arch-arm: Add initial support for SVE contiguous loads/stores
[gem5.git] / src / arch / x86 / decoder.cc
1 /*
2 * Copyright (c) 2011 Google
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Authors: Gabe Black
29 */
30
31 #include "arch/x86/decoder.hh"
32
33 #include "arch/x86/regs/misc.hh"
34 #include "base/logging.hh"
35 #include "base/trace.hh"
36 #include "base/types.hh"
37 #include "debug/Decoder.hh"
38
39 namespace X86ISA
40 {
41
42 Decoder::State
43 Decoder::doResetState()
44 {
45 origPC = basePC + offset;
46 DPRINTF(Decoder, "Setting origPC to %#x\n", origPC);
47 instBytes = &decodePages->lookup(origPC);
48 chunkIdx = 0;
49
50 emi.rex = 0;
51 emi.legacy = 0;
52 emi.vex = 0;
53
54 emi.opcode.type = BadOpcode;
55 emi.opcode.op = 0;
56
57 immediateCollected = 0;
58 emi.immediate = 0;
59 emi.displacement = 0;
60 emi.dispSize = 0;
61
62 emi.modRM = 0;
63 emi.sib = 0;
64
65 if (instBytes->si) {
66 return FromCacheState;
67 } else {
68 instBytes->chunks.clear();
69 return PrefixState;
70 }
71 }
72
73 void
74 Decoder::process()
75 {
76 //This function drives the decoder state machine.
77
78 //Some sanity checks. You shouldn't try to process more bytes if
79 //there aren't any, and you shouldn't overwrite an already
80 //decoder ExtMachInst.
81 assert(!outOfBytes);
82 assert(!instDone);
83
84 if (state == ResetState)
85 state = doResetState();
86 if (state == FromCacheState) {
87 state = doFromCacheState();
88 } else {
89 instBytes->chunks.push_back(fetchChunk);
90 }
91
92 //While there's still something to do...
93 while (!instDone && !outOfBytes) {
94 uint8_t nextByte = getNextByte();
95 switch (state) {
96 case PrefixState:
97 state = doPrefixState(nextByte);
98 break;
99 case Vex2Of2State:
100 state = doVex2Of2State(nextByte);
101 break;
102 case Vex2Of3State:
103 state = doVex2Of3State(nextByte);
104 break;
105 case Vex3Of3State:
106 state = doVex3Of3State(nextByte);
107 break;
108 case VexOpcodeState:
109 state = doVexOpcodeState(nextByte);
110 break;
111 case OneByteOpcodeState:
112 state = doOneByteOpcodeState(nextByte);
113 break;
114 case TwoByteOpcodeState:
115 state = doTwoByteOpcodeState(nextByte);
116 break;
117 case ThreeByte0F38OpcodeState:
118 state = doThreeByte0F38OpcodeState(nextByte);
119 break;
120 case ThreeByte0F3AOpcodeState:
121 state = doThreeByte0F3AOpcodeState(nextByte);
122 break;
123 case ModRMState:
124 state = doModRMState(nextByte);
125 break;
126 case SIBState:
127 state = doSIBState(nextByte);
128 break;
129 case DisplacementState:
130 state = doDisplacementState();
131 break;
132 case ImmediateState:
133 state = doImmediateState();
134 break;
135 case ErrorState:
136 panic("Went to the error state in the decoder.\n");
137 default:
138 panic("Unrecognized state! %d\n", state);
139 }
140 }
141 }
142
143 Decoder::State
144 Decoder::doFromCacheState()
145 {
146 DPRINTF(Decoder, "Looking at cache state.\n");
147 if ((fetchChunk & instBytes->masks[chunkIdx]) !=
148 instBytes->chunks[chunkIdx]) {
149 DPRINTF(Decoder, "Decode cache miss.\n");
150 // The chached chunks didn't match what was fetched. Fall back to the
151 // predecoder.
152 instBytes->chunks[chunkIdx] = fetchChunk;
153 instBytes->chunks.resize(chunkIdx + 1);
154 instBytes->si = NULL;
155 chunkIdx = 0;
156 fetchChunk = instBytes->chunks[0];
157 offset = origPC % sizeof(MachInst);
158 basePC = origPC - offset;
159 return PrefixState;
160 } else if (chunkIdx == instBytes->chunks.size() - 1) {
161 // We matched the cache, so use its value.
162 instDone = true;
163 offset = instBytes->lastOffset;
164 if (offset == sizeof(MachInst))
165 outOfBytes = true;
166 return ResetState;
167 } else {
168 // We matched so far, but need to check more chunks.
169 chunkIdx++;
170 outOfBytes = true;
171 return FromCacheState;
172 }
173 }
174
175 //Either get a prefix and record it in the ExtMachInst, or send the
176 //state machine on to get the opcode(s).
177 Decoder::State
178 Decoder::doPrefixState(uint8_t nextByte)
179 {
180 uint8_t prefix = Prefixes[nextByte];
181 State nextState = PrefixState;
182 // REX prefixes are only recognized in 64 bit mode.
183 if (prefix == RexPrefix && emi.mode.submode != SixtyFourBitMode)
184 prefix = 0;
185 if (prefix)
186 consumeByte();
187 switch(prefix)
188 {
189 //Operand size override prefixes
190 case OperandSizeOverride:
191 DPRINTF(Decoder, "Found operand size override prefix.\n");
192 emi.legacy.op = true;
193 break;
194 case AddressSizeOverride:
195 DPRINTF(Decoder, "Found address size override prefix.\n");
196 emi.legacy.addr = true;
197 break;
198 //Segment override prefixes
199 case CSOverride:
200 case DSOverride:
201 case ESOverride:
202 case FSOverride:
203 case GSOverride:
204 case SSOverride:
205 DPRINTF(Decoder, "Found segment override.\n");
206 emi.legacy.seg = prefix;
207 break;
208 case Lock:
209 DPRINTF(Decoder, "Found lock prefix.\n");
210 emi.legacy.lock = true;
211 break;
212 case Rep:
213 DPRINTF(Decoder, "Found rep prefix.\n");
214 emi.legacy.rep = true;
215 break;
216 case Repne:
217 DPRINTF(Decoder, "Found repne prefix.\n");
218 emi.legacy.repne = true;
219 break;
220 case RexPrefix:
221 DPRINTF(Decoder, "Found Rex prefix %#x.\n", nextByte);
222 emi.rex = nextByte;
223 break;
224 case Vex2Prefix:
225 DPRINTF(Decoder, "Found VEX two-byte prefix %#x.\n", nextByte);
226 emi.vex.present = 1;
227 nextState = Vex2Of2State;
228 break;
229 case Vex3Prefix:
230 DPRINTF(Decoder, "Found VEX three-byte prefix %#x.\n", nextByte);
231 emi.vex.present = 1;
232 nextState = Vex2Of3State;
233 break;
234 case 0:
235 nextState = OneByteOpcodeState;
236 break;
237
238 default:
239 panic("Unrecognized prefix %#x\n", nextByte);
240 }
241 return nextState;
242 }
243
244 Decoder::State
245 Decoder::doVex2Of2State(uint8_t nextByte)
246 {
247 consumeByte();
248 Vex2Of2 vex = nextByte;
249
250 emi.rex.r = !vex.r;
251
252 emi.vex.l = vex.l;
253 emi.vex.v = ~vex.v;
254
255 switch (vex.p) {
256 case 0:
257 break;
258 case 1:
259 emi.legacy.op = 1;
260 break;
261 case 2:
262 emi.legacy.rep = 1;
263 break;
264 case 3:
265 emi.legacy.repne = 1;
266 break;
267 }
268
269 emi.opcode.type = TwoByteOpcode;
270
271 return VexOpcodeState;
272 }
273
274 Decoder::State
275 Decoder::doVex2Of3State(uint8_t nextByte)
276 {
277 if (emi.mode.submode != SixtyFourBitMode && bits(nextByte, 7, 6) == 0x3) {
278 // This was actually an LDS instruction. Reroute to that path.
279 emi.vex.present = 0;
280 emi.opcode.type = OneByteOpcode;
281 emi.opcode.op = 0xC4;
282 return processOpcode(ImmediateTypeOneByte, UsesModRMOneByte,
283 nextByte >= 0xA0 && nextByte <= 0xA3);
284 }
285
286 consumeByte();
287 Vex2Of3 vex = nextByte;
288
289 emi.rex.r = !vex.r;
290 emi.rex.x = !vex.x;
291 emi.rex.b = !vex.b;
292
293 switch (vex.m) {
294 case 1:
295 emi.opcode.type = TwoByteOpcode;
296 break;
297 case 2:
298 emi.opcode.type = ThreeByte0F38Opcode;
299 break;
300 case 3:
301 emi.opcode.type = ThreeByte0F3AOpcode;
302 break;
303 default:
304 // These encodings are reserved. Pretend this was an undefined
305 // instruction so the main decoder will behave correctly, and stop
306 // trying to interpret bytes.
307 emi.opcode.type = TwoByteOpcode;
308 emi.opcode.op = 0x0B;
309 instDone = true;
310 return ResetState;
311 }
312 return Vex3Of3State;
313 }
314
315 Decoder::State
316 Decoder::doVex3Of3State(uint8_t nextByte)
317 {
318 if (emi.mode.submode != SixtyFourBitMode && bits(nextByte, 7, 6) == 0x3) {
319 // This was actually an LES instruction. Reroute to that path.
320 emi.vex.present = 0;
321 emi.opcode.type = OneByteOpcode;
322 emi.opcode.op = 0xC5;
323 return processOpcode(ImmediateTypeOneByte, UsesModRMOneByte,
324 nextByte >= 0xA0 && nextByte <= 0xA3);
325 }
326
327 consumeByte();
328 Vex3Of3 vex = nextByte;
329
330 emi.rex.w = vex.w;
331
332 emi.vex.l = vex.l;
333 emi.vex.v = ~vex.v;
334
335 switch (vex.p) {
336 case 0:
337 break;
338 case 1:
339 emi.legacy.op = 1;
340 break;
341 case 2:
342 emi.legacy.rep = 1;
343 break;
344 case 3:
345 emi.legacy.repne = 1;
346 break;
347 }
348
349 return VexOpcodeState;
350 }
351
352 Decoder::State
353 Decoder::doVexOpcodeState(uint8_t nextByte)
354 {
355 DPRINTF(Decoder, "Found VEX opcode %#x.\n", nextByte);
356
357 emi.opcode.op = nextByte;
358 consumeByte();
359
360 switch (emi.opcode.type) {
361 case TwoByteOpcode:
362 return processOpcode(ImmediateTypeTwoByte, UsesModRMTwoByte);
363 case ThreeByte0F38Opcode:
364 return processOpcode(ImmediateTypeThreeByte0F38,
365 UsesModRMThreeByte0F38);
366 case ThreeByte0F3AOpcode:
367 return processOpcode(ImmediateTypeThreeByte0F3A,
368 UsesModRMThreeByte0F3A);
369 default:
370 panic("Unrecognized opcode type %d.\n", emi.opcode.type);
371 }
372 }
373
374 // Load the first opcode byte. Determine if there are more opcode bytes, and
375 // if not, what immediate and/or ModRM is needed.
376 Decoder::State
377 Decoder::doOneByteOpcodeState(uint8_t nextByte)
378 {
379 State nextState = ErrorState;
380 consumeByte();
381
382 if (nextByte == 0x0f) {
383 DPRINTF(Decoder, "Found opcode escape byte %#x.\n", nextByte);
384 nextState = TwoByteOpcodeState;
385 } else {
386 DPRINTF(Decoder, "Found one byte opcode %#x.\n", nextByte);
387 emi.opcode.type = OneByteOpcode;
388 emi.opcode.op = nextByte;
389
390 nextState = processOpcode(ImmediateTypeOneByte, UsesModRMOneByte,
391 nextByte >= 0xA0 && nextByte <= 0xA3);
392 }
393 return nextState;
394 }
395
396 // Load the second opcode byte. Determine if there are more opcode bytes, and
397 // if not, what immediate and/or ModRM is needed.
398 Decoder::State
399 Decoder::doTwoByteOpcodeState(uint8_t nextByte)
400 {
401 State nextState = ErrorState;
402 consumeByte();
403 if (nextByte == 0x38) {
404 nextState = ThreeByte0F38OpcodeState;
405 DPRINTF(Decoder, "Found opcode escape byte %#x.\n", nextByte);
406 } else if (nextByte == 0x3a) {
407 nextState = ThreeByte0F3AOpcodeState;
408 DPRINTF(Decoder, "Found opcode escape byte %#x.\n", nextByte);
409 } else {
410 DPRINTF(Decoder, "Found two byte opcode %#x.\n", nextByte);
411 emi.opcode.type = TwoByteOpcode;
412 emi.opcode.op = nextByte;
413
414 nextState = processOpcode(ImmediateTypeTwoByte, UsesModRMTwoByte);
415 }
416 return nextState;
417 }
418
419 // Load the third opcode byte and determine what immediate and/or ModRM is
420 // needed.
421 Decoder::State
422 Decoder::doThreeByte0F38OpcodeState(uint8_t nextByte)
423 {
424 consumeByte();
425
426 DPRINTF(Decoder, "Found three byte 0F38 opcode %#x.\n", nextByte);
427 emi.opcode.type = ThreeByte0F38Opcode;
428 emi.opcode.op = nextByte;
429
430 return processOpcode(ImmediateTypeThreeByte0F38, UsesModRMThreeByte0F38);
431 }
432
433 // Load the third opcode byte and determine what immediate and/or ModRM is
434 // needed.
435 Decoder::State
436 Decoder::doThreeByte0F3AOpcodeState(uint8_t nextByte)
437 {
438 consumeByte();
439
440 DPRINTF(Decoder, "Found three byte 0F3A opcode %#x.\n", nextByte);
441 emi.opcode.type = ThreeByte0F3AOpcode;
442 emi.opcode.op = nextByte;
443
444 return processOpcode(ImmediateTypeThreeByte0F3A, UsesModRMThreeByte0F3A);
445 }
446
447 // Generic opcode processing which determines the immediate size, and whether
448 // or not there's a modrm byte.
449 Decoder::State
450 Decoder::processOpcode(ByteTable &immTable, ByteTable &modrmTable,
451 bool addrSizedImm)
452 {
453 State nextState = ErrorState;
454 const uint8_t opcode = emi.opcode.op;
455
456 //Figure out the effective operand size. This can be overriden to
457 //a fixed value at the decoder level.
458 int logOpSize;
459 if (emi.rex.w)
460 logOpSize = 3; // 64 bit operand size
461 else if (emi.legacy.op)
462 logOpSize = altOp;
463 else
464 logOpSize = defOp;
465
466 //Set the actual op size
467 emi.opSize = 1 << logOpSize;
468
469 //Figure out the effective address size. This can be overriden to
470 //a fixed value at the decoder level.
471 int logAddrSize;
472 if (emi.legacy.addr)
473 logAddrSize = altAddr;
474 else
475 logAddrSize = defAddr;
476
477 //Set the actual address size
478 emi.addrSize = 1 << logAddrSize;
479
480 //Figure out the effective stack width. This can be overriden to
481 //a fixed value at the decoder level.
482 emi.stackSize = 1 << stack;
483
484 //Figure out how big of an immediate we'll retreive based
485 //on the opcode.
486 int immType = immTable[opcode];
487 if (addrSizedImm)
488 immediateSize = SizeTypeToSize[logAddrSize - 1][immType];
489 else
490 immediateSize = SizeTypeToSize[logOpSize - 1][immType];
491
492 //Determine what to expect next
493 if (modrmTable[opcode]) {
494 nextState = ModRMState;
495 } else {
496 if (immediateSize) {
497 nextState = ImmediateState;
498 } else {
499 instDone = true;
500 nextState = ResetState;
501 }
502 }
503 return nextState;
504 }
505
506 //Get the ModRM byte and determine what displacement, if any, there is.
507 //Also determine whether or not to get the SIB byte, displacement, or
508 //immediate next.
509 Decoder::State
510 Decoder::doModRMState(uint8_t nextByte)
511 {
512 State nextState = ErrorState;
513 ModRM modRM = nextByte;
514 DPRINTF(Decoder, "Found modrm byte %#x.\n", nextByte);
515 if (defOp == 1) {
516 //figure out 16 bit displacement size
517 if ((modRM.mod == 0 && modRM.rm == 6) || modRM.mod == 2)
518 displacementSize = 2;
519 else if (modRM.mod == 1)
520 displacementSize = 1;
521 else
522 displacementSize = 0;
523 } else {
524 //figure out 32/64 bit displacement size
525 if ((modRM.mod == 0 && modRM.rm == 5) || modRM.mod == 2)
526 displacementSize = 4;
527 else if (modRM.mod == 1)
528 displacementSize = 1;
529 else
530 displacementSize = 0;
531 }
532
533 // The "test" instruction in group 3 needs an immediate, even though
534 // the other instructions with the same actual opcode don't.
535 if (emi.opcode.type == OneByteOpcode && (modRM.reg & 0x6) == 0) {
536 if (emi.opcode.op == 0xF6)
537 immediateSize = 1;
538 else if (emi.opcode.op == 0xF7)
539 immediateSize = (emi.opSize == 8) ? 4 : emi.opSize;
540 }
541
542 //If there's an SIB, get that next.
543 //There is no SIB in 16 bit mode.
544 if (modRM.rm == 4 && modRM.mod != 3) {
545 // && in 32/64 bit mode)
546 nextState = SIBState;
547 } else if (displacementSize) {
548 nextState = DisplacementState;
549 } else if (immediateSize) {
550 nextState = ImmediateState;
551 } else {
552 instDone = true;
553 nextState = ResetState;
554 }
555 //The ModRM byte is consumed no matter what
556 consumeByte();
557 emi.modRM = modRM;
558 return nextState;
559 }
560
561 //Get the SIB byte. We don't do anything with it at this point, other
562 //than storing it in the ExtMachInst. Determine if we need to get a
563 //displacement or immediate next.
564 Decoder::State
565 Decoder::doSIBState(uint8_t nextByte)
566 {
567 State nextState = ErrorState;
568 emi.sib = nextByte;
569 DPRINTF(Decoder, "Found SIB byte %#x.\n", nextByte);
570 consumeByte();
571 if (emi.modRM.mod == 0 && emi.sib.base == 5)
572 displacementSize = 4;
573 if (displacementSize) {
574 nextState = DisplacementState;
575 } else if (immediateSize) {
576 nextState = ImmediateState;
577 } else {
578 instDone = true;
579 nextState = ResetState;
580 }
581 return nextState;
582 }
583
584 //Gather up the displacement, or at least as much of it
585 //as we can get.
586 Decoder::State
587 Decoder::doDisplacementState()
588 {
589 State nextState = ErrorState;
590
591 getImmediate(immediateCollected,
592 emi.displacement,
593 displacementSize);
594
595 DPRINTF(Decoder, "Collecting %d byte displacement, got %d bytes.\n",
596 displacementSize, immediateCollected);
597
598 if (displacementSize == immediateCollected) {
599 //Reset this for other immediates.
600 immediateCollected = 0;
601 //Sign extend the displacement
602 switch(displacementSize)
603 {
604 case 1:
605 emi.displacement = sext<8>(emi.displacement);
606 break;
607 case 2:
608 emi.displacement = sext<16>(emi.displacement);
609 break;
610 case 4:
611 emi.displacement = sext<32>(emi.displacement);
612 break;
613 default:
614 panic("Undefined displacement size!\n");
615 }
616 DPRINTF(Decoder, "Collected displacement %#x.\n",
617 emi.displacement);
618 if (immediateSize) {
619 nextState = ImmediateState;
620 } else {
621 instDone = true;
622 nextState = ResetState;
623 }
624
625 emi.dispSize = displacementSize;
626 }
627 else
628 nextState = DisplacementState;
629 return nextState;
630 }
631
632 //Gather up the immediate, or at least as much of it
633 //as we can get
634 Decoder::State
635 Decoder::doImmediateState()
636 {
637 State nextState = ErrorState;
638
639 getImmediate(immediateCollected,
640 emi.immediate,
641 immediateSize);
642
643 DPRINTF(Decoder, "Collecting %d byte immediate, got %d bytes.\n",
644 immediateSize, immediateCollected);
645
646 if (immediateSize == immediateCollected)
647 {
648 //Reset this for other immediates.
649 immediateCollected = 0;
650
651 //XXX Warning! The following is an observed pattern and might
652 //not always be true!
653
654 //Instructions which use 64 bit operands but 32 bit immediates
655 //need to have the immediate sign extended to 64 bits.
656 //Instructions which use true 64 bit immediates won't be
657 //affected, and instructions that use true 32 bit immediates
658 //won't notice.
659 switch(immediateSize)
660 {
661 case 4:
662 emi.immediate = sext<32>(emi.immediate);
663 break;
664 case 1:
665 emi.immediate = sext<8>(emi.immediate);
666 }
667
668 DPRINTF(Decoder, "Collected immediate %#x.\n",
669 emi.immediate);
670 instDone = true;
671 nextState = ResetState;
672 }
673 else
674 nextState = ImmediateState;
675 return nextState;
676 }
677
678 Decoder::InstBytes Decoder::dummy;
679 Decoder::InstCacheMap Decoder::instCacheMap;
680
681 StaticInstPtr
682 Decoder::decode(ExtMachInst mach_inst, Addr addr)
683 {
684 auto iter = instMap->find(mach_inst);
685 if (iter != instMap->end())
686 return iter->second;
687
688 StaticInstPtr si = decodeInst(mach_inst);
689 (*instMap)[mach_inst] = si;
690 return si;
691 }
692
693 StaticInstPtr
694 Decoder::decode(PCState &nextPC)
695 {
696 if (!instDone)
697 return NULL;
698 instDone = false;
699 updateNPC(nextPC);
700
701 StaticInstPtr &si = instBytes->si;
702 if (si)
703 return si;
704
705 // We didn't match in the AddrMap, but we still populated an entry. Fix
706 // up its byte masks.
707 const int chunkSize = sizeof(MachInst);
708
709 instBytes->lastOffset = offset;
710
711 Addr firstBasePC = basePC - (instBytes->chunks.size() - 1) * chunkSize;
712 Addr firstOffset = origPC - firstBasePC;
713 Addr totalSize = instBytes->lastOffset - firstOffset +
714 (instBytes->chunks.size() - 1) * chunkSize;
715 int start = firstOffset;
716 instBytes->masks.clear();
717
718 while (totalSize) {
719 int end = start + totalSize;
720 end = (chunkSize < end) ? chunkSize : end;
721 int size = end - start;
722 int idx = instBytes->masks.size();
723
724 MachInst maskVal = mask(size * 8) << (start * 8);
725 assert(maskVal);
726
727 instBytes->masks.push_back(maskVal);
728 instBytes->chunks[idx] &= instBytes->masks[idx];
729 totalSize -= size;
730 start = 0;
731 }
732
733 si = decode(emi, origPC);
734 return si;
735 }
736
737 }