timing.cc (8832:247fee427324) timing.cc (8850:ed91b534ed04)
1/*
2 * Copyright (c) 2010 ARM Limited
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2002-2005 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Steve Reinhardt
41 */
42
43#include "arch/locked_mem.hh"
44#include "arch/mmapped_ipr.hh"
45#include "arch/utility.hh"
46#include "base/bigint.hh"
47#include "config/the_isa.hh"
48#include "cpu/simple/timing.hh"
49#include "cpu/exetrace.hh"
50#include "debug/Config.hh"
51#include "debug/ExecFaulting.hh"
52#include "debug/SimpleCPU.hh"
53#include "mem/packet.hh"
54#include "mem/packet_access.hh"
55#include "params/TimingSimpleCPU.hh"
56#include "sim/faults.hh"
57#include "sim/full_system.hh"
58#include "sim/system.hh"
59
60using namespace std;
61using namespace TheISA;
62
1/*
2 * Copyright (c) 2010 ARM Limited
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2002-2005 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Steve Reinhardt
41 */
42
43#include "arch/locked_mem.hh"
44#include "arch/mmapped_ipr.hh"
45#include "arch/utility.hh"
46#include "base/bigint.hh"
47#include "config/the_isa.hh"
48#include "cpu/simple/timing.hh"
49#include "cpu/exetrace.hh"
50#include "debug/Config.hh"
51#include "debug/ExecFaulting.hh"
52#include "debug/SimpleCPU.hh"
53#include "mem/packet.hh"
54#include "mem/packet_access.hh"
55#include "params/TimingSimpleCPU.hh"
56#include "sim/faults.hh"
57#include "sim/full_system.hh"
58#include "sim/system.hh"
59
60using namespace std;
61using namespace TheISA;
62
63Port *
64TimingSimpleCPU::getPort(const std::string &if_name, int idx)
65{
66 if (if_name == "dcache_port")
67 return &dcachePort;
68 else if (if_name == "icache_port")
69 return &icachePort;
70 else
71 panic("No Such Port\n");
72}
73
74void
75TimingSimpleCPU::init()
76{
77 BaseCPU::init();
78 if (FullSystem) {
79 for (int i = 0; i < threadContexts.size(); ++i) {
80 ThreadContext *tc = threadContexts[i];
81 // initialize CPU, including PC
82 TheISA::initCPU(tc, _cpuId);
83 }
84 }
85
86 // Initialise the ThreadContext's memory proxies
87 tcBase()->initMemProxies(tcBase());
88}
89
90void
91TimingSimpleCPU::TimingCPUPort::TickEvent::schedule(PacketPtr _pkt, Tick t)
92{
93 pkt = _pkt;
94 cpu->schedule(this, t);
95}
96
97TimingSimpleCPU::TimingSimpleCPU(TimingSimpleCPUParams *p)
98 : BaseSimpleCPU(p), fetchTranslation(this), icachePort(this),
99 dcachePort(this), fetchEvent(this)
100{
101 _status = Idle;
102
103 ifetch_pkt = dcache_pkt = NULL;
104 drainEvent = NULL;
105 previousTick = 0;
106 changeState(SimObject::Running);
107 system->totalNumInsts = 0;
108}
109
110
111TimingSimpleCPU::~TimingSimpleCPU()
112{
113}
114
115void
116TimingSimpleCPU::serialize(ostream &os)
117{
118 SimObject::State so_state = SimObject::getState();
119 SERIALIZE_ENUM(so_state);
120 BaseSimpleCPU::serialize(os);
121}
122
123void
124TimingSimpleCPU::unserialize(Checkpoint *cp, const string &section)
125{
126 SimObject::State so_state;
127 UNSERIALIZE_ENUM(so_state);
128 BaseSimpleCPU::unserialize(cp, section);
129}
130
131unsigned int
132TimingSimpleCPU::drain(Event *drain_event)
133{
134 // TimingSimpleCPU is ready to drain if it's not waiting for
135 // an access to complete.
136 if (_status == Idle || _status == Running || _status == SwitchedOut) {
137 changeState(SimObject::Drained);
138 return 0;
139 } else {
140 changeState(SimObject::Draining);
141 drainEvent = drain_event;
142 return 1;
143 }
144}
145
146void
147TimingSimpleCPU::resume()
148{
149 DPRINTF(SimpleCPU, "Resume\n");
150 if (_status != SwitchedOut && _status != Idle) {
151 assert(system->getMemoryMode() == Enums::timing);
152
153 if (fetchEvent.scheduled())
154 deschedule(fetchEvent);
155
156 schedule(fetchEvent, nextCycle());
157 }
158
159 changeState(SimObject::Running);
160}
161
162void
163TimingSimpleCPU::switchOut()
164{
165 assert(_status == Running || _status == Idle);
166 _status = SwitchedOut;
167 numCycles += tickToCycles(curTick() - previousTick);
168
169 // If we've been scheduled to resume but are then told to switch out,
170 // we'll need to cancel it.
171 if (fetchEvent.scheduled())
172 deschedule(fetchEvent);
173}
174
175
176void
177TimingSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
178{
179 BaseCPU::takeOverFrom(oldCPU);
180
181 // if any of this CPU's ThreadContexts are active, mark the CPU as
182 // running and schedule its tick event.
183 for (int i = 0; i < threadContexts.size(); ++i) {
184 ThreadContext *tc = threadContexts[i];
185 if (tc->status() == ThreadContext::Active && _status != Running) {
186 _status = Running;
187 break;
188 }
189 }
190
191 if (_status != Running) {
192 _status = Idle;
193 }
194 assert(threadContexts.size() == 1);
195 previousTick = curTick();
196}
197
198
199void
200TimingSimpleCPU::activateContext(ThreadID thread_num, int delay)
201{
202 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);
203
204 assert(thread_num == 0);
205 assert(thread);
206
207 assert(_status == Idle);
208
209 notIdleFraction++;
210 _status = Running;
211
212 // kick things off by initiating the fetch of the next instruction
213 schedule(fetchEvent, nextCycle(curTick() + ticks(delay)));
214}
215
216
217void
218TimingSimpleCPU::suspendContext(ThreadID thread_num)
219{
220 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);
221
222 assert(thread_num == 0);
223 assert(thread);
224
225 if (_status == Idle)
226 return;
227
228 assert(_status == Running);
229
230 // just change status to Idle... if status != Running,
231 // completeInst() will not initiate fetch of next instruction.
232
233 notIdleFraction--;
234 _status = Idle;
235}
236
237bool
238TimingSimpleCPU::handleReadPacket(PacketPtr pkt)
239{
240 RequestPtr req = pkt->req;
241 if (req->isMmappedIpr()) {
242 Tick delay;
243 delay = TheISA::handleIprRead(thread->getTC(), pkt);
244 new IprEvent(pkt, this, nextCycle(curTick() + delay));
245 _status = DcacheWaitResponse;
246 dcache_pkt = NULL;
247 } else if (!dcachePort.sendTiming(pkt)) {
248 _status = DcacheRetry;
249 dcache_pkt = pkt;
250 } else {
251 _status = DcacheWaitResponse;
252 // memory system takes ownership of packet
253 dcache_pkt = NULL;
254 }
255 return dcache_pkt == NULL;
256}
257
258void
259TimingSimpleCPU::sendData(RequestPtr req, uint8_t *data, uint64_t *res,
260 bool read)
261{
262 PacketPtr pkt;
263 buildPacket(pkt, req, read);
264 pkt->dataDynamicArray<uint8_t>(data);
265 if (req->getFlags().isSet(Request::NO_ACCESS)) {
266 assert(!dcache_pkt);
267 pkt->makeResponse();
268 completeDataAccess(pkt);
269 } else if (read) {
270 handleReadPacket(pkt);
271 } else {
272 bool do_access = true; // flag to suppress cache access
273
274 if (req->isLLSC()) {
275 do_access = TheISA::handleLockedWrite(thread, req);
276 } else if (req->isCondSwap()) {
277 assert(res);
278 req->setExtraData(*res);
279 }
280
281 if (do_access) {
282 dcache_pkt = pkt;
283 handleWritePacket();
284 } else {
285 _status = DcacheWaitResponse;
286 completeDataAccess(pkt);
287 }
288 }
289}
290
291void
292TimingSimpleCPU::sendSplitData(RequestPtr req1, RequestPtr req2,
293 RequestPtr req, uint8_t *data, bool read)
294{
295 PacketPtr pkt1, pkt2;
296 buildSplitPacket(pkt1, pkt2, req1, req2, req, data, read);
297 if (req->getFlags().isSet(Request::NO_ACCESS)) {
298 assert(!dcache_pkt);
299 pkt1->makeResponse();
300 completeDataAccess(pkt1);
301 } else if (read) {
302 SplitFragmentSenderState * send_state =
303 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
304 if (handleReadPacket(pkt1)) {
305 send_state->clearFromParent();
306 send_state = dynamic_cast<SplitFragmentSenderState *>(
307 pkt2->senderState);
308 if (handleReadPacket(pkt2)) {
309 send_state->clearFromParent();
310 }
311 }
312 } else {
313 dcache_pkt = pkt1;
314 SplitFragmentSenderState * send_state =
315 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
316 if (handleWritePacket()) {
317 send_state->clearFromParent();
318 dcache_pkt = pkt2;
319 send_state = dynamic_cast<SplitFragmentSenderState *>(
320 pkt2->senderState);
321 if (handleWritePacket()) {
322 send_state->clearFromParent();
323 }
324 }
325 }
326}
327
328void
329TimingSimpleCPU::translationFault(Fault fault)
330{
331 // fault may be NoFault in cases where a fault is suppressed,
332 // for instance prefetches.
333 numCycles += tickToCycles(curTick() - previousTick);
334 previousTick = curTick();
335
336 if (traceData) {
337 // Since there was a fault, we shouldn't trace this instruction.
338 delete traceData;
339 traceData = NULL;
340 }
341
342 postExecute();
343
344 if (getState() == SimObject::Draining) {
345 advancePC(fault);
346 completeDrain();
347 } else {
348 advanceInst(fault);
349 }
350}
351
352void
353TimingSimpleCPU::buildPacket(PacketPtr &pkt, RequestPtr req, bool read)
354{
355 MemCmd cmd;
356 if (read) {
357 cmd = MemCmd::ReadReq;
358 if (req->isLLSC())
359 cmd = MemCmd::LoadLockedReq;
360 } else {
361 cmd = MemCmd::WriteReq;
362 if (req->isLLSC()) {
363 cmd = MemCmd::StoreCondReq;
364 } else if (req->isSwap()) {
365 cmd = MemCmd::SwapReq;
366 }
367 }
368 pkt = new Packet(req, cmd, Packet::Broadcast);
369}
370
371void
372TimingSimpleCPU::buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
373 RequestPtr req1, RequestPtr req2, RequestPtr req,
374 uint8_t *data, bool read)
375{
376 pkt1 = pkt2 = NULL;
377
378 assert(!req1->isMmappedIpr() && !req2->isMmappedIpr());
379
380 if (req->getFlags().isSet(Request::NO_ACCESS)) {
381 buildPacket(pkt1, req, read);
382 return;
383 }
384
385 buildPacket(pkt1, req1, read);
386 buildPacket(pkt2, req2, read);
387
388 req->setPhys(req1->getPaddr(), req->getSize(), req1->getFlags(), dataMasterId());
389 PacketPtr pkt = new Packet(req, pkt1->cmd.responseCommand(),
390 Packet::Broadcast);
391
392 pkt->dataDynamicArray<uint8_t>(data);
393 pkt1->dataStatic<uint8_t>(data);
394 pkt2->dataStatic<uint8_t>(data + req1->getSize());
395
396 SplitMainSenderState * main_send_state = new SplitMainSenderState;
397 pkt->senderState = main_send_state;
398 main_send_state->fragments[0] = pkt1;
399 main_send_state->fragments[1] = pkt2;
400 main_send_state->outstanding = 2;
401 pkt1->senderState = new SplitFragmentSenderState(pkt, 0);
402 pkt2->senderState = new SplitFragmentSenderState(pkt, 1);
403}
404
405Fault
406TimingSimpleCPU::readMem(Addr addr, uint8_t *data,
407 unsigned size, unsigned flags)
408{
409 Fault fault;
410 const int asid = 0;
411 const ThreadID tid = 0;
412 const Addr pc = thread->instAddr();
413 unsigned block_size = dcachePort.peerBlockSize();
414 BaseTLB::Mode mode = BaseTLB::Read;
415
416 if (traceData) {
417 traceData->setAddr(addr);
418 }
419
420 RequestPtr req = new Request(asid, addr, size,
421 flags, dataMasterId(), pc, _cpuId, tid);
422
423 Addr split_addr = roundDown(addr + size - 1, block_size);
424 assert(split_addr <= addr || split_addr - addr < block_size);
425
426 _status = DTBWaitResponse;
427 if (split_addr > addr) {
428 RequestPtr req1, req2;
429 assert(!req->isLLSC() && !req->isSwap());
430 req->splitOnVaddr(split_addr, req1, req2);
431
432 WholeTranslationState *state =
433 new WholeTranslationState(req, req1, req2, new uint8_t[size],
434 NULL, mode);
435 DataTranslation<TimingSimpleCPU *> *trans1 =
436 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
437 DataTranslation<TimingSimpleCPU *> *trans2 =
438 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
439
440 thread->dtb->translateTiming(req1, tc, trans1, mode);
441 thread->dtb->translateTiming(req2, tc, trans2, mode);
442 } else {
443 WholeTranslationState *state =
444 new WholeTranslationState(req, new uint8_t[size], NULL, mode);
445 DataTranslation<TimingSimpleCPU *> *translation
446 = new DataTranslation<TimingSimpleCPU *>(this, state);
447 thread->dtb->translateTiming(req, tc, translation, mode);
448 }
449
450 return NoFault;
451}
452
453bool
454TimingSimpleCPU::handleWritePacket()
455{
456 RequestPtr req = dcache_pkt->req;
457 if (req->isMmappedIpr()) {
458 Tick delay;
459 delay = TheISA::handleIprWrite(thread->getTC(), dcache_pkt);
460 new IprEvent(dcache_pkt, this, nextCycle(curTick() + delay));
461 _status = DcacheWaitResponse;
462 dcache_pkt = NULL;
463 } else if (!dcachePort.sendTiming(dcache_pkt)) {
464 _status = DcacheRetry;
465 } else {
466 _status = DcacheWaitResponse;
467 // memory system takes ownership of packet
468 dcache_pkt = NULL;
469 }
470 return dcache_pkt == NULL;
471}
472
473Fault
474TimingSimpleCPU::writeMem(uint8_t *data, unsigned size,
475 Addr addr, unsigned flags, uint64_t *res)
476{
477 uint8_t *newData = new uint8_t[size];
478 memcpy(newData, data, size);
479
480 const int asid = 0;
481 const ThreadID tid = 0;
482 const Addr pc = thread->instAddr();
483 unsigned block_size = dcachePort.peerBlockSize();
484 BaseTLB::Mode mode = BaseTLB::Write;
485
486 if (traceData) {
487 traceData->setAddr(addr);
488 }
489
490 RequestPtr req = new Request(asid, addr, size,
491 flags, dataMasterId(), pc, _cpuId, tid);
492
493 Addr split_addr = roundDown(addr + size - 1, block_size);
494 assert(split_addr <= addr || split_addr - addr < block_size);
495
496 _status = DTBWaitResponse;
497 if (split_addr > addr) {
498 RequestPtr req1, req2;
499 assert(!req->isLLSC() && !req->isSwap());
500 req->splitOnVaddr(split_addr, req1, req2);
501
502 WholeTranslationState *state =
503 new WholeTranslationState(req, req1, req2, newData, res, mode);
504 DataTranslation<TimingSimpleCPU *> *trans1 =
505 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
506 DataTranslation<TimingSimpleCPU *> *trans2 =
507 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
508
509 thread->dtb->translateTiming(req1, tc, trans1, mode);
510 thread->dtb->translateTiming(req2, tc, trans2, mode);
511 } else {
512 WholeTranslationState *state =
513 new WholeTranslationState(req, newData, res, mode);
514 DataTranslation<TimingSimpleCPU *> *translation =
515 new DataTranslation<TimingSimpleCPU *>(this, state);
516 thread->dtb->translateTiming(req, tc, translation, mode);
517 }
518
519 // Translation faults will be returned via finishTranslation()
520 return NoFault;
521}
522
523
524void
525TimingSimpleCPU::finishTranslation(WholeTranslationState *state)
526{
527 _status = Running;
528
529 if (state->getFault() != NoFault) {
530 if (state->isPrefetch()) {
531 state->setNoFault();
532 }
533 delete [] state->data;
534 state->deleteReqs();
535 translationFault(state->getFault());
536 } else {
537 if (!state->isSplit) {
538 sendData(state->mainReq, state->data, state->res,
539 state->mode == BaseTLB::Read);
540 } else {
541 sendSplitData(state->sreqLow, state->sreqHigh, state->mainReq,
542 state->data, state->mode == BaseTLB::Read);
543 }
544 }
545
546 delete state;
547}
548
549
550void
551TimingSimpleCPU::fetch()
552{
553 DPRINTF(SimpleCPU, "Fetch\n");
554
555 if (!curStaticInst || !curStaticInst->isDelayedCommit())
556 checkForInterrupts();
557
558 checkPcEventQueue();
559
560 // We must have just got suspended by a PC event
561 if (_status == Idle)
562 return;
563
564 TheISA::PCState pcState = thread->pcState();
565 bool needToFetch = !isRomMicroPC(pcState.microPC()) && !curMacroStaticInst;
566
567 if (needToFetch) {
568 _status = Running;
569 Request *ifetch_req = new Request();
570 ifetch_req->setThreadContext(_cpuId, /* thread ID */ 0);
571 setupFetchRequest(ifetch_req);
572 DPRINTF(SimpleCPU, "Translating address %#x\n", ifetch_req->getVaddr());
573 thread->itb->translateTiming(ifetch_req, tc, &fetchTranslation,
574 BaseTLB::Execute);
575 } else {
576 _status = IcacheWaitResponse;
577 completeIfetch(NULL);
578
579 numCycles += tickToCycles(curTick() - previousTick);
580 previousTick = curTick();
581 }
582}
583
584
585void
586TimingSimpleCPU::sendFetch(Fault fault, RequestPtr req, ThreadContext *tc)
587{
588 if (fault == NoFault) {
589 DPRINTF(SimpleCPU, "Sending fetch for addr %#x(pa: %#x)\n",
590 req->getVaddr(), req->getPaddr());
591 ifetch_pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast);
592 ifetch_pkt->dataStatic(&inst);
593 DPRINTF(SimpleCPU, " -- pkt addr: %#x\n", ifetch_pkt->getAddr());
594
595 if (!icachePort.sendTiming(ifetch_pkt)) {
596 // Need to wait for retry
597 _status = IcacheRetry;
598 } else {
599 // Need to wait for cache to respond
600 _status = IcacheWaitResponse;
601 // ownership of packet transferred to memory system
602 ifetch_pkt = NULL;
603 }
604 } else {
605 DPRINTF(SimpleCPU, "Translation of addr %#x faulted\n", req->getVaddr());
606 delete req;
607 // fetch fault: advance directly to next instruction (fault handler)
608 _status = Running;
609 advanceInst(fault);
610 }
611
612 numCycles += tickToCycles(curTick() - previousTick);
613 previousTick = curTick();
614}
615
616
617void
618TimingSimpleCPU::advanceInst(Fault fault)
619{
620
621 if (_status == Faulting)
622 return;
623
624 if (fault != NoFault) {
625 advancePC(fault);
626 DPRINTF(SimpleCPU, "Fault occured, scheduling fetch event\n");
627 reschedule(fetchEvent, nextCycle(), true);
628 _status = Faulting;
629 return;
630 }
631
632
633 if (!stayAtPC)
634 advancePC(fault);
635
636 if (_status == Running) {
637 // kick off fetch of next instruction... callback from icache
638 // response will cause that instruction to be executed,
639 // keeping the CPU running.
640 fetch();
641 }
642}
643
644
645void
646TimingSimpleCPU::completeIfetch(PacketPtr pkt)
647{
648 DPRINTF(SimpleCPU, "Complete ICache Fetch for addr %#x\n", pkt ?
649 pkt->getAddr() : 0);
650
651 // received a response from the icache: execute the received
652 // instruction
653
654 assert(!pkt || !pkt->isError());
655 assert(_status == IcacheWaitResponse);
656
657 _status = Running;
658
659 numCycles += tickToCycles(curTick() - previousTick);
660 previousTick = curTick();
661
662 if (getState() == SimObject::Draining) {
663 if (pkt) {
664 delete pkt->req;
665 delete pkt;
666 }
667
668 completeDrain();
669 return;
670 }
671
672 preExecute();
673 if (curStaticInst && curStaticInst->isMemRef()) {
674 // load or store: just send to dcache
675 Fault fault = curStaticInst->initiateAcc(this, traceData);
676
677 // If we're not running now the instruction will complete in a dcache
678 // response callback or the instruction faulted and has started an
679 // ifetch
680 if (_status == Running) {
681 if (fault != NoFault && traceData) {
682 // If there was a fault, we shouldn't trace this instruction.
683 delete traceData;
684 traceData = NULL;
685 }
686
687 postExecute();
688 // @todo remove me after debugging with legion done
689 if (curStaticInst && (!curStaticInst->isMicroop() ||
690 curStaticInst->isFirstMicroop()))
691 instCnt++;
692 advanceInst(fault);
693 }
694 } else if (curStaticInst) {
695 // non-memory instruction: execute completely now
696 Fault fault = curStaticInst->execute(this, traceData);
697
698 // keep an instruction count
699 if (fault == NoFault)
700 countInst();
701 else if (traceData && !DTRACE(ExecFaulting)) {
702 delete traceData;
703 traceData = NULL;
704 }
705
706 postExecute();
707 // @todo remove me after debugging with legion done
708 if (curStaticInst && (!curStaticInst->isMicroop() ||
709 curStaticInst->isFirstMicroop()))
710 instCnt++;
711 advanceInst(fault);
712 } else {
713 advanceInst(NoFault);
714 }
715
716 if (pkt) {
717 delete pkt->req;
718 delete pkt;
719 }
720}
721
722void
723TimingSimpleCPU::IcachePort::ITickEvent::process()
724{
725 cpu->completeIfetch(pkt);
726}
727
728bool
729TimingSimpleCPU::IcachePort::recvTiming(PacketPtr pkt)
730{
731 if (pkt->isResponse() && !pkt->wasNacked()) {
732 DPRINTF(SimpleCPU, "Received timing response %#x\n", pkt->getAddr());
733 // delay processing of returned data until next CPU clock edge
734 Tick next_tick = cpu->nextCycle(curTick());
735
736 if (next_tick == curTick())
737 cpu->completeIfetch(pkt);
738 else
739 tickEvent.schedule(pkt, next_tick);
740
741 return true;
742 } else if (pkt->wasNacked()) {
743 assert(cpu->_status == IcacheWaitResponse);
744 pkt->reinitNacked();
745 if (!sendTiming(pkt)) {
746 cpu->_status = IcacheRetry;
747 cpu->ifetch_pkt = pkt;
748 }
749 }
750 //Snooping a Coherence Request, do nothing
751 return true;
752}
753
754void
755TimingSimpleCPU::IcachePort::recvRetry()
756{
757 // we shouldn't get a retry unless we have a packet that we're
758 // waiting to transmit
759 assert(cpu->ifetch_pkt != NULL);
760 assert(cpu->_status == IcacheRetry);
761 PacketPtr tmp = cpu->ifetch_pkt;
762 if (sendTiming(tmp)) {
763 cpu->_status = IcacheWaitResponse;
764 cpu->ifetch_pkt = NULL;
765 }
766}
767
768void
769TimingSimpleCPU::completeDataAccess(PacketPtr pkt)
770{
771 // received a response from the dcache: complete the load or store
772 // instruction
773 assert(!pkt->isError());
774 assert(_status == DcacheWaitResponse || _status == DTBWaitResponse ||
775 pkt->req->getFlags().isSet(Request::NO_ACCESS));
776
777 numCycles += tickToCycles(curTick() - previousTick);
778 previousTick = curTick();
779
780 if (pkt->senderState) {
781 SplitFragmentSenderState * send_state =
782 dynamic_cast<SplitFragmentSenderState *>(pkt->senderState);
783 assert(send_state);
784 delete pkt->req;
785 delete pkt;
786 PacketPtr big_pkt = send_state->bigPkt;
787 delete send_state;
788
789 SplitMainSenderState * main_send_state =
790 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
791 assert(main_send_state);
792 // Record the fact that this packet is no longer outstanding.
793 assert(main_send_state->outstanding != 0);
794 main_send_state->outstanding--;
795
796 if (main_send_state->outstanding) {
797 return;
798 } else {
799 delete main_send_state;
800 big_pkt->senderState = NULL;
801 pkt = big_pkt;
802 }
803 }
804
805 _status = Running;
806
807 Fault fault = curStaticInst->completeAcc(pkt, this, traceData);
808
809 // keep an instruction count
810 if (fault == NoFault)
811 countInst();
812 else if (traceData) {
813 // If there was a fault, we shouldn't trace this instruction.
814 delete traceData;
815 traceData = NULL;
816 }
817
818 // the locked flag may be cleared on the response packet, so check
819 // pkt->req and not pkt to see if it was a load-locked
820 if (pkt->isRead() && pkt->req->isLLSC()) {
821 TheISA::handleLockedRead(thread, pkt->req);
822 }
823
824 delete pkt->req;
825 delete pkt;
826
827 postExecute();
828
829 if (getState() == SimObject::Draining) {
830 advancePC(fault);
831 completeDrain();
832
833 return;
834 }
835
836 advanceInst(fault);
837}
838
839
840void
841TimingSimpleCPU::completeDrain()
842{
843 DPRINTF(Config, "Done draining\n");
844 changeState(SimObject::Drained);
845 drainEvent->process();
846}
847
848bool
849TimingSimpleCPU::DcachePort::recvTiming(PacketPtr pkt)
850{
851 if (pkt->isResponse() && !pkt->wasNacked()) {
852 // delay processing of returned data until next CPU clock edge
853 Tick next_tick = cpu->nextCycle(curTick());
854
855 if (next_tick == curTick()) {
856 cpu->completeDataAccess(pkt);
857 } else {
858 if (!tickEvent.scheduled()) {
859 tickEvent.schedule(pkt, next_tick);
860 } else {
861 // In the case of a split transaction and a cache that is
862 // faster than a CPU we could get two responses before
863 // next_tick expires
864 if (!retryEvent.scheduled())
865 cpu->schedule(retryEvent, next_tick);
866 return false;
867 }
868 }
869
870 return true;
871 }
872 else if (pkt->wasNacked()) {
873 assert(cpu->_status == DcacheWaitResponse);
874 pkt->reinitNacked();
875 if (!sendTiming(pkt)) {
876 cpu->_status = DcacheRetry;
877 cpu->dcache_pkt = pkt;
878 }
879 }
880 //Snooping a Coherence Request, do nothing
881 return true;
882}
883
884void
885TimingSimpleCPU::DcachePort::DTickEvent::process()
886{
887 cpu->completeDataAccess(pkt);
888}
889
890void
891TimingSimpleCPU::DcachePort::recvRetry()
892{
893 // we shouldn't get a retry unless we have a packet that we're
894 // waiting to transmit
895 assert(cpu->dcache_pkt != NULL);
896 assert(cpu->_status == DcacheRetry);
897 PacketPtr tmp = cpu->dcache_pkt;
898 if (tmp->senderState) {
899 // This is a packet from a split access.
900 SplitFragmentSenderState * send_state =
901 dynamic_cast<SplitFragmentSenderState *>(tmp->senderState);
902 assert(send_state);
903 PacketPtr big_pkt = send_state->bigPkt;
904
905 SplitMainSenderState * main_send_state =
906 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
907 assert(main_send_state);
908
909 if (sendTiming(tmp)) {
910 // If we were able to send without retrying, record that fact
911 // and try sending the other fragment.
912 send_state->clearFromParent();
913 int other_index = main_send_state->getPendingFragment();
914 if (other_index > 0) {
915 tmp = main_send_state->fragments[other_index];
916 cpu->dcache_pkt = tmp;
917 if ((big_pkt->isRead() && cpu->handleReadPacket(tmp)) ||
918 (big_pkt->isWrite() && cpu->handleWritePacket())) {
919 main_send_state->fragments[other_index] = NULL;
920 }
921 } else {
922 cpu->_status = DcacheWaitResponse;
923 // memory system takes ownership of packet
924 cpu->dcache_pkt = NULL;
925 }
926 }
927 } else if (sendTiming(tmp)) {
928 cpu->_status = DcacheWaitResponse;
929 // memory system takes ownership of packet
930 cpu->dcache_pkt = NULL;
931 }
932}
933
934TimingSimpleCPU::IprEvent::IprEvent(Packet *_pkt, TimingSimpleCPU *_cpu,
935 Tick t)
936 : pkt(_pkt), cpu(_cpu)
937{
938 cpu->schedule(this, t);
939}
940
941void
942TimingSimpleCPU::IprEvent::process()
943{
944 cpu->completeDataAccess(pkt);
945}
946
947const char *
948TimingSimpleCPU::IprEvent::description() const
949{
950 return "Timing Simple CPU Delay IPR event";
951}
952
953
954void
955TimingSimpleCPU::printAddr(Addr a)
956{
957 dcachePort.printAddr(a);
958}
959
960
961////////////////////////////////////////////////////////////////////////
962//
963// TimingSimpleCPU Simulation Object
964//
965TimingSimpleCPU *
966TimingSimpleCPUParams::create()
967{
968 numThreads = 1;
969 if (!FullSystem && workload.size() != 1)
970 panic("only one workload allowed");
971 return new TimingSimpleCPU(this);
972}
63void
64TimingSimpleCPU::init()
65{
66 BaseCPU::init();
67 if (FullSystem) {
68 for (int i = 0; i < threadContexts.size(); ++i) {
69 ThreadContext *tc = threadContexts[i];
70 // initialize CPU, including PC
71 TheISA::initCPU(tc, _cpuId);
72 }
73 }
74
75 // Initialise the ThreadContext's memory proxies
76 tcBase()->initMemProxies(tcBase());
77}
78
79void
80TimingSimpleCPU::TimingCPUPort::TickEvent::schedule(PacketPtr _pkt, Tick t)
81{
82 pkt = _pkt;
83 cpu->schedule(this, t);
84}
85
86TimingSimpleCPU::TimingSimpleCPU(TimingSimpleCPUParams *p)
87 : BaseSimpleCPU(p), fetchTranslation(this), icachePort(this),
88 dcachePort(this), fetchEvent(this)
89{
90 _status = Idle;
91
92 ifetch_pkt = dcache_pkt = NULL;
93 drainEvent = NULL;
94 previousTick = 0;
95 changeState(SimObject::Running);
96 system->totalNumInsts = 0;
97}
98
99
100TimingSimpleCPU::~TimingSimpleCPU()
101{
102}
103
104void
105TimingSimpleCPU::serialize(ostream &os)
106{
107 SimObject::State so_state = SimObject::getState();
108 SERIALIZE_ENUM(so_state);
109 BaseSimpleCPU::serialize(os);
110}
111
112void
113TimingSimpleCPU::unserialize(Checkpoint *cp, const string &section)
114{
115 SimObject::State so_state;
116 UNSERIALIZE_ENUM(so_state);
117 BaseSimpleCPU::unserialize(cp, section);
118}
119
120unsigned int
121TimingSimpleCPU::drain(Event *drain_event)
122{
123 // TimingSimpleCPU is ready to drain if it's not waiting for
124 // an access to complete.
125 if (_status == Idle || _status == Running || _status == SwitchedOut) {
126 changeState(SimObject::Drained);
127 return 0;
128 } else {
129 changeState(SimObject::Draining);
130 drainEvent = drain_event;
131 return 1;
132 }
133}
134
135void
136TimingSimpleCPU::resume()
137{
138 DPRINTF(SimpleCPU, "Resume\n");
139 if (_status != SwitchedOut && _status != Idle) {
140 assert(system->getMemoryMode() == Enums::timing);
141
142 if (fetchEvent.scheduled())
143 deschedule(fetchEvent);
144
145 schedule(fetchEvent, nextCycle());
146 }
147
148 changeState(SimObject::Running);
149}
150
151void
152TimingSimpleCPU::switchOut()
153{
154 assert(_status == Running || _status == Idle);
155 _status = SwitchedOut;
156 numCycles += tickToCycles(curTick() - previousTick);
157
158 // If we've been scheduled to resume but are then told to switch out,
159 // we'll need to cancel it.
160 if (fetchEvent.scheduled())
161 deschedule(fetchEvent);
162}
163
164
165void
166TimingSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
167{
168 BaseCPU::takeOverFrom(oldCPU);
169
170 // if any of this CPU's ThreadContexts are active, mark the CPU as
171 // running and schedule its tick event.
172 for (int i = 0; i < threadContexts.size(); ++i) {
173 ThreadContext *tc = threadContexts[i];
174 if (tc->status() == ThreadContext::Active && _status != Running) {
175 _status = Running;
176 break;
177 }
178 }
179
180 if (_status != Running) {
181 _status = Idle;
182 }
183 assert(threadContexts.size() == 1);
184 previousTick = curTick();
185}
186
187
188void
189TimingSimpleCPU::activateContext(ThreadID thread_num, int delay)
190{
191 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);
192
193 assert(thread_num == 0);
194 assert(thread);
195
196 assert(_status == Idle);
197
198 notIdleFraction++;
199 _status = Running;
200
201 // kick things off by initiating the fetch of the next instruction
202 schedule(fetchEvent, nextCycle(curTick() + ticks(delay)));
203}
204
205
206void
207TimingSimpleCPU::suspendContext(ThreadID thread_num)
208{
209 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);
210
211 assert(thread_num == 0);
212 assert(thread);
213
214 if (_status == Idle)
215 return;
216
217 assert(_status == Running);
218
219 // just change status to Idle... if status != Running,
220 // completeInst() will not initiate fetch of next instruction.
221
222 notIdleFraction--;
223 _status = Idle;
224}
225
226bool
227TimingSimpleCPU::handleReadPacket(PacketPtr pkt)
228{
229 RequestPtr req = pkt->req;
230 if (req->isMmappedIpr()) {
231 Tick delay;
232 delay = TheISA::handleIprRead(thread->getTC(), pkt);
233 new IprEvent(pkt, this, nextCycle(curTick() + delay));
234 _status = DcacheWaitResponse;
235 dcache_pkt = NULL;
236 } else if (!dcachePort.sendTiming(pkt)) {
237 _status = DcacheRetry;
238 dcache_pkt = pkt;
239 } else {
240 _status = DcacheWaitResponse;
241 // memory system takes ownership of packet
242 dcache_pkt = NULL;
243 }
244 return dcache_pkt == NULL;
245}
246
247void
248TimingSimpleCPU::sendData(RequestPtr req, uint8_t *data, uint64_t *res,
249 bool read)
250{
251 PacketPtr pkt;
252 buildPacket(pkt, req, read);
253 pkt->dataDynamicArray<uint8_t>(data);
254 if (req->getFlags().isSet(Request::NO_ACCESS)) {
255 assert(!dcache_pkt);
256 pkt->makeResponse();
257 completeDataAccess(pkt);
258 } else if (read) {
259 handleReadPacket(pkt);
260 } else {
261 bool do_access = true; // flag to suppress cache access
262
263 if (req->isLLSC()) {
264 do_access = TheISA::handleLockedWrite(thread, req);
265 } else if (req->isCondSwap()) {
266 assert(res);
267 req->setExtraData(*res);
268 }
269
270 if (do_access) {
271 dcache_pkt = pkt;
272 handleWritePacket();
273 } else {
274 _status = DcacheWaitResponse;
275 completeDataAccess(pkt);
276 }
277 }
278}
279
280void
281TimingSimpleCPU::sendSplitData(RequestPtr req1, RequestPtr req2,
282 RequestPtr req, uint8_t *data, bool read)
283{
284 PacketPtr pkt1, pkt2;
285 buildSplitPacket(pkt1, pkt2, req1, req2, req, data, read);
286 if (req->getFlags().isSet(Request::NO_ACCESS)) {
287 assert(!dcache_pkt);
288 pkt1->makeResponse();
289 completeDataAccess(pkt1);
290 } else if (read) {
291 SplitFragmentSenderState * send_state =
292 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
293 if (handleReadPacket(pkt1)) {
294 send_state->clearFromParent();
295 send_state = dynamic_cast<SplitFragmentSenderState *>(
296 pkt2->senderState);
297 if (handleReadPacket(pkt2)) {
298 send_state->clearFromParent();
299 }
300 }
301 } else {
302 dcache_pkt = pkt1;
303 SplitFragmentSenderState * send_state =
304 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
305 if (handleWritePacket()) {
306 send_state->clearFromParent();
307 dcache_pkt = pkt2;
308 send_state = dynamic_cast<SplitFragmentSenderState *>(
309 pkt2->senderState);
310 if (handleWritePacket()) {
311 send_state->clearFromParent();
312 }
313 }
314 }
315}
316
317void
318TimingSimpleCPU::translationFault(Fault fault)
319{
320 // fault may be NoFault in cases where a fault is suppressed,
321 // for instance prefetches.
322 numCycles += tickToCycles(curTick() - previousTick);
323 previousTick = curTick();
324
325 if (traceData) {
326 // Since there was a fault, we shouldn't trace this instruction.
327 delete traceData;
328 traceData = NULL;
329 }
330
331 postExecute();
332
333 if (getState() == SimObject::Draining) {
334 advancePC(fault);
335 completeDrain();
336 } else {
337 advanceInst(fault);
338 }
339}
340
341void
342TimingSimpleCPU::buildPacket(PacketPtr &pkt, RequestPtr req, bool read)
343{
344 MemCmd cmd;
345 if (read) {
346 cmd = MemCmd::ReadReq;
347 if (req->isLLSC())
348 cmd = MemCmd::LoadLockedReq;
349 } else {
350 cmd = MemCmd::WriteReq;
351 if (req->isLLSC()) {
352 cmd = MemCmd::StoreCondReq;
353 } else if (req->isSwap()) {
354 cmd = MemCmd::SwapReq;
355 }
356 }
357 pkt = new Packet(req, cmd, Packet::Broadcast);
358}
359
360void
361TimingSimpleCPU::buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
362 RequestPtr req1, RequestPtr req2, RequestPtr req,
363 uint8_t *data, bool read)
364{
365 pkt1 = pkt2 = NULL;
366
367 assert(!req1->isMmappedIpr() && !req2->isMmappedIpr());
368
369 if (req->getFlags().isSet(Request::NO_ACCESS)) {
370 buildPacket(pkt1, req, read);
371 return;
372 }
373
374 buildPacket(pkt1, req1, read);
375 buildPacket(pkt2, req2, read);
376
377 req->setPhys(req1->getPaddr(), req->getSize(), req1->getFlags(), dataMasterId());
378 PacketPtr pkt = new Packet(req, pkt1->cmd.responseCommand(),
379 Packet::Broadcast);
380
381 pkt->dataDynamicArray<uint8_t>(data);
382 pkt1->dataStatic<uint8_t>(data);
383 pkt2->dataStatic<uint8_t>(data + req1->getSize());
384
385 SplitMainSenderState * main_send_state = new SplitMainSenderState;
386 pkt->senderState = main_send_state;
387 main_send_state->fragments[0] = pkt1;
388 main_send_state->fragments[1] = pkt2;
389 main_send_state->outstanding = 2;
390 pkt1->senderState = new SplitFragmentSenderState(pkt, 0);
391 pkt2->senderState = new SplitFragmentSenderState(pkt, 1);
392}
393
394Fault
395TimingSimpleCPU::readMem(Addr addr, uint8_t *data,
396 unsigned size, unsigned flags)
397{
398 Fault fault;
399 const int asid = 0;
400 const ThreadID tid = 0;
401 const Addr pc = thread->instAddr();
402 unsigned block_size = dcachePort.peerBlockSize();
403 BaseTLB::Mode mode = BaseTLB::Read;
404
405 if (traceData) {
406 traceData->setAddr(addr);
407 }
408
409 RequestPtr req = new Request(asid, addr, size,
410 flags, dataMasterId(), pc, _cpuId, tid);
411
412 Addr split_addr = roundDown(addr + size - 1, block_size);
413 assert(split_addr <= addr || split_addr - addr < block_size);
414
415 _status = DTBWaitResponse;
416 if (split_addr > addr) {
417 RequestPtr req1, req2;
418 assert(!req->isLLSC() && !req->isSwap());
419 req->splitOnVaddr(split_addr, req1, req2);
420
421 WholeTranslationState *state =
422 new WholeTranslationState(req, req1, req2, new uint8_t[size],
423 NULL, mode);
424 DataTranslation<TimingSimpleCPU *> *trans1 =
425 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
426 DataTranslation<TimingSimpleCPU *> *trans2 =
427 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
428
429 thread->dtb->translateTiming(req1, tc, trans1, mode);
430 thread->dtb->translateTiming(req2, tc, trans2, mode);
431 } else {
432 WholeTranslationState *state =
433 new WholeTranslationState(req, new uint8_t[size], NULL, mode);
434 DataTranslation<TimingSimpleCPU *> *translation
435 = new DataTranslation<TimingSimpleCPU *>(this, state);
436 thread->dtb->translateTiming(req, tc, translation, mode);
437 }
438
439 return NoFault;
440}
441
442bool
443TimingSimpleCPU::handleWritePacket()
444{
445 RequestPtr req = dcache_pkt->req;
446 if (req->isMmappedIpr()) {
447 Tick delay;
448 delay = TheISA::handleIprWrite(thread->getTC(), dcache_pkt);
449 new IprEvent(dcache_pkt, this, nextCycle(curTick() + delay));
450 _status = DcacheWaitResponse;
451 dcache_pkt = NULL;
452 } else if (!dcachePort.sendTiming(dcache_pkt)) {
453 _status = DcacheRetry;
454 } else {
455 _status = DcacheWaitResponse;
456 // memory system takes ownership of packet
457 dcache_pkt = NULL;
458 }
459 return dcache_pkt == NULL;
460}
461
462Fault
463TimingSimpleCPU::writeMem(uint8_t *data, unsigned size,
464 Addr addr, unsigned flags, uint64_t *res)
465{
466 uint8_t *newData = new uint8_t[size];
467 memcpy(newData, data, size);
468
469 const int asid = 0;
470 const ThreadID tid = 0;
471 const Addr pc = thread->instAddr();
472 unsigned block_size = dcachePort.peerBlockSize();
473 BaseTLB::Mode mode = BaseTLB::Write;
474
475 if (traceData) {
476 traceData->setAddr(addr);
477 }
478
479 RequestPtr req = new Request(asid, addr, size,
480 flags, dataMasterId(), pc, _cpuId, tid);
481
482 Addr split_addr = roundDown(addr + size - 1, block_size);
483 assert(split_addr <= addr || split_addr - addr < block_size);
484
485 _status = DTBWaitResponse;
486 if (split_addr > addr) {
487 RequestPtr req1, req2;
488 assert(!req->isLLSC() && !req->isSwap());
489 req->splitOnVaddr(split_addr, req1, req2);
490
491 WholeTranslationState *state =
492 new WholeTranslationState(req, req1, req2, newData, res, mode);
493 DataTranslation<TimingSimpleCPU *> *trans1 =
494 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
495 DataTranslation<TimingSimpleCPU *> *trans2 =
496 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
497
498 thread->dtb->translateTiming(req1, tc, trans1, mode);
499 thread->dtb->translateTiming(req2, tc, trans2, mode);
500 } else {
501 WholeTranslationState *state =
502 new WholeTranslationState(req, newData, res, mode);
503 DataTranslation<TimingSimpleCPU *> *translation =
504 new DataTranslation<TimingSimpleCPU *>(this, state);
505 thread->dtb->translateTiming(req, tc, translation, mode);
506 }
507
508 // Translation faults will be returned via finishTranslation()
509 return NoFault;
510}
511
512
513void
514TimingSimpleCPU::finishTranslation(WholeTranslationState *state)
515{
516 _status = Running;
517
518 if (state->getFault() != NoFault) {
519 if (state->isPrefetch()) {
520 state->setNoFault();
521 }
522 delete [] state->data;
523 state->deleteReqs();
524 translationFault(state->getFault());
525 } else {
526 if (!state->isSplit) {
527 sendData(state->mainReq, state->data, state->res,
528 state->mode == BaseTLB::Read);
529 } else {
530 sendSplitData(state->sreqLow, state->sreqHigh, state->mainReq,
531 state->data, state->mode == BaseTLB::Read);
532 }
533 }
534
535 delete state;
536}
537
538
539void
540TimingSimpleCPU::fetch()
541{
542 DPRINTF(SimpleCPU, "Fetch\n");
543
544 if (!curStaticInst || !curStaticInst->isDelayedCommit())
545 checkForInterrupts();
546
547 checkPcEventQueue();
548
549 // We must have just got suspended by a PC event
550 if (_status == Idle)
551 return;
552
553 TheISA::PCState pcState = thread->pcState();
554 bool needToFetch = !isRomMicroPC(pcState.microPC()) && !curMacroStaticInst;
555
556 if (needToFetch) {
557 _status = Running;
558 Request *ifetch_req = new Request();
559 ifetch_req->setThreadContext(_cpuId, /* thread ID */ 0);
560 setupFetchRequest(ifetch_req);
561 DPRINTF(SimpleCPU, "Translating address %#x\n", ifetch_req->getVaddr());
562 thread->itb->translateTiming(ifetch_req, tc, &fetchTranslation,
563 BaseTLB::Execute);
564 } else {
565 _status = IcacheWaitResponse;
566 completeIfetch(NULL);
567
568 numCycles += tickToCycles(curTick() - previousTick);
569 previousTick = curTick();
570 }
571}
572
573
574void
575TimingSimpleCPU::sendFetch(Fault fault, RequestPtr req, ThreadContext *tc)
576{
577 if (fault == NoFault) {
578 DPRINTF(SimpleCPU, "Sending fetch for addr %#x(pa: %#x)\n",
579 req->getVaddr(), req->getPaddr());
580 ifetch_pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast);
581 ifetch_pkt->dataStatic(&inst);
582 DPRINTF(SimpleCPU, " -- pkt addr: %#x\n", ifetch_pkt->getAddr());
583
584 if (!icachePort.sendTiming(ifetch_pkt)) {
585 // Need to wait for retry
586 _status = IcacheRetry;
587 } else {
588 // Need to wait for cache to respond
589 _status = IcacheWaitResponse;
590 // ownership of packet transferred to memory system
591 ifetch_pkt = NULL;
592 }
593 } else {
594 DPRINTF(SimpleCPU, "Translation of addr %#x faulted\n", req->getVaddr());
595 delete req;
596 // fetch fault: advance directly to next instruction (fault handler)
597 _status = Running;
598 advanceInst(fault);
599 }
600
601 numCycles += tickToCycles(curTick() - previousTick);
602 previousTick = curTick();
603}
604
605
606void
607TimingSimpleCPU::advanceInst(Fault fault)
608{
609
610 if (_status == Faulting)
611 return;
612
613 if (fault != NoFault) {
614 advancePC(fault);
615 DPRINTF(SimpleCPU, "Fault occured, scheduling fetch event\n");
616 reschedule(fetchEvent, nextCycle(), true);
617 _status = Faulting;
618 return;
619 }
620
621
622 if (!stayAtPC)
623 advancePC(fault);
624
625 if (_status == Running) {
626 // kick off fetch of next instruction... callback from icache
627 // response will cause that instruction to be executed,
628 // keeping the CPU running.
629 fetch();
630 }
631}
632
633
634void
635TimingSimpleCPU::completeIfetch(PacketPtr pkt)
636{
637 DPRINTF(SimpleCPU, "Complete ICache Fetch for addr %#x\n", pkt ?
638 pkt->getAddr() : 0);
639
640 // received a response from the icache: execute the received
641 // instruction
642
643 assert(!pkt || !pkt->isError());
644 assert(_status == IcacheWaitResponse);
645
646 _status = Running;
647
648 numCycles += tickToCycles(curTick() - previousTick);
649 previousTick = curTick();
650
651 if (getState() == SimObject::Draining) {
652 if (pkt) {
653 delete pkt->req;
654 delete pkt;
655 }
656
657 completeDrain();
658 return;
659 }
660
661 preExecute();
662 if (curStaticInst && curStaticInst->isMemRef()) {
663 // load or store: just send to dcache
664 Fault fault = curStaticInst->initiateAcc(this, traceData);
665
666 // If we're not running now the instruction will complete in a dcache
667 // response callback or the instruction faulted and has started an
668 // ifetch
669 if (_status == Running) {
670 if (fault != NoFault && traceData) {
671 // If there was a fault, we shouldn't trace this instruction.
672 delete traceData;
673 traceData = NULL;
674 }
675
676 postExecute();
677 // @todo remove me after debugging with legion done
678 if (curStaticInst && (!curStaticInst->isMicroop() ||
679 curStaticInst->isFirstMicroop()))
680 instCnt++;
681 advanceInst(fault);
682 }
683 } else if (curStaticInst) {
684 // non-memory instruction: execute completely now
685 Fault fault = curStaticInst->execute(this, traceData);
686
687 // keep an instruction count
688 if (fault == NoFault)
689 countInst();
690 else if (traceData && !DTRACE(ExecFaulting)) {
691 delete traceData;
692 traceData = NULL;
693 }
694
695 postExecute();
696 // @todo remove me after debugging with legion done
697 if (curStaticInst && (!curStaticInst->isMicroop() ||
698 curStaticInst->isFirstMicroop()))
699 instCnt++;
700 advanceInst(fault);
701 } else {
702 advanceInst(NoFault);
703 }
704
705 if (pkt) {
706 delete pkt->req;
707 delete pkt;
708 }
709}
710
711void
712TimingSimpleCPU::IcachePort::ITickEvent::process()
713{
714 cpu->completeIfetch(pkt);
715}
716
717bool
718TimingSimpleCPU::IcachePort::recvTiming(PacketPtr pkt)
719{
720 if (pkt->isResponse() && !pkt->wasNacked()) {
721 DPRINTF(SimpleCPU, "Received timing response %#x\n", pkt->getAddr());
722 // delay processing of returned data until next CPU clock edge
723 Tick next_tick = cpu->nextCycle(curTick());
724
725 if (next_tick == curTick())
726 cpu->completeIfetch(pkt);
727 else
728 tickEvent.schedule(pkt, next_tick);
729
730 return true;
731 } else if (pkt->wasNacked()) {
732 assert(cpu->_status == IcacheWaitResponse);
733 pkt->reinitNacked();
734 if (!sendTiming(pkt)) {
735 cpu->_status = IcacheRetry;
736 cpu->ifetch_pkt = pkt;
737 }
738 }
739 //Snooping a Coherence Request, do nothing
740 return true;
741}
742
743void
744TimingSimpleCPU::IcachePort::recvRetry()
745{
746 // we shouldn't get a retry unless we have a packet that we're
747 // waiting to transmit
748 assert(cpu->ifetch_pkt != NULL);
749 assert(cpu->_status == IcacheRetry);
750 PacketPtr tmp = cpu->ifetch_pkt;
751 if (sendTiming(tmp)) {
752 cpu->_status = IcacheWaitResponse;
753 cpu->ifetch_pkt = NULL;
754 }
755}
756
757void
758TimingSimpleCPU::completeDataAccess(PacketPtr pkt)
759{
760 // received a response from the dcache: complete the load or store
761 // instruction
762 assert(!pkt->isError());
763 assert(_status == DcacheWaitResponse || _status == DTBWaitResponse ||
764 pkt->req->getFlags().isSet(Request::NO_ACCESS));
765
766 numCycles += tickToCycles(curTick() - previousTick);
767 previousTick = curTick();
768
769 if (pkt->senderState) {
770 SplitFragmentSenderState * send_state =
771 dynamic_cast<SplitFragmentSenderState *>(pkt->senderState);
772 assert(send_state);
773 delete pkt->req;
774 delete pkt;
775 PacketPtr big_pkt = send_state->bigPkt;
776 delete send_state;
777
778 SplitMainSenderState * main_send_state =
779 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
780 assert(main_send_state);
781 // Record the fact that this packet is no longer outstanding.
782 assert(main_send_state->outstanding != 0);
783 main_send_state->outstanding--;
784
785 if (main_send_state->outstanding) {
786 return;
787 } else {
788 delete main_send_state;
789 big_pkt->senderState = NULL;
790 pkt = big_pkt;
791 }
792 }
793
794 _status = Running;
795
796 Fault fault = curStaticInst->completeAcc(pkt, this, traceData);
797
798 // keep an instruction count
799 if (fault == NoFault)
800 countInst();
801 else if (traceData) {
802 // If there was a fault, we shouldn't trace this instruction.
803 delete traceData;
804 traceData = NULL;
805 }
806
807 // the locked flag may be cleared on the response packet, so check
808 // pkt->req and not pkt to see if it was a load-locked
809 if (pkt->isRead() && pkt->req->isLLSC()) {
810 TheISA::handleLockedRead(thread, pkt->req);
811 }
812
813 delete pkt->req;
814 delete pkt;
815
816 postExecute();
817
818 if (getState() == SimObject::Draining) {
819 advancePC(fault);
820 completeDrain();
821
822 return;
823 }
824
825 advanceInst(fault);
826}
827
828
829void
830TimingSimpleCPU::completeDrain()
831{
832 DPRINTF(Config, "Done draining\n");
833 changeState(SimObject::Drained);
834 drainEvent->process();
835}
836
837bool
838TimingSimpleCPU::DcachePort::recvTiming(PacketPtr pkt)
839{
840 if (pkt->isResponse() && !pkt->wasNacked()) {
841 // delay processing of returned data until next CPU clock edge
842 Tick next_tick = cpu->nextCycle(curTick());
843
844 if (next_tick == curTick()) {
845 cpu->completeDataAccess(pkt);
846 } else {
847 if (!tickEvent.scheduled()) {
848 tickEvent.schedule(pkt, next_tick);
849 } else {
850 // In the case of a split transaction and a cache that is
851 // faster than a CPU we could get two responses before
852 // next_tick expires
853 if (!retryEvent.scheduled())
854 cpu->schedule(retryEvent, next_tick);
855 return false;
856 }
857 }
858
859 return true;
860 }
861 else if (pkt->wasNacked()) {
862 assert(cpu->_status == DcacheWaitResponse);
863 pkt->reinitNacked();
864 if (!sendTiming(pkt)) {
865 cpu->_status = DcacheRetry;
866 cpu->dcache_pkt = pkt;
867 }
868 }
869 //Snooping a Coherence Request, do nothing
870 return true;
871}
872
873void
874TimingSimpleCPU::DcachePort::DTickEvent::process()
875{
876 cpu->completeDataAccess(pkt);
877}
878
879void
880TimingSimpleCPU::DcachePort::recvRetry()
881{
882 // we shouldn't get a retry unless we have a packet that we're
883 // waiting to transmit
884 assert(cpu->dcache_pkt != NULL);
885 assert(cpu->_status == DcacheRetry);
886 PacketPtr tmp = cpu->dcache_pkt;
887 if (tmp->senderState) {
888 // This is a packet from a split access.
889 SplitFragmentSenderState * send_state =
890 dynamic_cast<SplitFragmentSenderState *>(tmp->senderState);
891 assert(send_state);
892 PacketPtr big_pkt = send_state->bigPkt;
893
894 SplitMainSenderState * main_send_state =
895 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
896 assert(main_send_state);
897
898 if (sendTiming(tmp)) {
899 // If we were able to send without retrying, record that fact
900 // and try sending the other fragment.
901 send_state->clearFromParent();
902 int other_index = main_send_state->getPendingFragment();
903 if (other_index > 0) {
904 tmp = main_send_state->fragments[other_index];
905 cpu->dcache_pkt = tmp;
906 if ((big_pkt->isRead() && cpu->handleReadPacket(tmp)) ||
907 (big_pkt->isWrite() && cpu->handleWritePacket())) {
908 main_send_state->fragments[other_index] = NULL;
909 }
910 } else {
911 cpu->_status = DcacheWaitResponse;
912 // memory system takes ownership of packet
913 cpu->dcache_pkt = NULL;
914 }
915 }
916 } else if (sendTiming(tmp)) {
917 cpu->_status = DcacheWaitResponse;
918 // memory system takes ownership of packet
919 cpu->dcache_pkt = NULL;
920 }
921}
922
923TimingSimpleCPU::IprEvent::IprEvent(Packet *_pkt, TimingSimpleCPU *_cpu,
924 Tick t)
925 : pkt(_pkt), cpu(_cpu)
926{
927 cpu->schedule(this, t);
928}
929
930void
931TimingSimpleCPU::IprEvent::process()
932{
933 cpu->completeDataAccess(pkt);
934}
935
936const char *
937TimingSimpleCPU::IprEvent::description() const
938{
939 return "Timing Simple CPU Delay IPR event";
940}
941
942
943void
944TimingSimpleCPU::printAddr(Addr a)
945{
946 dcachePort.printAddr(a);
947}
948
949
950////////////////////////////////////////////////////////////////////////
951//
952// TimingSimpleCPU Simulation Object
953//
954TimingSimpleCPU *
955TimingSimpleCPUParams::create()
956{
957 numThreads = 1;
958 if (!FullSystem && workload.size() != 1)
959 panic("only one workload allowed");
960 return new TimingSimpleCPU(this);
961}