atomic.cc revision 5169
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() 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#if FULL_SYSTEM 83 for (int i = 0; i < threadContexts.size(); ++i) { 84 ThreadContext *tc = threadContexts[i]; 85 86 // initialize CPU, including PC 87 TheISA::initCPU(tc, tc->readCpuId()); 88 } 89#endif 90 if (hasPhysMemPort) { 91 bool snoop = false; 92 AddrRangeList pmAddrList; 93 physmemPort.getPeerAddressRanges(pmAddrList, snoop); 94 physMemAddr = *pmAddrList.begin(); 95 } 96} 97 98bool 99AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt) 100{ 101 panic("AtomicSimpleCPU doesn't expect recvTiming callback!"); 102 return true; 103} 104 105Tick 106AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt) 107{ 108 //Snooping a coherence request, just return 109 return 0; 110} 111 112void 113AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt) 114{ 115 //No internal storage to update, just return 116 return; 117} 118 119void 120AtomicSimpleCPU::CpuPort::recvStatusChange(Status status) 121{ 122 if (status == RangeChange) { 123 if (!snoopRangeSent) { 124 snoopRangeSent = true; 125 sendStatusChange(Port::RangeChange); 126 } 127 return; 128 } 129 130 panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!"); 131} 132 133void 134AtomicSimpleCPU::CpuPort::recvRetry() 135{ 136 panic("AtomicSimpleCPU doesn't expect recvRetry callback!"); 137} 138 139void 140AtomicSimpleCPU::DcachePort::setPeer(Port *port) 141{ 142 Port::setPeer(port); 143 144#if FULL_SYSTEM 145 // Update the ThreadContext's memory ports (Functional/Virtual 146 // Ports) 147 cpu->tcBase()->connectMemPorts(); 148#endif 149} 150 151AtomicSimpleCPU::AtomicSimpleCPU(Params *p) 152 : BaseSimpleCPU(p), tickEvent(this), 153 width(p->width), simulate_stalls(p->simulate_stalls), 154 icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this), 155 physmemPort(name() + "-iport", this), hasPhysMemPort(false) 156{ 157 _status = Idle; 158 159 icachePort.snoopRangeSent = false; 160 dcachePort.snoopRangeSent = false; 161 162 ifetch_req.setThreadContext(cpuId, 0); // Add thread ID if we add MT 163 data_read_req.setThreadContext(cpuId, 0); // Add thread ID here too 164 data_write_req.setThreadContext(cpuId, 0); // Add thread ID here too 165} 166 167 168AtomicSimpleCPU::~AtomicSimpleCPU() 169{ 170} 171 172void 173AtomicSimpleCPU::serialize(ostream &os) 174{ 175 SimObject::State so_state = SimObject::getState(); 176 SERIALIZE_ENUM(so_state); 177 Status _status = status(); 178 SERIALIZE_ENUM(_status); 179 BaseSimpleCPU::serialize(os); 180 nameOut(os, csprintf("%s.tickEvent", name())); 181 tickEvent.serialize(os); 182} 183 184void 185AtomicSimpleCPU::unserialize(Checkpoint *cp, const string §ion) 186{ 187 SimObject::State so_state; 188 UNSERIALIZE_ENUM(so_state); 189 UNSERIALIZE_ENUM(_status); 190 BaseSimpleCPU::unserialize(cp, section); 191 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 192} 193 194void 195AtomicSimpleCPU::resume() 196{ 197 DPRINTF(SimpleCPU, "Resume\n"); 198 if (_status != SwitchedOut && _status != Idle) { 199 assert(system->getMemoryMode() == Enums::atomic); 200 201 changeState(SimObject::Running); 202 if (thread->status() == ThreadContext::Active) { 203 if (!tickEvent.scheduled()) { 204 tickEvent.schedule(nextCycle()); 205 } 206 } 207 } 208} 209 210void 211AtomicSimpleCPU::switchOut() 212{ 213 assert(status() == Running || status() == Idle); 214 _status = SwitchedOut; 215 216 tickEvent.squash(); 217} 218 219 220void 221AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU) 222{ 223 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort); 224 225 assert(!tickEvent.scheduled()); 226 227 // if any of this CPU's ThreadContexts are active, mark the CPU as 228 // running and schedule its tick event. 229 for (int i = 0; i < threadContexts.size(); ++i) { 230 ThreadContext *tc = threadContexts[i]; 231 if (tc->status() == ThreadContext::Active && _status != Running) { 232 _status = Running; 233 tickEvent.schedule(nextCycle()); 234 break; 235 } 236 } 237 if (_status != Running) { 238 _status = Idle; 239 } 240 assert(threadContexts.size() == 1); 241 cpuId = tc->readCpuId(); 242} 243 244 245void 246AtomicSimpleCPU::activateContext(int thread_num, int delay) 247{ 248 DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay); 249 250 assert(thread_num == 0); 251 assert(thread); 252 253 assert(_status == Idle); 254 assert(!tickEvent.scheduled()); 255 256 notIdleFraction++; 257 numCycles += tickToCycles(thread->lastActivate - thread->lastSuspend); 258 259 //Make sure ticks are still on multiples of cycles 260 tickEvent.schedule(nextCycle(curTick + ticks(delay))); 261 _status = Running; 262} 263 264 265void 266AtomicSimpleCPU::suspendContext(int thread_num) 267{ 268 DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num); 269 270 assert(thread_num == 0); 271 assert(thread); 272 273 assert(_status == Running); 274 275 // tick event may not be scheduled if this gets called from inside 276 // an instruction's execution, e.g. "quiesce" 277 if (tickEvent.scheduled()) 278 tickEvent.deschedule(); 279 280 notIdleFraction--; 281 _status = Idle; 282} 283 284 285template <class T> 286Fault 287AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) 288{ 289 // use the CPU's statically allocated read request and packet objects 290 Request *req = &data_read_req; 291 292 if (traceData) { 293 traceData->setAddr(addr); 294 } 295 296 //The block size of our peer. 297 int blockSize = dcachePort.peerBlockSize(); 298 //The size of the data we're trying to read. 299 int dataSize = sizeof(T); 300 301 uint8_t * dataPtr = (uint8_t *)&data; 302 303 //The address of the second part of this access if it needs to be split 304 //across a cache line boundary. 305 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize); 306 307 if(secondAddr > addr) 308 dataSize = secondAddr - addr; 309 310 dcache_latency = 0; 311 312 while(1) { 313 req->setVirt(0, addr, dataSize, flags, thread->readPC()); 314 315 // translate to physical address 316 Fault fault = thread->translateDataReadReq(req); 317 318 // Now do the access. 319 if (fault == NoFault) { 320 Packet pkt = Packet(req, 321 req->isLocked() ? MemCmd::LoadLockedReq : MemCmd::ReadReq, 322 Packet::Broadcast); 323 pkt.dataStatic(dataPtr); 324 325 if (req->isMmapedIpr()) 326 dcache_latency += TheISA::handleIprRead(thread->getTC(), &pkt); 327 else { 328 if (hasPhysMemPort && pkt.getAddr() == physMemAddr) 329 dcache_latency += physmemPort.sendAtomic(&pkt); 330 else 331 dcache_latency += dcachePort.sendAtomic(&pkt); 332 } 333 dcache_access = true; 334 335 assert(!pkt.isError()); 336 337 if (req->isLocked()) { 338 TheISA::handleLockedRead(thread, req); 339 } 340 } 341 342 // This will need a new way to tell if it has a dcache attached. 343 if (req->isUncacheable()) 344 recordEvent("Uncached Read"); 345 346 //If there's a fault, return it 347 if (fault != NoFault) 348 return fault; 349 //If we don't need to access a second cache line, stop now. 350 if (secondAddr <= addr) 351 { 352 data = gtoh(data); 353 return fault; 354 } 355 356 /* 357 * Set up for accessing the second cache line. 358 */ 359 360 //Move the pointer we're reading into to the correct location. 361 dataPtr += dataSize; 362 //Adjust the size to get the remaining bytes. 363 dataSize = addr + sizeof(T) - secondAddr; 364 //And access the right address. 365 addr = secondAddr; 366 } 367} 368 369#ifndef DOXYGEN_SHOULD_SKIP_THIS 370 371template 372Fault 373AtomicSimpleCPU::read(Addr addr, Twin32_t &data, unsigned flags); 374 375template 376Fault 377AtomicSimpleCPU::read(Addr addr, Twin64_t &data, unsigned flags); 378 379template 380Fault 381AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags); 382 383template 384Fault 385AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags); 386 387template 388Fault 389AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags); 390 391template 392Fault 393AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags); 394 395#endif //DOXYGEN_SHOULD_SKIP_THIS 396 397template<> 398Fault 399AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags) 400{ 401 return read(addr, *(uint64_t*)&data, flags); 402} 403 404template<> 405Fault 406AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags) 407{ 408 return read(addr, *(uint32_t*)&data, flags); 409} 410 411 412template<> 413Fault 414AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags) 415{ 416 return read(addr, (uint32_t&)data, flags); 417} 418 419 420template <class T> 421Fault 422AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) 423{ 424 // use the CPU's statically allocated write request and packet objects 425 Request *req = &data_write_req; 426 427 if (traceData) { 428 traceData->setAddr(addr); 429 } 430 431 //The block size of our peer. 432 int blockSize = dcachePort.peerBlockSize(); 433 //The size of the data we're trying to read. 434 int dataSize = sizeof(T); 435 436 uint8_t * dataPtr = (uint8_t *)&data; 437 438 //The address of the second part of this access if it needs to be split 439 //across a cache line boundary. 440 Addr secondAddr = roundDown(addr + dataSize - 1, blockSize); 441 442 if(secondAddr > addr) 443 dataSize = secondAddr - addr; 444 445 dcache_latency = 0; 446 447 while(1) { 448 req->setVirt(0, addr, dataSize, flags, thread->readPC()); 449 450 // translate to physical address 451 Fault fault = thread->translateDataWriteReq(req); 452 453 // Now do the access. 454 if (fault == NoFault) { 455 MemCmd cmd = MemCmd::WriteReq; // default 456 bool do_access = true; // flag to suppress cache access 457 458 if (req->isLocked()) { 459 cmd = MemCmd::StoreCondReq; 460 do_access = TheISA::handleLockedWrite(thread, req); 461 } else if (req->isSwap()) { 462 cmd = MemCmd::SwapReq; 463 if (req->isCondSwap()) { 464 assert(res); 465 req->setExtraData(*res); 466 } 467 } 468 469 if (do_access) { 470 Packet pkt = Packet(req, cmd, Packet::Broadcast); 471 pkt.dataStatic(dataPtr); 472 473 if (req->isMmapedIpr()) { 474 dcache_latency += 475 TheISA::handleIprWrite(thread->getTC(), &pkt); 476 } else { 477 //XXX This needs to be outside of the loop in order to 478 //work properly for cache line boundary crossing 479 //accesses in transendian simulations. 480 data = htog(data); 481 if (hasPhysMemPort && pkt.getAddr() == physMemAddr) 482 dcache_latency += physmemPort.sendAtomic(&pkt); 483 else 484 dcache_latency += dcachePort.sendAtomic(&pkt); 485 } 486 dcache_access = true; 487 assert(!pkt.isError()); 488 489 if (req->isSwap()) { 490 assert(res); 491 *res = pkt.get<T>(); 492 } 493 } 494 495 if (res && !req->isSwap()) { 496 *res = req->getExtraData(); 497 } 498 } 499 500 // This will need a new way to tell if it's hooked up to a cache or not. 501 if (req->isUncacheable()) 502 recordEvent("Uncached Write"); 503 504 //If there's a fault or we don't need to access a second cache line, 505 //stop now. 506 if (fault != NoFault || secondAddr <= addr) 507 { 508 // If the write needs to have a fault on the access, consider 509 // calling changeStatus() and changing it to "bad addr write" 510 // or something. 511 return fault; 512 } 513 514 /* 515 * Set up for accessing the second cache line. 516 */ 517 518 //Move the pointer we're reading into to the correct location. 519 dataPtr += dataSize; 520 //Adjust the size to get the remaining bytes. 521 dataSize = addr + sizeof(T) - secondAddr; 522 //And access the right address. 523 addr = secondAddr; 524 } 525} 526 527 528#ifndef DOXYGEN_SHOULD_SKIP_THIS 529 530template 531Fault 532AtomicSimpleCPU::write(Twin32_t data, Addr addr, 533 unsigned flags, uint64_t *res); 534 535template 536Fault 537AtomicSimpleCPU::write(Twin64_t data, Addr addr, 538 unsigned flags, uint64_t *res); 539 540template 541Fault 542AtomicSimpleCPU::write(uint64_t data, Addr addr, 543 unsigned flags, uint64_t *res); 544 545template 546Fault 547AtomicSimpleCPU::write(uint32_t data, Addr addr, 548 unsigned flags, uint64_t *res); 549 550template 551Fault 552AtomicSimpleCPU::write(uint16_t data, Addr addr, 553 unsigned flags, uint64_t *res); 554 555template 556Fault 557AtomicSimpleCPU::write(uint8_t data, Addr addr, 558 unsigned flags, uint64_t *res); 559 560#endif //DOXYGEN_SHOULD_SKIP_THIS 561 562template<> 563Fault 564AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res) 565{ 566 return write(*(uint64_t*)&data, addr, flags, res); 567} 568 569template<> 570Fault 571AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res) 572{ 573 return write(*(uint32_t*)&data, addr, flags, res); 574} 575 576 577template<> 578Fault 579AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res) 580{ 581 return write((uint32_t)data, addr, flags, res); 582} 583 584 585void 586AtomicSimpleCPU::tick() 587{ 588 DPRINTF(SimpleCPU, "Tick\n"); 589 590 Tick latency = ticks(1); // instruction takes one cycle by default 591 592 for (int i = 0; i < width; ++i) { 593 numCycles++; 594 595 if (!curStaticInst || !curStaticInst->isDelayedCommit()) 596 checkForInterrupts(); 597 598 Fault fault = setupFetchRequest(&ifetch_req); 599 600 if (fault == NoFault) { 601 Tick icache_latency = 0; 602 bool icache_access = false; 603 dcache_access = false; // assume no dcache access 604 605 //Fetch more instruction memory if necessary 606 //if(predecoder.needMoreBytes()) 607 //{ 608 icache_access = true; 609 Packet ifetch_pkt = Packet(&ifetch_req, MemCmd::ReadReq, 610 Packet::Broadcast); 611 ifetch_pkt.dataStatic(&inst); 612 613 if (hasPhysMemPort && ifetch_pkt.getAddr() == physMemAddr) 614 icache_latency = physmemPort.sendAtomic(&ifetch_pkt); 615 else 616 icache_latency = icachePort.sendAtomic(&ifetch_pkt); 617 618 assert(!ifetch_pkt.isError()); 619 620 // ifetch_req is initialized to read the instruction directly 621 // into the CPU object's inst field. 622 //} 623 624 preExecute(); 625 626 if (curStaticInst) { 627 fault = curStaticInst->execute(this, traceData); 628 629 // keep an instruction count 630 if (fault == NoFault) 631 countInst(); 632 else if (traceData) { 633 // If there was a fault, we should trace this instruction. 634 delete traceData; 635 traceData = NULL; 636 } 637 638 postExecute(); 639 } 640 641 // @todo remove me after debugging with legion done 642 if (curStaticInst && (!curStaticInst->isMicroop() || 643 curStaticInst->isFirstMicroop())) 644 instCnt++; 645 646 if (simulate_stalls) { 647 Tick icache_stall = 648 icache_access ? icache_latency - ticks(1) : 0; 649 Tick dcache_stall = 650 dcache_access ? dcache_latency - ticks(1) : 0; 651 Tick stall_cycles = (icache_stall + dcache_stall) / ticks(1); 652 if (ticks(stall_cycles) < (icache_stall + dcache_stall)) 653 latency += ticks(stall_cycles+1); 654 else 655 latency += ticks(stall_cycles); 656 } 657 658 } 659 if(fault != NoFault || !stayAtPC) 660 advancePC(fault); 661 } 662 663 if (_status != Idle) 664 tickEvent.schedule(curTick + latency); 665} 666 667 668//////////////////////////////////////////////////////////////////////// 669// 670// AtomicSimpleCPU Simulation Object 671// 672AtomicSimpleCPU * 673AtomicSimpleCPUParams::create() 674{ 675 AtomicSimpleCPU::Params *params = new AtomicSimpleCPU::Params(); 676 params->name = name; 677 params->numberOfThreads = 1; 678 params->max_insts_any_thread = max_insts_any_thread; 679 params->max_insts_all_threads = max_insts_all_threads; 680 params->max_loads_any_thread = max_loads_any_thread; 681 params->max_loads_all_threads = max_loads_all_threads; 682 params->progress_interval = progress_interval; 683 params->deferRegistration = defer_registration; 684 params->phase = phase; 685 params->clock = clock; 686 params->functionTrace = function_trace; 687 params->functionTraceStart = function_trace_start; 688 params->width = width; 689 params->simulate_stalls = simulate_stalls; 690 params->system = system; 691 params->cpu_id = cpu_id; 692 params->tracer = tracer; 693 694 params->itb = itb; 695 params->dtb = dtb; 696#if FULL_SYSTEM 697 params->profile = profile; 698 params->do_quiesce = do_quiesce; 699 params->do_checkpoint_insts = do_checkpoint_insts; 700 params->do_statistics_insts = do_statistics_insts; 701#else 702 if (workload.size() != 1) 703 panic("only one workload allowed"); 704 params->process = workload[0]; 705#endif 706 707 AtomicSimpleCPU *cpu = new AtomicSimpleCPU(params); 708 return cpu; 709} 710