 int dlb2_parse_params(const char *params,
685 struct dlb2_devargs *dlb2_args,
689 extern struct process_local_port_data dlb2_port[][DLB2_NUM_PORT_TYPES];
691 #endif /* _DLB2_PRIV_H_ */