atomic.cc revision 4022:c422464ca16e 
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 "cpu/exetrace.hh"
35#include "cpu/simple/atomic.hh"
36#include "mem/packet.hh"
37#include "mem/packet_access.hh"
38#include "sim/builder.hh"
39#include "sim/system.hh"
40
41using namespace std;
42using namespace TheISA;
43
44AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c)
45    : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c)
46{
47}
48
49
50void
51AtomicSimpleCPU::TickEvent::process()
52{
53    cpu->tick();
54}
55
56const char *
57AtomicSimpleCPU::TickEvent::description()
58{
59    return "AtomicSimpleCPU tick event";
60}
61
62Port *
63AtomicSimpleCPU::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
74AtomicSimpleCPU::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, tc->readCpuId());
83    }
84#endif
85}
86
87bool
88AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt)
89{
90    panic("AtomicSimpleCPU doesn't expect recvTiming callback!");
91    return true;
92}
93
94Tick
95AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt)
96{
97    //Snooping a coherence request, just return
98    return 0;
99}
100
101void
102AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt)
103{
104    //No internal storage to update, just return
105    return;
106}
107
108void
109AtomicSimpleCPU::CpuPort::recvStatusChange(Status status)
110{
111    if (status == RangeChange) {
112        if (!snoopRangeSent) {
113            snoopRangeSent = true;
114            sendStatusChange(Port::RangeChange);
115        }
116        return;
117    }
118
119    panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!");
120}
121
122void
123AtomicSimpleCPU::CpuPort::recvRetry()
124{
125    panic("AtomicSimpleCPU doesn't expect recvRetry callback!");
126}
127
128
129AtomicSimpleCPU::AtomicSimpleCPU(Params *p)
130    : BaseSimpleCPU(p), tickEvent(this),
131      width(p->width), simulate_stalls(p->simulate_stalls),
132      icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this)
133{
134    _status = Idle;
135
136    icachePort.snoopRangeSent = false;
137    dcachePort.snoopRangeSent = false;
138
139    ifetch_req = new Request();
140    ifetch_req->setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT
141    ifetch_pkt = new Packet(ifetch_req, MemCmd::ReadReq, Packet::Broadcast);
142    ifetch_pkt->dataStatic(&inst);
143
144    data_read_req = new Request();
145    data_read_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too
146    data_read_pkt = new Packet(data_read_req, MemCmd::ReadReq,
147                               Packet::Broadcast);
148    data_read_pkt->dataStatic(&dataReg);
149
150    data_write_req = new Request();
151    data_write_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too
152    data_write_pkt = new Packet(data_write_req, MemCmd::WriteReq,
153                                Packet::Broadcast);
154}
155
156
157AtomicSimpleCPU::~AtomicSimpleCPU()
158{
159}
160
161void
162AtomicSimpleCPU::serialize(ostream &os)
163{
164    SimObject::State so_state = SimObject::getState();
165    SERIALIZE_ENUM(so_state);
166    Status _status = status();
167    SERIALIZE_ENUM(_status);
168    BaseSimpleCPU::serialize(os);
169    nameOut(os, csprintf("%s.tickEvent", name()));
170    tickEvent.serialize(os);
171}
172
173void
174AtomicSimpleCPU::unserialize(Checkpoint *cp, const string &section)
175{
176    SimObject::State so_state;
177    UNSERIALIZE_ENUM(so_state);
178    UNSERIALIZE_ENUM(_status);
179    BaseSimpleCPU::unserialize(cp, section);
180    tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
181}
182
183void
184AtomicSimpleCPU::resume()
185{
186    if (_status != SwitchedOut && _status != Idle) {
187        assert(system->getMemoryMode() == System::Atomic);
188
189        changeState(SimObject::Running);
190        if (thread->status() == ThreadContext::Active) {
191            if (!tickEvent.scheduled()) {
192                tickEvent.schedule(nextCycle());
193            }
194        }
195    }
196}
197
198void
199AtomicSimpleCPU::switchOut()
200{
201    assert(status() == Running || status() == Idle);
202    _status = SwitchedOut;
203
204    tickEvent.squash();
205}
206
207
208void
209AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
210{
211    BaseCPU::takeOverFrom(oldCPU);
212
213    assert(!tickEvent.scheduled());
214
215    // if any of this CPU's ThreadContexts are active, mark the CPU as
216    // running and schedule its tick event.
217    for (int i = 0; i < threadContexts.size(); ++i) {
218        ThreadContext *tc = threadContexts[i];
219        if (tc->status() == ThreadContext::Active && _status != Running) {
220            _status = Running;
221            tickEvent.schedule(nextCycle());
222            break;
223        }
224    }
225    if (_status != Running) {
226        _status = Idle;
227    }
228}
229
230
231void
232AtomicSimpleCPU::activateContext(int thread_num, int delay)
233{
234    assert(thread_num == 0);
235    assert(thread);
236
237    assert(_status == Idle);
238    assert(!tickEvent.scheduled());
239
240    notIdleFraction++;
241
242#if FULL_SYSTEM
243    // Connect the ThreadContext's memory ports (Functional/Virtual
244    // Ports)
245    tc->connectMemPorts();
246#endif
247
248    //Make sure ticks are still on multiples of cycles
249    tickEvent.schedule(nextCycle(curTick + cycles(delay)));
250    _status = Running;
251}
252
253
254void
255AtomicSimpleCPU::suspendContext(int thread_num)
256{
257    assert(thread_num == 0);
258    assert(thread);
259
260    assert(_status == Running);
261
262    // tick event may not be scheduled if this gets called from inside
263    // an instruction's execution, e.g. "quiesce"
264    if (tickEvent.scheduled())
265        tickEvent.deschedule();
266
267    notIdleFraction--;
268    _status = Idle;
269}
270
271
272template <class T>
273Fault
274AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags)
275{
276    // use the CPU's statically allocated read request and packet objects
277    Request *req = data_read_req;
278    PacketPtr pkt = data_read_pkt;
279
280    req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
281
282    if (traceData) {
283        traceData->setAddr(addr);
284    }
285
286    // translate to physical address
287    Fault fault = thread->translateDataReadReq(req);
288
289    // Now do the access.
290    if (fault == NoFault) {
291        pkt->reinitFromRequest();
292
293        if (req->isMmapedIpr())
294            dcache_latency = TheISA::handleIprRead(thread->getTC(),pkt);
295        else
296            dcache_latency = dcachePort.sendAtomic(pkt);
297        dcache_access = true;
298#if !defined(NDEBUG)
299        if (pkt->result != Packet::Success)
300            panic("Unable to find responder for address pa = %#X va = %#X\n",
301                    pkt->req->getPaddr(), pkt->req->getVaddr());
302#endif
303        data = pkt->get<T>();
304
305        if (req->isLocked()) {
306            TheISA::handleLockedRead(thread, req);
307        }
308    }
309
310    // This will need a new way to tell if it has a dcache attached.
311    if (req->isUncacheable())
312        recordEvent("Uncached Read");
313
314    return fault;
315}
316
317#ifndef DOXYGEN_SHOULD_SKIP_THIS
318
319template
320Fault
321AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);
322
323template
324Fault
325AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);
326
327template
328Fault
329AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);
330
331template
332Fault
333AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);
334
335#endif //DOXYGEN_SHOULD_SKIP_THIS
336
337template<>
338Fault
339AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags)
340{
341    return read(addr, *(uint64_t*)&data, flags);
342}
343
344template<>
345Fault
346AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags)
347{
348    return read(addr, *(uint32_t*)&data, flags);
349}
350
351
352template<>
353Fault
354AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
355{
356    return read(addr, (uint32_t&)data, flags);
357}
358
359
360template <class T>
361Fault
362AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
363{
364    // use the CPU's statically allocated write request and packet objects
365    Request *req = data_write_req;
366    PacketPtr pkt = data_write_pkt;
367
368    req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
369
370    if (traceData) {
371        traceData->setAddr(addr);
372    }
373
374    // translate to physical address
375    Fault fault = thread->translateDataWriteReq(req);
376
377    // Now do the access.
378    if (fault == NoFault) {
379        bool do_access = true;  // flag to suppress cache access
380
381        if (req->isLocked()) {
382            do_access = TheISA::handleLockedWrite(thread, req);
383        }
384
385        if (do_access) {
386            pkt->reinitFromRequest();
387            pkt->dataStatic(&data);
388
389            if (req->isMmapedIpr()) {
390                dcache_latency = TheISA::handleIprWrite(thread->getTC(), pkt);
391            } else {
392                data = htog(data);
393                dcache_latency = dcachePort.sendAtomic(pkt);
394            }
395            dcache_access = true;
396
397#if !defined(NDEBUG)
398            if (pkt->result != Packet::Success)
399                panic("Unable to find responder for address pa = %#X va = %#X\n",
400                        pkt->req->getPaddr(), pkt->req->getVaddr());
401#endif
402        }
403
404        if (req->isLocked()) {
405            uint64_t scResult = req->getScResult();
406            if (scResult != 0) {
407                // clear failure counter
408                thread->setStCondFailures(0);
409            }
410            if (res) {
411                *res = req->getScResult();
412            }
413        }
414    }
415
416    // This will need a new way to tell if it's hooked up to a cache or not.
417    if (req->isUncacheable())
418        recordEvent("Uncached Write");
419
420    // If the write needs to have a fault on the access, consider calling
421    // changeStatus() and changing it to "bad addr write" or something.
422    return fault;
423}
424
425
426#ifndef DOXYGEN_SHOULD_SKIP_THIS
427template
428Fault
429AtomicSimpleCPU::write(uint64_t data, Addr addr,
430                       unsigned flags, uint64_t *res);
431
432template
433Fault
434AtomicSimpleCPU::write(uint32_t data, Addr addr,
435                       unsigned flags, uint64_t *res);
436
437template
438Fault
439AtomicSimpleCPU::write(uint16_t data, Addr addr,
440                       unsigned flags, uint64_t *res);
441
442template
443Fault
444AtomicSimpleCPU::write(uint8_t data, Addr addr,
445                       unsigned flags, uint64_t *res);
446
447#endif //DOXYGEN_SHOULD_SKIP_THIS
448
449template<>
450Fault
451AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
452{
453    return write(*(uint64_t*)&data, addr, flags, res);
454}
455
456template<>
457Fault
458AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
459{
460    return write(*(uint32_t*)&data, addr, flags, res);
461}
462
463
464template<>
465Fault
466AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
467{
468    return write((uint32_t)data, addr, flags, res);
469}
470
471
472void
473AtomicSimpleCPU::tick()
474{
475    Tick latency = cycles(1); // instruction takes one cycle by default
476
477    for (int i = 0; i < width; ++i) {
478        numCycles++;
479
480        if (!curStaticInst || !curStaticInst->isDelayedCommit())
481            checkForInterrupts();
482
483        Fault fault = setupFetchRequest(ifetch_req);
484
485        if (fault == NoFault) {
486            ifetch_pkt->reinitFromRequest();
487
488            Tick icache_latency = icachePort.sendAtomic(ifetch_pkt);
489            // ifetch_req is initialized to read the instruction directly
490            // into the CPU object's inst field.
491
492            dcache_access = false; // assume no dcache access
493            preExecute();
494
495            fault = curStaticInst->execute(this, traceData);
496            postExecute();
497
498            // @todo remove me after debugging with legion done
499            if (curStaticInst && (!curStaticInst->isMicroOp() ||
500                        curStaticInst->isFirstMicroOp()))
501                instCnt++;
502
503            if (simulate_stalls) {
504                Tick icache_stall = icache_latency - cycles(1);
505                Tick dcache_stall =
506                    dcache_access ? dcache_latency - cycles(1) : 0;
507                Tick stall_cycles = (icache_stall + dcache_stall) / cycles(1);
508                if (cycles(stall_cycles) < (icache_stall + dcache_stall))
509                    latency += cycles(stall_cycles+1);
510                else
511                    latency += cycles(stall_cycles);
512            }
513
514        }
515
516        advancePC(fault);
517    }
518
519    if (_status != Idle)
520        tickEvent.schedule(curTick + latency);
521}
522
523
524////////////////////////////////////////////////////////////////////////
525//
526//  AtomicSimpleCPU Simulation Object
527//
528BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
529
530    Param<Counter> max_insts_any_thread;
531    Param<Counter> max_insts_all_threads;
532    Param<Counter> max_loads_any_thread;
533    Param<Counter> max_loads_all_threads;
534    Param<Tick> progress_interval;
535    SimObjectParam<System *> system;
536    Param<int> cpu_id;
537
538#if FULL_SYSTEM
539    SimObjectParam<TheISA::ITB *> itb;
540    SimObjectParam<TheISA::DTB *> dtb;
541    Param<Tick> profile;
542
543    Param<bool> do_quiesce;
544    Param<bool> do_checkpoint_insts;
545    Param<bool> do_statistics_insts;
546#else
547    SimObjectParam<Process *> workload;
548#endif // FULL_SYSTEM
549
550    Param<int> clock;
551    Param<int> phase;
552
553    Param<bool> defer_registration;
554    Param<int> width;
555    Param<bool> function_trace;
556    Param<Tick> function_trace_start;
557    Param<bool> simulate_stalls;
558
559END_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
560
561BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
562
563    INIT_PARAM(max_insts_any_thread,
564               "terminate when any thread reaches this inst count"),
565    INIT_PARAM(max_insts_all_threads,
566               "terminate when all threads have reached this inst count"),
567    INIT_PARAM(max_loads_any_thread,
568               "terminate when any thread reaches this load count"),
569    INIT_PARAM(max_loads_all_threads,
570               "terminate when all threads have reached this load count"),
571    INIT_PARAM(progress_interval, "Progress interval"),
572    INIT_PARAM(system, "system object"),
573    INIT_PARAM(cpu_id, "processor ID"),
574
575#if FULL_SYSTEM
576    INIT_PARAM(itb, "Instruction TLB"),
577    INIT_PARAM(dtb, "Data TLB"),
578    INIT_PARAM(profile, ""),
579    INIT_PARAM(do_quiesce, ""),
580    INIT_PARAM(do_checkpoint_insts, ""),
581    INIT_PARAM(do_statistics_insts, ""),
582#else
583    INIT_PARAM(workload, "processes to run"),
584#endif // FULL_SYSTEM
585
586    INIT_PARAM(clock, "clock speed"),
587    INIT_PARAM_DFLT(phase, "clock phase", 0),
588    INIT_PARAM(defer_registration, "defer system registration (for sampling)"),
589    INIT_PARAM(width, "cpu width"),
590    INIT_PARAM(function_trace, "Enable function trace"),
591    INIT_PARAM(function_trace_start, "Cycle to start function trace"),
592    INIT_PARAM(simulate_stalls, "Simulate cache stall cycles")
593
594END_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
595
596
597CREATE_SIM_OBJECT(AtomicSimpleCPU)
598{
599    AtomicSimpleCPU::Params *params = new AtomicSimpleCPU::Params();
600    params->name = getInstanceName();
601    params->numberOfThreads = 1;
602    params->max_insts_any_thread = max_insts_any_thread;
603    params->max_insts_all_threads = max_insts_all_threads;
604    params->max_loads_any_thread = max_loads_any_thread;
605    params->max_loads_all_threads = max_loads_all_threads;
606    params->progress_interval = progress_interval;
607    params->deferRegistration = defer_registration;
608    params->phase = phase;
609    params->clock = clock;
610    params->functionTrace = function_trace;
611    params->functionTraceStart = function_trace_start;
612    params->width = width;
613    params->simulate_stalls = simulate_stalls;
614    params->system = system;
615    params->cpu_id = cpu_id;
616
617#if FULL_SYSTEM
618    params->itb = itb;
619    params->dtb = dtb;
620    params->profile = profile;
621    params->do_quiesce = do_quiesce;
622    params->do_checkpoint_insts = do_checkpoint_insts;
623    params->do_statistics_insts = do_statistics_insts;
624#else
625    params->process = workload;
626#endif
627
628    AtomicSimpleCPU *cpu = new AtomicSimpleCPU(params);
629    return cpu;
630}
631
632REGISTER_SIM_OBJECT("AtomicSimpleCPU", AtomicSimpleCPU)
633
634