atomic.cc revision 3807:1455bc719432 
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, Packet::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, Packet::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, Packet::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
299        assert(pkt->result == Packet::Success);
300        data = pkt->get<T>();
301
302        if (req->isLocked()) {
303            TheISA::handleLockedRead(thread, req);
304        }
305    }
306
307    // This will need a new way to tell if it has a dcache attached.
308    if (req->isUncacheable())
309        recordEvent("Uncached Read");
310
311    return fault;
312}
313
314#ifndef DOXYGEN_SHOULD_SKIP_THIS
315
316template
317Fault
318AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);
319
320template
321Fault
322AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);
323
324template
325Fault
326AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);
327
328template
329Fault
330AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);
331
332#endif //DOXYGEN_SHOULD_SKIP_THIS
333
334template<>
335Fault
336AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags)
337{
338    return read(addr, *(uint64_t*)&data, flags);
339}
340
341template<>
342Fault
343AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags)
344{
345    return read(addr, *(uint32_t*)&data, flags);
346}
347
348
349template<>
350Fault
351AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
352{
353    return read(addr, (uint32_t&)data, flags);
354}
355
356
357template <class T>
358Fault
359AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
360{
361    // use the CPU's statically allocated write request and packet objects
362    Request *req = data_write_req;
363    PacketPtr pkt = data_write_pkt;
364
365    req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
366
367    if (traceData) {
368        traceData->setAddr(addr);
369    }
370
371    // translate to physical address
372    Fault fault = thread->translateDataWriteReq(req);
373
374    // Now do the access.
375    if (fault == NoFault) {
376        bool do_access = true;  // flag to suppress cache access
377
378        if (req->isLocked()) {
379            do_access = TheISA::handleLockedWrite(thread, req);
380        }
381
382        if (do_access) {
383            pkt->reinitFromRequest();
384            pkt->dataStatic(&data);
385
386            if (req->isMmapedIpr()) {
387                dcache_latency = TheISA::handleIprWrite(thread->getTC(), pkt);
388            } else {
389                data = htog(data);
390                dcache_latency = dcachePort.sendAtomic(pkt);
391            }
392            dcache_access = true;
393
394            assert(pkt->result == Packet::Success);
395        }
396
397        if (req->isLocked()) {
398            uint64_t scResult = req->getScResult();
399            if (scResult != 0) {
400                // clear failure counter
401                thread->setStCondFailures(0);
402            }
403            if (res) {
404                *res = req->getScResult();
405            }
406        }
407    }
408
409    // This will need a new way to tell if it's hooked up to a cache or not.
410    if (req->isUncacheable())
411        recordEvent("Uncached Write");
412
413    // If the write needs to have a fault on the access, consider calling
414    // changeStatus() and changing it to "bad addr write" or something.
415    return fault;
416}
417
418
419#ifndef DOXYGEN_SHOULD_SKIP_THIS
420template
421Fault
422AtomicSimpleCPU::write(uint64_t data, Addr addr,
423                       unsigned flags, uint64_t *res);
424
425template
426Fault
427AtomicSimpleCPU::write(uint32_t data, Addr addr,
428                       unsigned flags, uint64_t *res);
429
430template
431Fault
432AtomicSimpleCPU::write(uint16_t data, Addr addr,
433                       unsigned flags, uint64_t *res);
434
435template
436Fault
437AtomicSimpleCPU::write(uint8_t data, Addr addr,
438                       unsigned flags, uint64_t *res);
439
440#endif //DOXYGEN_SHOULD_SKIP_THIS
441
442template<>
443Fault
444AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
445{
446    return write(*(uint64_t*)&data, addr, flags, res);
447}
448
449template<>
450Fault
451AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
452{
453    return write(*(uint32_t*)&data, addr, flags, res);
454}
455
456
457template<>
458Fault
459AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
460{
461    return write((uint32_t)data, addr, flags, res);
462}
463
464
465void
466AtomicSimpleCPU::tick()
467{
468    Tick latency = cycles(1); // instruction takes one cycle by default
469
470    for (int i = 0; i < width; ++i) {
471        numCycles++;
472
473        if (!curStaticInst || !curStaticInst->isDelayedCommit())
474            checkForInterrupts();
475
476        Fault fault = setupFetchRequest(ifetch_req);
477
478        if (fault == NoFault) {
479            ifetch_pkt->reinitFromRequest();
480
481            Tick icache_latency = icachePort.sendAtomic(ifetch_pkt);
482            // ifetch_req is initialized to read the instruction directly
483            // into the CPU object's inst field.
484
485            dcache_access = false; // assume no dcache access
486            preExecute();
487            fault = curStaticInst->execute(this, traceData);
488            postExecute();
489
490            if (simulate_stalls) {
491                Tick icache_stall = icache_latency - cycles(1);
492                Tick dcache_stall =
493                    dcache_access ? dcache_latency - cycles(1) : 0;
494                Tick stall_cycles = (icache_stall + dcache_stall) / cycles(1);
495                if (cycles(stall_cycles) < (icache_stall + dcache_stall))
496                    latency += cycles(stall_cycles+1);
497                else
498                    latency += cycles(stall_cycles);
499            }
500
501        }
502
503        advancePC(fault);
504    }
505
506    if (_status != Idle)
507        tickEvent.schedule(curTick + latency);
508}
509
510
511////////////////////////////////////////////////////////////////////////
512//
513//  AtomicSimpleCPU Simulation Object
514//
515BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
516
517    Param<Counter> max_insts_any_thread;
518    Param<Counter> max_insts_all_threads;
519    Param<Counter> max_loads_any_thread;
520    Param<Counter> max_loads_all_threads;
521    Param<Tick> progress_interval;
522    SimObjectParam<System *> system;
523    Param<int> cpu_id;
524
525#if FULL_SYSTEM
526    SimObjectParam<TheISA::ITB *> itb;
527    SimObjectParam<TheISA::DTB *> dtb;
528    Param<Tick> profile;
529
530    Param<bool> do_quiesce;
531    Param<bool> do_checkpoint_insts;
532    Param<bool> do_statistics_insts;
533#else
534    SimObjectParam<Process *> workload;
535#endif // FULL_SYSTEM
536
537    Param<int> clock;
538    Param<int> phase;
539
540    Param<bool> defer_registration;
541    Param<int> width;
542    Param<bool> function_trace;
543    Param<Tick> function_trace_start;
544    Param<bool> simulate_stalls;
545
546END_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
547
548BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
549
550    INIT_PARAM(max_insts_any_thread,
551               "terminate when any thread reaches this inst count"),
552    INIT_PARAM(max_insts_all_threads,
553               "terminate when all threads have reached this inst count"),
554    INIT_PARAM(max_loads_any_thread,
555               "terminate when any thread reaches this load count"),
556    INIT_PARAM(max_loads_all_threads,
557               "terminate when all threads have reached this load count"),
558    INIT_PARAM(progress_interval, "Progress interval"),
559    INIT_PARAM(system, "system object"),
560    INIT_PARAM(cpu_id, "processor ID"),
561
562#if FULL_SYSTEM
563    INIT_PARAM(itb, "Instruction TLB"),
564    INIT_PARAM(dtb, "Data TLB"),
565    INIT_PARAM(profile, ""),
566    INIT_PARAM(do_quiesce, ""),
567    INIT_PARAM(do_checkpoint_insts, ""),
568    INIT_PARAM(do_statistics_insts, ""),
569#else
570    INIT_PARAM(workload, "processes to run"),
571#endif // FULL_SYSTEM
572
573    INIT_PARAM(clock, "clock speed"),
574    INIT_PARAM_DFLT(phase, "clock phase", 0),
575    INIT_PARAM(defer_registration, "defer system registration (for sampling)"),
576    INIT_PARAM(width, "cpu width"),
577    INIT_PARAM(function_trace, "Enable function trace"),
578    INIT_PARAM(function_trace_start, "Cycle to start function trace"),
579    INIT_PARAM(simulate_stalls, "Simulate cache stall cycles")
580
581END_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
582
583
584CREATE_SIM_OBJECT(AtomicSimpleCPU)
585{
586    AtomicSimpleCPU::Params *params = new AtomicSimpleCPU::Params();
587    params->name = getInstanceName();
588    params->numberOfThreads = 1;
589    params->max_insts_any_thread = max_insts_any_thread;
590    params->max_insts_all_threads = max_insts_all_threads;
591    params->max_loads_any_thread = max_loads_any_thread;
592    params->max_loads_all_threads = max_loads_all_threads;
593    params->progress_interval = progress_interval;
594    params->deferRegistration = defer_registration;
595    params->phase = phase;
596    params->clock = clock;
597    params->functionTrace = function_trace;
598    params->functionTraceStart = function_trace_start;
599    params->width = width;
600    params->simulate_stalls = simulate_stalls;
601    params->system = system;
602    params->cpu_id = cpu_id;
603
604#if FULL_SYSTEM
605    params->itb = itb;
606    params->dtb = dtb;
607    params->profile = profile;
608    params->do_quiesce = do_quiesce;
609    params->do_checkpoint_insts = do_checkpoint_insts;
610    params->do_statistics_insts = do_statistics_insts;
611#else
612    params->process = workload;
613#endif
614
615    AtomicSimpleCPU *cpu = new AtomicSimpleCPU(params);
616    return cpu;
617}
618
619REGISTER_SIM_OBJECT("AtomicSimpleCPU", AtomicSimpleCPU)
620
621