Mercurial > hg > truffle
comparison graal/com.oracle.max.asm.amd64/src/com/oracle/max/asm/amd64/X86InstructionDecoder.java @ 6493:85c1b84f8fd9
moved ADM64-specific assembler code into separate project
author | Doug Simon <doug.simon@oracle.com> |
---|---|
date | Tue, 02 Oct 2012 22:22:06 +0200 |
parents | graal/com.oracle.max.asm/src/com/oracle/max/asm/target/amd64/X86InstructionDecoder.java@2f2c6347fce4 |
children |
comparison
equal
deleted
inserted
replaced
6492:dc409418cc2c | 6493:85c1b84f8fd9 |
---|---|
1 /* | |
2 * Copyright (c) 2009, 2011, Oracle and/or its affiliates. All rights reserved. | |
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA | |
20 * or visit www.oracle.com if you need additional information or have any | |
21 * questions. | |
22 */ | |
23 package com.oracle.max.asm.amd64; | |
24 | |
25 | |
26 public final class X86InstructionDecoder { | |
27 | |
28 private boolean targetIs64Bit; | |
29 private byte[] code; | |
30 private int currentEndOfInstruction; | |
31 private int currentDisplacementPosition; | |
32 | |
33 private static class Prefix { | |
34 | |
35 // segment overrides | |
36 public static final int CSSegment = 0x2e; | |
37 public static final int SSSegment = 0x36; | |
38 public static final int DSSegment = 0x3e; | |
39 public static final int ESSegment = 0x26; | |
40 public static final int FSSegment = 0x64; | |
41 public static final int GSSegment = 0x65; | |
42 public static final int REX = 0x40; | |
43 public static final int REXB = 0x41; | |
44 public static final int REXX = 0x42; | |
45 public static final int REXXB = 0x43; | |
46 public static final int REXR = 0x44; | |
47 public static final int REXRB = 0x45; | |
48 public static final int REXRX = 0x46; | |
49 public static final int REXRXB = 0x47; | |
50 public static final int REXW = 0x48; | |
51 public static final int REXWB = 0x49; | |
52 public static final int REXWX = 0x4A; | |
53 public static final int REXWXB = 0x4B; | |
54 public static final int REXWR = 0x4C; | |
55 public static final int REXWRB = 0x4D; | |
56 public static final int REXWRX = 0x4E; | |
57 public static final int REXWRXB = 0x4F; | |
58 } | |
59 | |
60 private X86InstructionDecoder(byte[] code, boolean targetIs64Bit) { | |
61 this.code = code; | |
62 this.targetIs64Bit = targetIs64Bit; | |
63 } | |
64 | |
65 public int currentEndOfInstruction() { | |
66 return currentEndOfInstruction; | |
67 } | |
68 | |
69 public int currentDisplacementPosition() { | |
70 return currentDisplacementPosition; | |
71 } | |
72 | |
73 public void decodePosition(int inst) { | |
74 | |
75 assert inst >= 0 && inst < code.length; | |
76 | |
77 // Decode the given instruction, and return the Pointer of | |
78 // an embedded 32-bit operand word. | |
79 | |
80 // If "which" is WhichOperand.disp32operand, selects the displacement portion | |
81 // of an effective Pointer specifier. | |
82 // If "which" is imm64Operand, selects the trailing immediate constant. | |
83 // If "which" is WhichOperand.call32operand, selects the displacement of a call or jump. | |
84 // Caller is responsible for ensuring that there is such an operand, | |
85 // and that it is 32/64 bits wide. | |
86 | |
87 // If "which" is endPcOperand, find the end of the instruction. | |
88 | |
89 int ip = inst; | |
90 boolean is64bit = false; | |
91 | |
92 boolean hasDisp32 = false; | |
93 int tailSize = 0; // other random bytes (#32, #16, etc.) at end of insn | |
94 | |
95 boolean againAfterPrefix = true; | |
96 | |
97 while (againAfterPrefix) { | |
98 againAfterPrefix = false; | |
99 switch (0xFF & code[ip++]) { | |
100 | |
101 // These convenience macros generate groups of "case" labels for the switch. | |
102 | |
103 case Prefix.CSSegment: | |
104 case Prefix.SSSegment: | |
105 case Prefix.DSSegment: | |
106 case Prefix.ESSegment: | |
107 case Prefix.FSSegment: | |
108 case Prefix.GSSegment: | |
109 // Seems dubious | |
110 assert !targetIs64Bit : "shouldn't have that prefix"; | |
111 assert ip == inst + 1 : "only one prefix allowed"; | |
112 againAfterPrefix = true; | |
113 break; | |
114 | |
115 case 0x67: | |
116 case Prefix.REX: | |
117 case Prefix.REXB: | |
118 case Prefix.REXX: | |
119 case Prefix.REXXB: | |
120 case Prefix.REXR: | |
121 case Prefix.REXRB: | |
122 case Prefix.REXRX: | |
123 case Prefix.REXRXB: | |
124 assert targetIs64Bit : "64bit prefixes"; | |
125 againAfterPrefix = true; | |
126 break; | |
127 | |
128 case Prefix.REXW: | |
129 case Prefix.REXWB: | |
130 case Prefix.REXWX: | |
131 case Prefix.REXWXB: | |
132 case Prefix.REXWR: | |
133 case Prefix.REXWRB: | |
134 case Prefix.REXWRX: | |
135 case Prefix.REXWRXB: | |
136 assert targetIs64Bit : "64bit prefixes"; | |
137 is64bit = true; | |
138 againAfterPrefix = true; | |
139 break; | |
140 | |
141 case 0xFF: // pushq a; decl a; incl a; call a; jmp a | |
142 case 0x88: // movb a, r | |
143 case 0x89: // movl a, r | |
144 case 0x8A: // movb r, a | |
145 case 0x8B: // movl r, a | |
146 case 0x8F: // popl a | |
147 hasDisp32 = true; | |
148 break; | |
149 | |
150 case 0x68: // pushq #32 | |
151 currentEndOfInstruction = ip + 4; | |
152 currentDisplacementPosition = ip; | |
153 return; // not produced by emitOperand | |
154 | |
155 case 0x66: // movw ... (size prefix) | |
156 boolean againAfterSizePrefix2 = true; | |
157 while (againAfterSizePrefix2) { | |
158 againAfterSizePrefix2 = false; | |
159 switch (0xFF & code[ip++]) { | |
160 case Prefix.REX: | |
161 case Prefix.REXB: | |
162 case Prefix.REXX: | |
163 case Prefix.REXXB: | |
164 case Prefix.REXR: | |
165 case Prefix.REXRB: | |
166 case Prefix.REXRX: | |
167 case Prefix.REXRXB: | |
168 case Prefix.REXW: | |
169 case Prefix.REXWB: | |
170 case Prefix.REXWX: | |
171 case Prefix.REXWXB: | |
172 case Prefix.REXWR: | |
173 case Prefix.REXWRB: | |
174 case Prefix.REXWRX: | |
175 case Prefix.REXWRXB: | |
176 assert targetIs64Bit : "64bit prefix found"; | |
177 againAfterSizePrefix2 = true; | |
178 break; | |
179 case 0x8B: // movw r, a | |
180 case 0x89: // movw a, r | |
181 hasDisp32 = true; | |
182 break; | |
183 case 0xC7: // movw a, #16 | |
184 hasDisp32 = true; | |
185 tailSize = 2; // the imm16 | |
186 break; | |
187 case 0x0F: // several SSE/SSE2 variants | |
188 ip--; // reparse the 0x0F | |
189 againAfterPrefix = true; | |
190 break; | |
191 default: | |
192 throw new InternalError("should not reach here"); | |
193 } | |
194 } | |
195 break; | |
196 | |
197 case 0xB8: // movl/q r, #32/#64(oop?) | |
198 case 0xB9: | |
199 case 0xBA: | |
200 case 0xBB: | |
201 case 0xBC: | |
202 case 0xBD: | |
203 case 0xBE: | |
204 case 0xBF: | |
205 currentEndOfInstruction = ip + (is64bit ? 8 : 4); | |
206 currentDisplacementPosition = ip; | |
207 return; | |
208 | |
209 case 0x69: // imul r, a, #32 | |
210 case 0xC7: // movl a, #32(oop?) | |
211 tailSize = 4; | |
212 hasDisp32 = true; // has both kinds of operands! | |
213 break; | |
214 | |
215 case 0x0F: // movx..., etc. | |
216 switch (0xFF & code[ip++]) { | |
217 case 0x12: // movlps | |
218 case 0x28: // movaps | |
219 case 0x2E: // ucomiss | |
220 case 0x2F: // comiss | |
221 case 0x54: // andps | |
222 case 0x55: // andnps | |
223 case 0x56: // orps | |
224 case 0x57: // xorps | |
225 case 0x6E: // movd | |
226 case 0x7E: // movd | |
227 case 0xAE: // ldmxcsr a | |
228 // 64bit side says it these have both operands but that doesn't | |
229 // appear to be true | |
230 hasDisp32 = true; | |
231 break; | |
232 | |
233 case 0xAD: // shrd r, a, %cl | |
234 case 0xAF: // imul r, a | |
235 case 0xBE: // movsbl r, a (movsxb) | |
236 case 0xBF: // movswl r, a (movsxw) | |
237 case 0xB6: // movzbl r, a (movzxb) | |
238 case 0xB7: // movzwl r, a (movzxw) | |
239 case 0x40: // cmovl cc, r, a | |
240 case 0x41: | |
241 case 0x42: | |
242 case 0x43: | |
243 case 0x44: | |
244 case 0x45: | |
245 case 0x46: | |
246 case 0x47: | |
247 case 0x48: | |
248 case 0x49: | |
249 case 0x4A: | |
250 case 0x4B: | |
251 case 0x4C: | |
252 case 0x4D: | |
253 case 0x4E: | |
254 case 0x4F: | |
255 case 0xB0: // cmpxchgb | |
256 case 0xB1: // cmpxchg | |
257 case 0xC1: // xaddl | |
258 case 0xC7: // cmpxchg8 | |
259 case 0x90: // setcc a | |
260 case 0x91: | |
261 case 0x92: | |
262 case 0x93: | |
263 case 0x94: | |
264 case 0x95: | |
265 case 0x96: | |
266 case 0x97: | |
267 case 0x98: | |
268 case 0x99: | |
269 case 0x9A: | |
270 case 0x9B: | |
271 case 0x9C: | |
272 case 0x9D: | |
273 case 0x9E: | |
274 case 0x9F: | |
275 hasDisp32 = true; | |
276 // fall out of the switch to decode the Pointer | |
277 break; | |
278 | |
279 case 0xAC: // shrd r, a, #8 | |
280 hasDisp32 = true; | |
281 tailSize = 1; // the imm8 | |
282 break; | |
283 | |
284 case 0x80: // jcc rdisp32 | |
285 case 0x81: | |
286 case 0x82: | |
287 case 0x83: | |
288 case 0x84: | |
289 case 0x85: | |
290 case 0x86: | |
291 case 0x87: | |
292 case 0x88: | |
293 case 0x89: | |
294 case 0x8A: | |
295 case 0x8B: | |
296 case 0x8C: | |
297 case 0x8D: | |
298 case 0x8E: | |
299 case 0x8F: | |
300 currentEndOfInstruction = ip + 4; | |
301 currentDisplacementPosition = ip; | |
302 return; | |
303 default: | |
304 throw new InternalError("should not reach here"); | |
305 } | |
306 break; | |
307 | |
308 case 0x81: // addl a, #32; addl r, #32 | |
309 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl | |
310 // on 32bit in the case of cmpl, the imm might be an oop | |
311 tailSize = 4; | |
312 hasDisp32 = true; // has both kinds of operands! | |
313 break; | |
314 | |
315 case 0x83: // addl a, #8; addl r, #8 | |
316 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl | |
317 hasDisp32 = true; // has both kinds of operands! | |
318 tailSize = 1; | |
319 break; | |
320 | |
321 case 0x9B: | |
322 switch (0xFF & code[ip++]) { | |
323 case 0xD9: // fnstcw a | |
324 hasDisp32 = true; | |
325 break; | |
326 default: | |
327 throw new InternalError("should not reach here"); | |
328 } | |
329 break; | |
330 | |
331 case 0x00: // addb a, r; addl a, r; addb r, a; addl r, a | |
332 case 0x01: | |
333 case 0x02: | |
334 case 0x03: | |
335 case 0x10: // adc... | |
336 case 0x11: | |
337 case 0x12: | |
338 case 0x13: | |
339 case 0x20: // and... | |
340 case 0x21: | |
341 case 0x22: | |
342 case 0x23: | |
343 case 0x30: // xor... | |
344 case 0x31: | |
345 case 0x32: | |
346 case 0x33: | |
347 case 0x08: // or... | |
348 case 0x09: | |
349 case 0x0a: | |
350 case 0x0b: | |
351 case 0x18: // sbb... | |
352 case 0x19: | |
353 case 0x1a: | |
354 case 0x1b: | |
355 case 0x28: // sub... | |
356 case 0x29: | |
357 case 0x2a: | |
358 case 0x2b: | |
359 case 0xF7: // mull a | |
360 case 0x8D: // lea r, a | |
361 case 0x87: // xchg r, a | |
362 case 0x38: // cmp... | |
363 case 0x39: | |
364 case 0x3a: | |
365 case 0x3b: | |
366 case 0x85: // test r, a | |
367 hasDisp32 = true; // has both kinds of operands! | |
368 break; | |
369 | |
370 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8 | |
371 case 0xC6: // movb a, #8 | |
372 case 0x80: // cmpb a, #8 | |
373 case 0x6B: // imul r, a, #8 | |
374 hasDisp32 = true; // has both kinds of operands! | |
375 tailSize = 1; // the imm8 | |
376 break; | |
377 | |
378 case 0xE8: // call rdisp32 | |
379 case 0xE9: // jmp rdisp32 | |
380 currentEndOfInstruction = ip + 4; | |
381 currentDisplacementPosition = ip; | |
382 return; | |
383 | |
384 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1 | |
385 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl | |
386 case 0xD9: // fldS a; fstS a; fstpS a; fldcw a | |
387 case 0xDD: // fldD a; fstD a; fstpD a | |
388 case 0xDB: // fildS a; fistpS a; fldX a; fstpX a | |
389 case 0xDF: // fildD a; fistpD a | |
390 case 0xD8: // faddS a; fsubrS a; fmulS a; fdivrS a; fcompS a | |
391 case 0xDC: // faddD a; fsubrD a; fmulD a; fdivrD a; fcompD a | |
392 case 0xDE: // faddpD a; fsubrpD a; fmulpD a; fdivrpD a; fcomppD a | |
393 hasDisp32 = true; | |
394 break; | |
395 | |
396 case 0xF0: // Lock | |
397 againAfterPrefix = true; | |
398 break; | |
399 | |
400 case 0xF3: // For SSE | |
401 case 0xF2: // For SSE2 | |
402 switch (0xFF & code[ip++]) { | |
403 case Prefix.REX: | |
404 case Prefix.REXB: | |
405 case Prefix.REXX: | |
406 case Prefix.REXXB: | |
407 case Prefix.REXR: | |
408 case Prefix.REXRB: | |
409 case Prefix.REXRX: | |
410 case Prefix.REXRXB: | |
411 case Prefix.REXW: | |
412 case Prefix.REXWB: | |
413 case Prefix.REXWX: | |
414 case Prefix.REXWXB: | |
415 case Prefix.REXWR: | |
416 case Prefix.REXWRB: | |
417 case Prefix.REXWRX: | |
418 case Prefix.REXWRXB: | |
419 assert targetIs64Bit : "found 64bit prefix"; | |
420 ip++; | |
421 // fall through | |
422 default: | |
423 ip++; | |
424 } | |
425 hasDisp32 = true; // has both kinds of operands! | |
426 break; | |
427 | |
428 default: | |
429 throw new InternalError("should not reach here"); | |
430 } | |
431 } | |
432 | |
433 assert hasDisp32 : "(thomaswue) not sure if this holds: instruction has no disp32 field"; | |
434 | |
435 // parse the output of emitOperand | |
436 int op2 = 0xFF & code[ip++]; | |
437 int base = op2 & 0x07; | |
438 int op3 = -1; | |
439 int b100 = 4; | |
440 int b101 = 5; | |
441 if (base == b100 && (op2 >> 6) != 3) { | |
442 op3 = 0xFF & code[ip++]; | |
443 base = op3 & 0x07; // refetch the base | |
444 } | |
445 // now ip points at the disp (if any) | |
446 | |
447 switch (op2 >> 6) { | |
448 case 0: | |
449 // [00 reg 100][ss index base] | |
450 // [00 reg 100][00 100 esp] | |
451 // [00 reg base] | |
452 // [00 reg 100][ss index 101][disp32] | |
453 // [00 reg 101] [disp32] | |
454 | |
455 if (base == b101) { | |
456 | |
457 currentDisplacementPosition = ip; | |
458 ip += 4; // skip the disp32 | |
459 } | |
460 break; | |
461 | |
462 case 1: | |
463 // [01 reg 100][ss index base][disp8] | |
464 // [01 reg 100][00 100 esp][disp8] | |
465 // [01 reg base] [disp8] | |
466 ip += 1; // skip the disp8 | |
467 break; | |
468 | |
469 case 2: | |
470 // [10 reg 100][ss index base][disp32] | |
471 // [10 reg 100][00 100 esp][disp32] | |
472 // [10 reg base] [disp32] | |
473 currentDisplacementPosition = ip; | |
474 ip += 4; // skip the disp32 | |
475 break; | |
476 | |
477 case 3: | |
478 // [11 reg base] (not a memory addressing mode) | |
479 break; | |
480 } | |
481 | |
482 currentEndOfInstruction = ip + tailSize; | |
483 } | |
484 | |
485 public static void patchRelativeInstruction(byte[] code, int codePos, int relative) { | |
486 X86InstructionDecoder decoder = new X86InstructionDecoder(code, true); | |
487 decoder.decodePosition(codePos); | |
488 int patchPos = decoder.currentDisplacementPosition(); | |
489 int endOfInstruction = decoder.currentEndOfInstruction(); | |
490 int offset = relative - endOfInstruction + codePos; | |
491 patchDisp32(code, patchPos, offset); | |
492 } | |
493 | |
494 private static void patchDisp32(byte[] code, int pos, int offset) { | |
495 assert pos + 4 <= code.length; | |
496 | |
497 assert code[pos] == 0; | |
498 assert code[pos + 1] == 0; | |
499 assert code[pos + 2] == 0; | |
500 assert code[pos + 3] == 0; | |
501 | |
502 code[pos] = (byte) (offset & 0xFF); | |
503 code[pos + 1] = (byte) ((offset >> 8) & 0xFF); | |
504 code[pos + 2] = (byte) ((offset >> 16) & 0xFF); | |
505 code[pos + 3] = (byte) ((offset >> 24) & 0xFF); | |
506 } | |
507 } |