src/cpu/base.cc

   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  *          Nathan Binkert
  30  */
  31
  32 #include <iostream>
  33 #include <string>
  34 #include <sstream>
  35
  36 #include "base/cprintf.hh"
  37 #include "base/loader/symtab.hh"
  38 #include "base/misc.hh"
  39 #include "base/output.hh"
  40 #include "base/trace.hh"
  41 #include "cpu/base.hh"
  42 #include "cpu/cpuevent.hh"
  43 #include "cpu/thread_context.hh"
  44 #include "cpu/profile.hh"
  45 #include "params/BaseCPU.hh"
  46 #include "sim/sim_exit.hh"
  47 #include "sim/process.hh"
  48 #include "sim/sim_events.hh"
  49 #include "sim/system.hh"
  50
  51 // Hack
  52 #include "sim/stat_control.hh"
  53
  54 using namespace std;
  55
  56 vector<BaseCPU *> BaseCPU::cpuList;
  57
  58 // This variable reflects the max number of threads in any CPU.  Be
  59 // careful to only use it once all the CPUs that you care about have
  60 // been initialized
  61 int maxThreadsPerCPU = 1;
  62
  63 CPUProgressEvent::CPUProgressEvent(BaseCPU *_cpu, Tick ival)
  64     : Event(Event::Progress_Event_Pri), interval(ival), lastNumInst(0),
  65       cpu(_cpu)
  66 {
  67     if (interval)
  68         cpu->schedule(this, curTick + interval);
  69 }
  70
  71 void
  72 CPUProgressEvent::process()
  73 {
  74     Counter temp = cpu->totalInstructions();
  75 #ifndef NDEBUG
  76     double ipc = double(temp - lastNumInst) / (interval / cpu->ticks(1));
  77
  78     DPRINTFN("%s progress event, instructions committed: %lli, IPC: %0.8d\n",
  79              cpu->name(), temp - lastNumInst, ipc);
  80     ipc = 0.0;
  81 #else
  82     cprintf("%lli: %s progress event, instructions committed: %lli\n",
  83             curTick, cpu->name(), temp - lastNumInst);
  84 #endif
  85     lastNumInst = temp;
  86     cpu->schedule(this, curTick + interval);
  87 }
  88
  89 const char *
  90 CPUProgressEvent::description() const
  91 {
  92     return "CPU Progress";
  93 }
  94
  95 #if FULL_SYSTEM
  96 BaseCPU::BaseCPU(Params *p)
  97     : MemObject(p), clock(p->clock), instCnt(0), _cpuId(p->cpu_id),
  98       interrupts(p->interrupts),
  99       number_of_threads(p->numThreads), system(p->system),
 100       phase(p->phase)
 101 #else
 102 BaseCPU::BaseCPU(Params *p)
 103     : MemObject(p), clock(p->clock), _cpuId(p->cpu_id),
 104       number_of_threads(p->numThreads), system(p->system),
 105       phase(p->phase)
 106 #endif
 107 {
 108 //    currentTick = curTick;
 109
 110     // if Python did not provide a valid ID, do it here
 111     if (_cpuId == -1 ) {
 112         _cpuId = cpuList.size();
 113     }
 114
 115     // add self to global list of CPUs
 116     cpuList.push_back(this);
 117
 118     DPRINTF(SyscallVerbose, "Constructing CPU with id %d\n", _cpuId);
 119
 120     if (number_of_threads > maxThreadsPerCPU)
 121         maxThreadsPerCPU = number_of_threads;
 122
 123     // allocate per-thread instruction-based event queues
 124     comInstEventQueue = new EventQueue *[number_of_threads];
 125     for (int i = 0; i < number_of_threads; ++i)
 126         comInstEventQueue[i] = new EventQueue("instruction-based event queue");
 127
 128     //
 129     // set up instruction-count-based termination events, if any
 130     //
 131     if (p->max_insts_any_thread != 0) {
 132         const char *cause = "a thread reached the max instruction count";
 133         for (int i = 0; i < number_of_threads; ++i) {
 134             Event *event = new SimLoopExitEvent(cause, 0);
 135             comInstEventQueue[i]->schedule(event, p->max_insts_any_thread);
 136         }
 137     }
 138
 139     if (p->max_insts_all_threads != 0) {
 140         const char *cause = "all threads reached the max instruction count";
 141
 142         // allocate & initialize shared downcounter: each event will
 143         // decrement this when triggered; simulation will terminate
 144         // when counter reaches 0
 145         int *counter = new int;
 146         *counter = number_of_threads;
 147         for (int i = 0; i < number_of_threads; ++i) {
 148             Event *event = new CountedExitEvent(cause, *counter);
 149             comInstEventQueue[i]->schedule(event, p->max_insts_any_thread);
 150         }
 151     }
 152
 153     // allocate per-thread load-based event queues
 154     comLoadEventQueue = new EventQueue *[number_of_threads];
 155     for (int i = 0; i < number_of_threads; ++i)
 156         comLoadEventQueue[i] = new EventQueue("load-based event queue");
 157
 158     //
 159     // set up instruction-count-based termination events, if any
 160     //
 161     if (p->max_loads_any_thread != 0) {
 162         const char *cause = "a thread reached the max load count";
 163         for (int i = 0; i < number_of_threads; ++i) {
 164             Event *event = new SimLoopExitEvent(cause, 0);
 165             comLoadEventQueue[i]->schedule(event, p->max_loads_any_thread);
 166         }
 167     }
 168
 169     if (p->max_loads_all_threads != 0) {
 170         const char *cause = "all threads reached the max load count";
 171         // allocate & initialize shared downcounter: each event will
 172         // decrement this when triggered; simulation will terminate
 173         // when counter reaches 0
 174         int *counter = new int;
 175         *counter = number_of_threads;
 176         for (int i = 0; i < number_of_threads; ++i) {
 177             Event *event = new CountedExitEvent(cause, *counter);
 178             comLoadEventQueue[i]->schedule(event, p->max_loads_all_threads);
 179         }
 180     }
 181
 182     functionTracingEnabled = false;
 183     if (p->function_trace) {
 184         functionTraceStream = simout.find(csprintf("ftrace.%s", name()));
 185         currentFunctionStart = currentFunctionEnd = 0;
 186         functionEntryTick = p->function_trace_start;
 187
 188         if (p->function_trace_start == 0) {
 189             functionTracingEnabled = true;
 190         } else {
 191             typedef EventWrapper<BaseCPU, &BaseCPU::enableFunctionTrace> wrap;
 192             Event *event = new wrap(this, true);
 193             schedule(event, p->function_trace_start);
 194         }
 195     }
 196 #if FULL_SYSTEM
 197     interrupts->setCPU(this);
 198
 199     profileEvent = NULL;
 200     if (params()->profile)
 201         profileEvent = new ProfileEvent(this, params()->profile);
 202 #endif
 203     tracer = params()->tracer;
 204 }
 205
 206 void
 207 BaseCPU::enableFunctionTrace()
 208 {
 209     functionTracingEnabled = true;
 210 }
 211
 212 BaseCPU::~BaseCPU()
 213 {
 214 }
 215
 216 void
 217 BaseCPU::init()
 218 {
 219     if (!params()->defer_registration)
 220         registerThreadContexts();
 221 }
 222
 223 void
 224 BaseCPU::startup()
 225 {
 226 #if FULL_SYSTEM
 227     if (!params()->defer_registration && profileEvent)
 228         schedule(profileEvent, curTick);
 229 #endif
 230
 231     if (params()->progress_interval) {
 232         Tick num_ticks = ticks(params()->progress_interval);
 233         Event *event = new CPUProgressEvent(this, num_ticks);
 234         schedule(event, curTick + num_ticks);
 235     }
 236 }
 237
 238
 239 void
 240 BaseCPU::regStats()
 241 {
 242     using namespace Stats;
 243
 244     numCycles
 245         .name(name() + ".numCycles")
 246         .desc("number of cpu cycles simulated")
 247         ;
 248
 249     int size = threadContexts.size();
 250     if (size > 1) {
 251         for (int i = 0; i < size; ++i) {
 252             stringstream namestr;
 253             ccprintf(namestr, "%s.ctx%d", name(), i);
 254             threadContexts[i]->regStats(namestr.str());
 255         }
 256     } else if (size == 1)
 257         threadContexts[0]->regStats(name());
 258
 259 #if FULL_SYSTEM
 260 #endif
 261 }
 262
 263 Tick
 264 BaseCPU::nextCycle()
 265 {
 266     Tick next_tick = curTick - phase + clock - 1;
 267     next_tick -= (next_tick % clock);
 268     next_tick += phase;
 269     return next_tick;
 270 }
 271
 272 Tick
 273 BaseCPU::nextCycle(Tick begin_tick)
 274 {
 275     Tick next_tick = begin_tick;
 276     if (next_tick % clock != 0)
 277         next_tick = next_tick - (next_tick % clock) + clock;
 278     next_tick += phase;
 279
 280     assert(next_tick >= curTick);
 281     return next_tick;
 282 }
 283
 284 void
 285 BaseCPU::registerThreadContexts()
 286 {
 287     for (int i = 0; i < threadContexts.size(); ++i) {
 288         ThreadContext *tc = threadContexts[i];
 289
 290         /** This is so that contextId and cpuId match where there is a
 291          * 1cpu:1context relationship.  Otherwise, the order of registration
 292          * could affect the assignment and cpu 1 could have context id 3, for
 293          * example.  We may even want to do something like this for SMT so that
 294          * cpu 0 has the lowest thread contexts and cpu N has the highest, but
 295          * I'll just do this for now
 296          */
 297         if (number_of_threads == 1)
 298             tc->setContextId(system->registerThreadContext(tc, _cpuId));
 299         else
 300             tc->setContextId(system->registerThreadContext(tc));
 301 #if !FULL_SYSTEM
 302         tc->getProcessPtr()->assignThreadContext(tc->contextId());
 303 #endif
 304     }
 305 }
 306
 307
 308 int
 309 BaseCPU::findContext(ThreadContext *tc)
 310 {
 311     for (int i = 0; i < threadContexts.size(); ++i) {
 312         if (tc == threadContexts[i])
 313             return i;
 314     }
 315     return 0;
 316 }
 317
 318 void
 319 BaseCPU::switchOut()
 320 {
 321 //    panic("This CPU doesn't support sampling!");
 322 #if FULL_SYSTEM
 323     if (profileEvent && profileEvent->scheduled())
 324         deschedule(profileEvent);
 325 #endif
 326 }
 327
 328 void
 329 BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc)
 330 {
 331     assert(threadContexts.size() == oldCPU->threadContexts.size());
 332
 333     _cpuId = oldCPU->cpuId();
 334
 335     for (int i = 0; i < threadContexts.size(); ++i) {
 336         ThreadContext *newTC = threadContexts[i];
 337         ThreadContext *oldTC = oldCPU->threadContexts[i];
 338
 339         newTC->takeOverFrom(oldTC);
 340
 341         CpuEvent::replaceThreadContext(oldTC, newTC);
 342
 343         assert(newTC->contextId() == oldTC->contextId());
 344         assert(newTC->threadId() == oldTC->threadId());
 345         system->replaceThreadContext(newTC, newTC->contextId());
 346
 347         /* This code no longer works since the zero register (e.g.,
 348          * r31 on Alpha) doesn't necessarily contain zero at this
 349          * point.
 350            if (DTRACE(Context))
 351             ThreadContext::compare(oldTC, newTC);
 352         */
 353     }
 354
 355 #if FULL_SYSTEM
 356     interrupts = oldCPU->interrupts;
 357     interrupts->setCPU(this);
 358
 359     for (int i = 0; i < threadContexts.size(); ++i)
 360         threadContexts[i]->profileClear();
 361
 362     if (profileEvent)
 363         schedule(profileEvent, curTick);
 364 #endif
 365
 366     // Connect new CPU to old CPU's memory only if new CPU isn't
 367     // connected to anything.  Also connect old CPU's memory to new
 368     // CPU.
 369     if (!ic->isConnected()) {
 370         Port *peer = oldCPU->getPort("icache_port")->getPeer();
 371         ic->setPeer(peer);
 372         peer->setPeer(ic);
 373     }
 374
 375     if (!dc->isConnected()) {
 376         Port *peer = oldCPU->getPort("dcache_port")->getPeer();
 377         dc->setPeer(peer);
 378         peer->setPeer(dc);
 379     }
 380 }
 381
 382
 383 #if FULL_SYSTEM
 384 BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, Tick _interval)
 385     : cpu(_cpu), interval(_interval)
 386 { }
 387
 388 void
 389 BaseCPU::ProfileEvent::process()
 390 {
 391     for (int i = 0, size = cpu->threadContexts.size(); i < size; ++i) {
 392         ThreadContext *tc = cpu->threadContexts[i];
 393         tc->profileSample();
 394     }
 395
 396     cpu->schedule(this, curTick + interval);
 397 }
 398
 399 void
 400 BaseCPU::serialize(std::ostream &os)
 401 {
 402     SERIALIZE_SCALAR(instCnt);
 403     interrupts->serialize(os);
 404 }
 405
 406 void
 407 BaseCPU::unserialize(Checkpoint *cp, const std::string &section)
 408 {
 409     UNSERIALIZE_SCALAR(instCnt);
 410     interrupts->unserialize(cp, section);
 411 }
 412
 413 #endif // FULL_SYSTEM
 414
 415 void
 416 BaseCPU::traceFunctionsInternal(Addr pc)
 417 {
 418     if (!debugSymbolTable)
 419         return;
 420
 421     // if pc enters different function, print new function symbol and
 422     // update saved range.  Otherwise do nothing.
 423     if (pc < currentFunctionStart || pc >= currentFunctionEnd) {
 424         string sym_str;
 425         bool found = debugSymbolTable->findNearestSymbol(pc, sym_str,
 426                                                          currentFunctionStart,
 427                                                          currentFunctionEnd);
 428
 429         if (!found) {
 430             // no symbol found: use addr as label
 431             sym_str = csprintf("0x%x", pc);
 432             currentFunctionStart = pc;
 433             currentFunctionEnd = pc + 1;
 434         }
 435
 436         ccprintf(*functionTraceStream, " (%d)\n%d: %s",
 437                  curTick - functionEntryTick, curTick, sym_str);
 438         functionEntryTick = curTick;
 439     }
 440 }