?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qapi: fix block-latency-histogram-set description and examples
[qemu/ar7.git] / tcg / ppc / tcg-target.inc.c
blob773690f1d98870b8e3cdd998d15c831118e2687b
1 /*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "elf.h"
26 #include "tcg-pool.inc.c"
27
28 #if defined _CALL_DARWIN || defined __APPLE__
29 #define TCG_TARGET_CALL_DARWIN
30 #endif
31 #ifdef _CALL_SYSV
32 # define TCG_TARGET_CALL_ALIGN_ARGS 1
33 #endif
34
35 /* For some memory operations, we need a scratch that isn't R0. For the AIX
36 calling convention, we can re-use the TOC register since we'll be reloading
37 it at every call. Otherwise R12 will do nicely as neither a call-saved
38 register nor a parameter register. */
39 #ifdef _CALL_AIX
40 # define TCG_REG_TMP1 TCG_REG_R2
41 #else
42 # define TCG_REG_TMP1 TCG_REG_R12
43 #endif
44
45 #define TCG_REG_TB TCG_REG_R31
46 #define USE_REG_TB (TCG_TARGET_REG_BITS == 64)
47
48 /* Shorthand for size of a pointer. Avoid promotion to unsigned. */
49 #define SZP ((int)sizeof(void *))
50
51 /* Shorthand for size of a register. */
52 #define SZR (TCG_TARGET_REG_BITS / 8)
53
54 #define TCG_CT_CONST_S16 0x100
55 #define TCG_CT_CONST_U16 0x200
56 #define TCG_CT_CONST_S32 0x400
57 #define TCG_CT_CONST_U32 0x800
58 #define TCG_CT_CONST_ZERO 0x1000
59 #define TCG_CT_CONST_MONE 0x2000
60 #define TCG_CT_CONST_WSZ 0x4000
61
62 static tcg_insn_unit *tb_ret_addr;
63
64 bool have_isa_2_06;
65 bool have_isa_3_00;
66
67 #define HAVE_ISA_2_06 have_isa_2_06
68 #define HAVE_ISEL have_isa_2_06
69
70 #ifndef CONFIG_SOFTMMU
71 #define TCG_GUEST_BASE_REG 30
72 #endif
73
74 #ifdef CONFIG_DEBUG_TCG
75 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
76 "r0",
77 "r1",
78 "r2",
79 "r3",
80 "r4",
81 "r5",
82 "r6",
83 "r7",
84 "r8",
85 "r9",
86 "r10",
87 "r11",
88 "r12",
89 "r13",
90 "r14",
91 "r15",
92 "r16",
93 "r17",
94 "r18",
95 "r19",
96 "r20",
97 "r21",
98 "r22",
99 "r23",
100 "r24",
101 "r25",
102 "r26",
103 "r27",
104 "r28",
105 "r29",
106 "r30",
107 "r31"
108 };
109 #endif
110
111 static const int tcg_target_reg_alloc_order[] = {
112 TCG_REG_R14, /* call saved registers */
113 TCG_REG_R15,
114 TCG_REG_R16,
115 TCG_REG_R17,
116 TCG_REG_R18,
117 TCG_REG_R19,
118 TCG_REG_R20,
119 TCG_REG_R21,
120 TCG_REG_R22,
121 TCG_REG_R23,
122 TCG_REG_R24,
123 TCG_REG_R25,
124 TCG_REG_R26,
125 TCG_REG_R27,
126 TCG_REG_R28,
127 TCG_REG_R29,
128 TCG_REG_R30,
129 TCG_REG_R31,
130 TCG_REG_R12, /* call clobbered, non-arguments */
131 TCG_REG_R11,
132 TCG_REG_R2,
133 TCG_REG_R13,
134 TCG_REG_R10, /* call clobbered, arguments */
135 TCG_REG_R9,
136 TCG_REG_R8,
137 TCG_REG_R7,
138 TCG_REG_R6,
139 TCG_REG_R5,
140 TCG_REG_R4,
141 TCG_REG_R3,
142 };
143
144 static const int tcg_target_call_iarg_regs[] = {
145 TCG_REG_R3,
146 TCG_REG_R4,
147 TCG_REG_R5,
148 TCG_REG_R6,
149 TCG_REG_R7,
150 TCG_REG_R8,
151 TCG_REG_R9,
152 TCG_REG_R10
153 };
154
155 static const int tcg_target_call_oarg_regs[] = {
156 TCG_REG_R3,
157 TCG_REG_R4
158 };
159
160 static const int tcg_target_callee_save_regs[] = {
161 #ifdef TCG_TARGET_CALL_DARWIN
162 TCG_REG_R11,
163 #endif
164 TCG_REG_R14,
165 TCG_REG_R15,
166 TCG_REG_R16,
167 TCG_REG_R17,
168 TCG_REG_R18,
169 TCG_REG_R19,
170 TCG_REG_R20,
171 TCG_REG_R21,
172 TCG_REG_R22,
173 TCG_REG_R23,
174 TCG_REG_R24,
175 TCG_REG_R25,
176 TCG_REG_R26,
177 TCG_REG_R27, /* currently used for the global env */
178 TCG_REG_R28,
179 TCG_REG_R29,
180 TCG_REG_R30,
181 TCG_REG_R31
182 };
183
184 static inline bool in_range_b(tcg_target_long target)
185 {
186 return target == sextract64(target, 0, 26);
187 }
188
189 static uint32_t reloc_pc24_val(tcg_insn_unit *pc, tcg_insn_unit *target)
190 {
191 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
192 tcg_debug_assert(in_range_b(disp));
193 return disp & 0x3fffffc;
194 }
195
196 static bool reloc_pc24(tcg_insn_unit *pc, tcg_insn_unit *target)
197 {
198 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
199 if (in_range_b(disp)) {
200 *pc = (*pc & ~0x3fffffc) | (disp & 0x3fffffc);
201 return true;
202 }
203 return false;
204 }
205
206 static uint16_t reloc_pc14_val(tcg_insn_unit *pc, tcg_insn_unit *target)
207 {
208 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
209 tcg_debug_assert(disp == (int16_t) disp);
210 return disp & 0xfffc;
211 }
212
213 static bool reloc_pc14(tcg_insn_unit *pc, tcg_insn_unit *target)
214 {
215 ptrdiff_t disp = tcg_ptr_byte_diff(target, pc);
216 if (disp == (int16_t) disp) {
217 *pc = (*pc & ~0xfffc) | (disp & 0xfffc);
218 return true;
219 }
220 return false;
221 }
222
223 /* parse target specific constraints */
224 static const char *target_parse_constraint(TCGArgConstraint *ct,
225 const char *ct_str, TCGType type)
226 {
227 switch (*ct_str++) {
228 case 'A': case 'B': case 'C': case 'D':
229 ct->ct |= TCG_CT_REG;
230 tcg_regset_set_reg(ct->u.regs, 3 + ct_str[0] - 'A');
231 break;
232 case 'r':
233 ct->ct |= TCG_CT_REG;
234 ct->u.regs = 0xffffffff;
235 break;
236 case 'L': /* qemu_ld constraint */
237 ct->ct |= TCG_CT_REG;
238 ct->u.regs = 0xffffffff;
239 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
240 #ifdef CONFIG_SOFTMMU
241 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4);
242 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5);
243 #endif
244 break;
245 case 'S': /* qemu_st constraint */
246 ct->ct |= TCG_CT_REG;
247 ct->u.regs = 0xffffffff;
248 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
249 #ifdef CONFIG_SOFTMMU
250 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4);
251 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5);
252 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R6);
253 #endif
254 break;
255 case 'I':
256 ct->ct |= TCG_CT_CONST_S16;
257 break;
258 case 'J':
259 ct->ct |= TCG_CT_CONST_U16;
260 break;
261 case 'M':
262 ct->ct |= TCG_CT_CONST_MONE;
263 break;
264 case 'T':
265 ct->ct |= TCG_CT_CONST_S32;
266 break;
267 case 'U':
268 ct->ct |= TCG_CT_CONST_U32;
269 break;
270 case 'W':
271 ct->ct |= TCG_CT_CONST_WSZ;
272 break;
273 case 'Z':
274 ct->ct |= TCG_CT_CONST_ZERO;
275 break;
276 default:
277 return NULL;
278 }
279 return ct_str;
280 }
281
282 /* test if a constant matches the constraint */
283 static int tcg_target_const_match(tcg_target_long val, TCGType type,
284 const TCGArgConstraint *arg_ct)
285 {
286 int ct = arg_ct->ct;
287 if (ct & TCG_CT_CONST) {
288 return 1;
289 }
290
291 /* The only 32-bit constraint we use aside from
292 TCG_CT_CONST is TCG_CT_CONST_S16. */
293 if (type == TCG_TYPE_I32) {
294 val = (int32_t)val;
295 }
296
297 if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
298 return 1;
299 } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
300 return 1;
301 } else if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
302 return 1;
303 } else if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
304 return 1;
305 } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
306 return 1;
307 } else if ((ct & TCG_CT_CONST_MONE) && val == -1) {
308 return 1;
309 } else if ((ct & TCG_CT_CONST_WSZ)
310 && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
311 return 1;
312 }
313 return 0;
314 }
315
316 #define OPCD(opc) ((opc)<<26)
317 #define XO19(opc) (OPCD(19)|((opc)<<1))
318 #define MD30(opc) (OPCD(30)|((opc)<<2))
319 #define MDS30(opc) (OPCD(30)|((opc)<<1))
320 #define XO31(opc) (OPCD(31)|((opc)<<1))
321 #define XO58(opc) (OPCD(58)|(opc))
322 #define XO62(opc) (OPCD(62)|(opc))
323
324 #define B OPCD( 18)
325 #define BC OPCD( 16)
326 #define LBZ OPCD( 34)
327 #define LHZ OPCD( 40)
328 #define LHA OPCD( 42)
329 #define LWZ OPCD( 32)
330 #define LWZUX XO31( 55)
331 #define STB OPCD( 38)
332 #define STH OPCD( 44)
333 #define STW OPCD( 36)
334
335 #define STD XO62( 0)
336 #define STDU XO62( 1)
337 #define STDX XO31(149)
338
339 #define LD XO58( 0)
340 #define LDX XO31( 21)
341 #define LDU XO58( 1)
342 #define LDUX XO31( 53)
343 #define LWA XO58( 2)
344 #define LWAX XO31(341)
345
346 #define ADDIC OPCD( 12)
347 #define ADDI OPCD( 14)
348 #define ADDIS OPCD( 15)
349 #define ORI OPCD( 24)
350 #define ORIS OPCD( 25)
351 #define XORI OPCD( 26)
352 #define XORIS OPCD( 27)
353 #define ANDI OPCD( 28)
354 #define ANDIS OPCD( 29)
355 #define MULLI OPCD( 7)
356 #define CMPLI OPCD( 10)
357 #define CMPI OPCD( 11)
358 #define SUBFIC OPCD( 8)
359
360 #define LWZU OPCD( 33)
361 #define STWU OPCD( 37)
362
363 #define RLWIMI OPCD( 20)
364 #define RLWINM OPCD( 21)
365 #define RLWNM OPCD( 23)
366
367 #define RLDICL MD30( 0)
368 #define RLDICR MD30( 1)
369 #define RLDIMI MD30( 3)
370 #define RLDCL MDS30( 8)
371
372 #define BCLR XO19( 16)
373 #define BCCTR XO19(528)
374 #define CRAND XO19(257)
375 #define CRANDC XO19(129)
376 #define CRNAND XO19(225)
377 #define CROR XO19(449)
378 #define CRNOR XO19( 33)
379
380 #define EXTSB XO31(954)
381 #define EXTSH XO31(922)
382 #define EXTSW XO31(986)
383 #define ADD XO31(266)
384 #define ADDE XO31(138)
385 #define ADDME XO31(234)
386 #define ADDZE XO31(202)
387 #define ADDC XO31( 10)
388 #define AND XO31( 28)
389 #define SUBF XO31( 40)
390 #define SUBFC XO31( 8)
391 #define SUBFE XO31(136)
392 #define SUBFME XO31(232)
393 #define SUBFZE XO31(200)
394 #define OR XO31(444)
395 #define XOR XO31(316)
396 #define MULLW XO31(235)
397 #define MULHW XO31( 75)
398 #define MULHWU XO31( 11)
399 #define DIVW XO31(491)
400 #define DIVWU XO31(459)
401 #define CMP XO31( 0)
402 #define CMPL XO31( 32)
403 #define LHBRX XO31(790)
404 #define LWBRX XO31(534)
405 #define LDBRX XO31(532)
406 #define STHBRX XO31(918)
407 #define STWBRX XO31(662)
408 #define STDBRX XO31(660)
409 #define MFSPR XO31(339)
410 #define MTSPR XO31(467)
411 #define SRAWI XO31(824)
412 #define NEG XO31(104)
413 #define MFCR XO31( 19)
414 #define MFOCRF (MFCR | (1u << 20))
415 #define NOR XO31(124)
416 #define CNTLZW XO31( 26)
417 #define CNTLZD XO31( 58)
418 #define CNTTZW XO31(538)
419 #define CNTTZD XO31(570)
420 #define CNTPOPW XO31(378)
421 #define CNTPOPD XO31(506)
422 #define ANDC XO31( 60)
423 #define ORC XO31(412)
424 #define EQV XO31(284)
425 #define NAND XO31(476)
426 #define ISEL XO31( 15)
427
428 #define MULLD XO31(233)
429 #define MULHD XO31( 73)
430 #define MULHDU XO31( 9)
431 #define DIVD XO31(489)
432 #define DIVDU XO31(457)
433
434 #define LBZX XO31( 87)
435 #define LHZX XO31(279)
436 #define LHAX XO31(343)
437 #define LWZX XO31( 23)
438 #define STBX XO31(215)
439 #define STHX XO31(407)
440 #define STWX XO31(151)
441
442 #define EIEIO XO31(854)
443 #define HWSYNC XO31(598)
444 #define LWSYNC (HWSYNC | (1u << 21))
445
446 #define SPR(a, b) ((((a)<<5)|(b))<<11)
447 #define LR SPR(8, 0)
448 #define CTR SPR(9, 0)
449
450 #define SLW XO31( 24)
451 #define SRW XO31(536)
452 #define SRAW XO31(792)
453
454 #define SLD XO31( 27)
455 #define SRD XO31(539)
456 #define SRAD XO31(794)
457 #define SRADI XO31(413<<1)
458
459 #define TW XO31( 4)
460 #define TRAP (TW | TO(31))
461
462 #define NOP ORI /* ori 0,0,0 */
463
464 #define RT(r) ((r)<<21)
465 #define RS(r) ((r)<<21)
466 #define RA(r) ((r)<<16)
467 #define RB(r) ((r)<<11)
468 #define TO(t) ((t)<<21)
469 #define SH(s) ((s)<<11)
470 #define MB(b) ((b)<<6)
471 #define ME(e) ((e)<<1)
472 #define BO(o) ((o)<<21)
473 #define MB64(b) ((b)<<5)
474 #define FXM(b) (1 << (19 - (b)))
475
476 #define LK 1
477
478 #define TAB(t, a, b) (RT(t) | RA(a) | RB(b))
479 #define SAB(s, a, b) (RS(s) | RA(a) | RB(b))
480 #define TAI(s, a, i) (RT(s) | RA(a) | ((i) & 0xffff))
481 #define SAI(s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))
482
483 #define BF(n) ((n)<<23)
484 #define BI(n, c) (((c)+((n)*4))<<16)
485 #define BT(n, c) (((c)+((n)*4))<<21)
486 #define BA(n, c) (((c)+((n)*4))<<16)
487 #define BB(n, c) (((c)+((n)*4))<<11)
488 #define BC_(n, c) (((c)+((n)*4))<<6)
489
490 #define BO_COND_TRUE BO(12)
491 #define BO_COND_FALSE BO( 4)
492 #define BO_ALWAYS BO(20)
493
494 enum {
495 CR_LT,
496 CR_GT,
497 CR_EQ,
498 CR_SO
499 };
500
501 static const uint32_t tcg_to_bc[] = {
502 [TCG_COND_EQ] = BC | BI(7, CR_EQ) | BO_COND_TRUE,
503 [TCG_COND_NE] = BC | BI(7, CR_EQ) | BO_COND_FALSE,
504 [TCG_COND_LT] = BC | BI(7, CR_LT) | BO_COND_TRUE,
505 [TCG_COND_GE] = BC | BI(7, CR_LT) | BO_COND_FALSE,
506 [TCG_COND_LE] = BC | BI(7, CR_GT) | BO_COND_FALSE,
507 [TCG_COND_GT] = BC | BI(7, CR_GT) | BO_COND_TRUE,
508 [TCG_COND_LTU] = BC | BI(7, CR_LT) | BO_COND_TRUE,
509 [TCG_COND_GEU] = BC | BI(7, CR_LT) | BO_COND_FALSE,
510 [TCG_COND_LEU] = BC | BI(7, CR_GT) | BO_COND_FALSE,
511 [TCG_COND_GTU] = BC | BI(7, CR_GT) | BO_COND_TRUE,
512 };
513
514 /* The low bit here is set if the RA and RB fields must be inverted. */
515 static const uint32_t tcg_to_isel[] = {
516 [TCG_COND_EQ] = ISEL | BC_(7, CR_EQ),
517 [TCG_COND_NE] = ISEL | BC_(7, CR_EQ) | 1,
518 [TCG_COND_LT] = ISEL | BC_(7, CR_LT),
519 [TCG_COND_GE] = ISEL | BC_(7, CR_LT) | 1,
520 [TCG_COND_LE] = ISEL | BC_(7, CR_GT) | 1,
521 [TCG_COND_GT] = ISEL | BC_(7, CR_GT),
522 [TCG_COND_LTU] = ISEL | BC_(7, CR_LT),
523 [TCG_COND_GEU] = ISEL | BC_(7, CR_LT) | 1,
524 [TCG_COND_LEU] = ISEL | BC_(7, CR_GT) | 1,
525 [TCG_COND_GTU] = ISEL | BC_(7, CR_GT),
526 };
527
528 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
529 intptr_t value, intptr_t addend)
530 {
531 tcg_insn_unit *target;
532 tcg_insn_unit old;
533
534 value += addend;
535 target = (tcg_insn_unit *)value;
536
537 switch (type) {
538 case R_PPC_REL14:
539 return reloc_pc14(code_ptr, target);
540 case R_PPC_REL24:
541 return reloc_pc24(code_ptr, target);
542 case R_PPC_ADDR16:
543 /* We are abusing this relocation type. This points to a pair
544 of insns, addis + load. If the displacement is small, we
545 can nop out the addis. */
546 if (value == (int16_t)value) {
547 code_ptr[0] = NOP;
548 old = deposit32(code_ptr[1], 0, 16, value);
549 code_ptr[1] = deposit32(old, 16, 5, TCG_REG_TB);
550 } else {
551 int16_t lo = value;
552 int hi = value - lo;
553 if (hi + lo != value) {
554 return false;
555 }
556 code_ptr[0] = deposit32(code_ptr[0], 0, 16, hi >> 16);
557 code_ptr[1] = deposit32(code_ptr[1], 0, 16, lo);
558 }
559 break;
560 default:
561 g_assert_not_reached();
562 }
563 return true;
564 }
565
566 static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
567 TCGReg base, tcg_target_long offset);
568
569 static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
570 {
571 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
572 if (ret != arg) {
573 tcg_out32(s, OR | SAB(arg, ret, arg));
574 }
575 }
576
577 static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,
578 int sh, int mb)
579 {
580 tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
581 sh = SH(sh & 0x1f) | (((sh >> 5) & 1) << 1);
582 mb = MB64((mb >> 5) | ((mb << 1) & 0x3f));
583 tcg_out32(s, op | RA(ra) | RS(rs) | sh | mb);
584 }
585
586 static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,
587 int sh, int mb, int me)
588 {
589 tcg_out32(s, op | RA(ra) | RS(rs) | SH(sh) | MB(mb) | ME(me));
590 }
591
592 static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)
593 {
594 tcg_out_rld(s, RLDICL, dst, src, 0, 32);
595 }
596
597 static inline void tcg_out_shli32(TCGContext *s, TCGReg dst, TCGReg src, int c)
598 {
599 tcg_out_rlw(s, RLWINM, dst, src, c, 0, 31 - c);
600 }
601
602 static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)
603 {
604 tcg_out_rld(s, RLDICR, dst, src, c, 63 - c);
605 }
606
607 static inline void tcg_out_shri32(TCGContext *s, TCGReg dst, TCGReg src, int c)
608 {
609 tcg_out_rlw(s, RLWINM, dst, src, 32 - c, c, 31);
610 }
611
612 static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)
613 {
614 tcg_out_rld(s, RLDICL, dst, src, 64 - c, c);
615 }
616
617 /* Emit a move into ret of arg, if it can be done in one insn. */
618 static bool tcg_out_movi_one(TCGContext *s, TCGReg ret, tcg_target_long arg)
619 {
620 if (arg == (int16_t)arg) {
621 tcg_out32(s, ADDI | TAI(ret, 0, arg));
622 return true;
623 }
624 if (arg == (int32_t)arg && (arg & 0xffff) == 0) {
625 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
626 return true;
627 }
628 return false;
629 }
630
631 static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
632 tcg_target_long arg, bool in_prologue)
633 {
634 intptr_t tb_diff;
635 tcg_target_long tmp;
636 int shift;
637
638 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
639
640 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
641 arg = (int32_t)arg;
642 }
643
644 /* Load 16-bit immediates with one insn. */
645 if (tcg_out_movi_one(s, ret, arg)) {
646 return;
647 }
648
649 /* Load addresses within the TB with one insn. */
650 tb_diff = arg - (intptr_t)s->code_gen_ptr;
651 if (!in_prologue && USE_REG_TB && tb_diff == (int16_t)tb_diff) {
652 tcg_out32(s, ADDI | TAI(ret, TCG_REG_TB, tb_diff));
653 return;
654 }
655
656 /* Load 32-bit immediates with two insns. Note that we've already
657 eliminated bare ADDIS, so we know both insns are required. */
658 if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
659 tcg_out32(s, ADDIS | TAI(ret, 0, arg >> 16));
660 tcg_out32(s, ORI | SAI(ret, ret, arg));
661 return;
662 }
663 if (arg == (uint32_t)arg && !(arg & 0x8000)) {
664 tcg_out32(s, ADDI | TAI(ret, 0, arg));
665 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
666 return;
667 }
668
669 /* Load masked 16-bit value. */
670 if (arg > 0 && (arg & 0x8000)) {
671 tmp = arg | 0x7fff;
672 if ((tmp & (tmp + 1)) == 0) {
673 int mb = clz64(tmp + 1) + 1;
674 tcg_out32(s, ADDI | TAI(ret, 0, arg));
675 tcg_out_rld(s, RLDICL, ret, ret, 0, mb);
676 return;
677 }
678 }
679
680 /* Load common masks with 2 insns. */
681 shift = ctz64(arg);
682 tmp = arg >> shift;
683 if (tmp == (int16_t)tmp) {
684 tcg_out32(s, ADDI | TAI(ret, 0, tmp));
685 tcg_out_shli64(s, ret, ret, shift);
686 return;
687 }
688 shift = clz64(arg);
689 if (tcg_out_movi_one(s, ret, arg << shift)) {
690 tcg_out_shri64(s, ret, ret, shift);
691 return;
692 }
693
694 /* Load addresses within 2GB of TB with 2 (or rarely 3) insns. */
695 if (!in_prologue && USE_REG_TB && tb_diff == (int32_t)tb_diff) {
696 tcg_out_mem_long(s, ADDI, ADD, ret, TCG_REG_TB, tb_diff);
697 return;
698 }
699
700 /* Use the constant pool, if possible. */
701 if (!in_prologue && USE_REG_TB) {
702 new_pool_label(s, arg, R_PPC_ADDR16, s->code_ptr,
703 -(intptr_t)s->code_gen_ptr);
704 tcg_out32(s, ADDIS | TAI(ret, TCG_REG_TB, 0));
705 tcg_out32(s, LD | TAI(ret, ret, 0));
706 return;
707 }
708
709 tmp = arg >> 31 >> 1;
710 tcg_out_movi(s, TCG_TYPE_I32, ret, tmp);
711 if (tmp) {
712 tcg_out_shli64(s, ret, ret, 32);
713 }
714 if (arg & 0xffff0000) {
715 tcg_out32(s, ORIS | SAI(ret, ret, arg >> 16));
716 }
717 if (arg & 0xffff) {
718 tcg_out32(s, ORI | SAI(ret, ret, arg));
719 }
720 }
721
722 static inline void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,
723 tcg_target_long arg)
724 {
725 tcg_out_movi_int(s, type, ret, arg, false);
726 }
727
728 static bool mask_operand(uint32_t c, int *mb, int *me)
729 {
730 uint32_t lsb, test;
731
732 /* Accept a bit pattern like:
733 0....01....1
734 1....10....0
735 0..01..10..0
736 Keep track of the transitions. */
737 if (c == 0 || c == -1) {
738 return false;
739 }
740 test = c;
741 lsb = test & -test;
742 test += lsb;
743 if (test & (test - 1)) {
744 return false;
745 }
746
747 *me = clz32(lsb);
748 *mb = test ? clz32(test & -test) + 1 : 0;
749 return true;
750 }
751
752 static bool mask64_operand(uint64_t c, int *mb, int *me)
753 {
754 uint64_t lsb;
755
756 if (c == 0) {
757 return false;
758 }
759
760 lsb = c & -c;
761 /* Accept 1..10..0. */
762 if (c == -lsb) {
763 *mb = 0;
764 *me = clz64(lsb);
765 return true;
766 }
767 /* Accept 0..01..1. */
768 if (lsb == 1 && (c & (c + 1)) == 0) {
769 *mb = clz64(c + 1) + 1;
770 *me = 63;
771 return true;
772 }
773 return false;
774 }
775
776 static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
777 {
778 int mb, me;
779
780 if (mask_operand(c, &mb, &me)) {
781 tcg_out_rlw(s, RLWINM, dst, src, 0, mb, me);
782 } else if ((c & 0xffff) == c) {
783 tcg_out32(s, ANDI | SAI(src, dst, c));
784 return;
785 } else if ((c & 0xffff0000) == c) {
786 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
787 return;
788 } else {
789 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R0, c);
790 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
791 }
792 }
793
794 static void tcg_out_andi64(TCGContext *s, TCGReg dst, TCGReg src, uint64_t c)
795 {
796 int mb, me;
797
798 tcg_debug_assert(TCG_TARGET_REG_BITS == 64);
799 if (mask64_operand(c, &mb, &me)) {
800 if (mb == 0) {
801 tcg_out_rld(s, RLDICR, dst, src, 0, me);
802 } else {
803 tcg_out_rld(s, RLDICL, dst, src, 0, mb);
804 }
805 } else if ((c & 0xffff) == c) {
806 tcg_out32(s, ANDI | SAI(src, dst, c));
807 return;
808 } else if ((c & 0xffff0000) == c) {
809 tcg_out32(s, ANDIS | SAI(src, dst, c >> 16));
810 return;
811 } else {
812 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, c);
813 tcg_out32(s, AND | SAB(src, dst, TCG_REG_R0));
814 }
815 }
816
817 static void tcg_out_zori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c,
818 int op_lo, int op_hi)
819 {
820 if (c >> 16) {
821 tcg_out32(s, op_hi | SAI(src, dst, c >> 16));
822 src = dst;
823 }
824 if (c & 0xffff) {
825 tcg_out32(s, op_lo | SAI(src, dst, c));
826 src = dst;
827 }
828 }
829
830 static void tcg_out_ori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
831 {
832 tcg_out_zori32(s, dst, src, c, ORI, ORIS);
833 }
834
835 static void tcg_out_xori32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c)
836 {
837 tcg_out_zori32(s, dst, src, c, XORI, XORIS);
838 }
839
840 static void tcg_out_b(TCGContext *s, int mask, tcg_insn_unit *target)
841 {
842 ptrdiff_t disp = tcg_pcrel_diff(s, target);
843 if (in_range_b(disp)) {
844 tcg_out32(s, B | (disp & 0x3fffffc) | mask);
845 } else {
846 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, (uintptr_t)target);
847 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | CTR);
848 tcg_out32(s, BCCTR | BO_ALWAYS | mask);
849 }
850 }
851
852 static void tcg_out_mem_long(TCGContext *s, int opi, int opx, TCGReg rt,
853 TCGReg base, tcg_target_long offset)
854 {
855 tcg_target_long orig = offset, l0, l1, extra = 0, align = 0;
856 bool is_store = false;
857 TCGReg rs = TCG_REG_TMP1;
858
859 switch (opi) {
860 case LD: case LWA:
861 align = 3;
862 /* FALLTHRU */
863 default:
864 if (rt != TCG_REG_R0) {
865 rs = rt;
866 break;
867 }
868 break;
869 case STD:
870 align = 3;
871 /* FALLTHRU */
872 case STB: case STH: case STW:
873 is_store = true;
874 break;
875 }
876
877 /* For unaligned, or very large offsets, use the indexed form. */
878 if (offset & align || offset != (int32_t)offset) {
879 if (rs == base) {
880 rs = TCG_REG_R0;
881 }
882 tcg_debug_assert(!is_store || rs != rt);
883 tcg_out_movi(s, TCG_TYPE_PTR, rs, orig);
884 tcg_out32(s, opx | TAB(rt, base, rs));
885 return;
886 }
887
888 l0 = (int16_t)offset;
889 offset = (offset - l0) >> 16;
890 l1 = (int16_t)offset;
891
892 if (l1 < 0 && orig >= 0) {
893 extra = 0x4000;
894 l1 = (int16_t)(offset - 0x4000);
895 }
896 if (l1) {
897 tcg_out32(s, ADDIS | TAI(rs, base, l1));
898 base = rs;
899 }
900 if (extra) {
901 tcg_out32(s, ADDIS | TAI(rs, base, extra));
902 base = rs;
903 }
904 if (opi != ADDI || base != rt || l0 != 0) {
905 tcg_out32(s, opi | TAI(rt, base, l0));
906 }
907 }
908
909 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
910 TCGReg arg1, intptr_t arg2)
911 {
912 int opi, opx;
913
914 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
915 if (type == TCG_TYPE_I32) {
916 opi = LWZ, opx = LWZX;
917 } else {
918 opi = LD, opx = LDX;
919 }
920 tcg_out_mem_long(s, opi, opx, ret, arg1, arg2);
921 }
922
923 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
924 TCGReg arg1, intptr_t arg2)
925 {
926 int opi, opx;
927
928 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
929 if (type == TCG_TYPE_I32) {
930 opi = STW, opx = STWX;
931 } else {
932 opi = STD, opx = STDX;
933 }
934 tcg_out_mem_long(s, opi, opx, arg, arg1, arg2);
935 }
936
937 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
938 TCGReg base, intptr_t ofs)
939 {
940 return false;
941 }
942
943 static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,
944 int const_arg2, int cr, TCGType type)
945 {
946 int imm;
947 uint32_t op;
948
949 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
950
951 /* Simplify the comparisons below wrt CMPI. */
952 if (type == TCG_TYPE_I32) {
953 arg2 = (int32_t)arg2;
954 }
955
956 switch (cond) {
957 case TCG_COND_EQ:
958 case TCG_COND_NE:
959 if (const_arg2) {
960 if ((int16_t) arg2 == arg2) {
961 op = CMPI;
962 imm = 1;
963 break;
964 } else if ((uint16_t) arg2 == arg2) {
965 op = CMPLI;
966 imm = 1;
967 break;
968 }
969 }
970 op = CMPL;
971 imm = 0;
972 break;
973
974 case TCG_COND_LT:
975 case TCG_COND_GE:
976 case TCG_COND_LE:
977 case TCG_COND_GT:
978 if (const_arg2) {
979 if ((int16_t) arg2 == arg2) {
980 op = CMPI;
981 imm = 1;
982 break;
983 }
984 }
985 op = CMP;
986 imm = 0;
987 break;
988
989 case TCG_COND_LTU:
990 case TCG_COND_GEU:
991 case TCG_COND_LEU:
992 case TCG_COND_GTU:
993 if (const_arg2) {
994 if ((uint16_t) arg2 == arg2) {
995 op = CMPLI;
996 imm = 1;
997 break;
998 }
999 }
1000 op = CMPL;
1001 imm = 0;
1002 break;
1003
1004 default:
1005 tcg_abort();
1006 }
1007 op |= BF(cr) | ((type == TCG_TYPE_I64) << 21);
1008
1009 if (imm) {
1010 tcg_out32(s, op | RA(arg1) | (arg2 & 0xffff));
1011 } else {
1012 if (const_arg2) {
1013 tcg_out_movi(s, type, TCG_REG_R0, arg2);
1014 arg2 = TCG_REG_R0;
1015 }
1016 tcg_out32(s, op | RA(arg1) | RB(arg2));
1017 }
1018 }
1019
1020 static void tcg_out_setcond_eq0(TCGContext *s, TCGType type,
1021 TCGReg dst, TCGReg src)
1022 {
1023 if (type == TCG_TYPE_I32) {
1024 tcg_out32(s, CNTLZW | RS(src) | RA(dst));
1025 tcg_out_shri32(s, dst, dst, 5);
1026 } else {
1027 tcg_out32(s, CNTLZD | RS(src) | RA(dst));
1028 tcg_out_shri64(s, dst, dst, 6);
1029 }
1030 }
1031
1032 static void tcg_out_setcond_ne0(TCGContext *s, TCGReg dst, TCGReg src)
1033 {
1034 /* X != 0 implies X + -1 generates a carry. Extra addition
1035 trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */
1036 if (dst != src) {
1037 tcg_out32(s, ADDIC | TAI(dst, src, -1));
1038 tcg_out32(s, SUBFE | TAB(dst, dst, src));
1039 } else {
1040 tcg_out32(s, ADDIC | TAI(TCG_REG_R0, src, -1));
1041 tcg_out32(s, SUBFE | TAB(dst, TCG_REG_R0, src));
1042 }
1043 }
1044
1045 static TCGReg tcg_gen_setcond_xor(TCGContext *s, TCGReg arg1, TCGArg arg2,
1046 bool const_arg2)
1047 {
1048 if (const_arg2) {
1049 if ((uint32_t)arg2 == arg2) {
1050 tcg_out_xori32(s, TCG_REG_R0, arg1, arg2);
1051 } else {
1052 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_R0, arg2);
1053 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, TCG_REG_R0));
1054 }
1055 } else {
1056 tcg_out32(s, XOR | SAB(arg1, TCG_REG_R0, arg2));
1057 }
1058 return TCG_REG_R0;
1059 }
1060
1061 static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
1062 TCGArg arg0, TCGArg arg1, TCGArg arg2,
1063 int const_arg2)
1064 {
1065 int crop, sh;
1066
1067 tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32);
1068
1069 /* Ignore high bits of a potential constant arg2. */
1070 if (type == TCG_TYPE_I32) {
1071 arg2 = (uint32_t)arg2;
1072 }
1073
1074 /* Handle common and trivial cases before handling anything else. */
1075 if (arg2 == 0) {
1076 switch (cond) {
1077 case TCG_COND_EQ:
1078 tcg_out_setcond_eq0(s, type, arg0, arg1);
1079 return;
1080 case TCG_COND_NE:
1081 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1082 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1083 arg1 = TCG_REG_R0;
1084 }
1085 tcg_out_setcond_ne0(s, arg0, arg1);
1086 return;
1087 case TCG_COND_GE:
1088 tcg_out32(s, NOR | SAB(arg1, arg0, arg1));
1089 arg1 = arg0;
1090 /* FALLTHRU */
1091 case TCG_COND_LT:
1092 /* Extract the sign bit. */
1093 if (type == TCG_TYPE_I32) {
1094 tcg_out_shri32(s, arg0, arg1, 31);
1095 } else {
1096 tcg_out_shri64(s, arg0, arg1, 63);
1097 }
1098 return;
1099 default:
1100 break;
1101 }
1102 }
1103
1104 /* If we have ISEL, we can implement everything with 3 or 4 insns.
1105 All other cases below are also at least 3 insns, so speed up the
1106 code generator by not considering them and always using ISEL. */
1107 if (HAVE_ISEL) {
1108 int isel, tab;
1109
1110 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1111
1112 isel = tcg_to_isel[cond];
1113
1114 tcg_out_movi(s, type, arg0, 1);
1115 if (isel & 1) {
1116 /* arg0 = (bc ? 0 : 1) */
1117 tab = TAB(arg0, 0, arg0);
1118 isel &= ~1;
1119 } else {
1120 /* arg0 = (bc ? 1 : 0) */
1121 tcg_out_movi(s, type, TCG_REG_R0, 0);
1122 tab = TAB(arg0, arg0, TCG_REG_R0);
1123 }
1124 tcg_out32(s, isel | tab);
1125 return;
1126 }
1127
1128 switch (cond) {
1129 case TCG_COND_EQ:
1130 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1131 tcg_out_setcond_eq0(s, type, arg0, arg1);
1132 return;
1133
1134 case TCG_COND_NE:
1135 arg1 = tcg_gen_setcond_xor(s, arg1, arg2, const_arg2);
1136 /* Discard the high bits only once, rather than both inputs. */
1137 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) {
1138 tcg_out_ext32u(s, TCG_REG_R0, arg1);
1139 arg1 = TCG_REG_R0;
1140 }
1141 tcg_out_setcond_ne0(s, arg0, arg1);
1142 return;
1143
1144 case TCG_COND_GT:
1145 case TCG_COND_GTU:
1146 sh = 30;
1147 crop = 0;
1148 goto crtest;
1149
1150 case TCG_COND_LT:
1151 case TCG_COND_LTU:
1152 sh = 29;
1153 crop = 0;
1154 goto crtest;
1155
1156 case TCG_COND_GE:
1157 case TCG_COND_GEU:
1158 sh = 31;
1159 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_LT) | BB(7, CR_LT);
1160 goto crtest;
1161
1162 case TCG_COND_LE:
1163 case TCG_COND_LEU:
1164 sh = 31;
1165 crop = CRNOR | BT(7, CR_EQ) | BA(7, CR_GT) | BB(7, CR_GT);
1166 crtest:
1167 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1168 if (crop) {
1169 tcg_out32(s, crop);
1170 }
1171 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1172 tcg_out_rlw(s, RLWINM, arg0, TCG_REG_R0, sh, 31, 31);
1173 break;
1174
1175 default:
1176 tcg_abort();
1177 }
1178 }
1179
1180 static void tcg_out_bc(TCGContext *s, int bc, TCGLabel *l)
1181 {
1182 if (l->has_value) {
1183 bc |= reloc_pc14_val(s->code_ptr, l->u.value_ptr);
1184 } else {
1185 tcg_out_reloc(s, s->code_ptr, R_PPC_REL14, l, 0);
1186 }
1187 tcg_out32(s, bc);
1188 }
1189
1190 static void tcg_out_brcond(TCGContext *s, TCGCond cond,
1191 TCGArg arg1, TCGArg arg2, int const_arg2,
1192 TCGLabel *l, TCGType type)
1193 {
1194 tcg_out_cmp(s, cond, arg1, arg2, const_arg2, 7, type);
1195 tcg_out_bc(s, tcg_to_bc[cond], l);
1196 }
1197
1198 static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1199 TCGArg dest, TCGArg c1, TCGArg c2, TCGArg v1,
1200 TCGArg v2, bool const_c2)
1201 {
1202 /* If for some reason both inputs are zero, don't produce bad code. */
1203 if (v1 == 0 && v2 == 0) {
1204 tcg_out_movi(s, type, dest, 0);
1205 return;
1206 }
1207
1208 tcg_out_cmp(s, cond, c1, c2, const_c2, 7, type);
1209
1210 if (HAVE_ISEL) {
1211 int isel = tcg_to_isel[cond];
1212
1213 /* Swap the V operands if the operation indicates inversion. */
1214 if (isel & 1) {
1215 int t = v1;
1216 v1 = v2;
1217 v2 = t;
1218 isel &= ~1;
1219 }
1220 /* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */
1221 if (v2 == 0) {
1222 tcg_out_movi(s, type, TCG_REG_R0, 0);
1223 }
1224 tcg_out32(s, isel | TAB(dest, v1, v2));
1225 } else {
1226 if (dest == v2) {
1227 cond = tcg_invert_cond(cond);
1228 v2 = v1;
1229 } else if (dest != v1) {
1230 if (v1 == 0) {
1231 tcg_out_movi(s, type, dest, 0);
1232 } else {
1233 tcg_out_mov(s, type, dest, v1);
1234 }
1235 }
1236 /* Branch forward over one insn */
1237 tcg_out32(s, tcg_to_bc[cond] | 8);
1238 if (v2 == 0) {
1239 tcg_out_movi(s, type, dest, 0);
1240 } else {
1241 tcg_out_mov(s, type, dest, v2);
1242 }
1243 }
1244 }
1245
1246 static void tcg_out_cntxz(TCGContext *s, TCGType type, uint32_t opc,
1247 TCGArg a0, TCGArg a1, TCGArg a2, bool const_a2)
1248 {
1249 if (const_a2 && a2 == (type == TCG_TYPE_I32 ? 32 : 64)) {
1250 tcg_out32(s, opc | RA(a0) | RS(a1));
1251 } else {
1252 tcg_out_cmp(s, TCG_COND_EQ, a1, 0, 1, 7, type);
1253 /* Note that the only other valid constant for a2 is 0. */
1254 if (HAVE_ISEL) {
1255 tcg_out32(s, opc | RA(TCG_REG_R0) | RS(a1));
1256 tcg_out32(s, tcg_to_isel[TCG_COND_EQ] | TAB(a0, a2, TCG_REG_R0));
1257 } else if (!const_a2 && a0 == a2) {
1258 tcg_out32(s, tcg_to_bc[TCG_COND_EQ] | 8);
1259 tcg_out32(s, opc | RA(a0) | RS(a1));
1260 } else {
1261 tcg_out32(s, opc | RA(a0) | RS(a1));
1262 tcg_out32(s, tcg_to_bc[TCG_COND_NE] | 8);
1263 if (const_a2) {
1264 tcg_out_movi(s, type, a0, 0);
1265 } else {
1266 tcg_out_mov(s, type, a0, a2);
1267 }
1268 }
1269 }
1270 }
1271
1272 static void tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1273 const int *const_args)
1274 {
1275 static const struct { uint8_t bit1, bit2; } bits[] = {
1276 [TCG_COND_LT ] = { CR_LT, CR_LT },
1277 [TCG_COND_LE ] = { CR_LT, CR_GT },
1278 [TCG_COND_GT ] = { CR_GT, CR_GT },
1279 [TCG_COND_GE ] = { CR_GT, CR_LT },
1280 [TCG_COND_LTU] = { CR_LT, CR_LT },
1281 [TCG_COND_LEU] = { CR_LT, CR_GT },
1282 [TCG_COND_GTU] = { CR_GT, CR_GT },
1283 [TCG_COND_GEU] = { CR_GT, CR_LT },
1284 };
1285
1286 TCGCond cond = args[4], cond2;
1287 TCGArg al, ah, bl, bh;
1288 int blconst, bhconst;
1289 int op, bit1, bit2;
1290
1291 al = args[0];
1292 ah = args[1];
1293 bl = args[2];
1294 bh = args[3];
1295 blconst = const_args[2];
1296 bhconst = const_args[3];
1297
1298 switch (cond) {
1299 case TCG_COND_EQ:
1300 op = CRAND;
1301 goto do_equality;
1302 case TCG_COND_NE:
1303 op = CRNAND;
1304 do_equality:
1305 tcg_out_cmp(s, cond, al, bl, blconst, 6, TCG_TYPE_I32);
1306 tcg_out_cmp(s, cond, ah, bh, bhconst, 7, TCG_TYPE_I32);
1307 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1308 break;
1309
1310 case TCG_COND_LT:
1311 case TCG_COND_LE:
1312 case TCG_COND_GT:
1313 case TCG_COND_GE:
1314 case TCG_COND_LTU:
1315 case TCG_COND_LEU:
1316 case TCG_COND_GTU:
1317 case TCG_COND_GEU:
1318 bit1 = bits[cond].bit1;
1319 bit2 = bits[cond].bit2;
1320 op = (bit1 != bit2 ? CRANDC : CRAND);
1321 cond2 = tcg_unsigned_cond(cond);
1322
1323 tcg_out_cmp(s, cond, ah, bh, bhconst, 6, TCG_TYPE_I32);
1324 tcg_out_cmp(s, cond2, al, bl, blconst, 7, TCG_TYPE_I32);
1325 tcg_out32(s, op | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, bit2));
1326 tcg_out32(s, CROR | BT(7, CR_EQ) | BA(6, bit1) | BB(7, CR_EQ));
1327 break;
1328
1329 default:
1330 tcg_abort();
1331 }
1332 }
1333
1334 static void tcg_out_setcond2(TCGContext *s, const TCGArg *args,
1335 const int *const_args)
1336 {
1337 tcg_out_cmp2(s, args + 1, const_args + 1);
1338 tcg_out32(s, MFOCRF | RT(TCG_REG_R0) | FXM(7));
1339 tcg_out_rlw(s, RLWINM, args[0], TCG_REG_R0, 31, 31, 31);
1340 }
1341
1342 static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
1343 const int *const_args)
1344 {
1345 tcg_out_cmp2(s, args, const_args);
1346 tcg_out_bc(s, BC | BI(7, CR_EQ) | BO_COND_TRUE, arg_label(args[5]));
1347 }
1348
1349 static void tcg_out_mb(TCGContext *s, TCGArg a0)
1350 {
1351 uint32_t insn = HWSYNC;
1352 a0 &= TCG_MO_ALL;
1353 if (a0 == TCG_MO_LD_LD) {
1354 insn = LWSYNC;
1355 } else if (a0 == TCG_MO_ST_ST) {
1356 insn = EIEIO;
1357 }
1358 tcg_out32(s, insn);
1359 }
1360
1361 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
1362 uintptr_t addr)
1363 {
1364 if (TCG_TARGET_REG_BITS == 64) {
1365 tcg_insn_unit i1, i2;
1366 intptr_t tb_diff = addr - tc_ptr;
1367 intptr_t br_diff = addr - (jmp_addr + 4);
1368 uint64_t pair;
1369
1370 /* This does not exercise the range of the branch, but we do
1371 still need to be able to load the new value of TCG_REG_TB.
1372 But this does still happen quite often. */
1373 if (tb_diff == (int16_t)tb_diff) {
1374 i1 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, tb_diff);
1375 i2 = B | (br_diff & 0x3fffffc);
1376 } else {
1377 intptr_t lo = (int16_t)tb_diff;
1378 intptr_t hi = (int32_t)(tb_diff - lo);
1379 assert(tb_diff == hi + lo);
1380 i1 = ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, hi >> 16);
1381 i2 = ADDI | TAI(TCG_REG_TB, TCG_REG_TB, lo);
1382 }
1383 #ifdef HOST_WORDS_BIGENDIAN
1384 pair = (uint64_t)i1 << 32 | i2;
1385 #else
1386 pair = (uint64_t)i2 << 32 | i1;
1387 #endif
1388
1389 /* As per the enclosing if, this is ppc64. Avoid the _Static_assert
1390 within atomic_set that would fail to build a ppc32 host. */
1391 atomic_set__nocheck((uint64_t *)jmp_addr, pair);
1392 flush_icache_range(jmp_addr, jmp_addr + 8);
1393 } else {
1394 intptr_t diff = addr - jmp_addr;
1395 tcg_debug_assert(in_range_b(diff));
1396 atomic_set((uint32_t *)jmp_addr, B | (diff & 0x3fffffc));
1397 flush_icache_range(jmp_addr, jmp_addr + 4);
1398 }
1399 }
1400
1401 static void tcg_out_call(TCGContext *s, tcg_insn_unit *target)
1402 {
1403 #ifdef _CALL_AIX
1404 /* Look through the descriptor. If the branch is in range, and we
1405 don't have to spend too much effort on building the toc. */
1406 void *tgt = ((void **)target)[0];
1407 uintptr_t toc = ((uintptr_t *)target)[1];
1408 intptr_t diff = tcg_pcrel_diff(s, tgt);
1409
1410 if (in_range_b(diff) && toc == (uint32_t)toc) {
1411 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, toc);
1412 tcg_out_b(s, LK, tgt);
1413 } else {
1414 /* Fold the low bits of the constant into the addresses below. */
1415 intptr_t arg = (intptr_t)target;
1416 int ofs = (int16_t)arg;
1417
1418 if (ofs + 8 < 0x8000) {
1419 arg -= ofs;
1420 } else {
1421 ofs = 0;
1422 }
1423 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP1, arg);
1424 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_TMP1, ofs);
1425 tcg_out32(s, MTSPR | RA(TCG_REG_R0) | CTR);
1426 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_REG_TMP1, ofs + SZP);
1427 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1428 }
1429 #elif defined(_CALL_ELF) && _CALL_ELF == 2
1430 intptr_t diff;
1431
1432 /* In the ELFv2 ABI, we have to set up r12 to contain the destination
1433 address, which the callee uses to compute its TOC address. */
1434 /* FIXME: when the branch is in range, we could avoid r12 load if we
1435 knew that the destination uses the same TOC, and what its local
1436 entry point offset is. */
1437 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R12, (intptr_t)target);
1438
1439 diff = tcg_pcrel_diff(s, target);
1440 if (in_range_b(diff)) {
1441 tcg_out_b(s, LK, target);
1442 } else {
1443 tcg_out32(s, MTSPR | RS(TCG_REG_R12) | CTR);
1444 tcg_out32(s, BCCTR | BO_ALWAYS | LK);
1445 }
1446 #else
1447 tcg_out_b(s, LK, target);
1448 #endif
1449 }
1450
1451 static const uint32_t qemu_ldx_opc[16] = {
1452 [MO_UB] = LBZX,
1453 [MO_UW] = LHZX,
1454 [MO_UL] = LWZX,
1455 [MO_Q] = LDX,
1456 [MO_SW] = LHAX,
1457 [MO_SL] = LWAX,
1458 [MO_BSWAP | MO_UB] = LBZX,
1459 [MO_BSWAP | MO_UW] = LHBRX,
1460 [MO_BSWAP | MO_UL] = LWBRX,
1461 [MO_BSWAP | MO_Q] = LDBRX,
1462 };
1463
1464 static const uint32_t qemu_stx_opc[16] = {
1465 [MO_UB] = STBX,
1466 [MO_UW] = STHX,
1467 [MO_UL] = STWX,
1468 [MO_Q] = STDX,
1469 [MO_BSWAP | MO_UB] = STBX,
1470 [MO_BSWAP | MO_UW] = STHBRX,
1471 [MO_BSWAP | MO_UL] = STWBRX,
1472 [MO_BSWAP | MO_Q] = STDBRX,
1473 };
1474
1475 static const uint32_t qemu_exts_opc[4] = {
1476 EXTSB, EXTSH, EXTSW, 0
1477 };
1478
1479 #if defined (CONFIG_SOFTMMU)
1480 #include "tcg-ldst.inc.c"
1481
1482 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
1483 * int mmu_idx, uintptr_t ra)
1484 */
1485 static void * const qemu_ld_helpers[16] = {
1486 [MO_UB] = helper_ret_ldub_mmu,
1487 [MO_LEUW] = helper_le_lduw_mmu,
1488 [MO_LEUL] = helper_le_ldul_mmu,
1489 [MO_LEQ] = helper_le_ldq_mmu,
1490 [MO_BEUW] = helper_be_lduw_mmu,
1491 [MO_BEUL] = helper_be_ldul_mmu,
1492 [MO_BEQ] = helper_be_ldq_mmu,
1493 };
1494
1495 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1496 * uintxx_t val, int mmu_idx, uintptr_t ra)
1497 */
1498 static void * const qemu_st_helpers[16] = {
1499 [MO_UB] = helper_ret_stb_mmu,
1500 [MO_LEUW] = helper_le_stw_mmu,
1501 [MO_LEUL] = helper_le_stl_mmu,
1502 [MO_LEQ] = helper_le_stq_mmu,
1503 [MO_BEUW] = helper_be_stw_mmu,
1504 [MO_BEUL] = helper_be_stl_mmu,
1505 [MO_BEQ] = helper_be_stq_mmu,
1506 };
1507
1508 /* We expect tlb_mask to be before tlb_table. */
1509 QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
1510 offsetof(CPUArchState, tlb_mask));
1511
1512 /* Perform the TLB load and compare. Places the result of the comparison
1513 in CR7, loads the addend of the TLB into R3, and returns the register
1514 containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
1515
1516 static TCGReg tcg_out_tlb_read(TCGContext *s, TCGMemOp opc,
1517 TCGReg addrlo, TCGReg addrhi,
1518 int mem_index, bool is_read)
1519 {
1520 int cmp_off
1521 = (is_read
1522 ? offsetof(CPUTLBEntry, addr_read)
1523 : offsetof(CPUTLBEntry, addr_write));
1524 int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
1525 int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
1526 TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
1527 unsigned s_bits = opc & MO_SIZE;
1528 unsigned a_bits = get_alignment_bits(opc);
1529
1530 if (table_off > 0x7fff) {
1531 int mask_hi = mask_off - (int16_t)mask_off;
1532 int table_hi = table_off - (int16_t)table_off;
1533
1534 table_base = TCG_REG_R4;
1535 if (mask_hi == table_hi) {
1536 mask_base = table_base;
1537 } else if (mask_hi) {
1538 mask_base = TCG_REG_R3;
1539 tcg_out32(s, ADDIS | TAI(mask_base, TCG_AREG0, mask_hi >> 16));
1540 }
1541 tcg_out32(s, ADDIS | TAI(table_base, TCG_AREG0, table_hi >> 16));
1542 mask_off -= mask_hi;
1543 table_off -= table_hi;
1544 }
1545
1546 /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
1547 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, mask_base, mask_off);
1548 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, table_base, table_off);
1549
1550 /* Extract the page index, shifted into place for tlb index. */
1551 if (TCG_TARGET_REG_BITS == 32) {
1552 tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
1553 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1554 } else {
1555 tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
1556 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1557 }
1558 tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
1559
1560 /* Load the TLB comparator. */
1561 if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
1562 uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
1563 ? LWZUX : LDUX);
1564 tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
1565 } else {
1566 tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
1567 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1568 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
1569 tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
1570 } else {
1571 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
1572 }
1573 }
1574
1575 /* Load the TLB addend for use on the fast path. Do this asap
1576 to minimize any load use delay. */
1577 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
1578 offsetof(CPUTLBEntry, addend));
1579
1580 /* Clear the non-page, non-alignment bits from the address */
1581 if (TCG_TARGET_REG_BITS == 32) {
1582 /* We don't support unaligned accesses on 32-bits.
1583 * Preserve the bottom bits and thus trigger a comparison
1584 * failure on unaligned accesses.
1585 */
1586 if (a_bits < s_bits) {
1587 a_bits = s_bits;
1588 }
1589 tcg_out_rlw(s, RLWINM, TCG_REG_R0, addrlo, 0,
1590 (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1591 } else {
1592 TCGReg t = addrlo;
1593
1594 /* If the access is unaligned, we need to make sure we fail if we
1595 * cross a page boundary. The trick is to add the access size-1
1596 * to the address before masking the low bits. That will make the
1597 * address overflow to the next page if we cross a page boundary,
1598 * which will then force a mismatch of the TLB compare.
1599 */
1600 if (a_bits < s_bits) {
1601 unsigned a_mask = (1 << a_bits) - 1;
1602 unsigned s_mask = (1 << s_bits) - 1;
1603 tcg_out32(s, ADDI | TAI(TCG_REG_R0, t, s_mask - a_mask));
1604 t = TCG_REG_R0;
1605 }
1606
1607 /* Mask the address for the requested alignment. */
1608 if (TARGET_LONG_BITS == 32) {
1609 tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
1610 (32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
1611 /* Zero-extend the address for use in the final address. */
1612 tcg_out_ext32u(s, TCG_REG_R4, addrlo);
1613 addrlo = TCG_REG_R4;
1614 } else if (a_bits == 0) {
1615 tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
1616 } else {
1617 tcg_out_rld(s, RLDICL, TCG_REG_R0, t,
1618 64 - TARGET_PAGE_BITS, TARGET_PAGE_BITS - a_bits);
1619 tcg_out_rld(s, RLDICL, TCG_REG_R0, TCG_REG_R0, TARGET_PAGE_BITS, 0);
1620 }
1621 }
1622
1623 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1624 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1625 0, 7, TCG_TYPE_I32);
1626 tcg_out_cmp(s, TCG_COND_EQ, addrhi, TCG_REG_R4, 0, 6, TCG_TYPE_I32);
1627 tcg_out32(s, CRAND | BT(7, CR_EQ) | BA(6, CR_EQ) | BB(7, CR_EQ));
1628 } else {
1629 tcg_out_cmp(s, TCG_COND_EQ, TCG_REG_R0, TCG_REG_TMP1,
1630 0, 7, TCG_TYPE_TL);
1631 }
1632
1633 return addrlo;
1634 }
1635
1636 /* Record the context of a call to the out of line helper code for the slow
1637 path for a load or store, so that we can later generate the correct
1638 helper code. */
1639 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1640 TCGReg datalo_reg, TCGReg datahi_reg,
1641 TCGReg addrlo_reg, TCGReg addrhi_reg,
1642 tcg_insn_unit *raddr, tcg_insn_unit *lptr)
1643 {
1644 TCGLabelQemuLdst *label = new_ldst_label(s);
1645
1646 label->is_ld = is_ld;
1647 label->oi = oi;
1648 label->datalo_reg = datalo_reg;
1649 label->datahi_reg = datahi_reg;
1650 label->addrlo_reg = addrlo_reg;
1651 label->addrhi_reg = addrhi_reg;
1652 label->raddr = raddr;
1653 label->label_ptr[0] = lptr;
1654 }
1655
1656 static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1657 {
1658 TCGMemOpIdx oi = lb->oi;
1659 TCGMemOp opc = get_memop(oi);
1660 TCGReg hi, lo, arg = TCG_REG_R3;
1661
1662 **lb->label_ptr |= reloc_pc14_val(*lb->label_ptr, s->code_ptr);
1663
1664 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1665
1666 lo = lb->addrlo_reg;
1667 hi = lb->addrhi_reg;
1668 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1669 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1670 arg |= 1;
1671 #endif
1672 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1673 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1674 } else {
1675 /* If the address needed to be zero-extended, we'll have already
1676 placed it in R4. The only remaining case is 64-bit guest. */
1677 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1678 }
1679
1680 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1681 tcg_out32(s, MFSPR | RT(arg) | LR);
1682
1683 tcg_out_call(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1684
1685 lo = lb->datalo_reg;
1686 hi = lb->datahi_reg;
1687 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) {
1688 tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_REG_R4);
1689 tcg_out_mov(s, TCG_TYPE_I32, hi, TCG_REG_R3);
1690 } else if (opc & MO_SIGN) {
1691 uint32_t insn = qemu_exts_opc[opc & MO_SIZE];
1692 tcg_out32(s, insn | RA(lo) | RS(TCG_REG_R3));
1693 } else {
1694 tcg_out_mov(s, TCG_TYPE_REG, lo, TCG_REG_R3);
1695 }
1696
1697 tcg_out_b(s, 0, lb->raddr);
1698 }
1699
1700 static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1701 {
1702 TCGMemOpIdx oi = lb->oi;
1703 TCGMemOp opc = get_memop(oi);
1704 TCGMemOp s_bits = opc & MO_SIZE;
1705 TCGReg hi, lo, arg = TCG_REG_R3;
1706
1707 **lb->label_ptr |= reloc_pc14_val(*lb->label_ptr, s->code_ptr);
1708
1709 tcg_out_mov(s, TCG_TYPE_PTR, arg++, TCG_AREG0);
1710
1711 lo = lb->addrlo_reg;
1712 hi = lb->addrhi_reg;
1713 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
1714 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1715 arg |= 1;
1716 #endif
1717 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1718 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1719 } else {
1720 /* If the address needed to be zero-extended, we'll have already
1721 placed it in R4. The only remaining case is 64-bit guest. */
1722 tcg_out_mov(s, TCG_TYPE_TL, arg++, lo);
1723 }
1724
1725 lo = lb->datalo_reg;
1726 hi = lb->datahi_reg;
1727 if (TCG_TARGET_REG_BITS == 32) {
1728 switch (s_bits) {
1729 case MO_64:
1730 #ifdef TCG_TARGET_CALL_ALIGN_ARGS
1731 arg |= 1;
1732 #endif
1733 tcg_out_mov(s, TCG_TYPE_I32, arg++, hi);
1734 /* FALLTHRU */
1735 case MO_32:
1736 tcg_out_mov(s, TCG_TYPE_I32, arg++, lo);
1737 break;
1738 default:
1739 tcg_out_rlw(s, RLWINM, arg++, lo, 0, 32 - (8 << s_bits), 31);
1740 break;
1741 }
1742 } else {
1743 if (s_bits == MO_64) {
1744 tcg_out_mov(s, TCG_TYPE_I64, arg++, lo);
1745 } else {
1746 tcg_out_rld(s, RLDICL, arg++, lo, 0, 64 - (8 << s_bits));
1747 }
1748 }
1749
1750 tcg_out_movi(s, TCG_TYPE_I32, arg++, oi);
1751 tcg_out32(s, MFSPR | RT(arg) | LR);
1752
1753 tcg_out_call(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1754
1755 tcg_out_b(s, 0, lb->raddr);
1756 }
1757 #endif /* SOFTMMU */
1758
1759 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
1760 {
1761 TCGReg datalo, datahi, addrlo, rbase;
1762 TCGReg addrhi __attribute__((unused));
1763 TCGMemOpIdx oi;
1764 TCGMemOp opc, s_bits;
1765 #ifdef CONFIG_SOFTMMU
1766 int mem_index;
1767 tcg_insn_unit *label_ptr;
1768 #endif
1769
1770 datalo = *args++;
1771 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1772 addrlo = *args++;
1773 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1774 oi = *args++;
1775 opc = get_memop(oi);
1776 s_bits = opc & MO_SIZE;
1777
1778 #ifdef CONFIG_SOFTMMU
1779 mem_index = get_mmuidx(oi);
1780 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, true);
1781
1782 /* Load a pointer into the current opcode w/conditional branch-link. */
1783 label_ptr = s->code_ptr;
1784 tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1785
1786 rbase = TCG_REG_R3;
1787 #else /* !CONFIG_SOFTMMU */
1788 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1789 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1790 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1791 addrlo = TCG_REG_TMP1;
1792 }
1793 #endif
1794
1795 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1796 if (opc & MO_BSWAP) {
1797 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1798 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1799 tcg_out32(s, LWBRX | TAB(datahi, rbase, TCG_REG_R0));
1800 } else if (rbase != 0) {
1801 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1802 tcg_out32(s, LWZX | TAB(datahi, rbase, addrlo));
1803 tcg_out32(s, LWZX | TAB(datalo, rbase, TCG_REG_R0));
1804 } else if (addrlo == datahi) {
1805 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1806 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1807 } else {
1808 tcg_out32(s, LWZ | TAI(datahi, addrlo, 0));
1809 tcg_out32(s, LWZ | TAI(datalo, addrlo, 4));
1810 }
1811 } else {
1812 uint32_t insn = qemu_ldx_opc[opc & (MO_BSWAP | MO_SSIZE)];
1813 if (!HAVE_ISA_2_06 && insn == LDBRX) {
1814 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1815 tcg_out32(s, LWBRX | TAB(datalo, rbase, addrlo));
1816 tcg_out32(s, LWBRX | TAB(TCG_REG_R0, rbase, TCG_REG_R0));
1817 tcg_out_rld(s, RLDIMI, datalo, TCG_REG_R0, 32, 0);
1818 } else if (insn) {
1819 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1820 } else {
1821 insn = qemu_ldx_opc[opc & (MO_SIZE | MO_BSWAP)];
1822 tcg_out32(s, insn | TAB(datalo, rbase, addrlo));
1823 insn = qemu_exts_opc[s_bits];
1824 tcg_out32(s, insn | RA(datalo) | RS(datalo));
1825 }
1826 }
1827
1828 #ifdef CONFIG_SOFTMMU
1829 add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1830 s->code_ptr, label_ptr);
1831 #endif
1832 }
1833
1834 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
1835 {
1836 TCGReg datalo, datahi, addrlo, rbase;
1837 TCGReg addrhi __attribute__((unused));
1838 TCGMemOpIdx oi;
1839 TCGMemOp opc, s_bits;
1840 #ifdef CONFIG_SOFTMMU
1841 int mem_index;
1842 tcg_insn_unit *label_ptr;
1843 #endif
1844
1845 datalo = *args++;
1846 datahi = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0);
1847 addrlo = *args++;
1848 addrhi = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0);
1849 oi = *args++;
1850 opc = get_memop(oi);
1851 s_bits = opc & MO_SIZE;
1852
1853 #ifdef CONFIG_SOFTMMU
1854 mem_index = get_mmuidx(oi);
1855 addrlo = tcg_out_tlb_read(s, opc, addrlo, addrhi, mem_index, false);
1856
1857 /* Load a pointer into the current opcode w/conditional branch-link. */
1858 label_ptr = s->code_ptr;
1859 tcg_out32(s, BC | BI(7, CR_EQ) | BO_COND_FALSE | LK);
1860
1861 rbase = TCG_REG_R3;
1862 #else /* !CONFIG_SOFTMMU */
1863 rbase = guest_base ? TCG_GUEST_BASE_REG : 0;
1864 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {
1865 tcg_out_ext32u(s, TCG_REG_TMP1, addrlo);
1866 addrlo = TCG_REG_TMP1;
1867 }
1868 #endif
1869
1870 if (TCG_TARGET_REG_BITS == 32 && s_bits == MO_64) {
1871 if (opc & MO_BSWAP) {
1872 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1873 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1874 tcg_out32(s, STWBRX | SAB(datahi, rbase, TCG_REG_R0));
1875 } else if (rbase != 0) {
1876 tcg_out32(s, ADDI | TAI(TCG_REG_R0, addrlo, 4));
1877 tcg_out32(s, STWX | SAB(datahi, rbase, addrlo));
1878 tcg_out32(s, STWX | SAB(datalo, rbase, TCG_REG_R0));
1879 } else {
1880 tcg_out32(s, STW | TAI(datahi, addrlo, 0));
1881 tcg_out32(s, STW | TAI(datalo, addrlo, 4));
1882 }
1883 } else {
1884 uint32_t insn = qemu_stx_opc[opc & (MO_BSWAP | MO_SIZE)];
1885 if (!HAVE_ISA_2_06 && insn == STDBRX) {
1886 tcg_out32(s, STWBRX | SAB(datalo, rbase, addrlo));
1887 tcg_out32(s, ADDI | TAI(TCG_REG_TMP1, addrlo, 4));
1888 tcg_out_shri64(s, TCG_REG_R0, datalo, 32);
1889 tcg_out32(s, STWBRX | SAB(TCG_REG_R0, rbase, TCG_REG_TMP1));
1890 } else {
1891 tcg_out32(s, insn | SAB(datalo, rbase, addrlo));
1892 }
1893 }
1894
1895 #ifdef CONFIG_SOFTMMU
1896 add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1897 s->code_ptr, label_ptr);
1898 #endif
1899 }
1900
1901 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1902 {
1903 int i;
1904 for (i = 0; i < count; ++i) {
1905 p[i] = NOP;
1906 }
1907 }
1908
1909 /* Parameters for function call generation, used in tcg.c. */
1910 #define TCG_TARGET_STACK_ALIGN 16
1911 #define TCG_TARGET_EXTEND_ARGS 1
1912
1913 #ifdef _CALL_AIX
1914 # define LINK_AREA_SIZE (6 * SZR)
1915 # define LR_OFFSET (1 * SZR)
1916 # define TCG_TARGET_CALL_STACK_OFFSET (LINK_AREA_SIZE + 8 * SZR)
1917 #elif defined(TCG_TARGET_CALL_DARWIN)
1918 # define LINK_AREA_SIZE (6 * SZR)
1919 # define LR_OFFSET (2 * SZR)
1920 #elif TCG_TARGET_REG_BITS == 64
1921 # if defined(_CALL_ELF) && _CALL_ELF == 2
1922 # define LINK_AREA_SIZE (4 * SZR)
1923 # define LR_OFFSET (1 * SZR)
1924 # endif
1925 #else /* TCG_TARGET_REG_BITS == 32 */
1926 # if defined(_CALL_SYSV)
1927 # define LINK_AREA_SIZE (2 * SZR)
1928 # define LR_OFFSET (1 * SZR)
1929 # endif
1930 #endif
1931 #ifndef LR_OFFSET
1932 # error "Unhandled abi"
1933 #endif
1934 #ifndef TCG_TARGET_CALL_STACK_OFFSET
1935 # define TCG_TARGET_CALL_STACK_OFFSET LINK_AREA_SIZE
1936 #endif
1937
1938 #define CPU_TEMP_BUF_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
1939 #define REG_SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * SZR)
1940
1941 #define FRAME_SIZE ((TCG_TARGET_CALL_STACK_OFFSET \
1942 + TCG_STATIC_CALL_ARGS_SIZE \
1943 + CPU_TEMP_BUF_SIZE \
1944 + REG_SAVE_SIZE \
1945 + TCG_TARGET_STACK_ALIGN - 1) \
1946 & -TCG_TARGET_STACK_ALIGN)
1947
1948 #define REG_SAVE_BOT (FRAME_SIZE - REG_SAVE_SIZE)
1949
1950 static void tcg_target_qemu_prologue(TCGContext *s)
1951 {
1952 int i;
1953
1954 #ifdef _CALL_AIX
1955 void **desc = (void **)s->code_ptr;
1956 desc[0] = desc + 2; /* entry point */
1957 desc[1] = 0; /* environment pointer */
1958 s->code_ptr = (void *)(desc + 2); /* skip over descriptor */
1959 #endif
1960
1961 tcg_set_frame(s, TCG_REG_CALL_STACK, REG_SAVE_BOT - CPU_TEMP_BUF_SIZE,
1962 CPU_TEMP_BUF_SIZE);
1963
1964 /* Prologue */
1965 tcg_out32(s, MFSPR | RT(TCG_REG_R0) | LR);
1966 tcg_out32(s, (SZR == 8 ? STDU : STWU)
1967 | SAI(TCG_REG_R1, TCG_REG_R1, -FRAME_SIZE));
1968
1969 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1970 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1971 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1972 }
1973 tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1974
1975 #ifndef CONFIG_SOFTMMU
1976 if (guest_base) {
1977 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
1978 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1979 }
1980 #endif
1981
1982 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1983 tcg_out32(s, MTSPR | RS(tcg_target_call_iarg_regs[1]) | CTR);
1984 if (USE_REG_TB) {
1985 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, tcg_target_call_iarg_regs[1]);
1986 }
1987 tcg_out32(s, BCCTR | BO_ALWAYS);
1988
1989 /* Epilogue */
1990 s->code_gen_epilogue = tb_ret_addr = s->code_ptr;
1991
1992 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R0, TCG_REG_R1, FRAME_SIZE+LR_OFFSET);
1993 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i) {
1994 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
1995 TCG_REG_R1, REG_SAVE_BOT + i * SZR);
1996 }
1997 tcg_out32(s, MTSPR | RS(TCG_REG_R0) | LR);
1998 tcg_out32(s, ADDI | TAI(TCG_REG_R1, TCG_REG_R1, FRAME_SIZE));
1999 tcg_out32(s, BCLR | BO_ALWAYS);
2000 }
2001
2002 static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
2003 const int *const_args)
2004 {
2005 TCGArg a0, a1, a2;
2006 int c;
2007
2008 switch (opc) {
2009 case INDEX_op_exit_tb:
2010 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R3, args[0]);
2011 tcg_out_b(s, 0, tb_ret_addr);
2012 break;
2013 case INDEX_op_goto_tb:
2014 if (s->tb_jmp_insn_offset) {
2015 /* Direct jump. */
2016 if (TCG_TARGET_REG_BITS == 64) {
2017 /* Ensure the next insns are 8-byte aligned. */
2018 if ((uintptr_t)s->code_ptr & 7) {
2019 tcg_out32(s, NOP);
2020 }
2021 s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2022 tcg_out32(s, ADDIS | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2023 tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, 0));
2024 } else {
2025 s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
2026 tcg_out32(s, B);
2027 s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
2028 break;
2029 }
2030 } else {
2031 /* Indirect jump. */
2032 tcg_debug_assert(s->tb_jmp_insn_offset == NULL);
2033 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, 0,
2034 (intptr_t)(s->tb_jmp_insn_offset + args[0]));
2035 }
2036 tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
2037 tcg_out32(s, BCCTR | BO_ALWAYS);
2038 set_jmp_reset_offset(s, args[0]);
2039 if (USE_REG_TB) {
2040 /* For the unlinked case, need to reset TCG_REG_TB. */
2041 c = -tcg_current_code_size(s);
2042 assert(c == (int16_t)c);
2043 tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, c));
2044 }
2045 break;
2046 case INDEX_op_goto_ptr:
2047 tcg_out32(s, MTSPR | RS(args[0]) | CTR);
2048 if (USE_REG_TB) {
2049 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_TB, args[0]);
2050 }
2051 tcg_out32(s, ADDI | TAI(TCG_REG_R3, 0, 0));
2052 tcg_out32(s, BCCTR | BO_ALWAYS);
2053 break;
2054 case INDEX_op_br:
2055 {
2056 TCGLabel *l = arg_label(args[0]);
2057 uint32_t insn = B;
2058
2059 if (l->has_value) {
2060 insn |= reloc_pc24_val(s->code_ptr, l->u.value_ptr);
2061 } else {
2062 tcg_out_reloc(s, s->code_ptr, R_PPC_REL24, l, 0);
2063 }
2064 tcg_out32(s, insn);
2065 }
2066 break;
2067 case INDEX_op_ld8u_i32:
2068 case INDEX_op_ld8u_i64:
2069 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2070 break;
2071 case INDEX_op_ld8s_i32:
2072 case INDEX_op_ld8s_i64:
2073 tcg_out_mem_long(s, LBZ, LBZX, args[0], args[1], args[2]);
2074 tcg_out32(s, EXTSB | RS(args[0]) | RA(args[0]));
2075 break;
2076 case INDEX_op_ld16u_i32:
2077 case INDEX_op_ld16u_i64:
2078 tcg_out_mem_long(s, LHZ, LHZX, args[0], args[1], args[2]);
2079 break;
2080 case INDEX_op_ld16s_i32:
2081 case INDEX_op_ld16s_i64:
2082 tcg_out_mem_long(s, LHA, LHAX, args[0], args[1], args[2]);
2083 break;
2084 case INDEX_op_ld_i32:
2085 case INDEX_op_ld32u_i64:
2086 tcg_out_mem_long(s, LWZ, LWZX, args[0], args[1], args[2]);
2087 break;
2088 case INDEX_op_ld32s_i64:
2089 tcg_out_mem_long(s, LWA, LWAX, args[0], args[1], args[2]);
2090 break;
2091 case INDEX_op_ld_i64:
2092 tcg_out_mem_long(s, LD, LDX, args[0], args[1], args[2]);
2093 break;
2094 case INDEX_op_st8_i32:
2095 case INDEX_op_st8_i64:
2096 tcg_out_mem_long(s, STB, STBX, args[0], args[1], args[2]);
2097 break;
2098 case INDEX_op_st16_i32:
2099 case INDEX_op_st16_i64:
2100 tcg_out_mem_long(s, STH, STHX, args[0], args[1], args[2]);
2101 break;
2102 case INDEX_op_st_i32:
2103 case INDEX_op_st32_i64:
2104 tcg_out_mem_long(s, STW, STWX, args[0], args[1], args[2]);
2105 break;
2106 case INDEX_op_st_i64:
2107 tcg_out_mem_long(s, STD, STDX, args[0], args[1], args[2]);
2108 break;
2109
2110 case INDEX_op_add_i32:
2111 a0 = args[0], a1 = args[1], a2 = args[2];
2112 if (const_args[2]) {
2113 do_addi_32:
2114 tcg_out_mem_long(s, ADDI, ADD, a0, a1, (int32_t)a2);
2115 } else {
2116 tcg_out32(s, ADD | TAB(a0, a1, a2));
2117 }
2118 break;
2119 case INDEX_op_sub_i32:
2120 a0 = args[0], a1 = args[1], a2 = args[2];
2121 if (const_args[1]) {
2122 if (const_args[2]) {
2123 tcg_out_movi(s, TCG_TYPE_I32, a0, a1 - a2);
2124 } else {
2125 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2126 }
2127 } else if (const_args[2]) {
2128 a2 = -a2;
2129 goto do_addi_32;
2130 } else {
2131 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2132 }
2133 break;
2134
2135 case INDEX_op_and_i32:
2136 a0 = args[0], a1 = args[1], a2 = args[2];
2137 if (const_args[2]) {
2138 tcg_out_andi32(s, a0, a1, a2);
2139 } else {
2140 tcg_out32(s, AND | SAB(a1, a0, a2));
2141 }
2142 break;
2143 case INDEX_op_and_i64:
2144 a0 = args[0], a1 = args[1], a2 = args[2];
2145 if (const_args[2]) {
2146 tcg_out_andi64(s, a0, a1, a2);
2147 } else {
2148 tcg_out32(s, AND | SAB(a1, a0, a2));
2149 }
2150 break;
2151 case INDEX_op_or_i64:
2152 case INDEX_op_or_i32:
2153 a0 = args[0], a1 = args[1], a2 = args[2];
2154 if (const_args[2]) {
2155 tcg_out_ori32(s, a0, a1, a2);
2156 } else {
2157 tcg_out32(s, OR | SAB(a1, a0, a2));
2158 }
2159 break;
2160 case INDEX_op_xor_i64:
2161 case INDEX_op_xor_i32:
2162 a0 = args[0], a1 = args[1], a2 = args[2];
2163 if (const_args[2]) {
2164 tcg_out_xori32(s, a0, a1, a2);
2165 } else {
2166 tcg_out32(s, XOR | SAB(a1, a0, a2));
2167 }
2168 break;
2169 case INDEX_op_andc_i32:
2170 a0 = args[0], a1 = args[1], a2 = args[2];
2171 if (const_args[2]) {
2172 tcg_out_andi32(s, a0, a1, ~a2);
2173 } else {
2174 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2175 }
2176 break;
2177 case INDEX_op_andc_i64:
2178 a0 = args[0], a1 = args[1], a2 = args[2];
2179 if (const_args[2]) {
2180 tcg_out_andi64(s, a0, a1, ~a2);
2181 } else {
2182 tcg_out32(s, ANDC | SAB(a1, a0, a2));
2183 }
2184 break;
2185 case INDEX_op_orc_i32:
2186 if (const_args[2]) {
2187 tcg_out_ori32(s, args[0], args[1], ~args[2]);
2188 break;
2189 }
2190 /* FALLTHRU */
2191 case INDEX_op_orc_i64:
2192 tcg_out32(s, ORC | SAB(args[1], args[0], args[2]));
2193 break;
2194 case INDEX_op_eqv_i32:
2195 if (const_args[2]) {
2196 tcg_out_xori32(s, args[0], args[1], ~args[2]);
2197 break;
2198 }
2199 /* FALLTHRU */
2200 case INDEX_op_eqv_i64:
2201 tcg_out32(s, EQV | SAB(args[1], args[0], args[2]));
2202 break;
2203 case INDEX_op_nand_i32:
2204 case INDEX_op_nand_i64:
2205 tcg_out32(s, NAND | SAB(args[1], args[0], args[2]));
2206 break;
2207 case INDEX_op_nor_i32:
2208 case INDEX_op_nor_i64:
2209 tcg_out32(s, NOR | SAB(args[1], args[0], args[2]));
2210 break;
2211
2212 case INDEX_op_clz_i32:
2213 tcg_out_cntxz(s, TCG_TYPE_I32, CNTLZW, args[0], args[1],
2214 args[2], const_args[2]);
2215 break;
2216 case INDEX_op_ctz_i32:
2217 tcg_out_cntxz(s, TCG_TYPE_I32, CNTTZW, args[0], args[1],
2218 args[2], const_args[2]);
2219 break;
2220 case INDEX_op_ctpop_i32:
2221 tcg_out32(s, CNTPOPW | SAB(args[1], args[0], 0));
2222 break;
2223
2224 case INDEX_op_clz_i64:
2225 tcg_out_cntxz(s, TCG_TYPE_I64, CNTLZD, args[0], args[1],
2226 args[2], const_args[2]);
2227 break;
2228 case INDEX_op_ctz_i64:
2229 tcg_out_cntxz(s, TCG_TYPE_I64, CNTTZD, args[0], args[1],
2230 args[2], const_args[2]);
2231 break;
2232 case INDEX_op_ctpop_i64:
2233 tcg_out32(s, CNTPOPD | SAB(args[1], args[0], 0));
2234 break;
2235
2236 case INDEX_op_mul_i32:
2237 a0 = args[0], a1 = args[1], a2 = args[2];
2238 if (const_args[2]) {
2239 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2240 } else {
2241 tcg_out32(s, MULLW | TAB(a0, a1, a2));
2242 }
2243 break;
2244
2245 case INDEX_op_div_i32:
2246 tcg_out32(s, DIVW | TAB(args[0], args[1], args[2]));
2247 break;
2248
2249 case INDEX_op_divu_i32:
2250 tcg_out32(s, DIVWU | TAB(args[0], args[1], args[2]));
2251 break;
2252
2253 case INDEX_op_shl_i32:
2254 if (const_args[2]) {
2255 tcg_out_shli32(s, args[0], args[1], args[2]);
2256 } else {
2257 tcg_out32(s, SLW | SAB(args[1], args[0], args[2]));
2258 }
2259 break;
2260 case INDEX_op_shr_i32:
2261 if (const_args[2]) {
2262 tcg_out_shri32(s, args[0], args[1], args[2]);
2263 } else {
2264 tcg_out32(s, SRW | SAB(args[1], args[0], args[2]));
2265 }
2266 break;
2267 case INDEX_op_sar_i32:
2268 if (const_args[2]) {
2269 tcg_out32(s, SRAWI | RS(args[1]) | RA(args[0]) | SH(args[2]));
2270 } else {
2271 tcg_out32(s, SRAW | SAB(args[1], args[0], args[2]));
2272 }
2273 break;
2274 case INDEX_op_rotl_i32:
2275 if (const_args[2]) {
2276 tcg_out_rlw(s, RLWINM, args[0], args[1], args[2], 0, 31);
2277 } else {
2278 tcg_out32(s, RLWNM | SAB(args[1], args[0], args[2])
2279 | MB(0) | ME(31));
2280 }
2281 break;
2282 case INDEX_op_rotr_i32:
2283 if (const_args[2]) {
2284 tcg_out_rlw(s, RLWINM, args[0], args[1], 32 - args[2], 0, 31);
2285 } else {
2286 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 32));
2287 tcg_out32(s, RLWNM | SAB(args[1], args[0], TCG_REG_R0)
2288 | MB(0) | ME(31));
2289 }
2290 break;
2291
2292 case INDEX_op_brcond_i32:
2293 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2294 arg_label(args[3]), TCG_TYPE_I32);
2295 break;
2296 case INDEX_op_brcond_i64:
2297 tcg_out_brcond(s, args[2], args[0], args[1], const_args[1],
2298 arg_label(args[3]), TCG_TYPE_I64);
2299 break;
2300 case INDEX_op_brcond2_i32:
2301 tcg_out_brcond2(s, args, const_args);
2302 break;
2303
2304 case INDEX_op_neg_i32:
2305 case INDEX_op_neg_i64:
2306 tcg_out32(s, NEG | RT(args[0]) | RA(args[1]));
2307 break;
2308
2309 case INDEX_op_not_i32:
2310 case INDEX_op_not_i64:
2311 tcg_out32(s, NOR | SAB(args[1], args[0], args[1]));
2312 break;
2313
2314 case INDEX_op_add_i64:
2315 a0 = args[0], a1 = args[1], a2 = args[2];
2316 if (const_args[2]) {
2317 do_addi_64:
2318 tcg_out_mem_long(s, ADDI, ADD, a0, a1, a2);
2319 } else {
2320 tcg_out32(s, ADD | TAB(a0, a1, a2));
2321 }
2322 break;
2323 case INDEX_op_sub_i64:
2324 a0 = args[0], a1 = args[1], a2 = args[2];
2325 if (const_args[1]) {
2326 if (const_args[2]) {
2327 tcg_out_movi(s, TCG_TYPE_I64, a0, a1 - a2);
2328 } else {
2329 tcg_out32(s, SUBFIC | TAI(a0, a2, a1));
2330 }
2331 } else if (const_args[2]) {
2332 a2 = -a2;
2333 goto do_addi_64;
2334 } else {
2335 tcg_out32(s, SUBF | TAB(a0, a2, a1));
2336 }
2337 break;
2338
2339 case INDEX_op_shl_i64:
2340 if (const_args[2]) {
2341 tcg_out_shli64(s, args[0], args[1], args[2]);
2342 } else {
2343 tcg_out32(s, SLD | SAB(args[1], args[0], args[2]));
2344 }
2345 break;
2346 case INDEX_op_shr_i64:
2347 if (const_args[2]) {
2348 tcg_out_shri64(s, args[0], args[1], args[2]);
2349 } else {
2350 tcg_out32(s, SRD | SAB(args[1], args[0], args[2]));
2351 }
2352 break;
2353 case INDEX_op_sar_i64:
2354 if (const_args[2]) {
2355 int sh = SH(args[2] & 0x1f) | (((args[2] >> 5) & 1) << 1);
2356 tcg_out32(s, SRADI | RA(args[0]) | RS(args[1]) | sh);
2357 } else {
2358 tcg_out32(s, SRAD | SAB(args[1], args[0], args[2]));
2359 }
2360 break;
2361 case INDEX_op_rotl_i64:
2362 if (const_args[2]) {
2363 tcg_out_rld(s, RLDICL, args[0], args[1], args[2], 0);
2364 } else {
2365 tcg_out32(s, RLDCL | SAB(args[1], args[0], args[2]) | MB64(0));
2366 }
2367 break;
2368 case INDEX_op_rotr_i64:
2369 if (const_args[2]) {
2370 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 0);
2371 } else {
2372 tcg_out32(s, SUBFIC | TAI(TCG_REG_R0, args[2], 64));
2373 tcg_out32(s, RLDCL | SAB(args[1], args[0], TCG_REG_R0) | MB64(0));
2374 }
2375 break;
2376
2377 case INDEX_op_mul_i64:
2378 a0 = args[0], a1 = args[1], a2 = args[2];
2379 if (const_args[2]) {
2380 tcg_out32(s, MULLI | TAI(a0, a1, a2));
2381 } else {
2382 tcg_out32(s, MULLD | TAB(a0, a1, a2));
2383 }
2384 break;
2385 case INDEX_op_div_i64:
2386 tcg_out32(s, DIVD | TAB(args[0], args[1], args[2]));
2387 break;
2388 case INDEX_op_divu_i64:
2389 tcg_out32(s, DIVDU | TAB(args[0], args[1], args[2]));
2390 break;
2391
2392 case INDEX_op_qemu_ld_i32:
2393 tcg_out_qemu_ld(s, args, false);
2394 break;
2395 case INDEX_op_qemu_ld_i64:
2396 tcg_out_qemu_ld(s, args, true);
2397 break;
2398 case INDEX_op_qemu_st_i32:
2399 tcg_out_qemu_st(s, args, false);
2400 break;
2401 case INDEX_op_qemu_st_i64:
2402 tcg_out_qemu_st(s, args, true);
2403 break;
2404
2405 case INDEX_op_ext8s_i32:
2406 case INDEX_op_ext8s_i64:
2407 c = EXTSB;
2408 goto gen_ext;
2409 case INDEX_op_ext16s_i32:
2410 case INDEX_op_ext16s_i64:
2411 c = EXTSH;
2412 goto gen_ext;
2413 case INDEX_op_ext_i32_i64:
2414 case INDEX_op_ext32s_i64:
2415 c = EXTSW;
2416 goto gen_ext;
2417 gen_ext:
2418 tcg_out32(s, c | RS(args[1]) | RA(args[0]));
2419 break;
2420 case INDEX_op_extu_i32_i64:
2421 tcg_out_ext32u(s, args[0], args[1]);
2422 break;
2423
2424 case INDEX_op_setcond_i32:
2425 tcg_out_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], args[2],
2426 const_args[2]);
2427 break;
2428 case INDEX_op_setcond_i64:
2429 tcg_out_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], args[2],
2430 const_args[2]);
2431 break;
2432 case INDEX_op_setcond2_i32:
2433 tcg_out_setcond2(s, args, const_args);
2434 break;
2435
2436 case INDEX_op_bswap16_i32:
2437 case INDEX_op_bswap16_i64:
2438 a0 = args[0], a1 = args[1];
2439 /* a1 = abcd */
2440 if (a0 != a1) {
2441 /* a0 = (a1 r<< 24) & 0xff # 000c */
2442 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2443 /* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */
2444 tcg_out_rlw(s, RLWIMI, a0, a1, 8, 16, 23);
2445 } else {
2446 /* r0 = (a1 r<< 8) & 0xff00 # 00d0 */
2447 tcg_out_rlw(s, RLWINM, TCG_REG_R0, a1, 8, 16, 23);
2448 /* a0 = (a1 r<< 24) & 0xff # 000c */
2449 tcg_out_rlw(s, RLWINM, a0, a1, 24, 24, 31);
2450 /* a0 = a0 | r0 # 00dc */
2451 tcg_out32(s, OR | SAB(TCG_REG_R0, a0, a0));
2452 }
2453 break;
2454
2455 case INDEX_op_bswap32_i32:
2456 case INDEX_op_bswap32_i64:
2457 /* Stolen from gcc's builtin_bswap32 */
2458 a1 = args[1];
2459 a0 = args[0] == a1 ? TCG_REG_R0 : args[0];
2460
2461 /* a1 = args[1] # abcd */
2462 /* a0 = rotate_left (a1, 8) # bcda */
2463 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2464 /* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */
2465 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2466 /* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */
2467 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2468
2469 if (a0 == TCG_REG_R0) {
2470 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2471 }
2472 break;
2473
2474 case INDEX_op_bswap64_i64:
2475 a0 = args[0], a1 = args[1], a2 = TCG_REG_R0;
2476 if (a0 == a1) {
2477 a0 = TCG_REG_R0;
2478 a2 = a1;
2479 }
2480
2481 /* a1 = # abcd efgh */
2482 /* a0 = rl32(a1, 8) # 0000 fghe */
2483 tcg_out_rlw(s, RLWINM, a0, a1, 8, 0, 31);
2484 /* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */
2485 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 0, 7);
2486 /* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */
2487 tcg_out_rlw(s, RLWIMI, a0, a1, 24, 16, 23);
2488
2489 /* a0 = rl64(a0, 32) # hgfe 0000 */
2490 /* a2 = rl64(a1, 32) # efgh abcd */
2491 tcg_out_rld(s, RLDICL, a0, a0, 32, 0);
2492 tcg_out_rld(s, RLDICL, a2, a1, 32, 0);
2493
2494 /* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */
2495 tcg_out_rlw(s, RLWIMI, a0, a2, 8, 0, 31);
2496 /* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */
2497 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 0, 7);
2498 /* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */
2499 tcg_out_rlw(s, RLWIMI, a0, a2, 24, 16, 23);
2500
2501 if (a0 == 0) {
2502 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2503 }
2504 break;
2505
2506 case INDEX_op_deposit_i32:
2507 if (const_args[2]) {
2508 uint32_t mask = ((2u << (args[4] - 1)) - 1) << args[3];
2509 tcg_out_andi32(s, args[0], args[0], ~mask);
2510 } else {
2511 tcg_out_rlw(s, RLWIMI, args[0], args[2], args[3],
2512 32 - args[3] - args[4], 31 - args[3]);
2513 }
2514 break;
2515 case INDEX_op_deposit_i64:
2516 if (const_args[2]) {
2517 uint64_t mask = ((2ull << (args[4] - 1)) - 1) << args[3];
2518 tcg_out_andi64(s, args[0], args[0], ~mask);
2519 } else {
2520 tcg_out_rld(s, RLDIMI, args[0], args[2], args[3],
2521 64 - args[3] - args[4]);
2522 }
2523 break;
2524
2525 case INDEX_op_extract_i32:
2526 tcg_out_rlw(s, RLWINM, args[0], args[1],
2527 32 - args[2], 32 - args[3], 31);
2528 break;
2529 case INDEX_op_extract_i64:
2530 tcg_out_rld(s, RLDICL, args[0], args[1], 64 - args[2], 64 - args[3]);
2531 break;
2532
2533 case INDEX_op_movcond_i32:
2534 tcg_out_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], args[2],
2535 args[3], args[4], const_args[2]);
2536 break;
2537 case INDEX_op_movcond_i64:
2538 tcg_out_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], args[2],
2539 args[3], args[4], const_args[2]);
2540 break;
2541
2542 #if TCG_TARGET_REG_BITS == 64
2543 case INDEX_op_add2_i64:
2544 #else
2545 case INDEX_op_add2_i32:
2546 #endif
2547 /* Note that the CA bit is defined based on the word size of the
2548 environment. So in 64-bit mode it's always carry-out of bit 63.
2549 The fallback code using deposit works just as well for 32-bit. */
2550 a0 = args[0], a1 = args[1];
2551 if (a0 == args[3] || (!const_args[5] && a0 == args[5])) {
2552 a0 = TCG_REG_R0;
2553 }
2554 if (const_args[4]) {
2555 tcg_out32(s, ADDIC | TAI(a0, args[2], args[4]));
2556 } else {
2557 tcg_out32(s, ADDC | TAB(a0, args[2], args[4]));
2558 }
2559 if (const_args[5]) {
2560 tcg_out32(s, (args[5] ? ADDME : ADDZE) | RT(a1) | RA(args[3]));
2561 } else {
2562 tcg_out32(s, ADDE | TAB(a1, args[3], args[5]));
2563 }
2564 if (a0 != args[0]) {
2565 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2566 }
2567 break;
2568
2569 #if TCG_TARGET_REG_BITS == 64
2570 case INDEX_op_sub2_i64:
2571 #else
2572 case INDEX_op_sub2_i32:
2573 #endif
2574 a0 = args[0], a1 = args[1];
2575 if (a0 == args[5] || (!const_args[3] && a0 == args[3])) {
2576 a0 = TCG_REG_R0;
2577 }
2578 if (const_args[2]) {
2579 tcg_out32(s, SUBFIC | TAI(a0, args[4], args[2]));
2580 } else {
2581 tcg_out32(s, SUBFC | TAB(a0, args[4], args[2]));
2582 }
2583 if (const_args[3]) {
2584 tcg_out32(s, (args[3] ? SUBFME : SUBFZE) | RT(a1) | RA(args[5]));
2585 } else {
2586 tcg_out32(s, SUBFE | TAB(a1, args[5], args[3]));
2587 }
2588 if (a0 != args[0]) {
2589 tcg_out_mov(s, TCG_TYPE_REG, args[0], a0);
2590 }
2591 break;
2592
2593 case INDEX_op_muluh_i32:
2594 tcg_out32(s, MULHWU | TAB(args[0], args[1], args[2]));
2595 break;
2596 case INDEX_op_mulsh_i32:
2597 tcg_out32(s, MULHW | TAB(args[0], args[1], args[2]));
2598 break;
2599 case INDEX_op_muluh_i64:
2600 tcg_out32(s, MULHDU | TAB(args[0], args[1], args[2]));
2601 break;
2602 case INDEX_op_mulsh_i64:
2603 tcg_out32(s, MULHD | TAB(args[0], args[1], args[2]));
2604 break;
2605
2606 case INDEX_op_mb:
2607 tcg_out_mb(s, args[0]);
2608 break;
2609
2610 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2611 case INDEX_op_mov_i64:
2612 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2613 case INDEX_op_movi_i64:
2614 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2615 default:
2616 tcg_abort();
2617 }
2618 }
2619
2620 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
2621 {
2622 static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
2623 static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
2624 static const TCGTargetOpDef r_L = { .args_ct_str = { "r", "L" } };
2625 static const TCGTargetOpDef S_S = { .args_ct_str = { "S", "S" } };
2626 static const TCGTargetOpDef r_ri = { .args_ct_str = { "r", "ri" } };
2627 static const TCGTargetOpDef r_r_r = { .args_ct_str = { "r", "r", "r" } };
2628 static const TCGTargetOpDef r_L_L = { .args_ct_str = { "r", "L", "L" } };
2629 static const TCGTargetOpDef L_L_L = { .args_ct_str = { "L", "L", "L" } };
2630 static const TCGTargetOpDef S_S_S = { .args_ct_str = { "S", "S", "S" } };
2631 static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
2632 static const TCGTargetOpDef r_r_rI = { .args_ct_str = { "r", "r", "rI" } };
2633 static const TCGTargetOpDef r_r_rT = { .args_ct_str = { "r", "r", "rT" } };
2634 static const TCGTargetOpDef r_r_rU = { .args_ct_str = { "r", "r", "rU" } };
2635 static const TCGTargetOpDef r_rI_ri
2636 = { .args_ct_str = { "r", "rI", "ri" } };
2637 static const TCGTargetOpDef r_rI_rT
2638 = { .args_ct_str = { "r", "rI", "rT" } };
2639 static const TCGTargetOpDef r_r_rZW
2640 = { .args_ct_str = { "r", "r", "rZW" } };
2641 static const TCGTargetOpDef L_L_L_L
2642 = { .args_ct_str = { "L", "L", "L", "L" } };
2643 static const TCGTargetOpDef S_S_S_S
2644 = { .args_ct_str = { "S", "S", "S", "S" } };
2645 static const TCGTargetOpDef movc
2646 = { .args_ct_str = { "r", "r", "ri", "rZ", "rZ" } };
2647 static const TCGTargetOpDef dep
2648 = { .args_ct_str = { "r", "0", "rZ" } };
2649 static const TCGTargetOpDef br2
2650 = { .args_ct_str = { "r", "r", "ri", "ri" } };
2651 static const TCGTargetOpDef setc2
2652 = { .args_ct_str = { "r", "r", "r", "ri", "ri" } };
2653 static const TCGTargetOpDef add2
2654 = { .args_ct_str = { "r", "r", "r", "r", "rI", "rZM" } };
2655 static const TCGTargetOpDef sub2
2656 = { .args_ct_str = { "r", "r", "rI", "rZM", "r", "r" } };
2657
2658 switch (op) {
2659 case INDEX_op_goto_ptr:
2660 return &r;
2661
2662 case INDEX_op_ld8u_i32:
2663 case INDEX_op_ld8s_i32:
2664 case INDEX_op_ld16u_i32:
2665 case INDEX_op_ld16s_i32:
2666 case INDEX_op_ld_i32:
2667 case INDEX_op_st8_i32:
2668 case INDEX_op_st16_i32:
2669 case INDEX_op_st_i32:
2670 case INDEX_op_ctpop_i32:
2671 case INDEX_op_neg_i32:
2672 case INDEX_op_not_i32:
2673 case INDEX_op_ext8s_i32:
2674 case INDEX_op_ext16s_i32:
2675 case INDEX_op_bswap16_i32:
2676 case INDEX_op_bswap32_i32:
2677 case INDEX_op_extract_i32:
2678 case INDEX_op_ld8u_i64:
2679 case INDEX_op_ld8s_i64:
2680 case INDEX_op_ld16u_i64:
2681 case INDEX_op_ld16s_i64:
2682 case INDEX_op_ld32u_i64:
2683 case INDEX_op_ld32s_i64:
2684 case INDEX_op_ld_i64:
2685 case INDEX_op_st8_i64:
2686 case INDEX_op_st16_i64:
2687 case INDEX_op_st32_i64:
2688 case INDEX_op_st_i64:
2689 case INDEX_op_ctpop_i64:
2690 case INDEX_op_neg_i64:
2691 case INDEX_op_not_i64:
2692 case INDEX_op_ext8s_i64:
2693 case INDEX_op_ext16s_i64:
2694 case INDEX_op_ext32s_i64:
2695 case INDEX_op_ext_i32_i64:
2696 case INDEX_op_extu_i32_i64:
2697 case INDEX_op_bswap16_i64:
2698 case INDEX_op_bswap32_i64:
2699 case INDEX_op_bswap64_i64:
2700 case INDEX_op_extract_i64:
2701 return &r_r;
2702
2703 case INDEX_op_add_i32:
2704 case INDEX_op_and_i32:
2705 case INDEX_op_or_i32:
2706 case INDEX_op_xor_i32:
2707 case INDEX_op_andc_i32:
2708 case INDEX_op_orc_i32:
2709 case INDEX_op_eqv_i32:
2710 case INDEX_op_shl_i32:
2711 case INDEX_op_shr_i32:
2712 case INDEX_op_sar_i32:
2713 case INDEX_op_rotl_i32:
2714 case INDEX_op_rotr_i32:
2715 case INDEX_op_setcond_i32:
2716 case INDEX_op_and_i64:
2717 case INDEX_op_andc_i64:
2718 case INDEX_op_shl_i64:
2719 case INDEX_op_shr_i64:
2720 case INDEX_op_sar_i64:
2721 case INDEX_op_rotl_i64:
2722 case INDEX_op_rotr_i64:
2723 case INDEX_op_setcond_i64:
2724 return &r_r_ri;
2725 case INDEX_op_mul_i32:
2726 case INDEX_op_mul_i64:
2727 return &r_r_rI;
2728 case INDEX_op_div_i32:
2729 case INDEX_op_divu_i32:
2730 case INDEX_op_nand_i32:
2731 case INDEX_op_nor_i32:
2732 case INDEX_op_muluh_i32:
2733 case INDEX_op_mulsh_i32:
2734 case INDEX_op_orc_i64:
2735 case INDEX_op_eqv_i64:
2736 case INDEX_op_nand_i64:
2737 case INDEX_op_nor_i64:
2738 case INDEX_op_div_i64:
2739 case INDEX_op_divu_i64:
2740 case INDEX_op_mulsh_i64:
2741 case INDEX_op_muluh_i64:
2742 return &r_r_r;
2743 case INDEX_op_sub_i32:
2744 return &r_rI_ri;
2745 case INDEX_op_add_i64:
2746 return &r_r_rT;
2747 case INDEX_op_or_i64:
2748 case INDEX_op_xor_i64:
2749 return &r_r_rU;
2750 case INDEX_op_sub_i64:
2751 return &r_rI_rT;
2752 case INDEX_op_clz_i32:
2753 case INDEX_op_ctz_i32:
2754 case INDEX_op_clz_i64:
2755 case INDEX_op_ctz_i64:
2756 return &r_r_rZW;
2757
2758 case INDEX_op_brcond_i32:
2759 case INDEX_op_brcond_i64:
2760 return &r_ri;
2761
2762 case INDEX_op_movcond_i32:
2763 case INDEX_op_movcond_i64:
2764 return &movc;
2765 case INDEX_op_deposit_i32:
2766 case INDEX_op_deposit_i64:
2767 return &dep;
2768 case INDEX_op_brcond2_i32:
2769 return &br2;
2770 case INDEX_op_setcond2_i32:
2771 return &setc2;
2772 case INDEX_op_add2_i64:
2773 case INDEX_op_add2_i32:
2774 return &add2;
2775 case INDEX_op_sub2_i64:
2776 case INDEX_op_sub2_i32:
2777 return &sub2;
2778
2779 case INDEX_op_qemu_ld_i32:
2780 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
2781 ? &r_L : &r_L_L);
2782 case INDEX_op_qemu_st_i32:
2783 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32
2784 ? &S_S : &S_S_S);
2785 case INDEX_op_qemu_ld_i64:
2786 return (TCG_TARGET_REG_BITS == 64 ? &r_L
2787 : TARGET_LONG_BITS == 32 ? &L_L_L : &L_L_L_L);
2788 case INDEX_op_qemu_st_i64:
2789 return (TCG_TARGET_REG_BITS == 64 ? &S_S
2790 : TARGET_LONG_BITS == 32 ? &S_S_S : &S_S_S_S);
2791
2792 default:
2793 return NULL;
2794 }
2795 }
2796
2797 static void tcg_target_init(TCGContext *s)
2798 {
2799 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2800 unsigned long hwcap2 = qemu_getauxval(AT_HWCAP2);
2801
2802 if (hwcap & PPC_FEATURE_ARCH_2_06) {
2803 have_isa_2_06 = true;
2804 }
2805 #ifdef PPC_FEATURE2_ARCH_3_00
2806 if (hwcap2 & PPC_FEATURE2_ARCH_3_00) {
2807 have_isa_3_00 = true;
2808 }
2809 #endif
2810
2811 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
2812 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
2813
2814 tcg_target_call_clobber_regs = 0;
2815 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2816 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2817 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2818 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
2819 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
2820 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
2821 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R7);
2822 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R8);
2823 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R9);
2824 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R10);
2825 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R11);
2826 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2827
2828 s->reserved_regs = 0;
2829 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); /* tcg temp */
2830 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R1); /* stack pointer */
2831 #if defined(_CALL_SYSV)
2832 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R2); /* toc pointer */
2833 #endif
2834 #if defined(_CALL_SYSV) || TCG_TARGET_REG_BITS == 64
2835 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R13); /* thread pointer */
2836 #endif
2837 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1); /* mem temp */
2838 if (USE_REG_TB) {
2839 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB); /* tb->tc_ptr */
2840 }
2841 }
2842
2843 #ifdef __ELF__
2844 typedef struct {
2845 DebugFrameCIE cie;
2846 DebugFrameFDEHeader fde;
2847 uint8_t fde_def_cfa[4];
2848 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2 + 3];
2849 } DebugFrame;
2850
2851 /* We're expecting a 2 byte uleb128 encoded value. */
2852 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2853
2854 #if TCG_TARGET_REG_BITS == 64
2855 # define ELF_HOST_MACHINE EM_PPC64
2856 #else
2857 # define ELF_HOST_MACHINE EM_PPC
2858 #endif
2859
2860 static DebugFrame debug_frame = {
2861 .cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2862 .cie.id = -1,
2863 .cie.version = 1,
2864 .cie.code_align = 1,
2865 .cie.data_align = (-SZR & 0x7f), /* sleb128 -SZR */
2866 .cie.return_column = 65,
2867
2868 /* Total FDE size does not include the "len" member. */
2869 .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset),
2870
2871 .fde_def_cfa = {
2872 12, TCG_REG_R1, /* DW_CFA_def_cfa r1, ... */
2873 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2874 (FRAME_SIZE >> 7)
2875 },
2876 .fde_reg_ofs = {
2877 /* DW_CFA_offset_extended_sf, lr, LR_OFFSET */
2878 0x11, 65, (LR_OFFSET / -SZR) & 0x7f,
2879 }
2880 };
2881
2882 void tcg_register_jit(void *buf, size_t buf_size)
2883 {
2884 uint8_t *p = &debug_frame.fde_reg_ofs[3];
2885 int i;
2886
2887 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); ++i, p += 2) {
2888 p[0] = 0x80 + tcg_target_callee_save_regs[i];
2889 p[1] = (FRAME_SIZE - (REG_SAVE_BOT + i * SZR)) / SZR;
2890 }
2891
2892 debug_frame.fde.func_start = (uintptr_t)buf;
2893 debug_frame.fde.func_len = buf_size;
2894
2895 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2896 }
2897 #endif /* __ELF__ */
2898
2899 void flush_icache_range(uintptr_t start, uintptr_t stop)
2900 {
2901 uintptr_t p, start1, stop1;
2902 size_t dsize = qemu_dcache_linesize;
2903 size_t isize = qemu_icache_linesize;
2904
2905 start1 = start & ~(dsize - 1);
2906 stop1 = (stop + dsize - 1) & ~(dsize - 1);
2907 for (p = start1; p < stop1; p += dsize) {
2908 asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
2909 }
2910 asm volatile ("sync" : : : "memory");
2911
2912 start &= start & ~(isize - 1);
2913 stop1 = (stop + isize - 1) & ~(isize - 1);
2914 for (p = start1; p < stop1; p += isize) {
2915 asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
2916 }
2917 asm volatile ("sync" : : : "memory");
2918 asm volatile ("isync" : : : "memory");
2919 }
AR7 emulation for QEMU