timing.cc (8706:b1838faf3bcc) timing.cc (8707:489489c67fd9)
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/system.hh"
58
59using namespace std;
60using namespace TheISA;
61
62Port *
63TimingSimpleCPU::getPort(const std::string &if_name, int idx)
64{
65 if (if_name == "dcache_port")
66 return &dcachePort;
67 else if (if_name == "icache_port")
68 return &icachePort;
69 else
70 panic("No Such Port\n");
71}
72
73void
74TimingSimpleCPU::init()
75{
76 BaseCPU::init();
77#if FULL_SYSTEM
78 for (int i = 0; i < threadContexts.size(); ++i) {
79 ThreadContext *tc = threadContexts[i];
80
81 // initialize CPU, including PC
82 TheISA::initCPU(tc, _cpuId);
83 }
84
85 // Initialise the ThreadContext's memory proxies
86 tcBase()->initMemProxies(tcBase());
87#endif
88}
89
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/system.hh"
58
59using namespace std;
60using namespace TheISA;
61
62Port *
63TimingSimpleCPU::getPort(const std::string &if_name, int idx)
64{
65 if (if_name == "dcache_port")
66 return &dcachePort;
67 else if (if_name == "icache_port")
68 return &icachePort;
69 else
70 panic("No Such Port\n");
71}
72
73void
74TimingSimpleCPU::init()
75{
76 BaseCPU::init();
77#if FULL_SYSTEM
78 for (int i = 0; i < threadContexts.size(); ++i) {
79 ThreadContext *tc = threadContexts[i];
80
81 // initialize CPU, including PC
82 TheISA::initCPU(tc, _cpuId);
83 }
84
85 // Initialise the ThreadContext's memory proxies
86 tcBase()->initMemProxies(tcBase());
87#endif
88}
89
90Tick
91TimingSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt)
92{
93 panic("TimingSimpleCPU doesn't expect recvAtomic callback!");
94 return curTick();
95}
96
97void
90void
98TimingSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt)
91TimingSimpleCPU::TimingCPUPort::TickEvent::schedule(PacketPtr _pkt, Tick t)
99{
92{
100 //No internal storage to update, jusst return
101 return;
102}
103
104void
105TimingSimpleCPU::CpuPort::recvStatusChange(Status status)
106{
107 if (status == RangeChange) {
108 if (!snoopRangeSent) {
109 snoopRangeSent = true;
110 sendStatusChange(Port::RangeChange);
111 }
112 return;
113 }
114
115 panic("TimingSimpleCPU doesn't expect recvStatusChange callback!");
116}
117
118
119void
120TimingSimpleCPU::CpuPort::TickEvent::schedule(PacketPtr _pkt, Tick t)
121{
122 pkt = _pkt;
123 cpu->schedule(this, t);
124}
125
126TimingSimpleCPU::TimingSimpleCPU(TimingSimpleCPUParams *p)
93 pkt = _pkt;
94 cpu->schedule(this, t);
95}
96
97TimingSimpleCPU::TimingSimpleCPU(TimingSimpleCPUParams *p)
127 : BaseSimpleCPU(p), fetchTranslation(this), icachePort(this, p->clock),
128 dcachePort(this, p->clock), fetchEvent(this)
98 : BaseSimpleCPU(p), fetchTranslation(this), icachePort(this),
99 dcachePort(this), fetchEvent(this)
129{
130 _status = Idle;
131
100{
101 _status = Idle;
102
132 icachePort.snoopRangeSent = false;
133 dcachePort.snoopRangeSent = false;
134
135 ifetch_pkt = dcache_pkt = NULL;
136 drainEvent = NULL;
137 previousTick = 0;
138 changeState(SimObject::Running);
139 system->totalNumInsts = 0;
140}
141
142
143TimingSimpleCPU::~TimingSimpleCPU()
144{
145}
146
147void
148TimingSimpleCPU::serialize(ostream &os)
149{
150 SimObject::State so_state = SimObject::getState();
151 SERIALIZE_ENUM(so_state);
152 BaseSimpleCPU::serialize(os);
153}
154
155void
156TimingSimpleCPU::unserialize(Checkpoint *cp, const string &section)
157{
158 SimObject::State so_state;
159 UNSERIALIZE_ENUM(so_state);
160 BaseSimpleCPU::unserialize(cp, section);
161}
162
163unsigned int
164TimingSimpleCPU::drain(Event *drain_event)
165{
166 // TimingSimpleCPU is ready to drain if it's not waiting for
167 // an access to complete.
168 if (_status == Idle || _status == Running || _status == SwitchedOut) {
169 changeState(SimObject::Drained);
170 return 0;
171 } else {
172 changeState(SimObject::Draining);
173 drainEvent = drain_event;
174 return 1;
175 }
176}
177
178void
179TimingSimpleCPU::resume()
180{
181 DPRINTF(SimpleCPU, "Resume\n");
182 if (_status != SwitchedOut && _status != Idle) {
183 assert(system->getMemoryMode() == Enums::timing);
184
185 if (fetchEvent.scheduled())
186 deschedule(fetchEvent);
187
188 schedule(fetchEvent, nextCycle());
189 }
190
191 changeState(SimObject::Running);
192}
193
194void
195TimingSimpleCPU::switchOut()
196{
197 assert(_status == Running || _status == Idle);
198 _status = SwitchedOut;
199 numCycles += tickToCycles(curTick() - previousTick);
200
201 // If we've been scheduled to resume but are then told to switch out,
202 // we'll need to cancel it.
203 if (fetchEvent.scheduled())
204 deschedule(fetchEvent);
205}
206
207
208void
209TimingSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
210{
211 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort);
212
213 // if any of this CPU's ThreadContexts are active, mark the CPU as
214 // running and schedule its tick event.
215 for (int i = 0; i < threadContexts.size(); ++i) {
216 ThreadContext *tc = threadContexts[i];
217 if (tc->status() == ThreadContext::Active && _status != Running) {
218 _status = Running;
219 break;
220 }
221 }
222
223 if (_status != Running) {
224 _status = Idle;
225 }
226 assert(threadContexts.size() == 1);
227 previousTick = curTick();
228}
229
230
231void
232TimingSimpleCPU::activateContext(int thread_num, int delay)
233{
234 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);
235
236 assert(thread_num == 0);
237 assert(thread);
238
239 assert(_status == Idle);
240
241 notIdleFraction++;
242 _status = Running;
243
244 // kick things off by initiating the fetch of the next instruction
245 schedule(fetchEvent, nextCycle(curTick() + ticks(delay)));
246}
247
248
249void
250TimingSimpleCPU::suspendContext(int thread_num)
251{
252 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);
253
254 assert(thread_num == 0);
255 assert(thread);
256
257 if (_status == Idle)
258 return;
259
260 assert(_status == Running);
261
262 // just change status to Idle... if status != Running,
263 // completeInst() will not initiate fetch of next instruction.
264
265 notIdleFraction--;
266 _status = Idle;
267}
268
269bool
270TimingSimpleCPU::handleReadPacket(PacketPtr pkt)
271{
272 RequestPtr req = pkt->req;
273 if (req->isMmappedIpr()) {
274 Tick delay;
275 delay = TheISA::handleIprRead(thread->getTC(), pkt);
276 new IprEvent(pkt, this, nextCycle(curTick() + delay));
277 _status = DcacheWaitResponse;
278 dcache_pkt = NULL;
279 } else if (!dcachePort.sendTiming(pkt)) {
280 _status = DcacheRetry;
281 dcache_pkt = pkt;
282 } else {
283 _status = DcacheWaitResponse;
284 // memory system takes ownership of packet
285 dcache_pkt = NULL;
286 }
287 return dcache_pkt == NULL;
288}
289
290void
291TimingSimpleCPU::sendData(RequestPtr req, uint8_t *data, uint64_t *res,
292 bool read)
293{
294 PacketPtr pkt;
295 buildPacket(pkt, req, read);
296 pkt->dataDynamicArray<uint8_t>(data);
297 if (req->getFlags().isSet(Request::NO_ACCESS)) {
298 assert(!dcache_pkt);
299 pkt->makeResponse();
300 completeDataAccess(pkt);
301 } else if (read) {
302 handleReadPacket(pkt);
303 } else {
304 bool do_access = true; // flag to suppress cache access
305
306 if (req->isLLSC()) {
307 do_access = TheISA::handleLockedWrite(thread, req);
308 } else if (req->isCondSwap()) {
309 assert(res);
310 req->setExtraData(*res);
311 }
312
313 if (do_access) {
314 dcache_pkt = pkt;
315 handleWritePacket();
316 } else {
317 _status = DcacheWaitResponse;
318 completeDataAccess(pkt);
319 }
320 }
321}
322
323void
324TimingSimpleCPU::sendSplitData(RequestPtr req1, RequestPtr req2,
325 RequestPtr req, uint8_t *data, bool read)
326{
327 PacketPtr pkt1, pkt2;
328 buildSplitPacket(pkt1, pkt2, req1, req2, req, data, read);
329 if (req->getFlags().isSet(Request::NO_ACCESS)) {
330 assert(!dcache_pkt);
331 pkt1->makeResponse();
332 completeDataAccess(pkt1);
333 } else if (read) {
334 SplitFragmentSenderState * send_state =
335 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
336 if (handleReadPacket(pkt1)) {
337 send_state->clearFromParent();
338 send_state = dynamic_cast<SplitFragmentSenderState *>(
339 pkt2->senderState);
340 if (handleReadPacket(pkt2)) {
341 send_state->clearFromParent();
342 }
343 }
344 } else {
345 dcache_pkt = pkt1;
346 SplitFragmentSenderState * send_state =
347 dynamic_cast<SplitFragmentSenderState *>(pkt1->senderState);
348 if (handleWritePacket()) {
349 send_state->clearFromParent();
350 dcache_pkt = pkt2;
351 send_state = dynamic_cast<SplitFragmentSenderState *>(
352 pkt2->senderState);
353 if (handleWritePacket()) {
354 send_state->clearFromParent();
355 }
356 }
357 }
358}
359
360void
361TimingSimpleCPU::translationFault(Fault fault)
362{
363 // fault may be NoFault in cases where a fault is suppressed,
364 // for instance prefetches.
365 numCycles += tickToCycles(curTick() - previousTick);
366 previousTick = curTick();
367
368 if (traceData) {
369 // Since there was a fault, we shouldn't trace this instruction.
370 delete traceData;
371 traceData = NULL;
372 }
373
374 postExecute();
375
376 if (getState() == SimObject::Draining) {
377 advancePC(fault);
378 completeDrain();
379 } else {
380 advanceInst(fault);
381 }
382}
383
384void
385TimingSimpleCPU::buildPacket(PacketPtr &pkt, RequestPtr req, bool read)
386{
387 MemCmd cmd;
388 if (read) {
389 cmd = MemCmd::ReadReq;
390 if (req->isLLSC())
391 cmd = MemCmd::LoadLockedReq;
392 } else {
393 cmd = MemCmd::WriteReq;
394 if (req->isLLSC()) {
395 cmd = MemCmd::StoreCondReq;
396 } else if (req->isSwap()) {
397 cmd = MemCmd::SwapReq;
398 }
399 }
400 pkt = new Packet(req, cmd, Packet::Broadcast);
401}
402
403void
404TimingSimpleCPU::buildSplitPacket(PacketPtr &pkt1, PacketPtr &pkt2,
405 RequestPtr req1, RequestPtr req2, RequestPtr req,
406 uint8_t *data, bool read)
407{
408 pkt1 = pkt2 = NULL;
409
410 assert(!req1->isMmappedIpr() && !req2->isMmappedIpr());
411
412 if (req->getFlags().isSet(Request::NO_ACCESS)) {
413 buildPacket(pkt1, req, read);
414 return;
415 }
416
417 buildPacket(pkt1, req1, read);
418 buildPacket(pkt2, req2, read);
419
420 req->setPhys(req1->getPaddr(), req->getSize(), req1->getFlags());
421 PacketPtr pkt = new Packet(req, pkt1->cmd.responseCommand(),
422 Packet::Broadcast);
423
424 pkt->dataDynamicArray<uint8_t>(data);
425 pkt1->dataStatic<uint8_t>(data);
426 pkt2->dataStatic<uint8_t>(data + req1->getSize());
427
428 SplitMainSenderState * main_send_state = new SplitMainSenderState;
429 pkt->senderState = main_send_state;
430 main_send_state->fragments[0] = pkt1;
431 main_send_state->fragments[1] = pkt2;
432 main_send_state->outstanding = 2;
433 pkt1->senderState = new SplitFragmentSenderState(pkt, 0);
434 pkt2->senderState = new SplitFragmentSenderState(pkt, 1);
435}
436
437Fault
438TimingSimpleCPU::readMem(Addr addr, uint8_t *data,
439 unsigned size, unsigned flags)
440{
441 Fault fault;
442 const int asid = 0;
443 const ThreadID tid = 0;
444 const Addr pc = thread->instAddr();
445 unsigned block_size = dcachePort.peerBlockSize();
446 BaseTLB::Mode mode = BaseTLB::Read;
447
448 if (traceData) {
449 traceData->setAddr(addr);
450 }
451
452 RequestPtr req = new Request(asid, addr, size,
453 flags, pc, _cpuId, tid);
454
455 Addr split_addr = roundDown(addr + size - 1, block_size);
456 assert(split_addr <= addr || split_addr - addr < block_size);
457
458 _status = DTBWaitResponse;
459 if (split_addr > addr) {
460 RequestPtr req1, req2;
461 assert(!req->isLLSC() && !req->isSwap());
462 req->splitOnVaddr(split_addr, req1, req2);
463
464 WholeTranslationState *state =
465 new WholeTranslationState(req, req1, req2, new uint8_t[size],
466 NULL, mode);
467 DataTranslation<TimingSimpleCPU *> *trans1 =
468 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
469 DataTranslation<TimingSimpleCPU *> *trans2 =
470 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
471
472 thread->dtb->translateTiming(req1, tc, trans1, mode);
473 thread->dtb->translateTiming(req2, tc, trans2, mode);
474 } else {
475 WholeTranslationState *state =
476 new WholeTranslationState(req, new uint8_t[size], NULL, mode);
477 DataTranslation<TimingSimpleCPU *> *translation
478 = new DataTranslation<TimingSimpleCPU *>(this, state);
479 thread->dtb->translateTiming(req, tc, translation, mode);
480 }
481
482 return NoFault;
483}
484
485bool
486TimingSimpleCPU::handleWritePacket()
487{
488 RequestPtr req = dcache_pkt->req;
489 if (req->isMmappedIpr()) {
490 Tick delay;
491 delay = TheISA::handleIprWrite(thread->getTC(), dcache_pkt);
492 new IprEvent(dcache_pkt, this, nextCycle(curTick() + delay));
493 _status = DcacheWaitResponse;
494 dcache_pkt = NULL;
495 } else if (!dcachePort.sendTiming(dcache_pkt)) {
496 _status = DcacheRetry;
497 } else {
498 _status = DcacheWaitResponse;
499 // memory system takes ownership of packet
500 dcache_pkt = NULL;
501 }
502 return dcache_pkt == NULL;
503}
504
505Fault
506TimingSimpleCPU::writeMem(uint8_t *data, unsigned size,
507 Addr addr, unsigned flags, uint64_t *res)
508{
509 uint8_t *newData = new uint8_t[size];
510 memcpy(newData, data, size);
511
512 const int asid = 0;
513 const ThreadID tid = 0;
514 const Addr pc = thread->instAddr();
515 unsigned block_size = dcachePort.peerBlockSize();
516 BaseTLB::Mode mode = BaseTLB::Write;
517
518 if (traceData) {
519 traceData->setAddr(addr);
520 }
521
522 RequestPtr req = new Request(asid, addr, size,
523 flags, pc, _cpuId, tid);
524
525 Addr split_addr = roundDown(addr + size - 1, block_size);
526 assert(split_addr <= addr || split_addr - addr < block_size);
527
528 _status = DTBWaitResponse;
529 if (split_addr > addr) {
530 RequestPtr req1, req2;
531 assert(!req->isLLSC() && !req->isSwap());
532 req->splitOnVaddr(split_addr, req1, req2);
533
534 WholeTranslationState *state =
535 new WholeTranslationState(req, req1, req2, newData, res, mode);
536 DataTranslation<TimingSimpleCPU *> *trans1 =
537 new DataTranslation<TimingSimpleCPU *>(this, state, 0);
538 DataTranslation<TimingSimpleCPU *> *trans2 =
539 new DataTranslation<TimingSimpleCPU *>(this, state, 1);
540
541 thread->dtb->translateTiming(req1, tc, trans1, mode);
542 thread->dtb->translateTiming(req2, tc, trans2, mode);
543 } else {
544 WholeTranslationState *state =
545 new WholeTranslationState(req, newData, res, mode);
546 DataTranslation<TimingSimpleCPU *> *translation =
547 new DataTranslation<TimingSimpleCPU *>(this, state);
548 thread->dtb->translateTiming(req, tc, translation, mode);
549 }
550
551 // Translation faults will be returned via finishTranslation()
552 return NoFault;
553}
554
555
556void
557TimingSimpleCPU::finishTranslation(WholeTranslationState *state)
558{
559 _status = Running;
560
561 if (state->getFault() != NoFault) {
562 if (state->isPrefetch()) {
563 state->setNoFault();
564 }
565 delete [] state->data;
566 state->deleteReqs();
567 translationFault(state->getFault());
568 } else {
569 if (!state->isSplit) {
570 sendData(state->mainReq, state->data, state->res,
571 state->mode == BaseTLB::Read);
572 } else {
573 sendSplitData(state->sreqLow, state->sreqHigh, state->mainReq,
574 state->data, state->mode == BaseTLB::Read);
575 }
576 }
577
578 delete state;
579}
580
581
582void
583TimingSimpleCPU::fetch()
584{
585 DPRINTF(SimpleCPU, "Fetch\n");
586
587 if (!curStaticInst || !curStaticInst->isDelayedCommit())
588 checkForInterrupts();
589
590 checkPcEventQueue();
591
592 // We must have just got suspended by a PC event
593 if (_status == Idle)
594 return;
595
596 TheISA::PCState pcState = thread->pcState();
597 bool needToFetch = !isRomMicroPC(pcState.microPC()) && !curMacroStaticInst;
598
599 if (needToFetch) {
600 _status = Running;
601 Request *ifetch_req = new Request();
602 ifetch_req->setThreadContext(_cpuId, /* thread ID */ 0);
603 setupFetchRequest(ifetch_req);
604 DPRINTF(SimpleCPU, "Translating address %#x\n", ifetch_req->getVaddr());
605 thread->itb->translateTiming(ifetch_req, tc, &fetchTranslation,
606 BaseTLB::Execute);
607 } else {
608 _status = IcacheWaitResponse;
609 completeIfetch(NULL);
610
611 numCycles += tickToCycles(curTick() - previousTick);
612 previousTick = curTick();
613 }
614}
615
616
617void
618TimingSimpleCPU::sendFetch(Fault fault, RequestPtr req, ThreadContext *tc)
619{
620 if (fault == NoFault) {
621 DPRINTF(SimpleCPU, "Sending fetch for addr %#x(pa: %#x)\n",
622 req->getVaddr(), req->getPaddr());
623 ifetch_pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast);
624 ifetch_pkt->dataStatic(&inst);
625 DPRINTF(SimpleCPU, " -- pkt addr: %#x\n", ifetch_pkt->getAddr());
626
627 if (!icachePort.sendTiming(ifetch_pkt)) {
628 // Need to wait for retry
629 _status = IcacheRetry;
630 } else {
631 // Need to wait for cache to respond
632 _status = IcacheWaitResponse;
633 // ownership of packet transferred to memory system
634 ifetch_pkt = NULL;
635 }
636 } else {
637 DPRINTF(SimpleCPU, "Translation of addr %#x faulted\n", req->getVaddr());
638 delete req;
639 // fetch fault: advance directly to next instruction (fault handler)
640 _status = Running;
641 advanceInst(fault);
642 }
643
644 numCycles += tickToCycles(curTick() - previousTick);
645 previousTick = curTick();
646}
647
648
649void
650TimingSimpleCPU::advanceInst(Fault fault)
651{
652
653 if (_status == Faulting)
654 return;
655
656 if (fault != NoFault) {
657 advancePC(fault);
658 DPRINTF(SimpleCPU, "Fault occured, scheduling fetch event\n");
659 reschedule(fetchEvent, nextCycle(), true);
660 _status = Faulting;
661 return;
662 }
663
664
665 if (!stayAtPC)
666 advancePC(fault);
667
668 if (_status == Running) {
669 // kick off fetch of next instruction... callback from icache
670 // response will cause that instruction to be executed,
671 // keeping the CPU running.
672 fetch();
673 }
674}
675
676
677void
678TimingSimpleCPU::completeIfetch(PacketPtr pkt)
679{
680 DPRINTF(SimpleCPU, "Complete ICache Fetch for addr %#x\n", pkt ?
681 pkt->getAddr() : 0);
682
683 // received a response from the icache: execute the received
684 // instruction
685
686 assert(!pkt || !pkt->isError());
687 assert(_status == IcacheWaitResponse);
688
689 _status = Running;
690
691 numCycles += tickToCycles(curTick() - previousTick);
692 previousTick = curTick();
693
694 if (getState() == SimObject::Draining) {
695 if (pkt) {
696 delete pkt->req;
697 delete pkt;
698 }
699
700 completeDrain();
701 return;
702 }
703
704 preExecute();
705 if (curStaticInst && curStaticInst->isMemRef()) {
706 // load or store: just send to dcache
707 Fault fault = curStaticInst->initiateAcc(this, traceData);
708
709 // If we're not running now the instruction will complete in a dcache
710 // response callback or the instruction faulted and has started an
711 // ifetch
712 if (_status == Running) {
713 if (fault != NoFault && traceData) {
714 // If there was a fault, we shouldn't trace this instruction.
715 delete traceData;
716 traceData = NULL;
717 }
718
719 postExecute();
720 // @todo remove me after debugging with legion done
721 if (curStaticInst && (!curStaticInst->isMicroop() ||
722 curStaticInst->isFirstMicroop()))
723 instCnt++;
724 advanceInst(fault);
725 }
726 } else if (curStaticInst) {
727 // non-memory instruction: execute completely now
728 Fault fault = curStaticInst->execute(this, traceData);
729
730 // keep an instruction count
731 if (fault == NoFault)
732 countInst();
733 else if (traceData && !DTRACE(ExecFaulting)) {
734 delete traceData;
735 traceData = NULL;
736 }
737
738 postExecute();
739 // @todo remove me after debugging with legion done
740 if (curStaticInst && (!curStaticInst->isMicroop() ||
741 curStaticInst->isFirstMicroop()))
742 instCnt++;
743 advanceInst(fault);
744 } else {
745 advanceInst(NoFault);
746 }
747
748 if (pkt) {
749 delete pkt->req;
750 delete pkt;
751 }
752}
753
754void
755TimingSimpleCPU::IcachePort::ITickEvent::process()
756{
757 cpu->completeIfetch(pkt);
758}
759
760bool
761TimingSimpleCPU::IcachePort::recvTiming(PacketPtr pkt)
762{
763 if (pkt->isResponse() && !pkt->wasNacked()) {
764 DPRINTF(SimpleCPU, "Received timing response %#x\n", pkt->getAddr());
765 // delay processing of returned data until next CPU clock edge
766 Tick next_tick = cpu->nextCycle(curTick());
767
768 if (next_tick == curTick())
769 cpu->completeIfetch(pkt);
770 else
771 tickEvent.schedule(pkt, next_tick);
772
773 return true;
774 } else if (pkt->wasNacked()) {
775 assert(cpu->_status == IcacheWaitResponse);
776 pkt->reinitNacked();
777 if (!sendTiming(pkt)) {
778 cpu->_status = IcacheRetry;
779 cpu->ifetch_pkt = pkt;
780 }
781 }
782 //Snooping a Coherence Request, do nothing
783 return true;
784}
785
786void
787TimingSimpleCPU::IcachePort::recvRetry()
788{
789 // we shouldn't get a retry unless we have a packet that we're
790 // waiting to transmit
791 assert(cpu->ifetch_pkt != NULL);
792 assert(cpu->_status == IcacheRetry);
793 PacketPtr tmp = cpu->ifetch_pkt;
794 if (sendTiming(tmp)) {
795 cpu->_status = IcacheWaitResponse;
796 cpu->ifetch_pkt = NULL;
797 }
798}
799
800void
801TimingSimpleCPU::completeDataAccess(PacketPtr pkt)
802{
803 // received a response from the dcache: complete the load or store
804 // instruction
805 assert(!pkt->isError());
806 assert(_status == DcacheWaitResponse || _status == DTBWaitResponse ||
807 pkt->req->getFlags().isSet(Request::NO_ACCESS));
808
809 numCycles += tickToCycles(curTick() - previousTick);
810 previousTick = curTick();
811
812 if (pkt->senderState) {
813 SplitFragmentSenderState * send_state =
814 dynamic_cast<SplitFragmentSenderState *>(pkt->senderState);
815 assert(send_state);
816 delete pkt->req;
817 delete pkt;
818 PacketPtr big_pkt = send_state->bigPkt;
819 delete send_state;
820
821 SplitMainSenderState * main_send_state =
822 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
823 assert(main_send_state);
824 // Record the fact that this packet is no longer outstanding.
825 assert(main_send_state->outstanding != 0);
826 main_send_state->outstanding--;
827
828 if (main_send_state->outstanding) {
829 return;
830 } else {
831 delete main_send_state;
832 big_pkt->senderState = NULL;
833 pkt = big_pkt;
834 }
835 }
836
837 _status = Running;
838
839 Fault fault = curStaticInst->completeAcc(pkt, this, traceData);
840
841 // keep an instruction count
842 if (fault == NoFault)
843 countInst();
844 else if (traceData) {
845 // If there was a fault, we shouldn't trace this instruction.
846 delete traceData;
847 traceData = NULL;
848 }
849
850 // the locked flag may be cleared on the response packet, so check
851 // pkt->req and not pkt to see if it was a load-locked
852 if (pkt->isRead() && pkt->req->isLLSC()) {
853 TheISA::handleLockedRead(thread, pkt->req);
854 }
855
856 delete pkt->req;
857 delete pkt;
858
859 postExecute();
860
861 if (getState() == SimObject::Draining) {
862 advancePC(fault);
863 completeDrain();
864
865 return;
866 }
867
868 advanceInst(fault);
869}
870
871
872void
873TimingSimpleCPU::completeDrain()
874{
875 DPRINTF(Config, "Done draining\n");
876 changeState(SimObject::Drained);
877 drainEvent->process();
878}
879
880bool
881TimingSimpleCPU::DcachePort::recvTiming(PacketPtr pkt)
882{
883 if (pkt->isResponse() && !pkt->wasNacked()) {
884 // delay processing of returned data until next CPU clock edge
885 Tick next_tick = cpu->nextCycle(curTick());
886
887 if (next_tick == curTick()) {
888 cpu->completeDataAccess(pkt);
889 } else {
890 if (!tickEvent.scheduled()) {
891 tickEvent.schedule(pkt, next_tick);
892 } else {
893 // In the case of a split transaction and a cache that is
894 // faster than a CPU we could get two responses before
895 // next_tick expires
896 if (!retryEvent.scheduled())
897 schedule(retryEvent, next_tick);
898 return false;
899 }
900 }
901
902 return true;
903 }
904 else if (pkt->wasNacked()) {
905 assert(cpu->_status == DcacheWaitResponse);
906 pkt->reinitNacked();
907 if (!sendTiming(pkt)) {
908 cpu->_status = DcacheRetry;
909 cpu->dcache_pkt = pkt;
910 }
911 }
912 //Snooping a Coherence Request, do nothing
913 return true;
914}
915
916void
917TimingSimpleCPU::DcachePort::DTickEvent::process()
918{
919 cpu->completeDataAccess(pkt);
920}
921
922void
923TimingSimpleCPU::DcachePort::recvRetry()
924{
925 // we shouldn't get a retry unless we have a packet that we're
926 // waiting to transmit
927 assert(cpu->dcache_pkt != NULL);
928 assert(cpu->_status == DcacheRetry);
929 PacketPtr tmp = cpu->dcache_pkt;
930 if (tmp->senderState) {
931 // This is a packet from a split access.
932 SplitFragmentSenderState * send_state =
933 dynamic_cast<SplitFragmentSenderState *>(tmp->senderState);
934 assert(send_state);
935 PacketPtr big_pkt = send_state->bigPkt;
936
937 SplitMainSenderState * main_send_state =
938 dynamic_cast<SplitMainSenderState *>(big_pkt->senderState);
939 assert(main_send_state);
940
941 if (sendTiming(tmp)) {
942 // If we were able to send without retrying, record that fact
943 // and try sending the other fragment.
944 send_state->clearFromParent();
945 int other_index = main_send_state->getPendingFragment();
946 if (other_index > 0) {
947 tmp = main_send_state->fragments[other_index];
948 cpu->dcache_pkt = tmp;
949 if ((big_pkt->isRead() && cpu->handleReadPacket(tmp)) ||
950 (big_pkt->isWrite() && cpu->handleWritePacket())) {
951 main_send_state->fragments[other_index] = NULL;
952 }
953 } else {
954 cpu->_status = DcacheWaitResponse;
955 // memory system takes ownership of packet
956 cpu->dcache_pkt = NULL;
957 }
958 }
959 } else if (sendTiming(tmp)) {
960 cpu->_status = DcacheWaitResponse;
961 // memory system takes ownership of packet
962 cpu->dcache_pkt = NULL;
963 }
964}
965
966TimingSimpleCPU::IprEvent::IprEvent(Packet *_pkt, TimingSimpleCPU *_cpu,
967 Tick t)
968 : pkt(_pkt), cpu(_cpu)
969{
970 cpu->schedule(this, t);
971}
972
973void
974TimingSimpleCPU::IprEvent::process()
975{
976 cpu->completeDataAccess(pkt);
977}
978
979const char *
980TimingSimpleCPU::IprEvent::description() const
981{
982 return "Timing Simple CPU Delay IPR event";
983}
984
985
986void
987TimingSimpleCPU::printAddr(Addr a)
988{
989 dcachePort.printAddr(a);
990}
991
992
993////////////////////////////////////////////////////////////////////////
994//
995// TimingSimpleCPU Simulation Object
996//
997TimingSimpleCPU *
998TimingSimpleCPUParams::create()
999{
1000 numThreads = 1;
1001#if !FULL_SYSTEM
1002 if (workload.size() != 1)
1003 panic("only one workload allowed");
1004#endif
1005 return new TimingSimpleCPU(this);
1006}
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, &icachePort, &dcachePort);
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(int 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(int 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());
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, 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, 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 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 !FULL_SYSTEM
970 if (workload.size() != 1)
971 panic("only one workload allowed");
972#endif
973 return new TimingSimpleCPU(this);
974}