atomic.cc revision 3393:43e1a001a7ce
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/utility.hh" 33#include "cpu/exetrace.hh" 34#include "cpu/simple/atomic.hh" 35#include "mem/packet.hh" 36#include "mem/packet_access.hh" 37#include "sim/builder.hh" 38#include "sim/system.hh" 39 40using namespace std; 41using namespace TheISA; 42 43AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c) 44 : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) 45{ 46} 47 48 49void 50AtomicSimpleCPU::TickEvent::process() 51{ 52 cpu->tick(); 53} 54 55const char * 56AtomicSimpleCPU::TickEvent::description() 57{ 58 return "AtomicSimpleCPU tick event"; 59} 60 61Port * 62AtomicSimpleCPU::getPort(const std::string &if_name, int idx) 63{ 64 if (if_name == "dcache_port") 65 return &dcachePort; 66 else if (if_name == "icache_port") 67 return &icachePort; 68 else 69 panic("No Such Port\n"); 70} 71 72void 73AtomicSimpleCPU::init() 74{ 75 //Create Memory Ports (conect them up) 76// Port *mem_dport = mem->getPort(""); 77// dcachePort.setPeer(mem_dport); 78// mem_dport->setPeer(&dcachePort); 79 80// Port *mem_iport = mem->getPort(""); 81// icachePort.setPeer(mem_iport); 82// mem_iport->setPeer(&icachePort); 83 84 BaseCPU::init(); 85#if FULL_SYSTEM 86 for (int i = 0; i < threadContexts.size(); ++i) { 87 ThreadContext *tc = threadContexts[i]; 88 89 // initialize CPU, including PC 90 TheISA::initCPU(tc, tc->readCpuId()); 91 } 92#endif 93} 94 95bool 96AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt) 97{ 98 panic("AtomicSimpleCPU doesn't expect recvTiming callback!"); 99 return true; 100} 101 102Tick 103AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt) 104{ 105 //Snooping a coherence request, just return 106 return curTick; 107} 108 109void 110AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt) 111{ 112 //No internal storage to update, just return 113 return; 114} 115 116void 117AtomicSimpleCPU::CpuPort::recvStatusChange(Status status) 118{ 119 if (status == RangeChange) 120 return; 121 122 panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!"); 123} 124 125void 126AtomicSimpleCPU::CpuPort::recvRetry() 127{ 128 panic("AtomicSimpleCPU doesn't expect recvRetry callback!"); 129} 130 131 132AtomicSimpleCPU::AtomicSimpleCPU(Params *p) 133 : BaseSimpleCPU(p), tickEvent(this), 134 width(p->width), simulate_stalls(p->simulate_stalls), 135 icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this) 136{ 137 _status = Idle; 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 §ion) 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(curTick); 193 } 194 } 195} 196 197void 198AtomicSimpleCPU::switchOut() 199{ 200 assert(status() == Running || status() == Idle); 201 _status = SwitchedOut; 202 203 tickEvent.squash(); 204} 205 206 207void 208AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU) 209{ 210 BaseCPU::takeOverFrom(oldCPU); 211 212 assert(!tickEvent.scheduled()); 213 214 // if any of this CPU's ThreadContexts are active, mark the CPU as 215 // running and schedule its tick event. 216 for (int i = 0; i < threadContexts.size(); ++i) { 217 ThreadContext *tc = threadContexts[i]; 218 if (tc->status() == ThreadContext::Active && _status != Running) { 219 _status = Running; 220 tickEvent.schedule(curTick); 221 break; 222 } 223 } 224} 225 226 227void 228AtomicSimpleCPU::activateContext(int thread_num, int delay) 229{ 230 assert(thread_num == 0); 231 assert(thread); 232 233 assert(_status == Idle); 234 assert(!tickEvent.scheduled()); 235 236 notIdleFraction++; 237 tickEvent.schedule(curTick + cycles(delay)); 238 _status = Running; 239} 240 241 242void 243AtomicSimpleCPU::suspendContext(int thread_num) 244{ 245 assert(thread_num == 0); 246 assert(thread); 247 248 assert(_status == Running); 249 250 // tick event may not be scheduled if this gets called from inside 251 // an instruction's execution, e.g. "quiesce" 252 if (tickEvent.scheduled()) 253 tickEvent.deschedule(); 254 255 notIdleFraction--; 256 _status = Idle; 257} 258 259 260template <class T> 261Fault 262AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) 263{ 264 // use the CPU's statically allocated read request and packet objects 265 Request *req = data_read_req; 266 PacketPtr pkt = data_read_pkt; 267 268 req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); 269 270 if (traceData) { 271 traceData->setAddr(addr); 272 } 273 274 // translate to physical address 275 Fault fault = thread->translateDataReadReq(req); 276 277 // Now do the access. 278 if (fault == NoFault) { 279 pkt->reinitFromRequest(); 280 281 dcache_latency = dcachePort.sendAtomic(pkt); 282 dcache_access = true; 283 284 assert(pkt->result == Packet::Success); 285 data = pkt->get<T>(); 286 287 if (req->isLocked()) { 288 TheISA::handleLockedRead(thread, req); 289 } 290 } 291 292 // This will need a new way to tell if it has a dcache attached. 293 if (req->isUncacheable()) 294 recordEvent("Uncached Read"); 295 296 return fault; 297} 298 299#ifndef DOXYGEN_SHOULD_SKIP_THIS 300 301template 302Fault 303AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags); 304 305template 306Fault 307AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags); 308 309template 310Fault 311AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags); 312 313template 314Fault 315AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags); 316 317#endif //DOXYGEN_SHOULD_SKIP_THIS 318 319template<> 320Fault 321AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags) 322{ 323 return read(addr, *(uint64_t*)&data, flags); 324} 325 326template<> 327Fault 328AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags) 329{ 330 return read(addr, *(uint32_t*)&data, flags); 331} 332 333 334template<> 335Fault 336AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags) 337{ 338 return read(addr, (uint32_t&)data, flags); 339} 340 341 342template <class T> 343Fault 344AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) 345{ 346 // use the CPU's statically allocated write request and packet objects 347 Request *req = data_write_req; 348 PacketPtr pkt = data_write_pkt; 349 350 req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); 351 352 if (traceData) { 353 traceData->setAddr(addr); 354 } 355 356 // translate to physical address 357 Fault fault = thread->translateDataWriteReq(req); 358 359 // Now do the access. 360 if (fault == NoFault) { 361 bool do_access = true; // flag to suppress cache access 362 363 if (req->isLocked()) { 364 do_access = TheISA::handleLockedWrite(thread, req); 365 } 366 367 if (do_access) { 368 data = htog(data); 369 pkt->reinitFromRequest(); 370 pkt->dataStatic(&data); 371 372 dcache_latency = dcachePort.sendAtomic(pkt); 373 dcache_access = true; 374 375 assert(pkt->result == Packet::Success); 376 } 377 378 if (req->isLocked()) { 379 uint64_t scResult = req->getScResult(); 380 if (scResult != 0) { 381 // clear failure counter 382 thread->setStCondFailures(0); 383 } 384 if (res) { 385 *res = req->getScResult(); 386 } 387 } 388 } 389 390 // This will need a new way to tell if it's hooked up to a cache or not. 391 if (req->isUncacheable()) 392 recordEvent("Uncached Write"); 393 394 // If the write needs to have a fault on the access, consider calling 395 // changeStatus() and changing it to "bad addr write" or something. 396 return fault; 397} 398 399 400#ifndef DOXYGEN_SHOULD_SKIP_THIS 401template 402Fault 403AtomicSimpleCPU::write(uint64_t data, Addr addr, 404 unsigned flags, uint64_t *res); 405 406template 407Fault 408AtomicSimpleCPU::write(uint32_t data, Addr addr, 409 unsigned flags, uint64_t *res); 410 411template 412Fault 413AtomicSimpleCPU::write(uint16_t data, Addr addr, 414 unsigned flags, uint64_t *res); 415 416template 417Fault 418AtomicSimpleCPU::write(uint8_t data, Addr addr, 419 unsigned flags, uint64_t *res); 420 421#endif //DOXYGEN_SHOULD_SKIP_THIS 422 423template<> 424Fault 425AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res) 426{ 427 return write(*(uint64_t*)&data, addr, flags, res); 428} 429 430template<> 431Fault 432AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res) 433{ 434 return write(*(uint32_t*)&data, addr, flags, res); 435} 436 437 438template<> 439Fault 440AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res) 441{ 442 return write((uint32_t)data, addr, flags, res); 443} 444 445 446void 447AtomicSimpleCPU::tick() 448{ 449 Tick latency = cycles(1); // instruction takes one cycle by default 450 451 for (int i = 0; i < width; ++i) { 452 numCycles++; 453 454 if (!curStaticInst || !curStaticInst->isDelayedCommit()) 455 checkForInterrupts(); 456 457 Fault fault = setupFetchRequest(ifetch_req); 458 459 if (fault == NoFault) { 460 ifetch_pkt->reinitFromRequest(); 461 462 Tick icache_latency = icachePort.sendAtomic(ifetch_pkt); 463 // ifetch_req is initialized to read the instruction directly 464 // into the CPU object's inst field. 465 466 dcache_access = false; // assume no dcache access 467 preExecute(); 468 fault = curStaticInst->execute(this, traceData); 469 postExecute(); 470 471 if (simulate_stalls) { 472 Tick icache_stall = icache_latency - cycles(1); 473 Tick dcache_stall = 474 dcache_access ? dcache_latency - cycles(1) : 0; 475 Tick stall_cycles = (icache_stall + dcache_stall) / cycles(1); 476 if (cycles(stall_cycles) < (icache_stall + dcache_stall)) 477 latency += cycles(stall_cycles+1); 478 else 479 latency += cycles(stall_cycles); 480 } 481 482 } 483 484 advancePC(fault); 485 } 486 487 if (_status != Idle) 488 tickEvent.schedule(curTick + latency); 489} 490 491 492//////////////////////////////////////////////////////////////////////// 493// 494// AtomicSimpleCPU Simulation Object 495// 496BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 497 498 Param<Counter> max_insts_any_thread; 499 Param<Counter> max_insts_all_threads; 500 Param<Counter> max_loads_any_thread; 501 Param<Counter> max_loads_all_threads; 502 Param<Tick> progress_interval; 503 SimObjectParam<MemObject *> mem; 504 SimObjectParam<System *> system; 505 Param<int> cpu_id; 506 507#if FULL_SYSTEM 508 SimObjectParam<AlphaITB *> itb; 509 SimObjectParam<AlphaDTB *> dtb; 510 Param<Tick> profile; 511#else 512 SimObjectParam<Process *> workload; 513#endif // FULL_SYSTEM 514 515 Param<int> clock; 516 517 Param<bool> defer_registration; 518 Param<int> width; 519 Param<bool> function_trace; 520 Param<Tick> function_trace_start; 521 Param<bool> simulate_stalls; 522 523END_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 524 525BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 526 527 INIT_PARAM(max_insts_any_thread, 528 "terminate when any thread reaches this inst count"), 529 INIT_PARAM(max_insts_all_threads, 530 "terminate when all threads have reached this inst count"), 531 INIT_PARAM(max_loads_any_thread, 532 "terminate when any thread reaches this load count"), 533 INIT_PARAM(max_loads_all_threads, 534 "terminate when all threads have reached this load count"), 535 INIT_PARAM(progress_interval, "Progress interval"), 536 INIT_PARAM(mem, "memory"), 537 INIT_PARAM(system, "system object"), 538 INIT_PARAM(cpu_id, "processor ID"), 539 540#if FULL_SYSTEM 541 INIT_PARAM(itb, "Instruction TLB"), 542 INIT_PARAM(dtb, "Data TLB"), 543 INIT_PARAM(profile, ""), 544#else 545 INIT_PARAM(workload, "processes to run"), 546#endif // FULL_SYSTEM 547 548 INIT_PARAM(clock, "clock speed"), 549 INIT_PARAM(defer_registration, "defer system registration (for sampling)"), 550 INIT_PARAM(width, "cpu width"), 551 INIT_PARAM(function_trace, "Enable function trace"), 552 INIT_PARAM(function_trace_start, "Cycle to start function trace"), 553 INIT_PARAM(simulate_stalls, "Simulate cache stall cycles") 554 555END_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU) 556 557 558CREATE_SIM_OBJECT(AtomicSimpleCPU) 559{ 560 AtomicSimpleCPU::Params *params = new AtomicSimpleCPU::Params(); 561 params->name = getInstanceName(); 562 params->numberOfThreads = 1; 563 params->max_insts_any_thread = max_insts_any_thread; 564 params->max_insts_all_threads = max_insts_all_threads; 565 params->max_loads_any_thread = max_loads_any_thread; 566 params->max_loads_all_threads = max_loads_all_threads; 567 params->progress_interval = progress_interval; 568 params->deferRegistration = defer_registration; 569 params->clock = clock; 570 params->functionTrace = function_trace; 571 params->functionTraceStart = function_trace_start; 572 params->width = width; 573 params->simulate_stalls = simulate_stalls; 574 params->mem = mem; 575 params->system = system; 576 params->cpu_id = cpu_id; 577 578#if FULL_SYSTEM 579 params->itb = itb; 580 params->dtb = dtb; 581 params->profile = profile; 582#else 583 params->process = workload; 584#endif 585 586 AtomicSimpleCPU *cpu = new AtomicSimpleCPU(params); 587 return cpu; 588} 589 590REGISTER_SIM_OBJECT("AtomicSimpleCPU", AtomicSimpleCPU) 591 592