lsq_unit_impl.hh revision 3326
1/* 2 * Copyright (c) 2004-2005 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: Kevin Lim 29 * Korey Sewell 30 */ 31 32#include "arch/locked_mem.hh" 33#include "config/use_checker.hh" 34 35#include "cpu/o3/lsq.hh" 36#include "cpu/o3/lsq_unit.hh" 37#include "base/str.hh" 38#include "mem/packet.hh" 39#include "mem/request.hh" 40 41#if USE_CHECKER 42#include "cpu/checker/cpu.hh" 43#endif 44 45template<class Impl> 46LSQUnit<Impl>::WritebackEvent::WritebackEvent(DynInstPtr &_inst, PacketPtr _pkt, 47 LSQUnit *lsq_ptr) 48 : Event(&mainEventQueue), inst(_inst), pkt(_pkt), lsqPtr(lsq_ptr) 49{ 50 this->setFlags(Event::AutoDelete); 51} 52 53template<class Impl> 54void 55LSQUnit<Impl>::WritebackEvent::process() 56{ 57 if (!lsqPtr->isSwitchedOut()) { 58 lsqPtr->writeback(inst, pkt); 59 } 60 delete pkt; 61} 62 63template<class Impl> 64const char * 65LSQUnit<Impl>::WritebackEvent::description() 66{ 67 return "Store writeback event"; 68} 69 70template<class Impl> 71void 72LSQUnit<Impl>::completeDataAccess(PacketPtr pkt) 73{ 74 LSQSenderState *state = dynamic_cast<LSQSenderState *>(pkt->senderState); 75 DynInstPtr inst = state->inst; 76 DPRINTF(IEW, "Writeback event [sn:%lli]\n", inst->seqNum); 77 DPRINTF(Activity, "Activity: Writeback event [sn:%lli]\n", inst->seqNum); 78 79 //iewStage->ldstQueue.removeMSHR(inst->threadNumber,inst->seqNum); 80 81 if (isSwitchedOut() || inst->isSquashed()) { 82 iewStage->decrWb(inst->seqNum); 83 delete state; 84 delete pkt; 85 return; 86 } else { 87 if (!state->noWB) { 88 writeback(inst, pkt); 89 } 90 91 if (inst->isStore()) { 92 completeStore(state->idx); 93 } 94 } 95 96 delete state; 97 delete pkt; 98} 99 100template <class Impl> 101LSQUnit<Impl>::LSQUnit() 102 : loads(0), stores(0), storesToWB(0), stalled(false), 103 isStoreBlocked(false), isLoadBlocked(false), 104 loadBlockedHandled(false) 105{ 106} 107 108template<class Impl> 109void 110LSQUnit<Impl>::init(Params *params, LSQ *lsq_ptr, unsigned maxLQEntries, 111 unsigned maxSQEntries, unsigned id) 112{ 113 DPRINTF(LSQUnit, "Creating LSQUnit%i object.\n",id); 114 115 switchedOut = false; 116 117 lsq = lsq_ptr; 118 119 lsqID = id; 120 121 // Add 1 for the sentinel entry (they are circular queues). 122 LQEntries = maxLQEntries + 1; 123 SQEntries = maxSQEntries + 1; 124 125 loadQueue.resize(LQEntries); 126 storeQueue.resize(SQEntries); 127 128 loadHead = loadTail = 0; 129 130 storeHead = storeWBIdx = storeTail = 0; 131 132 usedPorts = 0; 133 cachePorts = params->cachePorts; 134 135 memDepViolator = NULL; 136 137 blockedLoadSeqNum = 0; 138} 139 140template<class Impl> 141void 142LSQUnit<Impl>::setCPU(O3CPU *cpu_ptr) 143{ 144 cpu = cpu_ptr; 145 146#if USE_CHECKER 147 if (cpu->checker) { 148 cpu->checker->setDcachePort(dcachePort); 149 } 150#endif 151} 152 153template<class Impl> 154std::string 155LSQUnit<Impl>::name() const 156{ 157 if (Impl::MaxThreads == 1) { 158 return iewStage->name() + ".lsq"; 159 } else { 160 return iewStage->name() + ".lsq.thread." + to_string(lsqID); 161 } 162} 163 164template<class Impl> 165void 166LSQUnit<Impl>::regStats() 167{ 168 lsqForwLoads 169 .name(name() + ".forwLoads") 170 .desc("Number of loads that had data forwarded from stores"); 171 172 invAddrLoads 173 .name(name() + ".invAddrLoads") 174 .desc("Number of loads ignored due to an invalid address"); 175 176 lsqSquashedLoads 177 .name(name() + ".squashedLoads") 178 .desc("Number of loads squashed"); 179 180 lsqIgnoredResponses 181 .name(name() + ".ignoredResponses") 182 .desc("Number of memory responses ignored because the instruction is squashed"); 183 184 lsqMemOrderViolation 185 .name(name() + ".memOrderViolation") 186 .desc("Number of memory ordering violations"); 187 188 lsqSquashedStores 189 .name(name() + ".squashedStores") 190 .desc("Number of stores squashed"); 191 192 invAddrSwpfs 193 .name(name() + ".invAddrSwpfs") 194 .desc("Number of software prefetches ignored due to an invalid address"); 195 196 lsqBlockedLoads 197 .name(name() + ".blockedLoads") 198 .desc("Number of blocked loads due to partial load-store forwarding"); 199 200 lsqRescheduledLoads 201 .name(name() + ".rescheduledLoads") 202 .desc("Number of loads that were rescheduled"); 203 204 lsqCacheBlocked 205 .name(name() + ".cacheBlocked") 206 .desc("Number of times an access to memory failed due to the cache being blocked"); 207} 208 209template<class Impl> 210void 211LSQUnit<Impl>::clearLQ() 212{ 213 loadQueue.clear(); 214} 215 216template<class Impl> 217void 218LSQUnit<Impl>::clearSQ() 219{ 220 storeQueue.clear(); 221} 222 223template<class Impl> 224void 225LSQUnit<Impl>::switchOut() 226{ 227 switchedOut = true; 228 for (int i = 0; i < loadQueue.size(); ++i) { 229 assert(!loadQueue[i]); 230 loadQueue[i] = NULL; 231 } 232 233 assert(storesToWB == 0); 234} 235 236template<class Impl> 237void 238LSQUnit<Impl>::takeOverFrom() 239{ 240 switchedOut = false; 241 loads = stores = storesToWB = 0; 242 243 loadHead = loadTail = 0; 244 245 storeHead = storeWBIdx = storeTail = 0; 246 247 usedPorts = 0; 248 249 memDepViolator = NULL; 250 251 blockedLoadSeqNum = 0; 252 253 stalled = false; 254 isLoadBlocked = false; 255 loadBlockedHandled = false; 256} 257 258template<class Impl> 259void 260LSQUnit<Impl>::resizeLQ(unsigned size) 261{ 262 unsigned size_plus_sentinel = size + 1; 263 assert(size_plus_sentinel >= LQEntries); 264 265 if (size_plus_sentinel > LQEntries) { 266 while (size_plus_sentinel > loadQueue.size()) { 267 DynInstPtr dummy; 268 loadQueue.push_back(dummy); 269 LQEntries++; 270 } 271 } else { 272 LQEntries = size_plus_sentinel; 273 } 274 275} 276 277template<class Impl> 278void 279LSQUnit<Impl>::resizeSQ(unsigned size) 280{ 281 unsigned size_plus_sentinel = size + 1; 282 if (size_plus_sentinel > SQEntries) { 283 while (size_plus_sentinel > storeQueue.size()) { 284 SQEntry dummy; 285 storeQueue.push_back(dummy); 286 SQEntries++; 287 } 288 } else { 289 SQEntries = size_plus_sentinel; 290 } 291} 292 293template <class Impl> 294void 295LSQUnit<Impl>::insert(DynInstPtr &inst) 296{ 297 assert(inst->isMemRef()); 298 299 assert(inst->isLoad() || inst->isStore()); 300 301 if (inst->isLoad()) { 302 insertLoad(inst); 303 } else { 304 insertStore(inst); 305 } 306 307 inst->setInLSQ(); 308} 309 310template <class Impl> 311void 312LSQUnit<Impl>::insertLoad(DynInstPtr &load_inst) 313{ 314 assert((loadTail + 1) % LQEntries != loadHead); 315 assert(loads < LQEntries); 316 317 DPRINTF(LSQUnit, "Inserting load PC %#x, idx:%i [sn:%lli]\n", 318 load_inst->readPC(), loadTail, load_inst->seqNum); 319 320 load_inst->lqIdx = loadTail; 321 322 if (stores == 0) { 323 load_inst->sqIdx = -1; 324 } else { 325 load_inst->sqIdx = storeTail; 326 } 327 328 loadQueue[loadTail] = load_inst; 329 330 incrLdIdx(loadTail); 331 332 ++loads; 333} 334 335template <class Impl> 336void 337LSQUnit<Impl>::insertStore(DynInstPtr &store_inst) 338{ 339 // Make sure it is not full before inserting an instruction. 340 assert((storeTail + 1) % SQEntries != storeHead); 341 assert(stores < SQEntries); 342 343 DPRINTF(LSQUnit, "Inserting store PC %#x, idx:%i [sn:%lli]\n", 344 store_inst->readPC(), storeTail, store_inst->seqNum); 345 346 store_inst->sqIdx = storeTail; 347 store_inst->lqIdx = loadTail; 348 349 storeQueue[storeTail] = SQEntry(store_inst); 350 351 incrStIdx(storeTail); 352 353 ++stores; 354} 355 356template <class Impl> 357typename Impl::DynInstPtr 358LSQUnit<Impl>::getMemDepViolator() 359{ 360 DynInstPtr temp = memDepViolator; 361 362 memDepViolator = NULL; 363 364 return temp; 365} 366 367template <class Impl> 368unsigned 369LSQUnit<Impl>::numFreeEntries() 370{ 371 unsigned free_lq_entries = LQEntries - loads; 372 unsigned free_sq_entries = SQEntries - stores; 373 374 // Both the LQ and SQ entries have an extra dummy entry to differentiate 375 // empty/full conditions. Subtract 1 from the free entries. 376 if (free_lq_entries < free_sq_entries) { 377 return free_lq_entries - 1; 378 } else { 379 return free_sq_entries - 1; 380 } 381} 382 383template <class Impl> 384int 385LSQUnit<Impl>::numLoadsReady() 386{ 387 int load_idx = loadHead; 388 int retval = 0; 389 390 while (load_idx != loadTail) { 391 assert(loadQueue[load_idx]); 392 393 if (loadQueue[load_idx]->readyToIssue()) { 394 ++retval; 395 } 396 } 397 398 return retval; 399} 400 401template <class Impl> 402Fault 403LSQUnit<Impl>::executeLoad(DynInstPtr &inst) 404{ 405 // Execute a specific load. 406 Fault load_fault = NoFault; 407 408 DPRINTF(LSQUnit, "Executing load PC %#x, [sn:%lli]\n", 409 inst->readPC(),inst->seqNum); 410 411 load_fault = inst->initiateAcc(); 412 413 // If the instruction faulted, then we need to send it along to commit 414 // without the instruction completing. 415 if (load_fault != NoFault) { 416 // Send this instruction to commit, also make sure iew stage 417 // realizes there is activity. 418 // Mark it as executed unless it is an uncached load that 419 // needs to hit the head of commit. 420 if (!(inst->req->isUncacheable()) || inst->isAtCommit()) { 421 inst->setExecuted(); 422 } 423 iewStage->instToCommit(inst); 424 iewStage->activityThisCycle(); 425 } 426 427 return load_fault; 428} 429 430template <class Impl> 431Fault 432LSQUnit<Impl>::executeStore(DynInstPtr &store_inst) 433{ 434 using namespace TheISA; 435 // Make sure that a store exists. 436 assert(stores != 0); 437 438 int store_idx = store_inst->sqIdx; 439 440 DPRINTF(LSQUnit, "Executing store PC %#x [sn:%lli]\n", 441 store_inst->readPC(), store_inst->seqNum); 442 443 // Check the recently completed loads to see if any match this store's 444 // address. If so, then we have a memory ordering violation. 445 int load_idx = store_inst->lqIdx; 446 447 Fault store_fault = store_inst->initiateAcc(); 448 449 if (storeQueue[store_idx].size == 0) { 450 DPRINTF(LSQUnit,"Fault on Store PC %#x, [sn:%lli],Size = 0\n", 451 store_inst->readPC(),store_inst->seqNum); 452 453 return store_fault; 454 } 455 456 assert(store_fault == NoFault); 457 458 if (store_inst->isStoreConditional()) { 459 // Store conditionals need to set themselves as able to 460 // writeback if we haven't had a fault by here. 461 storeQueue[store_idx].canWB = true; 462 463 ++storesToWB; 464 } 465 466 if (!memDepViolator) { 467 while (load_idx != loadTail) { 468 // Really only need to check loads that have actually executed 469 // It's safe to check all loads because effAddr is set to 470 // InvalAddr when the dyn inst is created. 471 472 // @todo: For now this is extra conservative, detecting a 473 // violation if the addresses match assuming all accesses 474 // are quad word accesses. 475 476 // @todo: Fix this, magic number being used here 477 if ((loadQueue[load_idx]->effAddr >> 8) == 478 (store_inst->effAddr >> 8)) { 479 // A load incorrectly passed this store. Squash and refetch. 480 // For now return a fault to show that it was unsuccessful. 481 memDepViolator = loadQueue[load_idx]; 482 ++lsqMemOrderViolation; 483 484 return genMachineCheckFault(); 485 } 486 487 incrLdIdx(load_idx); 488 } 489 490 // If we've reached this point, there was no violation. 491 memDepViolator = NULL; 492 } 493 494 return store_fault; 495} 496 497template <class Impl> 498void 499LSQUnit<Impl>::commitLoad() 500{ 501 assert(loadQueue[loadHead]); 502 503 DPRINTF(LSQUnit, "Committing head load instruction, PC %#x\n", 504 loadQueue[loadHead]->readPC()); 505 506 loadQueue[loadHead] = NULL; 507 508 incrLdIdx(loadHead); 509 510 --loads; 511} 512 513template <class Impl> 514void 515LSQUnit<Impl>::commitLoads(InstSeqNum &youngest_inst) 516{ 517 assert(loads == 0 || loadQueue[loadHead]); 518 519 while (loads != 0 && loadQueue[loadHead]->seqNum <= youngest_inst) { 520 commitLoad(); 521 } 522} 523 524template <class Impl> 525void 526LSQUnit<Impl>::commitStores(InstSeqNum &youngest_inst) 527{ 528 assert(stores == 0 || storeQueue[storeHead].inst); 529 530 int store_idx = storeHead; 531 532 while (store_idx != storeTail) { 533 assert(storeQueue[store_idx].inst); 534 // Mark any stores that are now committed and have not yet 535 // been marked as able to write back. 536 if (!storeQueue[store_idx].canWB) { 537 if (storeQueue[store_idx].inst->seqNum > youngest_inst) { 538 break; 539 } 540 DPRINTF(LSQUnit, "Marking store as able to write back, PC " 541 "%#x [sn:%lli]\n", 542 storeQueue[store_idx].inst->readPC(), 543 storeQueue[store_idx].inst->seqNum); 544 545 storeQueue[store_idx].canWB = true; 546 547 ++storesToWB; 548 } 549 550 incrStIdx(store_idx); 551 } 552} 553 554template <class Impl> 555void 556LSQUnit<Impl>::writebackStores() 557{ 558 while (storesToWB > 0 && 559 storeWBIdx != storeTail && 560 storeQueue[storeWBIdx].inst && 561 storeQueue[storeWBIdx].canWB && 562 usedPorts < cachePorts) { 563 564 if (isStoreBlocked || lsq->cacheBlocked()) { 565 DPRINTF(LSQUnit, "Unable to write back any more stores, cache" 566 " is blocked!\n"); 567 break; 568 } 569 570 // Store didn't write any data so no need to write it back to 571 // memory. 572 if (storeQueue[storeWBIdx].size == 0) { 573 completeStore(storeWBIdx); 574 575 incrStIdx(storeWBIdx); 576 577 continue; 578 } 579 580 ++usedPorts; 581 582 if (storeQueue[storeWBIdx].inst->isDataPrefetch()) { 583 incrStIdx(storeWBIdx); 584 585 continue; 586 } 587 588 assert(storeQueue[storeWBIdx].req); 589 assert(!storeQueue[storeWBIdx].committed); 590 591 DynInstPtr inst = storeQueue[storeWBIdx].inst; 592 593 Request *req = storeQueue[storeWBIdx].req; 594 storeQueue[storeWBIdx].committed = true; 595 596 assert(!inst->memData); 597 inst->memData = new uint8_t[64]; 598 memcpy(inst->memData, (uint8_t *)&storeQueue[storeWBIdx].data, 599 req->getSize()); 600 601 PacketPtr data_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); 602 data_pkt->dataStatic(inst->memData); 603 604 LSQSenderState *state = new LSQSenderState; 605 state->isLoad = false; 606 state->idx = storeWBIdx; 607 state->inst = inst; 608 data_pkt->senderState = state; 609 610 DPRINTF(LSQUnit, "D-Cache: Writing back store idx:%i PC:%#x " 611 "to Addr:%#x, data:%#x [sn:%lli]\n", 612 storeWBIdx, inst->readPC(), 613 req->getPaddr(), *(inst->memData), 614 inst->seqNum); 615 616 // @todo: Remove this SC hack once the memory system handles it. 617 if (req->isLocked()) { 618 // Disable recording the result temporarily. Writing to 619 // misc regs normally updates the result, but this is not 620 // the desired behavior when handling store conditionals. 621 inst->recordResult = false; 622 bool success = TheISA::handleLockedWrite(inst.get(), req); 623 inst->recordResult = true; 624 625 if (!success) { 626 // Instantly complete this store. 627 DPRINTF(LSQUnit, "Store conditional [sn:%lli] failed. " 628 "Instantly completing it.\n", 629 inst->seqNum); 630 WritebackEvent *wb = new WritebackEvent(inst, data_pkt, this); 631 wb->schedule(curTick + 1); 632 delete state; 633 completeStore(storeWBIdx); 634 incrStIdx(storeWBIdx); 635 continue; 636 } 637 } else { 638 // Non-store conditionals do not need a writeback. 639 state->noWB = true; 640 } 641 642 if (!dcachePort->sendTiming(data_pkt)) { 643 if (data_pkt->result == Packet::BadAddress) { 644 panic("LSQ sent out a bad address for a completed store!"); 645 } 646 // Need to handle becoming blocked on a store. 647 DPRINTF(IEW, "D-Cache became blcoked when writing [sn:%lli], will" 648 "retry later\n", 649 inst->seqNum); 650 isStoreBlocked = true; 651 ++lsqCacheBlocked; 652 assert(retryPkt == NULL); 653 retryPkt = data_pkt; 654 lsq->setRetryTid(lsqID); 655 } else { 656 storePostSend(data_pkt); 657 } 658 } 659 660 // Not sure this should set it to 0. 661 usedPorts = 0; 662 663 assert(stores >= 0 && storesToWB >= 0); 664} 665 666/*template <class Impl> 667void 668LSQUnit<Impl>::removeMSHR(InstSeqNum seqNum) 669{ 670 list<InstSeqNum>::iterator mshr_it = find(mshrSeqNums.begin(), 671 mshrSeqNums.end(), 672 seqNum); 673 674 if (mshr_it != mshrSeqNums.end()) { 675 mshrSeqNums.erase(mshr_it); 676 DPRINTF(LSQUnit, "Removing MSHR. count = %i\n",mshrSeqNums.size()); 677 } 678}*/ 679 680template <class Impl> 681void 682LSQUnit<Impl>::squash(const InstSeqNum &squashed_num) 683{ 684 DPRINTF(LSQUnit, "Squashing until [sn:%lli]!" 685 "(Loads:%i Stores:%i)\n", squashed_num, loads, stores); 686 687 int load_idx = loadTail; 688 decrLdIdx(load_idx); 689 690 while (loads != 0 && loadQueue[load_idx]->seqNum > squashed_num) { 691 DPRINTF(LSQUnit,"Load Instruction PC %#x squashed, " 692 "[sn:%lli]\n", 693 loadQueue[load_idx]->readPC(), 694 loadQueue[load_idx]->seqNum); 695 696 if (isStalled() && load_idx == stallingLoadIdx) { 697 stalled = false; 698 stallingStoreIsn = 0; 699 stallingLoadIdx = 0; 700 } 701 702 // Clear the smart pointer to make sure it is decremented. 703 loadQueue[load_idx]->setSquashed(); 704 loadQueue[load_idx] = NULL; 705 --loads; 706 707 // Inefficient! 708 loadTail = load_idx; 709 710 decrLdIdx(load_idx); 711 ++lsqSquashedLoads; 712 } 713 714 if (isLoadBlocked) { 715 if (squashed_num < blockedLoadSeqNum) { 716 isLoadBlocked = false; 717 loadBlockedHandled = false; 718 blockedLoadSeqNum = 0; 719 } 720 } 721 722 int store_idx = storeTail; 723 decrStIdx(store_idx); 724 725 while (stores != 0 && 726 storeQueue[store_idx].inst->seqNum > squashed_num) { 727 // Instructions marked as can WB are already committed. 728 if (storeQueue[store_idx].canWB) { 729 break; 730 } 731 732 DPRINTF(LSQUnit,"Store Instruction PC %#x squashed, " 733 "idx:%i [sn:%lli]\n", 734 storeQueue[store_idx].inst->readPC(), 735 store_idx, storeQueue[store_idx].inst->seqNum); 736 737 // I don't think this can happen. It should have been cleared 738 // by the stalling load. 739 if (isStalled() && 740 storeQueue[store_idx].inst->seqNum == stallingStoreIsn) { 741 panic("Is stalled should have been cleared by stalling load!\n"); 742 stalled = false; 743 stallingStoreIsn = 0; 744 } 745 746 // Clear the smart pointer to make sure it is decremented. 747 storeQueue[store_idx].inst->setSquashed(); 748 storeQueue[store_idx].inst = NULL; 749 storeQueue[store_idx].canWB = 0; 750 751 storeQueue[store_idx].req = NULL; 752 --stores; 753 754 // Inefficient! 755 storeTail = store_idx; 756 757 decrStIdx(store_idx); 758 ++lsqSquashedStores; 759 } 760} 761 762template <class Impl> 763void 764LSQUnit<Impl>::storePostSend(Packet *pkt) 765{ 766 if (isStalled() && 767 storeQueue[storeWBIdx].inst->seqNum == stallingStoreIsn) { 768 DPRINTF(LSQUnit, "Unstalling, stalling store [sn:%lli] " 769 "load idx:%i\n", 770 stallingStoreIsn, stallingLoadIdx); 771 stalled = false; 772 stallingStoreIsn = 0; 773 iewStage->replayMemInst(loadQueue[stallingLoadIdx]); 774 } 775 776 if (!storeQueue[storeWBIdx].inst->isStoreConditional()) { 777 // The store is basically completed at this time. This 778 // only works so long as the checker doesn't try to 779 // verify the value in memory for stores. 780 storeQueue[storeWBIdx].inst->setCompleted(); 781#if USE_CHECKER 782 if (cpu->checker) { 783 cpu->checker->verify(storeQueue[storeWBIdx].inst); 784 } 785#endif 786 } 787 788 if (pkt->result != Packet::Success) { 789 DPRINTF(LSQUnit,"D-Cache Write Miss on idx:%i!\n", 790 storeWBIdx); 791 792 DPRINTF(Activity, "Active st accessing mem miss [sn:%lli]\n", 793 storeQueue[storeWBIdx].inst->seqNum); 794 795 //mshrSeqNums.push_back(storeQueue[storeWBIdx].inst->seqNum); 796 797 //DPRINTF(LSQUnit, "Added MSHR. count = %i\n",mshrSeqNums.size()); 798 799 // @todo: Increment stat here. 800 } else { 801 DPRINTF(LSQUnit,"D-Cache: Write Hit on idx:%i !\n", 802 storeWBIdx); 803 804 DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n", 805 storeQueue[storeWBIdx].inst->seqNum); 806 } 807 808 incrStIdx(storeWBIdx); 809} 810 811template <class Impl> 812void 813LSQUnit<Impl>::writeback(DynInstPtr &inst, PacketPtr pkt) 814{ 815 iewStage->wakeCPU(); 816 817 // Squashed instructions do not need to complete their access. 818 if (inst->isSquashed()) { 819 iewStage->decrWb(inst->seqNum); 820 assert(!inst->isStore()); 821 ++lsqIgnoredResponses; 822 return; 823 } 824 825 if (!inst->isExecuted()) { 826 inst->setExecuted(); 827 828 // Complete access to copy data to proper place. 829 inst->completeAcc(pkt); 830 } 831 832 // Need to insert instruction into queue to commit 833 iewStage->instToCommit(inst); 834 835 iewStage->activityThisCycle(); 836} 837 838template <class Impl> 839void 840LSQUnit<Impl>::completeStore(int store_idx) 841{ 842 assert(storeQueue[store_idx].inst); 843 storeQueue[store_idx].completed = true; 844 --storesToWB; 845 // A bit conservative because a store completion may not free up entries, 846 // but hopefully avoids two store completions in one cycle from making 847 // the CPU tick twice. 848 cpu->wakeCPU(); 849 cpu->activityThisCycle(); 850 851 if (store_idx == storeHead) { 852 do { 853 incrStIdx(storeHead); 854 855 --stores; 856 } while (storeQueue[storeHead].completed && 857 storeHead != storeTail); 858 859 iewStage->updateLSQNextCycle = true; 860 } 861 862 DPRINTF(LSQUnit, "Completing store [sn:%lli], idx:%i, store head " 863 "idx:%i\n", 864 storeQueue[store_idx].inst->seqNum, store_idx, storeHead); 865 866 if (isStalled() && 867 storeQueue[store_idx].inst->seqNum == stallingStoreIsn) { 868 DPRINTF(LSQUnit, "Unstalling, stalling store [sn:%lli] " 869 "load idx:%i\n", 870 stallingStoreIsn, stallingLoadIdx); 871 stalled = false; 872 stallingStoreIsn = 0; 873 iewStage->replayMemInst(loadQueue[stallingLoadIdx]); 874 } 875 876 storeQueue[store_idx].inst->setCompleted(); 877 878 // Tell the checker we've completed this instruction. Some stores 879 // may get reported twice to the checker, but the checker can 880 // handle that case. 881#if USE_CHECKER 882 if (cpu->checker) { 883 cpu->checker->verify(storeQueue[store_idx].inst); 884 } 885#endif 886} 887 888template <class Impl> 889void 890LSQUnit<Impl>::recvRetry() 891{ 892 if (isStoreBlocked) { 893 assert(retryPkt != NULL); 894 895 if (dcachePort->sendTiming(retryPkt)) { 896 if (retryPkt->result == Packet::BadAddress) { 897 panic("LSQ sent out a bad address for a completed store!"); 898 } 899 storePostSend(retryPkt); 900 retryPkt = NULL; 901 isStoreBlocked = false; 902 lsq->setRetryTid(-1); 903 } else { 904 // Still blocked! 905 ++lsqCacheBlocked; 906 lsq->setRetryTid(lsqID); 907 } 908 } else if (isLoadBlocked) { 909 DPRINTF(LSQUnit, "Loads squash themselves and all younger insts, " 910 "no need to resend packet.\n"); 911 } else { 912 DPRINTF(LSQUnit, "Retry received but LSQ is no longer blocked.\n"); 913 } 914} 915 916template <class Impl> 917inline void 918LSQUnit<Impl>::incrStIdx(int &store_idx) 919{ 920 if (++store_idx >= SQEntries) 921 store_idx = 0; 922} 923 924template <class Impl> 925inline void 926LSQUnit<Impl>::decrStIdx(int &store_idx) 927{ 928 if (--store_idx < 0) 929 store_idx += SQEntries; 930} 931 932template <class Impl> 933inline void 934LSQUnit<Impl>::incrLdIdx(int &load_idx) 935{ 936 if (++load_idx >= LQEntries) 937 load_idx = 0; 938} 939 940template <class Impl> 941inline void 942LSQUnit<Impl>::decrLdIdx(int &load_idx) 943{ 944 if (--load_idx < 0) 945 load_idx += LQEntries; 946} 947 948template <class Impl> 949void 950LSQUnit<Impl>::dumpInsts() 951{ 952 cprintf("Load store queue: Dumping instructions.\n"); 953 cprintf("Load queue size: %i\n", loads); 954 cprintf("Load queue: "); 955 956 int load_idx = loadHead; 957 958 while (load_idx != loadTail && loadQueue[load_idx]) { 959 cprintf("%#x ", loadQueue[load_idx]->readPC()); 960 961 incrLdIdx(load_idx); 962 } 963 964 cprintf("Store queue size: %i\n", stores); 965 cprintf("Store queue: "); 966 967 int store_idx = storeHead; 968 969 while (store_idx != storeTail && storeQueue[store_idx].inst) { 970 cprintf("%#x ", storeQueue[store_idx].inst->readPC()); 971 972 incrStIdx(store_idx); 973 } 974 975 cprintf("\n"); 976} 977