mem-cache: Refactor the cache recvTimingReq function
[gem5.git] / src / mem / cache / cache.hh
1 /*
2 * Copyright (c) 2012-2018 ARM Limited
3 * All rights reserved.
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2002-2005 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Erik Hallnor
41 * Dave Greene
42 * Steve Reinhardt
43 * Ron Dreslinski
44 * Andreas Hansson
45 */
46
47 /**
48 * @file
49 * Describes a cache based on template policies.
50 */
51
52 #ifndef __MEM_CACHE_CACHE_HH__
53 #define __MEM_CACHE_CACHE_HH__
54
55 #include <unordered_set>
56
57 #include "base/logging.hh" // fatal, panic, and warn
58 #include "enums/Clusivity.hh"
59 #include "mem/cache/base.hh"
60 #include "mem/cache/blk.hh"
61 #include "mem/cache/mshr.hh"
62 #include "mem/cache/tags/base.hh"
63 #include "params/Cache.hh"
64 #include "sim/eventq.hh"
65
66 //Forward decleration
67 class BasePrefetcher;
68
69 /**
70 * A template-policy based cache. The behavior of the cache can be altered by
71 * supplying different template policies. TagStore handles all tag and data
72 * storage @sa TagStore, \ref gem5MemorySystem "gem5 Memory System"
73 */
74 class Cache : public BaseCache
75 {
76 protected:
77
78 /**
79 * The CPU-side port extends the base cache slave port with access
80 * functions for functional, atomic and timing requests.
81 */
82 class CpuSidePort : public CacheSlavePort
83 {
84 private:
85
86 // a pointer to our specific cache implementation
87 Cache *cache;
88
89 protected:
90
91 virtual bool recvTimingSnoopResp(PacketPtr pkt);
92
93 virtual bool tryTiming(PacketPtr pkt);
94
95 virtual bool recvTimingReq(PacketPtr pkt);
96
97 virtual Tick recvAtomic(PacketPtr pkt);
98
99 virtual void recvFunctional(PacketPtr pkt);
100
101 virtual AddrRangeList getAddrRanges() const;
102
103 public:
104
105 CpuSidePort(const std::string &_name, Cache *_cache,
106 const std::string &_label);
107
108 };
109
110 /**
111 * Override the default behaviour of sendDeferredPacket to enable
112 * the memory-side cache port to also send requests based on the
113 * current MSHR status. This queue has a pointer to our specific
114 * cache implementation and is used by the MemSidePort.
115 */
116 class CacheReqPacketQueue : public ReqPacketQueue
117 {
118
119 protected:
120
121 Cache &cache;
122 SnoopRespPacketQueue &snoopRespQueue;
123
124 public:
125
126 CacheReqPacketQueue(Cache &cache, MasterPort &port,
127 SnoopRespPacketQueue &snoop_resp_queue,
128 const std::string &label) :
129 ReqPacketQueue(cache, port, label), cache(cache),
130 snoopRespQueue(snoop_resp_queue) { }
131
132 /**
133 * Override the normal sendDeferredPacket and do not only
134 * consider the transmit list (used for responses), but also
135 * requests.
136 */
137 virtual void sendDeferredPacket();
138
139 /**
140 * Check if there is a conflicting snoop response about to be
141 * send out, and if so simply stall any requests, and schedule
142 * a send event at the same time as the next snoop response is
143 * being sent out.
144 */
145 bool checkConflictingSnoop(Addr addr)
146 {
147 if (snoopRespQueue.hasAddr(addr)) {
148 DPRINTF(CachePort, "Waiting for snoop response to be "
149 "sent\n");
150 Tick when = snoopRespQueue.deferredPacketReadyTime();
151 schedSendEvent(when);
152 return true;
153 }
154 return false;
155 }
156 };
157
158 /**
159 * The memory-side port extends the base cache master port with
160 * access functions for functional, atomic and timing snoops.
161 */
162 class MemSidePort : public CacheMasterPort
163 {
164 private:
165
166 /** The cache-specific queue. */
167 CacheReqPacketQueue _reqQueue;
168
169 SnoopRespPacketQueue _snoopRespQueue;
170
171 // a pointer to our specific cache implementation
172 Cache *cache;
173
174 protected:
175
176 virtual void recvTimingSnoopReq(PacketPtr pkt);
177
178 virtual bool recvTimingResp(PacketPtr pkt);
179
180 virtual Tick recvAtomicSnoop(PacketPtr pkt);
181
182 virtual void recvFunctionalSnoop(PacketPtr pkt);
183
184 public:
185
186 MemSidePort(const std::string &_name, Cache *_cache,
187 const std::string &_label);
188 };
189
190 /** Tag and data Storage */
191 BaseTags *tags;
192
193 /** Prefetcher */
194 BasePrefetcher *prefetcher;
195
196 /** Temporary cache block for occasional transitory use */
197 CacheBlk *tempBlock;
198
199 /**
200 * This cache should allocate a block on a line-sized write miss.
201 */
202 const bool doFastWrites;
203
204 /**
205 * Turn line-sized writes into WriteInvalidate transactions.
206 */
207 void promoteWholeLineWrites(PacketPtr pkt);
208
209 /**
210 * Notify the prefetcher on every access, not just misses.
211 */
212 const bool prefetchOnAccess;
213
214 /**
215 * Clusivity with respect to the upstream cache, determining if we
216 * fill into both this cache and the cache above on a miss. Note
217 * that we currently do not support strict clusivity policies.
218 */
219 const Enums::Clusivity clusivity;
220
221 /**
222 * Determine if clean lines should be written back or not. In
223 * cases where a downstream cache is mostly inclusive we likely
224 * want it to act as a victim cache also for lines that have not
225 * been modified. Hence, we cannot simply drop the line (or send a
226 * clean evict), but rather need to send the actual data.
227 */
228 const bool writebackClean;
229
230 /**
231 * Upstream caches need this packet until true is returned, so
232 * hold it for deletion until a subsequent call
233 */
234 std::unique_ptr<Packet> pendingDelete;
235
236 /**
237 * Writebacks from the tempBlock, resulting on the response path
238 * in atomic mode, must happen after the call to recvAtomic has
239 * finished (for the right ordering of the packets). We therefore
240 * need to hold on to the packets, and have a method and an event
241 * to send them.
242 */
243 PacketPtr tempBlockWriteback;
244
245 /**
246 * Send the outstanding tempBlock writeback. To be called after
247 * recvAtomic finishes in cases where the block we filled is in
248 * fact the tempBlock, and now needs to be written back.
249 */
250 void writebackTempBlockAtomic() {
251 assert(tempBlockWriteback != nullptr);
252 PacketList writebacks{tempBlockWriteback};
253 doWritebacksAtomic(writebacks);
254 tempBlockWriteback = nullptr;
255 }
256
257 /**
258 * An event to writeback the tempBlock after recvAtomic
259 * finishes. To avoid other calls to recvAtomic getting in
260 * between, we create this event with a higher priority.
261 */
262 EventFunctionWrapper writebackTempBlockAtomicEvent;
263
264 /**
265 * Store the outstanding requests that we are expecting snoop
266 * responses from so we can determine which snoop responses we
267 * generated and which ones were merely forwarded.
268 */
269 std::unordered_set<RequestPtr> outstandingSnoop;
270
271 /**
272 * Does all the processing necessary to perform the provided request.
273 * @param pkt The memory request to perform.
274 * @param blk The cache block to be updated.
275 * @param lat The latency of the access.
276 * @param writebacks List for any writebacks that need to be performed.
277 * @return Boolean indicating whether the request was satisfied.
278 */
279 bool access(PacketPtr pkt, CacheBlk *&blk,
280 Cycles &lat, PacketList &writebacks);
281
282 /**
283 *Handle doing the Compare and Swap function for SPARC.
284 */
285 void cmpAndSwap(CacheBlk *blk, PacketPtr pkt);
286
287 /**
288 * Find a block frame for new block at address addr targeting the
289 * given security space, assuming that the block is not currently
290 * in the cache. Append writebacks if any to provided packet
291 * list. Return free block frame. May return nullptr if there are
292 * no replaceable blocks at the moment.
293 */
294 CacheBlk *allocateBlock(Addr addr, bool is_secure, PacketList &writebacks);
295
296 /**
297 * Invalidate a cache block.
298 *
299 * @param blk Block to invalidate
300 */
301 void invalidateBlock(CacheBlk *blk);
302
303 /**
304 * Maintain the clusivity of this cache by potentially
305 * invalidating a block. This method works in conjunction with
306 * satisfyRequest, but is separate to allow us to handle all MSHR
307 * targets before potentially dropping a block.
308 *
309 * @param from_cache Whether we have dealt with a packet from a cache
310 * @param blk The block that should potentially be dropped
311 */
312 void maintainClusivity(bool from_cache, CacheBlk *blk);
313
314 /**
315 * Populates a cache block and handles all outstanding requests for the
316 * satisfied fill request. This version takes two memory requests. One
317 * contains the fill data, the other is an optional target to satisfy.
318 * @param pkt The memory request with the fill data.
319 * @param blk The cache block if it already exists.
320 * @param writebacks List for any writebacks that need to be performed.
321 * @param allocate Whether to allocate a block or use the temp block
322 * @return Pointer to the new cache block.
323 */
324 CacheBlk *handleFill(PacketPtr pkt, CacheBlk *blk,
325 PacketList &writebacks, bool allocate);
326
327 /**
328 * Determine whether we should allocate on a fill or not. If this
329 * cache is mostly inclusive with regards to the upstream cache(s)
330 * we always allocate (for any non-forwarded and cacheable
331 * requests). In the case of a mostly exclusive cache, we allocate
332 * on fill if the packet did not come from a cache, thus if we:
333 * are dealing with a whole-line write (the latter behaves much
334 * like a writeback), the original target packet came from a
335 * non-caching source, or if we are performing a prefetch or LLSC.
336 *
337 * @param cmd Command of the incoming requesting packet
338 * @return Whether we should allocate on the fill
339 */
340 inline bool allocOnFill(MemCmd cmd) const override
341 {
342 return clusivity == Enums::mostly_incl ||
343 cmd == MemCmd::WriteLineReq ||
344 cmd == MemCmd::ReadReq ||
345 cmd == MemCmd::WriteReq ||
346 cmd.isPrefetch() ||
347 cmd.isLLSC();
348 }
349
350 /*
351 * Handle a timing request that hit in the cache
352 *
353 * @param ptk The request packet
354 * @param blk The referenced block
355 * @param request_time The tick at which the block lookup is compete
356 */
357 void handleTimingReqHit(PacketPtr pkt, CacheBlk *blk, Tick request_time);
358
359 /*
360 * Handle a timing request that missed in the cache
361 *
362 * @param ptk The request packet
363 * @param blk The referenced block
364 * @param forward_time The tick at which we can process dependent requests
365 * @param request_time The tick at which the block lookup is compete
366 */
367 void handleTimingReqMiss(PacketPtr pkt, CacheBlk *blk, Tick forward_time,
368 Tick request_time);
369
370 /**
371 * Performs the access specified by the request.
372 * @param pkt The request to perform.
373 */
374 void recvTimingReq(PacketPtr pkt);
375
376 /**
377 * Insert writebacks into the write buffer
378 */
379 void doWritebacks(PacketList& writebacks, Tick forward_time);
380
381 /**
382 * Send writebacks down the memory hierarchy in atomic mode
383 */
384 void doWritebacksAtomic(PacketList& writebacks);
385
386 /**
387 * Handling the special case of uncacheable write responses to
388 * make recvTimingResp less cluttered.
389 */
390 void handleUncacheableWriteResp(PacketPtr pkt);
391
392 /**
393 * Service non-deferred MSHR targets using the received response
394 *
395 * Iterates through the list of targets that can be serviced with
396 * the current response. Any writebacks that need to performed
397 * must be appended to the writebacks parameter.
398 *
399 * @param mshr The MSHR that corresponds to the reponse
400 * @param pkt The response packet
401 * @param blk The reference block
402 * @param writebacks List of writebacks that need to be performed
403 */
404 void serviceMSHRTargets(MSHR *mshr, const PacketPtr pkt, CacheBlk *blk,
405 PacketList& writebacks);
406
407 /**
408 * Handles a response (cache line fill/write ack) from the bus.
409 * @param pkt The response packet
410 */
411 void recvTimingResp(PacketPtr pkt);
412
413 /**
414 * Snoops bus transactions to maintain coherence.
415 * @param pkt The current bus transaction.
416 */
417 void recvTimingSnoopReq(PacketPtr pkt);
418
419 /**
420 * Handle a snoop response.
421 * @param pkt Snoop response packet
422 */
423 void recvTimingSnoopResp(PacketPtr pkt);
424
425 /**
426 * Performs the access specified by the request.
427 * @param pkt The request to perform.
428 * @return The number of ticks required for the access.
429 */
430 Tick recvAtomic(PacketPtr pkt);
431
432 /**
433 * Snoop for the provided request in the cache and return the estimated
434 * time taken.
435 * @param pkt The memory request to snoop
436 * @return The number of ticks required for the snoop.
437 */
438 Tick recvAtomicSnoop(PacketPtr pkt);
439
440 /**
441 * Performs the access specified by the request.
442 * @param pkt The request to perform.
443 * @param fromCpuSide from the CPU side port or the memory side port
444 */
445 void functionalAccess(PacketPtr pkt, bool fromCpuSide);
446
447 /**
448 * Perform any necessary updates to the block and perform any data
449 * exchange between the packet and the block. The flags of the
450 * packet are also set accordingly.
451 *
452 * @param pkt Request packet from upstream that hit a block
453 * @param blk Cache block that the packet hit
454 * @param deferred_response Whether this hit is to block that
455 * originally missed
456 * @param pending_downgrade Whether the writable flag is to be removed
457 *
458 * @return True if the block is to be invalidated
459 */
460 void satisfyRequest(PacketPtr pkt, CacheBlk *blk,
461 bool deferred_response = false,
462 bool pending_downgrade = false);
463
464 void doTimingSupplyResponse(PacketPtr req_pkt, const uint8_t *blk_data,
465 bool already_copied, bool pending_inval);
466
467 /**
468 * Perform an upward snoop if needed, and update the block state
469 * (possibly invalidating the block). Also create a response if required.
470 *
471 * @param pkt Snoop packet
472 * @param blk Cache block being snooped
473 * @param is_timing Timing or atomic for the response
474 * @param is_deferred Is this a deferred snoop or not?
475 * @param pending_inval Do we have a pending invalidation?
476 *
477 * @return The snoop delay incurred by the upwards snoop
478 */
479 uint32_t handleSnoop(PacketPtr pkt, CacheBlk *blk,
480 bool is_timing, bool is_deferred, bool pending_inval);
481
482 /**
483 * Create a writeback request for the given block.
484 * @param blk The block to writeback.
485 * @return The writeback request for the block.
486 */
487 PacketPtr writebackBlk(CacheBlk *blk);
488
489 /**
490 * Create a writeclean request for the given block.
491 * @param blk The block to write clean
492 * @param dest The destination of this clean operation
493 * @return The write clean packet for the block.
494 */
495 PacketPtr writecleanBlk(CacheBlk *blk, Request::Flags dest, PacketId id);
496
497 /**
498 * Create a CleanEvict request for the given block.
499 * @param blk The block to evict.
500 * @return The CleanEvict request for the block.
501 */
502 PacketPtr cleanEvictBlk(CacheBlk *blk);
503
504
505 void memWriteback() override;
506 void memInvalidate() override;
507 bool isDirty() const override;
508
509 /**
510 * Cache block visitor that writes back dirty cache blocks using
511 * functional writes.
512 *
513 * \return Always returns true.
514 */
515 bool writebackVisitor(CacheBlk &blk);
516 /**
517 * Cache block visitor that invalidates all blocks in the cache.
518 *
519 * @warn Dirty cache lines will not be written back to memory.
520 *
521 * \return Always returns true.
522 */
523 bool invalidateVisitor(CacheBlk &blk);
524
525 /**
526 * Create an appropriate downstream bus request packet for the
527 * given parameters.
528 * @param cpu_pkt The miss that needs to be satisfied.
529 * @param blk The block currently in the cache corresponding to
530 * cpu_pkt (nullptr if none).
531 * @param needsWritable Indicates that the block must be writable
532 * even if the request in cpu_pkt doesn't indicate that.
533 * @return A new Packet containing the request, or nullptr if the
534 * current request in cpu_pkt should just be forwarded on.
535 */
536 PacketPtr createMissPacket(PacketPtr cpu_pkt, CacheBlk *blk,
537 bool needsWritable) const;
538
539 /**
540 * Return the next queue entry to service, either a pending miss
541 * from the MSHR queue, a buffered write from the write buffer, or
542 * something from the prefetcher. This function is responsible
543 * for prioritizing among those sources on the fly.
544 */
545 QueueEntry* getNextQueueEntry();
546
547 /**
548 * Send up a snoop request and find cached copies. If cached copies are
549 * found, set the BLOCK_CACHED flag in pkt.
550 */
551 bool isCachedAbove(PacketPtr pkt, bool is_timing = true) const;
552
553 /**
554 * Return whether there are any outstanding misses.
555 */
556 bool outstandingMisses() const
557 {
558 return !mshrQueue.isEmpty();
559 }
560
561 CacheBlk *findBlock(Addr addr, bool is_secure) const {
562 return tags->findBlock(addr, is_secure);
563 }
564
565 bool inCache(Addr addr, bool is_secure) const override {
566 return (tags->findBlock(addr, is_secure) != 0);
567 }
568
569 bool inMissQueue(Addr addr, bool is_secure) const override {
570 return (mshrQueue.findMatch(addr, is_secure) != 0);
571 }
572
573 /**
574 * Find next request ready time from among possible sources.
575 */
576 Tick nextQueueReadyTime() const;
577
578 public:
579 /** Instantiates a basic cache object. */
580 Cache(const CacheParams *p);
581
582 /** Non-default destructor is needed to deallocate memory. */
583 virtual ~Cache();
584
585 void regStats() override;
586
587 /**
588 * Take an MSHR, turn it into a suitable downstream packet, and
589 * send it out. This construct allows a queue entry to choose a suitable
590 * approach based on its type.
591 *
592 * @param mshr The MSHR to turn into a packet and send
593 * @return True if the port is waiting for a retry
594 */
595 bool sendMSHRQueuePacket(MSHR* mshr);
596
597 /**
598 * Similar to sendMSHR, but for a write-queue entry
599 * instead. Create the packet, and send it, and if successful also
600 * mark the entry in service.
601 *
602 * @param wq_entry The write-queue entry to turn into a packet and send
603 * @return True if the port is waiting for a retry
604 */
605 bool sendWriteQueuePacket(WriteQueueEntry* wq_entry);
606
607 /** serialize the state of the caches
608 * We currently don't support checkpointing cache state, so this panics.
609 */
610 void serialize(CheckpointOut &cp) const override;
611 void unserialize(CheckpointIn &cp) override;
612 };
613
614 /**
615 * Wrap a method and present it as a cache block visitor.
616 *
617 * For example the forEachBlk method in the tag arrays expects a
618 * callable object/function as their parameter. This class wraps a
619 * method in an object and presents callable object that adheres to
620 * the cache block visitor protocol.
621 */
622 class CacheBlkVisitorWrapper : public CacheBlkVisitor
623 {
624 public:
625 typedef bool (Cache::*VisitorPtr)(CacheBlk &blk);
626
627 CacheBlkVisitorWrapper(Cache &_cache, VisitorPtr _visitor)
628 : cache(_cache), visitor(_visitor) {}
629
630 bool operator()(CacheBlk &blk) override {
631 return (cache.*visitor)(blk);
632 }
633
634 private:
635 Cache &cache;
636 VisitorPtr visitor;
637 };
638
639 /**
640 * Cache block visitor that determines if there are dirty blocks in a
641 * cache.
642 *
643 * Use with the forEachBlk method in the tag array to determine if the
644 * array contains dirty blocks.
645 */
646 class CacheBlkIsDirtyVisitor : public CacheBlkVisitor
647 {
648 public:
649 CacheBlkIsDirtyVisitor()
650 : _isDirty(false) {}
651
652 bool operator()(CacheBlk &blk) override {
653 if (blk.isDirty()) {
654 _isDirty = true;
655 return false;
656 } else {
657 return true;
658 }
659 }
660
661 /**
662 * Does the array contain a dirty line?
663 *
664 * \return true if yes, false otherwise.
665 */
666 bool isDirty() const { return _isDirty; };
667
668 private:
669 bool _isDirty;
670 };
671
672 #endif // __MEM_CACHE_CACHE_HH__