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 if (pkt->result == Packet::Nacked) {
308 //pkt->reinitFromRequest();
309 warn("NACKs from devices not connected to the same bus not implemented\n");
310 return;
311 }
312 if (pkt->result == Packet::BadAddress) {
313 //Make the response a Bad address and send it
314 }
315 // MemDebug::cacheResponse(pkt);
316 DPRINTF(Cache, "Handling reponse to %x, blk addr: %x\n",pkt->getAddr(),
317 pkt->getAddr() & (((ULL(1))<<48)-1));
318
319 if (pkt->isCacheFill() && !pkt->isNoAllocate()) {
320 blk = tags->findBlock(pkt);
321 CacheBlk::State old_state = (blk) ? blk->status : 0;
322 PacketList writebacks;
323 CacheBlk::State new_state = coherence->getNewState(pkt,old_state);
324 if (old_state != new_state)
325 DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
326 pkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
327 blk = tags->handleFill(blk, (MSHR*)pkt->senderState,
328 new_state, writebacks, pkt);
329 while (!writebacks.empty()) {
330 missQueue->doWriteback(writebacks.front());
331 writebacks.pop_front();
332 }
333 }
334 missQueue->handleResponse(pkt, curTick + hitLatency);
335 }
336 }
337
338 template<class TagStore, class Buffering, class Coherence>
339 void
340 Cache<TagStore,Buffering,Coherence>::pseudoFill(Addr addr)
341 {
342 // Need to temporarily move this blk into MSHRs
343 MSHR *mshr = missQueue->allocateTargetList(addr);
344 int lat;
345 PacketList dummy;
346 // Read the data into the mshr
347 BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
348 assert(dummy.empty());
349 assert(mshr->pkt->flags & SATISFIED);
350 // can overload order since it isn't used on non pending blocks
351 mshr->order = blk->status;
352 // temporarily remove the block from the cache.
353 tags->invalidateBlk(addr);
354 }
355
356 template<class TagStore, class Buffering, class Coherence>
357 void
358 Cache<TagStore,Buffering,Coherence>::pseudoFill(MSHR *mshr)
359 {
360 // Need to temporarily move this blk into MSHRs
361 assert(mshr->pkt->cmd == Packet::ReadReq);
362 int lat;
363 PacketList dummy;
364 // Read the data into the mshr
365 BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
366 assert(dummy.empty());
367 assert(mshr->pkt->flags & SATISFIED);
368 // can overload order since it isn't used on non pending blocks
369 mshr->order = blk->status;
370 // temporarily remove the block from the cache.
371 tags->invalidateBlk(mshr->pkt->getAddr());
372 }
373
374
375 template<class TagStore, class Buffering, class Coherence>
376 Packet *
377 Cache<TagStore,Buffering,Coherence>::getCoherencePacket()
378 {
379 return coherence->getPacket();
380 }
381
382
383 template<class TagStore, class Buffering, class Coherence>
384 void
385 Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
386 {
387 if (pkt->req->isUncacheable()) {
388 //Can't get a hit on an uncacheable address
389 //Revisit this for multi level coherence
390 return;
391 }
392 Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
393 BlkType *blk = tags->findBlock(pkt);
394 MSHR *mshr = missQueue->findMSHR(blk_addr);
395 if (coherence->hasProtocol()) { //@todo Move this into handle bus req
396 //If we find an mshr, and it is in service, we need to NACK or invalidate
397 if (mshr) {
398 if (mshr->inService) {
399 if ((mshr->pkt->isInvalidate() || !mshr->pkt->isCacheFill())
400 && (pkt->cmd != Packet::InvalidateReq && pkt->cmd != Packet::WriteInvalidateReq)) {
401 //If the outstanding request was an invalidate (upgrade,readex,..)
402 //Then we need to ACK the request until we get the data
403 //Also NACK if the outstanding request is not a cachefill (writeback)
404 assert(!(pkt->flags & SATISFIED));
405 pkt->flags |= SATISFIED;
406 pkt->flags |= NACKED_LINE;
407 ///@todo NACK's from other levels
408 //warn("NACKs from devices not connected to the same bus not implemented\n");
409 //respondToSnoop(pkt, curTick + hitLatency);
410 return;
411 }
412 else {
413 //The supplier will be someone else, because we are waiting for
414 //the data. This should cause this cache to be forced to go to
415 //the shared state, not the exclusive even though the shared line
416 //won't be asserted. But for now we will just invlidate ourselves
417 //and allow the other cache to go into the exclusive state.
418 //@todo Make it so a read to a pending read doesn't invalidate.
419 //@todo Make it so that a read to a pending read can't be exclusive now.
420
421 //Set the address so find match works
422 //panic("Don't have invalidates yet\n");
423 invalidatePkt->addrOverride(pkt->getAddr());
424
425 //Append the invalidate on
426 missQueue->addTarget(mshr,invalidatePkt);
427 DPRINTF(Cache, "Appending Invalidate to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
428 return;
429 }
430 }
431 }
432 //We also need to check the writeback buffers and handle those
433 std::vector<MSHR *> writebacks;
434 if (missQueue->findWrites(blk_addr, writebacks)) {
435 DPRINTF(Cache, "Snoop hit in writeback to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
436
437 //Look through writebacks for any non-uncachable writes, use that
438 for (int i=0; i<writebacks.size(); i++) {
439 mshr = writebacks[i];
440
441 if (!mshr->pkt->req->isUncacheable()) {
442 if (pkt->isRead()) {
443 //Only Upgrades don't get here
444 //Supply the data
445 assert(!(pkt->flags & SATISFIED));
446 pkt->flags |= SATISFIED;
447
448 //If we are in an exclusive protocol, make it ask again
449 //to get write permissions (upgrade), signal shared
450 pkt->flags |= SHARED_LINE;
451
452 assert(pkt->isRead());
453 Addr offset = pkt->getAddr() & (blkSize - 1);
454 assert(offset < blkSize);
455 assert(pkt->getSize() <= blkSize);
456 assert(offset + pkt->getSize() <=blkSize);
457 memcpy(pkt->getPtr<uint8_t>(), mshr->pkt->getPtr<uint8_t>() + offset, pkt->getSize());
458
459 respondToSnoop(pkt, curTick + hitLatency);
460 }
461
462 if (pkt->isInvalidate()) {
463 //This must be an upgrade or other cache will take ownership
464 missQueue->markInService(mshr->pkt, mshr);
465 }
466 return;
467 }
468 }
469 }
470 }
471 CacheBlk::State new_state;
472 bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
473 if (satisfy) {
474 DPRINTF(Cache, "Cache snooped a %s request for addr %x and now supplying data,"
475 "new state is %i\n",
476 pkt->cmdString(), blk_addr, new_state);
477
478 tags->handleSnoop(blk, new_state, pkt);
479 respondToSnoop(pkt, curTick + hitLatency);
480 return;
481 }
482 if (blk) DPRINTF(Cache, "Cache snooped a %s request for addr %x, new state is %i\n",
483 pkt->cmdString(), blk_addr, new_state);
484 tags->handleSnoop(blk, new_state);
485 }
486
487 template<class TagStore, class Buffering, class Coherence>
488 void
489 Cache<TagStore,Buffering,Coherence>::snoopResponse(Packet * &pkt)
490 {
491 //Need to handle the response, if NACKED
492 if (pkt->flags & NACKED_LINE) {
493 //Need to mark it as not in service, and retry for bus
494 assert(0); //Yeah, we saw a NACK come through
495
496 //For now this should never get called, we return false when we see a NACK
497 //instead, by doing this we allow the bus_blocked mechanism to handle the retry
498 //For now it retrys in just 2 cycles, need to figure out how to change that
499 //Eventually we will want to also have success come in as a parameter
500 //Need to make sure that we handle the functionality that happens on successufl
501 //return of the sendAddr function
502 }
503 }
504
505 template<class TagStore, class Buffering, class Coherence>
506 void
507 Cache<TagStore,Buffering,Coherence>::invalidateBlk(Addr addr)
508 {
509 tags->invalidateBlk(addr);
510 }
511
512
513 /**
514 * @todo Fix to not assume write allocate
515 */
516 template<class TagStore, class Buffering, class Coherence>
517 Tick
518 Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort* otherSidePort)
519 {
520 // MemDebug::cacheProbe(pkt);
521 if (!pkt->req->isUncacheable()) {
522 if (pkt->isInvalidate() && !pkt->isRead()
523 && !pkt->isWrite()) {
524 //Upgrade or Invalidate, satisfy it, don't forward
525 DPRINTF(Cache, "%s %x ? blk_addr: %x\n", pkt->cmdString(),
526 pkt->getAddr() & (((ULL(1))<<48)-1),
527 pkt->getAddr() & ~((Addr)blkSize - 1));
528 pkt->flags |= SATISFIED;
529 return 0;
530 }
531 }
532
533 PacketList writebacks;
534 int lat;
535 BlkType *blk = tags->handleAccess(pkt, lat, writebacks, update);
536
537 DPRINTF(Cache, "%s %x %s blk_addr: %x\n", pkt->cmdString(),
538 pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss",
539 pkt->getAddr() & ~((Addr)blkSize - 1));
540
541 if (!blk) {
542 // Need to check for outstanding misses and writes
543 Addr blk_addr = pkt->getAddr() & ~(blkSize - 1);
544
545 // There can only be one matching outstanding miss.
546 MSHR* mshr = missQueue->findMSHR(blk_addr);
547
548 // There can be many matching outstanding writes.
549 std::vector<MSHR*> writes;
550 missQueue->findWrites(blk_addr, writes);
551
552 if (!update) {
553 otherSidePort->sendFunctional(pkt);
554
555 // Check for data in MSHR and writebuffer.
556 if (mshr) {
557 warn("Found outstanding miss on an non-update probe");
558 MSHR::TargetList *targets = mshr->getTargetList();
559 MSHR::TargetList::iterator i = targets->begin();
560 MSHR::TargetList::iterator end = targets->end();
561 for (; i != end; ++i) {
562 Packet * target = *i;
563 // If the target contains data, and it overlaps the
564 // probed request, need to update data
565 if (target->isWrite() && target->intersect(pkt)) {
566 uint8_t* pkt_data;
567 uint8_t* write_data;
568 int data_size;
569 if (target->getAddr() < pkt->getAddr()) {
570 int offset = pkt->getAddr() - target->getAddr();
571 pkt_data = pkt->getPtr<uint8_t>();
572 write_data = target->getPtr<uint8_t>() + offset;
573 data_size = target->getSize() - offset;
574 assert(data_size > 0);
575 if (data_size > pkt->getSize())
576 data_size = pkt->getSize();
577 } else {
578 int offset = target->getAddr() - pkt->getAddr();
579 pkt_data = pkt->getPtr<uint8_t>() + offset;
580 write_data = target->getPtr<uint8_t>();
581 data_size = pkt->getSize() - offset;
582 assert(data_size > pkt->getSize());
583 if (data_size > target->getSize())
584 data_size = target->getSize();
585 }
586
587 if (pkt->isWrite()) {
588 memcpy(pkt_data, write_data, data_size);
589 } else {
590 memcpy(write_data, pkt_data, data_size);
591 }
592 }
593 }
594 }
595 for (int i = 0; i < writes.size(); ++i) {
596 Packet * write = writes[i]->pkt;
597 if (write->intersect(pkt)) {
598 warn("Found outstanding write on an non-update probe");
599 uint8_t* pkt_data;
600 uint8_t* write_data;
601 int data_size;
602 if (write->getAddr() < pkt->getAddr()) {
603 int offset = pkt->getAddr() - write->getAddr();
604 pkt_data = pkt->getPtr<uint8_t>();
605 write_data = write->getPtr<uint8_t>() + offset;
606 data_size = write->getSize() - offset;
607 assert(data_size > 0);
608 if (data_size > pkt->getSize())
609 data_size = pkt->getSize();
610 } else {
611 int offset = write->getAddr() - pkt->getAddr();
612 pkt_data = pkt->getPtr<uint8_t>() + offset;
613 write_data = write->getPtr<uint8_t>();
614 data_size = pkt->getSize() - offset;
615 assert(data_size > pkt->getSize());
616 if (data_size > write->getSize())
617 data_size = write->getSize();
618 }
619
620 if (pkt->isWrite()) {
621 memcpy(pkt_data, write_data, data_size);
622 } else {
623 memcpy(write_data, pkt_data, data_size);
624 }
625
626 }
627 }
628 return 0;
629 } else {
630 // update the cache state and statistics
631 if (mshr || !writes.empty()){
632 // Can't handle it, return pktuest unsatisfied.
633 panic("Atomic access ran into outstanding MSHR's or WB's!");
634 }
635 if (!pkt->req->isUncacheable()) {
636 // Fetch the cache block to fill
637 BlkType *blk = tags->findBlock(pkt);
638 Packet::Command temp_cmd = coherence->getBusCmd(pkt->cmd,
639 (blk)? blk->status : 0);
640
641 Packet * busPkt = new Packet(pkt->req,temp_cmd, -1, blkSize);
642
643 busPkt->allocate();
644
645 busPkt->time = curTick;
646
647 DPRINTF(Cache, "Sending a atomic %s for %x blk_addr: %x\n",
648 busPkt->cmdString(),
649 busPkt->getAddr() & (((ULL(1))<<48)-1),
650 busPkt->getAddr() & ~((Addr)blkSize - 1));
651
652 lat = memSidePort->sendAtomic(busPkt);
653
654 //Be sure to flip the response to a request for coherence
655 if (busPkt->needsResponse()) {
656 busPkt->makeAtomicResponse();
657 }
658
659 /* if (!(busPkt->flags & SATISFIED)) {
660 // blocked at a higher level, just return
661 return 0;
662 }
663
664 */ misses[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
665
666 CacheBlk::State old_state = (blk) ? blk->status : 0;
667 CacheBlk::State new_state = coherence->getNewState(busPkt, old_state);
668 DPRINTF(Cache, "Receive response:%s for blk addr %x in state %i\n",
669 busPkt->cmdString(),
670 busPkt->getAddr() & (((ULL(1))<<48)-1), old_state);
671 if (old_state != new_state)
672 DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
673 busPkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
674
675 tags->handleFill(blk, busPkt,
676 new_state,
677 writebacks, pkt);
678 //Free the packet
679 delete busPkt;
680
681 // Handle writebacks if needed
682 while (!writebacks.empty()){
683 Packet *wbPkt = writebacks.front();
684 memSidePort->sendAtomic(wbPkt);
685 writebacks.pop_front();
686 delete wbPkt;
687 }
688 return lat + hitLatency;
689 } else {
690 return memSidePort->sendAtomic(pkt);
691 }
692 }
693 } else {
694 // There was a cache hit.
695 // Handle writebacks if needed
696 while (!writebacks.empty()){
697 memSidePort->sendAtomic(writebacks.front());
698 writebacks.pop_front();
699 }
700
701 if (update) {
702 hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
703 } else if (pkt->isWrite()) {
704 // Still need to change data in all locations.
705 otherSidePort->sendFunctional(pkt);
706 }
707 return hitLatency;
708 }
709 fatal("Probe not handled.\n");
710 return 0;
711 }
712
713 template<class TagStore, class Buffering, class Coherence>
714 Tick
715 Cache<TagStore,Buffering,Coherence>::snoopProbe(PacketPtr &pkt)
716 {
717 Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
718 BlkType *blk = tags->findBlock(pkt);
719 MSHR *mshr = missQueue->findMSHR(blk_addr);
720 CacheBlk::State new_state = 0;
721 bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
722 if (satisfy) {
723 DPRINTF(Cache, "Cache snooped a %s request for addr %x and now supplying data,"
724 "new state is %i\n",
725 pkt->cmdString(), blk_addr, new_state);
726
727 tags->handleSnoop(blk, new_state, pkt);
728 return hitLatency;
729 }
730 if (blk) DPRINTF(Cache, "Cache snooped a %s request for addr %x, new state is %i\n",
731 pkt->cmdString(), blk_addr, new_state);
732 tags->handleSnoop(blk, new_state);
733 return 0;
734 }
735