base.hh (10713:eddb533708cb) base.hh (10714:9ba5e70964a4)
1/*
2 * Copyright (c) 2012-2013, 2015 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) 2003-2005 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 * Steve Reinhardt
42 * Ron Dreslinski
43 */
44
45/**
46 * @file
47 * Declares a basic cache interface BaseCache.
48 */
49
50#ifndef __BASE_CACHE_HH__
51#define __BASE_CACHE_HH__
52
53#include <algorithm>
54#include <list>
55#include <string>
56#include <vector>
57
58#include "base/misc.hh"
59#include "base/statistics.hh"
60#include "base/trace.hh"
61#include "base/types.hh"
62#include "debug/Cache.hh"
63#include "debug/CachePort.hh"
64#include "mem/cache/mshr_queue.hh"
65#include "mem/mem_object.hh"
66#include "mem/packet.hh"
67#include "mem/qport.hh"
68#include "mem/request.hh"
69#include "params/BaseCache.hh"
70#include "sim/eventq.hh"
71#include "sim/full_system.hh"
72#include "sim/sim_exit.hh"
73#include "sim/system.hh"
74
75class MSHR;
76/**
77 * A basic cache interface. Implements some common functions for speed.
78 */
79class BaseCache : public MemObject
80{
81 /**
82 * Indexes to enumerate the MSHR queues.
83 */
84 enum MSHRQueueIndex {
85 MSHRQueue_MSHRs,
86 MSHRQueue_WriteBuffer
87 };
88
89 public:
90 /**
91 * Reasons for caches to be blocked.
92 */
93 enum BlockedCause {
94 Blocked_NoMSHRs = MSHRQueue_MSHRs,
95 Blocked_NoWBBuffers = MSHRQueue_WriteBuffer,
96 Blocked_NoTargets,
97 NUM_BLOCKED_CAUSES
98 };
99
100 /**
101 * Reasons for cache to request a bus.
102 */
103 enum RequestCause {
104 Request_MSHR = MSHRQueue_MSHRs,
105 Request_WB = MSHRQueue_WriteBuffer,
106 Request_PF,
107 NUM_REQUEST_CAUSES
108 };
109
110 protected:
111
112 /**
113 * A cache master port is used for the memory-side port of the
114 * cache, and in addition to the basic timing port that only sends
115 * response packets through a transmit list, it also offers the
116 * ability to schedule and send request packets (requests &
117 * writebacks). The send event is scheduled through requestBus,
118 * and the sendDeferredPacket of the timing port is modified to
119 * consider both the transmit list and the requests from the MSHR.
120 */
121 class CacheMasterPort : public QueuedMasterPort
122 {
123
124 public:
125
126 /**
127 * Schedule a send of a request packet (from the MSHR). Note
128 * that we could already have a retry outstanding.
129 */
130 void requestBus(RequestCause cause, Tick time)
131 {
1/*
2 * Copyright (c) 2012-2013, 2015 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) 2003-2005 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 * Steve Reinhardt
42 * Ron Dreslinski
43 */
44
45/**
46 * @file
47 * Declares a basic cache interface BaseCache.
48 */
49
50#ifndef __BASE_CACHE_HH__
51#define __BASE_CACHE_HH__
52
53#include <algorithm>
54#include <list>
55#include <string>
56#include <vector>
57
58#include "base/misc.hh"
59#include "base/statistics.hh"
60#include "base/trace.hh"
61#include "base/types.hh"
62#include "debug/Cache.hh"
63#include "debug/CachePort.hh"
64#include "mem/cache/mshr_queue.hh"
65#include "mem/mem_object.hh"
66#include "mem/packet.hh"
67#include "mem/qport.hh"
68#include "mem/request.hh"
69#include "params/BaseCache.hh"
70#include "sim/eventq.hh"
71#include "sim/full_system.hh"
72#include "sim/sim_exit.hh"
73#include "sim/system.hh"
74
75class MSHR;
76/**
77 * A basic cache interface. Implements some common functions for speed.
78 */
79class BaseCache : public MemObject
80{
81 /**
82 * Indexes to enumerate the MSHR queues.
83 */
84 enum MSHRQueueIndex {
85 MSHRQueue_MSHRs,
86 MSHRQueue_WriteBuffer
87 };
88
89 public:
90 /**
91 * Reasons for caches to be blocked.
92 */
93 enum BlockedCause {
94 Blocked_NoMSHRs = MSHRQueue_MSHRs,
95 Blocked_NoWBBuffers = MSHRQueue_WriteBuffer,
96 Blocked_NoTargets,
97 NUM_BLOCKED_CAUSES
98 };
99
100 /**
101 * Reasons for cache to request a bus.
102 */
103 enum RequestCause {
104 Request_MSHR = MSHRQueue_MSHRs,
105 Request_WB = MSHRQueue_WriteBuffer,
106 Request_PF,
107 NUM_REQUEST_CAUSES
108 };
109
110 protected:
111
112 /**
113 * A cache master port is used for the memory-side port of the
114 * cache, and in addition to the basic timing port that only sends
115 * response packets through a transmit list, it also offers the
116 * ability to schedule and send request packets (requests &
117 * writebacks). The send event is scheduled through requestBus,
118 * and the sendDeferredPacket of the timing port is modified to
119 * consider both the transmit list and the requests from the MSHR.
120 */
121 class CacheMasterPort : public QueuedMasterPort
122 {
123
124 public:
125
126 /**
127 * Schedule a send of a request packet (from the MSHR). Note
128 * that we could already have a retry outstanding.
129 */
130 void requestBus(RequestCause cause, Tick time)
131 {
132 DPRINTF(CachePort, "Asserting bus request for cause %d\n", cause);
132 DPRINTF(CachePort, "Scheduling request at %llu due to %d\n",
133 time, cause);
133 reqQueue.schedSendEvent(time);
134 }
135
136 protected:
137
138 CacheMasterPort(const std::string &_name, BaseCache *_cache,
139 ReqPacketQueue &_reqQueue,
140 SnoopRespPacketQueue &_snoopRespQueue) :
141 QueuedMasterPort(_name, _cache, _reqQueue, _snoopRespQueue)
142 { }
143
144 /**
145 * Memory-side port always snoops.
146 *
147 * @return always true
148 */
149 virtual bool isSnooping() const { return true; }
150 };
151
152 /**
153 * A cache slave port is used for the CPU-side port of the cache,
154 * and it is basically a simple timing port that uses a transmit
155 * list for responses to the CPU (or connected master). In
156 * addition, it has the functionality to block the port for
157 * incoming requests. If blocked, the port will issue a retry once
158 * unblocked.
159 */
160 class CacheSlavePort : public QueuedSlavePort
161 {
162
163 public:
164
165 /** Do not accept any new requests. */
166 void setBlocked();
167
168 /** Return to normal operation and accept new requests. */
169 void clearBlocked();
170
171 bool isBlocked() const { return blocked; }
172
173 protected:
174
175 CacheSlavePort(const std::string &_name, BaseCache *_cache,
176 const std::string &_label);
177
178 /** A normal packet queue used to store responses. */
179 RespPacketQueue queue;
180
181 bool blocked;
182
183 bool mustSendRetry;
184
185 private:
186
187 void processSendRetry();
188
189 EventWrapper<CacheSlavePort,
190 &CacheSlavePort::processSendRetry> sendRetryEvent;
191
192 };
193
194 CacheSlavePort *cpuSidePort;
195 CacheMasterPort *memSidePort;
196
197 protected:
198
199 /** Miss status registers */
200 MSHRQueue mshrQueue;
201
202 /** Write/writeback buffer */
203 MSHRQueue writeBuffer;
204
205 /**
206 * Allocate a buffer, passing the time indicating when schedule an
207 * event to the queued port to go and ask the MSHR and write queue
208 * if they have packets to send.
209 *
210 * allocateBufferInternal() function is called in:
211 * - MSHR allocateWriteBuffer (unchached write forwarded to WriteBuffer);
212 * - MSHR allocateMissBuffer (cacheable miss in MSHR queue);
213 * - MSHR allocateUncachedReadBuffer (unchached read allocated in MSHR
214 * queue)
215 */
216 MSHR *allocateBufferInternal(MSHRQueue *mq, Addr addr, int size,
217 PacketPtr pkt, Tick time, bool requestBus)
218 {
219 MSHR *mshr = mq->allocate(addr, size, pkt, time, order++);
220
221 if (mq->isFull()) {
222 setBlocked((BlockedCause)mq->index);
223 }
224
225 if (requestBus) {
226 requestMemSideBus((RequestCause)mq->index, time);
227 }
228
229 return mshr;
230 }
231
232 void markInServiceInternal(MSHR *mshr, bool pending_dirty_resp)
233 {
234 MSHRQueue *mq = mshr->queue;
235 bool wasFull = mq->isFull();
236 mq->markInService(mshr, pending_dirty_resp);
237 if (wasFull && !mq->isFull()) {
238 clearBlocked((BlockedCause)mq->index);
239 }
240 }
241
242 /**
243 * Write back dirty blocks in the cache using functional accesses.
244 */
245 virtual void memWriteback() = 0;
246 /**
247 * Invalidates all blocks in the cache.
248 *
249 * @warn Dirty cache lines will not be written back to
250 * memory. Make sure to call functionalWriteback() first if you
251 * want the to write them to memory.
252 */
253 virtual void memInvalidate() = 0;
254 /**
255 * Determine if there are any dirty blocks in the cache.
256 *
257 * \return true if at least one block is dirty, false otherwise.
258 */
259 virtual bool isDirty() const = 0;
260
261 /** Block size of this cache */
262 const unsigned blkSize;
263
264 /**
265 * The latency of tag lookup of a cache. It occurs when there is
266 * an access to the cache.
267 */
268 const Cycles lookupLatency;
269
270 /**
271 * This is the forward latency of the cache. It occurs when there
272 * is a cache miss and a request is forwarded downstream, in
273 * particular an outbound miss.
274 */
275 const Cycles forwardLatency;
276
277 /** The latency to fill a cache block */
278 const Cycles fillLatency;
279
280 /**
281 * The latency of sending reponse to its upper level cache/core on
282 * a linefill. The responseLatency parameter captures this
283 * latency.
284 */
285 const Cycles responseLatency;
286
287 /** The number of targets for each MSHR. */
288 const int numTarget;
289
290 /** Do we forward snoops from mem side port through to cpu side port? */
291 const bool forwardSnoops;
292
293 /** Is this cache a toplevel cache (e.g. L1, I/O cache). If so we should
294 * never try to forward ownership and similar optimizations to the cpu
295 * side */
296 const bool isTopLevel;
297
298 /**
299 * Bit vector of the blocking reasons for the access path.
300 * @sa #BlockedCause
301 */
302 uint8_t blocked;
303
304 /** Increasing order number assigned to each incoming request. */
305 uint64_t order;
306
307 /** Stores time the cache blocked for statistics. */
308 Cycles blockedCycle;
309
310 /** Pointer to the MSHR that has no targets. */
311 MSHR *noTargetMSHR;
312
313 /** The number of misses to trigger an exit event. */
314 Counter missCount;
315
316 /**
317 * The address range to which the cache responds on the CPU side.
318 * Normally this is all possible memory addresses. */
319 const AddrRangeList addrRanges;
320
321 public:
322 /** System we are currently operating in. */
323 System *system;
324
325 // Statistics
326 /**
327 * @addtogroup CacheStatistics
328 * @{
329 */
330
331 /** Number of hits per thread for each type of command. @sa Packet::Command */
332 Stats::Vector hits[MemCmd::NUM_MEM_CMDS];
333 /** Number of hits for demand accesses. */
334 Stats::Formula demandHits;
335 /** Number of hit for all accesses. */
336 Stats::Formula overallHits;
337
338 /** Number of misses per thread for each type of command. @sa Packet::Command */
339 Stats::Vector misses[MemCmd::NUM_MEM_CMDS];
340 /** Number of misses for demand accesses. */
341 Stats::Formula demandMisses;
342 /** Number of misses for all accesses. */
343 Stats::Formula overallMisses;
344
345 /**
346 * Total number of cycles per thread/command spent waiting for a miss.
347 * Used to calculate the average miss latency.
348 */
349 Stats::Vector missLatency[MemCmd::NUM_MEM_CMDS];
350 /** Total number of cycles spent waiting for demand misses. */
351 Stats::Formula demandMissLatency;
352 /** Total number of cycles spent waiting for all misses. */
353 Stats::Formula overallMissLatency;
354
355 /** The number of accesses per command and thread. */
356 Stats::Formula accesses[MemCmd::NUM_MEM_CMDS];
357 /** The number of demand accesses. */
358 Stats::Formula demandAccesses;
359 /** The number of overall accesses. */
360 Stats::Formula overallAccesses;
361
362 /** The miss rate per command and thread. */
363 Stats::Formula missRate[MemCmd::NUM_MEM_CMDS];
364 /** The miss rate of all demand accesses. */
365 Stats::Formula demandMissRate;
366 /** The miss rate for all accesses. */
367 Stats::Formula overallMissRate;
368
369 /** The average miss latency per command and thread. */
370 Stats::Formula avgMissLatency[MemCmd::NUM_MEM_CMDS];
371 /** The average miss latency for demand misses. */
372 Stats::Formula demandAvgMissLatency;
373 /** The average miss latency for all misses. */
374 Stats::Formula overallAvgMissLatency;
375
376 /** The total number of cycles blocked for each blocked cause. */
377 Stats::Vector blocked_cycles;
378 /** The number of times this cache blocked for each blocked cause. */
379 Stats::Vector blocked_causes;
380
381 /** The average number of cycles blocked for each blocked cause. */
382 Stats::Formula avg_blocked;
383
384 /** The number of fast writes (WH64) performed. */
385 Stats::Scalar fastWrites;
386
387 /** The number of cache copies performed. */
388 Stats::Scalar cacheCopies;
389
390 /** Number of blocks written back per thread. */
391 Stats::Vector writebacks;
392
393 /** Number of misses that hit in the MSHRs per command and thread. */
394 Stats::Vector mshr_hits[MemCmd::NUM_MEM_CMDS];
395 /** Demand misses that hit in the MSHRs. */
396 Stats::Formula demandMshrHits;
397 /** Total number of misses that hit in the MSHRs. */
398 Stats::Formula overallMshrHits;
399
400 /** Number of misses that miss in the MSHRs, per command and thread. */
401 Stats::Vector mshr_misses[MemCmd::NUM_MEM_CMDS];
402 /** Demand misses that miss in the MSHRs. */
403 Stats::Formula demandMshrMisses;
404 /** Total number of misses that miss in the MSHRs. */
405 Stats::Formula overallMshrMisses;
406
407 /** Number of misses that miss in the MSHRs, per command and thread. */
408 Stats::Vector mshr_uncacheable[MemCmd::NUM_MEM_CMDS];
409 /** Total number of misses that miss in the MSHRs. */
410 Stats::Formula overallMshrUncacheable;
411
412 /** Total cycle latency of each MSHR miss, per command and thread. */
413 Stats::Vector mshr_miss_latency[MemCmd::NUM_MEM_CMDS];
414 /** Total cycle latency of demand MSHR misses. */
415 Stats::Formula demandMshrMissLatency;
416 /** Total cycle latency of overall MSHR misses. */
417 Stats::Formula overallMshrMissLatency;
418
419 /** Total cycle latency of each MSHR miss, per command and thread. */
420 Stats::Vector mshr_uncacheable_lat[MemCmd::NUM_MEM_CMDS];
421 /** Total cycle latency of overall MSHR misses. */
422 Stats::Formula overallMshrUncacheableLatency;
423
424#if 0
425 /** The total number of MSHR accesses per command and thread. */
426 Stats::Formula mshrAccesses[MemCmd::NUM_MEM_CMDS];
427 /** The total number of demand MSHR accesses. */
428 Stats::Formula demandMshrAccesses;
429 /** The total number of MSHR accesses. */
430 Stats::Formula overallMshrAccesses;
431#endif
432
433 /** The miss rate in the MSHRs pre command and thread. */
434 Stats::Formula mshrMissRate[MemCmd::NUM_MEM_CMDS];
435 /** The demand miss rate in the MSHRs. */
436 Stats::Formula demandMshrMissRate;
437 /** The overall miss rate in the MSHRs. */
438 Stats::Formula overallMshrMissRate;
439
440 /** The average latency of an MSHR miss, per command and thread. */
441 Stats::Formula avgMshrMissLatency[MemCmd::NUM_MEM_CMDS];
442 /** The average latency of a demand MSHR miss. */
443 Stats::Formula demandAvgMshrMissLatency;
444 /** The average overall latency of an MSHR miss. */
445 Stats::Formula overallAvgMshrMissLatency;
446
447 /** The average latency of an MSHR miss, per command and thread. */
448 Stats::Formula avgMshrUncacheableLatency[MemCmd::NUM_MEM_CMDS];
449 /** The average overall latency of an MSHR miss. */
450 Stats::Formula overallAvgMshrUncacheableLatency;
451
452 /** The number of times a thread hit its MSHR cap. */
453 Stats::Vector mshr_cap_events;
454 /** The number of times software prefetches caused the MSHR to block. */
455 Stats::Vector soft_prefetch_mshr_full;
456
457 Stats::Scalar mshr_no_allocate_misses;
458
459 /**
460 * @}
461 */
462
463 /**
464 * Register stats for this object.
465 */
466 virtual void regStats();
467
468 public:
469 typedef BaseCacheParams Params;
470 BaseCache(const Params *p);
471 ~BaseCache() {}
472
473 virtual void init();
474
475 virtual BaseMasterPort &getMasterPort(const std::string &if_name,
476 PortID idx = InvalidPortID);
477 virtual BaseSlavePort &getSlavePort(const std::string &if_name,
478 PortID idx = InvalidPortID);
479
480 /**
481 * Query block size of a cache.
482 * @return The block size
483 */
484 unsigned
485 getBlockSize() const
486 {
487 return blkSize;
488 }
489
490
491 Addr blockAlign(Addr addr) const { return (addr & ~(Addr(blkSize - 1))); }
492
493
494 const AddrRangeList &getAddrRanges() const { return addrRanges; }
495
496 MSHR *allocateMissBuffer(PacketPtr pkt, Tick time, bool requestBus)
497 {
498 assert(!pkt->req->isUncacheable());
499 return allocateBufferInternal(&mshrQueue,
500 blockAlign(pkt->getAddr()), blkSize,
501 pkt, time, requestBus);
502 }
503
504 MSHR *allocateWriteBuffer(PacketPtr pkt, Tick time, bool requestBus)
505 {
506 assert(pkt->isWrite() && !pkt->isRead());
507 return allocateBufferInternal(&writeBuffer,
508 pkt->getAddr(), pkt->getSize(),
509 pkt, time, requestBus);
510 }
511
512 MSHR *allocateUncachedReadBuffer(PacketPtr pkt, Tick time, bool requestBus)
513 {
514 assert(pkt->req->isUncacheable());
515 assert(pkt->isRead());
516 return allocateBufferInternal(&mshrQueue,
517 pkt->getAddr(), pkt->getSize(),
518 pkt, time, requestBus);
519 }
520
521 /**
522 * Returns true if the cache is blocked for accesses.
523 */
524 bool isBlocked() const
525 {
526 return blocked != 0;
527 }
528
529 /**
530 * Marks the access path of the cache as blocked for the given cause. This
531 * also sets the blocked flag in the slave interface.
532 * @param cause The reason for the cache blocking.
533 */
534 void setBlocked(BlockedCause cause)
535 {
536 uint8_t flag = 1 << cause;
537 if (blocked == 0) {
538 blocked_causes[cause]++;
539 blockedCycle = curCycle();
540 cpuSidePort->setBlocked();
541 }
542 blocked |= flag;
543 DPRINTF(Cache,"Blocking for cause %d, mask=%d\n", cause, blocked);
544 }
545
546 /**
547 * Marks the cache as unblocked for the given cause. This also clears the
548 * blocked flags in the appropriate interfaces.
549 * @param cause The newly unblocked cause.
550 * @warning Calling this function can cause a blocked request on the bus to
551 * access the cache. The cache must be in a state to handle that request.
552 */
553 void clearBlocked(BlockedCause cause)
554 {
555 uint8_t flag = 1 << cause;
556 blocked &= ~flag;
557 DPRINTF(Cache,"Unblocking for cause %d, mask=%d\n", cause, blocked);
558 if (blocked == 0) {
559 blocked_cycles[cause] += curCycle() - blockedCycle;
560 cpuSidePort->clearBlocked();
561 }
562 }
563
564 /**
565 * Request the master bus for the given cause and time.
566 * @param cause The reason for the request.
567 * @param time The time to make the request.
568 */
569 void requestMemSideBus(RequestCause cause, Tick time)
570 {
571 memSidePort->requestBus(cause, time);
572 }
573
574 /**
575 * Clear the master bus request for the given cause.
576 * @param cause The request reason to clear.
577 */
578 void deassertMemSideBusRequest(RequestCause cause)
579 {
580 // Obsolete... we no longer signal bus requests explicitly so
581 // we can't deassert them. Leaving this in as a no-op since
582 // the prefetcher calls it to indicate that it no longer wants
583 // to request a prefetch, and someday that might be
584 // interesting again.
585 }
586
587 virtual unsigned int drain(DrainManager *dm);
588
589 virtual bool inCache(Addr addr, bool is_secure) const = 0;
590
591 virtual bool inMissQueue(Addr addr, bool is_secure) const = 0;
592
593 void incMissCount(PacketPtr pkt)
594 {
595 assert(pkt->req->masterId() < system->maxMasters());
596 misses[pkt->cmdToIndex()][pkt->req->masterId()]++;
597 pkt->req->incAccessDepth();
598 if (missCount) {
599 --missCount;
600 if (missCount == 0)
601 exitSimLoop("A cache reached the maximum miss count");
602 }
603 }
604 void incHitCount(PacketPtr pkt)
605 {
606 assert(pkt->req->masterId() < system->maxMasters());
607 hits[pkt->cmdToIndex()][pkt->req->masterId()]++;
608
609 }
610
611};
612
613#endif //__BASE_CACHE_HH__
134 reqQueue.schedSendEvent(time);
135 }
136
137 protected:
138
139 CacheMasterPort(const std::string &_name, BaseCache *_cache,
140 ReqPacketQueue &_reqQueue,
141 SnoopRespPacketQueue &_snoopRespQueue) :
142 QueuedMasterPort(_name, _cache, _reqQueue, _snoopRespQueue)
143 { }
144
145 /**
146 * Memory-side port always snoops.
147 *
148 * @return always true
149 */
150 virtual bool isSnooping() const { return true; }
151 };
152
153 /**
154 * A cache slave port is used for the CPU-side port of the cache,
155 * and it is basically a simple timing port that uses a transmit
156 * list for responses to the CPU (or connected master). In
157 * addition, it has the functionality to block the port for
158 * incoming requests. If blocked, the port will issue a retry once
159 * unblocked.
160 */
161 class CacheSlavePort : public QueuedSlavePort
162 {
163
164 public:
165
166 /** Do not accept any new requests. */
167 void setBlocked();
168
169 /** Return to normal operation and accept new requests. */
170 void clearBlocked();
171
172 bool isBlocked() const { return blocked; }
173
174 protected:
175
176 CacheSlavePort(const std::string &_name, BaseCache *_cache,
177 const std::string &_label);
178
179 /** A normal packet queue used to store responses. */
180 RespPacketQueue queue;
181
182 bool blocked;
183
184 bool mustSendRetry;
185
186 private:
187
188 void processSendRetry();
189
190 EventWrapper<CacheSlavePort,
191 &CacheSlavePort::processSendRetry> sendRetryEvent;
192
193 };
194
195 CacheSlavePort *cpuSidePort;
196 CacheMasterPort *memSidePort;
197
198 protected:
199
200 /** Miss status registers */
201 MSHRQueue mshrQueue;
202
203 /** Write/writeback buffer */
204 MSHRQueue writeBuffer;
205
206 /**
207 * Allocate a buffer, passing the time indicating when schedule an
208 * event to the queued port to go and ask the MSHR and write queue
209 * if they have packets to send.
210 *
211 * allocateBufferInternal() function is called in:
212 * - MSHR allocateWriteBuffer (unchached write forwarded to WriteBuffer);
213 * - MSHR allocateMissBuffer (cacheable miss in MSHR queue);
214 * - MSHR allocateUncachedReadBuffer (unchached read allocated in MSHR
215 * queue)
216 */
217 MSHR *allocateBufferInternal(MSHRQueue *mq, Addr addr, int size,
218 PacketPtr pkt, Tick time, bool requestBus)
219 {
220 MSHR *mshr = mq->allocate(addr, size, pkt, time, order++);
221
222 if (mq->isFull()) {
223 setBlocked((BlockedCause)mq->index);
224 }
225
226 if (requestBus) {
227 requestMemSideBus((RequestCause)mq->index, time);
228 }
229
230 return mshr;
231 }
232
233 void markInServiceInternal(MSHR *mshr, bool pending_dirty_resp)
234 {
235 MSHRQueue *mq = mshr->queue;
236 bool wasFull = mq->isFull();
237 mq->markInService(mshr, pending_dirty_resp);
238 if (wasFull && !mq->isFull()) {
239 clearBlocked((BlockedCause)mq->index);
240 }
241 }
242
243 /**
244 * Write back dirty blocks in the cache using functional accesses.
245 */
246 virtual void memWriteback() = 0;
247 /**
248 * Invalidates all blocks in the cache.
249 *
250 * @warn Dirty cache lines will not be written back to
251 * memory. Make sure to call functionalWriteback() first if you
252 * want the to write them to memory.
253 */
254 virtual void memInvalidate() = 0;
255 /**
256 * Determine if there are any dirty blocks in the cache.
257 *
258 * \return true if at least one block is dirty, false otherwise.
259 */
260 virtual bool isDirty() const = 0;
261
262 /** Block size of this cache */
263 const unsigned blkSize;
264
265 /**
266 * The latency of tag lookup of a cache. It occurs when there is
267 * an access to the cache.
268 */
269 const Cycles lookupLatency;
270
271 /**
272 * This is the forward latency of the cache. It occurs when there
273 * is a cache miss and a request is forwarded downstream, in
274 * particular an outbound miss.
275 */
276 const Cycles forwardLatency;
277
278 /** The latency to fill a cache block */
279 const Cycles fillLatency;
280
281 /**
282 * The latency of sending reponse to its upper level cache/core on
283 * a linefill. The responseLatency parameter captures this
284 * latency.
285 */
286 const Cycles responseLatency;
287
288 /** The number of targets for each MSHR. */
289 const int numTarget;
290
291 /** Do we forward snoops from mem side port through to cpu side port? */
292 const bool forwardSnoops;
293
294 /** Is this cache a toplevel cache (e.g. L1, I/O cache). If so we should
295 * never try to forward ownership and similar optimizations to the cpu
296 * side */
297 const bool isTopLevel;
298
299 /**
300 * Bit vector of the blocking reasons for the access path.
301 * @sa #BlockedCause
302 */
303 uint8_t blocked;
304
305 /** Increasing order number assigned to each incoming request. */
306 uint64_t order;
307
308 /** Stores time the cache blocked for statistics. */
309 Cycles blockedCycle;
310
311 /** Pointer to the MSHR that has no targets. */
312 MSHR *noTargetMSHR;
313
314 /** The number of misses to trigger an exit event. */
315 Counter missCount;
316
317 /**
318 * The address range to which the cache responds on the CPU side.
319 * Normally this is all possible memory addresses. */
320 const AddrRangeList addrRanges;
321
322 public:
323 /** System we are currently operating in. */
324 System *system;
325
326 // Statistics
327 /**
328 * @addtogroup CacheStatistics
329 * @{
330 */
331
332 /** Number of hits per thread for each type of command. @sa Packet::Command */
333 Stats::Vector hits[MemCmd::NUM_MEM_CMDS];
334 /** Number of hits for demand accesses. */
335 Stats::Formula demandHits;
336 /** Number of hit for all accesses. */
337 Stats::Formula overallHits;
338
339 /** Number of misses per thread for each type of command. @sa Packet::Command */
340 Stats::Vector misses[MemCmd::NUM_MEM_CMDS];
341 /** Number of misses for demand accesses. */
342 Stats::Formula demandMisses;
343 /** Number of misses for all accesses. */
344 Stats::Formula overallMisses;
345
346 /**
347 * Total number of cycles per thread/command spent waiting for a miss.
348 * Used to calculate the average miss latency.
349 */
350 Stats::Vector missLatency[MemCmd::NUM_MEM_CMDS];
351 /** Total number of cycles spent waiting for demand misses. */
352 Stats::Formula demandMissLatency;
353 /** Total number of cycles spent waiting for all misses. */
354 Stats::Formula overallMissLatency;
355
356 /** The number of accesses per command and thread. */
357 Stats::Formula accesses[MemCmd::NUM_MEM_CMDS];
358 /** The number of demand accesses. */
359 Stats::Formula demandAccesses;
360 /** The number of overall accesses. */
361 Stats::Formula overallAccesses;
362
363 /** The miss rate per command and thread. */
364 Stats::Formula missRate[MemCmd::NUM_MEM_CMDS];
365 /** The miss rate of all demand accesses. */
366 Stats::Formula demandMissRate;
367 /** The miss rate for all accesses. */
368 Stats::Formula overallMissRate;
369
370 /** The average miss latency per command and thread. */
371 Stats::Formula avgMissLatency[MemCmd::NUM_MEM_CMDS];
372 /** The average miss latency for demand misses. */
373 Stats::Formula demandAvgMissLatency;
374 /** The average miss latency for all misses. */
375 Stats::Formula overallAvgMissLatency;
376
377 /** The total number of cycles blocked for each blocked cause. */
378 Stats::Vector blocked_cycles;
379 /** The number of times this cache blocked for each blocked cause. */
380 Stats::Vector blocked_causes;
381
382 /** The average number of cycles blocked for each blocked cause. */
383 Stats::Formula avg_blocked;
384
385 /** The number of fast writes (WH64) performed. */
386 Stats::Scalar fastWrites;
387
388 /** The number of cache copies performed. */
389 Stats::Scalar cacheCopies;
390
391 /** Number of blocks written back per thread. */
392 Stats::Vector writebacks;
393
394 /** Number of misses that hit in the MSHRs per command and thread. */
395 Stats::Vector mshr_hits[MemCmd::NUM_MEM_CMDS];
396 /** Demand misses that hit in the MSHRs. */
397 Stats::Formula demandMshrHits;
398 /** Total number of misses that hit in the MSHRs. */
399 Stats::Formula overallMshrHits;
400
401 /** Number of misses that miss in the MSHRs, per command and thread. */
402 Stats::Vector mshr_misses[MemCmd::NUM_MEM_CMDS];
403 /** Demand misses that miss in the MSHRs. */
404 Stats::Formula demandMshrMisses;
405 /** Total number of misses that miss in the MSHRs. */
406 Stats::Formula overallMshrMisses;
407
408 /** Number of misses that miss in the MSHRs, per command and thread. */
409 Stats::Vector mshr_uncacheable[MemCmd::NUM_MEM_CMDS];
410 /** Total number of misses that miss in the MSHRs. */
411 Stats::Formula overallMshrUncacheable;
412
413 /** Total cycle latency of each MSHR miss, per command and thread. */
414 Stats::Vector mshr_miss_latency[MemCmd::NUM_MEM_CMDS];
415 /** Total cycle latency of demand MSHR misses. */
416 Stats::Formula demandMshrMissLatency;
417 /** Total cycle latency of overall MSHR misses. */
418 Stats::Formula overallMshrMissLatency;
419
420 /** Total cycle latency of each MSHR miss, per command and thread. */
421 Stats::Vector mshr_uncacheable_lat[MemCmd::NUM_MEM_CMDS];
422 /** Total cycle latency of overall MSHR misses. */
423 Stats::Formula overallMshrUncacheableLatency;
424
425#if 0
426 /** The total number of MSHR accesses per command and thread. */
427 Stats::Formula mshrAccesses[MemCmd::NUM_MEM_CMDS];
428 /** The total number of demand MSHR accesses. */
429 Stats::Formula demandMshrAccesses;
430 /** The total number of MSHR accesses. */
431 Stats::Formula overallMshrAccesses;
432#endif
433
434 /** The miss rate in the MSHRs pre command and thread. */
435 Stats::Formula mshrMissRate[MemCmd::NUM_MEM_CMDS];
436 /** The demand miss rate in the MSHRs. */
437 Stats::Formula demandMshrMissRate;
438 /** The overall miss rate in the MSHRs. */
439 Stats::Formula overallMshrMissRate;
440
441 /** The average latency of an MSHR miss, per command and thread. */
442 Stats::Formula avgMshrMissLatency[MemCmd::NUM_MEM_CMDS];
443 /** The average latency of a demand MSHR miss. */
444 Stats::Formula demandAvgMshrMissLatency;
445 /** The average overall latency of an MSHR miss. */
446 Stats::Formula overallAvgMshrMissLatency;
447
448 /** The average latency of an MSHR miss, per command and thread. */
449 Stats::Formula avgMshrUncacheableLatency[MemCmd::NUM_MEM_CMDS];
450 /** The average overall latency of an MSHR miss. */
451 Stats::Formula overallAvgMshrUncacheableLatency;
452
453 /** The number of times a thread hit its MSHR cap. */
454 Stats::Vector mshr_cap_events;
455 /** The number of times software prefetches caused the MSHR to block. */
456 Stats::Vector soft_prefetch_mshr_full;
457
458 Stats::Scalar mshr_no_allocate_misses;
459
460 /**
461 * @}
462 */
463
464 /**
465 * Register stats for this object.
466 */
467 virtual void regStats();
468
469 public:
470 typedef BaseCacheParams Params;
471 BaseCache(const Params *p);
472 ~BaseCache() {}
473
474 virtual void init();
475
476 virtual BaseMasterPort &getMasterPort(const std::string &if_name,
477 PortID idx = InvalidPortID);
478 virtual BaseSlavePort &getSlavePort(const std::string &if_name,
479 PortID idx = InvalidPortID);
480
481 /**
482 * Query block size of a cache.
483 * @return The block size
484 */
485 unsigned
486 getBlockSize() const
487 {
488 return blkSize;
489 }
490
491
492 Addr blockAlign(Addr addr) const { return (addr & ~(Addr(blkSize - 1))); }
493
494
495 const AddrRangeList &getAddrRanges() const { return addrRanges; }
496
497 MSHR *allocateMissBuffer(PacketPtr pkt, Tick time, bool requestBus)
498 {
499 assert(!pkt->req->isUncacheable());
500 return allocateBufferInternal(&mshrQueue,
501 blockAlign(pkt->getAddr()), blkSize,
502 pkt, time, requestBus);
503 }
504
505 MSHR *allocateWriteBuffer(PacketPtr pkt, Tick time, bool requestBus)
506 {
507 assert(pkt->isWrite() && !pkt->isRead());
508 return allocateBufferInternal(&writeBuffer,
509 pkt->getAddr(), pkt->getSize(),
510 pkt, time, requestBus);
511 }
512
513 MSHR *allocateUncachedReadBuffer(PacketPtr pkt, Tick time, bool requestBus)
514 {
515 assert(pkt->req->isUncacheable());
516 assert(pkt->isRead());
517 return allocateBufferInternal(&mshrQueue,
518 pkt->getAddr(), pkt->getSize(),
519 pkt, time, requestBus);
520 }
521
522 /**
523 * Returns true if the cache is blocked for accesses.
524 */
525 bool isBlocked() const
526 {
527 return blocked != 0;
528 }
529
530 /**
531 * Marks the access path of the cache as blocked for the given cause. This
532 * also sets the blocked flag in the slave interface.
533 * @param cause The reason for the cache blocking.
534 */
535 void setBlocked(BlockedCause cause)
536 {
537 uint8_t flag = 1 << cause;
538 if (blocked == 0) {
539 blocked_causes[cause]++;
540 blockedCycle = curCycle();
541 cpuSidePort->setBlocked();
542 }
543 blocked |= flag;
544 DPRINTF(Cache,"Blocking for cause %d, mask=%d\n", cause, blocked);
545 }
546
547 /**
548 * Marks the cache as unblocked for the given cause. This also clears the
549 * blocked flags in the appropriate interfaces.
550 * @param cause The newly unblocked cause.
551 * @warning Calling this function can cause a blocked request on the bus to
552 * access the cache. The cache must be in a state to handle that request.
553 */
554 void clearBlocked(BlockedCause cause)
555 {
556 uint8_t flag = 1 << cause;
557 blocked &= ~flag;
558 DPRINTF(Cache,"Unblocking for cause %d, mask=%d\n", cause, blocked);
559 if (blocked == 0) {
560 blocked_cycles[cause] += curCycle() - blockedCycle;
561 cpuSidePort->clearBlocked();
562 }
563 }
564
565 /**
566 * Request the master bus for the given cause and time.
567 * @param cause The reason for the request.
568 * @param time The time to make the request.
569 */
570 void requestMemSideBus(RequestCause cause, Tick time)
571 {
572 memSidePort->requestBus(cause, time);
573 }
574
575 /**
576 * Clear the master bus request for the given cause.
577 * @param cause The request reason to clear.
578 */
579 void deassertMemSideBusRequest(RequestCause cause)
580 {
581 // Obsolete... we no longer signal bus requests explicitly so
582 // we can't deassert them. Leaving this in as a no-op since
583 // the prefetcher calls it to indicate that it no longer wants
584 // to request a prefetch, and someday that might be
585 // interesting again.
586 }
587
588 virtual unsigned int drain(DrainManager *dm);
589
590 virtual bool inCache(Addr addr, bool is_secure) const = 0;
591
592 virtual bool inMissQueue(Addr addr, bool is_secure) const = 0;
593
594 void incMissCount(PacketPtr pkt)
595 {
596 assert(pkt->req->masterId() < system->maxMasters());
597 misses[pkt->cmdToIndex()][pkt->req->masterId()]++;
598 pkt->req->incAccessDepth();
599 if (missCount) {
600 --missCount;
601 if (missCount == 0)
602 exitSimLoop("A cache reached the maximum miss count");
603 }
604 }
605 void incHitCount(PacketPtr pkt)
606 {
607 assert(pkt->req->masterId() < system->maxMasters());
608 hits[pkt->cmdToIndex()][pkt->req->masterId()]++;
609
610 }
611
612};
613
614#endif //__BASE_CACHE_HH__