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
|
/*
* Copyright (c) 2015 ARM Limited
* All rights reserved
*
* The license below extends only to copyright in the software and shall
* not be construed as granting a license to any other intellectual
* property including but not limited to intellectual property relating
* to a hardware implementation of the functionality of the software
* licensed hereunder. You may use the software subject to the license
* terms below provided that you ensure that this notice is replicated
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Authors: Gabor Dozsa
*/
/* @file
* The interface class for dist-gem5 simulations.
*/
#include "dev/net/dist_iface.hh"
#include <queue>
#include <thread>
#include "base/random.hh"
#include "base/trace.hh"
#include "debug/DistEthernet.hh"
#include "debug/DistEthernetPkt.hh"
#include "dev/net/etherpkt.hh"
#include "sim/sim_exit.hh"
#include "sim/sim_object.hh"
using namespace std;
DistIface::Sync *DistIface::sync = nullptr;
DistIface::SyncEvent *DistIface::syncEvent = nullptr;
unsigned DistIface::distIfaceNum = 0;
unsigned DistIface::recvThreadsNum = 0;
DistIface *DistIface::master = nullptr;
void
DistIface::Sync::init(Tick start_tick, Tick repeat_tick)
{
if (start_tick < firstAt) {
firstAt = start_tick;
inform("Next dist synchronisation tick is changed to %lu.\n", nextAt);
}
if (repeat_tick == 0)
panic("Dist synchronisation interval must be greater than zero");
if (repeat_tick < nextRepeat) {
nextRepeat = repeat_tick;
inform("Dist synchronisation interval is changed to %lu.\n",
nextRepeat);
}
}
DistIface::SyncSwitch::SyncSwitch(int num_nodes)
{
numNodes = num_nodes;
waitNum = num_nodes;
numExitReq = 0;
numCkptReq = 0;
doExit = false;
doCkpt = false;
firstAt = std::numeric_limits<Tick>::max();
nextAt = 0;
nextRepeat = std::numeric_limits<Tick>::max();
}
DistIface::SyncNode::SyncNode()
{
waitNum = 0;
needExit = ReqType::none;
needCkpt = ReqType::none;
doExit = false;
doCkpt = false;
firstAt = std::numeric_limits<Tick>::max();
nextAt = 0;
nextRepeat = std::numeric_limits<Tick>::max();
}
void
DistIface::SyncNode::run(bool same_tick)
{
std::unique_lock<std::mutex> sync_lock(lock);
Header header;
assert(waitNum == 0);
waitNum = DistIface::recvThreadsNum;
// initiate the global synchronisation
header.msgType = MsgType::cmdSyncReq;
header.sendTick = curTick();
header.syncRepeat = nextRepeat;
header.needCkpt = needCkpt;
if (needCkpt != ReqType::none)
needCkpt = ReqType::pending;
header.needExit = needExit;
if (needExit != ReqType::none)
needExit = ReqType::pending;
DistIface::master->sendCmd(header);
// now wait until all receiver threads complete the synchronisation
auto lf = [this]{ return waitNum == 0; };
cv.wait(sync_lock, lf);
// global synchronisation is done
assert(!same_tick || (nextAt == curTick()));
}
void
DistIface::SyncSwitch::run(bool same_tick)
{
std::unique_lock<std::mutex> sync_lock(lock);
Header header;
// Wait for the sync requests from the nodes
if (waitNum > 0) {
auto lf = [this]{ return waitNum == 0; };
cv.wait(sync_lock, lf);
}
assert(waitNum == 0);
assert(!same_tick || (nextAt == curTick()));
waitNum = numNodes;
// Complete the global synchronisation
header.msgType = MsgType::cmdSyncAck;
header.sendTick = nextAt;
header.syncRepeat = nextRepeat;
if (doCkpt || numCkptReq == numNodes) {
doCkpt = true;
header.needCkpt = ReqType::immediate;
numCkptReq = 0;
} else {
header.needCkpt = ReqType::none;
}
if (doExit || numExitReq == numNodes) {
doExit = true;
header.needExit = ReqType::immediate;
} else {
header.needExit = ReqType::none;
}
DistIface::master->sendCmd(header);
}
void
DistIface::SyncSwitch::progress(Tick send_tick,
Tick sync_repeat,
ReqType need_ckpt,
ReqType need_exit)
{
std::unique_lock<std::mutex> sync_lock(lock);
assert(waitNum > 0);
if (send_tick > nextAt)
nextAt = send_tick;
if (nextRepeat > sync_repeat)
nextRepeat = sync_repeat;
if (need_ckpt == ReqType::collective)
numCkptReq++;
else if (need_ckpt == ReqType::immediate)
doCkpt = true;
if (need_exit == ReqType::collective)
numExitReq++;
else if (need_exit == ReqType::immediate)
doExit = true;
waitNum--;
// Notify the simulation thread if the on-going sync is complete
if (waitNum == 0) {
sync_lock.unlock();
cv.notify_one();
}
}
void
DistIface::SyncNode::progress(Tick max_send_tick,
Tick next_repeat,
ReqType do_ckpt,
ReqType do_exit)
{
std::unique_lock<std::mutex> sync_lock(lock);
assert(waitNum > 0);
nextAt = max_send_tick;
nextRepeat = next_repeat;
doCkpt = (do_ckpt != ReqType::none);
doExit = (do_exit != ReqType::none);
waitNum--;
// Notify the simulation thread if the on-going sync is complete
if (waitNum == 0) {
sync_lock.unlock();
cv.notify_one();
}
}
void
DistIface::SyncNode::requestCkpt(ReqType req)
{
std::lock_guard<std::mutex> sync_lock(lock);
assert(req != ReqType::none);
if (needCkpt != ReqType::none)
warn("Ckpt requested multiple times (req:%d)\n", static_cast<int>(req));
if (needCkpt == ReqType::none || req == ReqType::immediate)
needCkpt = req;
}
void
DistIface::SyncNode::requestExit(ReqType req)
{
std::lock_guard<std::mutex> sync_lock(lock);
assert(req != ReqType::none);
if (needExit != ReqType::none)
warn("Exit requested multiple times (req:%d)\n", static_cast<int>(req));
if (needExit == ReqType::none || req == ReqType::immediate)
needExit = req;
}
void
DistIface::Sync::drainComplete()
{
if (doCkpt) {
// The first DistIface object called this right before writing the
// checkpoint. We need to drain the underlying physical network here.
// Note that other gem5 peers may enter this barrier at different
// ticks due to draining.
run(false);
// Only the "first" DistIface object has to perform the sync
doCkpt = false;
}
}
void
DistIface::SyncNode::serialize(CheckpointOut &cp) const
{
int need_exit = static_cast<int>(needExit);
SERIALIZE_SCALAR(need_exit);
}
void
DistIface::SyncNode::unserialize(CheckpointIn &cp)
{
int need_exit;
UNSERIALIZE_SCALAR(need_exit);
needExit = static_cast<ReqType>(need_exit);
}
void
DistIface::SyncSwitch::serialize(CheckpointOut &cp) const
{
SERIALIZE_SCALAR(numExitReq);
}
void
DistIface::SyncSwitch::unserialize(CheckpointIn &cp)
{
UNSERIALIZE_SCALAR(numExitReq);
}
void
DistIface::SyncEvent::start()
{
// Note that this may be called either from startup() or drainResume()
// At this point, all DistIface objects has already called Sync::init() so
// we have a local minimum of the start tick and repeat for the periodic
// sync.
Tick firstAt = DistIface::sync->firstAt;
repeat = DistIface::sync->nextRepeat;
// Do a global barrier to agree on a common repeat value (the smallest
// one from all participating nodes
DistIface::sync->run(curTick() == 0);
assert(!DistIface::sync->doCkpt);
assert(!DistIface::sync->doExit);
assert(DistIface::sync->nextAt >= curTick());
assert(DistIface::sync->nextRepeat <= repeat);
// if this is called at tick 0 then we use the config start param otherwise
// the maximum of the current tick of all participating nodes
if (curTick() == 0) {
assert(!scheduled());
assert(DistIface::sync->nextAt == 0);
schedule(firstAt);
} else {
if (scheduled())
reschedule(DistIface::sync->nextAt);
else
schedule(DistIface::sync->nextAt);
}
inform("Dist sync scheduled at %lu and repeats %lu\n", when(),
DistIface::sync->nextRepeat);
}
void
DistIface::SyncEvent::process()
{
// We may not start a global periodic sync while draining before taking a
// checkpoint. This is due to the possibility that peer gem5 processes
// may not hit the same periodic sync before they complete draining and
// that would make this periodic sync clash with sync called from
// DistIface::serialize() by other gem5 processes.
// We would need a 'distributed drain' solution to eliminate this
// restriction.
// Note that if draining was not triggered by checkpointing then we are
// fine since no extra global sync will happen (i.e. all peer gem5 will
// hit this periodic sync eventually).
panic_if(_draining && DistIface::sync->doCkpt,
"Distributed sync is hit while draining");
/*
* Note that this is a global event so this process method will be called
* by only exactly one thread.
*/
/*
* We hold the eventq lock at this point but the receiver thread may
* need the lock to schedule new recv events while waiting for the
* dist sync to complete.
* Note that the other simulation threads also release their eventq
* locks while waiting for us due to the global event semantics.
*/
{
EventQueue::ScopedRelease sr(curEventQueue());
// we do a global sync here that is supposed to happen at the same
// tick in all gem5 peers
DistIface::sync->run(true);
// global sync completed
}
if (DistIface::sync->doCkpt)
exitSimLoop("checkpoint");
if (DistIface::sync->doExit)
exitSimLoop("exit request from gem5 peers");
// schedule the next periodic sync
repeat = DistIface::sync->nextRepeat;
schedule(curTick() + repeat);
}
void
DistIface::RecvScheduler::init(Event *recv_done, Tick link_delay)
{
// This is called from the receiver thread when it starts running. The new
// receiver thread shares the event queue with the simulation thread
// (associated with the simulated Ethernet link).
curEventQueue(eventManager->eventQueue());
recvDone = recv_done;
linkDelay = link_delay;
}
Tick
DistIface::RecvScheduler::calcReceiveTick(Tick send_tick,
Tick send_delay,
Tick prev_recv_tick)
{
Tick recv_tick = send_tick + send_delay + linkDelay;
// sanity check (we need atleast a send delay long window)
assert(recv_tick >= prev_recv_tick + send_delay);
panic_if(prev_recv_tick + send_delay > recv_tick,
"Receive window is smaller than send delay");
panic_if(recv_tick <= curTick(),
"Simulators out of sync - missed packet receive by %llu ticks"
"(rev_recv_tick: %lu send_tick: %lu send_delay: %lu "
"linkDelay: %lu )",
curTick() - recv_tick, prev_recv_tick, send_tick, send_delay,
linkDelay);
return recv_tick;
}
void
DistIface::RecvScheduler::resumeRecvTicks()
{
// Schedule pending packets asap in case link speed/delay changed when
// restoring from the checkpoint.
// This may be done during unserialize except that curTick() is unknown
// so we call this during drainResume().
// If we are not restoring from a checkppint then link latency could not
// change so we just return.
if (!ckptRestore)
return;
std::vector<Desc> v;
while (!descQueue.empty()) {
Desc d = descQueue.front();
descQueue.pop();
d.sendTick = curTick();
d.sendDelay = d.packet->simLength; // assume 1 tick/byte max link speed
v.push_back(d);
}
for (auto &d : v)
descQueue.push(d);
if (recvDone->scheduled()) {
assert(!descQueue.empty());
eventManager->reschedule(recvDone, curTick());
} else {
assert(descQueue.empty() && v.empty());
}
ckptRestore = false;
}
void
DistIface::RecvScheduler::pushPacket(EthPacketPtr new_packet,
Tick send_tick,
Tick send_delay)
{
// Note : this is called from the receiver thread
curEventQueue()->lock();
Tick recv_tick = calcReceiveTick(send_tick, send_delay, prevRecvTick);
DPRINTF(DistEthernetPkt, "DistIface::recvScheduler::pushPacket "
"send_tick:%llu send_delay:%llu link_delay:%llu recv_tick:%llu\n",
send_tick, send_delay, linkDelay, recv_tick);
// Every packet must be sent and arrive in the same quantum
assert(send_tick > master->syncEvent->when() -
master->syncEvent->repeat);
// No packet may be scheduled for receive in the arrival quantum
assert(send_tick + send_delay + linkDelay > master->syncEvent->when());
// Now we are about to schedule a recvDone event for the new data packet.
// We use the same recvDone object for all incoming data packets. Packet
// descriptors are saved in the ordered queue. The currently scheduled
// packet is always on the top of the queue.
// NOTE: we use the event queue lock to protect the receive desc queue,
// too, which is accessed both by the receiver thread and the simulation
// thread.
descQueue.emplace(new_packet, send_tick, send_delay);
if (descQueue.size() == 1) {
assert(!recvDone->scheduled());
eventManager->schedule(recvDone, recv_tick);
} else {
assert(recvDone->scheduled());
panic_if(descQueue.front().sendTick + descQueue.front().sendDelay > recv_tick,
"Out of order packet received (recv_tick: %lu top(): %lu\n",
recv_tick, descQueue.front().sendTick + descQueue.front().sendDelay);
}
curEventQueue()->unlock();
}
EthPacketPtr
DistIface::RecvScheduler::popPacket()
{
// Note : this is called from the simulation thread when a receive done
// event is being processed for the link. We assume that the thread holds
// the event queue queue lock when this is called!
EthPacketPtr next_packet = descQueue.front().packet;
descQueue.pop();
if (descQueue.size() > 0) {
Tick recv_tick = calcReceiveTick(descQueue.front().sendTick,
descQueue.front().sendDelay,
curTick());
eventManager->schedule(recvDone, recv_tick);
}
prevRecvTick = curTick();
return next_packet;
}
void
DistIface::RecvScheduler::Desc::serialize(CheckpointOut &cp) const
{
SERIALIZE_SCALAR(sendTick);
SERIALIZE_SCALAR(sendDelay);
packet->serialize("rxPacket", cp);
}
void
DistIface::RecvScheduler::Desc::unserialize(CheckpointIn &cp)
{
UNSERIALIZE_SCALAR(sendTick);
UNSERIALIZE_SCALAR(sendDelay);
packet = std::make_shared<EthPacketData>();
packet->unserialize("rxPacket", cp);
}
void
DistIface::RecvScheduler::serialize(CheckpointOut &cp) const
{
SERIALIZE_SCALAR(prevRecvTick);
// serialize the receive desc queue
std::queue<Desc> tmp_queue(descQueue);
unsigned n_desc_queue = descQueue.size();
assert(tmp_queue.size() == descQueue.size());
SERIALIZE_SCALAR(n_desc_queue);
for (int i = 0; i < n_desc_queue; i++) {
tmp_queue.front().serializeSection(cp, csprintf("rxDesc_%d", i));
tmp_queue.pop();
}
assert(tmp_queue.empty());
}
void
DistIface::RecvScheduler::unserialize(CheckpointIn &cp)
{
assert(descQueue.size() == 0);
assert(!recvDone->scheduled());
assert(!ckptRestore);
UNSERIALIZE_SCALAR(prevRecvTick);
// unserialize the receive desc queue
unsigned n_desc_queue;
UNSERIALIZE_SCALAR(n_desc_queue);
for (int i = 0; i < n_desc_queue; i++) {
Desc recv_desc;
recv_desc.unserializeSection(cp, csprintf("rxDesc_%d", i));
descQueue.push(recv_desc);
}
ckptRestore = true;
}
DistIface::DistIface(unsigned dist_rank,
unsigned dist_size,
Tick sync_start,
Tick sync_repeat,
EventManager *em,
bool is_switch, int num_nodes) :
syncStart(sync_start), syncRepeat(sync_repeat),
recvThread(nullptr), recvScheduler(em),
rank(dist_rank), size(dist_size)
{
DPRINTF(DistEthernet, "DistIface() ctor rank:%d\n",dist_rank);
isMaster = false;
if (master == nullptr) {
assert(sync == nullptr);
assert(syncEvent == nullptr);
if (is_switch)
sync = new SyncSwitch(num_nodes);
else
sync = new SyncNode();
syncEvent = new SyncEvent();
master = this;
isMaster = true;
}
distIfaceId = distIfaceNum;
distIfaceNum++;
}
DistIface::~DistIface()
{
assert(recvThread);
delete recvThread;
if (this == master) {
assert(syncEvent);
delete syncEvent;
assert(sync);
delete sync;
master = nullptr;
}
}
void
DistIface::packetOut(EthPacketPtr pkt, Tick send_delay)
{
Header header;
// Prepare a dist header packet for the Ethernet packet we want to
// send out.
header.msgType = MsgType::dataDescriptor;
header.sendTick = curTick();
header.sendDelay = send_delay;
header.dataPacketLength = pkt->length;
header.simLength = pkt->simLength;
// Send out the packet and the meta info.
sendPacket(header, pkt);
DPRINTF(DistEthernetPkt,
"DistIface::sendDataPacket() done size:%d send_delay:%llu\n",
pkt->length, send_delay);
}
void
DistIface::recvThreadFunc(Event *recv_done, Tick link_delay)
{
EthPacketPtr new_packet;
DistHeaderPkt::Header header;
// Initialize receive scheduler parameters
recvScheduler.init(recv_done, link_delay);
// Main loop to wait for and process any incoming message.
for (;;) {
// recvHeader() blocks until the next dist header packet comes in.
if (!recvHeader(header)) {
// We lost connection to the peer gem5 processes most likely
// because one of them called m5 exit. So we stop here.
// Grab the eventq lock to stop the simulation thread
curEventQueue()->lock();
exitSimLoop("Message server closed connection, simulator "
"is exiting");
curEventQueue()->unlock();
break;
}
// We got a valid dist header packet, let's process it
if (header.msgType == MsgType::dataDescriptor) {
recvPacket(header, new_packet);
recvScheduler.pushPacket(new_packet,
header.sendTick,
header.sendDelay);
} else {
// everything else must be synchronisation related command
sync->progress(header.sendTick,
header.syncRepeat,
header.needCkpt,
header.needExit);
}
}
}
void
DistIface::spawnRecvThread(const Event *recv_done, Tick link_delay)
{
assert(recvThread == nullptr);
recvThread = new std::thread(&DistIface::recvThreadFunc,
this,
const_cast<Event *>(recv_done),
link_delay);
recvThreadsNum++;
}
DrainState
DistIface::drain()
{
DPRINTF(DistEthernet,"DistIFace::drain() called\n");
// This can be called multiple times in the same drain cycle.
if (this == master)
syncEvent->draining(true);
return DrainState::Drained;
}
void
DistIface::drainResume() {
DPRINTF(DistEthernet,"DistIFace::drainResume() called\n");
if (this == master)
syncEvent->draining(false);
recvScheduler.resumeRecvTicks();
}
void
DistIface::serialize(CheckpointOut &cp) const
{
// Drain the dist interface before the checkpoint is taken. We cannot call
// this as part of the normal drain cycle because this dist sync has to be
// called exactly once after the system is fully drained.
sync->drainComplete();
unsigned rank_orig = rank, dist_iface_id_orig = distIfaceId;
SERIALIZE_SCALAR(rank_orig);
SERIALIZE_SCALAR(dist_iface_id_orig);
recvScheduler.serializeSection(cp, "recvScheduler");
if (this == master) {
sync->serializeSection(cp, "Sync");
}
}
void
DistIface::unserialize(CheckpointIn &cp)
{
unsigned rank_orig, dist_iface_id_orig;
UNSERIALIZE_SCALAR(rank_orig);
UNSERIALIZE_SCALAR(dist_iface_id_orig);
panic_if(rank != rank_orig, "Rank mismatch at resume (rank=%d, orig=%d)",
rank, rank_orig);
panic_if(distIfaceId != dist_iface_id_orig, "Dist iface ID mismatch "
"at resume (distIfaceId=%d, orig=%d)", distIfaceId,
dist_iface_id_orig);
recvScheduler.unserializeSection(cp, "recvScheduler");
if (this == master) {
sync->unserializeSection(cp, "Sync");
}
}
void
DistIface::init(const Event *done_event, Tick link_delay)
{
// Init hook for the underlaying message transport to setup/finalize
// communication channels
initTransport();
// Spawn a new receiver thread that will process messages
// coming in from peer gem5 processes.
// The receive thread will also schedule a (receive) doneEvent
// for each incoming data packet.
spawnRecvThread(done_event, link_delay);
// Adjust the periodic sync start and interval. Different DistIface
// might have different requirements. The singleton sync object
// will select the minimum values for both params.
assert(sync != nullptr);
sync->init(syncStart, syncRepeat);
// Initialize the seed for random generator to avoid the same sequence
// in all gem5 peer processes
assert(master != nullptr);
if (this == master)
random_mt.init(5489 * (rank+1) + 257);
}
void
DistIface::startup()
{
DPRINTF(DistEthernet, "DistIface::startup() started\n");
if (this == master)
syncEvent->start();
DPRINTF(DistEthernet, "DistIface::startup() done\n");
}
bool
DistIface::readyToCkpt(Tick delay, Tick period)
{
bool ret = true;
DPRINTF(DistEthernet, "DistIface::readyToCkpt() called, delay:%lu "
"period:%lu\n", delay, period);
if (master) {
if (delay == 0) {
inform("m5 checkpoint called with zero delay => triggering collaborative "
"checkpoint\n");
sync->requestCkpt(ReqType::collective);
} else {
inform("m5 checkpoint called with non-zero delay => triggering immediate "
"checkpoint (at the next sync)\n");
sync->requestCkpt(ReqType::immediate);
}
if (period != 0)
inform("Non-zero period for m5_ckpt is ignored in "
"distributed gem5 runs\n");
ret = false;
}
return ret;
}
bool
DistIface::readyToExit(Tick delay)
{
bool ret = true;
DPRINTF(DistEthernet, "DistIface::readyToExit() called, delay:%lu\n",
delay);
if (master) {
if (delay == 0) {
inform("m5 exit called with zero delay => triggering collaborative "
"exit\n");
sync->requestExit(ReqType::collective);
} else {
inform("m5 exit called with non-zero delay => triggering immediate "
"exit (at the next sync)\n");
sync->requestExit(ReqType::immediate);
}
ret = false;
}
return ret;
}
uint64_t
DistIface::rankParam()
{
uint64_t val;
if (master) {
val = master->rank;
} else {
warn("Dist-rank parameter is queried in single gem5 simulation.");
val = 0;
}
return val;
}
uint64_t
DistIface::sizeParam()
{
uint64_t val;
if (master) {
val = master->size;
} else {
warn("Dist-size parameter is queried in single gem5 simulation.");
val = 1;
}
return val;
}
|