Merge zizzer:/bk/newmem
[gem5.git] / src / mem / cache / cache_impl.hh
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 * Dave Greene
30 * Nathan Binkert
31 */
32
33 /**
34 * @file
35 * Cache definitions.
36 */
37
38 #include <assert.h>
39 #include <math.h>
40
41 #include <cassert>
42 #include <iostream>
43 #include <string>
44
45 #include "sim/host.hh"
46 #include "base/misc.hh"
47 #include "cpu/smt.hh"
48
49 #include "mem/cache/cache.hh"
50 #include "mem/cache/cache_blk.hh"
51 #include "mem/cache/miss/mshr.hh"
52 #include "mem/cache/prefetch/prefetcher.hh"
53
54 #include "sim/sim_exit.hh" // for SimExitEvent
55
56 template<class TagStore, class Buffering, class Coherence>
57 bool
58 Cache<TagStore,Buffering,Coherence>::
59 doTimingAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide)
60 {
61 if (isCpuSide)
62 {
63 if (pkt->isWrite() && (pkt->req->isLocked())) {
64 pkt->req->setScResult(1);
65 }
66 access(pkt);
67
68 }
69 else
70 {
71 if (pkt->isResponse())
72 handleResponse(pkt);
73 else {
74 //Check if we should do the snoop
75 if (pkt->flags & SNOOP_COMMIT)
76 snoop(pkt);
77 }
78 }
79 return true;
80 }
81
82 template<class TagStore, class Buffering, class Coherence>
83 Tick
84 Cache<TagStore,Buffering,Coherence>::
85 doAtomicAccess(Packet *pkt, bool isCpuSide)
86 {
87 if (isCpuSide)
88 {
89 //Temporary solution to LL/SC
90 if (pkt->isWrite() && (pkt->req->isLocked())) {
91 pkt->req->setScResult(1);
92 }
93
94 probe(pkt, true, NULL);
95 //TEMP ALWAYS SUCCES FOR NOW
96 pkt->result = Packet::Success;
97 }
98 else
99 {
100 if (pkt->isResponse())
101 handleResponse(pkt);
102 else
103 return snoopProbe(pkt);
104 }
105 //Fix this timing info
106 return hitLatency;
107 }
108
109 template<class TagStore, class Buffering, class Coherence>
110 void
111 Cache<TagStore,Buffering,Coherence>::
112 doFunctionalAccess(Packet *pkt, bool isCpuSide)
113 {
114 if (isCpuSide)
115 {
116 //TEMP USE CPU?THREAD 0 0
117 pkt->req->setThreadContext(0,0);
118
119 //Temporary solution to LL/SC
120 if (pkt->isWrite() && (pkt->req->isLocked())) {
121 assert("Can't handle LL/SC on functional path\n");
122 }
123
124 probe(pkt, false, memSidePort);
125 //TEMP ALWAYS SUCCESFUL FOR NOW
126 pkt->result = Packet::Success;
127 }
128 else
129 {
130 probe(pkt, false, cpuSidePort);
131 }
132 }
133
134 template<class TagStore, class Buffering, class Coherence>
135 void
136 Cache<TagStore,Buffering,Coherence>::
137 recvStatusChange(Port::Status status, bool isCpuSide)
138 {
139
140 }
141
142
143 template<class TagStore, class Buffering, class Coherence>
144 Cache<TagStore,Buffering,Coherence>::
145 Cache(const std::string &_name,
146 Cache<TagStore,Buffering,Coherence>::Params &params)
147 : BaseCache(_name, params.baseParams),
148 prefetchAccess(params.prefetchAccess),
149 tags(params.tags), missQueue(params.missQueue),
150 coherence(params.coherence), prefetcher(params.prefetcher),
151 doCopy(params.doCopy), blockOnCopy(params.blockOnCopy),
152 hitLatency(params.hitLatency)
153 {
154 tags->setCache(this);
155 tags->setPrefetcher(prefetcher);
156 missQueue->setCache(this);
157 missQueue->setPrefetcher(prefetcher);
158 coherence->setCache(this);
159 prefetcher->setCache(this);
160 prefetcher->setTags(tags);
161 prefetcher->setBuffer(missQueue);
162 invalidateReq = new Request((Addr) NULL, blkSize, 0);
163 invalidatePkt = new Packet(invalidateReq, Packet::InvalidateReq, 0);
164 }
165
166 template<class TagStore, class Buffering, class Coherence>
167 void
168 Cache<TagStore,Buffering,Coherence>::regStats()
169 {
170 BaseCache::regStats();
171 tags->regStats(name());
172 missQueue->regStats(name());
173 coherence->regStats(name());
174 prefetcher->regStats(name());
175 }
176
177 template<class TagStore, class Buffering, class Coherence>
178 bool
179 Cache<TagStore,Buffering,Coherence>::access(PacketPtr &pkt)
180 {
181 //@todo Add back in MemDebug Calls
182 // MemDebug::cacheAccess(pkt);
183 BlkType *blk = NULL;
184 PacketList writebacks;
185 int size = blkSize;
186 int lat = hitLatency;
187 if (prefetchAccess) {
188 //We are determining prefetches on access stream, call prefetcher
189 prefetcher->handleMiss(pkt, curTick);
190 }
191 if (!pkt->req->isUncacheable()) {
192 blk = tags->handleAccess(pkt, lat, writebacks);
193 } else {
194 size = pkt->getSize();
195 }
196 // If this is a block size write/hint (WH64) allocate the block here
197 // if the coherence protocol allows it.
198 /** @todo make the fast write alloc (wh64) work with coherence. */
199 /** @todo Do we want to do fast writes for writebacks as well? */
200 if (!blk && pkt->getSize() >= blkSize && coherence->allowFastWrites() &&
201 (pkt->cmd == Packet::WriteReq || pkt->cmd == Packet::WriteInvalidateReq) ) {
202 // not outstanding misses, can do this
203 MSHR* outstanding_miss = missQueue->findMSHR(pkt->getAddr());
204 if (pkt->cmd == Packet::WriteInvalidateReq || !outstanding_miss) {
205 if (outstanding_miss) {
206 warn("WriteInv doing a fastallocate"
207 "with an outstanding miss to the same address\n");
208 }
209 blk = tags->handleFill(NULL, pkt, BlkValid | BlkWritable,
210 writebacks);
211 ++fastWrites;
212 }
213 }
214 while (!writebacks.empty()) {
215 missQueue->doWriteback(writebacks.front());
216 writebacks.pop_front();
217 }
218 DPRINTF(Cache, "%s %x %s blk_addr: %x\n", pkt->cmdString(),
219 pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss",
220 pkt->getAddr() & ~((Addr)blkSize - 1));
221 if (blk) {
222 // Hit
223 hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
224 // clear dirty bit if write through
225 if (pkt->needsResponse())
226 respond(pkt, curTick+lat);
227 if (pkt->cmd == Packet::Writeback) {
228 //Signal that you can kill the pkt/req
229 pkt->flags |= SATISFIED;
230 }
231 return true;
232 }
233
234 // Miss
235 if (!pkt->req->isUncacheable()) {
236 misses[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
237 /** @todo Move miss count code into BaseCache */
238 if (missCount) {
239 --missCount;
240 if (missCount == 0)
241 exitSimLoop("A cache reached the maximum miss count");
242 }
243 }
244 missQueue->handleMiss(pkt, size, curTick + hitLatency);
245 // return MA_CACHE_MISS;
246 return true;
247 }
248
249
250 template<class TagStore, class Buffering, class Coherence>
251 Packet *
252 Cache<TagStore,Buffering,Coherence>::getPacket()
253 {
254 assert(missQueue->havePending());
255 Packet * pkt = missQueue->getPacket();
256 if (pkt) {
257 if (!pkt->req->isUncacheable()) {
258 if (pkt->cmd == Packet::HardPFReq) misses[Packet::HardPFReq][0/*pkt->req->getThreadNum()*/]++;
259 BlkType *blk = tags->findBlock(pkt);
260 Packet::Command cmd = coherence->getBusCmd(pkt->cmd,
261 (blk)? blk->status : 0);
262 missQueue->setBusCmd(pkt, cmd);
263 }
264 }
265
266 assert(!doMasterRequest() || missQueue->havePending());
267 assert(!pkt || pkt->time <= curTick);
268 return pkt;
269 }
270
271 template<class TagStore, class Buffering, class Coherence>
272 void
273 Cache<TagStore,Buffering,Coherence>::sendResult(PacketPtr &pkt, MSHR* mshr, bool success)
274 {
275 if (success && !(pkt->flags & NACKED_LINE)) {
276 missQueue->markInService(pkt, mshr);
277 //Temp Hack for UPGRADES
278 if (pkt->cmd == Packet::UpgradeReq) {
279 pkt->flags &= ~CACHE_LINE_FILL;
280 BlkType *blk = tags->findBlock(pkt);
281 CacheBlk::State old_state = (blk) ? blk->status : 0;
282 CacheBlk::State new_state = coherence->getNewState(pkt,old_state);
283 if (old_state != new_state)
284 DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
285 pkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
286 //Set the state on the upgrade
287 memcpy(pkt->getPtr<uint8_t>(), blk->data, blkSize);
288 PacketList writebacks;
289 tags->handleFill(blk, mshr, new_state, writebacks, pkt);
290 assert(writebacks.empty());
291 missQueue->handleResponse(pkt, curTick + hitLatency);
292 }
293 } else if (pkt && !pkt->req->isUncacheable()) {
294 pkt->flags &= ~NACKED_LINE;
295 pkt->flags &= ~SATISFIED;
296 pkt->flags &= ~SNOOP_COMMIT;
297 missQueue->restoreOrigCmd(pkt);
298 }
299 }
300
301 template<class TagStore, class Buffering, class Coherence>
302 void
303 Cache<TagStore,Buffering,Coherence>::handleResponse(Packet * &pkt)
304 {
305 BlkType *blk = NULL;
306 if (pkt->senderState) {
307 ((MSHR*)pkt->senderState)->pkt = pkt;
308 if (pkt->result == Packet::Nacked) {
309 //pkt->reinitFromRequest();
310 warn("NACKs from devices not connected to the same bus not implemented\n");
311 return;
312 }
313 if (pkt->result == Packet::BadAddress) {
314 //Make the response a Bad address and send it
315 }
316 // MemDebug::cacheResponse(pkt);
317 DPRINTF(Cache, "Handling reponse to %x, blk addr: %x\n",pkt->getAddr(),
318 pkt->getAddr() & (((ULL(1))<<48)-1));
319
320 if (pkt->isCacheFill() && !pkt->isNoAllocate()) {
321 blk = tags->findBlock(pkt);
322 CacheBlk::State old_state = (blk) ? blk->status : 0;
323 PacketList writebacks;
324 CacheBlk::State new_state = coherence->getNewState(pkt,old_state);
325 if (old_state != new_state)
326 DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
327 pkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
328 blk = tags->handleFill(blk, (MSHR*)pkt->senderState,
329 new_state, writebacks, pkt);
330 while (!writebacks.empty()) {
331 missQueue->doWriteback(writebacks.front());
332 writebacks.pop_front();
333 }
334 }
335 missQueue->handleResponse(pkt, curTick + hitLatency);
336 }
337 }
338
339 template<class TagStore, class Buffering, class Coherence>
340 void
341 Cache<TagStore,Buffering,Coherence>::pseudoFill(Addr addr)
342 {
343 // Need to temporarily move this blk into MSHRs
344 MSHR *mshr = missQueue->allocateTargetList(addr);
345 int lat;
346 PacketList dummy;
347 // Read the data into the mshr
348 BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
349 assert(dummy.empty());
350 assert(mshr->pkt->flags & SATISFIED);
351 // can overload order since it isn't used on non pending blocks
352 mshr->order = blk->status;
353 // temporarily remove the block from the cache.
354 tags->invalidateBlk(addr);
355 }
356
357 template<class TagStore, class Buffering, class Coherence>
358 void
359 Cache<TagStore,Buffering,Coherence>::pseudoFill(MSHR *mshr)
360 {
361 // Need to temporarily move this blk into MSHRs
362 assert(mshr->pkt->cmd == Packet::ReadReq);
363 int lat;
364 PacketList dummy;
365 // Read the data into the mshr
366 BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
367 assert(dummy.empty());
368 assert(mshr->pkt->flags & SATISFIED);
369 // can overload order since it isn't used on non pending blocks
370 mshr->order = blk->status;
371 // temporarily remove the block from the cache.
372 tags->invalidateBlk(mshr->pkt->getAddr());
373 }
374
375
376 template<class TagStore, class Buffering, class Coherence>
377 Packet *
378 Cache<TagStore,Buffering,Coherence>::getCoherencePacket()
379 {
380 return coherence->getPacket();
381 }
382
383 template<class TagStore, class Buffering, class Coherence>
384 void
385 Cache<TagStore,Buffering,Coherence>::sendCoherenceResult(Packet* &pkt,
386 MSHR *cshr,
387 bool success)
388 {
389 coherence->sendResult(pkt, cshr, success);
390 }
391
392
393 template<class TagStore, class Buffering, class Coherence>
394 void
395 Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
396 {
397 if (pkt->req->isUncacheable()) {
398 //Can't get a hit on an uncacheable address
399 //Revisit this for multi level coherence
400 return;
401 }
402 Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
403 BlkType *blk = tags->findBlock(pkt);
404 MSHR *mshr = missQueue->findMSHR(blk_addr);
405 if (coherence->hasProtocol()) { //@todo Move this into handle bus req
406 //If we find an mshr, and it is in service, we need to NACK or invalidate
407 if (mshr) {
408 if (mshr->inService) {
409 if ((mshr->pkt->isInvalidate() || !mshr->pkt->isCacheFill())
410 && (pkt->cmd != Packet::InvalidateReq && pkt->cmd != Packet::WriteInvalidateReq)) {
411 //If the outstanding request was an invalidate (upgrade,readex,..)
412 //Then we need to ACK the request until we get the data
413 //Also NACK if the outstanding request is not a cachefill (writeback)
414 assert(!(pkt->flags & SATISFIED));
415 pkt->flags |= SATISFIED;
416 pkt->flags |= NACKED_LINE;
417 ///@todo NACK's from other levels
418 //warn("NACKs from devices not connected to the same bus not implemented\n");
419 //respondToSnoop(pkt, curTick + hitLatency);
420 return;
421 }
422 else {
423 //The supplier will be someone else, because we are waiting for
424 //the data. This should cause this cache to be forced to go to
425 //the shared state, not the exclusive even though the shared line
426 //won't be asserted. But for now we will just invlidate ourselves
427 //and allow the other cache to go into the exclusive state.
428 //@todo Make it so a read to a pending read doesn't invalidate.
429 //@todo Make it so that a read to a pending read can't be exclusive now.
430
431 //Set the address so find match works
432 //panic("Don't have invalidates yet\n");
433 invalidatePkt->addrOverride(pkt->getAddr());
434
435 //Append the invalidate on
436 missQueue->addTarget(mshr,invalidatePkt);
437 DPRINTF(Cache, "Appending Invalidate to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
438 return;
439 }
440 }
441 }
442 //We also need to check the writeback buffers and handle those
443 std::vector<MSHR *> writebacks;
444 if (missQueue->findWrites(blk_addr, writebacks)) {
445 DPRINTF(Cache, "Snoop hit in writeback to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
446
447 //Look through writebacks for any non-uncachable writes, use that
448 for (int i=0; i<writebacks.size(); i++) {
449 mshr = writebacks[i];
450
451 if (!mshr->pkt->req->isUncacheable()) {
452 if (pkt->isRead()) {
453 //Only Upgrades don't get here
454 //Supply the data
455 assert(!(pkt->flags & SATISFIED));
456 pkt->flags |= SATISFIED;
457
458 //If we are in an exclusive protocol, make it ask again
459 //to get write permissions (upgrade), signal shared
460 pkt->flags |= SHARED_LINE;
461
462 assert(pkt->isRead());
463 Addr offset = pkt->getAddr() & (blkSize - 1);
464 assert(offset < blkSize);
465 assert(pkt->getSize() <= blkSize);
466 assert(offset + pkt->getSize() <=blkSize);
467 memcpy(pkt->getPtr<uint8_t>(), mshr->pkt->getPtr<uint8_t>() + offset, pkt->getSize());
468
469 respondToSnoop(pkt, curTick + hitLatency);
470 }
471
472 if (pkt->isInvalidate()) {
473 //This must be an upgrade or other cache will take ownership
474 missQueue->markInService(mshr->pkt, mshr);
475 }
476 return;
477 }
478 }
479 }
480 }
481 CacheBlk::State new_state;
482 bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
483 if (satisfy) {
484 DPRINTF(Cache, "Cache snooped a %s request for addr %x and now supplying data,"
485 "new state is %i\n",
486 pkt->cmdString(), blk_addr, new_state);
487
488 tags->handleSnoop(blk, new_state, pkt);
489 respondToSnoop(pkt, curTick + hitLatency);
490 return;
491 }
492 if (blk) DPRINTF(Cache, "Cache snooped a %s request for addr %x, new state is %i\n",
493 pkt->cmdString(), blk_addr, new_state);
494 tags->handleSnoop(blk, new_state);
495 }
496
497 template<class TagStore, class Buffering, class Coherence>
498 void
499 Cache<TagStore,Buffering,Coherence>::snoopResponse(Packet * &pkt)
500 {
501 //Need to handle the response, if NACKED
502 if (pkt->flags & NACKED_LINE) {
503 //Need to mark it as not in service, and retry for bus
504 assert(0); //Yeah, we saw a NACK come through
505
506 //For now this should never get called, we return false when we see a NACK
507 //instead, by doing this we allow the bus_blocked mechanism to handle the retry
508 //For now it retrys in just 2 cycles, need to figure out how to change that
509 //Eventually we will want to also have success come in as a parameter
510 //Need to make sure that we handle the functionality that happens on successufl
511 //return of the sendAddr function
512 }
513 }
514
515 template<class TagStore, class Buffering, class Coherence>
516 void
517 Cache<TagStore,Buffering,Coherence>::invalidateBlk(Addr addr)
518 {
519 tags->invalidateBlk(addr);
520 }
521
522
523 /**
524 * @todo Fix to not assume write allocate
525 */
526 template<class TagStore, class Buffering, class Coherence>
527 Tick
528 Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort* otherSidePort)
529 {
530 // MemDebug::cacheProbe(pkt);
531 if (!pkt->req->isUncacheable()) {
532 if (pkt->isInvalidate() && !pkt->isRead()
533 && !pkt->isWrite()) {
534 //Upgrade or Invalidate, satisfy it, don't forward
535 DPRINTF(Cache, "%s %x ? blk_addr: %x\n", pkt->cmdString(),
536 pkt->getAddr() & (((ULL(1))<<48)-1),
537 pkt->getAddr() & ~((Addr)blkSize - 1));
538 pkt->flags |= SATISFIED;
539 return 0;
540 }
541 }
542
543 PacketList writebacks;
544 int lat;
545 BlkType *blk = tags->handleAccess(pkt, lat, writebacks, update);
546
547 DPRINTF(Cache, "%s %x %s blk_addr: %x\n", pkt->cmdString(),
548 pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss",
549 pkt->getAddr() & ~((Addr)blkSize - 1));
550
551 if (!blk) {
552 // Need to check for outstanding misses and writes
553 Addr blk_addr = pkt->getAddr() & ~(blkSize - 1);
554
555 // There can only be one matching outstanding miss.
556 MSHR* mshr = missQueue->findMSHR(blk_addr);
557
558 // There can be many matching outstanding writes.
559 std::vector<MSHR*> writes;
560 missQueue->findWrites(blk_addr, writes);
561
562 if (!update) {
563 otherSidePort->sendFunctional(pkt);
564
565 // Check for data in MSHR and writebuffer.
566 if (mshr) {
567 warn("Found outstanding miss on an non-update probe");
568 MSHR::TargetList *targets = mshr->getTargetList();
569 MSHR::TargetList::iterator i = targets->begin();
570 MSHR::TargetList::iterator end = targets->end();
571 for (; i != end; ++i) {
572 Packet * target = *i;
573 // If the target contains data, and it overlaps the
574 // probed request, need to update data
575 if (target->isWrite() && target->intersect(pkt)) {
576 uint8_t* pkt_data;
577 uint8_t* write_data;
578 int data_size;
579 if (target->getAddr() < pkt->getAddr()) {
580 int offset = pkt->getAddr() - target->getAddr();
581 pkt_data = pkt->getPtr<uint8_t>();
582 write_data = target->getPtr<uint8_t>() + offset;
583 data_size = target->getSize() - offset;
584 assert(data_size > 0);
585 if (data_size > pkt->getSize())
586 data_size = pkt->getSize();
587 } else {
588 int offset = target->getAddr() - pkt->getAddr();
589 pkt_data = pkt->getPtr<uint8_t>() + offset;
590 write_data = target->getPtr<uint8_t>();
591 data_size = pkt->getSize() - offset;
592 assert(data_size > pkt->getSize());
593 if (data_size > target->getSize())
594 data_size = target->getSize();
595 }
596
597 if (pkt->isWrite()) {
598 memcpy(pkt_data, write_data, data_size);
599 } else {
600 memcpy(write_data, pkt_data, data_size);
601 }
602 }
603 }
604 }
605 for (int i = 0; i < writes.size(); ++i) {
606 Packet * write = writes[i]->pkt;
607 if (write->intersect(pkt)) {
608 warn("Found outstanding write on an non-update probe");
609 uint8_t* pkt_data;
610 uint8_t* write_data;
611 int data_size;
612 if (write->getAddr() < pkt->getAddr()) {
613 int offset = pkt->getAddr() - write->getAddr();
614 pkt_data = pkt->getPtr<uint8_t>();
615 write_data = write->getPtr<uint8_t>() + offset;
616 data_size = write->getSize() - offset;
617 assert(data_size > 0);
618 if (data_size > pkt->getSize())
619 data_size = pkt->getSize();
620 } else {
621 int offset = write->getAddr() - pkt->getAddr();
622 pkt_data = pkt->getPtr<uint8_t>() + offset;
623 write_data = write->getPtr<uint8_t>();
624 data_size = pkt->getSize() - offset;
625 assert(data_size > pkt->getSize());
626 if (data_size > write->getSize())
627 data_size = write->getSize();
628 }
629
630 if (pkt->isWrite()) {
631 memcpy(pkt_data, write_data, data_size);
632 } else {
633 memcpy(write_data, pkt_data, data_size);
634 }
635
636 }
637 }
638 return 0;
639 } else {
640 // update the cache state and statistics
641 if (mshr || !writes.empty()){
642 // Can't handle it, return pktuest unsatisfied.
643 panic("Atomic access ran into outstanding MSHR's or WB's!");
644 }
645 if (!pkt->req->isUncacheable()) {
646 // Fetch the cache block to fill
647 BlkType *blk = tags->findBlock(pkt);
648 Packet::Command temp_cmd = coherence->getBusCmd(pkt->cmd,
649 (blk)? blk->status : 0);
650
651 Packet * busPkt = new Packet(pkt->req,temp_cmd, -1, blkSize);
652
653 busPkt->allocate();
654
655 busPkt->time = curTick;
656
657 DPRINTF(Cache, "Sending a atomic %s for %x blk_addr: %x\n",
658 busPkt->cmdString(),
659 busPkt->getAddr() & (((ULL(1))<<48)-1),
660 busPkt->getAddr() & ~((Addr)blkSize - 1));
661
662 lat = memSidePort->sendAtomic(busPkt);
663
664 //Be sure to flip the response to a request for coherence
665 if (busPkt->needsResponse()) {
666 busPkt->makeAtomicResponse();
667 }
668
669 /* if (!(busPkt->flags & SATISFIED)) {
670 // blocked at a higher level, just return
671 return 0;
672 }
673
674 */ misses[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
675
676 CacheBlk::State old_state = (blk) ? blk->status : 0;
677 CacheBlk::State new_state = coherence->getNewState(busPkt, old_state);
678 DPRINTF(Cache, "Receive response:%s for blk addr %x in state %i\n",
679 busPkt->cmdString(),
680 busPkt->getAddr() & (((ULL(1))<<48)-1), old_state);
681 if (old_state != new_state)
682 DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
683 busPkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
684
685 tags->handleFill(blk, busPkt,
686 new_state,
687 writebacks, pkt);
688 //Free the packet
689 delete busPkt;
690
691 // Handle writebacks if needed
692 while (!writebacks.empty()){
693 Packet *wbPkt = writebacks.front();
694 memSidePort->sendAtomic(wbPkt);
695 writebacks.pop_front();
696 delete wbPkt;
697 }
698 return lat + hitLatency;
699 } else {
700 return memSidePort->sendAtomic(pkt);
701 }
702 }
703 } else {
704 // There was a cache hit.
705 // Handle writebacks if needed
706 while (!writebacks.empty()){
707 memSidePort->sendAtomic(writebacks.front());
708 writebacks.pop_front();
709 }
710
711 if (update) {
712 hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
713 } else if (pkt->isWrite()) {
714 // Still need to change data in all locations.
715 otherSidePort->sendFunctional(pkt);
716 }
717 return hitLatency;
718 }
719 fatal("Probe not handled.\n");
720 return 0;
721 }
722
723 template<class TagStore, class Buffering, class Coherence>
724 Tick
725 Cache<TagStore,Buffering,Coherence>::snoopProbe(PacketPtr &pkt)
726 {
727 Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
728 BlkType *blk = tags->findBlock(pkt);
729 MSHR *mshr = missQueue->findMSHR(blk_addr);
730 CacheBlk::State new_state = 0;
731 bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
732 if (satisfy) {
733 DPRINTF(Cache, "Cache snooped a %s request for addr %x and now supplying data,"
734 "new state is %i\n",
735 pkt->cmdString(), blk_addr, new_state);
736
737 tags->handleSnoop(blk, new_state, pkt);
738 return hitLatency;
739 }
740 if (blk) DPRINTF(Cache, "Cache snooped a %s request for addr %x, new state is %i\n",
741 pkt->cmdString(), blk_addr, new_state);
742 tags->handleSnoop(blk, new_state);
743 return 0;
744 }
745