base.hh revision 10764:b32578b2af99
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, "Scheduling request at %llu due to %d\n",
133                    time, cause);
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        // check that the address is block aligned since we rely on
221        // this in a number of places when checking for matches and
222        // overlap
223        assert(addr == blockAlign(addr));
224
225        MSHR *mshr = mq->allocate(addr, size, pkt, time, order++);
226
227        if (mq->isFull()) {
228            setBlocked((BlockedCause)mq->index);
229        }
230
231        if (requestBus) {
232            requestMemSideBus((RequestCause)mq->index, time);
233        }
234
235        return mshr;
236    }
237
238    void markInServiceInternal(MSHR *mshr, bool pending_dirty_resp)
239    {
240        MSHRQueue *mq = mshr->queue;
241        bool wasFull = mq->isFull();
242        mq->markInService(mshr, pending_dirty_resp);
243        if (wasFull && !mq->isFull()) {
244            clearBlocked((BlockedCause)mq->index);
245        }
246    }
247
248    /**
249     * Write back dirty blocks in the cache using functional accesses.
250     */
251    virtual void memWriteback() = 0;
252    /**
253     * Invalidates all blocks in the cache.
254     *
255     * @warn Dirty cache lines will not be written back to
256     * memory. Make sure to call functionalWriteback() first if you
257     * want the to write them to memory.
258     */
259    virtual void memInvalidate() = 0;
260    /**
261     * Determine if there are any dirty blocks in the cache.
262     *
263     * \return true if at least one block is dirty, false otherwise.
264     */
265    virtual bool isDirty() const = 0;
266
267    /** Block size of this cache */
268    const unsigned blkSize;
269
270    /**
271     * The latency of tag lookup of a cache. It occurs when there is
272     * an access to the cache.
273     */
274    const Cycles lookupLatency;
275
276    /**
277     * This is the forward latency of the cache. It occurs when there
278     * is a cache miss and a request is forwarded downstream, in
279     * particular an outbound miss.
280     */
281    const Cycles forwardLatency;
282
283    /** The latency to fill a cache block */
284    const Cycles fillLatency;
285
286    /**
287     * The latency of sending reponse to its upper level cache/core on
288     * a linefill. The responseLatency parameter captures this
289     * latency.
290     */
291    const Cycles responseLatency;
292
293    /** The number of targets for each MSHR. */
294    const int numTarget;
295
296    /** Do we forward snoops from mem side port through to cpu side port? */
297    const bool forwardSnoops;
298
299    /** Is this cache a toplevel cache (e.g. L1, I/O cache). If so we should
300     * never try to forward ownership and similar optimizations to the cpu
301     * side */
302    const bool isTopLevel;
303
304    /**
305     * Bit vector of the blocking reasons for the access path.
306     * @sa #BlockedCause
307     */
308    uint8_t blocked;
309
310    /** Increasing order number assigned to each incoming request. */
311    uint64_t order;
312
313    /** Stores time the cache blocked for statistics. */
314    Cycles blockedCycle;
315
316    /** Pointer to the MSHR that has no targets. */
317    MSHR *noTargetMSHR;
318
319    /** The number of misses to trigger an exit event. */
320    Counter missCount;
321
322    /**
323     * The address range to which the cache responds on the CPU side.
324     * Normally this is all possible memory addresses. */
325    const AddrRangeList addrRanges;
326
327  public:
328    /** System we are currently operating in. */
329    System *system;
330
331    // Statistics
332    /**
333     * @addtogroup CacheStatistics
334     * @{
335     */
336
337    /** Number of hits per thread for each type of command. @sa Packet::Command */
338    Stats::Vector hits[MemCmd::NUM_MEM_CMDS];
339    /** Number of hits for demand accesses. */
340    Stats::Formula demandHits;
341    /** Number of hit for all accesses. */
342    Stats::Formula overallHits;
343
344    /** Number of misses per thread for each type of command. @sa Packet::Command */
345    Stats::Vector misses[MemCmd::NUM_MEM_CMDS];
346    /** Number of misses for demand accesses. */
347    Stats::Formula demandMisses;
348    /** Number of misses for all accesses. */
349    Stats::Formula overallMisses;
350
351    /**
352     * Total number of cycles per thread/command spent waiting for a miss.
353     * Used to calculate the average miss latency.
354     */
355    Stats::Vector missLatency[MemCmd::NUM_MEM_CMDS];
356    /** Total number of cycles spent waiting for demand misses. */
357    Stats::Formula demandMissLatency;
358    /** Total number of cycles spent waiting for all misses. */
359    Stats::Formula overallMissLatency;
360
361    /** The number of accesses per command and thread. */
362    Stats::Formula accesses[MemCmd::NUM_MEM_CMDS];
363    /** The number of demand accesses. */
364    Stats::Formula demandAccesses;
365    /** The number of overall accesses. */
366    Stats::Formula overallAccesses;
367
368    /** The miss rate per command and thread. */
369    Stats::Formula missRate[MemCmd::NUM_MEM_CMDS];
370    /** The miss rate of all demand accesses. */
371    Stats::Formula demandMissRate;
372    /** The miss rate for all accesses. */
373    Stats::Formula overallMissRate;
374
375    /** The average miss latency per command and thread. */
376    Stats::Formula avgMissLatency[MemCmd::NUM_MEM_CMDS];
377    /** The average miss latency for demand misses. */
378    Stats::Formula demandAvgMissLatency;
379    /** The average miss latency for all misses. */
380    Stats::Formula overallAvgMissLatency;
381
382    /** The total number of cycles blocked for each blocked cause. */
383    Stats::Vector blocked_cycles;
384    /** The number of times this cache blocked for each blocked cause. */
385    Stats::Vector blocked_causes;
386
387    /** The average number of cycles blocked for each blocked cause. */
388    Stats::Formula avg_blocked;
389
390    /** The number of fast writes (WH64) performed. */
391    Stats::Scalar fastWrites;
392
393    /** The number of cache copies performed. */
394    Stats::Scalar cacheCopies;
395
396    /** Number of blocks written back per thread. */
397    Stats::Vector writebacks;
398
399    /** Number of misses that hit in the MSHRs per command and thread. */
400    Stats::Vector mshr_hits[MemCmd::NUM_MEM_CMDS];
401    /** Demand misses that hit in the MSHRs. */
402    Stats::Formula demandMshrHits;
403    /** Total number of misses that hit in the MSHRs. */
404    Stats::Formula overallMshrHits;
405
406    /** Number of misses that miss in the MSHRs, per command and thread. */
407    Stats::Vector mshr_misses[MemCmd::NUM_MEM_CMDS];
408    /** Demand misses that miss in the MSHRs. */
409    Stats::Formula demandMshrMisses;
410    /** Total number of misses that miss in the MSHRs. */
411    Stats::Formula overallMshrMisses;
412
413    /** Number of misses that miss in the MSHRs, per command and thread. */
414    Stats::Vector mshr_uncacheable[MemCmd::NUM_MEM_CMDS];
415    /** Total number of misses that miss in the MSHRs. */
416    Stats::Formula overallMshrUncacheable;
417
418    /** Total cycle latency of each MSHR miss, per command and thread. */
419    Stats::Vector mshr_miss_latency[MemCmd::NUM_MEM_CMDS];
420    /** Total cycle latency of demand MSHR misses. */
421    Stats::Formula demandMshrMissLatency;
422    /** Total cycle latency of overall MSHR misses. */
423    Stats::Formula overallMshrMissLatency;
424
425    /** Total cycle latency of each MSHR miss, per command and thread. */
426    Stats::Vector mshr_uncacheable_lat[MemCmd::NUM_MEM_CMDS];
427    /** Total cycle latency of overall MSHR misses. */
428    Stats::Formula overallMshrUncacheableLatency;
429
430#if 0
431    /** The total number of MSHR accesses per command and thread. */
432    Stats::Formula mshrAccesses[MemCmd::NUM_MEM_CMDS];
433    /** The total number of demand MSHR accesses. */
434    Stats::Formula demandMshrAccesses;
435    /** The total number of MSHR accesses. */
436    Stats::Formula overallMshrAccesses;
437#endif
438
439    /** The miss rate in the MSHRs pre command and thread. */
440    Stats::Formula mshrMissRate[MemCmd::NUM_MEM_CMDS];
441    /** The demand miss rate in the MSHRs. */
442    Stats::Formula demandMshrMissRate;
443    /** The overall miss rate in the MSHRs. */
444    Stats::Formula overallMshrMissRate;
445
446    /** The average latency of an MSHR miss, per command and thread. */
447    Stats::Formula avgMshrMissLatency[MemCmd::NUM_MEM_CMDS];
448    /** The average latency of a demand MSHR miss. */
449    Stats::Formula demandAvgMshrMissLatency;
450    /** The average overall latency of an MSHR miss. */
451    Stats::Formula overallAvgMshrMissLatency;
452
453    /** The average latency of an MSHR miss, per command and thread. */
454    Stats::Formula avgMshrUncacheableLatency[MemCmd::NUM_MEM_CMDS];
455    /** The average overall latency of an MSHR miss. */
456    Stats::Formula overallAvgMshrUncacheableLatency;
457
458    /** The number of times a thread hit its MSHR cap. */
459    Stats::Vector mshr_cap_events;
460    /** The number of times software prefetches caused the MSHR to block. */
461    Stats::Vector soft_prefetch_mshr_full;
462
463    Stats::Scalar mshr_no_allocate_misses;
464
465    /**
466     * @}
467     */
468
469    /**
470     * Register stats for this object.
471     */
472    virtual void regStats();
473
474  public:
475    typedef BaseCacheParams Params;
476    BaseCache(const Params *p);
477    ~BaseCache() {}
478
479    virtual void init();
480
481    virtual BaseMasterPort &getMasterPort(const std::string &if_name,
482                                          PortID idx = InvalidPortID);
483    virtual BaseSlavePort &getSlavePort(const std::string &if_name,
484                                        PortID idx = InvalidPortID);
485
486    /**
487     * Query block size of a cache.
488     * @return  The block size
489     */
490    unsigned
491    getBlockSize() const
492    {
493        return blkSize;
494    }
495
496
497    Addr blockAlign(Addr addr) const { return (addr & ~(Addr(blkSize - 1))); }
498
499
500    const AddrRangeList &getAddrRanges() const { return addrRanges; }
501
502    MSHR *allocateMissBuffer(PacketPtr pkt, Tick time, bool requestBus)
503    {
504        assert(!pkt->req->isUncacheable());
505        return allocateBufferInternal(&mshrQueue,
506                                      blockAlign(pkt->getAddr()), blkSize,
507                                      pkt, time, requestBus);
508    }
509
510    MSHR *allocateWriteBuffer(PacketPtr pkt, Tick time, bool requestBus)
511    {
512        assert(pkt->isWrite() && !pkt->isRead());
513        return allocateBufferInternal(&writeBuffer,
514                                      blockAlign(pkt->getAddr()), blkSize,
515                                      pkt, time, requestBus);
516    }
517
518    MSHR *allocateUncachedReadBuffer(PacketPtr pkt, Tick time, bool requestBus)
519    {
520        assert(pkt->req->isUncacheable());
521        assert(pkt->isRead());
522        return allocateBufferInternal(&mshrQueue,
523                                      blockAlign(pkt->getAddr()), blkSize,
524                                      pkt, time, requestBus);
525    }
526
527    /**
528     * Returns true if the cache is blocked for accesses.
529     */
530    bool isBlocked() const
531    {
532        return blocked != 0;
533    }
534
535    /**
536     * Marks the access path of the cache as blocked for the given cause. This
537     * also sets the blocked flag in the slave interface.
538     * @param cause The reason for the cache blocking.
539     */
540    void setBlocked(BlockedCause cause)
541    {
542        uint8_t flag = 1 << cause;
543        if (blocked == 0) {
544            blocked_causes[cause]++;
545            blockedCycle = curCycle();
546            cpuSidePort->setBlocked();
547        }
548        blocked |= flag;
549        DPRINTF(Cache,"Blocking for cause %d, mask=%d\n", cause, blocked);
550    }
551
552    /**
553     * Marks the cache as unblocked for the given cause. This also clears the
554     * blocked flags in the appropriate interfaces.
555     * @param cause The newly unblocked cause.
556     * @warning Calling this function can cause a blocked request on the bus to
557     * access the cache. The cache must be in a state to handle that request.
558     */
559    void clearBlocked(BlockedCause cause)
560    {
561        uint8_t flag = 1 << cause;
562        blocked &= ~flag;
563        DPRINTF(Cache,"Unblocking for cause %d, mask=%d\n", cause, blocked);
564        if (blocked == 0) {
565            blocked_cycles[cause] += curCycle() - blockedCycle;
566            cpuSidePort->clearBlocked();
567        }
568    }
569
570    /**
571     * Request the master bus for the given cause and time.
572     * @param cause The reason for the request.
573     * @param time The time to make the request.
574     */
575    void requestMemSideBus(RequestCause cause, Tick time)
576    {
577        memSidePort->requestBus(cause, time);
578    }
579
580    /**
581     * Clear the master bus request for the given cause.
582     * @param cause The request reason to clear.
583     */
584    void deassertMemSideBusRequest(RequestCause cause)
585    {
586        // Obsolete... we no longer signal bus requests explicitly so
587        // we can't deassert them.  Leaving this in as a no-op since
588        // the prefetcher calls it to indicate that it no longer wants
589        // to request a prefetch, and someday that might be
590        // interesting again.
591    }
592
593    virtual unsigned int drain(DrainManager *dm);
594
595    virtual bool inCache(Addr addr, bool is_secure) const = 0;
596
597    virtual bool inMissQueue(Addr addr, bool is_secure) const = 0;
598
599    void incMissCount(PacketPtr pkt)
600    {
601        assert(pkt->req->masterId() < system->maxMasters());
602        misses[pkt->cmdToIndex()][pkt->req->masterId()]++;
603        pkt->req->incAccessDepth();
604        if (missCount) {
605            --missCount;
606            if (missCount == 0)
607                exitSimLoop("A cache reached the maximum miss count");
608        }
609    }
610    void incHitCount(PacketPtr pkt)
611    {
612        assert(pkt->req->masterId() < system->maxMasters());
613        hits[pkt->cmdToIndex()][pkt->req->masterId()]++;
614
615    }
616
617};
618
619#endif //__BASE_CACHE_HH__
620