mem-cache: Avoid write merging if there are reads in between
[gem5.git] / src / mem / packet.cc
1 /*
2 * Copyright (c) 2011-2019 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) 2006 The Regents of The University of Michigan
15 * Copyright (c) 2010,2015 Advanced Micro Devices, Inc.
16 * All rights reserved.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 *
41 * Authors: Ali Saidi
42 * Steve Reinhardt
43 */
44
45 /**
46 * @file
47 * Definition of the Packet Class, a packet is a transaction occuring
48 * between a single level of the memory heirarchy (ie L1->L2).
49 */
50
51 #include "mem/packet.hh"
52
53 #include <algorithm>
54 #include <cstring>
55 #include <iostream>
56 #include <sstream>
57 #include <string>
58
59 #include "base/cprintf.hh"
60 #include "base/logging.hh"
61 #include "base/trace.hh"
62 #include "mem/packet_access.hh"
63
64 // The one downside to bitsets is that static initializers can get ugly.
65 #define SET1(a1) (1 << (a1))
66 #define SET2(a1, a2) (SET1(a1) | SET1(a2))
67 #define SET3(a1, a2, a3) (SET2(a1, a2) | SET1(a3))
68 #define SET4(a1, a2, a3, a4) (SET3(a1, a2, a3) | SET1(a4))
69 #define SET5(a1, a2, a3, a4, a5) (SET4(a1, a2, a3, a4) | SET1(a5))
70 #define SET6(a1, a2, a3, a4, a5, a6) (SET5(a1, a2, a3, a4, a5) | SET1(a6))
71 #define SET7(a1, a2, a3, a4, a5, a6, a7) (SET6(a1, a2, a3, a4, a5, a6) | \
72 SET1(a7))
73
74 const MemCmd::CommandInfo
75 MemCmd::commandInfo[] =
76 {
77 /* InvalidCmd */
78 { 0, InvalidCmd, "InvalidCmd" },
79 /* ReadReq - Read issued by a non-caching agent such as a CPU or
80 * device, with no restrictions on alignment. */
81 { SET3(IsRead, IsRequest, NeedsResponse), ReadResp, "ReadReq" },
82 /* ReadResp */
83 { SET3(IsRead, IsResponse, HasData), InvalidCmd, "ReadResp" },
84 /* ReadRespWithInvalidate */
85 { SET4(IsRead, IsResponse, HasData, IsInvalidate),
86 InvalidCmd, "ReadRespWithInvalidate" },
87 /* WriteReq */
88 { SET5(IsWrite, NeedsWritable, IsRequest, NeedsResponse, HasData),
89 WriteResp, "WriteReq" },
90 /* WriteResp */
91 { SET2(IsWrite, IsResponse), InvalidCmd, "WriteResp" },
92 /* WritebackDirty */
93 { SET5(IsWrite, IsRequest, IsEviction, HasData, FromCache),
94 InvalidCmd, "WritebackDirty" },
95 /* WritebackClean - This allows the upstream cache to writeback a
96 * line to the downstream cache without it being considered
97 * dirty. */
98 { SET5(IsWrite, IsRequest, IsEviction, HasData, FromCache),
99 InvalidCmd, "WritebackClean" },
100 /* WriteClean - This allows a cache to write a dirty block to a memory
101 below without evicting its copy. */
102 { SET4(IsWrite, IsRequest, HasData, FromCache), InvalidCmd, "WriteClean" },
103 /* CleanEvict */
104 { SET3(IsRequest, IsEviction, FromCache), InvalidCmd, "CleanEvict" },
105 /* SoftPFReq */
106 { SET4(IsRead, IsRequest, IsSWPrefetch, NeedsResponse),
107 SoftPFResp, "SoftPFReq" },
108 /* SoftPFExReq */
109 { SET6(IsRead, NeedsWritable, IsInvalidate, IsRequest,
110 IsSWPrefetch, NeedsResponse), SoftPFResp, "SoftPFExReq" },
111 /* HardPFReq */
112 { SET5(IsRead, IsRequest, IsHWPrefetch, NeedsResponse, FromCache),
113 HardPFResp, "HardPFReq" },
114 /* SoftPFResp */
115 { SET4(IsRead, IsResponse, IsSWPrefetch, HasData),
116 InvalidCmd, "SoftPFResp" },
117 /* HardPFResp */
118 { SET4(IsRead, IsResponse, IsHWPrefetch, HasData),
119 InvalidCmd, "HardPFResp" },
120 /* WriteLineReq */
121 { SET5(IsWrite, NeedsWritable, IsRequest, NeedsResponse, HasData),
122 WriteResp, "WriteLineReq" },
123 /* UpgradeReq */
124 { SET6(IsInvalidate, NeedsWritable, IsUpgrade, IsRequest, NeedsResponse,
125 FromCache),
126 UpgradeResp, "UpgradeReq" },
127 /* SCUpgradeReq: response could be UpgradeResp or UpgradeFailResp */
128 { SET7(IsInvalidate, NeedsWritable, IsUpgrade, IsLlsc,
129 IsRequest, NeedsResponse, FromCache),
130 UpgradeResp, "SCUpgradeReq" },
131 /* UpgradeResp */
132 { SET2(IsUpgrade, IsResponse),
133 InvalidCmd, "UpgradeResp" },
134 /* SCUpgradeFailReq: generates UpgradeFailResp but still gets the data */
135 { SET7(IsRead, NeedsWritable, IsInvalidate,
136 IsLlsc, IsRequest, NeedsResponse, FromCache),
137 UpgradeFailResp, "SCUpgradeFailReq" },
138 /* UpgradeFailResp - Behaves like a ReadExReq, but notifies an SC
139 * that it has failed, acquires line as Dirty*/
140 { SET3(IsRead, IsResponse, HasData),
141 InvalidCmd, "UpgradeFailResp" },
142 /* ReadExReq - Read issues by a cache, always cache-line aligned,
143 * and the response is guaranteed to be writeable (exclusive or
144 * even modified) */
145 { SET6(IsRead, NeedsWritable, IsInvalidate, IsRequest, NeedsResponse,
146 FromCache),
147 ReadExResp, "ReadExReq" },
148 /* ReadExResp - Response matching a read exclusive, as we check
149 * the need for exclusive also on responses */
150 { SET3(IsRead, IsResponse, HasData),
151 InvalidCmd, "ReadExResp" },
152 /* ReadCleanReq - Read issued by a cache, always cache-line
153 * aligned, and the response is guaranteed to not contain dirty data
154 * (exclusive or shared).*/
155 { SET4(IsRead, IsRequest, NeedsResponse, FromCache),
156 ReadResp, "ReadCleanReq" },
157 /* ReadSharedReq - Read issued by a cache, always cache-line
158 * aligned, response is shared, possibly exclusive, owned or even
159 * modified. */
160 { SET4(IsRead, IsRequest, NeedsResponse, FromCache),
161 ReadResp, "ReadSharedReq" },
162 /* LoadLockedReq: note that we use plain ReadResp as response, so that
163 * we can also use ReadRespWithInvalidate when needed */
164 { SET4(IsRead, IsLlsc, IsRequest, NeedsResponse),
165 ReadResp, "LoadLockedReq" },
166 /* StoreCondReq */
167 { SET6(IsWrite, NeedsWritable, IsLlsc,
168 IsRequest, NeedsResponse, HasData),
169 StoreCondResp, "StoreCondReq" },
170 /* StoreCondFailReq: generates failing StoreCondResp */
171 { SET6(IsWrite, NeedsWritable, IsLlsc,
172 IsRequest, NeedsResponse, HasData),
173 StoreCondResp, "StoreCondFailReq" },
174 /* StoreCondResp */
175 { SET3(IsWrite, IsLlsc, IsResponse),
176 InvalidCmd, "StoreCondResp" },
177 /* SwapReq -- for Swap ldstub type operations */
178 { SET6(IsRead, IsWrite, NeedsWritable, IsRequest, HasData, NeedsResponse),
179 SwapResp, "SwapReq" },
180 /* SwapResp -- for Swap ldstub type operations */
181 { SET4(IsRead, IsWrite, IsResponse, HasData),
182 InvalidCmd, "SwapResp" },
183 { 0, InvalidCmd, "Deprecated_MessageReq" },
184 { 0, InvalidCmd, "Deprecated_MessageResp" },
185 /* MemFenceReq -- for synchronization requests */
186 {SET2(IsRequest, NeedsResponse), MemFenceResp, "MemFenceReq"},
187 /* MemFenceResp -- for synchronization responses */
188 {SET1(IsResponse), InvalidCmd, "MemFenceResp"},
189 /* Cache Clean Request -- Update with the latest data all existing
190 copies of the block down to the point indicated by the
191 request */
192 { SET4(IsRequest, IsClean, NeedsResponse, FromCache),
193 CleanSharedResp, "CleanSharedReq" },
194 /* Cache Clean Response - Indicates that all caches up to the
195 specified point of reference have a up-to-date copy of the
196 cache block or no copy at all */
197 { SET2(IsResponse, IsClean), InvalidCmd, "CleanSharedResp" },
198 /* Cache Clean and Invalidate Request -- Invalidate all existing
199 copies down to the point indicated by the request */
200 { SET5(IsRequest, IsInvalidate, IsClean, NeedsResponse, FromCache),
201 CleanInvalidResp, "CleanInvalidReq" },
202 /* Cache Clean and Invalidate Respose -- Indicates that no cache
203 above the specified point holds the block and that the block
204 was written to a memory below the specified point. */
205 { SET3(IsResponse, IsInvalidate, IsClean),
206 InvalidCmd, "CleanInvalidResp" },
207 /* InvalidDestError -- packet dest field invalid */
208 { SET2(IsResponse, IsError), InvalidCmd, "InvalidDestError" },
209 /* BadAddressError -- memory address invalid */
210 { SET2(IsResponse, IsError), InvalidCmd, "BadAddressError" },
211 /* FunctionalReadError */
212 { SET3(IsRead, IsResponse, IsError), InvalidCmd, "FunctionalReadError" },
213 /* FunctionalWriteError */
214 { SET3(IsWrite, IsResponse, IsError), InvalidCmd, "FunctionalWriteError" },
215 /* PrintReq */
216 { SET2(IsRequest, IsPrint), InvalidCmd, "PrintReq" },
217 /* Flush Request */
218 { SET3(IsRequest, IsFlush, NeedsWritable), InvalidCmd, "FlushReq" },
219 /* Invalidation Request */
220 { SET5(IsInvalidate, IsRequest, NeedsWritable, NeedsResponse, FromCache),
221 InvalidateResp, "InvalidateReq" },
222 /* Invalidation Response */
223 { SET2(IsInvalidate, IsResponse),
224 InvalidCmd, "InvalidateResp" }
225 };
226
227 AddrRange
228 Packet::getAddrRange() const
229 {
230 return RangeSize(getAddr(), getSize());
231 }
232
233 bool
234 Packet::trySatisfyFunctional(Printable *obj, Addr addr, bool is_secure, int size,
235 uint8_t *_data)
236 {
237 const Addr func_start = getAddr();
238 const Addr func_end = getAddr() + getSize() - 1;
239 const Addr val_start = addr;
240 const Addr val_end = val_start + size - 1;
241
242 if (is_secure != _isSecure || func_start > val_end ||
243 val_start > func_end) {
244 // no intersection
245 return false;
246 }
247
248 // check print first since it doesn't require data
249 if (isPrint()) {
250 assert(!_data);
251 safe_cast<PrintReqState*>(senderState)->printObj(obj);
252 return false;
253 }
254
255 // we allow the caller to pass NULL to signify the other packet
256 // has no data
257 if (!_data) {
258 return false;
259 }
260
261 const Addr val_offset = func_start > val_start ?
262 func_start - val_start : 0;
263 const Addr func_offset = func_start < val_start ?
264 val_start - func_start : 0;
265 const Addr overlap_size = std::min(val_end, func_end)+1 -
266 std::max(val_start, func_start);
267
268 if (isRead()) {
269 std::memcpy(getPtr<uint8_t>() + func_offset,
270 _data + val_offset,
271 overlap_size);
272
273 // initialise the tracking of valid bytes if we have not
274 // used it already
275 if (bytesValid.empty())
276 bytesValid.resize(getSize(), false);
277
278 // track if we are done filling the functional access
279 bool all_bytes_valid = true;
280
281 int i = 0;
282
283 // check up to func_offset
284 for (; all_bytes_valid && i < func_offset; ++i)
285 all_bytes_valid &= bytesValid[i];
286
287 // update the valid bytes
288 for (i = func_offset; i < func_offset + overlap_size; ++i)
289 bytesValid[i] = true;
290
291 // check the bit after the update we just made
292 for (; all_bytes_valid && i < getSize(); ++i)
293 all_bytes_valid &= bytesValid[i];
294
295 return all_bytes_valid;
296 } else if (isWrite()) {
297 std::memcpy(_data + val_offset,
298 getConstPtr<uint8_t>() + func_offset,
299 overlap_size);
300 } else {
301 panic("Don't know how to handle command %s\n", cmdString());
302 }
303
304 // keep going with request by default
305 return false;
306 }
307
308 void
309 Packet::copyResponderFlags(const PacketPtr pkt)
310 {
311 assert(isRequest());
312 // If we have already found a responder, no other cache should
313 // commit to responding
314 assert(!pkt->cacheResponding() || !cacheResponding());
315 flags.set(pkt->flags & RESPONDER_FLAGS);
316 }
317
318 void
319 Packet::pushSenderState(Packet::SenderState *sender_state)
320 {
321 assert(sender_state != NULL);
322 sender_state->predecessor = senderState;
323 senderState = sender_state;
324 }
325
326 Packet::SenderState *
327 Packet::popSenderState()
328 {
329 assert(senderState != NULL);
330 SenderState *sender_state = senderState;
331 senderState = sender_state->predecessor;
332 sender_state->predecessor = NULL;
333 return sender_state;
334 }
335
336 uint64_t
337 Packet::getUintX(ByteOrder endian) const
338 {
339 switch(getSize()) {
340 case 1:
341 return (uint64_t)get<uint8_t>(endian);
342 case 2:
343 return (uint64_t)get<uint16_t>(endian);
344 case 4:
345 return (uint64_t)get<uint32_t>(endian);
346 case 8:
347 return (uint64_t)get<uint64_t>(endian);
348 default:
349 panic("%i isn't a supported word size.\n", getSize());
350 }
351 }
352
353 void
354 Packet::setUintX(uint64_t w, ByteOrder endian)
355 {
356 switch(getSize()) {
357 case 1:
358 set<uint8_t>((uint8_t)w, endian);
359 break;
360 case 2:
361 set<uint16_t>((uint16_t)w, endian);
362 break;
363 case 4:
364 set<uint32_t>((uint32_t)w, endian);
365 break;
366 case 8:
367 set<uint64_t>((uint64_t)w, endian);
368 break;
369 default:
370 panic("%i isn't a supported word size.\n", getSize());
371 }
372
373 }
374
375 void
376 Packet::print(std::ostream &o, const int verbosity,
377 const std::string &prefix) const
378 {
379 ccprintf(o, "%s%s [%x:%x]%s%s%s%s%s%s", prefix, cmdString(),
380 getAddr(), getAddr() + getSize() - 1,
381 req->isSecure() ? " (s)" : "",
382 req->isInstFetch() ? " IF" : "",
383 req->isUncacheable() ? " UC" : "",
384 isExpressSnoop() ? " ES" : "",
385 req->isToPOC() ? " PoC" : "",
386 req->isToPOU() ? " PoU" : "");
387 }
388
389 std::string
390 Packet::print() const {
391 std::ostringstream str;
392 print(str);
393 return str.str();
394 }
395
396 bool
397 Packet::matchBlockAddr(const Addr addr, const bool is_secure,
398 const int blk_size) const
399 {
400 return (getBlockAddr(blk_size) == addr) && (isSecure() == is_secure);
401 }
402
403 bool
404 Packet::matchBlockAddr(const PacketPtr pkt, const int blk_size) const
405 {
406 return matchBlockAddr(pkt->getBlockAddr(blk_size), pkt->isSecure(),
407 blk_size);
408 }
409
410 bool
411 Packet::matchAddr(const Addr addr, const bool is_secure) const
412 {
413 return (getAddr() == addr) && (isSecure() == is_secure);
414 }
415
416 bool
417 Packet::matchAddr(const PacketPtr pkt) const
418 {
419 return matchAddr(pkt->getAddr(), pkt->isSecure());
420 }
421
422 Packet::PrintReqState::PrintReqState(std::ostream &_os, int _verbosity)
423 : curPrefixPtr(new std::string("")), os(_os), verbosity(_verbosity)
424 {
425 labelStack.push_back(LabelStackEntry("", curPrefixPtr));
426 }
427
428 Packet::PrintReqState::~PrintReqState()
429 {
430 labelStack.pop_back();
431 assert(labelStack.empty());
432 delete curPrefixPtr;
433 }
434
435 Packet::PrintReqState::
436 LabelStackEntry::LabelStackEntry(const std::string &_label,
437 std::string *_prefix)
438 : label(_label), prefix(_prefix), labelPrinted(false)
439 {
440 }
441
442 void
443 Packet::PrintReqState::pushLabel(const std::string &lbl,
444 const std::string &prefix)
445 {
446 labelStack.push_back(LabelStackEntry(lbl, curPrefixPtr));
447 curPrefixPtr = new std::string(*curPrefixPtr);
448 *curPrefixPtr += prefix;
449 }
450
451 void
452 Packet::PrintReqState::popLabel()
453 {
454 delete curPrefixPtr;
455 curPrefixPtr = labelStack.back().prefix;
456 labelStack.pop_back();
457 assert(!labelStack.empty());
458 }
459
460 void
461 Packet::PrintReqState::printLabels()
462 {
463 if (!labelStack.back().labelPrinted) {
464 LabelStack::iterator i = labelStack.begin();
465 LabelStack::iterator end = labelStack.end();
466 while (i != end) {
467 if (!i->labelPrinted) {
468 ccprintf(os, "%s%s\n", *(i->prefix), i->label);
469 i->labelPrinted = true;
470 }
471 i++;
472 }
473 }
474 }
475
476
477 void
478 Packet::PrintReqState::printObj(Printable *obj)
479 {
480 printLabels();
481 obj->print(os, verbosity, curPrefix());
482 }