1/*
2 * Copyright (c) 2002-2005 The Regents of The University of Michigan
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Authors: Steve Reinhardt
29 */
30
31#include "arch/locked_mem.hh"
32#include "arch/mmaped_ipr.hh"
33#include "arch/utility.hh"
34#include "base/bigint.hh"
35#include "cpu/exetrace.hh"
36#include "cpu/simple/atomic.hh"
37#include "mem/packet.hh"
38#include "mem/packet_access.hh"
39#include "params/AtomicSimpleCPU.hh"
40#include "sim/system.hh"
41
42using namespace std;
43using namespace TheISA;
44
45AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c)
46 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c)
47{
48}
49
50
51void
52AtomicSimpleCPU::TickEvent::process()
53{
54 cpu->tick();
55}
56
57const char *
58AtomicSimpleCPU::TickEvent::description() const
59{
60 return "AtomicSimpleCPU tick";
61}
62
63Port *
64AtomicSimpleCPU::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 if (if_name == "physmem_port") {
71 hasPhysMemPort = true;
72 return &physmemPort;
73 }
74 else
75 panic("No Such Port\n");
76}
77
78void
79AtomicSimpleCPU::init()
80{
81 BaseCPU::init();
82 cpuId = tc->readCpuId();
83#if FULL_SYSTEM
84 for (int i = 0; i < threadContexts.size(); ++i) {
85 ThreadContext *tc = threadContexts[i];
86
87 // initialize CPU, including PC
88 TheISA::initCPU(tc, cpuId);
89 }
90#endif
91 if (hasPhysMemPort) {
92 bool snoop = false;
93 AddrRangeList pmAddrList;
94 physmemPort.getPeerAddressRanges(pmAddrList, snoop);
95 physMemAddr = *pmAddrList.begin();
96 }
97 ifetch_req.setThreadContext(cpuId, 0); // Add thread ID if we add MT
98 data_read_req.setThreadContext(cpuId, 0); // Add thread ID here too
99 data_write_req.setThreadContext(cpuId, 0); // Add thread ID here too
100}
101
102bool
103AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt)
104{
105 panic("AtomicSimpleCPU doesn't expect recvTiming callback!");
106 return true;
107}
108
109Tick
110AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt)
111{
112 //Snooping a coherence request, just return
113 return 0;
114}
115
116void
117AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt)
118{
119 //No internal storage to update, just return
120 return;
121}
122
123void
124AtomicSimpleCPU::CpuPort::recvStatusChange(Status status)
125{
126 if (status == RangeChange) {
127 if (!snoopRangeSent) {
128 snoopRangeSent = true;
129 sendStatusChange(Port::RangeChange);
130 }
131 return;
132 }
133
134 panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!");
135}
136
137void
138AtomicSimpleCPU::CpuPort::recvRetry()
139{
140 panic("AtomicSimpleCPU doesn't expect recvRetry callback!");
141}
142
143void
144AtomicSimpleCPU::DcachePort::setPeer(Port *port)
145{
146 Port::setPeer(port);
147
148#if FULL_SYSTEM
149 // Update the ThreadContext's memory ports (Functional/Virtual
150 // Ports)
151 cpu->tcBase()->connectMemPorts();
152#endif
153}
154
155AtomicSimpleCPU::AtomicSimpleCPU(Params *p)
156 : BaseSimpleCPU(p), tickEvent(this),
157 width(p->width), simulate_stalls(p->simulate_stalls),
158 icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this),
159 physmemPort(name() + "-iport", this), hasPhysMemPort(false)
160{
161 _status = Idle;
162
163 icachePort.snoopRangeSent = false;
164 dcachePort.snoopRangeSent = false;
165
166}
167
168
169AtomicSimpleCPU::~AtomicSimpleCPU()
170{
171}
172
173void
174AtomicSimpleCPU::serialize(ostream &os)
175{
176 SimObject::State so_state = SimObject::getState();
177 SERIALIZE_ENUM(so_state);
178 Status _status = status();
179 SERIALIZE_ENUM(_status);
180 BaseSimpleCPU::serialize(os);
181 nameOut(os, csprintf("%s.tickEvent", name()));
182 tickEvent.serialize(os);
183}
184
185void
186AtomicSimpleCPU::unserialize(Checkpoint *cp, const string §ion)
187{
188 SimObject::State so_state;
189 UNSERIALIZE_ENUM(so_state);
190 UNSERIALIZE_ENUM(_status);
191 BaseSimpleCPU::unserialize(cp, section);
192 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
193}
194
195void
196AtomicSimpleCPU::resume()
197{
198 if (_status == Idle || _status == SwitchedOut)
199 return;
200
201 DPRINTF(SimpleCPU, "Resume\n");
202 assert(system->getMemoryMode() == Enums::atomic);
203
204 changeState(SimObject::Running);
205 if (thread->status() == ThreadContext::Active) {
206 if (!tickEvent.scheduled()) {
207 tickEvent.schedule(nextCycle());
208 }
209 }
210}
211
212void
213AtomicSimpleCPU::switchOut()
214{
215 assert(status() == Running || status() == Idle);
216 _status = SwitchedOut;
217
218 tickEvent.squash();
219}
220
221
222void
223AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
224{
225 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort);
226
227 assert(!tickEvent.scheduled());
228
229 // if any of this CPU's ThreadContexts are active, mark the CPU as
230 // running and schedule its tick event.
231 for (int i = 0; i < threadContexts.size(); ++i) {
232 ThreadContext *tc = threadContexts[i];
233 if (tc->status() == ThreadContext::Active && _status != Running) {
234 _status = Running;
235 tickEvent.schedule(nextCycle());
236 break;
237 }
238 }
239 if (_status != Running) {
240 _status = Idle;
241 }
242 assert(threadContexts.size() == 1);
243 cpuId = tc->readCpuId();
244 ifetch_req.setThreadContext(cpuId, 0); // Add thread ID if we add MT
245 data_read_req.setThreadContext(cpuId, 0); // Add thread ID here too
246 data_write_req.setThreadContext(cpuId, 0); // Add thread ID here too
247}
248
249
250void
251AtomicSimpleCPU::activateContext(int thread_num, int delay)
252{
253 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);
254
255 assert(thread_num == 0);
256 assert(thread);
257
258 assert(_status == Idle);
259 assert(!tickEvent.scheduled());
260
261 notIdleFraction++;
262 numCycles += tickToCycles(thread->lastActivate - thread->lastSuspend);
263
264 //Make sure ticks are still on multiples of cycles
265 tickEvent.schedule(nextCycle(curTick + ticks(delay)));
266 _status = Running;
267}
268
269
270void
271AtomicSimpleCPU::suspendContext(int thread_num)
272{
273 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);
274
275 assert(thread_num == 0);
276 assert(thread);
277
278 assert(_status == Running);
279
280 // tick event may not be scheduled if this gets called from inside
281 // an instruction's execution, e.g. "quiesce"
282 if (tickEvent.scheduled())
283 tickEvent.deschedule();
284
285 notIdleFraction--;
286 _status = Idle;
287}
288
289
290template <class T>
291Fault
292AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags)
293{
294 // use the CPU's statically allocated read request and packet objects
295 Request *req = &data_read_req;
296
297 if (traceData) {
298 traceData->setAddr(addr);
299 }
300
301 //The block size of our peer.
302 int blockSize = dcachePort.peerBlockSize();
303 //The size of the data we're trying to read.
304 int dataSize = sizeof(T);
305
306 uint8_t * dataPtr = (uint8_t *)&data;
307
308 //The address of the second part of this access if it needs to be split
309 //across a cache line boundary.
310 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);
311
312 if(secondAddr > addr)
313 dataSize = secondAddr - addr;
314
315 dcache_latency = 0;
316
317 while(1) {
318 req->setVirt(0, addr, dataSize, flags, thread->readPC());
319
320 // translate to physical address
321 Fault fault = thread->translateDataReadReq(req);
322
323 // Now do the access.
324 if (fault == NoFault) {
325 Packet pkt = Packet(req,
326 req->isLocked() ? MemCmd::LoadLockedReq : MemCmd::ReadReq,
327 Packet::Broadcast);
328 pkt.dataStatic(dataPtr);
329
330 if (req->isMmapedIpr())
331 dcache_latency += TheISA::handleIprRead(thread->getTC(), &pkt);
332 else {
333 if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
334 dcache_latency += physmemPort.sendAtomic(&pkt);
335 else
336 dcache_latency += dcachePort.sendAtomic(&pkt);
337 }
338 dcache_access = true;
339
340 assert(!pkt.isError());
341
342 if (req->isLocked()) {
343 TheISA::handleLockedRead(thread, req);
344 }
345 }
346
347 // This will need a new way to tell if it has a dcache attached.
348 if (req->isUncacheable())
349 recordEvent("Uncached Read");
350
351 //If there's a fault, return it
352 if (fault != NoFault)
353 return fault;
354 //If we don't need to access a second cache line, stop now.
355 if (secondAddr <= addr)
356 {
357 data = gtoh(data);
358 return fault;
359 }
360
361 /*
362 * Set up for accessing the second cache line.
363 */
364
365 //Move the pointer we're reading into to the correct location.
366 dataPtr += dataSize;
367 //Adjust the size to get the remaining bytes.
368 dataSize = addr + sizeof(T) - secondAddr;
369 //And access the right address.
370 addr = secondAddr;
371 }
372}
373
374Fault
375AtomicSimpleCPU::translateDataReadAddr(Addr vaddr, Addr & paddr,
376 int size, unsigned flags)
377{
378 // use the CPU's statically allocated read request and packet objects
379 Request *req = &data_read_req;
380
381 if (traceData) {
382 traceData->setAddr(vaddr);
383 }
384
385 //The block size of our peer.
386 int blockSize = dcachePort.peerBlockSize();
387 //The size of the data we're trying to read.
388 int dataSize = size;
389
390 bool firstTimeThrough = true;
391
392 //The address of the second part of this access if it needs to be split
393 //across a cache line boundary.
394 Addr secondAddr = roundDown(vaddr + dataSize - 1, blockSize);
395
396 if(secondAddr > vaddr)
397 dataSize = secondAddr - vaddr;
398
399 while(1) {
400 req->setVirt(0, vaddr, dataSize, flags, thread->readPC());
401
402 // translate to physical address
403 Fault fault = thread->translateDataReadReq(req);
404
405 //If there's a fault, return it
406 if (fault != NoFault)
407 return fault;
408
409 if (firstTimeThrough) {
410 paddr = req->getPaddr();
411 firstTimeThrough = false;
412 }
413
414 //If we don't need to access a second cache line, stop now.
415 if (secondAddr <= vaddr)
416 return fault;
417
418 /*
419 * Set up for accessing the second cache line.
420 */
421
422 //Adjust the size to get the remaining bytes.
423 dataSize = vaddr + size - secondAddr;
424 //And access the right address.
425 vaddr = secondAddr;
426 }
427}
428
429#ifndef DOXYGEN_SHOULD_SKIP_THIS
430
431template
432Fault
433AtomicSimpleCPU::read(Addr addr, Twin32_t &data, unsigned flags);
434
435template
436Fault
437AtomicSimpleCPU::read(Addr addr, Twin64_t &data, unsigned flags);
438
439template
440Fault
441AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);
442
443template
444Fault
445AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);
446
447template
448Fault
449AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);
450
451template
452Fault
453AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);
454
455#endif //DOXYGEN_SHOULD_SKIP_THIS
456
457template<>
458Fault
459AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags)
460{
461 return read(addr, *(uint64_t*)&data, flags);
462}
463
464template<>
465Fault
466AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags)
467{
468 return read(addr, *(uint32_t*)&data, flags);
469}
470
471
472template<>
473Fault
474AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
475{
476 return read(addr, (uint32_t&)data, flags);
477}
478
479
480template <class T>
481Fault
482AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
483{
484 // use the CPU's statically allocated write request and packet objects
485 Request *req = &data_write_req;
486
487 if (traceData) {
488 traceData->setAddr(addr);
489 }
490
491 //The block size of our peer.
492 int blockSize = dcachePort.peerBlockSize();
493 //The size of the data we're trying to read.
494 int dataSize = sizeof(T);
495
496 uint8_t * dataPtr = (uint8_t *)&data;
497
498 //The address of the second part of this access if it needs to be split
499 //across a cache line boundary.
500 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);
501
502 if(secondAddr > addr)
503 dataSize = secondAddr - addr;
504
505 dcache_latency = 0;
506
507 while(1) {
508 req->setVirt(0, addr, dataSize, flags, thread->readPC());
509
510 // translate to physical address
511 Fault fault = thread->translateDataWriteReq(req);
512
513 // Now do the access.
514 if (fault == NoFault) {
515 MemCmd cmd = MemCmd::WriteReq; // default
516 bool do_access = true; // flag to suppress cache access
517
518 if (req->isLocked()) {
519 cmd = MemCmd::StoreCondReq;
520 do_access = TheISA::handleLockedWrite(thread, req);
521 } else if (req->isSwap()) {
522 cmd = MemCmd::SwapReq;
523 if (req->isCondSwap()) {
524 assert(res);
525 req->setExtraData(*res);
526 }
527 }
528
529 if (do_access) {
530 Packet pkt = Packet(req, cmd, Packet::Broadcast);
531 pkt.dataStatic(dataPtr);
532
533 if (req->isMmapedIpr()) {
534 dcache_latency +=
535 TheISA::handleIprWrite(thread->getTC(), &pkt);
536 } else {
537 //XXX This needs to be outside of the loop in order to
538 //work properly for cache line boundary crossing
539 //accesses in transendian simulations.
540 data = htog(data);
541 if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
542 dcache_latency += physmemPort.sendAtomic(&pkt);
543 else
544 dcache_latency += dcachePort.sendAtomic(&pkt);
545 }
546 dcache_access = true;
547 assert(!pkt.isError());
548
549 if (req->isSwap()) {
550 assert(res);
551 *res = pkt.get<T>();
552 }
553 }
554
555 if (res && !req->isSwap()) {
556 *res = req->getExtraData();
557 }
558 }
559
560 // This will need a new way to tell if it's hooked up to a cache or not.
561 if (req->isUncacheable())
562 recordEvent("Uncached Write");
563
564 //If there's a fault or we don't need to access a second cache line,
565 //stop now.
566 if (fault != NoFault || secondAddr <= addr)
567 {
568 // If the write needs to have a fault on the access, consider
569 // calling changeStatus() and changing it to "bad addr write"
570 // or something.
571 return fault;
572 }
573
574 /*
575 * Set up for accessing the second cache line.
576 */
577
578 //Move the pointer we're reading into to the correct location.
579 dataPtr += dataSize;
580 //Adjust the size to get the remaining bytes.
581 dataSize = addr + sizeof(T) - secondAddr;
582 //And access the right address.
583 addr = secondAddr;
584 }
585}
586
587Fault
588AtomicSimpleCPU::translateDataWriteAddr(Addr vaddr, Addr &paddr,
589 int size, unsigned flags)
590{
591 // use the CPU's statically allocated write request and packet objects
592 Request *req = &data_write_req;
593
594 if (traceData) {
595 traceData->setAddr(vaddr);
596 }
597
598 //The block size of our peer.
599 int blockSize = dcachePort.peerBlockSize();
600
601 //The address of the second part of this access if it needs to be split
602 //across a cache line boundary.
603 Addr secondAddr = roundDown(vaddr + size - 1, blockSize);
604
605 //The size of the data we're trying to read.
606 int dataSize = size;
607
608 bool firstTimeThrough = true;
609
610 if(secondAddr > vaddr)
611 dataSize = secondAddr - vaddr;
612
613 dcache_latency = 0;
614
615 while(1) {
616 req->setVirt(0, vaddr, dataSize, flags, thread->readPC());
617
618 // translate to physical address
619 Fault fault = thread->translateDataWriteReq(req);
620
621 //If there's a fault or we don't need to access a second cache line,
622 //stop now.
623 if (fault != NoFault)
624 return fault;
625
626 if (firstTimeThrough) {
627 paddr = req->getPaddr();
628 firstTimeThrough = false;
629 }
630
631 if (secondAddr <= vaddr)
632 return fault;
633
634 /*
635 * Set up for accessing the second cache line.
636 */
637
638 //Adjust the size to get the remaining bytes.
639 dataSize = vaddr + size - secondAddr;
640 //And access the right address.
641 vaddr = secondAddr;
642 }
643}
644
645
646#ifndef DOXYGEN_SHOULD_SKIP_THIS
647
648template
649Fault
650AtomicSimpleCPU::write(Twin32_t data, Addr addr,
651 unsigned flags, uint64_t *res);
652
653template
654Fault
655AtomicSimpleCPU::write(Twin64_t data, Addr addr,
656 unsigned flags, uint64_t *res);
657
658template
659Fault
660AtomicSimpleCPU::write(uint64_t data, Addr addr,
661 unsigned flags, uint64_t *res);
662
663template
664Fault
665AtomicSimpleCPU::write(uint32_t data, Addr addr,
666 unsigned flags, uint64_t *res);
667
668template
669Fault
670AtomicSimpleCPU::write(uint16_t data, Addr addr,
671 unsigned flags, uint64_t *res);
672
673template
674Fault
675AtomicSimpleCPU::write(uint8_t data, Addr addr,
676 unsigned flags, uint64_t *res);
677
678#endif //DOXYGEN_SHOULD_SKIP_THIS
679
680template<>
681Fault
682AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
683{
684 return write(*(uint64_t*)&data, addr, flags, res);
685}
686
687template<>
688Fault
689AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
690{
691 return write(*(uint32_t*)&data, addr, flags, res);
692}
693
694
695template<>
696Fault
697AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
698{
699 return write((uint32_t)data, addr, flags, res);
700}
701
702
703void
704AtomicSimpleCPU::tick()
705{
706 DPRINTF(SimpleCPU, "Tick\n");
707
708 Tick latency = ticks(1); // instruction takes one cycle by default
709
710 for (int i = 0; i < width; ++i) {
711 numCycles++;
712
713 if (!curStaticInst || !curStaticInst->isDelayedCommit())
714 checkForInterrupts();
715
| 1/*
2 * Copyright (c) 2002-2005 The Regents of The University of Michigan
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Authors: Steve Reinhardt
29 */
30
31#include "arch/locked_mem.hh"
32#include "arch/mmaped_ipr.hh"
33#include "arch/utility.hh"
34#include "base/bigint.hh"
35#include "cpu/exetrace.hh"
36#include "cpu/simple/atomic.hh"
37#include "mem/packet.hh"
38#include "mem/packet_access.hh"
39#include "params/AtomicSimpleCPU.hh"
40#include "sim/system.hh"
41
42using namespace std;
43using namespace TheISA;
44
45AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c)
46 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c)
47{
48}
49
50
51void
52AtomicSimpleCPU::TickEvent::process()
53{
54 cpu->tick();
55}
56
57const char *
58AtomicSimpleCPU::TickEvent::description() const
59{
60 return "AtomicSimpleCPU tick";
61}
62
63Port *
64AtomicSimpleCPU::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 if (if_name == "physmem_port") {
71 hasPhysMemPort = true;
72 return &physmemPort;
73 }
74 else
75 panic("No Such Port\n");
76}
77
78void
79AtomicSimpleCPU::init()
80{
81 BaseCPU::init();
82 cpuId = tc->readCpuId();
83#if FULL_SYSTEM
84 for (int i = 0; i < threadContexts.size(); ++i) {
85 ThreadContext *tc = threadContexts[i];
86
87 // initialize CPU, including PC
88 TheISA::initCPU(tc, cpuId);
89 }
90#endif
91 if (hasPhysMemPort) {
92 bool snoop = false;
93 AddrRangeList pmAddrList;
94 physmemPort.getPeerAddressRanges(pmAddrList, snoop);
95 physMemAddr = *pmAddrList.begin();
96 }
97 ifetch_req.setThreadContext(cpuId, 0); // Add thread ID if we add MT
98 data_read_req.setThreadContext(cpuId, 0); // Add thread ID here too
99 data_write_req.setThreadContext(cpuId, 0); // Add thread ID here too
100}
101
102bool
103AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt)
104{
105 panic("AtomicSimpleCPU doesn't expect recvTiming callback!");
106 return true;
107}
108
109Tick
110AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt)
111{
112 //Snooping a coherence request, just return
113 return 0;
114}
115
116void
117AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt)
118{
119 //No internal storage to update, just return
120 return;
121}
122
123void
124AtomicSimpleCPU::CpuPort::recvStatusChange(Status status)
125{
126 if (status == RangeChange) {
127 if (!snoopRangeSent) {
128 snoopRangeSent = true;
129 sendStatusChange(Port::RangeChange);
130 }
131 return;
132 }
133
134 panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!");
135}
136
137void
138AtomicSimpleCPU::CpuPort::recvRetry()
139{
140 panic("AtomicSimpleCPU doesn't expect recvRetry callback!");
141}
142
143void
144AtomicSimpleCPU::DcachePort::setPeer(Port *port)
145{
146 Port::setPeer(port);
147
148#if FULL_SYSTEM
149 // Update the ThreadContext's memory ports (Functional/Virtual
150 // Ports)
151 cpu->tcBase()->connectMemPorts();
152#endif
153}
154
155AtomicSimpleCPU::AtomicSimpleCPU(Params *p)
156 : BaseSimpleCPU(p), tickEvent(this),
157 width(p->width), simulate_stalls(p->simulate_stalls),
158 icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this),
159 physmemPort(name() + "-iport", this), hasPhysMemPort(false)
160{
161 _status = Idle;
162
163 icachePort.snoopRangeSent = false;
164 dcachePort.snoopRangeSent = false;
165
166}
167
168
169AtomicSimpleCPU::~AtomicSimpleCPU()
170{
171}
172
173void
174AtomicSimpleCPU::serialize(ostream &os)
175{
176 SimObject::State so_state = SimObject::getState();
177 SERIALIZE_ENUM(so_state);
178 Status _status = status();
179 SERIALIZE_ENUM(_status);
180 BaseSimpleCPU::serialize(os);
181 nameOut(os, csprintf("%s.tickEvent", name()));
182 tickEvent.serialize(os);
183}
184
185void
186AtomicSimpleCPU::unserialize(Checkpoint *cp, const string §ion)
187{
188 SimObject::State so_state;
189 UNSERIALIZE_ENUM(so_state);
190 UNSERIALIZE_ENUM(_status);
191 BaseSimpleCPU::unserialize(cp, section);
192 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
193}
194
195void
196AtomicSimpleCPU::resume()
197{
198 if (_status == Idle || _status == SwitchedOut)
199 return;
200
201 DPRINTF(SimpleCPU, "Resume\n");
202 assert(system->getMemoryMode() == Enums::atomic);
203
204 changeState(SimObject::Running);
205 if (thread->status() == ThreadContext::Active) {
206 if (!tickEvent.scheduled()) {
207 tickEvent.schedule(nextCycle());
208 }
209 }
210}
211
212void
213AtomicSimpleCPU::switchOut()
214{
215 assert(status() == Running || status() == Idle);
216 _status = SwitchedOut;
217
218 tickEvent.squash();
219}
220
221
222void
223AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
224{
225 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort);
226
227 assert(!tickEvent.scheduled());
228
229 // if any of this CPU's ThreadContexts are active, mark the CPU as
230 // running and schedule its tick event.
231 for (int i = 0; i < threadContexts.size(); ++i) {
232 ThreadContext *tc = threadContexts[i];
233 if (tc->status() == ThreadContext::Active && _status != Running) {
234 _status = Running;
235 tickEvent.schedule(nextCycle());
236 break;
237 }
238 }
239 if (_status != Running) {
240 _status = Idle;
241 }
242 assert(threadContexts.size() == 1);
243 cpuId = tc->readCpuId();
244 ifetch_req.setThreadContext(cpuId, 0); // Add thread ID if we add MT
245 data_read_req.setThreadContext(cpuId, 0); // Add thread ID here too
246 data_write_req.setThreadContext(cpuId, 0); // Add thread ID here too
247}
248
249
250void
251AtomicSimpleCPU::activateContext(int thread_num, int delay)
252{
253 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);
254
255 assert(thread_num == 0);
256 assert(thread);
257
258 assert(_status == Idle);
259 assert(!tickEvent.scheduled());
260
261 notIdleFraction++;
262 numCycles += tickToCycles(thread->lastActivate - thread->lastSuspend);
263
264 //Make sure ticks are still on multiples of cycles
265 tickEvent.schedule(nextCycle(curTick + ticks(delay)));
266 _status = Running;
267}
268
269
270void
271AtomicSimpleCPU::suspendContext(int thread_num)
272{
273 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);
274
275 assert(thread_num == 0);
276 assert(thread);
277
278 assert(_status == Running);
279
280 // tick event may not be scheduled if this gets called from inside
281 // an instruction's execution, e.g. "quiesce"
282 if (tickEvent.scheduled())
283 tickEvent.deschedule();
284
285 notIdleFraction--;
286 _status = Idle;
287}
288
289
290template <class T>
291Fault
292AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags)
293{
294 // use the CPU's statically allocated read request and packet objects
295 Request *req = &data_read_req;
296
297 if (traceData) {
298 traceData->setAddr(addr);
299 }
300
301 //The block size of our peer.
302 int blockSize = dcachePort.peerBlockSize();
303 //The size of the data we're trying to read.
304 int dataSize = sizeof(T);
305
306 uint8_t * dataPtr = (uint8_t *)&data;
307
308 //The address of the second part of this access if it needs to be split
309 //across a cache line boundary.
310 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);
311
312 if(secondAddr > addr)
313 dataSize = secondAddr - addr;
314
315 dcache_latency = 0;
316
317 while(1) {
318 req->setVirt(0, addr, dataSize, flags, thread->readPC());
319
320 // translate to physical address
321 Fault fault = thread->translateDataReadReq(req);
322
323 // Now do the access.
324 if (fault == NoFault) {
325 Packet pkt = Packet(req,
326 req->isLocked() ? MemCmd::LoadLockedReq : MemCmd::ReadReq,
327 Packet::Broadcast);
328 pkt.dataStatic(dataPtr);
329
330 if (req->isMmapedIpr())
331 dcache_latency += TheISA::handleIprRead(thread->getTC(), &pkt);
332 else {
333 if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
334 dcache_latency += physmemPort.sendAtomic(&pkt);
335 else
336 dcache_latency += dcachePort.sendAtomic(&pkt);
337 }
338 dcache_access = true;
339
340 assert(!pkt.isError());
341
342 if (req->isLocked()) {
343 TheISA::handleLockedRead(thread, req);
344 }
345 }
346
347 // This will need a new way to tell if it has a dcache attached.
348 if (req->isUncacheable())
349 recordEvent("Uncached Read");
350
351 //If there's a fault, return it
352 if (fault != NoFault)
353 return fault;
354 //If we don't need to access a second cache line, stop now.
355 if (secondAddr <= addr)
356 {
357 data = gtoh(data);
358 return fault;
359 }
360
361 /*
362 * Set up for accessing the second cache line.
363 */
364
365 //Move the pointer we're reading into to the correct location.
366 dataPtr += dataSize;
367 //Adjust the size to get the remaining bytes.
368 dataSize = addr + sizeof(T) - secondAddr;
369 //And access the right address.
370 addr = secondAddr;
371 }
372}
373
374Fault
375AtomicSimpleCPU::translateDataReadAddr(Addr vaddr, Addr & paddr,
376 int size, unsigned flags)
377{
378 // use the CPU's statically allocated read request and packet objects
379 Request *req = &data_read_req;
380
381 if (traceData) {
382 traceData->setAddr(vaddr);
383 }
384
385 //The block size of our peer.
386 int blockSize = dcachePort.peerBlockSize();
387 //The size of the data we're trying to read.
388 int dataSize = size;
389
390 bool firstTimeThrough = true;
391
392 //The address of the second part of this access if it needs to be split
393 //across a cache line boundary.
394 Addr secondAddr = roundDown(vaddr + dataSize - 1, blockSize);
395
396 if(secondAddr > vaddr)
397 dataSize = secondAddr - vaddr;
398
399 while(1) {
400 req->setVirt(0, vaddr, dataSize, flags, thread->readPC());
401
402 // translate to physical address
403 Fault fault = thread->translateDataReadReq(req);
404
405 //If there's a fault, return it
406 if (fault != NoFault)
407 return fault;
408
409 if (firstTimeThrough) {
410 paddr = req->getPaddr();
411 firstTimeThrough = false;
412 }
413
414 //If we don't need to access a second cache line, stop now.
415 if (secondAddr <= vaddr)
416 return fault;
417
418 /*
419 * Set up for accessing the second cache line.
420 */
421
422 //Adjust the size to get the remaining bytes.
423 dataSize = vaddr + size - secondAddr;
424 //And access the right address.
425 vaddr = secondAddr;
426 }
427}
428
429#ifndef DOXYGEN_SHOULD_SKIP_THIS
430
431template
432Fault
433AtomicSimpleCPU::read(Addr addr, Twin32_t &data, unsigned flags);
434
435template
436Fault
437AtomicSimpleCPU::read(Addr addr, Twin64_t &data, unsigned flags);
438
439template
440Fault
441AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);
442
443template
444Fault
445AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);
446
447template
448Fault
449AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);
450
451template
452Fault
453AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);
454
455#endif //DOXYGEN_SHOULD_SKIP_THIS
456
457template<>
458Fault
459AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags)
460{
461 return read(addr, *(uint64_t*)&data, flags);
462}
463
464template<>
465Fault
466AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags)
467{
468 return read(addr, *(uint32_t*)&data, flags);
469}
470
471
472template<>
473Fault
474AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
475{
476 return read(addr, (uint32_t&)data, flags);
477}
478
479
480template <class T>
481Fault
482AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
483{
484 // use the CPU's statically allocated write request and packet objects
485 Request *req = &data_write_req;
486
487 if (traceData) {
488 traceData->setAddr(addr);
489 }
490
491 //The block size of our peer.
492 int blockSize = dcachePort.peerBlockSize();
493 //The size of the data we're trying to read.
494 int dataSize = sizeof(T);
495
496 uint8_t * dataPtr = (uint8_t *)&data;
497
498 //The address of the second part of this access if it needs to be split
499 //across a cache line boundary.
500 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);
501
502 if(secondAddr > addr)
503 dataSize = secondAddr - addr;
504
505 dcache_latency = 0;
506
507 while(1) {
508 req->setVirt(0, addr, dataSize, flags, thread->readPC());
509
510 // translate to physical address
511 Fault fault = thread->translateDataWriteReq(req);
512
513 // Now do the access.
514 if (fault == NoFault) {
515 MemCmd cmd = MemCmd::WriteReq; // default
516 bool do_access = true; // flag to suppress cache access
517
518 if (req->isLocked()) {
519 cmd = MemCmd::StoreCondReq;
520 do_access = TheISA::handleLockedWrite(thread, req);
521 } else if (req->isSwap()) {
522 cmd = MemCmd::SwapReq;
523 if (req->isCondSwap()) {
524 assert(res);
525 req->setExtraData(*res);
526 }
527 }
528
529 if (do_access) {
530 Packet pkt = Packet(req, cmd, Packet::Broadcast);
531 pkt.dataStatic(dataPtr);
532
533 if (req->isMmapedIpr()) {
534 dcache_latency +=
535 TheISA::handleIprWrite(thread->getTC(), &pkt);
536 } else {
537 //XXX This needs to be outside of the loop in order to
538 //work properly for cache line boundary crossing
539 //accesses in transendian simulations.
540 data = htog(data);
541 if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
542 dcache_latency += physmemPort.sendAtomic(&pkt);
543 else
544 dcache_latency += dcachePort.sendAtomic(&pkt);
545 }
546 dcache_access = true;
547 assert(!pkt.isError());
548
549 if (req->isSwap()) {
550 assert(res);
551 *res = pkt.get<T>();
552 }
553 }
554
555 if (res && !req->isSwap()) {
556 *res = req->getExtraData();
557 }
558 }
559
560 // This will need a new way to tell if it's hooked up to a cache or not.
561 if (req->isUncacheable())
562 recordEvent("Uncached Write");
563
564 //If there's a fault or we don't need to access a second cache line,
565 //stop now.
566 if (fault != NoFault || secondAddr <= addr)
567 {
568 // If the write needs to have a fault on the access, consider
569 // calling changeStatus() and changing it to "bad addr write"
570 // or something.
571 return fault;
572 }
573
574 /*
575 * Set up for accessing the second cache line.
576 */
577
578 //Move the pointer we're reading into to the correct location.
579 dataPtr += dataSize;
580 //Adjust the size to get the remaining bytes.
581 dataSize = addr + sizeof(T) - secondAddr;
582 //And access the right address.
583 addr = secondAddr;
584 }
585}
586
587Fault
588AtomicSimpleCPU::translateDataWriteAddr(Addr vaddr, Addr &paddr,
589 int size, unsigned flags)
590{
591 // use the CPU's statically allocated write request and packet objects
592 Request *req = &data_write_req;
593
594 if (traceData) {
595 traceData->setAddr(vaddr);
596 }
597
598 //The block size of our peer.
599 int blockSize = dcachePort.peerBlockSize();
600
601 //The address of the second part of this access if it needs to be split
602 //across a cache line boundary.
603 Addr secondAddr = roundDown(vaddr + size - 1, blockSize);
604
605 //The size of the data we're trying to read.
606 int dataSize = size;
607
608 bool firstTimeThrough = true;
609
610 if(secondAddr > vaddr)
611 dataSize = secondAddr - vaddr;
612
613 dcache_latency = 0;
614
615 while(1) {
616 req->setVirt(0, vaddr, dataSize, flags, thread->readPC());
617
618 // translate to physical address
619 Fault fault = thread->translateDataWriteReq(req);
620
621 //If there's a fault or we don't need to access a second cache line,
622 //stop now.
623 if (fault != NoFault)
624 return fault;
625
626 if (firstTimeThrough) {
627 paddr = req->getPaddr();
628 firstTimeThrough = false;
629 }
630
631 if (secondAddr <= vaddr)
632 return fault;
633
634 /*
635 * Set up for accessing the second cache line.
636 */
637
638 //Adjust the size to get the remaining bytes.
639 dataSize = vaddr + size - secondAddr;
640 //And access the right address.
641 vaddr = secondAddr;
642 }
643}
644
645
646#ifndef DOXYGEN_SHOULD_SKIP_THIS
647
648template
649Fault
650AtomicSimpleCPU::write(Twin32_t data, Addr addr,
651 unsigned flags, uint64_t *res);
652
653template
654Fault
655AtomicSimpleCPU::write(Twin64_t data, Addr addr,
656 unsigned flags, uint64_t *res);
657
658template
659Fault
660AtomicSimpleCPU::write(uint64_t data, Addr addr,
661 unsigned flags, uint64_t *res);
662
663template
664Fault
665AtomicSimpleCPU::write(uint32_t data, Addr addr,
666 unsigned flags, uint64_t *res);
667
668template
669Fault
670AtomicSimpleCPU::write(uint16_t data, Addr addr,
671 unsigned flags, uint64_t *res);
672
673template
674Fault
675AtomicSimpleCPU::write(uint8_t data, Addr addr,
676 unsigned flags, uint64_t *res);
677
678#endif //DOXYGEN_SHOULD_SKIP_THIS
679
680template<>
681Fault
682AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
683{
684 return write(*(uint64_t*)&data, addr, flags, res);
685}
686
687template<>
688Fault
689AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
690{
691 return write(*(uint32_t*)&data, addr, flags, res);
692}
693
694
695template<>
696Fault
697AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
698{
699 return write((uint32_t)data, addr, flags, res);
700}
701
702
703void
704AtomicSimpleCPU::tick()
705{
706 DPRINTF(SimpleCPU, "Tick\n");
707
708 Tick latency = ticks(1); // instruction takes one cycle by default
709
710 for (int i = 0; i < width; ++i) {
711 numCycles++;
712
713 if (!curStaticInst || !curStaticInst->isDelayedCommit())
714 checkForInterrupts();
715
|