DPDK logo

Elixir Cross Referencer

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright(c) 2016-2017 Intel Corporation
 */

#include <rte_ring.h>
#include <rte_hash_crc.h>
#include <rte_event_ring.h>
#include "sw_evdev.h"
#include "iq_chunk.h"
#include "event_ring.h"

#define SW_IQS_MASK (SW_IQS_MAX-1)

/* Retrieve the highest priority IQ or -1 if no pkts available. Doing the
 * CLZ twice is faster than caching the value due to data dependencies
 */
#define PKT_MASK_TO_IQ(pkts) \
	(__builtin_ctz(pkts | (1 << SW_IQS_MAX)))

#if SW_IQS_MAX != 4
#error Misconfigured PRIO_TO_IQ caused by SW_IQS_MAX value change
#endif
#define PRIO_TO_IQ(prio) (prio >> 6)

#define MAX_PER_IQ_DEQUEUE 48
#define FLOWID_MASK (SW_QID_NUM_FIDS-1)
/* use cheap bit mixing, we only need to lose a few bits */
#define SW_HASH_FLOWID(f) (((f) ^ (f >> 10)) & FLOWID_MASK)


static inline uint32_t
sw_schedule_atomic_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
		uint32_t iq_num, unsigned int count)
{
	struct rte_event qes[MAX_PER_IQ_DEQUEUE]; /* count <= MAX */
	struct rte_event blocked_qes[MAX_PER_IQ_DEQUEUE];
	uint32_t nb_blocked = 0;
	uint32_t i;

	if (count > MAX_PER_IQ_DEQUEUE)
		count = MAX_PER_IQ_DEQUEUE;

	/* This is the QID ID. The QID ID is static, hence it can be
	 * used to identify the stage of processing in history lists etc
	 */
	uint32_t qid_id = qid->id;

	iq_dequeue_burst(sw, &qid->iq[iq_num], qes, count);
	for (i = 0; i < count; i++) {
		const struct rte_event *qe = &qes[i];
		const uint16_t flow_id = SW_HASH_FLOWID(qes[i].flow_id);
		struct sw_fid_t *fid = &qid->fids[flow_id];
		int cq = fid->cq;

		if (cq < 0) {
			uint32_t cq_idx;
			if (qid->cq_next_tx >= qid->cq_num_mapped_cqs)
				qid->cq_next_tx = 0;
			cq_idx = qid->cq_next_tx++;

			cq = qid->cq_map[cq_idx];

			/* find least used */
			int cq_free_cnt = sw->cq_ring_space[cq];
			for (cq_idx = 0; cq_idx < qid->cq_num_mapped_cqs;
					cq_idx++) {
				int test_cq = qid->cq_map[cq_idx];
				int test_cq_free = sw->cq_ring_space[test_cq];
				if (test_cq_free > cq_free_cnt) {
					cq = test_cq;
					cq_free_cnt = test_cq_free;
				}
			}

			fid->cq = cq; /* this pins early */
		}

		if (sw->cq_ring_space[cq] == 0 ||
				sw->ports[cq].inflights == SW_PORT_HIST_LIST) {
			blocked_qes[nb_blocked++] = *qe;
			continue;
		}

		struct sw_port *p = &sw->ports[cq];

		/* at this point we can queue up the packet on the cq_buf */
		fid->pcount++;
		p->cq_buf[p->cq_buf_count++] = *qe;
		p->inflights++;
		sw->cq_ring_space[cq]--;

		int head = (p->hist_head++ & (SW_PORT_HIST_LIST-1));
		p->hist_list[head].fid = flow_id;
		p->hist_list[head].qid = qid_id;

		p->stats.tx_pkts++;
		qid->stats.tx_pkts++;
		qid->to_port[cq]++;

		/* if we just filled in the last slot, flush the buffer */
		if (sw->cq_ring_space[cq] == 0) {
			struct rte_event_ring *worker = p->cq_worker_ring;
			rte_event_ring_enqueue_burst(worker, p->cq_buf,
					p->cq_buf_count,
					&sw->cq_ring_space[cq]);
			p->cq_buf_count = 0;
		}
	}
	iq_put_back(sw, &qid->iq[iq_num], blocked_qes, nb_blocked);

	return count - nb_blocked;
}

static inline uint32_t
sw_schedule_parallel_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
		uint32_t iq_num, unsigned int count, int keep_order)
{
	uint32_t i;
	uint32_t cq_idx = qid->cq_next_tx;

	/* This is the QID ID. The QID ID is static, hence it can be
	 * used to identify the stage of processing in history lists etc
	 */
	uint32_t qid_id = qid->id;

	if (count > MAX_PER_IQ_DEQUEUE)
		count = MAX_PER_IQ_DEQUEUE;

	if (keep_order)
		/* only schedule as many as we have reorder buffer entries */
		count = RTE_MIN(count,
				rob_ring_count(qid->reorder_buffer_freelist));

	for (i = 0; i < count; i++) {
		const struct rte_event *qe = iq_peek(&qid->iq[iq_num]);
		uint32_t cq_check_count = 0;
		uint32_t cq;

		/*
		 *  for parallel, just send to next available CQ in round-robin
		 * fashion. So scan for an available CQ. If all CQs are full
		 * just return and move on to next QID
		 */
		do {
			if (++cq_check_count > qid->cq_num_mapped_cqs)
				goto exit;
			if (cq_idx >= qid->cq_num_mapped_cqs)
				cq_idx = 0;
			cq = qid->cq_map[cq_idx++];

		} while (sw->ports[cq].inflights == SW_PORT_HIST_LIST ||
				rte_event_ring_free_count(
					sw->ports[cq].cq_worker_ring) == 0);

		struct sw_port *p = &sw->ports[cq];
		if (sw->cq_ring_space[cq] == 0 ||
				p->inflights == SW_PORT_HIST_LIST)
			break;

		sw->cq_ring_space[cq]--;

		qid->stats.tx_pkts++;

		const int head = (p->hist_head & (SW_PORT_HIST_LIST-1));
		p->hist_list[head].fid = SW_HASH_FLOWID(qe->flow_id);
		p->hist_list[head].qid = qid_id;

		if (keep_order)
			rob_ring_dequeue(qid->reorder_buffer_freelist,
					(void *)&p->hist_list[head].rob_entry);

		sw->ports[cq].cq_buf[sw->ports[cq].cq_buf_count++] = *qe;
		iq_pop(sw, &qid->iq[iq_num]);

		rte_compiler_barrier();
		p->inflights++;
		p->stats.tx_pkts++;
		p->hist_head++;
	}
exit:
	qid->cq_next_tx = cq_idx;
	return i;
}

static uint32_t
sw_schedule_dir_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
		uint32_t iq_num, unsigned int count __rte_unused)
{
	uint32_t cq_id = qid->cq_map[0];
	struct sw_port *port = &sw->ports[cq_id];

	/* get max burst enq size for cq_ring */
	uint32_t count_free = sw->cq_ring_space[cq_id];
	if (count_free == 0)
		return 0;

	/* burst dequeue from the QID IQ ring */
	struct sw_iq *iq = &qid->iq[iq_num];
	uint32_t ret = iq_dequeue_burst(sw, iq,
			&port->cq_buf[port->cq_buf_count], count_free);
	port->cq_buf_count += ret;

	/* Update QID, Port and Total TX stats */
	qid->stats.tx_pkts += ret;
	port->stats.tx_pkts += ret;

	/* Subtract credits from cached value */
	sw->cq_ring_space[cq_id] -= ret;

	return ret;
}

static uint32_t
sw_schedule_qid_to_cq(struct sw_evdev *sw)
{
	uint32_t pkts = 0;
	uint32_t qid_idx;

	sw->sched_cq_qid_called++;

	for (qid_idx = 0; qid_idx < sw->qid_count; qid_idx++) {
		struct sw_qid *qid = sw->qids_prioritized[qid_idx];

		int type = qid->type;
		int iq_num = PKT_MASK_TO_IQ(qid->iq_pkt_mask);

		/* zero mapped CQs indicates directed */
		if (iq_num >= SW_IQS_MAX || qid->cq_num_mapped_cqs == 0)
			continue;

		uint32_t pkts_done = 0;
		uint32_t count = iq_count(&qid->iq[iq_num]);

		if (count >= sw->sched_min_burst) {
			if (type == SW_SCHED_TYPE_DIRECT)
				pkts_done += sw_schedule_dir_to_cq(sw, qid,
						iq_num, count);
			else if (type == RTE_SCHED_TYPE_ATOMIC)
				pkts_done += sw_schedule_atomic_to_cq(sw, qid,
						iq_num, count);
			else
				pkts_done += sw_schedule_parallel_to_cq(sw, qid,
						iq_num, count,
						type == RTE_SCHED_TYPE_ORDERED);
		}

		/* Check if the IQ that was polled is now empty, and unset it
		 * in the IQ mask if its empty.
		 */
		int all_done = (pkts_done == count);

		qid->iq_pkt_mask &= ~(all_done << (iq_num));
		pkts += pkts_done;
	}

	return pkts;
}

/* This function will perform re-ordering of packets, and injecting into
 * the appropriate QID IQ. As LB and DIR QIDs are in the same array, but *NOT*
 * contiguous in that array, this function accepts a "range" of QIDs to scan.
 */
static uint16_t
sw_schedule_reorder(struct sw_evdev *sw, int qid_start, int qid_end)
{
	/* Perform egress reordering */
	struct rte_event *qe;
	uint32_t pkts_iter = 0;

	for (; qid_start < qid_end; qid_start++) {
		struct sw_qid *qid = &sw->qids[qid_start];
		unsigned int i, num_entries_in_use;

		if (qid->type != RTE_SCHED_TYPE_ORDERED)
			continue;

		num_entries_in_use = rob_ring_free_count(
					qid->reorder_buffer_freelist);

		if (num_entries_in_use < sw->sched_min_burst)
			num_entries_in_use = 0;

		for (i = 0; i < num_entries_in_use; i++) {
			struct reorder_buffer_entry *entry;
			int j;

			entry = &qid->reorder_buffer[qid->reorder_buffer_index];

			if (!entry->ready)
				break;

			for (j = 0; j < entry->num_fragments; j++) {
				uint16_t dest_qid;
				uint16_t dest_iq;

				int idx = entry->fragment_index + j;
				qe = &entry->fragments[idx];

				dest_qid = qe->queue_id;
				dest_iq  = PRIO_TO_IQ(qe->priority);

				if (dest_qid >= sw->qid_count) {
					sw->stats.rx_dropped++;
					continue;
				}

				pkts_iter++;

				struct sw_qid *q = &sw->qids[dest_qid];
				struct sw_iq *iq = &q->iq[dest_iq];

				/* we checked for space above, so enqueue must
				 * succeed
				 */
				iq_enqueue(sw, iq, qe);
				q->iq_pkt_mask |= (1 << (dest_iq));
				q->iq_pkt_count[dest_iq]++;
				q->stats.rx_pkts++;
			}

			entry->ready = (j != entry->num_fragments);
			entry->num_fragments -= j;
			entry->fragment_index += j;

			if (!entry->ready) {
				entry->fragment_index = 0;

				rob_ring_enqueue(
						qid->reorder_buffer_freelist,
						entry);

				qid->reorder_buffer_index++;
				qid->reorder_buffer_index %= qid->window_size;
			}
		}
	}
	return pkts_iter;
}

static __rte_always_inline void
sw_refill_pp_buf(struct sw_evdev *sw, struct sw_port *port)
{
	RTE_SET_USED(sw);
	struct rte_event_ring *worker = port->rx_worker_ring;
	port->pp_buf_start = 0;
	port->pp_buf_count = rte_event_ring_dequeue_burst(worker, port->pp_buf,
			sw->sched_deq_burst_size, NULL);
}

static __rte_always_inline uint32_t
__pull_port_lb(struct sw_evdev *sw, uint32_t port_id, int allow_reorder)
{
	static struct reorder_buffer_entry dummy_rob;
	uint32_t pkts_iter = 0;
	struct sw_port *port = &sw->ports[port_id];

	/* If shadow ring has 0 pkts, pull from worker ring */
	if (!sw->refill_once_per_iter && port->pp_buf_count == 0)
		sw_refill_pp_buf(sw, port);

	while (port->pp_buf_count) {
		const struct rte_event *qe = &port->pp_buf[port->pp_buf_start];
		struct sw_hist_list_entry *hist_entry = NULL;
		uint8_t flags = qe->op;
		const uint16_t eop = !(flags & QE_FLAG_NOT_EOP);
		int needs_reorder = 0;
		/* if no-reordering, having PARTIAL == NEW */
		if (!allow_reorder && !eop)
			flags = QE_FLAG_VALID;

		/*
		 * if we don't have space for this packet in an IQ,
		 * then move on to next queue. Technically, for a
		 * packet that needs reordering, we don't need to check
		 * here, but it simplifies things not to special-case
		 */
		uint32_t iq_num = PRIO_TO_IQ(qe->priority);
		struct sw_qid *qid = &sw->qids[qe->queue_id];

		/* now process based on flags. Note that for directed
		 * queues, the enqueue_flush masks off all but the
		 * valid flag. This makes FWD and PARTIAL enqueues just
		 * NEW type, and makes DROPS no-op calls.
		 */
		if ((flags & QE_FLAG_COMPLETE) && port->inflights > 0) {
			const uint32_t hist_tail = port->hist_tail &
					(SW_PORT_HIST_LIST - 1);

			hist_entry = &port->hist_list[hist_tail];
			const uint32_t hist_qid = hist_entry->qid;
			const uint32_t hist_fid = hist_entry->fid;

			struct sw_fid_t *fid =
				&sw->qids[hist_qid].fids[hist_fid];
			fid->pcount -= eop;
			if (fid->pcount == 0)
				fid->cq = -1;

			if (allow_reorder) {
				/* set reorder ready if an ordered QID */
				uintptr_t rob_ptr =
					(uintptr_t)hist_entry->rob_entry;
				const uintptr_t valid = (rob_ptr != 0);
				needs_reorder = valid;
				rob_ptr |=
					((valid - 1) & (uintptr_t)&dummy_rob);
				struct reorder_buffer_entry *tmp_rob_ptr =
					(struct reorder_buffer_entry *)rob_ptr;
				tmp_rob_ptr->ready = eop * needs_reorder;
			}

			port->inflights -= eop;
			port->hist_tail += eop;
		}
		if (flags & QE_FLAG_VALID) {
			port->stats.rx_pkts++;

			if (allow_reorder && needs_reorder) {
				struct reorder_buffer_entry *rob_entry =
						hist_entry->rob_entry;

				hist_entry->rob_entry = NULL;
				/* Although fragmentation not currently
				 * supported by eventdev API, we support it
				 * here. Open: How do we alert the user that
				 * they've exceeded max frags?
				 */
				int num_frag = rob_entry->num_fragments;
				if (num_frag == SW_FRAGMENTS_MAX)
					sw->stats.rx_dropped++;
				else {
					int idx = rob_entry->num_fragments++;
					rob_entry->fragments[idx] = *qe;
				}
				goto end_qe;
			}

			/* Use the iq_num from above to push the QE
			 * into the qid at the right priority
			 */

			qid->iq_pkt_mask |= (1 << (iq_num));
			iq_enqueue(sw, &qid->iq[iq_num], qe);
			qid->iq_pkt_count[iq_num]++;
			qid->stats.rx_pkts++;
			pkts_iter++;
		}

end_qe:
		port->pp_buf_start++;
		port->pp_buf_count--;
	} /* while (avail_qes) */

	return pkts_iter;
}

static uint32_t
sw_schedule_pull_port_lb(struct sw_evdev *sw, uint32_t port_id)
{
	return __pull_port_lb(sw, port_id, 1);
}

static uint32_t
sw_schedule_pull_port_no_reorder(struct sw_evdev *sw, uint32_t port_id)
{
	return __pull_port_lb(sw, port_id, 0);
}

static uint32_t
sw_schedule_pull_port_dir(struct sw_evdev *sw, uint32_t port_id)
{
	uint32_t pkts_iter = 0;
	struct sw_port *port = &sw->ports[port_id];

	/* If shadow ring has 0 pkts, pull from worker ring */
	if (!sw->refill_once_per_iter && port->pp_buf_count == 0)
		sw_refill_pp_buf(sw, port);

	while (port->pp_buf_count) {
		const struct rte_event *qe = &port->pp_buf[port->pp_buf_start];
		uint8_t flags = qe->op;

		if ((flags & QE_FLAG_VALID) == 0)
			goto end_qe;

		uint32_t iq_num = PRIO_TO_IQ(qe->priority);
		struct sw_qid *qid = &sw->qids[qe->queue_id];
		struct sw_iq *iq = &qid->iq[iq_num];

		port->stats.rx_pkts++;

		/* Use the iq_num from above to push the QE
		 * into the qid at the right priority
		 */
		qid->iq_pkt_mask |= (1 << (iq_num));
		iq_enqueue(sw, iq, qe);
		qid->iq_pkt_count[iq_num]++;
		qid->stats.rx_pkts++;
		pkts_iter++;

end_qe:
		port->pp_buf_start++;
		port->pp_buf_count--;
	} /* while port->pp_buf_count */

	return pkts_iter;
}

void
sw_event_schedule(struct rte_eventdev *dev)
{
	struct sw_evdev *sw = sw_pmd_priv(dev);
	uint32_t in_pkts, out_pkts;
	uint32_t out_pkts_total = 0, in_pkts_total = 0;
	int32_t sched_quanta = sw->sched_quanta;
	uint32_t i;

	sw->sched_called++;
	if (unlikely(!sw->started))
		return;

	do {
		uint32_t in_pkts_this_iteration = 0;

		/* Pull from rx_ring for ports */
		do {
			in_pkts = 0;
			for (i = 0; i < sw->port_count; i++) {
				/* ack the unlinks in progress as done */
				if (sw->ports[i].unlinks_in_progress)
					sw->ports[i].unlinks_in_progress = 0;

				if (sw->ports[i].is_directed)
					in_pkts += sw_schedule_pull_port_dir(sw, i);
				else if (sw->ports[i].num_ordered_qids > 0)
					in_pkts += sw_schedule_pull_port_lb(sw, i);
				else
					in_pkts += sw_schedule_pull_port_no_reorder(sw, i);
			}

			/* QID scan for re-ordered */
			in_pkts += sw_schedule_reorder(sw, 0,
					sw->qid_count);
			in_pkts_this_iteration += in_pkts;
		} while (in_pkts > 4 &&
				(int)in_pkts_this_iteration < sched_quanta);

		out_pkts = sw_schedule_qid_to_cq(sw);
		out_pkts_total += out_pkts;
		in_pkts_total += in_pkts_this_iteration;

		if (in_pkts == 0 && out_pkts == 0)
			break;
	} while ((int)out_pkts_total < sched_quanta);

	sw->stats.tx_pkts += out_pkts_total;
	sw->stats.rx_pkts += in_pkts_total;

	sw->sched_no_iq_enqueues += (in_pkts_total == 0);
	sw->sched_no_cq_enqueues += (out_pkts_total == 0);

	/* push all the internal buffered QEs in port->cq_ring to the
	 * worker cores: aka, do the ring transfers batched.
	 */
	int no_enq = 1;
	for (i = 0; i < sw->port_count; i++) {
		struct sw_port *port = &sw->ports[i];
		struct rte_event_ring *worker = port->cq_worker_ring;

		/* If shadow ring has 0 pkts, pull from worker ring */
		if (sw->refill_once_per_iter && port->pp_buf_count == 0)
			sw_refill_pp_buf(sw, port);

		if (port->cq_buf_count >= sw->sched_min_burst) {
			rte_event_ring_enqueue_burst(worker,
					port->cq_buf,
					port->cq_buf_count,
					&sw->cq_ring_space[i]);
			port->cq_buf_count = 0;
			no_enq = 0;
		} else {
			sw->cq_ring_space[i] =
					rte_event_ring_free_count(worker) -
					port->cq_buf_count;
		}
	}

	if (no_enq) {
		if (unlikely(sw->sched_flush_count > SCHED_NO_ENQ_CYCLE_FLUSH))
			sw->sched_min_burst = 1;
		else
			sw->sched_flush_count++;
	} else {
		if (sw->sched_flush_count)
			sw->sched_flush_count--;
		else
			sw->sched_min_burst = sw->sched_min_burst_size;
	}

}