memtest.cc revision 4579
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: Erik Hallnor 29 * Steve Reinhardt 30 */ 31 32// FIX ME: make trackBlkAddr use blocksize from actual cache, not hard coded 33 34#include <iomanip> 35#include <set> 36#include <string> 37#include <vector> 38 39#include "base/misc.hh" 40#include "base/statistics.hh" 41#include "cpu/memtest/memtest.hh" 42//#include "cpu/simple_thread.hh" 43//#include "mem/cache/base_cache.hh" 44#include "mem/mem_object.hh" 45#include "mem/port.hh" 46#include "mem/packet.hh" 47//#include "mem/physical.hh" 48#include "mem/request.hh" 49#include "sim/builder.hh" 50#include "sim/sim_events.hh" 51#include "sim/stats.hh" 52 53using namespace std; 54 55int TESTER_ALLOCATOR=0; 56 57bool 58MemTest::CpuPort::recvTiming(PacketPtr pkt) 59{ 60 memtest->completeRequest(pkt); 61 return true; 62} 63 64Tick 65MemTest::CpuPort::recvAtomic(PacketPtr pkt) 66{ 67 panic("MemTest doesn't expect recvAtomic callback!"); 68 return curTick; 69} 70 71void 72MemTest::CpuPort::recvFunctional(PacketPtr pkt) 73{ 74 //Do nothing if we see one come through 75// if (curTick != 0)//Supress warning durring initialization 76// warn("Functional Writes not implemented in MemTester\n"); 77 //Need to find any response values that intersect and update 78 return; 79} 80 81void 82MemTest::CpuPort::recvStatusChange(Status status) 83{ 84 if (status == RangeChange) { 85 if (!snoopRangeSent) { 86 snoopRangeSent = true; 87 sendStatusChange(Port::RangeChange); 88 } 89 return; 90 } 91 92 panic("MemTest doesn't expect recvStatusChange callback!"); 93} 94 95void 96MemTest::CpuPort::recvRetry() 97{ 98 memtest->doRetry(); 99} 100 101void 102MemTest::sendPkt(PacketPtr pkt) { 103 if (atomic) { 104 cachePort.sendAtomic(pkt); 105 pkt->makeAtomicResponse(); 106 completeRequest(pkt); 107 } 108 else if (!cachePort.sendTiming(pkt)) { 109 accessRetry = true; 110 retryPkt = pkt; 111 } 112 113} 114 115MemTest::MemTest(const string &name, 116// MemInterface *_cache_interface, 117// PhysicalMemory *main_mem, 118// PhysicalMemory *check_mem, 119 unsigned _memorySize, 120 unsigned _percentReads, 121 unsigned _percentFunctional, 122 unsigned _percentUncacheable, 123 unsigned _progressInterval, 124 unsigned _percentSourceUnaligned, 125 unsigned _percentDestUnaligned, 126 Addr _traceAddr, 127 Counter _max_loads, 128 bool _atomic) 129 : MemObject(name), 130 tickEvent(this), 131 cachePort("test", this), 132 funcPort("functional", this), 133 retryPkt(NULL), 134// mainMem(main_mem), 135// checkMem(check_mem), 136 size(_memorySize), 137 percentReads(_percentReads), 138 percentFunctional(_percentFunctional), 139 percentUncacheable(_percentUncacheable), 140 progressInterval(_progressInterval), 141 nextProgressMessage(_progressInterval), 142 percentSourceUnaligned(_percentSourceUnaligned), 143 percentDestUnaligned(percentDestUnaligned), 144 maxLoads(_max_loads), 145 atomic(_atomic) 146{ 147 vector<string> cmd; 148 cmd.push_back("/bin/ls"); 149 vector<string> null_vec; 150 // thread = new SimpleThread(NULL, 0, NULL, 0, mainMem); 151 curTick = 0; 152 153 cachePort.snoopRangeSent = false; 154 funcPort.snoopRangeSent = true; 155 156 // Needs to be masked off once we know the block size. 157 traceBlockAddr = _traceAddr; 158 baseAddr1 = 0x100000; 159 baseAddr2 = 0x400000; 160 uncacheAddr = 0x800000; 161 162 // set up counters 163 noResponseCycles = 0; 164 numReads = 0; 165 tickEvent.schedule(0); 166 167 id = TESTER_ALLOCATOR++; 168 if (TESTER_ALLOCATOR > 8) 169 panic("False sharing memtester only allows up to 8 testers"); 170 171 accessRetry = false; 172} 173 174Port * 175MemTest::getPort(const std::string &if_name, int idx) 176{ 177 if (if_name == "functional") 178 return &funcPort; 179 else if (if_name == "test") 180 return &cachePort; 181 else 182 panic("No Such Port\n"); 183} 184 185void 186MemTest::init() 187{ 188 // By the time init() is called, the ports should be hooked up. 189 blockSize = cachePort.peerBlockSize(); 190 blockAddrMask = blockSize - 1; 191 traceBlockAddr = blockAddr(traceBlockAddr); 192 193 // initial memory contents for both physical memory and functional 194 // memory should be 0; no need to initialize them. 195} 196 197static void 198printData(ostream &os, uint8_t *data, int nbytes) 199{ 200 os << hex << setfill('0'); 201 // assume little-endian: print bytes from highest address to lowest 202 for (uint8_t *dp = data + nbytes - 1; dp >= data; --dp) { 203 os << setw(2) << (unsigned)*dp; 204 } 205 os << dec; 206} 207 208void 209MemTest::completeRequest(PacketPtr pkt) 210{ 211 MemTestSenderState *state = 212 dynamic_cast<MemTestSenderState *>(pkt->senderState); 213 214 uint8_t *data = state->data; 215 uint8_t *pkt_data = pkt->getPtr<uint8_t>(); 216 Request *req = pkt->req; 217 218 //Remove the address from the list of outstanding 219 std::set<unsigned>::iterator removeAddr = outstandingAddrs.find(req->getPaddr()); 220 assert(removeAddr != outstandingAddrs.end()); 221 outstandingAddrs.erase(removeAddr); 222 223 switch (pkt->cmd.toInt()) { 224 case MemCmd::ReadResp: 225 226 if (memcmp(pkt_data, data, pkt->getSize()) != 0) { 227 cerr << name() << ": on read of 0x" << hex << req->getPaddr() 228 << " (0x" << hex << blockAddr(req->getPaddr()) << ")" 229 << "@ cycle " << dec << curTick 230 << ", cache returns 0x"; 231 printData(cerr, pkt_data, pkt->getSize()); 232 cerr << ", expected 0x"; 233 printData(cerr, data, pkt->getSize()); 234 cerr << endl; 235 fatal(""); 236 } 237 238 numReads++; 239 numReadsStat++; 240 241 if (numReads == nextProgressMessage) { 242 ccprintf(cerr, "%s: completed %d read accesses @%d\n", 243 name(), numReads, curTick); 244 nextProgressMessage += progressInterval; 245 } 246 247 if (numReads >= maxLoads) 248 exitSimLoop("Maximum number of loads reached!"); 249 break; 250 251 case MemCmd::WriteResp: 252 numWritesStat++; 253 break; 254/* 255 case Copy: 256 //Also remove dest from outstanding list 257 removeAddr = outstandingAddrs.find(req->dest); 258 assert(removeAddr != outstandingAddrs.end()); 259 outstandingAddrs.erase(removeAddr); 260 numCopiesStat++; 261 break; 262*/ 263 default: 264 panic("invalid command %s (%d)", pkt->cmdString(), pkt->cmd.toInt()); 265 } 266 267 if (blockAddr(req->getPaddr()) == traceBlockAddr) { 268 cerr << name() << ": completed " 269 << (pkt->isWrite() ? "write" : "read") 270 << " access of " 271 << dec << pkt->getSize() << " bytes at address 0x" 272 << hex << req->getPaddr() 273 << " (0x" << hex << blockAddr(req->getPaddr()) << ")" 274 << ", value = 0x"; 275 printData(cerr, pkt_data, pkt->getSize()); 276 cerr << " @ cycle " << dec << curTick; 277 278 cerr << endl; 279 } 280 281 noResponseCycles = 0; 282 delete state; 283 delete [] data; 284 delete pkt->req; 285 delete pkt; 286} 287 288void 289MemTest::regStats() 290{ 291 using namespace Stats; 292 293 numReadsStat 294 .name(name() + ".num_reads") 295 .desc("number of read accesses completed") 296 ; 297 298 numWritesStat 299 .name(name() + ".num_writes") 300 .desc("number of write accesses completed") 301 ; 302 303 numCopiesStat 304 .name(name() + ".num_copies") 305 .desc("number of copy accesses completed") 306 ; 307} 308 309void 310MemTest::tick() 311{ 312 if (!tickEvent.scheduled()) 313 tickEvent.schedule(curTick + cycles(1)); 314 315 if (++noResponseCycles >= 500000) { 316 cerr << name() << ": deadlocked at cycle " << curTick << endl; 317 fatal(""); 318 } 319 320 if (accessRetry) { 321 return; 322 } 323 324 //make new request 325 unsigned cmd = random() % 100; 326 unsigned offset = random() % size; 327 unsigned base = random() % 2; 328 uint64_t data = random(); 329 unsigned access_size = random() % 4; 330 unsigned cacheable = random() % 100; 331 332 //If we aren't doing copies, use id as offset, and do a false sharing 333 //mem tester 334 //We can eliminate the lower bits of the offset, and then use the id 335 //to offset within the blks 336 offset &= ~63; //Not the low order bits 337 offset += id; 338 access_size = 0; 339 340 Request *req = new Request(); 341 uint32_t flags = 0; 342 Addr paddr; 343 344 if (cacheable < percentUncacheable) { 345 flags |= UNCACHEABLE; 346 paddr = uncacheAddr + offset; 347 } else { 348 paddr = ((base) ? baseAddr1 : baseAddr2) + offset; 349 } 350 bool probe = (random() % 100 < percentFunctional) && !(flags & UNCACHEABLE); 351 //bool probe = false; 352 353 paddr &= ~((1 << access_size) - 1); 354 req->setPhys(paddr, 1 << access_size, flags); 355 req->setThreadContext(id,0); 356 357 uint8_t *result = new uint8_t[8]; 358 359 if (cmd < percentReads) { 360 // read 361 362 //For now we only allow one outstanding request per addreess per tester 363 //This means we assume CPU does write forwarding to reads that alias something 364 //in the cpu store buffer. 365 if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) { 366 delete [] result; 367 delete req; 368 return; 369 } 370 else outstandingAddrs.insert(paddr); 371 372 // ***** NOTE FOR RON: I'm not sure how to access checkMem. - Kevin 373 funcPort.readBlob(req->getPaddr(), result, req->getSize()); 374 375 if (blockAddr(paddr) == traceBlockAddr) { 376 cerr << name() 377 << ": initiating read " 378 << ((probe) ? "probe of " : "access of ") 379 << dec << req->getSize() << " bytes from addr 0x" 380 << hex << paddr 381 << " (0x" << hex << blockAddr(paddr) << ")" 382 << " at cycle " 383 << dec << curTick << endl; 384 } 385 386 PacketPtr pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast); 387 pkt->dataDynamicArray(new uint8_t[req->getSize()]); 388 MemTestSenderState *state = new MemTestSenderState(result); 389 pkt->senderState = state; 390 391 if (probe) { 392 cachePort.sendFunctional(pkt); 393 pkt->makeAtomicResponse(); 394 completeRequest(pkt); 395 } else { 396// req->completionEvent = new MemCompleteEvent(req, result, this); 397 sendPkt(pkt); 398 } 399 } else { 400 // write 401 402 //For now we only allow one outstanding request per addreess per tester 403 //This means we assume CPU does write forwarding to reads that alias something 404 //in the cpu store buffer. 405 if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) { 406 delete [] result; 407 delete req; 408 return; 409 } 410 411 else outstandingAddrs.insert(paddr); 412 413/* 414 if (blockAddr(req->getPaddr()) == traceBlockAddr) { 415 cerr << name() << ": initiating write " 416 << ((probe)?"probe of ":"access of ") 417 << dec << req->getSize() << " bytes (value = 0x"; 418 printData(cerr, data_pkt->getPtr(), req->getSize()); 419 cerr << ") to addr 0x" 420 << hex << req->getPaddr() 421 << " (0x" << hex << blockAddr(req->getPaddr()) << ")" 422 << " at cycle " 423 << dec << curTick << endl; 424 } 425*/ 426 PacketPtr pkt = new Packet(req, MemCmd::WriteReq, Packet::Broadcast); 427 uint8_t *pkt_data = new uint8_t[req->getSize()]; 428 pkt->dataDynamicArray(pkt_data); 429 memcpy(pkt_data, &data, req->getSize()); 430 MemTestSenderState *state = new MemTestSenderState(result); 431 pkt->senderState = state; 432 433 funcPort.writeBlob(req->getPaddr(), pkt_data, req->getSize()); 434 435 if (probe) { 436 cachePort.sendFunctional(pkt); 437 pkt->makeAtomicResponse(); 438 completeRequest(pkt); 439 } else { 440// req->completionEvent = new MemCompleteEvent(req, NULL, this); 441 sendPkt(pkt); 442 } 443 } 444/* else { 445 // copy 446 unsigned source_align = random() % 100; 447 unsigned dest_align = random() % 100; 448 unsigned offset2 = random() % size; 449 450 Addr source = ((base) ? baseAddr1 : baseAddr2) + offset; 451 Addr dest = ((base) ? baseAddr2 : baseAddr1) + offset2; 452 if (outstandingAddrs.find(source) != outstandingAddrs.end()) return; 453 else outstandingAddrs.insert(source); 454 if (outstandingAddrs.find(dest) != outstandingAddrs.end()) return; 455 else outstandingAddrs.insert(dest); 456 457 if (source_align >= percentSourceUnaligned) { 458 source = blockAddr(source); 459 } 460 if (dest_align >= percentDestUnaligned) { 461 dest = blockAddr(dest); 462 } 463 req->cmd = Copy; 464 req->flags &= ~UNCACHEABLE; 465 req->paddr = source; 466 req->dest = dest; 467 delete [] req->data; 468 req->data = new uint8_t[blockSize]; 469 req->size = blockSize; 470 if (source == traceBlockAddr || dest == traceBlockAddr) { 471 cerr << name() 472 << ": initiating copy of " 473 << dec << req->size << " bytes from addr 0x" 474 << hex << source 475 << " (0x" << hex << blockAddr(source) << ")" 476 << " to addr 0x" 477 << hex << dest 478 << " (0x" << hex << blockAddr(dest) << ")" 479 << " at cycle " 480 << dec << curTick << endl; 481 }* 482 cacheInterface->access(req); 483 uint8_t result[blockSize]; 484 checkMem->access(Read, source, &result, blockSize); 485 checkMem->access(Write, dest, &result, blockSize); 486 } 487*/ 488} 489 490void 491MemTest::doRetry() 492{ 493 if (cachePort.sendTiming(retryPkt)) { 494 accessRetry = false; 495 retryPkt = NULL; 496 } 497} 498 499BEGIN_DECLARE_SIM_OBJECT_PARAMS(MemTest) 500 501// SimObjectParam<BaseCache *> cache; 502// SimObjectParam<PhysicalMemory *> main_mem; 503// SimObjectParam<PhysicalMemory *> check_mem; 504 Param<unsigned> memory_size; 505 Param<unsigned> percent_reads; 506 Param<unsigned> percent_functional; 507 Param<unsigned> percent_uncacheable; 508 Param<unsigned> progress_interval; 509 Param<unsigned> percent_source_unaligned; 510 Param<unsigned> percent_dest_unaligned; 511 Param<Addr> trace_addr; 512 Param<Counter> max_loads; 513 Param<bool> atomic; 514 515END_DECLARE_SIM_OBJECT_PARAMS(MemTest) 516 517 518BEGIN_INIT_SIM_OBJECT_PARAMS(MemTest) 519 520// INIT_PARAM(cache, "L1 cache"), 521// INIT_PARAM(main_mem, "hierarchical memory"), 522// INIT_PARAM(check_mem, "check memory"), 523 INIT_PARAM(memory_size, "memory size"), 524 INIT_PARAM(percent_reads, "target read percentage"), 525 INIT_PARAM(percent_functional, "percentage of access that are functional"), 526 INIT_PARAM(percent_uncacheable, "target uncacheable percentage"), 527 INIT_PARAM(progress_interval, "progress report interval (in accesses)"), 528 INIT_PARAM(percent_source_unaligned, 529 "percent of copy source address that are unaligned"), 530 INIT_PARAM(percent_dest_unaligned, 531 "percent of copy dest address that are unaligned"), 532 INIT_PARAM(trace_addr, "address to trace"), 533 INIT_PARAM(max_loads, "terminate when we have reached this load count"), 534 INIT_PARAM(atomic, "Is the tester testing atomic mode (or timing)") 535 536END_INIT_SIM_OBJECT_PARAMS(MemTest) 537 538 539CREATE_SIM_OBJECT(MemTest) 540{ 541 return new MemTest(getInstanceName(), /*cache->getInterface(),*/ /*main_mem,*/ 542 /*check_mem,*/ memory_size, percent_reads, percent_functional, 543 percent_uncacheable, progress_interval, 544 percent_source_unaligned, percent_dest_unaligned, 545 trace_addr, max_loads, atomic); 546} 547 548REGISTER_SIM_OBJECT("MemTest", MemTest) 549