summaryrefslogtreecommitdiffstats
path: root/VNFs/DPPD-PROX/handle_lat.c
blob: 8285686b754d179d79043227457a19fc25ddaa1c (plain)
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
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
/*
// Copyright (c) 2010-2017 Intel Corporation
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
*/

//#define LAT_DEBUG

#include <rte_cycles.h>
#include <stdio.h>
#include <math.h>

#include "handle_gen.h"
#include "prox_malloc.h"
#include "mbuf_utils.h"
#include "handle_lat.h"
#include "log.h"
#include "task_init.h"
#include "task_base.h"
#include "stats.h"
#include "lconf.h"
#include "quit.h"
#include "eld.h"
#include "prox_shared.h"
#include "prox_port_cfg.h"

#define DEFAULT_BUCKET_SIZE	10
#define ACCURACY_BUFFER_SIZE	64

struct lat_info {
	uint32_t rx_packet_index;
	uint64_t tx_packet_index;
	uint32_t tx_err;
	uint32_t rx_err;
	uint64_t rx_time;
	uint64_t tx_time;
	uint16_t port_queue_id;
#ifdef LAT_DEBUG
	uint16_t id_in_bulk;
	uint16_t bulk_size;
	uint64_t begin;
	uint64_t after;
	uint64_t before;
#endif
};

struct delayed_latency_entry {
	uint32_t rx_packet_id;
	uint32_t tx_packet_id;
	uint32_t packet_id;
	uint8_t generator_id;
	uint64_t pkt_rx_time;
	uint64_t pkt_tx_time;
	uint64_t rx_time_err;
};

static struct delayed_latency_entry *delayed_latency_get(struct delayed_latency_entry **delayed_latency_entries, uint8_t generator_id, uint32_t packet_id)
{
	struct delayed_latency_entry *delayed_latency_entry = &delayed_latency_entries[generator_id][packet_id % ACCURACY_BUFFER_SIZE];
	if (delayed_latency_entry->packet_id == packet_id)
		return delayed_latency_entry;
	else
		return NULL;
}

static struct delayed_latency_entry *delayed_latency_create(struct delayed_latency_entry **delayed_latency_entries, uint8_t generator_id, uint32_t packet_id)
{
	struct delayed_latency_entry *delayed_latency_entry = &delayed_latency_entries[generator_id][packet_id % ACCURACY_BUFFER_SIZE];
	delayed_latency_entry->packet_id = packet_id;
	return delayed_latency_entry;
}

struct rx_pkt_meta_data {
	uint8_t  *hdr;
	uint32_t pkt_tx_time;
	uint32_t bytes_after_in_bulk;
};

struct task_lat {
	struct task_base base;
	uint64_t limit;
	uint64_t rx_packet_index;
	uint64_t last_pkts_tsc;
	struct delayed_latency_entry **delayed_latency_entries;
	struct lat_info *latency_buffer;
	uint32_t latency_buffer_idx;
	uint32_t latency_buffer_size;
	uint64_t begin;
	uint16_t lat_pos;
	uint16_t unique_id_pos;
	uint16_t accur_pos;
	uint16_t sig_pos;
	uint32_t sig;
	volatile uint16_t use_lt; /* which lt to use, */
	volatile uint16_t using_lt; /* 0 or 1 depending on which of the 2 measurements are used */
	struct lat_test lt[2];
	struct lat_test *lat_test;
	uint32_t generator_count;
	uint16_t min_pkt_len;
	struct early_loss_detect *eld;
	struct rx_pkt_meta_data *rx_pkt_meta;
	uint64_t link_speed;
	// Following fields are only used when starting or stopping, not in general runtime
	uint64_t *prev_tx_packet_index;
	FILE *fp_rx;
	FILE *fp_tx;
	struct prox_port_cfg *port;
};
/* This function calculate the difference between rx and tx_time
 * Both values are uint32_t (see handle_lat_bulk)
 * rx time should be higher than tx_time...except every UINT32_MAX
 * cycles, when rx_time overflows.
 * As the return value is also uint32_t, returning (rx_time - tx_time)
 * is also fine when it overflows.
 */
static uint32_t diff_time(uint32_t rx_time, uint32_t tx_time)
{
	return rx_time - tx_time;
}

struct lat_test *task_lat_get_latency_meassurement(struct task_lat *task)
{
	if (task->use_lt == task->using_lt)
		return &task->lt[!task->using_lt];
	return NULL;
}

void task_lat_use_other_latency_meassurement(struct task_lat *task)
{
	task->use_lt = !task->using_lt;
}

static void task_lat_update_lat_test(struct task_lat *task)
{
	if (task->use_lt != task->using_lt) {
		task->using_lt = task->use_lt;
		task->lat_test = &task->lt[task->using_lt];
		task->lat_test->accuracy_limit_tsc = task->limit;
	}
}

static int compare_tx_time(const void *val1, const void *val2)
{
	const struct lat_info *ptr1 = val1;
	const struct lat_info *ptr2 = val2;

	return ptr1->tx_time > ptr2->tx_time ? 1 : -1;
}

static int compare_tx_packet_index(const void *val1, const void *val2)
{
	const struct lat_info *ptr1 = val1;
	const struct lat_info *ptr2 = val2;

	return ptr1->tx_packet_index > ptr2->tx_packet_index ? 1 : -1;
}

static void fix_latency_buffer_tx_packet_index(struct lat_info *lat, uint32_t count)
{
	uint32_t tx_packet_index, old_tx_packet_index = lat->tx_packet_index, n_overflow = 0;
	uint32_t small = UINT32_MAX >> 1;

	lat++;

	/* Buffer is sorted so far by RX time.
	 * We might have packets being reordered by SUT.
	 *     => consider small differences as re-order and big ones as overflow of tx_packet_index.
	 * Note that:
	 *	- overflow only happens if receiving and storing 4 billions packets...
	 *	- a absolute difference of less than 2 billion packets is not considered as an overflow
	 */
	for (uint32_t i = 1; i < count; i++) {
		tx_packet_index = lat->tx_packet_index;
		if (tx_packet_index > old_tx_packet_index) {
			if (tx_packet_index - old_tx_packet_index < small) {
				// The diff is small => increasing index count
			} else {
				// The diff is big => it is more likely that the previous packet was overflow
				n_overflow--;
			}
		} else {
			if (old_tx_packet_index - tx_packet_index < small) {
				// The diff is small => packet reorder
			} else {
				// The diff is big => it is more likely that this is an overflow
				n_overflow++;
			}
		}
		lat->tx_packet_index += ((uint64_t)UINT32_MAX + 1) * n_overflow;
		old_tx_packet_index = tx_packet_index;
		lat++;
	}
}

static void fix_latency_buffer_tx_time(struct lat_info *lat, uint32_t count)
{
	uint32_t tx_time, old_tx_time = lat->tx_time, n_overflow = 0;
	uint32_t small = UINT32_MAX >> 1;
	lat++;

	/*
	 * Same algorithm as above, but with time.
	 * Note that:
	 *	- overflow happens after 4 billions "cycles" (shifted by LATENCY_ACCURACY) = ~4sec
	 *	- a absolute difference up to 2 billion (shifted) cycles (~=2sec) is not considered as an overflow
	 *		=> algorithm does not work if receiving less than 1 packet every 2 seconds
	 */
	for (uint32_t i = 1; i < count; i++) {
		tx_time = lat->tx_time;
		if (tx_time > old_tx_time) {
			if (tx_time - old_tx_time > small) {
				n_overflow--;
			}
		} else {
			if (old_tx_time - tx_time > small) {
				n_overflow++;
			}
		}
		lat->tx_time += ((uint64_t)UINT32_MAX + 1) * n_overflow;
		old_tx_time = tx_time;
		lat++;
	}
}

static void task_lat_count_remaining_lost_packets(struct task_lat *task)
{
	struct lat_test *lat_test = task->lat_test;

	for (uint32_t j = 0; j < task->generator_count; j++) {
		struct early_loss_detect *eld = &task->eld[j];

		lat_test->lost_packets += early_loss_detect_count_remaining_loss(eld);
	}
}

static void task_lat_reset_eld(struct task_lat *task)
{
	for (uint32_t j = 0; j < task->generator_count; j++) {
		early_loss_detect_reset(&task->eld[j]);
	}
}

static uint64_t lat_latency_buffer_get_min_tsc(struct task_lat *task)
{
	uint64_t min_tsc = UINT64_MAX;

	for (uint32_t i = 0; i < task->latency_buffer_idx; i++) {
		if (min_tsc > task->latency_buffer[i].tx_time)
			min_tsc = task->latency_buffer[i].tx_time;
	}

	return min_tsc << LATENCY_ACCURACY;
}

static uint64_t lat_info_get_lat_tsc(struct lat_info *lat_info)
{
	uint64_t lat = diff_time(lat_info->rx_time, lat_info->tx_time);

	return lat << LATENCY_ACCURACY;
}

static uint64_t lat_info_get_tx_err_tsc(const struct lat_info *lat_info)
{
	return ((uint64_t)lat_info->tx_err) << LATENCY_ACCURACY;
}

static uint64_t lat_info_get_rx_err_tsc(const struct lat_info *lat_info)
{
	return ((uint64_t)lat_info->rx_err) << LATENCY_ACCURACY;
}

static uint64_t lat_info_get_rx_tsc(const struct lat_info *lat_info)
{
	return ((uint64_t)lat_info->rx_time) << LATENCY_ACCURACY;
}

static uint64_t lat_info_get_tx_tsc(const struct lat_info *lat_info)
{
	return ((uint64_t)lat_info->tx_time) << LATENCY_ACCURACY;
}

static void lat_write_latency_to_file(struct task_lat *task)
{
	uint64_t min_tsc;
	uint64_t n_loss;

	min_tsc = lat_latency_buffer_get_min_tsc(task);

	// Dumping all packet statistics
	fprintf(task->fp_rx, "Latency stats for %u packets, ordered by rx time\n", task->latency_buffer_idx);
	fprintf(task->fp_rx, "rx index; queue; tx index; lat (nsec);tx time;\n");
	for (uint32_t i = 0; i < task->latency_buffer_idx ; i++) {
		struct lat_info *lat_info = &task->latency_buffer[i];
		uint64_t lat_tsc = lat_info_get_lat_tsc(lat_info);
		uint64_t rx_tsc = lat_info_get_rx_tsc(lat_info);
		uint64_t tx_tsc = lat_info_get_tx_tsc(lat_info);

		fprintf(task->fp_rx, "%u;%u;%lu;%lu;%lu;%lu\n",
			lat_info->rx_packet_index,
			lat_info->port_queue_id,
			lat_info->tx_packet_index,
			tsc_to_nsec(lat_tsc),
			tsc_to_nsec(rx_tsc - min_tsc),
			tsc_to_nsec(tx_tsc - min_tsc));
	}

	// To detect dropped packets, we need to sort them based on TX
	if (task->unique_id_pos) {
		plogx_info("Adapting tx_packet_index\n");
		fix_latency_buffer_tx_packet_index(task->latency_buffer, task->latency_buffer_idx);
		plogx_info("Sorting packets based on tx_packet_index\n");
		qsort (task->latency_buffer, task->latency_buffer_idx, sizeof(struct lat_info), compare_tx_packet_index);
		plogx_info("Sorted packets based on packet_index\n");
	} else {
		plogx_info("Adapting tx_time\n");
		fix_latency_buffer_tx_time(task->latency_buffer, task->latency_buffer_idx);
		plogx_info("Sorting packets based on tx_time\n");
		qsort (task->latency_buffer, task->latency_buffer_idx, sizeof(struct lat_info), compare_tx_time);
		plogx_info("Sorted packets based on packet_time\n");
	}

	// A packet is marked as dropped if 2 packets received from the same queue are not consecutive
	fprintf(task->fp_tx, "Latency stats for %u packets, sorted by tx time\n", task->latency_buffer_idx);
	fprintf(task->fp_tx, "queue;tx index; rx index; lat (nsec);tx time; rx time; tx_err;rx_err\n");

	for (uint32_t i = 0; i < task->generator_count;i++)
		task->prev_tx_packet_index[i] = -1;

	for (uint32_t i = 0; i < task->latency_buffer_idx; i++) {
		struct lat_info *lat_info = &task->latency_buffer[i];
		uint64_t lat_tsc = lat_info_get_lat_tsc(lat_info);
		uint64_t tx_err_tsc = lat_info_get_tx_err_tsc(lat_info);
		uint64_t rx_err_tsc = lat_info_get_rx_err_tsc(lat_info);
		uint64_t rx_tsc = lat_info_get_rx_tsc(lat_info);
		uint64_t tx_tsc = lat_info_get_tx_tsc(lat_info);

		/* Packet n + ACCURACY_BUFFER_SIZE delivers the TX error for packet n,
		   hence the last ACCURACY_BUFFER_SIZE packets do no have TX error. */
		if (i + ACCURACY_BUFFER_SIZE >= task->latency_buffer_idx) {
			tx_err_tsc = 0;
		}

		if (lat_info->port_queue_id >= task->generator_count) {
			plog_err("Unexpected generator id %u for packet %lu - skipping packet\n",
				lat_info->port_queue_id, lat_info->tx_packet_index);
			continue;
		}
		// Log dropped packet
		n_loss = lat_info->tx_packet_index - task->prev_tx_packet_index[lat_info->port_queue_id] - 1;
		if (n_loss)
			fprintf(task->fp_tx, "===> %u;%lu;0;0;0;0;0;0 lost %lu packets <===\n",
				lat_info->port_queue_id,
				lat_info->tx_packet_index - n_loss, n_loss);
		// Log next packet
		fprintf(task->fp_tx, "%u;%lu;%u;%lu;%lu;%lu;%lu;%lu",
			lat_info->port_queue_id,
			lat_info->tx_packet_index,
			lat_info->rx_packet_index,
			tsc_to_nsec(lat_tsc),
			tsc_to_nsec(tx_tsc - min_tsc),
			tsc_to_nsec(rx_tsc - min_tsc),
			tsc_to_nsec(tx_err_tsc),
			tsc_to_nsec(rx_err_tsc));
#ifdef LAT_DEBUG
		fprintf(task->fp_tx, ";%u from %u;%lu;%lu;%lu",
			lat_info->id_in_bulk,
			lat_info->bulk_size,
			tsc_to_nsec(lat_info->begin - min_tsc),
			tsc_to_nsec(lat_info->before - min_tsc),
			tsc_to_nsec(lat_info->after - min_tsc));
#endif
		fprintf(task->fp_tx, "\n");
		task->prev_tx_packet_index[lat_info->port_queue_id] = lat_info->tx_packet_index;
	}
	fflush(task->fp_rx);
	fflush(task->fp_tx);
	task->latency_buffer_idx = 0;
}

static void lat_stop(struct task_base *tbase)
{
	struct task_lat *task = (struct task_lat *)tbase;

	if (task->unique_id_pos) {
		task_lat_count_remaining_lost_packets(task);
		task_lat_reset_eld(task);
	}
	if (task->latency_buffer)
		lat_write_latency_to_file(task);
}

#ifdef LAT_DEBUG
static void task_lat_store_lat_debug(struct task_lat *task, uint32_t rx_packet_index, uint32_t id_in_bulk, uint32_t bulk_size)
{
	struct lat_info *lat_info = &task->latency_buffer[rx_packet_index];

	lat_info->bulk_size = bulk_size;
	lat_info->id_in_bulk = id_in_bulk;
	lat_info->begin = task->begin;
	lat_info->before = task->base.aux->tsc_rx.before;
	lat_info->after = task->base.aux->tsc_rx.after;
}
#endif

static void task_lat_store_lat_buf(struct task_lat *task, uint64_t rx_packet_index, uint64_t rx_time, uint64_t tx_time, uint64_t rx_err, uint64_t tx_err, uint32_t packet_id, uint8_t generator_id)
{
	struct lat_info *lat_info;

	/* If unique_id_pos is specified then latency is stored per
	   packet being sent. Lost packets are detected runtime, and
	   latency stored for those packets will be 0 */
	lat_info = &task->latency_buffer[task->latency_buffer_idx++];
	lat_info->rx_packet_index = rx_packet_index;
	lat_info->tx_packet_index = packet_id;
	lat_info->port_queue_id = generator_id;
	lat_info->rx_time = rx_time;
	lat_info->tx_time = tx_time;
	lat_info->rx_err = rx_err;
	lat_info->tx_err = tx_err;
}

static uint32_t task_lat_early_loss_detect(struct task_lat *task, uint32_t packet_id, uint8_t generator_id)
{
	struct early_loss_detect *eld = &task->eld[generator_id];
	return early_loss_detect_add(eld, packet_id);
}

static uint64_t tsc_extrapolate_backward(uint64_t link_speed, uint64_t tsc_from, uint64_t bytes, uint64_t tsc_minimum)
{
	uint64_t tsc = tsc_from - (rte_get_tsc_hz()*bytes)/link_speed;
	if (likely(tsc > tsc_minimum))
		return tsc;
	else
		return tsc_minimum;
}

static void lat_test_histogram_add(struct lat_test *lat_test, uint64_t lat_tsc)
{
	uint64_t bucket_id = (lat_tsc >> lat_test->bucket_size);
	size_t bucket_count = sizeof(lat_test->buckets)/sizeof(lat_test->buckets[0]);

	bucket_id = bucket_id < bucket_count? bucket_id : bucket_count;
	lat_test->buckets[bucket_id]++;
}

static void lat_test_add_lost(struct lat_test *lat_test, uint64_t lost_packets)
{
	lat_test->lost_packets += lost_packets;
}

static void lat_test_add_latency(struct lat_test *lat_test, uint64_t lat_tsc, uint64_t error)
{
	if (error > lat_test->accuracy_limit_tsc)
		return;
	lat_test->tot_pkts++;

	lat_test->tot_lat += lat_tsc;
	lat_test->tot_lat_error += error;

	/* (a +- b)^2 = a^2 +- (2ab + b^2) */
	lat_test->var_lat += lat_tsc * lat_tsc;
	lat_test->var_lat_error += 2 * lat_tsc * error;
	lat_test->var_lat_error += error * error;

	if (lat_tsc > lat_test->max_lat) {
		lat_test->max_lat = lat_tsc;
		lat_test->max_lat_error = error;
	}
	if (lat_tsc < lat_test->min_lat) {
		lat_test->min_lat = lat_tsc;
		lat_test->min_lat_error = error;
	}

#ifdef LATENCY_HISTOGRAM
	lat_test_histogram_add(lat_test, lat_tsc);
#endif
}

static int task_lat_can_store_latency(struct task_lat *task)
{
	return task->latency_buffer_idx < task->latency_buffer_size;
}

static void task_lat_store_lat(struct task_lat *task, uint64_t rx_packet_index, uint64_t rx_time, uint64_t tx_time, uint64_t rx_error, uint64_t tx_error, uint32_t packet_id, uint8_t generator_id)
{
	uint32_t lat_tsc = diff_time(rx_time, tx_time) << LATENCY_ACCURACY;

	lat_test_add_latency(task->lat_test, lat_tsc, rx_error + tx_error);

	if (task_lat_can_store_latency(task)) {
		task_lat_store_lat_buf(task, rx_packet_index, rx_time, tx_time, rx_error, tx_error, packet_id, generator_id);
	}
}

static int handle_lat_bulk(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
{
	struct task_lat *task = (struct task_lat *)tbase;

	// If link is down, link_speed is 0
	if (unlikely(task->link_speed == 0)) {
		if (task->port && task->port->link_speed != 0) {
			task->link_speed = task->port->link_speed * 125000L;
			plog_info("\tPort %u: link speed is %ld Mbps\n",
				(uint8_t)(task->port - prox_port_cfg), 8 * task->link_speed / 1000000);
		} else if (n_pkts) {
			return task->base.tx_pkt(&task->base, mbufs, n_pkts, NULL);
		} else {
			return 0;
		}
	}

	if (n_pkts == 0) {
		task->begin = tbase->aux->tsc_rx.before;
		return 0;
	}

	task_lat_update_lat_test(task);

	// Remember those packets with bad length or bad signature
	uint64_t pkt_bad_len_sig[(MAX_RX_PKT_ALL + 63) / 64];
#define BIT64_SET(a64, bit)	a64[bit / 64] |=  (((uint64_t)1) << (bit & 63))
#define BIT64_CLR(a64, bit)	a64[bit / 64] &= ~(((uint64_t)1) << (bit & 63))
#define BIT64_TEST(a64, bit)	a64[bit / 64]  &  (((uint64_t)1) << (bit & 63))

	/* Go once through all received packets and read them.  If
	   packet has just been modified by another core, the cost of
	   latency will be partialy amortized though the bulk size */
	for (uint16_t j = 0; j < n_pkts; ++j) {
		struct rte_mbuf *mbuf = mbufs[j];
		task->rx_pkt_meta[j].hdr = rte_pktmbuf_mtod(mbuf, uint8_t *);

		// Remember those packets which are too short to hold the values that we expect
		if (unlikely(rte_pktmbuf_pkt_len(mbuf) < task->min_pkt_len))
			BIT64_SET(pkt_bad_len_sig, j);
		else
			BIT64_CLR(pkt_bad_len_sig, j);
	}

	if (task->sig_pos) {
		for (uint16_t j = 0; j < n_pkts; ++j) {
			if (unlikely(BIT64_TEST(pkt_bad_len_sig, j)))
				continue;
			// Remember those packets with bad signature
			if (likely(*(uint32_t *)(task->rx_pkt_meta[j].hdr + task->sig_pos) == task->sig))
				task->rx_pkt_meta[j].pkt_tx_time = *(uint32_t *)(task->rx_pkt_meta[j].hdr + task->lat_pos);
			else
				BIT64_SET(pkt_bad_len_sig, j);
		}
	} else {
		for (uint16_t j = 0; j < n_pkts; ++j) {
			if (unlikely(BIT64_TEST(pkt_bad_len_sig, j)))
				continue;
			task->rx_pkt_meta[j].pkt_tx_time = *(uint32_t *)(task->rx_pkt_meta[j].hdr + task->lat_pos);
		}
	}

	uint32_t bytes_total_in_bulk = 0;
	// Find RX time of first packet, for RX accuracy
	for (uint16_t j = 0; j < n_pkts; ++j) {
		uint16_t flipped = n_pkts - 1 - j;

		task->rx_pkt_meta[flipped].bytes_after_in_bulk = bytes_total_in_bulk;
		bytes_total_in_bulk += mbuf_wire_size(mbufs[flipped]);
	}

	const uint64_t rx_tsc = tbase->aux->tsc_rx.after;

	uint64_t rx_time_err;
	uint64_t pkt_rx_time64 = tsc_extrapolate_backward(task->link_speed, rx_tsc, task->rx_pkt_meta[0].bytes_after_in_bulk, task->last_pkts_tsc) >> LATENCY_ACCURACY;
	if (unlikely((task->begin >> LATENCY_ACCURACY) > pkt_rx_time64)) {
		// Extrapolation went up to BEFORE begin => packets were stuck in the NIC but we were not seeing them
		rx_time_err = pkt_rx_time64 - (task->last_pkts_tsc >> LATENCY_ACCURACY);
	} else {
		rx_time_err = pkt_rx_time64 - (task->begin >> LATENCY_ACCURACY);
	}

	for (uint16_t j = 0; j < n_pkts; ++j) {
		// Used to display % of packets within accuracy limit vs. total number of packets (used_col)
		task->lat_test->tot_all_pkts++;

		// Skip those packets with bad length or bad signature
		if (unlikely(BIT64_TEST(pkt_bad_len_sig, j)))
			continue;

		struct rx_pkt_meta_data *rx_pkt_meta = &task->rx_pkt_meta[j];
		uint8_t *hdr = rx_pkt_meta->hdr;

		uint32_t pkt_rx_time = tsc_extrapolate_backward(task->link_speed, rx_tsc, rx_pkt_meta->bytes_after_in_bulk, task->last_pkts_tsc) >> LATENCY_ACCURACY;
		uint32_t pkt_tx_time = rx_pkt_meta->pkt_tx_time;

		uint8_t generator_id;
		uint32_t packet_id;
		if (task->unique_id_pos) {
			struct unique_id *unique_id = (struct unique_id *)(hdr + task->unique_id_pos);
			unique_id_get(unique_id, &generator_id, &packet_id);

			if (unlikely(generator_id >= task->generator_count)) {
				/* No need to remember unexpected packet at this stage
				BIT64_SET(pkt_bad_len_sig, j);
				*/
				// Skip unexpected packet
				continue;
			}

			lat_test_add_lost(task->lat_test, task_lat_early_loss_detect(task, packet_id, generator_id));
		} else {
			generator_id = 0;
			packet_id = task->rx_packet_index;
		}

		/* If accuracy is enabled, latency is reported with a
		   delay of ACCURACY_BUFFER_SIZE packets since the generator puts the
		   accuracy for packet N into packet N + ACCURACY_BUFFER_SIZE. The delay
		   ensures that all reported latencies have both rx
		   and tx error. */
		if (task->accur_pos) {
			uint32_t tx_time_err = *(uint32_t *)(hdr + task->accur_pos);

			struct delayed_latency_entry *delayed_latency_entry = delayed_latency_get(task->delayed_latency_entries, generator_id, packet_id - ACCURACY_BUFFER_SIZE);

			if (delayed_latency_entry) {
				task_lat_store_lat(task,
						   delayed_latency_entry->rx_packet_id,
						   delayed_latency_entry->pkt_rx_time,
						   delayed_latency_entry->pkt_tx_time,
						   delayed_latency_entry->rx_time_err,
						   tx_time_err,
						   delayed_latency_entry->tx_packet_id,
						   delayed_latency_entry->generator_id);
			}

			delayed_latency_entry = delayed_latency_create(task->delayed_latency_entries, generator_id, packet_id);
			delayed_latency_entry->pkt_rx_time = pkt_rx_time;
			delayed_latency_entry->pkt_tx_time = pkt_tx_time;
			delayed_latency_entry->rx_time_err = rx_time_err;
			delayed_latency_entry->rx_packet_id = task->rx_packet_index;
			delayed_latency_entry->tx_packet_id = packet_id;
			delayed_latency_entry->generator_id = generator_id;
		} else {
			task_lat_store_lat(task, task->rx_packet_index, pkt_rx_time, pkt_tx_time, 0, 0, packet_id, generator_id);
		}

		// Bad/unexpected packets do not need to be indexed
		task->rx_packet_index++;
	}

	task->begin = tbase->aux->tsc_rx.before;
	task->last_pkts_tsc = tbase->aux->tsc_rx.after;

	return task->base.tx_pkt(&task->base, mbufs, n_pkts, NULL);
}

static void init_task_lat_latency_buffer(struct task_lat *task, uint32_t core_id)
{
	const int socket_id = rte_lcore_to_socket_id(core_id);
	char name[256];
	size_t latency_buffer_mem_size = 0;

	if (task->latency_buffer_size > UINT32_MAX - MAX_RING_BURST)
		task->latency_buffer_size = UINT32_MAX - MAX_RING_BURST;

	latency_buffer_mem_size = sizeof(struct lat_info) * task->latency_buffer_size;

	task->latency_buffer = prox_zmalloc(latency_buffer_mem_size, socket_id);
	PROX_PANIC(task->latency_buffer == NULL, "Failed to allocate %zu kbytes for latency_buffer\n", latency_buffer_mem_size / 1024);

	sprintf(name, "latency.rx_%u.txt", core_id);
	task->fp_rx = fopen(name, "w+");
	PROX_PANIC(task->fp_rx == NULL, "Failed to open %s\n", name);

	sprintf(name, "latency.tx_%u.txt", core_id);
	task->fp_tx = fopen(name, "w+");
	PROX_PANIC(task->fp_tx == NULL, "Failed to open %s\n", name);

	task->prev_tx_packet_index = prox_zmalloc(sizeof(task->prev_tx_packet_index[0]) * task->generator_count, socket_id);
	PROX_PANIC(task->prev_tx_packet_index == NULL, "Failed to allocated prev_tx_packet_index\n");
}

static void task_init_generator_count(struct task_lat *task)
{
	uint8_t *generator_count = prox_sh_find_system("generator_count");

	if (generator_count == NULL) {
		task->generator_count = 1;
		plog_info("\tNo generators found, hard-coding to %u generators\n", task->generator_count);
	} else
		task->generator_count = *generator_count;
	plog_info("\tLatency using %u generators\n", task->generator_count);
}

static void task_lat_init_eld(struct task_lat *task, uint8_t socket_id)
{
	size_t eld_mem_size;

	eld_mem_size = sizeof(task->eld[0]) * task->generator_count;
	task->eld = prox_zmalloc(eld_mem_size, socket_id);
	PROX_PANIC(task->eld == NULL, "Failed to allocate eld\n");
}

void task_lat_set_accuracy_limit(struct task_lat *task, uint32_t accuracy_limit_nsec)
{
	task->limit = nsec_to_tsc(accuracy_limit_nsec);
}

static void lat_start(struct task_base *tbase)
{
	struct task_lat *task = (struct task_lat *)tbase;

	if (task->port) {
		// task->port->link_speed reports the link speed in Mbps e.g. 40k for a 40 Gbps NIC.
		// task->link_speed reports link speed in Bytes per sec.
		// It can be 0 if link is down, and must hence be updated in fast path.
		task->link_speed = task->port->link_speed * 125000L;
		if (task->link_speed)
			plog_info("\tPort %u: link speed is %ld Mbps\n",
				(uint8_t)(task->port - prox_port_cfg), 8 * task->link_speed / 1000000);
		else
			plog_info("\tPort %u: link speed is %ld Mbps - link might be down\n",
				(uint8_t)(task->port - prox_port_cfg), 8 * task->link_speed / 1000000);
	}
}

static void init_task_lat(struct task_base *tbase, struct task_args *targ)
{
	struct task_lat *task = (struct task_lat *)tbase;
	const int socket_id = rte_lcore_to_socket_id(targ->lconf->id);

	task->lat_pos = targ->lat_pos;
	task->accur_pos = targ->accur_pos;
	task->sig_pos = targ->sig_pos;
	task->sig = targ->sig;

	task->unique_id_pos = targ->packet_id_pos;
	task->latency_buffer_size = targ->latency_buffer_size;

	PROX_PANIC(task->lat_pos == 0, "Missing 'lat pos' parameter in config file\n");
	uint16_t min_pkt_len = task->lat_pos + sizeof(uint32_t);
	if (task->unique_id_pos && (
		min_pkt_len < task->unique_id_pos + sizeof(struct unique_id)))
		min_pkt_len = task->unique_id_pos + sizeof(struct unique_id);
	if (task->accur_pos && (
		min_pkt_len < task->accur_pos + sizeof(uint32_t)))
		min_pkt_len = task->accur_pos + sizeof(uint32_t);
	if (task->sig_pos && (
		min_pkt_len < task->sig_pos + sizeof(uint32_t)))
		min_pkt_len = task->sig_pos + sizeof(uint32_t);
	task->min_pkt_len = min_pkt_len;

	task_init_generator_count(task);

	if (task->latency_buffer_size) {
		init_task_lat_latency_buffer(task, targ->lconf->id);
	}

	if (targ->bucket_size < DEFAULT_BUCKET_SIZE) {
		targ->bucket_size = DEFAULT_BUCKET_SIZE;
	}

	if (task->accur_pos) {
		task->delayed_latency_entries = prox_zmalloc(sizeof(*task->delayed_latency_entries) * task->generator_count , socket_id);
		PROX_PANIC(task->delayed_latency_entries == NULL, "Failed to allocate array for storing delayed latency entries\n");
		for (uint i = 0; i < task->generator_count; i++) {
			task->delayed_latency_entries[i] = prox_zmalloc(sizeof(**task->delayed_latency_entries) * ACCURACY_BUFFER_SIZE, socket_id);
			PROX_PANIC(task->delayed_latency_entries[i] == NULL, "Failed to allocate array for storing delayed latency entries\n");
		}
		if (task->unique_id_pos == 0) {
			/* When using accuracy feature, the accuracy from TX is written ACCURACY_BUFFER_SIZE packets later
			* We can only retrieve the good packet if a packet id is written to it.
			* Otherwise we will use the packet RECEIVED ACCURACY_BUFFER_SIZE packets ago which is OK if
			* packets are not re-ordered. If packets are re-ordered, then the matching between
			* the tx accuracy znd the latency is wrong.
			*/
			plog_warn("\tWhen accuracy feature is used, a unique id should ideally also be used\n");
		}
	}

	task->lt[0].bucket_size = targ->bucket_size - LATENCY_ACCURACY;
	task->lt[1].bucket_size = targ->bucket_size - LATENCY_ACCURACY;
        if (task->unique_id_pos) {
		task_lat_init_eld(task, socket_id);
		task_lat_reset_eld(task);
        }
	task->lat_test = &task->lt[task->using_lt];

	task_lat_set_accuracy_limit(task, targ->accuracy_limit_nsec);
	task->rx_pkt_meta = prox_zmalloc(MAX_RX_PKT_ALL * sizeof(*task->rx_pkt_meta), socket_id);
	PROX_PANIC(task->rx_pkt_meta == NULL, "unable to allocate memory to store RX packet meta data");

	task->link_speed = UINT64_MAX;
	if (targ->nb_rxports) {
		// task->port structure is only used while starting handle_lat to get the link_speed.
		// link_speed can not be quiried at init as the port has not been initialized yet.
		struct prox_port_cfg *port = &prox_port_cfg[targ->rx_port_queue[0].port];
		task->port = port;
	}
}

static struct task_init task_init_lat = {
	.mode_str = "lat",
	.init = init_task_lat,
	.handle = handle_lat_bulk,
	.start = lat_start,
	.stop = lat_stop,
	.flag_features = TASK_FEATURE_TSC_RX | TASK_FEATURE_RX_ALL | TASK_FEATURE_ZERO_RX | TASK_FEATURE_NEVER_DISCARDS,
	.size = sizeof(struct task_lat)
};

__attribute__((constructor)) static void reg_task_lat(void)
{
	reg_task(&task_init_lat);
}