4045492ca53483f5a5e195d27f3ccce42511922c
[gem5.git] / src / cpu / o3 / fetch_impl.hh
1 /*
2 * Copyright (c) 2004-2006 The Regents of The University of Michigan
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Authors: Kevin Lim
29 * Korey Sewell
30 */
31
32 #include "config/use_checker.hh"
33
34 #include "arch/isa_traits.hh"
35 #include "arch/utility.hh"
36 #include "cpu/checker/cpu.hh"
37 #include "cpu/exetrace.hh"
38 #include "cpu/o3/fetch.hh"
39 #include "mem/packet.hh"
40 #include "mem/request.hh"
41 #include "sim/byteswap.hh"
42 #include "sim/host.hh"
43 #include "sim/root.hh"
44
45 #if FULL_SYSTEM
46 #include "arch/tlb.hh"
47 #include "arch/vtophys.hh"
48 #include "base/remote_gdb.hh"
49 #include "sim/system.hh"
50 #endif // FULL_SYSTEM
51
52 #include <algorithm>
53
54 using namespace std;
55 using namespace TheISA;
56
57 template<class Impl>
58 Tick
59 DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt)
60 {
61 panic("DefaultFetch doesn't expect recvAtomic callback!");
62 return curTick;
63 }
64
65 template<class Impl>
66 void
67 DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt)
68 {
69 panic("DefaultFetch doesn't expect recvFunctional callback!");
70 }
71
72 template<class Impl>
73 void
74 DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status)
75 {
76 if (status == RangeChange)
77 return;
78
79 panic("DefaultFetch doesn't expect recvStatusChange callback!");
80 }
81
82 template<class Impl>
83 bool
84 DefaultFetch<Impl>::IcachePort::recvTiming(Packet *pkt)
85 {
86 fetch->processCacheCompletion(pkt);
87 return true;
88 }
89
90 template<class Impl>
91 void
92 DefaultFetch<Impl>::IcachePort::recvRetry()
93 {
94 fetch->recvRetry();
95 }
96
97 template<class Impl>
98 DefaultFetch<Impl>::DefaultFetch(Params *params)
99 : mem(params->mem),
100 branchPred(params),
101 decodeToFetchDelay(params->decodeToFetchDelay),
102 renameToFetchDelay(params->renameToFetchDelay),
103 iewToFetchDelay(params->iewToFetchDelay),
104 commitToFetchDelay(params->commitToFetchDelay),
105 fetchWidth(params->fetchWidth),
106 cacheBlocked(false),
107 retryPkt(NULL),
108 retryTid(-1),
109 numThreads(params->numberOfThreads),
110 numFetchingThreads(params->smtNumFetchingThreads),
111 interruptPending(false),
112 drainPending(false),
113 switchedOut(false)
114 {
115 if (numThreads > Impl::MaxThreads)
116 fatal("numThreads is not a valid value\n");
117
118 // Set fetch stage's status to inactive.
119 _status = Inactive;
120
121 string policy = params->smtFetchPolicy;
122
123 // Convert string to lowercase
124 std::transform(policy.begin(), policy.end(), policy.begin(),
125 (int(*)(int)) tolower);
126
127 // Figure out fetch policy
128 if (policy == "singlethread") {
129 fetchPolicy = SingleThread;
130 if (numThreads > 1)
131 panic("Invalid Fetch Policy for a SMT workload.");
132 } else if (policy == "roundrobin") {
133 fetchPolicy = RoundRobin;
134 DPRINTF(Fetch, "Fetch policy set to Round Robin\n");
135 } else if (policy == "branch") {
136 fetchPolicy = Branch;
137 DPRINTF(Fetch, "Fetch policy set to Branch Count\n");
138 } else if (policy == "iqcount") {
139 fetchPolicy = IQ;
140 DPRINTF(Fetch, "Fetch policy set to IQ count\n");
141 } else if (policy == "lsqcount") {
142 fetchPolicy = LSQ;
143 DPRINTF(Fetch, "Fetch policy set to LSQ count\n");
144 } else {
145 fatal("Invalid Fetch Policy. Options Are: {SingleThread,"
146 " RoundRobin,LSQcount,IQcount}\n");
147 }
148
149 // Size of cache block.
150 cacheBlkSize = 64;
151
152 // Create mask to get rid of offset bits.
153 cacheBlkMask = (cacheBlkSize - 1);
154
155 for (int tid=0; tid < numThreads; tid++) {
156
157 fetchStatus[tid] = Running;
158
159 priorityList.push_back(tid);
160
161 memReq[tid] = NULL;
162
163 // Create space to store a cache line.
164 cacheData[tid] = new uint8_t[cacheBlkSize];
165
166 stalls[tid].decode = 0;
167 stalls[tid].rename = 0;
168 stalls[tid].iew = 0;
169 stalls[tid].commit = 0;
170 }
171
172 // Get the size of an instruction.
173 instSize = sizeof(MachInst);
174 }
175
176 template <class Impl>
177 std::string
178 DefaultFetch<Impl>::name() const
179 {
180 return cpu->name() + ".fetch";
181 }
182
183 template <class Impl>
184 void
185 DefaultFetch<Impl>::regStats()
186 {
187 icacheStallCycles
188 .name(name() + ".icacheStallCycles")
189 .desc("Number of cycles fetch is stalled on an Icache miss")
190 .prereq(icacheStallCycles);
191
192 fetchedInsts
193 .name(name() + ".Insts")
194 .desc("Number of instructions fetch has processed")
195 .prereq(fetchedInsts);
196
197 fetchedBranches
198 .name(name() + ".Branches")
199 .desc("Number of branches that fetch encountered")
200 .prereq(fetchedBranches);
201
202 predictedBranches
203 .name(name() + ".predictedBranches")
204 .desc("Number of branches that fetch has predicted taken")
205 .prereq(predictedBranches);
206
207 fetchCycles
208 .name(name() + ".Cycles")
209 .desc("Number of cycles fetch has run and was not squashing or"
210 " blocked")
211 .prereq(fetchCycles);
212
213 fetchSquashCycles
214 .name(name() + ".SquashCycles")
215 .desc("Number of cycles fetch has spent squashing")
216 .prereq(fetchSquashCycles);
217
218 fetchIdleCycles
219 .name(name() + ".IdleCycles")
220 .desc("Number of cycles fetch was idle")
221 .prereq(fetchIdleCycles);
222
223 fetchBlockedCycles
224 .name(name() + ".BlockedCycles")
225 .desc("Number of cycles fetch has spent blocked")
226 .prereq(fetchBlockedCycles);
227
228 fetchedCacheLines
229 .name(name() + ".CacheLines")
230 .desc("Number of cache lines fetched")
231 .prereq(fetchedCacheLines);
232
233 fetchMiscStallCycles
234 .name(name() + ".MiscStallCycles")
235 .desc("Number of cycles fetch has spent waiting on interrupts, or "
236 "bad addresses, or out of MSHRs")
237 .prereq(fetchMiscStallCycles);
238
239 fetchIcacheSquashes
240 .name(name() + ".IcacheSquashes")
241 .desc("Number of outstanding Icache misses that were squashed")
242 .prereq(fetchIcacheSquashes);
243
244 fetchNisnDist
245 .init(/* base value */ 0,
246 /* last value */ fetchWidth,
247 /* bucket size */ 1)
248 .name(name() + ".rateDist")
249 .desc("Number of instructions fetched each cycle (Total)")
250 .flags(Stats::pdf);
251
252 idleRate
253 .name(name() + ".idleRate")
254 .desc("Percent of cycles fetch was idle")
255 .prereq(idleRate);
256 idleRate = fetchIdleCycles * 100 / cpu->numCycles;
257
258 branchRate
259 .name(name() + ".branchRate")
260 .desc("Number of branch fetches per cycle")
261 .flags(Stats::total);
262 branchRate = fetchedBranches / cpu->numCycles;
263
264 fetchRate
265 .name(name() + ".rate")
266 .desc("Number of inst fetches per cycle")
267 .flags(Stats::total);
268 fetchRate = fetchedInsts / cpu->numCycles;
269
270 branchPred.regStats();
271 }
272
273 template<class Impl>
274 void
275 DefaultFetch<Impl>::setCPU(O3CPU *cpu_ptr)
276 {
277 DPRINTF(Fetch, "Setting the CPU pointer.\n");
278 cpu = cpu_ptr;
279
280 // Name is finally available, so create the port.
281 icachePort = new IcachePort(this);
282
283 #if USE_CHECKER
284 if (cpu->checker) {
285 cpu->checker->setIcachePort(icachePort);
286 }
287 #endif
288
289 // Fetch needs to start fetching instructions at the very beginning,
290 // so it must start up in active state.
291 switchToActive();
292 }
293
294 template<class Impl>
295 void
296 DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer)
297 {
298 DPRINTF(Fetch, "Setting the time buffer pointer.\n");
299 timeBuffer = time_buffer;
300
301 // Create wires to get information from proper places in time buffer.
302 fromDecode = timeBuffer->getWire(-decodeToFetchDelay);
303 fromRename = timeBuffer->getWire(-renameToFetchDelay);
304 fromIEW = timeBuffer->getWire(-iewToFetchDelay);
305 fromCommit = timeBuffer->getWire(-commitToFetchDelay);
306 }
307
308 template<class Impl>
309 void
310 DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr)
311 {
312 DPRINTF(Fetch, "Setting active threads list pointer.\n");
313 activeThreads = at_ptr;
314 }
315
316 template<class Impl>
317 void
318 DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr)
319 {
320 DPRINTF(Fetch, "Setting the fetch queue pointer.\n");
321 fetchQueue = fq_ptr;
322
323 // Create wire to write information to proper place in fetch queue.
324 toDecode = fetchQueue->getWire(0);
325 }
326
327 template<class Impl>
328 void
329 DefaultFetch<Impl>::initStage()
330 {
331 // Setup PC and nextPC with initial state.
332 for (int tid = 0; tid < numThreads; tid++) {
333 PC[tid] = cpu->readPC(tid);
334 nextPC[tid] = cpu->readNextPC(tid);
335 #if THE_ISA != ALPHA_ISA
336 nextNPC[tid] = cpu->readNextNPC(tid);
337 #endif
338 }
339 }
340
341 template<class Impl>
342 void
343 DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt)
344 {
345 unsigned tid = pkt->req->getThreadNum();
346
347 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid);
348
349 // Only change the status if it's still waiting on the icache access
350 // to return.
351 if (fetchStatus[tid] != IcacheWaitResponse ||
352 pkt->req != memReq[tid] ||
353 isSwitchedOut()) {
354 ++fetchIcacheSquashes;
355 delete pkt->req;
356 delete pkt;
357 return;
358 }
359
360 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize);
361
362 if (!drainPending) {
363 // Wake up the CPU (if it went to sleep and was waiting on
364 // this completion event).
365 cpu->wakeCPU();
366
367 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n",
368 tid);
369
370 switchToActive();
371 }
372
373 // Only switch to IcacheAccessComplete if we're not stalled as well.
374 if (checkStall(tid)) {
375 fetchStatus[tid] = Blocked;
376 } else {
377 fetchStatus[tid] = IcacheAccessComplete;
378 }
379
380 // Reset the mem req to NULL.
381 delete pkt->req;
382 delete pkt;
383 memReq[tid] = NULL;
384 }
385
386 template <class Impl>
387 bool
388 DefaultFetch<Impl>::drain()
389 {
390 // Fetch is ready to drain at any time.
391 cpu->signalDrained();
392 drainPending = true;
393 return true;
394 }
395
396 template <class Impl>
397 void
398 DefaultFetch<Impl>::resume()
399 {
400 drainPending = false;
401 }
402
403 template <class Impl>
404 void
405 DefaultFetch<Impl>::switchOut()
406 {
407 switchedOut = true;
408 // Branch predictor needs to have its state cleared.
409 branchPred.switchOut();
410 }
411
412 template <class Impl>
413 void
414 DefaultFetch<Impl>::takeOverFrom()
415 {
416 // Reset all state
417 for (int i = 0; i < Impl::MaxThreads; ++i) {
418 stalls[i].decode = 0;
419 stalls[i].rename = 0;
420 stalls[i].iew = 0;
421 stalls[i].commit = 0;
422 PC[i] = cpu->readPC(i);
423 nextPC[i] = cpu->readNextPC(i);
424 #if THE_ISA != ALPHA_ISA
425 nextNPC[i] = cpu->readNextNPC(i);
426 #endif
427 fetchStatus[i] = Running;
428 }
429 numInst = 0;
430 wroteToTimeBuffer = false;
431 _status = Inactive;
432 switchedOut = false;
433 branchPred.takeOverFrom();
434 }
435
436 template <class Impl>
437 void
438 DefaultFetch<Impl>::wakeFromQuiesce()
439 {
440 DPRINTF(Fetch, "Waking up from quiesce\n");
441 // Hopefully this is safe
442 // @todo: Allow other threads to wake from quiesce.
443 fetchStatus[0] = Running;
444 }
445
446 template <class Impl>
447 inline void
448 DefaultFetch<Impl>::switchToActive()
449 {
450 if (_status == Inactive) {
451 DPRINTF(Activity, "Activating stage.\n");
452
453 cpu->activateStage(O3CPU::FetchIdx);
454
455 _status = Active;
456 }
457 }
458
459 template <class Impl>
460 inline void
461 DefaultFetch<Impl>::switchToInactive()
462 {
463 if (_status == Active) {
464 DPRINTF(Activity, "Deactivating stage.\n");
465
466 cpu->deactivateStage(O3CPU::FetchIdx);
467
468 _status = Inactive;
469 }
470 }
471
472 template <class Impl>
473 bool
474 DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC)
475 {
476 // Do branch prediction check here.
477 // A bit of a misnomer...next_PC is actually the current PC until
478 // this function updates it.
479 bool predict_taken;
480
481 if (!inst->isControl()) {
482 next_PC = next_PC + instSize;
483 inst->setPredTarg(next_PC);
484 return false;
485 }
486
487 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber);
488
489 ++fetchedBranches;
490
491 if (predict_taken) {
492 ++predictedBranches;
493 }
494
495 return predict_taken;
496 }
497
498 template <class Impl>
499 bool
500 DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid)
501 {
502 Fault fault = NoFault;
503
504 #if FULL_SYSTEM
505 // Flag to say whether or not address is physical addr.
506 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0;
507 #else
508 unsigned flags = 0;
509 #endif // FULL_SYSTEM
510
511 if (cacheBlocked || (interruptPending && flags == 0)) {
512 // Hold off fetch from getting new instructions when:
513 // Cache is blocked, or
514 // while an interrupt is pending and we're not in PAL mode, or
515 // fetch is switched out.
516 return false;
517 }
518
519 // Align the fetch PC so it's at the start of a cache block.
520 fetch_PC = icacheBlockAlignPC(fetch_PC);
521
522 // If we've already got the block, no need to try to fetch it again.
523 if (fetch_PC == cacheDataPC[tid]) {
524 return true;
525 }
526
527 // Setup the memReq to do a read of the first instruction's address.
528 // Set the appropriate read size and flags as well.
529 // Build request here.
530 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags,
531 fetch_PC, cpu->readCpuId(), tid);
532
533 memReq[tid] = mem_req;
534
535 // Translate the instruction request.
536 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]);
537
538 // In the case of faults, the fetch stage may need to stall and wait
539 // for the ITB miss to be handled.
540
541 // If translation was successful, attempt to read the first
542 // instruction.
543 if (fault == NoFault) {
544 #if 0
545 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) ||
546 memReq[tid]->flags & UNCACHEABLE) {
547 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a "
548 "misspeculating path)!",
549 memReq[tid]->paddr);
550 ret_fault = TheISA::genMachineCheckFault();
551 return false;
552 }
553 #endif
554
555 // Build packet here.
556 PacketPtr data_pkt = new Packet(mem_req,
557 Packet::ReadReq, Packet::Broadcast);
558 data_pkt->dataDynamic(new uint8_t[cacheBlkSize]);
559
560 cacheDataPC[tid] = fetch_PC;
561
562 DPRINTF(Fetch, "Fetch: Doing instruction read.\n");
563
564 fetchedCacheLines++;
565
566 // Now do the timing access to see whether or not the instruction
567 // exists within the cache.
568 if (!icachePort->sendTiming(data_pkt)) {
569 assert(retryPkt == NULL);
570 assert(retryTid == -1);
571 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid);
572 fetchStatus[tid] = IcacheWaitRetry;
573 retryPkt = data_pkt;
574 retryTid = tid;
575 cacheBlocked = true;
576 return false;
577 }
578
579 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid);
580
581 lastIcacheStall[tid] = curTick;
582
583 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache "
584 "response.\n", tid);
585
586 fetchStatus[tid] = IcacheWaitResponse;
587 } else {
588 delete mem_req;
589 memReq[tid] = NULL;
590 }
591
592 ret_fault = fault;
593 return true;
594 }
595
596 template <class Impl>
597 inline void
598 DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid)
599 {
600 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n",
601 tid, new_PC);
602
603 PC[tid] = new_PC;
604 nextPC[tid] = new_PC + instSize;
605
606 // Clear the icache miss if it's outstanding.
607 if (fetchStatus[tid] == IcacheWaitResponse) {
608 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n",
609 tid);
610 memReq[tid] = NULL;
611 }
612
613 // Get rid of the retrying packet if it was from this thread.
614 if (retryTid == tid) {
615 assert(cacheBlocked);
616 cacheBlocked = false;
617 retryTid = -1;
618 retryPkt = NULL;
619 delete retryPkt->req;
620 delete retryPkt;
621 }
622
623 fetchStatus[tid] = Squashing;
624
625 ++fetchSquashCycles;
626 }
627
628 template<class Impl>
629 void
630 DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC,
631 const InstSeqNum &seq_num,
632 unsigned tid)
633 {
634 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid);
635
636 doSquash(new_PC, tid);
637
638 // Tell the CPU to remove any instructions that are in flight between
639 // fetch and decode.
640 cpu->removeInstsUntil(seq_num, tid);
641 }
642
643 template<class Impl>
644 bool
645 DefaultFetch<Impl>::checkStall(unsigned tid) const
646 {
647 bool ret_val = false;
648
649 if (cpu->contextSwitch) {
650 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid);
651 ret_val = true;
652 } else if (stalls[tid].decode) {
653 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid);
654 ret_val = true;
655 } else if (stalls[tid].rename) {
656 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid);
657 ret_val = true;
658 } else if (stalls[tid].iew) {
659 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid);
660 ret_val = true;
661 } else if (stalls[tid].commit) {
662 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid);
663 ret_val = true;
664 }
665
666 return ret_val;
667 }
668
669 template<class Impl>
670 typename DefaultFetch<Impl>::FetchStatus
671 DefaultFetch<Impl>::updateFetchStatus()
672 {
673 //Check Running
674 list<unsigned>::iterator threads = (*activeThreads).begin();
675
676 while (threads != (*activeThreads).end()) {
677
678 unsigned tid = *threads++;
679
680 if (fetchStatus[tid] == Running ||
681 fetchStatus[tid] == Squashing ||
682 fetchStatus[tid] == IcacheAccessComplete) {
683
684 if (_status == Inactive) {
685 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid);
686
687 if (fetchStatus[tid] == IcacheAccessComplete) {
688 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache"
689 "completion\n",tid);
690 }
691
692 cpu->activateStage(O3CPU::FetchIdx);
693 }
694
695 return Active;
696 }
697 }
698
699 // Stage is switching from active to inactive, notify CPU of it.
700 if (_status == Active) {
701 DPRINTF(Activity, "Deactivating stage.\n");
702
703 cpu->deactivateStage(O3CPU::FetchIdx);
704 }
705
706 return Inactive;
707 }
708
709 template <class Impl>
710 void
711 DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid)
712 {
713 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid);
714
715 doSquash(new_PC, tid);
716
717 // Tell the CPU to remove any instructions that are not in the ROB.
718 cpu->removeInstsNotInROB(tid);
719 }
720
721 template <class Impl>
722 void
723 DefaultFetch<Impl>::tick()
724 {
725 list<unsigned>::iterator threads = (*activeThreads).begin();
726 bool status_change = false;
727
728 wroteToTimeBuffer = false;
729
730 while (threads != (*activeThreads).end()) {
731 unsigned tid = *threads++;
732
733 // Check the signals for each thread to determine the proper status
734 // for each thread.
735 bool updated_status = checkSignalsAndUpdate(tid);
736 status_change = status_change || updated_status;
737 }
738
739 DPRINTF(Fetch, "Running stage.\n");
740
741 // Reset the number of the instruction we're fetching.
742 numInst = 0;
743
744 #if FULL_SYSTEM
745 if (fromCommit->commitInfo[0].interruptPending) {
746 interruptPending = true;
747 }
748
749 if (fromCommit->commitInfo[0].clearInterrupt) {
750 interruptPending = false;
751 }
752 #endif
753
754 for (threadFetched = 0; threadFetched < numFetchingThreads;
755 threadFetched++) {
756 // Fetch each of the actively fetching threads.
757 fetch(status_change);
758 }
759
760 // Record number of instructions fetched this cycle for distribution.
761 fetchNisnDist.sample(numInst);
762
763 if (status_change) {
764 // Change the fetch stage status if there was a status change.
765 _status = updateFetchStatus();
766 }
767
768 // If there was activity this cycle, inform the CPU of it.
769 if (wroteToTimeBuffer || cpu->contextSwitch) {
770 DPRINTF(Activity, "Activity this cycle.\n");
771
772 cpu->activityThisCycle();
773 }
774 }
775
776 template <class Impl>
777 bool
778 DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid)
779 {
780 // Update the per thread stall statuses.
781 if (fromDecode->decodeBlock[tid]) {
782 stalls[tid].decode = true;
783 }
784
785 if (fromDecode->decodeUnblock[tid]) {
786 assert(stalls[tid].decode);
787 assert(!fromDecode->decodeBlock[tid]);
788 stalls[tid].decode = false;
789 }
790
791 if (fromRename->renameBlock[tid]) {
792 stalls[tid].rename = true;
793 }
794
795 if (fromRename->renameUnblock[tid]) {
796 assert(stalls[tid].rename);
797 assert(!fromRename->renameBlock[tid]);
798 stalls[tid].rename = false;
799 }
800
801 if (fromIEW->iewBlock[tid]) {
802 stalls[tid].iew = true;
803 }
804
805 if (fromIEW->iewUnblock[tid]) {
806 assert(stalls[tid].iew);
807 assert(!fromIEW->iewBlock[tid]);
808 stalls[tid].iew = false;
809 }
810
811 if (fromCommit->commitBlock[tid]) {
812 stalls[tid].commit = true;
813 }
814
815 if (fromCommit->commitUnblock[tid]) {
816 assert(stalls[tid].commit);
817 assert(!fromCommit->commitBlock[tid]);
818 stalls[tid].commit = false;
819 }
820
821 // Check squash signals from commit.
822 if (fromCommit->commitInfo[tid].squash) {
823
824 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
825 "from commit.\n",tid);
826
827 // In any case, squash.
828 squash(fromCommit->commitInfo[tid].nextPC,tid);
829
830 // Also check if there's a mispredict that happened.
831 if (fromCommit->commitInfo[tid].branchMispredict) {
832 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
833 fromCommit->commitInfo[tid].nextPC,
834 fromCommit->commitInfo[tid].branchTaken,
835 tid);
836 } else {
837 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
838 tid);
839 }
840
841 return true;
842 } else if (fromCommit->commitInfo[tid].doneSeqNum) {
843 // Update the branch predictor if it wasn't a squashed instruction
844 // that was broadcasted.
845 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid);
846 }
847
848 // Check ROB squash signals from commit.
849 if (fromCommit->commitInfo[tid].robSquashing) {
850 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid);
851
852 // Continue to squash.
853 fetchStatus[tid] = Squashing;
854
855 return true;
856 }
857
858 // Check squash signals from decode.
859 if (fromDecode->decodeInfo[tid].squash) {
860 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
861 "from decode.\n",tid);
862
863 // Update the branch predictor.
864 if (fromDecode->decodeInfo[tid].branchMispredict) {
865 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
866 fromDecode->decodeInfo[tid].nextPC,
867 fromDecode->decodeInfo[tid].branchTaken,
868 tid);
869 } else {
870 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
871 tid);
872 }
873
874 if (fetchStatus[tid] != Squashing) {
875 // Squash unless we're already squashing
876 squashFromDecode(fromDecode->decodeInfo[tid].nextPC,
877 fromDecode->decodeInfo[tid].doneSeqNum,
878 tid);
879
880 return true;
881 }
882 }
883
884 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) {
885 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid);
886
887 fetchStatus[tid] = Blocked;
888
889 return true;
890 }
891
892 if (fetchStatus[tid] == Blocked ||
893 fetchStatus[tid] == Squashing) {
894 // Switch status to running if fetch isn't being told to block or
895 // squash this cycle.
896 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n",
897 tid);
898
899 fetchStatus[tid] = Running;
900
901 return true;
902 }
903
904 // If we've reached this point, we have not gotten any signals that
905 // cause fetch to change its status. Fetch remains the same as before.
906 return false;
907 }
908
909 template<class Impl>
910 void
911 DefaultFetch<Impl>::fetch(bool &status_change)
912 {
913 //////////////////////////////////////////
914 // Start actual fetch
915 //////////////////////////////////////////
916 int tid = getFetchingThread(fetchPolicy);
917
918 if (tid == -1 || drainPending) {
919 DPRINTF(Fetch,"There are no more threads available to fetch from.\n");
920
921 // Breaks looping condition in tick()
922 threadFetched = numFetchingThreads;
923 return;
924 }
925
926 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid);
927
928 // The current PC.
929 Addr &fetch_PC = PC[tid];
930
931 // Fault code for memory access.
932 Fault fault = NoFault;
933
934 // If returning from the delay of a cache miss, then update the status
935 // to running, otherwise do the cache access. Possibly move this up
936 // to tick() function.
937 if (fetchStatus[tid] == IcacheAccessComplete) {
938 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n",
939 tid);
940
941 fetchStatus[tid] = Running;
942 status_change = true;
943 } else if (fetchStatus[tid] == Running) {
944 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read "
945 "instruction, starting at PC %08p.\n",
946 tid, fetch_PC);
947
948 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid);
949 if (!fetch_success) {
950 if (cacheBlocked) {
951 ++icacheStallCycles;
952 } else {
953 ++fetchMiscStallCycles;
954 }
955 return;
956 }
957 } else {
958 if (fetchStatus[tid] == Idle) {
959 ++fetchIdleCycles;
960 } else if (fetchStatus[tid] == Blocked) {
961 ++fetchBlockedCycles;
962 } else if (fetchStatus[tid] == Squashing) {
963 ++fetchSquashCycles;
964 } else if (fetchStatus[tid] == IcacheWaitResponse) {
965 ++icacheStallCycles;
966 }
967
968 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so
969 // fetch should do nothing.
970 return;
971 }
972
973 ++fetchCycles;
974
975 // If we had a stall due to an icache miss, then return.
976 if (fetchStatus[tid] == IcacheWaitResponse) {
977 ++icacheStallCycles;
978 status_change = true;
979 return;
980 }
981
982 Addr next_PC = fetch_PC;
983 InstSeqNum inst_seq;
984 MachInst inst;
985 ExtMachInst ext_inst;
986 // @todo: Fix this hack.
987 unsigned offset = (fetch_PC & cacheBlkMask) & ~3;
988
989 if (fault == NoFault) {
990 // If the read of the first instruction was successful, then grab the
991 // instructions from the rest of the cache line and put them into the
992 // queue heading to decode.
993
994 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to "
995 "decode.\n",tid);
996
997 // Need to keep track of whether or not a predicted branch
998 // ended this fetch block.
999 bool predicted_branch = false;
1000
1001 for (;
1002 offset < cacheBlkSize &&
1003 numInst < fetchWidth &&
1004 !predicted_branch;
1005 ++numInst) {
1006
1007 // Get a sequence number.
1008 inst_seq = cpu->getAndIncrementInstSeq();
1009
1010 // Make sure this is a valid index.
1011 assert(offset <= cacheBlkSize - instSize);
1012
1013 // Get the instruction from the array of the cache line.
1014 inst = gtoh(*reinterpret_cast<MachInst *>
1015 (&cacheData[tid][offset]));
1016
1017 ext_inst = TheISA::makeExtMI(inst, fetch_PC);
1018
1019 // Create a new DynInst from the instruction fetched.
1020 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC,
1021 next_PC,
1022 inst_seq, cpu);
1023 instruction->setTid(tid);
1024
1025 instruction->setASID(tid);
1026
1027 instruction->setThreadState(cpu->thread[tid]);
1028
1029 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created "
1030 "[sn:%lli]\n",
1031 tid, instruction->readPC(), inst_seq);
1032
1033 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n",
1034 tid, instruction->staticInst->disassemble(fetch_PC));
1035
1036 instruction->traceData =
1037 Trace::getInstRecord(curTick, cpu->tcBase(tid), cpu,
1038 instruction->staticInst,
1039 instruction->readPC(),tid);
1040
1041 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC);
1042
1043 // Add instruction to the CPU's list of instructions.
1044 instruction->setInstListIt(cpu->addInst(instruction));
1045
1046 // Write the instruction to the first slot in the queue
1047 // that heads to decode.
1048 toDecode->insts[numInst] = instruction;
1049
1050 toDecode->size++;
1051
1052 // Increment stat of fetched instructions.
1053 ++fetchedInsts;
1054
1055 // Move to the next instruction, unless we have a branch.
1056 fetch_PC = next_PC;
1057
1058 if (instruction->isQuiesce()) {
1059 warn("cycle %lli: Quiesce instruction encountered, halting fetch!",
1060 curTick);
1061 fetchStatus[tid] = QuiescePending;
1062 ++numInst;
1063 status_change = true;
1064 break;
1065 }
1066
1067 offset+= instSize;
1068 }
1069 }
1070
1071 if (numInst > 0) {
1072 wroteToTimeBuffer = true;
1073 }
1074
1075 // Now that fetching is completed, update the PC to signify what the next
1076 // cycle will be.
1077 if (fault == NoFault) {
1078 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC);
1079
1080 #if THE_ISA == ALPHA_ISA
1081 PC[tid] = next_PC;
1082 nextPC[tid] = next_PC + instSize;
1083 #else
1084 PC[tid] = next_PC;
1085 nextPC[tid] = next_PC + instSize;
1086 nextPC[tid] = next_PC + instSize;
1087
1088 thread->setNextPC(thread->readNextNPC());
1089 thread->setNextNPC(thread->readNextNPC() + sizeof(MachInst));
1090 #endif
1091 } else {
1092 // We shouldn't be in an icache miss and also have a fault (an ITB
1093 // miss)
1094 if (fetchStatus[tid] == IcacheWaitResponse) {
1095 panic("Fetch should have exited prior to this!");
1096 }
1097
1098 // Send the fault to commit. This thread will not do anything
1099 // until commit handles the fault. The only other way it can
1100 // wake up is if a squash comes along and changes the PC.
1101 #if FULL_SYSTEM
1102 assert(numInst != fetchWidth);
1103 // Get a sequence number.
1104 inst_seq = cpu->getAndIncrementInstSeq();
1105 // We will use a nop in order to carry the fault.
1106 ext_inst = TheISA::NoopMachInst;
1107
1108 // Create a new DynInst from the dummy nop.
1109 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC,
1110 next_PC,
1111 inst_seq, cpu);
1112 instruction->setPredTarg(next_PC + instSize);
1113 instruction->setTid(tid);
1114
1115 instruction->setASID(tid);
1116
1117 instruction->setThreadState(cpu->thread[tid]);
1118
1119 instruction->traceData = NULL;
1120
1121 instruction->setInstListIt(cpu->addInst(instruction));
1122
1123 instruction->fault = fault;
1124
1125 toDecode->insts[numInst] = instruction;
1126 toDecode->size++;
1127
1128 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid);
1129
1130 fetchStatus[tid] = TrapPending;
1131 status_change = true;
1132
1133 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]);
1134 #else // !FULL_SYSTEM
1135 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]);
1136 #endif // FULL_SYSTEM
1137 }
1138 }
1139
1140 template<class Impl>
1141 void
1142 DefaultFetch<Impl>::recvRetry()
1143 {
1144 assert(cacheBlocked);
1145 if (retryPkt != NULL) {
1146 assert(retryTid != -1);
1147 assert(fetchStatus[retryTid] == IcacheWaitRetry);
1148
1149 if (icachePort->sendTiming(retryPkt)) {
1150 fetchStatus[retryTid] = IcacheWaitResponse;
1151 retryPkt = NULL;
1152 retryTid = -1;
1153 cacheBlocked = false;
1154 }
1155 } else {
1156 assert(retryTid == -1);
1157 // Access has been squashed since it was sent out. Just clear
1158 // the cache being blocked.
1159 cacheBlocked = false;
1160 }
1161 }
1162
1163 ///////////////////////////////////////
1164 // //
1165 // SMT FETCH POLICY MAINTAINED HERE //
1166 // //
1167 ///////////////////////////////////////
1168 template<class Impl>
1169 int
1170 DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority)
1171 {
1172 if (numThreads > 1) {
1173 switch (fetch_priority) {
1174
1175 case SingleThread:
1176 return 0;
1177
1178 case RoundRobin:
1179 return roundRobin();
1180
1181 case IQ:
1182 return iqCount();
1183
1184 case LSQ:
1185 return lsqCount();
1186
1187 case Branch:
1188 return branchCount();
1189
1190 default:
1191 return -1;
1192 }
1193 } else {
1194 int tid = *((*activeThreads).begin());
1195
1196 if (fetchStatus[tid] == Running ||
1197 fetchStatus[tid] == IcacheAccessComplete ||
1198 fetchStatus[tid] == Idle) {
1199 return tid;
1200 } else {
1201 return -1;
1202 }
1203 }
1204
1205 }
1206
1207
1208 template<class Impl>
1209 int
1210 DefaultFetch<Impl>::roundRobin()
1211 {
1212 list<unsigned>::iterator pri_iter = priorityList.begin();
1213 list<unsigned>::iterator end = priorityList.end();
1214
1215 int high_pri;
1216
1217 while (pri_iter != end) {
1218 high_pri = *pri_iter;
1219
1220 assert(high_pri <= numThreads);
1221
1222 if (fetchStatus[high_pri] == Running ||
1223 fetchStatus[high_pri] == IcacheAccessComplete ||
1224 fetchStatus[high_pri] == Idle) {
1225
1226 priorityList.erase(pri_iter);
1227 priorityList.push_back(high_pri);
1228
1229 return high_pri;
1230 }
1231
1232 pri_iter++;
1233 }
1234
1235 return -1;
1236 }
1237
1238 template<class Impl>
1239 int
1240 DefaultFetch<Impl>::iqCount()
1241 {
1242 priority_queue<unsigned> PQ;
1243
1244 list<unsigned>::iterator threads = (*activeThreads).begin();
1245
1246 while (threads != (*activeThreads).end()) {
1247 unsigned tid = *threads++;
1248
1249 PQ.push(fromIEW->iewInfo[tid].iqCount);
1250 }
1251
1252 while (!PQ.empty()) {
1253
1254 unsigned high_pri = PQ.top();
1255
1256 if (fetchStatus[high_pri] == Running ||
1257 fetchStatus[high_pri] == IcacheAccessComplete ||
1258 fetchStatus[high_pri] == Idle)
1259 return high_pri;
1260 else
1261 PQ.pop();
1262
1263 }
1264
1265 return -1;
1266 }
1267
1268 template<class Impl>
1269 int
1270 DefaultFetch<Impl>::lsqCount()
1271 {
1272 priority_queue<unsigned> PQ;
1273
1274
1275 list<unsigned>::iterator threads = (*activeThreads).begin();
1276
1277 while (threads != (*activeThreads).end()) {
1278 unsigned tid = *threads++;
1279
1280 PQ.push(fromIEW->iewInfo[tid].ldstqCount);
1281 }
1282
1283 while (!PQ.empty()) {
1284
1285 unsigned high_pri = PQ.top();
1286
1287 if (fetchStatus[high_pri] == Running ||
1288 fetchStatus[high_pri] == IcacheAccessComplete ||
1289 fetchStatus[high_pri] == Idle)
1290 return high_pri;
1291 else
1292 PQ.pop();
1293
1294 }
1295
1296 return -1;
1297 }
1298
1299 template<class Impl>
1300 int
1301 DefaultFetch<Impl>::branchCount()
1302 {
1303 list<unsigned>::iterator threads = (*activeThreads).begin();
1304 panic("Branch Count Fetch policy unimplemented\n");
1305 return *threads;
1306 }