?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qapi: add QAPI visitor core
[qemu/aliguori.git] / tcg / s390 / tcg-target.c
blob2fc564640075389529fd90b3f61dc77fcd950d28
1 /*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2009 Ulrich Hecht <uli@suse.de>
5 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
6 * Copyright (c) 2010 Richard Henderson <rth@twiddle.net>
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 */
26
27 /* ??? The translation blocks produced by TCG are generally small enough to
28 be entirely reachable with a 16-bit displacement. Leaving the option for
29 a 32-bit displacement here Just In Case. */
30 #define USE_LONG_BRANCHES 0
31
32 #define TCG_CT_CONST_32 0x0100
33 #define TCG_CT_CONST_NEG 0x0200
34 #define TCG_CT_CONST_ADDI 0x0400
35 #define TCG_CT_CONST_MULI 0x0800
36 #define TCG_CT_CONST_ANDI 0x1000
37 #define TCG_CT_CONST_ORI 0x2000
38 #define TCG_CT_CONST_XORI 0x4000
39 #define TCG_CT_CONST_CMPI 0x8000
40
41 /* Several places within the instruction set 0 means "no register"
42 rather than TCG_REG_R0. */
43 #define TCG_REG_NONE 0
44
45 /* A scratch register that may be be used throughout the backend. */
46 #define TCG_TMP0 TCG_REG_R14
47
48 #ifdef CONFIG_USE_GUEST_BASE
49 #define TCG_GUEST_BASE_REG TCG_REG_R13
50 #else
51 #define TCG_GUEST_BASE_REG TCG_REG_R0
52 #endif
53
54 #ifndef GUEST_BASE
55 #define GUEST_BASE 0
56 #endif
57
58
59 /* All of the following instructions are prefixed with their instruction
60 format, and are defined as 8- or 16-bit quantities, even when the two
61 halves of the 16-bit quantity may appear 32 bits apart in the insn.
62 This makes it easy to copy the values from the tables in Appendix B. */
63 typedef enum S390Opcode {
64 RIL_AFI = 0xc209,
65 RIL_AGFI = 0xc208,
66 RIL_ALGFI = 0xc20a,
67 RIL_BRASL = 0xc005,
68 RIL_BRCL = 0xc004,
69 RIL_CFI = 0xc20d,
70 RIL_CGFI = 0xc20c,
71 RIL_CLFI = 0xc20f,
72 RIL_CLGFI = 0xc20e,
73 RIL_IIHF = 0xc008,
74 RIL_IILF = 0xc009,
75 RIL_LARL = 0xc000,
76 RIL_LGFI = 0xc001,
77 RIL_LGRL = 0xc408,
78 RIL_LLIHF = 0xc00e,
79 RIL_LLILF = 0xc00f,
80 RIL_LRL = 0xc40d,
81 RIL_MSFI = 0xc201,
82 RIL_MSGFI = 0xc200,
83 RIL_NIHF = 0xc00a,
84 RIL_NILF = 0xc00b,
85 RIL_OIHF = 0xc00c,
86 RIL_OILF = 0xc00d,
87 RIL_XIHF = 0xc006,
88 RIL_XILF = 0xc007,
89
90 RI_AGHI = 0xa70b,
91 RI_AHI = 0xa70a,
92 RI_BRC = 0xa704,
93 RI_IIHH = 0xa500,
94 RI_IIHL = 0xa501,
95 RI_IILH = 0xa502,
96 RI_IILL = 0xa503,
97 RI_LGHI = 0xa709,
98 RI_LLIHH = 0xa50c,
99 RI_LLIHL = 0xa50d,
100 RI_LLILH = 0xa50e,
101 RI_LLILL = 0xa50f,
102 RI_MGHI = 0xa70d,
103 RI_MHI = 0xa70c,
104 RI_NIHH = 0xa504,
105 RI_NIHL = 0xa505,
106 RI_NILH = 0xa506,
107 RI_NILL = 0xa507,
108 RI_OIHH = 0xa508,
109 RI_OIHL = 0xa509,
110 RI_OILH = 0xa50a,
111 RI_OILL = 0xa50b,
112
113 RIE_CGIJ = 0xec7c,
114 RIE_CGRJ = 0xec64,
115 RIE_CIJ = 0xec7e,
116 RIE_CLGRJ = 0xec65,
117 RIE_CLIJ = 0xec7f,
118 RIE_CLGIJ = 0xec7d,
119 RIE_CLRJ = 0xec77,
120 RIE_CRJ = 0xec76,
121
122 RRE_AGR = 0xb908,
123 RRE_CGR = 0xb920,
124 RRE_CLGR = 0xb921,
125 RRE_DLGR = 0xb987,
126 RRE_DLR = 0xb997,
127 RRE_DSGFR = 0xb91d,
128 RRE_DSGR = 0xb90d,
129 RRE_LGBR = 0xb906,
130 RRE_LCGR = 0xb903,
131 RRE_LGFR = 0xb914,
132 RRE_LGHR = 0xb907,
133 RRE_LGR = 0xb904,
134 RRE_LLGCR = 0xb984,
135 RRE_LLGFR = 0xb916,
136 RRE_LLGHR = 0xb985,
137 RRE_LRVR = 0xb91f,
138 RRE_LRVGR = 0xb90f,
139 RRE_LTGR = 0xb902,
140 RRE_MSGR = 0xb90c,
141 RRE_MSR = 0xb252,
142 RRE_NGR = 0xb980,
143 RRE_OGR = 0xb981,
144 RRE_SGR = 0xb909,
145 RRE_XGR = 0xb982,
146
147 RR_AR = 0x1a,
148 RR_BASR = 0x0d,
149 RR_BCR = 0x07,
150 RR_CLR = 0x15,
151 RR_CR = 0x19,
152 RR_DR = 0x1d,
153 RR_LCR = 0x13,
154 RR_LR = 0x18,
155 RR_LTR = 0x12,
156 RR_NR = 0x14,
157 RR_OR = 0x16,
158 RR_SR = 0x1b,
159 RR_XR = 0x17,
160
161 RSY_RLL = 0xeb1d,
162 RSY_RLLG = 0xeb1c,
163 RSY_SLLG = 0xeb0d,
164 RSY_SRAG = 0xeb0a,
165 RSY_SRLG = 0xeb0c,
166
167 RS_SLL = 0x89,
168 RS_SRA = 0x8a,
169 RS_SRL = 0x88,
170
171 RXY_AG = 0xe308,
172 RXY_AY = 0xe35a,
173 RXY_CG = 0xe320,
174 RXY_CY = 0xe359,
175 RXY_LB = 0xe376,
176 RXY_LG = 0xe304,
177 RXY_LGB = 0xe377,
178 RXY_LGF = 0xe314,
179 RXY_LGH = 0xe315,
180 RXY_LHY = 0xe378,
181 RXY_LLGC = 0xe390,
182 RXY_LLGF = 0xe316,
183 RXY_LLGH = 0xe391,
184 RXY_LMG = 0xeb04,
185 RXY_LRV = 0xe31e,
186 RXY_LRVG = 0xe30f,
187 RXY_LRVH = 0xe31f,
188 RXY_LY = 0xe358,
189 RXY_STCY = 0xe372,
190 RXY_STG = 0xe324,
191 RXY_STHY = 0xe370,
192 RXY_STMG = 0xeb24,
193 RXY_STRV = 0xe33e,
194 RXY_STRVG = 0xe32f,
195 RXY_STRVH = 0xe33f,
196 RXY_STY = 0xe350,
197
198 RX_A = 0x5a,
199 RX_C = 0x59,
200 RX_L = 0x58,
201 RX_LH = 0x48,
202 RX_ST = 0x50,
203 RX_STC = 0x42,
204 RX_STH = 0x40,
205 } S390Opcode;
206
207 #define LD_SIGNED 0x04
208 #define LD_UINT8 0x00
209 #define LD_INT8 (LD_UINT8 | LD_SIGNED)
210 #define LD_UINT16 0x01
211 #define LD_INT16 (LD_UINT16 | LD_SIGNED)
212 #define LD_UINT32 0x02
213 #define LD_INT32 (LD_UINT32 | LD_SIGNED)
214 #define LD_UINT64 0x03
215 #define LD_INT64 (LD_UINT64 | LD_SIGNED)
216
217 #ifndef NDEBUG
218 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
219 "%r0", "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7",
220 "%r8", "%r9", "%r10" "%r11" "%r12" "%r13" "%r14" "%r15"
221 };
222 #endif
223
224 /* Since R6 is a potential argument register, choose it last of the
225 call-saved registers. Likewise prefer the call-clobbered registers
226 in reverse order to maximize the chance of avoiding the arguments. */
227 static const int tcg_target_reg_alloc_order[] = {
228 TCG_REG_R13,
229 TCG_REG_R12,
230 TCG_REG_R11,
231 TCG_REG_R10,
232 TCG_REG_R9,
233 TCG_REG_R8,
234 TCG_REG_R7,
235 TCG_REG_R6,
236 TCG_REG_R14,
237 TCG_REG_R0,
238 TCG_REG_R1,
239 TCG_REG_R5,
240 TCG_REG_R4,
241 TCG_REG_R3,
242 TCG_REG_R2,
243 };
244
245 static const int tcg_target_call_iarg_regs[] = {
246 TCG_REG_R2,
247 TCG_REG_R3,
248 TCG_REG_R4,
249 TCG_REG_R5,
250 TCG_REG_R6,
251 };
252
253 static const int tcg_target_call_oarg_regs[] = {
254 TCG_REG_R2,
255 TCG_REG_R3,
256 };
257
258 #define S390_CC_EQ 8
259 #define S390_CC_LT 4
260 #define S390_CC_GT 2
261 #define S390_CC_OV 1
262 #define S390_CC_NE (S390_CC_LT | S390_CC_GT)
263 #define S390_CC_LE (S390_CC_LT | S390_CC_EQ)
264 #define S390_CC_GE (S390_CC_GT | S390_CC_EQ)
265 #define S390_CC_NEVER 0
266 #define S390_CC_ALWAYS 15
267
268 /* Condition codes that result from a COMPARE and COMPARE LOGICAL. */
269 static const uint8_t tcg_cond_to_s390_cond[10] = {
270 [TCG_COND_EQ] = S390_CC_EQ,
271 [TCG_COND_NE] = S390_CC_NE,
272 [TCG_COND_LT] = S390_CC_LT,
273 [TCG_COND_LE] = S390_CC_LE,
274 [TCG_COND_GT] = S390_CC_GT,
275 [TCG_COND_GE] = S390_CC_GE,
276 [TCG_COND_LTU] = S390_CC_LT,
277 [TCG_COND_LEU] = S390_CC_LE,
278 [TCG_COND_GTU] = S390_CC_GT,
279 [TCG_COND_GEU] = S390_CC_GE,
280 };
281
282 /* Condition codes that result from a LOAD AND TEST. Here, we have no
283 unsigned instruction variation, however since the test is vs zero we
284 can re-map the outcomes appropriately. */
285 static const uint8_t tcg_cond_to_ltr_cond[10] = {
286 [TCG_COND_EQ] = S390_CC_EQ,
287 [TCG_COND_NE] = S390_CC_NE,
288 [TCG_COND_LT] = S390_CC_LT,
289 [TCG_COND_LE] = S390_CC_LE,
290 [TCG_COND_GT] = S390_CC_GT,
291 [TCG_COND_GE] = S390_CC_GE,
292 [TCG_COND_LTU] = S390_CC_NEVER,
293 [TCG_COND_LEU] = S390_CC_EQ,
294 [TCG_COND_GTU] = S390_CC_NE,
295 [TCG_COND_GEU] = S390_CC_ALWAYS,
296 };
297
298 #ifdef CONFIG_SOFTMMU
299
300 #include "../../softmmu_defs.h"
301
302 static void *qemu_ld_helpers[4] = {
303 __ldb_mmu,
304 __ldw_mmu,
305 __ldl_mmu,
306 __ldq_mmu,
307 };
308
309 static void *qemu_st_helpers[4] = {
310 __stb_mmu,
311 __stw_mmu,
312 __stl_mmu,
313 __stq_mmu,
314 };
315 #endif
316
317 static uint8_t *tb_ret_addr;
318
319 /* A list of relevant facilities used by this translator. Some of these
320 are required for proper operation, and these are checked at startup. */
321
322 #define FACILITY_ZARCH_ACTIVE (1ULL << (63 - 2))
323 #define FACILITY_LONG_DISP (1ULL << (63 - 18))
324 #define FACILITY_EXT_IMM (1ULL << (63 - 21))
325 #define FACILITY_GEN_INST_EXT (1ULL << (63 - 34))
326
327 static uint64_t facilities;
328
329 static void patch_reloc(uint8_t *code_ptr, int type,
330 tcg_target_long value, tcg_target_long addend)
331 {
332 tcg_target_long code_ptr_tl = (tcg_target_long)code_ptr;
333 tcg_target_long pcrel2;
334
335 /* ??? Not the usual definition of "addend". */
336 pcrel2 = (value - (code_ptr_tl + addend)) >> 1;
337
338 switch (type) {
339 case R_390_PC16DBL:
340 assert(pcrel2 == (int16_t)pcrel2);
341 *(int16_t *)code_ptr = pcrel2;
342 break;
343 case R_390_PC32DBL:
344 assert(pcrel2 == (int32_t)pcrel2);
345 *(int32_t *)code_ptr = pcrel2;
346 break;
347 default:
348 tcg_abort();
349 break;
350 }
351 }
352
353 static int tcg_target_get_call_iarg_regs_count(int flags)
354 {
355 return sizeof(tcg_target_call_iarg_regs) / sizeof(int);
356 }
357
358 /* parse target specific constraints */
359 static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
360 {
361 const char *ct_str = *pct_str;
362
363 switch (ct_str[0]) {
364 case 'r': /* all registers */
365 ct->ct |= TCG_CT_REG;
366 tcg_regset_set32(ct->u.regs, 0, 0xffff);
367 break;
368 case 'R': /* not R0 */
369 ct->ct |= TCG_CT_REG;
370 tcg_regset_set32(ct->u.regs, 0, 0xffff);
371 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
372 break;
373 case 'L': /* qemu_ld/st constraint */
374 ct->ct |= TCG_CT_REG;
375 tcg_regset_set32(ct->u.regs, 0, 0xffff);
376 tcg_regset_reset_reg (ct->u.regs, TCG_REG_R2);
377 tcg_regset_reset_reg (ct->u.regs, TCG_REG_R3);
378 break;
379 case 'a': /* force R2 for division */
380 ct->ct |= TCG_CT_REG;
381 tcg_regset_clear(ct->u.regs);
382 tcg_regset_set_reg(ct->u.regs, TCG_REG_R2);
383 break;
384 case 'b': /* force R3 for division */
385 ct->ct |= TCG_CT_REG;
386 tcg_regset_clear(ct->u.regs);
387 tcg_regset_set_reg(ct->u.regs, TCG_REG_R3);
388 break;
389 case 'N': /* force immediate negate */
390 ct->ct |= TCG_CT_CONST_NEG;
391 break;
392 case 'W': /* force 32-bit ("word") immediate */
393 ct->ct |= TCG_CT_CONST_32;
394 break;
395 case 'I':
396 ct->ct |= TCG_CT_CONST_ADDI;
397 break;
398 case 'K':
399 ct->ct |= TCG_CT_CONST_MULI;
400 break;
401 case 'A':
402 ct->ct |= TCG_CT_CONST_ANDI;
403 break;
404 case 'O':
405 ct->ct |= TCG_CT_CONST_ORI;
406 break;
407 case 'X':
408 ct->ct |= TCG_CT_CONST_XORI;
409 break;
410 case 'C':
411 ct->ct |= TCG_CT_CONST_CMPI;
412 break;
413 default:
414 return -1;
415 }
416 ct_str++;
417 *pct_str = ct_str;
418
419 return 0;
420 }
421
422 /* Immediates to be used with logical AND. This is an optimization only,
423 since a full 64-bit immediate AND can always be performed with 4 sequential
424 NI[LH][LH] instructions. What we're looking for is immediates that we
425 can load efficiently, and the immediate load plus the reg-reg AND is
426 smaller than the sequential NI's. */
427
428 static int tcg_match_andi(int ct, tcg_target_ulong val)
429 {
430 int i;
431
432 if (facilities & FACILITY_EXT_IMM) {
433 if (ct & TCG_CT_CONST_32) {
434 /* All 32-bit ANDs can be performed with 1 48-bit insn. */
435 return 1;
436 }
437
438 /* Zero-extensions. */
439 if (val == 0xff || val == 0xffff || val == 0xffffffff) {
440 return 1;
441 }
442 } else {
443 if (ct & TCG_CT_CONST_32) {
444 val = (uint32_t)val;
445 } else if (val == 0xffffffff) {
446 return 1;
447 }
448 }
449
450 /* Try all 32-bit insns that can perform it in one go. */
451 for (i = 0; i < 4; i++) {
452 tcg_target_ulong mask = ~(0xffffull << i*16);
453 if ((val & mask) == mask) {
454 return 1;
455 }
456 }
457
458 /* Look for 16-bit values performing the mask. These are better
459 to load with LLI[LH][LH]. */
460 for (i = 0; i < 4; i++) {
461 tcg_target_ulong mask = 0xffffull << i*16;
462 if ((val & mask) == val) {
463 return 0;
464 }
465 }
466
467 /* Look for 32-bit values performing the 64-bit mask. These
468 are better to load with LLI[LH]F, or if extended immediates
469 not available, with a pair of LLI insns. */
470 if ((ct & TCG_CT_CONST_32) == 0) {
471 if (val <= 0xffffffff || (val & 0xffffffff) == 0) {
472 return 0;
473 }
474 }
475
476 return 1;
477 }
478
479 /* Immediates to be used with logical OR. This is an optimization only,
480 since a full 64-bit immediate OR can always be performed with 4 sequential
481 OI[LH][LH] instructions. What we're looking for is immediates that we
482 can load efficiently, and the immediate load plus the reg-reg OR is
483 smaller than the sequential OI's. */
484
485 static int tcg_match_ori(int ct, tcg_target_long val)
486 {
487 if (facilities & FACILITY_EXT_IMM) {
488 if (ct & TCG_CT_CONST_32) {
489 /* All 32-bit ORs can be performed with 1 48-bit insn. */
490 return 1;
491 }
492 }
493
494 /* Look for negative values. These are best to load with LGHI. */
495 if (val < 0) {
496 if (val == (int16_t)val) {
497 return 0;
498 }
499 if (facilities & FACILITY_EXT_IMM) {
500 if (val == (int32_t)val) {
501 return 0;
502 }
503 }
504 }
505
506 return 1;
507 }
508
509 /* Immediates to be used with logical XOR. This is almost, but not quite,
510 only an optimization. XOR with immediate is only supported with the
511 extended-immediate facility. That said, there are a few patterns for
512 which it is better to load the value into a register first. */
513
514 static int tcg_match_xori(int ct, tcg_target_long val)
515 {
516 if ((facilities & FACILITY_EXT_IMM) == 0) {
517 return 0;
518 }
519
520 if (ct & TCG_CT_CONST_32) {
521 /* All 32-bit XORs can be performed with 1 48-bit insn. */
522 return 1;
523 }
524
525 /* Look for negative values. These are best to load with LGHI. */
526 if (val < 0 && val == (int32_t)val) {
527 return 0;
528 }
529
530 return 1;
531 }
532
533 /* Imediates to be used with comparisons. */
534
535 static int tcg_match_cmpi(int ct, tcg_target_long val)
536 {
537 if (facilities & FACILITY_EXT_IMM) {
538 /* The COMPARE IMMEDIATE instruction is available. */
539 if (ct & TCG_CT_CONST_32) {
540 /* We have a 32-bit immediate and can compare against anything. */
541 return 1;
542 } else {
543 /* ??? We have no insight here into whether the comparison is
544 signed or unsigned. The COMPARE IMMEDIATE insn uses a 32-bit
545 signed immediate, and the COMPARE LOGICAL IMMEDIATE insn uses
546 a 32-bit unsigned immediate. If we were to use the (semi)
547 obvious "val == (int32_t)val" we would be enabling unsigned
548 comparisons vs very large numbers. The only solution is to
549 take the intersection of the ranges. */
550 /* ??? Another possible solution is to simply lie and allow all
551 constants here and force the out-of-range values into a temp
552 register in tgen_cmp when we have knowledge of the actual
553 comparison code in use. */
554 return val >= 0 && val <= 0x7fffffff;
555 }
556 } else {
557 /* Only the LOAD AND TEST instruction is available. */
558 return val == 0;
559 }
560 }
561
562 /* Test if a constant matches the constraint. */
563 static int tcg_target_const_match(tcg_target_long val,
564 const TCGArgConstraint *arg_ct)
565 {
566 int ct = arg_ct->ct;
567
568 if (ct & TCG_CT_CONST) {
569 return 1;
570 }
571
572 /* Handle the modifiers. */
573 if (ct & TCG_CT_CONST_NEG) {
574 val = -val;
575 }
576 if (ct & TCG_CT_CONST_32) {
577 val = (int32_t)val;
578 }
579
580 /* The following are mutually exclusive. */
581 if (ct & TCG_CT_CONST_ADDI) {
582 /* Immediates that may be used with add. If we have the
583 extended-immediates facility then we have ADD IMMEDIATE
584 with signed and unsigned 32-bit, otherwise we have only
585 ADD HALFWORD IMMEDIATE with a signed 16-bit. */
586 if (facilities & FACILITY_EXT_IMM) {
587 return val == (int32_t)val || val == (uint32_t)val;
588 } else {
589 return val == (int16_t)val;
590 }
591 } else if (ct & TCG_CT_CONST_MULI) {
592 /* Immediates that may be used with multiply. If we have the
593 general-instruction-extensions, then we have MULTIPLY SINGLE
594 IMMEDIATE with a signed 32-bit, otherwise we have only
595 MULTIPLY HALFWORD IMMEDIATE, with a signed 16-bit. */
596 if (facilities & FACILITY_GEN_INST_EXT) {
597 return val == (int32_t)val;
598 } else {
599 return val == (int16_t)val;
600 }
601 } else if (ct & TCG_CT_CONST_ANDI) {
602 return tcg_match_andi(ct, val);
603 } else if (ct & TCG_CT_CONST_ORI) {
604 return tcg_match_ori(ct, val);
605 } else if (ct & TCG_CT_CONST_XORI) {
606 return tcg_match_xori(ct, val);
607 } else if (ct & TCG_CT_CONST_CMPI) {
608 return tcg_match_cmpi(ct, val);
609 }
610
611 return 0;
612 }
613
614 /* Emit instructions according to the given instruction format. */
615
616 static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
617 {
618 tcg_out16(s, (op << 8) | (r1 << 4) | r2);
619 }
620
621 static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
622 TCGReg r1, TCGReg r2)
623 {
624 tcg_out32(s, (op << 16) | (r1 << 4) | r2);
625 }
626
627 static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
628 {
629 tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
630 }
631
632 static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
633 {
634 tcg_out16(s, op | (r1 << 4));
635 tcg_out32(s, i2);
636 }
637
638 static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
639 TCGReg b2, TCGReg r3, int disp)
640 {
641 tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
642 | (disp & 0xfff));
643 }
644
645 static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
646 TCGReg b2, TCGReg r3, int disp)
647 {
648 tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
649 tcg_out32(s, (op & 0xff) | (b2 << 28)
650 | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
651 }
652
653 #define tcg_out_insn_RX tcg_out_insn_RS
654 #define tcg_out_insn_RXY tcg_out_insn_RSY
655
656 /* Emit an opcode with "type-checking" of the format. */
657 #define tcg_out_insn(S, FMT, OP, ...) \
658 glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
659
660
661 /* emit 64-bit shifts */
662 static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
663 TCGReg src, TCGReg sh_reg, int sh_imm)
664 {
665 tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
666 }
667
668 /* emit 32-bit shifts */
669 static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
670 TCGReg sh_reg, int sh_imm)
671 {
672 tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
673 }
674
675 static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
676 {
677 if (src != dst) {
678 if (type == TCG_TYPE_I32) {
679 tcg_out_insn(s, RR, LR, dst, src);
680 } else {
681 tcg_out_insn(s, RRE, LGR, dst, src);
682 }
683 }
684 }
685
686 /* load a register with an immediate value */
687 static void tcg_out_movi(TCGContext *s, TCGType type,
688 TCGReg ret, tcg_target_long sval)
689 {
690 static const S390Opcode lli_insns[4] = {
691 RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
692 };
693
694 tcg_target_ulong uval = sval;
695 int i;
696
697 if (type == TCG_TYPE_I32) {
698 uval = (uint32_t)sval;
699 sval = (int32_t)sval;
700 }
701
702 /* Try all 32-bit insns that can load it in one go. */
703 if (sval >= -0x8000 && sval < 0x8000) {
704 tcg_out_insn(s, RI, LGHI, ret, sval);
705 return;
706 }
707
708 for (i = 0; i < 4; i++) {
709 tcg_target_long mask = 0xffffull << i*16;
710 if ((uval & mask) == uval) {
711 tcg_out_insn_RI(s, lli_insns[i], ret, uval >> i*16);
712 return;
713 }
714 }
715
716 /* Try all 48-bit insns that can load it in one go. */
717 if (facilities & FACILITY_EXT_IMM) {
718 if (sval == (int32_t)sval) {
719 tcg_out_insn(s, RIL, LGFI, ret, sval);
720 return;
721 }
722 if (uval <= 0xffffffff) {
723 tcg_out_insn(s, RIL, LLILF, ret, uval);
724 return;
725 }
726 if ((uval & 0xffffffff) == 0) {
727 tcg_out_insn(s, RIL, LLIHF, ret, uval >> 31 >> 1);
728 return;
729 }
730 }
731
732 /* Try for PC-relative address load. */
733 if ((sval & 1) == 0) {
734 intptr_t off = (sval - (intptr_t)s->code_ptr) >> 1;
735 if (off == (int32_t)off) {
736 tcg_out_insn(s, RIL, LARL, ret, off);
737 return;
738 }
739 }
740
741 /* If extended immediates are not present, then we may have to issue
742 several instructions to load the low 32 bits. */
743 if (!(facilities & FACILITY_EXT_IMM)) {
744 /* A 32-bit unsigned value can be loaded in 2 insns. And given
745 that the lli_insns loop above did not succeed, we know that
746 both insns are required. */
747 if (uval <= 0xffffffff) {
748 tcg_out_insn(s, RI, LLILL, ret, uval);
749 tcg_out_insn(s, RI, IILH, ret, uval >> 16);
750 return;
751 }
752
753 /* If all high bits are set, the value can be loaded in 2 or 3 insns.
754 We first want to make sure that all the high bits get set. With
755 luck the low 16-bits can be considered negative to perform that for
756 free, otherwise we load an explicit -1. */
757 if (sval >> 31 >> 1 == -1) {
758 if (uval & 0x8000) {
759 tcg_out_insn(s, RI, LGHI, ret, uval);
760 } else {
761 tcg_out_insn(s, RI, LGHI, ret, -1);
762 tcg_out_insn(s, RI, IILL, ret, uval);
763 }
764 tcg_out_insn(s, RI, IILH, ret, uval >> 16);
765 return;
766 }
767 }
768
769 /* If we get here, both the high and low parts have non-zero bits. */
770
771 /* Recurse to load the lower 32-bits. */
772 tcg_out_movi(s, TCG_TYPE_I32, ret, sval);
773
774 /* Insert data into the high 32-bits. */
775 uval = uval >> 31 >> 1;
776 if (facilities & FACILITY_EXT_IMM) {
777 if (uval < 0x10000) {
778 tcg_out_insn(s, RI, IIHL, ret, uval);
779 } else if ((uval & 0xffff) == 0) {
780 tcg_out_insn(s, RI, IIHH, ret, uval >> 16);
781 } else {
782 tcg_out_insn(s, RIL, IIHF, ret, uval);
783 }
784 } else {
785 if (uval & 0xffff) {
786 tcg_out_insn(s, RI, IIHL, ret, uval);
787 }
788 if (uval & 0xffff0000) {
789 tcg_out_insn(s, RI, IIHH, ret, uval >> 16);
790 }
791 }
792 }
793
794
795 /* Emit a load/store type instruction. Inputs are:
796 DATA: The register to be loaded or stored.
797 BASE+OFS: The effective address.
798 OPC_RX: If the operation has an RX format opcode (e.g. STC), otherwise 0.
799 OPC_RXY: The RXY format opcode for the operation (e.g. STCY). */
800
801 static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
802 TCGReg data, TCGReg base, TCGReg index,
803 tcg_target_long ofs)
804 {
805 if (ofs < -0x80000 || ofs >= 0x80000) {
806 /* Combine the low 16 bits of the offset with the actual load insn;
807 the high 48 bits must come from an immediate load. */
808 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs & ~0xffff);
809 ofs &= 0xffff;
810
811 /* If we were already given an index register, add it in. */
812 if (index != TCG_REG_NONE) {
813 tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
814 }
815 index = TCG_TMP0;
816 }
817
818 if (opc_rx && ofs >= 0 && ofs < 0x1000) {
819 tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
820 } else {
821 tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
822 }
823 }
824
825
826 /* load data without address translation or endianness conversion */
827 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
828 TCGReg base, tcg_target_long ofs)
829 {
830 if (type == TCG_TYPE_I32) {
831 tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
832 } else {
833 tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
834 }
835 }
836
837 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
838 TCGReg base, tcg_target_long ofs)
839 {
840 if (type == TCG_TYPE_I32) {
841 tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
842 } else {
843 tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
844 }
845 }
846
847 /* load data from an absolute host address */
848 static void tcg_out_ld_abs(TCGContext *s, TCGType type, TCGReg dest, void *abs)
849 {
850 tcg_target_long addr = (tcg_target_long)abs;
851
852 if (facilities & FACILITY_GEN_INST_EXT) {
853 tcg_target_long disp = (addr - (tcg_target_long)s->code_ptr) >> 1;
854 if (disp == (int32_t)disp) {
855 if (type == TCG_TYPE_I32) {
856 tcg_out_insn(s, RIL, LRL, dest, disp);
857 } else {
858 tcg_out_insn(s, RIL, LGRL, dest, disp);
859 }
860 return;
861 }
862 }
863
864 tcg_out_movi(s, TCG_TYPE_PTR, dest, addr & ~0xffff);
865 tcg_out_ld(s, type, dest, dest, addr & 0xffff);
866 }
867
868 static void tgen_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
869 {
870 if (facilities & FACILITY_EXT_IMM) {
871 tcg_out_insn(s, RRE, LGBR, dest, src);
872 return;
873 }
874
875 if (type == TCG_TYPE_I32) {
876 if (dest == src) {
877 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 24);
878 } else {
879 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 24);
880 }
881 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 24);
882 } else {
883 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 56);
884 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 56);
885 }
886 }
887
888 static void tgen_ext8u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
889 {
890 if (facilities & FACILITY_EXT_IMM) {
891 tcg_out_insn(s, RRE, LLGCR, dest, src);
892 return;
893 }
894
895 if (dest == src) {
896 tcg_out_movi(s, type, TCG_TMP0, 0xff);
897 src = TCG_TMP0;
898 } else {
899 tcg_out_movi(s, type, dest, 0xff);
900 }
901 if (type == TCG_TYPE_I32) {
902 tcg_out_insn(s, RR, NR, dest, src);
903 } else {
904 tcg_out_insn(s, RRE, NGR, dest, src);
905 }
906 }
907
908 static void tgen_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
909 {
910 if (facilities & FACILITY_EXT_IMM) {
911 tcg_out_insn(s, RRE, LGHR, dest, src);
912 return;
913 }
914
915 if (type == TCG_TYPE_I32) {
916 if (dest == src) {
917 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 16);
918 } else {
919 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 16);
920 }
921 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 16);
922 } else {
923 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 48);
924 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 48);
925 }
926 }
927
928 static void tgen_ext16u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
929 {
930 if (facilities & FACILITY_EXT_IMM) {
931 tcg_out_insn(s, RRE, LLGHR, dest, src);
932 return;
933 }
934
935 if (dest == src) {
936 tcg_out_movi(s, type, TCG_TMP0, 0xffff);
937 src = TCG_TMP0;
938 } else {
939 tcg_out_movi(s, type, dest, 0xffff);
940 }
941 if (type == TCG_TYPE_I32) {
942 tcg_out_insn(s, RR, NR, dest, src);
943 } else {
944 tcg_out_insn(s, RRE, NGR, dest, src);
945 }
946 }
947
948 static inline void tgen_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
949 {
950 tcg_out_insn(s, RRE, LGFR, dest, src);
951 }
952
953 static inline void tgen_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
954 {
955 tcg_out_insn(s, RRE, LLGFR, dest, src);
956 }
957
958 static inline void tgen32_addi(TCGContext *s, TCGReg dest, int32_t val)
959 {
960 if (val == (int16_t)val) {
961 tcg_out_insn(s, RI, AHI, dest, val);
962 } else {
963 tcg_out_insn(s, RIL, AFI, dest, val);
964 }
965 }
966
967 static inline void tgen64_addi(TCGContext *s, TCGReg dest, int64_t val)
968 {
969 if (val == (int16_t)val) {
970 tcg_out_insn(s, RI, AGHI, dest, val);
971 } else if (val == (int32_t)val) {
972 tcg_out_insn(s, RIL, AGFI, dest, val);
973 } else if (val == (uint32_t)val) {
974 tcg_out_insn(s, RIL, ALGFI, dest, val);
975 } else {
976 tcg_abort();
977 }
978
979 }
980
981 static void tgen64_andi(TCGContext *s, TCGReg dest, tcg_target_ulong val)
982 {
983 static const S390Opcode ni_insns[4] = {
984 RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
985 };
986 static const S390Opcode nif_insns[2] = {
987 RIL_NILF, RIL_NIHF
988 };
989
990 int i;
991
992 /* Look for no-op. */
993 if (val == -1) {
994 return;
995 }
996
997 /* Look for the zero-extensions. */
998 if (val == 0xffffffff) {
999 tgen_ext32u(s, dest, dest);
1000 return;
1001 }
1002
1003 if (facilities & FACILITY_EXT_IMM) {
1004 if (val == 0xff) {
1005 tgen_ext8u(s, TCG_TYPE_I64, dest, dest);
1006 return;
1007 }
1008 if (val == 0xffff) {
1009 tgen_ext16u(s, TCG_TYPE_I64, dest, dest);
1010 return;
1011 }
1012
1013 /* Try all 32-bit insns that can perform it in one go. */
1014 for (i = 0; i < 4; i++) {
1015 tcg_target_ulong mask = ~(0xffffull << i*16);
1016 if ((val & mask) == mask) {
1017 tcg_out_insn_RI(s, ni_insns[i], dest, val >> i*16);
1018 return;
1019 }
1020 }
1021
1022 /* Try all 48-bit insns that can perform it in one go. */
1023 if (facilities & FACILITY_EXT_IMM) {
1024 for (i = 0; i < 2; i++) {
1025 tcg_target_ulong mask = ~(0xffffffffull << i*32);
1026 if ((val & mask) == mask) {
1027 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> i*32);
1028 return;
1029 }
1030 }
1031 }
1032
1033 /* Perform the AND via sequential modifications to the high and low
1034 parts. Do this via recursion to handle 16-bit vs 32-bit masks in
1035 each half. */
1036 tgen64_andi(s, dest, val | 0xffffffff00000000ull);
1037 tgen64_andi(s, dest, val | 0x00000000ffffffffull);
1038 } else {
1039 /* With no extended-immediate facility, just emit the sequence. */
1040 for (i = 0; i < 4; i++) {
1041 tcg_target_ulong mask = 0xffffull << i*16;
1042 if ((val & mask) != mask) {
1043 tcg_out_insn_RI(s, ni_insns[i], dest, val >> i*16);
1044 }
1045 }
1046 }
1047 }
1048
1049 static void tgen64_ori(TCGContext *s, TCGReg dest, tcg_target_ulong val)
1050 {
1051 static const S390Opcode oi_insns[4] = {
1052 RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
1053 };
1054 static const S390Opcode nif_insns[2] = {
1055 RIL_OILF, RIL_OIHF
1056 };
1057
1058 int i;
1059
1060 /* Look for no-op. */
1061 if (val == 0) {
1062 return;
1063 }
1064
1065 if (facilities & FACILITY_EXT_IMM) {
1066 /* Try all 32-bit insns that can perform it in one go. */
1067 for (i = 0; i < 4; i++) {
1068 tcg_target_ulong mask = (0xffffull << i*16);
1069 if ((val & mask) != 0 && (val & ~mask) == 0) {
1070 tcg_out_insn_RI(s, oi_insns[i], dest, val >> i*16);
1071 return;
1072 }
1073 }
1074
1075 /* Try all 48-bit insns that can perform it in one go. */
1076 for (i = 0; i < 2; i++) {
1077 tcg_target_ulong mask = (0xffffffffull << i*32);
1078 if ((val & mask) != 0 && (val & ~mask) == 0) {
1079 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> i*32);
1080 return;
1081 }
1082 }
1083
1084 /* Perform the OR via sequential modifications to the high and
1085 low parts. Do this via recursion to handle 16-bit vs 32-bit
1086 masks in each half. */
1087 tgen64_ori(s, dest, val & 0x00000000ffffffffull);
1088 tgen64_ori(s, dest, val & 0xffffffff00000000ull);
1089 } else {
1090 /* With no extended-immediate facility, we don't need to be so
1091 clever. Just iterate over the insns and mask in the constant. */
1092 for (i = 0; i < 4; i++) {
1093 tcg_target_ulong mask = (0xffffull << i*16);
1094 if ((val & mask) != 0) {
1095 tcg_out_insn_RI(s, oi_insns[i], dest, val >> i*16);
1096 }
1097 }
1098 }
1099 }
1100
1101 static void tgen64_xori(TCGContext *s, TCGReg dest, tcg_target_ulong val)
1102 {
1103 /* Perform the xor by parts. */
1104 if (val & 0xffffffff) {
1105 tcg_out_insn(s, RIL, XILF, dest, val);
1106 }
1107 if (val > 0xffffffff) {
1108 tcg_out_insn(s, RIL, XIHF, dest, val >> 31 >> 1);
1109 }
1110 }
1111
1112 static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1113 TCGArg c2, int c2const)
1114 {
1115 bool is_unsigned = (c > TCG_COND_GT);
1116 if (c2const) {
1117 if (c2 == 0) {
1118 if (type == TCG_TYPE_I32) {
1119 tcg_out_insn(s, RR, LTR, r1, r1);
1120 } else {
1121 tcg_out_insn(s, RRE, LTGR, r1, r1);
1122 }
1123 return tcg_cond_to_ltr_cond[c];
1124 } else {
1125 if (is_unsigned) {
1126 if (type == TCG_TYPE_I32) {
1127 tcg_out_insn(s, RIL, CLFI, r1, c2);
1128 } else {
1129 tcg_out_insn(s, RIL, CLGFI, r1, c2);
1130 }
1131 } else {
1132 if (type == TCG_TYPE_I32) {
1133 tcg_out_insn(s, RIL, CFI, r1, c2);
1134 } else {
1135 tcg_out_insn(s, RIL, CGFI, r1, c2);
1136 }
1137 }
1138 }
1139 } else {
1140 if (is_unsigned) {
1141 if (type == TCG_TYPE_I32) {
1142 tcg_out_insn(s, RR, CLR, r1, c2);
1143 } else {
1144 tcg_out_insn(s, RRE, CLGR, r1, c2);
1145 }
1146 } else {
1147 if (type == TCG_TYPE_I32) {
1148 tcg_out_insn(s, RR, CR, r1, c2);
1149 } else {
1150 tcg_out_insn(s, RRE, CGR, r1, c2);
1151 }
1152 }
1153 }
1154 return tcg_cond_to_s390_cond[c];
1155 }
1156
1157 static void tgen_setcond(TCGContext *s, TCGType type, TCGCond c,
1158 TCGReg dest, TCGReg r1, TCGArg c2, int c2const)
1159 {
1160 int cc = tgen_cmp(s, type, c, r1, c2, c2const);
1161
1162 /* Emit: r1 = 1; if (cc) goto over; r1 = 0; over: */
1163 tcg_out_movi(s, type, dest, 1);
1164 tcg_out_insn(s, RI, BRC, cc, (4 + 4) >> 1);
1165 tcg_out_movi(s, type, dest, 0);
1166 }
1167
1168 static void tgen_gotoi(TCGContext *s, int cc, tcg_target_long dest)
1169 {
1170 tcg_target_long off = (dest - (tcg_target_long)s->code_ptr) >> 1;
1171 if (off > -0x8000 && off < 0x7fff) {
1172 tcg_out_insn(s, RI, BRC, cc, off);
1173 } else if (off == (int32_t)off) {
1174 tcg_out_insn(s, RIL, BRCL, cc, off);
1175 } else {
1176 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, dest);
1177 tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1178 }
1179 }
1180
1181 static void tgen_branch(TCGContext *s, int cc, int labelno)
1182 {
1183 TCGLabel* l = &s->labels[labelno];
1184 if (l->has_value) {
1185 tgen_gotoi(s, cc, l->u.value);
1186 } else if (USE_LONG_BRANCHES) {
1187 tcg_out16(s, RIL_BRCL | (cc << 4));
1188 tcg_out_reloc(s, s->code_ptr, R_390_PC32DBL, labelno, -2);
1189 s->code_ptr += 4;
1190 } else {
1191 tcg_out16(s, RI_BRC | (cc << 4));
1192 tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, labelno, -2);
1193 s->code_ptr += 2;
1194 }
1195 }
1196
1197 static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1198 TCGReg r1, TCGReg r2, int labelno)
1199 {
1200 TCGLabel* l = &s->labels[labelno];
1201 tcg_target_long off;
1202
1203 if (l->has_value) {
1204 off = (l->u.value - (tcg_target_long)s->code_ptr) >> 1;
1205 } else {
1206 /* We need to keep the offset unchanged for retranslation. */
1207 off = ((int16_t *)s->code_ptr)[1];
1208 tcg_out_reloc(s, s->code_ptr + 2, R_390_PC16DBL, labelno, -2);
1209 }
1210
1211 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1212 tcg_out16(s, off);
1213 tcg_out16(s, cc << 12 | (opc & 0xff));
1214 }
1215
1216 static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1217 TCGReg r1, int i2, int labelno)
1218 {
1219 TCGLabel* l = &s->labels[labelno];
1220 tcg_target_long off;
1221
1222 if (l->has_value) {
1223 off = (l->u.value - (tcg_target_long)s->code_ptr) >> 1;
1224 } else {
1225 /* We need to keep the offset unchanged for retranslation. */
1226 off = ((int16_t *)s->code_ptr)[1];
1227 tcg_out_reloc(s, s->code_ptr + 2, R_390_PC16DBL, labelno, -2);
1228 }
1229
1230 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1231 tcg_out16(s, off);
1232 tcg_out16(s, (i2 << 8) | (opc & 0xff));
1233 }
1234
1235 static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1236 TCGReg r1, TCGArg c2, int c2const, int labelno)
1237 {
1238 int cc;
1239
1240 if (facilities & FACILITY_GEN_INST_EXT) {
1241 bool is_unsigned = (c > TCG_COND_GT);
1242 bool in_range;
1243 S390Opcode opc;
1244
1245 cc = tcg_cond_to_s390_cond[c];
1246
1247 if (!c2const) {
1248 opc = (type == TCG_TYPE_I32
1249 ? (is_unsigned ? RIE_CLRJ : RIE_CRJ)
1250 : (is_unsigned ? RIE_CLGRJ : RIE_CGRJ));
1251 tgen_compare_branch(s, opc, cc, r1, c2, labelno);
1252 return;
1253 }
1254
1255 /* COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1256 If the immediate we've been given does not fit that range, we'll
1257 fall back to separate compare and branch instructions using the
1258 larger comparison range afforded by COMPARE IMMEDIATE. */
1259 if (type == TCG_TYPE_I32) {
1260 if (is_unsigned) {
1261 opc = RIE_CLIJ;
1262 in_range = (uint32_t)c2 == (uint8_t)c2;
1263 } else {
1264 opc = RIE_CIJ;
1265 in_range = (int32_t)c2 == (int8_t)c2;
1266 }
1267 } else {
1268 if (is_unsigned) {
1269 opc = RIE_CLGIJ;
1270 in_range = (uint64_t)c2 == (uint8_t)c2;
1271 } else {
1272 opc = RIE_CGIJ;
1273 in_range = (int64_t)c2 == (int8_t)c2;
1274 }
1275 }
1276 if (in_range) {
1277 tgen_compare_imm_branch(s, opc, cc, r1, c2, labelno);
1278 return;
1279 }
1280 }
1281
1282 cc = tgen_cmp(s, type, c, r1, c2, c2const);
1283 tgen_branch(s, cc, labelno);
1284 }
1285
1286 static void tgen_calli(TCGContext *s, tcg_target_long dest)
1287 {
1288 tcg_target_long off = (dest - (tcg_target_long)s->code_ptr) >> 1;
1289 if (off == (int32_t)off) {
1290 tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1291 } else {
1292 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, dest);
1293 tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1294 }
1295 }
1296
1297 static void tcg_out_qemu_ld_direct(TCGContext *s, int opc, TCGReg data,
1298 TCGReg base, TCGReg index, int disp)
1299 {
1300 #ifdef TARGET_WORDS_BIGENDIAN
1301 const int bswap = 0;
1302 #else
1303 const int bswap = 1;
1304 #endif
1305 switch (opc) {
1306 case LD_UINT8:
1307 tcg_out_insn(s, RXY, LLGC, data, base, index, disp);
1308 break;
1309 case LD_INT8:
1310 tcg_out_insn(s, RXY, LGB, data, base, index, disp);
1311 break;
1312 case LD_UINT16:
1313 if (bswap) {
1314 /* swapped unsigned halfword load with upper bits zeroed */
1315 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1316 tgen_ext16u(s, TCG_TYPE_I64, data, data);
1317 } else {
1318 tcg_out_insn(s, RXY, LLGH, data, base, index, disp);
1319 }
1320 break;
1321 case LD_INT16:
1322 if (bswap) {
1323 /* swapped sign-extended halfword load */
1324 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1325 tgen_ext16s(s, TCG_TYPE_I64, data, data);
1326 } else {
1327 tcg_out_insn(s, RXY, LGH, data, base, index, disp);
1328 }
1329 break;
1330 case LD_UINT32:
1331 if (bswap) {
1332 /* swapped unsigned int load with upper bits zeroed */
1333 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1334 tgen_ext32u(s, data, data);
1335 } else {
1336 tcg_out_insn(s, RXY, LLGF, data, base, index, disp);
1337 }
1338 break;
1339 case LD_INT32:
1340 if (bswap) {
1341 /* swapped sign-extended int load */
1342 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1343 tgen_ext32s(s, data, data);
1344 } else {
1345 tcg_out_insn(s, RXY, LGF, data, base, index, disp);
1346 }
1347 break;
1348 case LD_UINT64:
1349 if (bswap) {
1350 tcg_out_insn(s, RXY, LRVG, data, base, index, disp);
1351 } else {
1352 tcg_out_insn(s, RXY, LG, data, base, index, disp);
1353 }
1354 break;
1355 default:
1356 tcg_abort();
1357 }
1358 }
1359
1360 static void tcg_out_qemu_st_direct(TCGContext *s, int opc, TCGReg data,
1361 TCGReg base, TCGReg index, int disp)
1362 {
1363 #ifdef TARGET_WORDS_BIGENDIAN
1364 const int bswap = 0;
1365 #else
1366 const int bswap = 1;
1367 #endif
1368 switch (opc) {
1369 case LD_UINT8:
1370 if (disp >= 0 && disp < 0x1000) {
1371 tcg_out_insn(s, RX, STC, data, base, index, disp);
1372 } else {
1373 tcg_out_insn(s, RXY, STCY, data, base, index, disp);
1374 }
1375 break;
1376 case LD_UINT16:
1377 if (bswap) {
1378 tcg_out_insn(s, RXY, STRVH, data, base, index, disp);
1379 } else if (disp >= 0 && disp < 0x1000) {
1380 tcg_out_insn(s, RX, STH, data, base, index, disp);
1381 } else {
1382 tcg_out_insn(s, RXY, STHY, data, base, index, disp);
1383 }
1384 break;
1385 case LD_UINT32:
1386 if (bswap) {
1387 tcg_out_insn(s, RXY, STRV, data, base, index, disp);
1388 } else if (disp >= 0 && disp < 0x1000) {
1389 tcg_out_insn(s, RX, ST, data, base, index, disp);
1390 } else {
1391 tcg_out_insn(s, RXY, STY, data, base, index, disp);
1392 }
1393 break;
1394 case LD_UINT64:
1395 if (bswap) {
1396 tcg_out_insn(s, RXY, STRVG, data, base, index, disp);
1397 } else {
1398 tcg_out_insn(s, RXY, STG, data, base, index, disp);
1399 }
1400 break;
1401 default:
1402 tcg_abort();
1403 }
1404 }
1405
1406 #if defined(CONFIG_SOFTMMU)
1407 static void tgen64_andi_tmp(TCGContext *s, TCGReg dest, tcg_target_ulong val)
1408 {
1409 if (tcg_match_andi(0, val)) {
1410 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, val);
1411 tcg_out_insn(s, RRE, NGR, dest, TCG_TMP0);
1412 } else {
1413 tgen64_andi(s, dest, val);
1414 }
1415 }
1416
1417 static void tcg_prepare_qemu_ldst(TCGContext* s, TCGReg data_reg,
1418 TCGReg addr_reg, int mem_index, int opc,
1419 uint16_t **label2_ptr_p, int is_store)
1420 {
1421 const TCGReg arg0 = TCG_REG_R2;
1422 const TCGReg arg1 = TCG_REG_R3;
1423 int s_bits = opc & 3;
1424 uint16_t *label1_ptr;
1425 tcg_target_long ofs;
1426
1427 if (TARGET_LONG_BITS == 32) {
1428 tgen_ext32u(s, arg0, addr_reg);
1429 } else {
1430 tcg_out_mov(s, TCG_TYPE_I64, arg0, addr_reg);
1431 }
1432
1433 tcg_out_sh64(s, RSY_SRLG, arg1, addr_reg, TCG_REG_NONE,
1434 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1435
1436 tgen64_andi_tmp(s, arg0, TARGET_PAGE_MASK | ((1 << s_bits) - 1));
1437 tgen64_andi_tmp(s, arg1, (CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS);
1438
1439 if (is_store) {
1440 ofs = offsetof(CPUState, tlb_table[mem_index][0].addr_write);
1441 } else {
1442 ofs = offsetof(CPUState, tlb_table[mem_index][0].addr_read);
1443 }
1444 assert(ofs < 0x80000);
1445
1446 if (TARGET_LONG_BITS == 32) {
1447 tcg_out_mem(s, RX_C, RXY_CY, arg0, arg1, TCG_AREG0, ofs);
1448 } else {
1449 tcg_out_mem(s, 0, RXY_CG, arg0, arg1, TCG_AREG0, ofs);
1450 }
1451
1452 if (TARGET_LONG_BITS == 32) {
1453 tgen_ext32u(s, arg0, addr_reg);
1454 } else {
1455 tcg_out_mov(s, TCG_TYPE_I64, arg0, addr_reg);
1456 }
1457
1458 label1_ptr = (uint16_t*)s->code_ptr;
1459
1460 /* je label1 (offset will be patched in later) */
1461 tcg_out_insn(s, RI, BRC, S390_CC_EQ, 0);
1462
1463 /* call load/store helper */
1464 if (is_store) {
1465 /* Make sure to zero-extend the value to the full register
1466 for the calling convention. */
1467 switch (opc) {
1468 case LD_UINT8:
1469 tgen_ext8u(s, TCG_TYPE_I64, arg1, data_reg);
1470 break;
1471 case LD_UINT16:
1472 tgen_ext16u(s, TCG_TYPE_I64, arg1, data_reg);
1473 break;
1474 case LD_UINT32:
1475 tgen_ext32u(s, arg1, data_reg);
1476 break;
1477 case LD_UINT64:
1478 tcg_out_mov(s, TCG_TYPE_I64, arg1, data_reg);
1479 break;
1480 default:
1481 tcg_abort();
1482 }
1483 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R4, mem_index);
1484 tgen_calli(s, (tcg_target_ulong)qemu_st_helpers[s_bits]);
1485 } else {
1486 tcg_out_movi(s, TCG_TYPE_I32, arg1, mem_index);
1487 tgen_calli(s, (tcg_target_ulong)qemu_ld_helpers[s_bits]);
1488
1489 /* sign extension */
1490 switch (opc) {
1491 case LD_INT8:
1492 tgen_ext8s(s, TCG_TYPE_I64, data_reg, arg0);
1493 break;
1494 case LD_INT16:
1495 tgen_ext16s(s, TCG_TYPE_I64, data_reg, arg0);
1496 break;
1497 case LD_INT32:
1498 tgen_ext32s(s, data_reg, arg0);
1499 break;
1500 default:
1501 /* unsigned -> just copy */
1502 tcg_out_mov(s, TCG_TYPE_I64, data_reg, arg0);
1503 break;
1504 }
1505 }
1506
1507 /* jump to label2 (end) */
1508 *label2_ptr_p = (uint16_t*)s->code_ptr;
1509
1510 tcg_out_insn(s, RI, BRC, S390_CC_ALWAYS, 0);
1511
1512 /* this is label1, patch branch */
1513 *(label1_ptr + 1) = ((unsigned long)s->code_ptr -
1514 (unsigned long)label1_ptr) >> 1;
1515
1516 ofs = offsetof(CPUState, tlb_table[mem_index][0].addend);
1517 assert(ofs < 0x80000);
1518
1519 tcg_out_mem(s, 0, RXY_AG, arg0, arg1, TCG_AREG0, ofs);
1520 }
1521
1522 static void tcg_finish_qemu_ldst(TCGContext* s, uint16_t *label2_ptr)
1523 {
1524 /* patch branch */
1525 *(label2_ptr + 1) = ((unsigned long)s->code_ptr -
1526 (unsigned long)label2_ptr) >> 1;
1527 }
1528 #else
1529 static void tcg_prepare_user_ldst(TCGContext *s, TCGReg *addr_reg,
1530 TCGReg *index_reg, tcg_target_long *disp)
1531 {
1532 if (TARGET_LONG_BITS == 32) {
1533 tgen_ext32u(s, TCG_TMP0, *addr_reg);
1534 *addr_reg = TCG_TMP0;
1535 }
1536 if (GUEST_BASE < 0x80000) {
1537 *index_reg = TCG_REG_NONE;
1538 *disp = GUEST_BASE;
1539 } else {
1540 *index_reg = TCG_GUEST_BASE_REG;
1541 *disp = 0;
1542 }
1543 }
1544 #endif /* CONFIG_SOFTMMU */
1545
1546 /* load data with address translation (if applicable)
1547 and endianness conversion */
1548 static void tcg_out_qemu_ld(TCGContext* s, const TCGArg* args, int opc)
1549 {
1550 TCGReg addr_reg, data_reg;
1551 #if defined(CONFIG_SOFTMMU)
1552 int mem_index;
1553 uint16_t *label2_ptr;
1554 #else
1555 TCGReg index_reg;
1556 tcg_target_long disp;
1557 #endif
1558
1559 data_reg = *args++;
1560 addr_reg = *args++;
1561
1562 #if defined(CONFIG_SOFTMMU)
1563 mem_index = *args;
1564
1565 tcg_prepare_qemu_ldst(s, data_reg, addr_reg, mem_index,
1566 opc, &label2_ptr, 0);
1567
1568 tcg_out_qemu_ld_direct(s, opc, data_reg, TCG_REG_R2, TCG_REG_NONE, 0);
1569
1570 tcg_finish_qemu_ldst(s, label2_ptr);
1571 #else
1572 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1573 tcg_out_qemu_ld_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1574 #endif
1575 }
1576
1577 static void tcg_out_qemu_st(TCGContext* s, const TCGArg* args, int opc)
1578 {
1579 TCGReg addr_reg, data_reg;
1580 #if defined(CONFIG_SOFTMMU)
1581 int mem_index;
1582 uint16_t *label2_ptr;
1583 #else
1584 TCGReg index_reg;
1585 tcg_target_long disp;
1586 #endif
1587
1588 data_reg = *args++;
1589 addr_reg = *args++;
1590
1591 #if defined(CONFIG_SOFTMMU)
1592 mem_index = *args;
1593
1594 tcg_prepare_qemu_ldst(s, data_reg, addr_reg, mem_index,
1595 opc, &label2_ptr, 1);
1596
1597 tcg_out_qemu_st_direct(s, opc, data_reg, TCG_REG_R2, TCG_REG_NONE, 0);
1598
1599 tcg_finish_qemu_ldst(s, label2_ptr);
1600 #else
1601 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1602 tcg_out_qemu_st_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1603 #endif
1604 }
1605
1606 #if TCG_TARGET_REG_BITS == 64
1607 # define OP_32_64(x) \
1608 case glue(glue(INDEX_op_,x),_i32): \
1609 case glue(glue(INDEX_op_,x),_i64)
1610 #else
1611 # define OP_32_64(x) \
1612 case glue(glue(INDEX_op_,x),_i32)
1613 #endif
1614
1615 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1616 const TCGArg *args, const int *const_args)
1617 {
1618 S390Opcode op;
1619
1620 switch (opc) {
1621 case INDEX_op_exit_tb:
1622 /* return value */
1623 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, args[0]);
1624 tgen_gotoi(s, S390_CC_ALWAYS, (unsigned long)tb_ret_addr);
1625 break;
1626
1627 case INDEX_op_goto_tb:
1628 if (s->tb_jmp_offset) {
1629 tcg_abort();
1630 } else {
1631 /* load address stored at s->tb_next + args[0] */
1632 tcg_out_ld_abs(s, TCG_TYPE_PTR, TCG_TMP0, s->tb_next + args[0]);
1633 /* and go there */
1634 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_TMP0);
1635 }
1636 s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;
1637 break;
1638
1639 case INDEX_op_call:
1640 if (const_args[0]) {
1641 tgen_calli(s, args[0]);
1642 } else {
1643 tcg_out_insn(s, RR, BASR, TCG_REG_R14, args[0]);
1644 }
1645 break;
1646
1647 case INDEX_op_mov_i32:
1648 tcg_out_mov(s, TCG_TYPE_I32, args[0], args[1]);
1649 break;
1650 case INDEX_op_movi_i32:
1651 tcg_out_movi(s, TCG_TYPE_I32, args[0], args[1]);
1652 break;
1653
1654 OP_32_64(ld8u):
1655 /* ??? LLC (RXY format) is only present with the extended-immediate
1656 facility, whereas LLGC is always present. */
1657 tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
1658 break;
1659
1660 OP_32_64(ld8s):
1661 /* ??? LB is no smaller than LGB, so no point to using it. */
1662 tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
1663 break;
1664
1665 OP_32_64(ld16u):
1666 /* ??? LLH (RXY format) is only present with the extended-immediate
1667 facility, whereas LLGH is always present. */
1668 tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
1669 break;
1670
1671 case INDEX_op_ld16s_i32:
1672 tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
1673 break;
1674
1675 case INDEX_op_ld_i32:
1676 tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1677 break;
1678
1679 OP_32_64(st8):
1680 tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
1681 TCG_REG_NONE, args[2]);
1682 break;
1683
1684 OP_32_64(st16):
1685 tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
1686 TCG_REG_NONE, args[2]);
1687 break;
1688
1689 case INDEX_op_st_i32:
1690 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1691 break;
1692
1693 case INDEX_op_add_i32:
1694 if (const_args[2]) {
1695 tgen32_addi(s, args[0], args[2]);
1696 } else {
1697 tcg_out_insn(s, RR, AR, args[0], args[2]);
1698 }
1699 break;
1700 case INDEX_op_sub_i32:
1701 if (const_args[2]) {
1702 tgen32_addi(s, args[0], -args[2]);
1703 } else {
1704 tcg_out_insn(s, RR, SR, args[0], args[2]);
1705 }
1706 break;
1707
1708 case INDEX_op_and_i32:
1709 if (const_args[2]) {
1710 tgen64_andi(s, args[0], args[2] | 0xffffffff00000000ull);
1711 } else {
1712 tcg_out_insn(s, RR, NR, args[0], args[2]);
1713 }
1714 break;
1715 case INDEX_op_or_i32:
1716 if (const_args[2]) {
1717 tgen64_ori(s, args[0], args[2] & 0xffffffff);
1718 } else {
1719 tcg_out_insn(s, RR, OR, args[0], args[2]);
1720 }
1721 break;
1722 case INDEX_op_xor_i32:
1723 if (const_args[2]) {
1724 tgen64_xori(s, args[0], args[2] & 0xffffffff);
1725 } else {
1726 tcg_out_insn(s, RR, XR, args[0], args[2]);
1727 }
1728 break;
1729
1730 case INDEX_op_neg_i32:
1731 tcg_out_insn(s, RR, LCR, args[0], args[1]);
1732 break;
1733
1734 case INDEX_op_mul_i32:
1735 if (const_args[2]) {
1736 if ((int32_t)args[2] == (int16_t)args[2]) {
1737 tcg_out_insn(s, RI, MHI, args[0], args[2]);
1738 } else {
1739 tcg_out_insn(s, RIL, MSFI, args[0], args[2]);
1740 }
1741 } else {
1742 tcg_out_insn(s, RRE, MSR, args[0], args[2]);
1743 }
1744 break;
1745
1746 case INDEX_op_div2_i32:
1747 tcg_out_insn(s, RR, DR, TCG_REG_R2, args[4]);
1748 break;
1749 case INDEX_op_divu2_i32:
1750 tcg_out_insn(s, RRE, DLR, TCG_REG_R2, args[4]);
1751 break;
1752
1753 case INDEX_op_shl_i32:
1754 op = RS_SLL;
1755 do_shift32:
1756 if (const_args[2]) {
1757 tcg_out_sh32(s, op, args[0], TCG_REG_NONE, args[2]);
1758 } else {
1759 tcg_out_sh32(s, op, args[0], args[2], 0);
1760 }
1761 break;
1762 case INDEX_op_shr_i32:
1763 op = RS_SRL;
1764 goto do_shift32;
1765 case INDEX_op_sar_i32:
1766 op = RS_SRA;
1767 goto do_shift32;
1768
1769 case INDEX_op_rotl_i32:
1770 /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol. */
1771 if (const_args[2]) {
1772 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
1773 } else {
1774 tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
1775 }
1776 break;
1777 case INDEX_op_rotr_i32:
1778 if (const_args[2]) {
1779 tcg_out_sh64(s, RSY_RLL, args[0], args[1],
1780 TCG_REG_NONE, (32 - args[2]) & 31);
1781 } else {
1782 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
1783 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
1784 }
1785 break;
1786
1787 case INDEX_op_ext8s_i32:
1788 tgen_ext8s(s, TCG_TYPE_I32, args[0], args[1]);
1789 break;
1790 case INDEX_op_ext16s_i32:
1791 tgen_ext16s(s, TCG_TYPE_I32, args[0], args[1]);
1792 break;
1793 case INDEX_op_ext8u_i32:
1794 tgen_ext8u(s, TCG_TYPE_I32, args[0], args[1]);
1795 break;
1796 case INDEX_op_ext16u_i32:
1797 tgen_ext16u(s, TCG_TYPE_I32, args[0], args[1]);
1798 break;
1799
1800 OP_32_64(bswap16):
1801 /* The TCG bswap definition requires bits 0-47 already be zero.
1802 Thus we don't need the G-type insns to implement bswap16_i64. */
1803 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
1804 tcg_out_sh32(s, RS_SRL, args[0], TCG_REG_NONE, 16);
1805 break;
1806 OP_32_64(bswap32):
1807 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
1808 break;
1809
1810 case INDEX_op_br:
1811 tgen_branch(s, S390_CC_ALWAYS, args[0]);
1812 break;
1813
1814 case INDEX_op_brcond_i32:
1815 tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
1816 args[1], const_args[1], args[3]);
1817 break;
1818 case INDEX_op_setcond_i32:
1819 tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
1820 args[2], const_args[2]);
1821 break;
1822
1823 case INDEX_op_qemu_ld8u:
1824 tcg_out_qemu_ld(s, args, LD_UINT8);
1825 break;
1826 case INDEX_op_qemu_ld8s:
1827 tcg_out_qemu_ld(s, args, LD_INT8);
1828 break;
1829 case INDEX_op_qemu_ld16u:
1830 tcg_out_qemu_ld(s, args, LD_UINT16);
1831 break;
1832 case INDEX_op_qemu_ld16s:
1833 tcg_out_qemu_ld(s, args, LD_INT16);
1834 break;
1835 case INDEX_op_qemu_ld32:
1836 /* ??? Technically we can use a non-extending instruction. */
1837 tcg_out_qemu_ld(s, args, LD_UINT32);
1838 break;
1839 case INDEX_op_qemu_ld64:
1840 tcg_out_qemu_ld(s, args, LD_UINT64);
1841 break;
1842
1843 case INDEX_op_qemu_st8:
1844 tcg_out_qemu_st(s, args, LD_UINT8);
1845 break;
1846 case INDEX_op_qemu_st16:
1847 tcg_out_qemu_st(s, args, LD_UINT16);
1848 break;
1849 case INDEX_op_qemu_st32:
1850 tcg_out_qemu_st(s, args, LD_UINT32);
1851 break;
1852 case INDEX_op_qemu_st64:
1853 tcg_out_qemu_st(s, args, LD_UINT64);
1854 break;
1855
1856 #if TCG_TARGET_REG_BITS == 64
1857 case INDEX_op_mov_i64:
1858 tcg_out_mov(s, TCG_TYPE_I64, args[0], args[1]);
1859 break;
1860 case INDEX_op_movi_i64:
1861 tcg_out_movi(s, TCG_TYPE_I64, args[0], args[1]);
1862 break;
1863
1864 case INDEX_op_ld16s_i64:
1865 tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
1866 break;
1867 case INDEX_op_ld32u_i64:
1868 tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
1869 break;
1870 case INDEX_op_ld32s_i64:
1871 tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
1872 break;
1873 case INDEX_op_ld_i64:
1874 tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
1875 break;
1876
1877 case INDEX_op_st32_i64:
1878 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1879 break;
1880 case INDEX_op_st_i64:
1881 tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
1882 break;
1883
1884 case INDEX_op_add_i64:
1885 if (const_args[2]) {
1886 tgen64_addi(s, args[0], args[2]);
1887 } else {
1888 tcg_out_insn(s, RRE, AGR, args[0], args[2]);
1889 }
1890 break;
1891 case INDEX_op_sub_i64:
1892 if (const_args[2]) {
1893 tgen64_addi(s, args[0], -args[2]);
1894 } else {
1895 tcg_out_insn(s, RRE, SGR, args[0], args[2]);
1896 }
1897 break;
1898
1899 case INDEX_op_and_i64:
1900 if (const_args[2]) {
1901 tgen64_andi(s, args[0], args[2]);
1902 } else {
1903 tcg_out_insn(s, RRE, NGR, args[0], args[2]);
1904 }
1905 break;
1906 case INDEX_op_or_i64:
1907 if (const_args[2]) {
1908 tgen64_ori(s, args[0], args[2]);
1909 } else {
1910 tcg_out_insn(s, RRE, OGR, args[0], args[2]);
1911 }
1912 break;
1913 case INDEX_op_xor_i64:
1914 if (const_args[2]) {
1915 tgen64_xori(s, args[0], args[2]);
1916 } else {
1917 tcg_out_insn(s, RRE, XGR, args[0], args[2]);
1918 }
1919 break;
1920
1921 case INDEX_op_neg_i64:
1922 tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
1923 break;
1924 case INDEX_op_bswap64_i64:
1925 tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
1926 break;
1927
1928 case INDEX_op_mul_i64:
1929 if (const_args[2]) {
1930 if (args[2] == (int16_t)args[2]) {
1931 tcg_out_insn(s, RI, MGHI, args[0], args[2]);
1932 } else {
1933 tcg_out_insn(s, RIL, MSGFI, args[0], args[2]);
1934 }
1935 } else {
1936 tcg_out_insn(s, RRE, MSGR, args[0], args[2]);
1937 }
1938 break;
1939
1940 case INDEX_op_div2_i64:
1941 /* ??? We get an unnecessary sign-extension of the dividend
1942 into R3 with this definition, but as we do in fact always
1943 produce both quotient and remainder using INDEX_op_div_i64
1944 instead requires jumping through even more hoops. */
1945 tcg_out_insn(s, RRE, DSGR, TCG_REG_R2, args[4]);
1946 break;
1947 case INDEX_op_divu2_i64:
1948 tcg_out_insn(s, RRE, DLGR, TCG_REG_R2, args[4]);
1949 break;
1950
1951 case INDEX_op_shl_i64:
1952 op = RSY_SLLG;
1953 do_shift64:
1954 if (const_args[2]) {
1955 tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
1956 } else {
1957 tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
1958 }
1959 break;
1960 case INDEX_op_shr_i64:
1961 op = RSY_SRLG;
1962 goto do_shift64;
1963 case INDEX_op_sar_i64:
1964 op = RSY_SRAG;
1965 goto do_shift64;
1966
1967 case INDEX_op_rotl_i64:
1968 if (const_args[2]) {
1969 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
1970 TCG_REG_NONE, args[2]);
1971 } else {
1972 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
1973 }
1974 break;
1975 case INDEX_op_rotr_i64:
1976 if (const_args[2]) {
1977 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
1978 TCG_REG_NONE, (64 - args[2]) & 63);
1979 } else {
1980 /* We can use the smaller 32-bit negate because only the
1981 low 6 bits are examined for the rotate. */
1982 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
1983 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
1984 }
1985 break;
1986
1987 case INDEX_op_ext8s_i64:
1988 tgen_ext8s(s, TCG_TYPE_I64, args[0], args[1]);
1989 break;
1990 case INDEX_op_ext16s_i64:
1991 tgen_ext16s(s, TCG_TYPE_I64, args[0], args[1]);
1992 break;
1993 case INDEX_op_ext32s_i64:
1994 tgen_ext32s(s, args[0], args[1]);
1995 break;
1996 case INDEX_op_ext8u_i64:
1997 tgen_ext8u(s, TCG_TYPE_I64, args[0], args[1]);
1998 break;
1999 case INDEX_op_ext16u_i64:
2000 tgen_ext16u(s, TCG_TYPE_I64, args[0], args[1]);
2001 break;
2002 case INDEX_op_ext32u_i64:
2003 tgen_ext32u(s, args[0], args[1]);
2004 break;
2005
2006 case INDEX_op_brcond_i64:
2007 tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2008 args[1], const_args[1], args[3]);
2009 break;
2010 case INDEX_op_setcond_i64:
2011 tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2012 args[2], const_args[2]);
2013 break;
2014
2015 case INDEX_op_qemu_ld32u:
2016 tcg_out_qemu_ld(s, args, LD_UINT32);
2017 break;
2018 case INDEX_op_qemu_ld32s:
2019 tcg_out_qemu_ld(s, args, LD_INT32);
2020 break;
2021 #endif /* TCG_TARGET_REG_BITS == 64 */
2022
2023 case INDEX_op_jmp:
2024 /* This one is obsolete and never emitted. */
2025 tcg_abort();
2026 break;
2027
2028 default:
2029 fprintf(stderr,"unimplemented opc 0x%x\n",opc);
2030 tcg_abort();
2031 }
2032 }
2033
2034 static const TCGTargetOpDef s390_op_defs[] = {
2035 { INDEX_op_exit_tb, { } },
2036 { INDEX_op_goto_tb, { } },
2037 { INDEX_op_call, { "ri" } },
2038 { INDEX_op_jmp, { "ri" } },
2039 { INDEX_op_br, { } },
2040
2041 { INDEX_op_mov_i32, { "r", "r" } },
2042 { INDEX_op_movi_i32, { "r" } },
2043
2044 { INDEX_op_ld8u_i32, { "r", "r" } },
2045 { INDEX_op_ld8s_i32, { "r", "r" } },
2046 { INDEX_op_ld16u_i32, { "r", "r" } },
2047 { INDEX_op_ld16s_i32, { "r", "r" } },
2048 { INDEX_op_ld_i32, { "r", "r" } },
2049 { INDEX_op_st8_i32, { "r", "r" } },
2050 { INDEX_op_st16_i32, { "r", "r" } },
2051 { INDEX_op_st_i32, { "r", "r" } },
2052
2053 { INDEX_op_add_i32, { "r", "0", "rWI" } },
2054 { INDEX_op_sub_i32, { "r", "0", "rWNI" } },
2055 { INDEX_op_mul_i32, { "r", "0", "rK" } },
2056
2057 { INDEX_op_div2_i32, { "b", "a", "0", "1", "r" } },
2058 { INDEX_op_divu2_i32, { "b", "a", "0", "1", "r" } },
2059
2060 { INDEX_op_and_i32, { "r", "0", "rWA" } },
2061 { INDEX_op_or_i32, { "r", "0", "rWO" } },
2062 { INDEX_op_xor_i32, { "r", "0", "rWX" } },
2063
2064 { INDEX_op_neg_i32, { "r", "r" } },
2065
2066 { INDEX_op_shl_i32, { "r", "0", "Ri" } },
2067 { INDEX_op_shr_i32, { "r", "0", "Ri" } },
2068 { INDEX_op_sar_i32, { "r", "0", "Ri" } },
2069
2070 { INDEX_op_rotl_i32, { "r", "r", "Ri" } },
2071 { INDEX_op_rotr_i32, { "r", "r", "Ri" } },
2072
2073 { INDEX_op_ext8s_i32, { "r", "r" } },
2074 { INDEX_op_ext8u_i32, { "r", "r" } },
2075 { INDEX_op_ext16s_i32, { "r", "r" } },
2076 { INDEX_op_ext16u_i32, { "r", "r" } },
2077
2078 { INDEX_op_bswap16_i32, { "r", "r" } },
2079 { INDEX_op_bswap32_i32, { "r", "r" } },
2080
2081 { INDEX_op_brcond_i32, { "r", "rWC" } },
2082 { INDEX_op_setcond_i32, { "r", "r", "rWC" } },
2083
2084 { INDEX_op_qemu_ld8u, { "r", "L" } },
2085 { INDEX_op_qemu_ld8s, { "r", "L" } },
2086 { INDEX_op_qemu_ld16u, { "r", "L" } },
2087 { INDEX_op_qemu_ld16s, { "r", "L" } },
2088 { INDEX_op_qemu_ld32, { "r", "L" } },
2089 { INDEX_op_qemu_ld64, { "r", "L" } },
2090
2091 { INDEX_op_qemu_st8, { "L", "L" } },
2092 { INDEX_op_qemu_st16, { "L", "L" } },
2093 { INDEX_op_qemu_st32, { "L", "L" } },
2094 { INDEX_op_qemu_st64, { "L", "L" } },
2095
2096 #if defined(__s390x__)
2097 { INDEX_op_mov_i64, { "r", "r" } },
2098 { INDEX_op_movi_i64, { "r" } },
2099
2100 { INDEX_op_ld8u_i64, { "r", "r" } },
2101 { INDEX_op_ld8s_i64, { "r", "r" } },
2102 { INDEX_op_ld16u_i64, { "r", "r" } },
2103 { INDEX_op_ld16s_i64, { "r", "r" } },
2104 { INDEX_op_ld32u_i64, { "r", "r" } },
2105 { INDEX_op_ld32s_i64, { "r", "r" } },
2106 { INDEX_op_ld_i64, { "r", "r" } },
2107
2108 { INDEX_op_st8_i64, { "r", "r" } },
2109 { INDEX_op_st16_i64, { "r", "r" } },
2110 { INDEX_op_st32_i64, { "r", "r" } },
2111 { INDEX_op_st_i64, { "r", "r" } },
2112
2113 { INDEX_op_add_i64, { "r", "0", "rI" } },
2114 { INDEX_op_sub_i64, { "r", "0", "rNI" } },
2115 { INDEX_op_mul_i64, { "r", "0", "rK" } },
2116
2117 { INDEX_op_div2_i64, { "b", "a", "0", "1", "r" } },
2118 { INDEX_op_divu2_i64, { "b", "a", "0", "1", "r" } },
2119
2120 { INDEX_op_and_i64, { "r", "0", "rA" } },
2121 { INDEX_op_or_i64, { "r", "0", "rO" } },
2122 { INDEX_op_xor_i64, { "r", "0", "rX" } },
2123
2124 { INDEX_op_neg_i64, { "r", "r" } },
2125
2126 { INDEX_op_shl_i64, { "r", "r", "Ri" } },
2127 { INDEX_op_shr_i64, { "r", "r", "Ri" } },
2128 { INDEX_op_sar_i64, { "r", "r", "Ri" } },
2129
2130 { INDEX_op_rotl_i64, { "r", "r", "Ri" } },
2131 { INDEX_op_rotr_i64, { "r", "r", "Ri" } },
2132
2133 { INDEX_op_ext8s_i64, { "r", "r" } },
2134 { INDEX_op_ext8u_i64, { "r", "r" } },
2135 { INDEX_op_ext16s_i64, { "r", "r" } },
2136 { INDEX_op_ext16u_i64, { "r", "r" } },
2137 { INDEX_op_ext32s_i64, { "r", "r" } },
2138 { INDEX_op_ext32u_i64, { "r", "r" } },
2139
2140 { INDEX_op_bswap16_i64, { "r", "r" } },
2141 { INDEX_op_bswap32_i64, { "r", "r" } },
2142 { INDEX_op_bswap64_i64, { "r", "r" } },
2143
2144 { INDEX_op_brcond_i64, { "r", "rC" } },
2145 { INDEX_op_setcond_i64, { "r", "r", "rC" } },
2146
2147 { INDEX_op_qemu_ld32u, { "r", "L" } },
2148 { INDEX_op_qemu_ld32s, { "r", "L" } },
2149 #endif
2150
2151 { -1 },
2152 };
2153
2154 /* ??? Linux kernels provide an AUXV entry AT_HWCAP that provides most of
2155 this information. However, getting at that entry is not easy this far
2156 away from main. Our options are: start searching from environ, but
2157 that fails as soon as someone does a setenv in between. Read the data
2158 from /proc/self/auxv. Or do the probing ourselves. The only thing
2159 extra that AT_HWCAP gives us is HWCAP_S390_HIGH_GPRS, which indicates
2160 that the kernel saves all 64-bits of the registers around traps while
2161 in 31-bit mode. But this is true of all "recent" kernels (ought to dig
2162 back and see from when this might not be true). */
2163
2164 #include <signal.h>
2165
2166 static volatile sig_atomic_t got_sigill;
2167
2168 static void sigill_handler(int sig)
2169 {
2170 got_sigill = 1;
2171 }
2172
2173 static void query_facilities(void)
2174 {
2175 struct sigaction sa_old, sa_new;
2176 register int r0 __asm__("0");
2177 register void *r1 __asm__("1");
2178 int fail;
2179
2180 memset(&sa_new, 0, sizeof(sa_new));
2181 sa_new.sa_handler = sigill_handler;
2182 sigaction(SIGILL, &sa_new, &sa_old);
2183
2184 /* First, try STORE FACILITY LIST EXTENDED. If this is present, then
2185 we need not do any more probing. Unfortunately, this itself is an
2186 extension and the original STORE FACILITY LIST instruction is
2187 kernel-only, storing its results at absolute address 200. */
2188 /* stfle 0(%r1) */
2189 r1 = &facilities;
2190 asm volatile(".word 0xb2b0,0x1000"
2191 : "=r"(r0) : "0"(0), "r"(r1) : "memory", "cc");
2192
2193 if (got_sigill) {
2194 /* STORE FACILITY EXTENDED is not available. Probe for one of each
2195 kind of instruction that we're interested in. */
2196 /* ??? Possibly some of these are in practice never present unless
2197 the store-facility-extended facility is also present. But since
2198 that isn't documented it's just better to probe for each. */
2199
2200 /* Test for z/Architecture. Required even in 31-bit mode. */
2201 got_sigill = 0;
2202 /* agr %r0,%r0 */
2203 asm volatile(".word 0xb908,0x0000" : "=r"(r0) : : "cc");
2204 if (!got_sigill) {
2205 facilities |= FACILITY_ZARCH_ACTIVE;
2206 }
2207
2208 /* Test for long displacement. */
2209 got_sigill = 0;
2210 /* ly %r0,0(%r1) */
2211 r1 = &facilities;
2212 asm volatile(".word 0xe300,0x1000,0x0058"
2213 : "=r"(r0) : "r"(r1) : "cc");
2214 if (!got_sigill) {
2215 facilities |= FACILITY_LONG_DISP;
2216 }
2217
2218 /* Test for extended immediates. */
2219 got_sigill = 0;
2220 /* afi %r0,0 */
2221 asm volatile(".word 0xc209,0x0000,0x0000" : : : "cc");
2222 if (!got_sigill) {
2223 facilities |= FACILITY_EXT_IMM;
2224 }
2225
2226 /* Test for general-instructions-extension. */
2227 got_sigill = 0;
2228 /* msfi %r0,1 */
2229 asm volatile(".word 0xc201,0x0000,0x0001");
2230 if (!got_sigill) {
2231 facilities |= FACILITY_GEN_INST_EXT;
2232 }
2233 }
2234
2235 sigaction(SIGILL, &sa_old, NULL);
2236
2237 /* The translator currently uses these extensions unconditionally.
2238 Pruning this back to the base ESA/390 architecture doesn't seem
2239 worthwhile, since even the KVM target requires z/Arch. */
2240 fail = 0;
2241 if ((facilities & FACILITY_ZARCH_ACTIVE) == 0) {
2242 fprintf(stderr, "TCG: z/Arch facility is required.\n");
2243 fprintf(stderr, "TCG: Boot with a 64-bit enabled kernel.\n");
2244 fail = 1;
2245 }
2246 if ((facilities & FACILITY_LONG_DISP) == 0) {
2247 fprintf(stderr, "TCG: long-displacement facility is required.\n");
2248 fail = 1;
2249 }
2250
2251 /* So far there's just enough support for 31-bit mode to let the
2252 compile succeed. This is good enough to run QEMU with KVM. */
2253 if (sizeof(void *) != 8) {
2254 fprintf(stderr, "TCG: 31-bit mode is not supported.\n");
2255 fail = 1;
2256 }
2257
2258 if (fail) {
2259 exit(-1);
2260 }
2261 }
2262
2263 static void tcg_target_init(TCGContext *s)
2264 {
2265 #if !defined(CONFIG_USER_ONLY)
2266 /* fail safe */
2267 if ((1 << CPU_TLB_ENTRY_BITS) != sizeof(CPUTLBEntry)) {
2268 tcg_abort();
2269 }
2270 #endif
2271
2272 query_facilities();
2273
2274 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffff);
2275 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, 0xffff);
2276
2277 tcg_regset_clear(tcg_target_call_clobber_regs);
2278 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2279 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2280 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2281 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2282 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
2283 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
2284 /* The return register can be considered call-clobbered. */
2285 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2286
2287 tcg_regset_clear(s->reserved_regs);
2288 tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
2289 /* XXX many insns can't be used with R0, so we better avoid it for now */
2290 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
2291 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2292
2293 tcg_add_target_add_op_defs(s390_op_defs);
2294 tcg_set_frame(s, TCG_AREG0, offsetof(CPUState, temp_buf),
2295 CPU_TEMP_BUF_NLONGS * sizeof(long));
2296 }
2297
2298 static void tcg_target_qemu_prologue(TCGContext *s)
2299 {
2300 /* stmg %r6,%r15,48(%r15) (save registers) */
2301 tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
2302
2303 /* aghi %r15,-160 (stack frame) */
2304 tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -160);
2305
2306 if (GUEST_BASE >= 0x80000) {
2307 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, GUEST_BASE);
2308 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2309 }
2310
2311 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2312 /* br %r3 (go to TB) */
2313 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
2314
2315 tb_ret_addr = s->code_ptr;
2316
2317 /* lmg %r6,%r15,208(%r15) (restore registers) */
2318 tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 208);
2319
2320 /* br %r14 (return) */
2321 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
2322 }
2323
2324 static inline void tcg_out_addi(TCGContext *s, int reg, tcg_target_long val)
2325 {
2326 tcg_abort();
2327 }
Anthony Liguori's QEMU tree