1/* 2 * Copyright (c) 2003-2005 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 * Ron Dreslinski 31 */ 32 33/** 34 * @file 35 * Declares a basic cache interface BaseCache. 36 */ 37 38#ifndef __BASE_CACHE_HH__ 39#define __BASE_CACHE_HH__ 40 41#include <vector> 42#include <string> 43#include <list> 44#include <algorithm> 45 46#include "base/misc.hh" 47#include "base/statistics.hh" 48#include "base/trace.hh" 49#include "base/types.hh"
|
50#include "config/full_system.hh" |
51#include "mem/cache/mshr_queue.hh" 52#include "mem/mem_object.hh" 53#include "mem/packet.hh" 54#include "mem/tport.hh" 55#include "mem/request.hh" 56#include "params/BaseCache.hh" 57#include "sim/eventq.hh" 58#include "sim/sim_exit.hh" 59 60class MSHR; 61/** 62 * A basic cache interface. Implements some common functions for speed. 63 */ 64class BaseCache : public MemObject 65{ 66 /** 67 * Indexes to enumerate the MSHR queues. 68 */ 69 enum MSHRQueueIndex { 70 MSHRQueue_MSHRs, 71 MSHRQueue_WriteBuffer 72 }; 73 74 /** 75 * Reasons for caches to be blocked. 76 */ 77 enum BlockedCause { 78 Blocked_NoMSHRs = MSHRQueue_MSHRs, 79 Blocked_NoWBBuffers = MSHRQueue_WriteBuffer, 80 Blocked_NoTargets, 81 NUM_BLOCKED_CAUSES 82 }; 83 84 public: 85 /** 86 * Reasons for cache to request a bus. 87 */ 88 enum RequestCause { 89 Request_MSHR = MSHRQueue_MSHRs, 90 Request_WB = MSHRQueue_WriteBuffer, 91 Request_PF, 92 NUM_REQUEST_CAUSES 93 }; 94 95 private: 96 97 class CachePort : public SimpleTimingPort 98 { 99 public: 100 BaseCache *cache; 101 102 protected: 103 CachePort(const std::string &_name, BaseCache *_cache, 104 const std::string &_label); 105 106 virtual void recvStatusChange(Status status); 107 108 virtual unsigned deviceBlockSize() const; 109 110 bool recvRetryCommon(); 111 112 typedef EventWrapper<Port, &Port::sendRetry> 113 SendRetryEvent; 114 115 const std::string label; 116 117 public: 118 void setOtherPort(CachePort *_otherPort) { otherPort = _otherPort; } 119 120 void setBlocked(); 121 122 void clearBlocked(); 123 124 bool checkFunctional(PacketPtr pkt); 125 126 CachePort *otherPort; 127 128 bool blocked; 129 130 bool mustSendRetry; 131 132 void requestBus(RequestCause cause, Tick time) 133 { 134 DPRINTF(CachePort, "Asserting bus request for cause %d\n", cause); 135 if (!waitingOnRetry) { 136 schedSendEvent(time); 137 } 138 } 139 140 void respond(PacketPtr pkt, Tick time) { 141 schedSendTiming(pkt, time); 142 } 143 }; 144 145 public: //Made public so coherence can get at it. 146 CachePort *cpuSidePort; 147 CachePort *memSidePort; 148 149 protected: 150 151 /** Miss status registers */ 152 MSHRQueue mshrQueue; 153 154 /** Write/writeback buffer */ 155 MSHRQueue writeBuffer; 156 157 MSHR *allocateBufferInternal(MSHRQueue *mq, Addr addr, int size, 158 PacketPtr pkt, Tick time, bool requestBus) 159 { 160 MSHR *mshr = mq->allocate(addr, size, pkt, time, order++); 161 162 if (mq->isFull()) { 163 setBlocked((BlockedCause)mq->index); 164 } 165 166 if (requestBus) { 167 requestMemSideBus((RequestCause)mq->index, time); 168 } 169 170 return mshr; 171 } 172 173 void markInServiceInternal(MSHR *mshr) 174 { 175 MSHRQueue *mq = mshr->queue; 176 bool wasFull = mq->isFull(); 177 mq->markInService(mshr); 178 if (wasFull && !mq->isFull()) { 179 clearBlocked((BlockedCause)mq->index); 180 } 181 } 182 183 /** Block size of this cache */ 184 const unsigned blkSize; 185 186 /** 187 * The latency of a hit in this device. 188 */ 189 int hitLatency; 190 191 /** The number of targets for each MSHR. */ 192 const int numTarget; 193 194 /** Do we forward snoops from mem side port through to cpu side port? */ 195 bool forwardSnoops; 196 197 /** 198 * Bit vector of the blocking reasons for the access path. 199 * @sa #BlockedCause 200 */ 201 uint8_t blocked; 202 203 /** Increasing order number assigned to each incoming request. */ 204 uint64_t order; 205 206 /** Stores time the cache blocked for statistics. */ 207 Tick blockedCycle; 208 209 /** Pointer to the MSHR that has no targets. */ 210 MSHR *noTargetMSHR; 211 212 /** The number of misses to trigger an exit event. */ 213 Counter missCount; 214 215 /** The drain event. */ 216 Event *drainEvent; 217 218 /** 219 * The address range to which the cache responds on the CPU side. 220 * Normally this is all possible memory addresses. */ 221 Range<Addr> addrRange; 222
|
223 /** number of cpus sharing this cache - from config file */ 224 int _numCpus; 225 |
226 public:
|
227 int numCpus() { return _numCpus; } |
228 // Statistics 229 /** 230 * @addtogroup CacheStatistics 231 * @{ 232 */ 233 234 /** Number of hits per thread for each type of command. @sa Packet::Command */ 235 Stats::Vector hits[MemCmd::NUM_MEM_CMDS]; 236 /** Number of hits for demand accesses. */ 237 Stats::Formula demandHits; 238 /** Number of hit for all accesses. */ 239 Stats::Formula overallHits; 240 241 /** Number of misses per thread for each type of command. @sa Packet::Command */ 242 Stats::Vector misses[MemCmd::NUM_MEM_CMDS]; 243 /** Number of misses for demand accesses. */ 244 Stats::Formula demandMisses; 245 /** Number of misses for all accesses. */ 246 Stats::Formula overallMisses; 247 248 /** 249 * Total number of cycles per thread/command spent waiting for a miss. 250 * Used to calculate the average miss latency. 251 */ 252 Stats::Vector missLatency[MemCmd::NUM_MEM_CMDS]; 253 /** Total number of cycles spent waiting for demand misses. */ 254 Stats::Formula demandMissLatency; 255 /** Total number of cycles spent waiting for all misses. */ 256 Stats::Formula overallMissLatency; 257 258 /** The number of accesses per command and thread. */ 259 Stats::Formula accesses[MemCmd::NUM_MEM_CMDS]; 260 /** The number of demand accesses. */ 261 Stats::Formula demandAccesses; 262 /** The number of overall accesses. */ 263 Stats::Formula overallAccesses; 264 265 /** The miss rate per command and thread. */ 266 Stats::Formula missRate[MemCmd::NUM_MEM_CMDS]; 267 /** The miss rate of all demand accesses. */ 268 Stats::Formula demandMissRate; 269 /** The miss rate for all accesses. */ 270 Stats::Formula overallMissRate; 271 272 /** The average miss latency per command and thread. */ 273 Stats::Formula avgMissLatency[MemCmd::NUM_MEM_CMDS]; 274 /** The average miss latency for demand misses. */ 275 Stats::Formula demandAvgMissLatency; 276 /** The average miss latency for all misses. */ 277 Stats::Formula overallAvgMissLatency; 278 279 /** The total number of cycles blocked for each blocked cause. */ 280 Stats::Vector blocked_cycles; 281 /** The number of times this cache blocked for each blocked cause. */ 282 Stats::Vector blocked_causes; 283 284 /** The average number of cycles blocked for each blocked cause. */ 285 Stats::Formula avg_blocked; 286 287 /** The number of fast writes (WH64) performed. */ 288 Stats::Scalar fastWrites; 289 290 /** The number of cache copies performed. */ 291 Stats::Scalar cacheCopies; 292 293 /** Number of blocks written back per thread. */ 294 Stats::Vector writebacks; 295 296 /** Number of misses that hit in the MSHRs per command and thread. */ 297 Stats::Vector mshr_hits[MemCmd::NUM_MEM_CMDS]; 298 /** Demand misses that hit in the MSHRs. */ 299 Stats::Formula demandMshrHits; 300 /** Total number of misses that hit in the MSHRs. */ 301 Stats::Formula overallMshrHits; 302 303 /** Number of misses that miss in the MSHRs, per command and thread. */ 304 Stats::Vector mshr_misses[MemCmd::NUM_MEM_CMDS]; 305 /** Demand misses that miss in the MSHRs. */ 306 Stats::Formula demandMshrMisses; 307 /** Total number of misses that miss in the MSHRs. */ 308 Stats::Formula overallMshrMisses; 309 310 /** Number of misses that miss in the MSHRs, per command and thread. */ 311 Stats::Vector mshr_uncacheable[MemCmd::NUM_MEM_CMDS]; 312 /** Total number of misses that miss in the MSHRs. */ 313 Stats::Formula overallMshrUncacheable; 314 315 /** Total cycle latency of each MSHR miss, per command and thread. */ 316 Stats::Vector mshr_miss_latency[MemCmd::NUM_MEM_CMDS]; 317 /** Total cycle latency of demand MSHR misses. */ 318 Stats::Formula demandMshrMissLatency; 319 /** Total cycle latency of overall MSHR misses. */ 320 Stats::Formula overallMshrMissLatency; 321 322 /** Total cycle latency of each MSHR miss, per command and thread. */ 323 Stats::Vector mshr_uncacheable_lat[MemCmd::NUM_MEM_CMDS]; 324 /** Total cycle latency of overall MSHR misses. */ 325 Stats::Formula overallMshrUncacheableLatency; 326 327 /** The total number of MSHR accesses per command and thread. */ 328 Stats::Formula mshrAccesses[MemCmd::NUM_MEM_CMDS]; 329 /** The total number of demand MSHR accesses. */ 330 Stats::Formula demandMshrAccesses; 331 /** The total number of MSHR accesses. */ 332 Stats::Formula overallMshrAccesses; 333 334 /** The miss rate in the MSHRs pre command and thread. */ 335 Stats::Formula mshrMissRate[MemCmd::NUM_MEM_CMDS]; 336 /** The demand miss rate in the MSHRs. */ 337 Stats::Formula demandMshrMissRate; 338 /** The overall miss rate in the MSHRs. */ 339 Stats::Formula overallMshrMissRate; 340 341 /** The average latency of an MSHR miss, per command and thread. */ 342 Stats::Formula avgMshrMissLatency[MemCmd::NUM_MEM_CMDS]; 343 /** The average latency of a demand MSHR miss. */ 344 Stats::Formula demandAvgMshrMissLatency; 345 /** The average overall latency of an MSHR miss. */ 346 Stats::Formula overallAvgMshrMissLatency; 347 348 /** The average latency of an MSHR miss, per command and thread. */ 349 Stats::Formula avgMshrUncacheableLatency[MemCmd::NUM_MEM_CMDS]; 350 /** The average overall latency of an MSHR miss. */ 351 Stats::Formula overallAvgMshrUncacheableLatency; 352 353 /** The number of times a thread hit its MSHR cap. */ 354 Stats::Vector mshr_cap_events; 355 /** The number of times software prefetches caused the MSHR to block. */ 356 Stats::Vector soft_prefetch_mshr_full; 357 358 Stats::Scalar mshr_no_allocate_misses; 359 360 /** 361 * @} 362 */ 363 364 /** 365 * Register stats for this object. 366 */ 367 virtual void regStats(); 368 369 public: 370 typedef BaseCacheParams Params; 371 BaseCache(const Params *p); 372 ~BaseCache() {} 373 374 virtual void init(); 375 376 /** 377 * Query block size of a cache. 378 * @return The block size 379 */ 380 unsigned 381 getBlockSize() const 382 { 383 return blkSize; 384 } 385 386 387 Addr blockAlign(Addr addr) const { return (addr & ~(Addr(blkSize - 1))); } 388 389 390 const Range<Addr> &getAddrRange() const { return addrRange; } 391 392 MSHR *allocateMissBuffer(PacketPtr pkt, Tick time, bool requestBus) 393 { 394 assert(!pkt->req->isUncacheable()); 395 return allocateBufferInternal(&mshrQueue, 396 blockAlign(pkt->getAddr()), blkSize, 397 pkt, time, requestBus); 398 } 399 400 MSHR *allocateWriteBuffer(PacketPtr pkt, Tick time, bool requestBus) 401 { 402 assert(pkt->isWrite() && !pkt->isRead()); 403 return allocateBufferInternal(&writeBuffer, 404 pkt->getAddr(), pkt->getSize(), 405 pkt, time, requestBus); 406 } 407 408 MSHR *allocateUncachedReadBuffer(PacketPtr pkt, Tick time, bool requestBus) 409 { 410 assert(pkt->req->isUncacheable()); 411 assert(pkt->isRead()); 412 return allocateBufferInternal(&mshrQueue, 413 pkt->getAddr(), pkt->getSize(), 414 pkt, time, requestBus); 415 } 416 417 /** 418 * Returns true if the cache is blocked for accesses. 419 */ 420 bool isBlocked() 421 { 422 return blocked != 0; 423 } 424 425 /** 426 * Marks the access path of the cache as blocked for the given cause. This 427 * also sets the blocked flag in the slave interface. 428 * @param cause The reason for the cache blocking. 429 */ 430 void setBlocked(BlockedCause cause) 431 { 432 uint8_t flag = 1 << cause; 433 if (blocked == 0) { 434 blocked_causes[cause]++; 435 blockedCycle = curTick; 436 cpuSidePort->setBlocked(); 437 } 438 blocked |= flag; 439 DPRINTF(Cache,"Blocking for cause %d, mask=%d\n", cause, blocked); 440 } 441 442 /** 443 * Marks the cache as unblocked for the given cause. This also clears the 444 * blocked flags in the appropriate interfaces. 445 * @param cause The newly unblocked cause. 446 * @warning Calling this function can cause a blocked request on the bus to 447 * access the cache. The cache must be in a state to handle that request. 448 */ 449 void clearBlocked(BlockedCause cause) 450 { 451 uint8_t flag = 1 << cause; 452 blocked &= ~flag; 453 DPRINTF(Cache,"Unblocking for cause %d, mask=%d\n", cause, blocked); 454 if (blocked == 0) { 455 blocked_cycles[cause] += curTick - blockedCycle; 456 cpuSidePort->clearBlocked(); 457 } 458 } 459 460 /** 461 * Request the master bus for the given cause and time. 462 * @param cause The reason for the request. 463 * @param time The time to make the request. 464 */ 465 void requestMemSideBus(RequestCause cause, Tick time) 466 { 467 memSidePort->requestBus(cause, time); 468 } 469 470 /** 471 * Clear the master bus request for the given cause. 472 * @param cause The request reason to clear. 473 */ 474 void deassertMemSideBusRequest(RequestCause cause) 475 { 476 // Obsolete... we no longer signal bus requests explicitly so 477 // we can't deassert them. Leaving this in as a no-op since 478 // the prefetcher calls it to indicate that it no longer wants 479 // to request a prefetch, and someday that might be 480 // interesting again. 481 } 482 483 virtual unsigned int drain(Event *de); 484 485 virtual bool inCache(Addr addr) = 0; 486 487 virtual bool inMissQueue(Addr addr) = 0; 488
|
484 void incMissCount(PacketPtr pkt)
|
489 void incMissCount(PacketPtr pkt, int id) |
490 {
|
486 misses[pkt->cmdToIndex()][0/*pkt->req->threadId()*/]++;
|
491
|
492 if (pkt->cmd == MemCmd::Writeback) { 493 assert(id == -1); 494 misses[pkt->cmdToIndex()][0]++; 495 /* same thing for writeback hits as misses - no context id 496 * available, meanwhile writeback hit/miss stats are not used 497 * in any aggregate hit/miss calculations, so just lump them all 498 * in bucket 0 */ 499#if FULL_SYSTEM 500 } else if (id == -1) { 501 // Device accesses have id -1 502 // lump device accesses into their own bucket 503 misses[pkt->cmdToIndex()][_numCpus]++; 504#endif 505 } else { 506 misses[pkt->cmdToIndex()][id % _numCpus]++; 507 } 508 |
509 if (missCount) { 510 --missCount; 511 if (missCount == 0) 512 exitSimLoop("A cache reached the maximum miss count"); 513 } 514 }
|
515 void incHitCount(PacketPtr pkt, int id) 516 { |
517
|
518 /* Writeback requests don't have a context id associated with 519 * them, so attributing a hit to a -1 context id is obviously a 520 * problem. I've noticed in the stats that hits are split into 521 * demand and non-demand hits - neither of which include writeback 522 * hits, so here, I'll just put the writeback hits into bucket 0 523 * since it won't mess with any other stats -hsul */ 524 if (pkt->cmd == MemCmd::Writeback) { 525 assert(id == -1); 526 hits[pkt->cmdToIndex()][0]++; 527#if FULL_SYSTEM 528 } else if (id == -1) { 529 // Device accesses have id -1 530 // lump device accesses into their own bucket 531 hits[pkt->cmdToIndex()][_numCpus]++; 532#endif 533 } else { 534 /* the % is necessary in case there are switch cpus */ 535 hits[pkt->cmdToIndex()][id % _numCpus]++; 536 } 537 } 538 |
539}; 540 541#endif //__BASE_CACHE_HH__
|