memory: Define API for MemoryRegionOps to take attrs and return status
[qemu.git] / tcg / s390 / tcg-target.c
blobbe984f5780a2c4d3df01db15f867d962e5be5ec2
1 /*
2 * Tiny Code Generator for QEMU
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>
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:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
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.
27 #include "tcg-be-ldst.h"
29 /* We only support generating code for 64-bit mode. */
30 #if TCG_TARGET_REG_BITS != 64
31 #error "unsupported code generation mode"
32 #endif
34 #include "elf.h"
36 /* ??? The translation blocks produced by TCG are generally small enough to
37 be entirely reachable with a 16-bit displacement. Leaving the option for
38 a 32-bit displacement here Just In Case. */
39 #define USE_LONG_BRANCHES 0
41 #define TCG_CT_CONST_MULI 0x100
42 #define TCG_CT_CONST_ORI 0x200
43 #define TCG_CT_CONST_XORI 0x400
44 #define TCG_CT_CONST_CMPI 0x800
45 #define TCG_CT_CONST_ADLI 0x1000
47 /* Several places within the instruction set 0 means "no register"
48 rather than TCG_REG_R0. */
49 #define TCG_REG_NONE 0
51 /* A scratch register that may be be used throughout the backend. */
52 #define TCG_TMP0 TCG_REG_R14
54 #ifdef CONFIG_USE_GUEST_BASE
55 #define TCG_GUEST_BASE_REG TCG_REG_R13
56 #else
57 #define TCG_GUEST_BASE_REG TCG_REG_R0
58 #endif
60 #ifndef GUEST_BASE
61 #define GUEST_BASE 0
62 #endif
65 /* All of the following instructions are prefixed with their instruction
66 format, and are defined as 8- or 16-bit quantities, even when the two
67 halves of the 16-bit quantity may appear 32 bits apart in the insn.
68 This makes it easy to copy the values from the tables in Appendix B. */
69 typedef enum S390Opcode {
70 RIL_AFI = 0xc209,
71 RIL_AGFI = 0xc208,
72 RIL_ALFI = 0xc20b,
73 RIL_ALGFI = 0xc20a,
74 RIL_BRASL = 0xc005,
75 RIL_BRCL = 0xc004,
76 RIL_CFI = 0xc20d,
77 RIL_CGFI = 0xc20c,
78 RIL_CLFI = 0xc20f,
79 RIL_CLGFI = 0xc20e,
80 RIL_IIHF = 0xc008,
81 RIL_IILF = 0xc009,
82 RIL_LARL = 0xc000,
83 RIL_LGFI = 0xc001,
84 RIL_LGRL = 0xc408,
85 RIL_LLIHF = 0xc00e,
86 RIL_LLILF = 0xc00f,
87 RIL_LRL = 0xc40d,
88 RIL_MSFI = 0xc201,
89 RIL_MSGFI = 0xc200,
90 RIL_NIHF = 0xc00a,
91 RIL_NILF = 0xc00b,
92 RIL_OIHF = 0xc00c,
93 RIL_OILF = 0xc00d,
94 RIL_SLFI = 0xc205,
95 RIL_SLGFI = 0xc204,
96 RIL_XIHF = 0xc006,
97 RIL_XILF = 0xc007,
99 RI_AGHI = 0xa70b,
100 RI_AHI = 0xa70a,
101 RI_BRC = 0xa704,
102 RI_IIHH = 0xa500,
103 RI_IIHL = 0xa501,
104 RI_IILH = 0xa502,
105 RI_IILL = 0xa503,
106 RI_LGHI = 0xa709,
107 RI_LLIHH = 0xa50c,
108 RI_LLIHL = 0xa50d,
109 RI_LLILH = 0xa50e,
110 RI_LLILL = 0xa50f,
111 RI_MGHI = 0xa70d,
112 RI_MHI = 0xa70c,
113 RI_NIHH = 0xa504,
114 RI_NIHL = 0xa505,
115 RI_NILH = 0xa506,
116 RI_NILL = 0xa507,
117 RI_OIHH = 0xa508,
118 RI_OIHL = 0xa509,
119 RI_OILH = 0xa50a,
120 RI_OILL = 0xa50b,
122 RIE_CGIJ = 0xec7c,
123 RIE_CGRJ = 0xec64,
124 RIE_CIJ = 0xec7e,
125 RIE_CLGRJ = 0xec65,
126 RIE_CLIJ = 0xec7f,
127 RIE_CLGIJ = 0xec7d,
128 RIE_CLRJ = 0xec77,
129 RIE_CRJ = 0xec76,
130 RIE_RISBG = 0xec55,
132 RRE_AGR = 0xb908,
133 RRE_ALGR = 0xb90a,
134 RRE_ALCR = 0xb998,
135 RRE_ALCGR = 0xb988,
136 RRE_CGR = 0xb920,
137 RRE_CLGR = 0xb921,
138 RRE_DLGR = 0xb987,
139 RRE_DLR = 0xb997,
140 RRE_DSGFR = 0xb91d,
141 RRE_DSGR = 0xb90d,
142 RRE_LGBR = 0xb906,
143 RRE_LCGR = 0xb903,
144 RRE_LGFR = 0xb914,
145 RRE_LGHR = 0xb907,
146 RRE_LGR = 0xb904,
147 RRE_LLGCR = 0xb984,
148 RRE_LLGFR = 0xb916,
149 RRE_LLGHR = 0xb985,
150 RRE_LRVR = 0xb91f,
151 RRE_LRVGR = 0xb90f,
152 RRE_LTGR = 0xb902,
153 RRE_MLGR = 0xb986,
154 RRE_MSGR = 0xb90c,
155 RRE_MSR = 0xb252,
156 RRE_NGR = 0xb980,
157 RRE_OGR = 0xb981,
158 RRE_SGR = 0xb909,
159 RRE_SLGR = 0xb90b,
160 RRE_SLBR = 0xb999,
161 RRE_SLBGR = 0xb989,
162 RRE_XGR = 0xb982,
164 RRF_LOCR = 0xb9f2,
165 RRF_LOCGR = 0xb9e2,
167 RR_AR = 0x1a,
168 RR_ALR = 0x1e,
169 RR_BASR = 0x0d,
170 RR_BCR = 0x07,
171 RR_CLR = 0x15,
172 RR_CR = 0x19,
173 RR_DR = 0x1d,
174 RR_LCR = 0x13,
175 RR_LR = 0x18,
176 RR_LTR = 0x12,
177 RR_NR = 0x14,
178 RR_OR = 0x16,
179 RR_SR = 0x1b,
180 RR_SLR = 0x1f,
181 RR_XR = 0x17,
183 RSY_RLL = 0xeb1d,
184 RSY_RLLG = 0xeb1c,
185 RSY_SLLG = 0xeb0d,
186 RSY_SRAG = 0xeb0a,
187 RSY_SRLG = 0xeb0c,
189 RS_SLL = 0x89,
190 RS_SRA = 0x8a,
191 RS_SRL = 0x88,
193 RXY_AG = 0xe308,
194 RXY_AY = 0xe35a,
195 RXY_CG = 0xe320,
196 RXY_CY = 0xe359,
197 RXY_LAY = 0xe371,
198 RXY_LB = 0xe376,
199 RXY_LG = 0xe304,
200 RXY_LGB = 0xe377,
201 RXY_LGF = 0xe314,
202 RXY_LGH = 0xe315,
203 RXY_LHY = 0xe378,
204 RXY_LLGC = 0xe390,
205 RXY_LLGF = 0xe316,
206 RXY_LLGH = 0xe391,
207 RXY_LMG = 0xeb04,
208 RXY_LRV = 0xe31e,
209 RXY_LRVG = 0xe30f,
210 RXY_LRVH = 0xe31f,
211 RXY_LY = 0xe358,
212 RXY_STCY = 0xe372,
213 RXY_STG = 0xe324,
214 RXY_STHY = 0xe370,
215 RXY_STMG = 0xeb24,
216 RXY_STRV = 0xe33e,
217 RXY_STRVG = 0xe32f,
218 RXY_STRVH = 0xe33f,
219 RXY_STY = 0xe350,
221 RX_A = 0x5a,
222 RX_C = 0x59,
223 RX_L = 0x58,
224 RX_LA = 0x41,
225 RX_LH = 0x48,
226 RX_ST = 0x50,
227 RX_STC = 0x42,
228 RX_STH = 0x40,
229 } S390Opcode;
231 #ifndef NDEBUG
232 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
233 "%r0", "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7",
234 "%r8", "%r9", "%r10" "%r11" "%r12" "%r13" "%r14" "%r15"
236 #endif
238 /* Since R6 is a potential argument register, choose it last of the
239 call-saved registers. Likewise prefer the call-clobbered registers
240 in reverse order to maximize the chance of avoiding the arguments. */
241 static const int tcg_target_reg_alloc_order[] = {
242 /* Call saved registers. */
243 TCG_REG_R13,
244 TCG_REG_R12,
245 TCG_REG_R11,
246 TCG_REG_R10,
247 TCG_REG_R9,
248 TCG_REG_R8,
249 TCG_REG_R7,
250 TCG_REG_R6,
251 /* Call clobbered registers. */
252 TCG_REG_R14,
253 TCG_REG_R0,
254 TCG_REG_R1,
255 /* Argument registers, in reverse order of allocation. */
256 TCG_REG_R5,
257 TCG_REG_R4,
258 TCG_REG_R3,
259 TCG_REG_R2,
262 static const int tcg_target_call_iarg_regs[] = {
263 TCG_REG_R2,
264 TCG_REG_R3,
265 TCG_REG_R4,
266 TCG_REG_R5,
267 TCG_REG_R6,
270 static const int tcg_target_call_oarg_regs[] = {
271 TCG_REG_R2,
274 #define S390_CC_EQ 8
275 #define S390_CC_LT 4
276 #define S390_CC_GT 2
277 #define S390_CC_OV 1
278 #define S390_CC_NE (S390_CC_LT | S390_CC_GT)
279 #define S390_CC_LE (S390_CC_LT | S390_CC_EQ)
280 #define S390_CC_GE (S390_CC_GT | S390_CC_EQ)
281 #define S390_CC_NEVER 0
282 #define S390_CC_ALWAYS 15
284 /* Condition codes that result from a COMPARE and COMPARE LOGICAL. */
285 static const uint8_t tcg_cond_to_s390_cond[] = {
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_LT,
293 [TCG_COND_LEU] = S390_CC_LE,
294 [TCG_COND_GTU] = S390_CC_GT,
295 [TCG_COND_GEU] = S390_CC_GE,
298 /* Condition codes that result from a LOAD AND TEST. Here, we have no
299 unsigned instruction variation, however since the test is vs zero we
300 can re-map the outcomes appropriately. */
301 static const uint8_t tcg_cond_to_ltr_cond[] = {
302 [TCG_COND_EQ] = S390_CC_EQ,
303 [TCG_COND_NE] = S390_CC_NE,
304 [TCG_COND_LT] = S390_CC_LT,
305 [TCG_COND_LE] = S390_CC_LE,
306 [TCG_COND_GT] = S390_CC_GT,
307 [TCG_COND_GE] = S390_CC_GE,
308 [TCG_COND_LTU] = S390_CC_NEVER,
309 [TCG_COND_LEU] = S390_CC_EQ,
310 [TCG_COND_GTU] = S390_CC_NE,
311 [TCG_COND_GEU] = S390_CC_ALWAYS,
314 #ifdef CONFIG_SOFTMMU
315 static void * const qemu_ld_helpers[16] = {
316 [MO_UB] = helper_ret_ldub_mmu,
317 [MO_SB] = helper_ret_ldsb_mmu,
318 [MO_LEUW] = helper_le_lduw_mmu,
319 [MO_LESW] = helper_le_ldsw_mmu,
320 [MO_LEUL] = helper_le_ldul_mmu,
321 [MO_LESL] = helper_le_ldsl_mmu,
322 [MO_LEQ] = helper_le_ldq_mmu,
323 [MO_BEUW] = helper_be_lduw_mmu,
324 [MO_BESW] = helper_be_ldsw_mmu,
325 [MO_BEUL] = helper_be_ldul_mmu,
326 [MO_BESL] = helper_be_ldsl_mmu,
327 [MO_BEQ] = helper_be_ldq_mmu,
330 static void * const qemu_st_helpers[16] = {
331 [MO_UB] = helper_ret_stb_mmu,
332 [MO_LEUW] = helper_le_stw_mmu,
333 [MO_LEUL] = helper_le_stl_mmu,
334 [MO_LEQ] = helper_le_stq_mmu,
335 [MO_BEUW] = helper_be_stw_mmu,
336 [MO_BEUL] = helper_be_stl_mmu,
337 [MO_BEQ] = helper_be_stq_mmu,
339 #endif
341 static tcg_insn_unit *tb_ret_addr;
343 /* A list of relevant facilities used by this translator. Some of these
344 are required for proper operation, and these are checked at startup. */
346 #define FACILITY_ZARCH_ACTIVE (1ULL << (63 - 2))
347 #define FACILITY_LONG_DISP (1ULL << (63 - 18))
348 #define FACILITY_EXT_IMM (1ULL << (63 - 21))
349 #define FACILITY_GEN_INST_EXT (1ULL << (63 - 34))
350 #define FACILITY_LOAD_ON_COND (1ULL << (63 - 45))
352 static uint64_t facilities;
354 static void patch_reloc(tcg_insn_unit *code_ptr, int type,
355 intptr_t value, intptr_t addend)
357 intptr_t pcrel2 = (tcg_insn_unit *)value - (code_ptr - 1);
358 assert(addend == -2);
360 switch (type) {
361 case R_390_PC16DBL:
362 assert(pcrel2 == (int16_t)pcrel2);
363 tcg_patch16(code_ptr, pcrel2);
364 break;
365 case R_390_PC32DBL:
366 assert(pcrel2 == (int32_t)pcrel2);
367 tcg_patch32(code_ptr, pcrel2);
368 break;
369 default:
370 tcg_abort();
371 break;
375 /* parse target specific constraints */
376 static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
378 const char *ct_str = *pct_str;
380 switch (ct_str[0]) {
381 case 'r': /* all registers */
382 ct->ct |= TCG_CT_REG;
383 tcg_regset_set32(ct->u.regs, 0, 0xffff);
384 break;
385 case 'R': /* not R0 */
386 ct->ct |= TCG_CT_REG;
387 tcg_regset_set32(ct->u.regs, 0, 0xffff);
388 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
389 break;
390 case 'L': /* qemu_ld/st constraint */
391 ct->ct |= TCG_CT_REG;
392 tcg_regset_set32(ct->u.regs, 0, 0xffff);
393 tcg_regset_reset_reg (ct->u.regs, TCG_REG_R2);
394 tcg_regset_reset_reg (ct->u.regs, TCG_REG_R3);
395 tcg_regset_reset_reg (ct->u.regs, TCG_REG_R4);
396 break;
397 case 'a': /* force R2 for division */
398 ct->ct |= TCG_CT_REG;
399 tcg_regset_clear(ct->u.regs);
400 tcg_regset_set_reg(ct->u.regs, TCG_REG_R2);
401 break;
402 case 'b': /* force R3 for division */
403 ct->ct |= TCG_CT_REG;
404 tcg_regset_clear(ct->u.regs);
405 tcg_regset_set_reg(ct->u.regs, TCG_REG_R3);
406 break;
407 case 'A':
408 ct->ct |= TCG_CT_CONST_ADLI;
409 break;
410 case 'K':
411 ct->ct |= TCG_CT_CONST_MULI;
412 break;
413 case 'O':
414 ct->ct |= TCG_CT_CONST_ORI;
415 break;
416 case 'X':
417 ct->ct |= TCG_CT_CONST_XORI;
418 break;
419 case 'C':
420 ct->ct |= TCG_CT_CONST_CMPI;
421 break;
422 default:
423 return -1;
425 ct_str++;
426 *pct_str = ct_str;
428 return 0;
431 /* Immediates to be used with logical OR. This is an optimization only,
432 since a full 64-bit immediate OR can always be performed with 4 sequential
433 OI[LH][LH] instructions. What we're looking for is immediates that we
434 can load efficiently, and the immediate load plus the reg-reg OR is
435 smaller than the sequential OI's. */
437 static int tcg_match_ori(TCGType type, tcg_target_long val)
439 if (facilities & FACILITY_EXT_IMM) {
440 if (type == TCG_TYPE_I32) {
441 /* All 32-bit ORs can be performed with 1 48-bit insn. */
442 return 1;
446 /* Look for negative values. These are best to load with LGHI. */
447 if (val < 0) {
448 if (val == (int16_t)val) {
449 return 0;
451 if (facilities & FACILITY_EXT_IMM) {
452 if (val == (int32_t)val) {
453 return 0;
458 return 1;
461 /* Immediates to be used with logical XOR. This is almost, but not quite,
462 only an optimization. XOR with immediate is only supported with the
463 extended-immediate facility. That said, there are a few patterns for
464 which it is better to load the value into a register first. */
466 static int tcg_match_xori(TCGType type, tcg_target_long val)
468 if ((facilities & FACILITY_EXT_IMM) == 0) {
469 return 0;
472 if (type == TCG_TYPE_I32) {
473 /* All 32-bit XORs can be performed with 1 48-bit insn. */
474 return 1;
477 /* Look for negative values. These are best to load with LGHI. */
478 if (val < 0 && val == (int32_t)val) {
479 return 0;
482 return 1;
485 /* Imediates to be used with comparisons. */
487 static int tcg_match_cmpi(TCGType type, tcg_target_long val)
489 if (facilities & FACILITY_EXT_IMM) {
490 /* The COMPARE IMMEDIATE instruction is available. */
491 if (type == TCG_TYPE_I32) {
492 /* We have a 32-bit immediate and can compare against anything. */
493 return 1;
494 } else {
495 /* ??? We have no insight here into whether the comparison is
496 signed or unsigned. The COMPARE IMMEDIATE insn uses a 32-bit
497 signed immediate, and the COMPARE LOGICAL IMMEDIATE insn uses
498 a 32-bit unsigned immediate. If we were to use the (semi)
499 obvious "val == (int32_t)val" we would be enabling unsigned
500 comparisons vs very large numbers. The only solution is to
501 take the intersection of the ranges. */
502 /* ??? Another possible solution is to simply lie and allow all
503 constants here and force the out-of-range values into a temp
504 register in tgen_cmp when we have knowledge of the actual
505 comparison code in use. */
506 return val >= 0 && val <= 0x7fffffff;
508 } else {
509 /* Only the LOAD AND TEST instruction is available. */
510 return val == 0;
514 /* Immediates to be used with add2/sub2. */
516 static int tcg_match_add2i(TCGType type, tcg_target_long val)
518 if (facilities & FACILITY_EXT_IMM) {
519 if (type == TCG_TYPE_I32) {
520 return 1;
521 } else if (val >= -0xffffffffll && val <= 0xffffffffll) {
522 return 1;
525 return 0;
528 /* Test if a constant matches the constraint. */
529 static int tcg_target_const_match(tcg_target_long val, TCGType type,
530 const TCGArgConstraint *arg_ct)
532 int ct = arg_ct->ct;
534 if (ct & TCG_CT_CONST) {
535 return 1;
538 if (type == TCG_TYPE_I32) {
539 val = (int32_t)val;
542 /* The following are mutually exclusive. */
543 if (ct & TCG_CT_CONST_MULI) {
544 /* Immediates that may be used with multiply. If we have the
545 general-instruction-extensions, then we have MULTIPLY SINGLE
546 IMMEDIATE with a signed 32-bit, otherwise we have only
547 MULTIPLY HALFWORD IMMEDIATE, with a signed 16-bit. */
548 if (facilities & FACILITY_GEN_INST_EXT) {
549 return val == (int32_t)val;
550 } else {
551 return val == (int16_t)val;
553 } else if (ct & TCG_CT_CONST_ADLI) {
554 return tcg_match_add2i(type, val);
555 } else if (ct & TCG_CT_CONST_ORI) {
556 return tcg_match_ori(type, val);
557 } else if (ct & TCG_CT_CONST_XORI) {
558 return tcg_match_xori(type, val);
559 } else if (ct & TCG_CT_CONST_CMPI) {
560 return tcg_match_cmpi(type, val);
563 return 0;
566 /* Emit instructions according to the given instruction format. */
568 static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
570 tcg_out16(s, (op << 8) | (r1 << 4) | r2);
573 static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
574 TCGReg r1, TCGReg r2)
576 tcg_out32(s, (op << 16) | (r1 << 4) | r2);
579 static void tcg_out_insn_RRF(TCGContext *s, S390Opcode op,
580 TCGReg r1, TCGReg r2, int m3)
582 tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
585 static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
587 tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
590 static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
592 tcg_out16(s, op | (r1 << 4));
593 tcg_out32(s, i2);
596 static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
597 TCGReg b2, TCGReg r3, int disp)
599 tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
600 | (disp & 0xfff));
603 static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
604 TCGReg b2, TCGReg r3, int disp)
606 tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
607 tcg_out32(s, (op & 0xff) | (b2 << 28)
608 | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
611 #define tcg_out_insn_RX tcg_out_insn_RS
612 #define tcg_out_insn_RXY tcg_out_insn_RSY
614 /* Emit an opcode with "type-checking" of the format. */
615 #define tcg_out_insn(S, FMT, OP, ...) \
616 glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
619 /* emit 64-bit shifts */
620 static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
621 TCGReg src, TCGReg sh_reg, int sh_imm)
623 tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
626 /* emit 32-bit shifts */
627 static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
628 TCGReg sh_reg, int sh_imm)
630 tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
633 static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
635 if (src != dst) {
636 if (type == TCG_TYPE_I32) {
637 tcg_out_insn(s, RR, LR, dst, src);
638 } else {
639 tcg_out_insn(s, RRE, LGR, dst, src);
644 /* load a register with an immediate value */
645 static void tcg_out_movi(TCGContext *s, TCGType type,
646 TCGReg ret, tcg_target_long sval)
648 static const S390Opcode lli_insns[4] = {
649 RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
652 tcg_target_ulong uval = sval;
653 int i;
655 if (type == TCG_TYPE_I32) {
656 uval = (uint32_t)sval;
657 sval = (int32_t)sval;
660 /* Try all 32-bit insns that can load it in one go. */
661 if (sval >= -0x8000 && sval < 0x8000) {
662 tcg_out_insn(s, RI, LGHI, ret, sval);
663 return;
666 for (i = 0; i < 4; i++) {
667 tcg_target_long mask = 0xffffull << i*16;
668 if ((uval & mask) == uval) {
669 tcg_out_insn_RI(s, lli_insns[i], ret, uval >> i*16);
670 return;
674 /* Try all 48-bit insns that can load it in one go. */
675 if (facilities & FACILITY_EXT_IMM) {
676 if (sval == (int32_t)sval) {
677 tcg_out_insn(s, RIL, LGFI, ret, sval);
678 return;
680 if (uval <= 0xffffffff) {
681 tcg_out_insn(s, RIL, LLILF, ret, uval);
682 return;
684 if ((uval & 0xffffffff) == 0) {
685 tcg_out_insn(s, RIL, LLIHF, ret, uval >> 31 >> 1);
686 return;
690 /* Try for PC-relative address load. */
691 if ((sval & 1) == 0) {
692 ptrdiff_t off = tcg_pcrel_diff(s, (void *)sval) >> 1;
693 if (off == (int32_t)off) {
694 tcg_out_insn(s, RIL, LARL, ret, off);
695 return;
699 /* If extended immediates are not present, then we may have to issue
700 several instructions to load the low 32 bits. */
701 if (!(facilities & FACILITY_EXT_IMM)) {
702 /* A 32-bit unsigned value can be loaded in 2 insns. And given
703 that the lli_insns loop above did not succeed, we know that
704 both insns are required. */
705 if (uval <= 0xffffffff) {
706 tcg_out_insn(s, RI, LLILL, ret, uval);
707 tcg_out_insn(s, RI, IILH, ret, uval >> 16);
708 return;
711 /* If all high bits are set, the value can be loaded in 2 or 3 insns.
712 We first want to make sure that all the high bits get set. With
713 luck the low 16-bits can be considered negative to perform that for
714 free, otherwise we load an explicit -1. */
715 if (sval >> 31 >> 1 == -1) {
716 if (uval & 0x8000) {
717 tcg_out_insn(s, RI, LGHI, ret, uval);
718 } else {
719 tcg_out_insn(s, RI, LGHI, ret, -1);
720 tcg_out_insn(s, RI, IILL, ret, uval);
722 tcg_out_insn(s, RI, IILH, ret, uval >> 16);
723 return;
727 /* If we get here, both the high and low parts have non-zero bits. */
729 /* Recurse to load the lower 32-bits. */
730 tcg_out_movi(s, TCG_TYPE_I64, ret, uval & 0xffffffff);
732 /* Insert data into the high 32-bits. */
733 uval = uval >> 31 >> 1;
734 if (facilities & FACILITY_EXT_IMM) {
735 if (uval < 0x10000) {
736 tcg_out_insn(s, RI, IIHL, ret, uval);
737 } else if ((uval & 0xffff) == 0) {
738 tcg_out_insn(s, RI, IIHH, ret, uval >> 16);
739 } else {
740 tcg_out_insn(s, RIL, IIHF, ret, uval);
742 } else {
743 if (uval & 0xffff) {
744 tcg_out_insn(s, RI, IIHL, ret, uval);
746 if (uval & 0xffff0000) {
747 tcg_out_insn(s, RI, IIHH, ret, uval >> 16);
753 /* Emit a load/store type instruction. Inputs are:
754 DATA: The register to be loaded or stored.
755 BASE+OFS: The effective address.
756 OPC_RX: If the operation has an RX format opcode (e.g. STC), otherwise 0.
757 OPC_RXY: The RXY format opcode for the operation (e.g. STCY). */
759 static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
760 TCGReg data, TCGReg base, TCGReg index,
761 tcg_target_long ofs)
763 if (ofs < -0x80000 || ofs >= 0x80000) {
764 /* Combine the low 20 bits of the offset with the actual load insn;
765 the high 44 bits must come from an immediate load. */
766 tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
767 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
768 ofs = low;
770 /* If we were already given an index register, add it in. */
771 if (index != TCG_REG_NONE) {
772 tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
774 index = TCG_TMP0;
777 if (opc_rx && ofs >= 0 && ofs < 0x1000) {
778 tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
779 } else {
780 tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
785 /* load data without address translation or endianness conversion */
786 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
787 TCGReg base, intptr_t ofs)
789 if (type == TCG_TYPE_I32) {
790 tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
791 } else {
792 tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
796 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
797 TCGReg base, intptr_t ofs)
799 if (type == TCG_TYPE_I32) {
800 tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
801 } else {
802 tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
806 /* load data from an absolute host address */
807 static void tcg_out_ld_abs(TCGContext *s, TCGType type, TCGReg dest, void *abs)
809 intptr_t addr = (intptr_t)abs;
811 if ((facilities & FACILITY_GEN_INST_EXT) && !(addr & 1)) {
812 ptrdiff_t disp = tcg_pcrel_diff(s, abs) >> 1;
813 if (disp == (int32_t)disp) {
814 if (type == TCG_TYPE_I32) {
815 tcg_out_insn(s, RIL, LRL, dest, disp);
816 } else {
817 tcg_out_insn(s, RIL, LGRL, dest, disp);
819 return;
823 tcg_out_movi(s, TCG_TYPE_PTR, dest, addr & ~0xffff);
824 tcg_out_ld(s, type, dest, dest, addr & 0xffff);
827 static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
828 int msb, int lsb, int ofs, int z)
830 /* Format RIE-f */
831 tcg_out16(s, (RIE_RISBG & 0xff00) | (dest << 4) | src);
832 tcg_out16(s, (msb << 8) | (z << 7) | lsb);
833 tcg_out16(s, (ofs << 8) | (RIE_RISBG & 0xff));
836 static void tgen_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
838 if (facilities & FACILITY_EXT_IMM) {
839 tcg_out_insn(s, RRE, LGBR, dest, src);
840 return;
843 if (type == TCG_TYPE_I32) {
844 if (dest == src) {
845 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 24);
846 } else {
847 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 24);
849 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 24);
850 } else {
851 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 56);
852 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 56);
856 static void tgen_ext8u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
858 if (facilities & FACILITY_EXT_IMM) {
859 tcg_out_insn(s, RRE, LLGCR, dest, src);
860 return;
863 if (dest == src) {
864 tcg_out_movi(s, type, TCG_TMP0, 0xff);
865 src = TCG_TMP0;
866 } else {
867 tcg_out_movi(s, type, dest, 0xff);
869 if (type == TCG_TYPE_I32) {
870 tcg_out_insn(s, RR, NR, dest, src);
871 } else {
872 tcg_out_insn(s, RRE, NGR, dest, src);
876 static void tgen_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
878 if (facilities & FACILITY_EXT_IMM) {
879 tcg_out_insn(s, RRE, LGHR, dest, src);
880 return;
883 if (type == TCG_TYPE_I32) {
884 if (dest == src) {
885 tcg_out_sh32(s, RS_SLL, dest, TCG_REG_NONE, 16);
886 } else {
887 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 16);
889 tcg_out_sh32(s, RS_SRA, dest, TCG_REG_NONE, 16);
890 } else {
891 tcg_out_sh64(s, RSY_SLLG, dest, src, TCG_REG_NONE, 48);
892 tcg_out_sh64(s, RSY_SRAG, dest, dest, TCG_REG_NONE, 48);
896 static void tgen_ext16u(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
898 if (facilities & FACILITY_EXT_IMM) {
899 tcg_out_insn(s, RRE, LLGHR, dest, src);
900 return;
903 if (dest == src) {
904 tcg_out_movi(s, type, TCG_TMP0, 0xffff);
905 src = TCG_TMP0;
906 } else {
907 tcg_out_movi(s, type, dest, 0xffff);
909 if (type == TCG_TYPE_I32) {
910 tcg_out_insn(s, RR, NR, dest, src);
911 } else {
912 tcg_out_insn(s, RRE, NGR, dest, src);
916 static inline void tgen_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
918 tcg_out_insn(s, RRE, LGFR, dest, src);
921 static inline void tgen_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
923 tcg_out_insn(s, RRE, LLGFR, dest, src);
926 /* Accept bit patterns like these:
927 0....01....1
928 1....10....0
929 1..10..01..1
930 0..01..10..0
931 Copied from gcc sources. */
932 static inline bool risbg_mask(uint64_t c)
934 uint64_t lsb;
935 /* We don't change the number of transitions by inverting,
936 so make sure we start with the LSB zero. */
937 if (c & 1) {
938 c = ~c;
940 /* Reject all zeros or all ones. */
941 if (c == 0) {
942 return false;
944 /* Find the first transition. */
945 lsb = c & -c;
946 /* Invert to look for a second transition. */
947 c = ~c;
948 /* Erase the first transition. */
949 c &= -lsb;
950 /* Find the second transition, if any. */
951 lsb = c & -c;
952 /* Match if all the bits are 1's, or if c is zero. */
953 return c == -lsb;
956 static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
958 int msb, lsb;
959 if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
960 /* Achieve wraparound by swapping msb and lsb. */
961 msb = 64 - ctz64(~val);
962 lsb = clz64(~val) - 1;
963 } else {
964 msb = clz64(val);
965 lsb = 63 - ctz64(val);
967 tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
970 static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
972 static const S390Opcode ni_insns[4] = {
973 RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
975 static const S390Opcode nif_insns[2] = {
976 RIL_NILF, RIL_NIHF
978 uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
979 int i;
981 /* Look for the zero-extensions. */
982 if ((val & valid) == 0xffffffff) {
983 tgen_ext32u(s, dest, dest);
984 return;
986 if (facilities & FACILITY_EXT_IMM) {
987 if ((val & valid) == 0xff) {
988 tgen_ext8u(s, TCG_TYPE_I64, dest, dest);
989 return;
991 if ((val & valid) == 0xffff) {
992 tgen_ext16u(s, TCG_TYPE_I64, dest, dest);
993 return;
997 /* Try all 32-bit insns that can perform it in one go. */
998 for (i = 0; i < 4; i++) {
999 tcg_target_ulong mask = ~(0xffffull << i*16);
1000 if (((val | ~valid) & mask) == mask) {
1001 tcg_out_insn_RI(s, ni_insns[i], dest, val >> i*16);
1002 return;
1006 /* Try all 48-bit insns that can perform it in one go. */
1007 if (facilities & FACILITY_EXT_IMM) {
1008 for (i = 0; i < 2; i++) {
1009 tcg_target_ulong mask = ~(0xffffffffull << i*32);
1010 if (((val | ~valid) & mask) == mask) {
1011 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> i*32);
1012 return;
1016 if ((facilities & FACILITY_GEN_INST_EXT) && risbg_mask(val)) {
1017 tgen_andi_risbg(s, dest, dest, val);
1018 return;
1021 /* Fall back to loading the constant. */
1022 tcg_out_movi(s, type, TCG_TMP0, val);
1023 if (type == TCG_TYPE_I32) {
1024 tcg_out_insn(s, RR, NR, dest, TCG_TMP0);
1025 } else {
1026 tcg_out_insn(s, RRE, NGR, dest, TCG_TMP0);
1030 static void tgen64_ori(TCGContext *s, TCGReg dest, tcg_target_ulong val)
1032 static const S390Opcode oi_insns[4] = {
1033 RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
1035 static const S390Opcode nif_insns[2] = {
1036 RIL_OILF, RIL_OIHF
1039 int i;
1041 /* Look for no-op. */
1042 if (val == 0) {
1043 return;
1046 if (facilities & FACILITY_EXT_IMM) {
1047 /* Try all 32-bit insns that can perform it in one go. */
1048 for (i = 0; i < 4; i++) {
1049 tcg_target_ulong mask = (0xffffull << i*16);
1050 if ((val & mask) != 0 && (val & ~mask) == 0) {
1051 tcg_out_insn_RI(s, oi_insns[i], dest, val >> i*16);
1052 return;
1056 /* Try all 48-bit insns that can perform it in one go. */
1057 for (i = 0; i < 2; i++) {
1058 tcg_target_ulong mask = (0xffffffffull << i*32);
1059 if ((val & mask) != 0 && (val & ~mask) == 0) {
1060 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> i*32);
1061 return;
1065 /* Perform the OR via sequential modifications to the high and
1066 low parts. Do this via recursion to handle 16-bit vs 32-bit
1067 masks in each half. */
1068 tgen64_ori(s, dest, val & 0x00000000ffffffffull);
1069 tgen64_ori(s, dest, val & 0xffffffff00000000ull);
1070 } else {
1071 /* With no extended-immediate facility, we don't need to be so
1072 clever. Just iterate over the insns and mask in the constant. */
1073 for (i = 0; i < 4; i++) {
1074 tcg_target_ulong mask = (0xffffull << i*16);
1075 if ((val & mask) != 0) {
1076 tcg_out_insn_RI(s, oi_insns[i], dest, val >> i*16);
1082 static void tgen64_xori(TCGContext *s, TCGReg dest, tcg_target_ulong val)
1084 /* Perform the xor by parts. */
1085 if (val & 0xffffffff) {
1086 tcg_out_insn(s, RIL, XILF, dest, val);
1088 if (val > 0xffffffff) {
1089 tcg_out_insn(s, RIL, XIHF, dest, val >> 31 >> 1);
1093 static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1094 TCGArg c2, int c2const)
1096 bool is_unsigned = is_unsigned_cond(c);
1097 if (c2const) {
1098 if (c2 == 0) {
1099 if (type == TCG_TYPE_I32) {
1100 tcg_out_insn(s, RR, LTR, r1, r1);
1101 } else {
1102 tcg_out_insn(s, RRE, LTGR, r1, r1);
1104 return tcg_cond_to_ltr_cond[c];
1105 } else {
1106 if (is_unsigned) {
1107 if (type == TCG_TYPE_I32) {
1108 tcg_out_insn(s, RIL, CLFI, r1, c2);
1109 } else {
1110 tcg_out_insn(s, RIL, CLGFI, r1, c2);
1112 } else {
1113 if (type == TCG_TYPE_I32) {
1114 tcg_out_insn(s, RIL, CFI, r1, c2);
1115 } else {
1116 tcg_out_insn(s, RIL, CGFI, r1, c2);
1120 } else {
1121 if (is_unsigned) {
1122 if (type == TCG_TYPE_I32) {
1123 tcg_out_insn(s, RR, CLR, r1, c2);
1124 } else {
1125 tcg_out_insn(s, RRE, CLGR, r1, c2);
1127 } else {
1128 if (type == TCG_TYPE_I32) {
1129 tcg_out_insn(s, RR, CR, r1, c2);
1130 } else {
1131 tcg_out_insn(s, RRE, CGR, r1, c2);
1135 return tcg_cond_to_s390_cond[c];
1138 static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1139 TCGReg dest, TCGReg c1, TCGArg c2, int c2const)
1141 int cc;
1143 switch (cond) {
1144 case TCG_COND_GTU:
1145 case TCG_COND_GT:
1146 do_greater:
1147 /* The result of a compare has CC=2 for GT and CC=3 unused.
1148 ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit. */
1149 tgen_cmp(s, type, cond, c1, c2, c2const);
1150 tcg_out_movi(s, type, dest, 0);
1151 tcg_out_insn(s, RRE, ALCGR, dest, dest);
1152 return;
1154 case TCG_COND_GEU:
1155 do_geu:
1156 /* We need "real" carry semantics, so use SUBTRACT LOGICAL
1157 instead of COMPARE LOGICAL. This needs an extra move. */
1158 tcg_out_mov(s, type, TCG_TMP0, c1);
1159 if (c2const) {
1160 tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1161 if (type == TCG_TYPE_I32) {
1162 tcg_out_insn(s, RIL, SLFI, TCG_TMP0, c2);
1163 } else {
1164 tcg_out_insn(s, RIL, SLGFI, TCG_TMP0, c2);
1166 } else {
1167 if (type == TCG_TYPE_I32) {
1168 tcg_out_insn(s, RR, SLR, TCG_TMP0, c2);
1169 } else {
1170 tcg_out_insn(s, RRE, SLGR, TCG_TMP0, c2);
1172 tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1174 tcg_out_insn(s, RRE, ALCGR, dest, dest);
1175 return;
1177 case TCG_COND_LEU:
1178 case TCG_COND_LTU:
1179 case TCG_COND_LT:
1180 /* Swap operands so that we can use GEU/GTU/GT. */
1181 if (c2const) {
1182 tcg_out_movi(s, type, TCG_TMP0, c2);
1183 c2 = c1;
1184 c2const = 0;
1185 c1 = TCG_TMP0;
1186 } else {
1187 TCGReg t = c1;
1188 c1 = c2;
1189 c2 = t;
1191 if (cond == TCG_COND_LEU) {
1192 goto do_geu;
1194 cond = tcg_swap_cond(cond);
1195 goto do_greater;
1197 case TCG_COND_NE:
1198 /* X != 0 is X > 0. */
1199 if (c2const && c2 == 0) {
1200 cond = TCG_COND_GTU;
1201 goto do_greater;
1203 break;
1205 case TCG_COND_EQ:
1206 /* X == 0 is X <= 0 is 0 >= X. */
1207 if (c2const && c2 == 0) {
1208 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0);
1209 c2 = c1;
1210 c2const = 0;
1211 c1 = TCG_TMP0;
1212 goto do_geu;
1214 break;
1216 default:
1217 break;
1220 cc = tgen_cmp(s, type, cond, c1, c2, c2const);
1221 if (facilities & FACILITY_LOAD_ON_COND) {
1222 /* Emit: d = 0, t = 1, d = (cc ? t : d). */
1223 tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1224 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 1);
1225 tcg_out_insn(s, RRF, LOCGR, dest, TCG_TMP0, cc);
1226 } else {
1227 /* Emit: d = 1; if (cc) goto over; d = 0; over: */
1228 tcg_out_movi(s, type, dest, 1);
1229 tcg_out_insn(s, RI, BRC, cc, (4 + 4) >> 1);
1230 tcg_out_movi(s, type, dest, 0);
1234 static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1235 TCGReg c1, TCGArg c2, int c2const, TCGReg r3)
1237 int cc;
1238 if (facilities & FACILITY_LOAD_ON_COND) {
1239 cc = tgen_cmp(s, type, c, c1, c2, c2const);
1240 tcg_out_insn(s, RRF, LOCGR, dest, r3, cc);
1241 } else {
1242 c = tcg_invert_cond(c);
1243 cc = tgen_cmp(s, type, c, c1, c2, c2const);
1245 /* Emit: if (cc) goto over; dest = r3; over: */
1246 tcg_out_insn(s, RI, BRC, cc, (4 + 4) >> 1);
1247 tcg_out_insn(s, RRE, LGR, dest, r3);
1251 bool tcg_target_deposit_valid(int ofs, int len)
1253 return (facilities & FACILITY_GEN_INST_EXT) != 0;
1256 static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1257 int ofs, int len)
1259 int lsb = (63 - ofs);
1260 int msb = lsb - (len - 1);
1261 tcg_out_risbg(s, dest, src, msb, lsb, ofs, 0);
1264 static void tgen_gotoi(TCGContext *s, int cc, tcg_insn_unit *dest)
1266 ptrdiff_t off = dest - s->code_ptr;
1267 if (off == (int16_t)off) {
1268 tcg_out_insn(s, RI, BRC, cc, off);
1269 } else if (off == (int32_t)off) {
1270 tcg_out_insn(s, RIL, BRCL, cc, off);
1271 } else {
1272 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1273 tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1277 static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1279 if (l->has_value) {
1280 tgen_gotoi(s, cc, l->u.value_ptr);
1281 } else if (USE_LONG_BRANCHES) {
1282 tcg_out16(s, RIL_BRCL | (cc << 4));
1283 tcg_out_reloc(s, s->code_ptr, R_390_PC32DBL, l, -2);
1284 s->code_ptr += 2;
1285 } else {
1286 tcg_out16(s, RI_BRC | (cc << 4));
1287 tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, -2);
1288 s->code_ptr += 1;
1292 static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1293 TCGReg r1, TCGReg r2, TCGLabel *l)
1295 intptr_t off;
1297 if (l->has_value) {
1298 off = l->u.value_ptr - s->code_ptr;
1299 } else {
1300 /* We need to keep the offset unchanged for retranslation. */
1301 off = s->code_ptr[1];
1302 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, -2);
1305 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1306 tcg_out16(s, off);
1307 tcg_out16(s, cc << 12 | (opc & 0xff));
1310 static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1311 TCGReg r1, int i2, TCGLabel *l)
1313 tcg_target_long off;
1315 if (l->has_value) {
1316 off = l->u.value_ptr - s->code_ptr;
1317 } else {
1318 /* We need to keep the offset unchanged for retranslation. */
1319 off = s->code_ptr[1];
1320 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, -2);
1323 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1324 tcg_out16(s, off);
1325 tcg_out16(s, (i2 << 8) | (opc & 0xff));
1328 static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1329 TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1331 int cc;
1333 if (facilities & FACILITY_GEN_INST_EXT) {
1334 bool is_unsigned = is_unsigned_cond(c);
1335 bool in_range;
1336 S390Opcode opc;
1338 cc = tcg_cond_to_s390_cond[c];
1340 if (!c2const) {
1341 opc = (type == TCG_TYPE_I32
1342 ? (is_unsigned ? RIE_CLRJ : RIE_CRJ)
1343 : (is_unsigned ? RIE_CLGRJ : RIE_CGRJ));
1344 tgen_compare_branch(s, opc, cc, r1, c2, l);
1345 return;
1348 /* COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1349 If the immediate we've been given does not fit that range, we'll
1350 fall back to separate compare and branch instructions using the
1351 larger comparison range afforded by COMPARE IMMEDIATE. */
1352 if (type == TCG_TYPE_I32) {
1353 if (is_unsigned) {
1354 opc = RIE_CLIJ;
1355 in_range = (uint32_t)c2 == (uint8_t)c2;
1356 } else {
1357 opc = RIE_CIJ;
1358 in_range = (int32_t)c2 == (int8_t)c2;
1360 } else {
1361 if (is_unsigned) {
1362 opc = RIE_CLGIJ;
1363 in_range = (uint64_t)c2 == (uint8_t)c2;
1364 } else {
1365 opc = RIE_CGIJ;
1366 in_range = (int64_t)c2 == (int8_t)c2;
1369 if (in_range) {
1370 tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1371 return;
1375 cc = tgen_cmp(s, type, c, r1, c2, c2const);
1376 tgen_branch(s, cc, l);
1379 static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
1381 ptrdiff_t off = dest - s->code_ptr;
1382 if (off == (int32_t)off) {
1383 tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1384 } else {
1385 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1386 tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1390 static void tcg_out_qemu_ld_direct(TCGContext *s, TCGMemOp opc, TCGReg data,
1391 TCGReg base, TCGReg index, int disp)
1393 switch (opc) {
1394 case MO_UB:
1395 tcg_out_insn(s, RXY, LLGC, data, base, index, disp);
1396 break;
1397 case MO_SB:
1398 tcg_out_insn(s, RXY, LGB, data, base, index, disp);
1399 break;
1401 case MO_UW | MO_BSWAP:
1402 /* swapped unsigned halfword load with upper bits zeroed */
1403 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1404 tgen_ext16u(s, TCG_TYPE_I64, data, data);
1405 break;
1406 case MO_UW:
1407 tcg_out_insn(s, RXY, LLGH, data, base, index, disp);
1408 break;
1410 case MO_SW | MO_BSWAP:
1411 /* swapped sign-extended halfword load */
1412 tcg_out_insn(s, RXY, LRVH, data, base, index, disp);
1413 tgen_ext16s(s, TCG_TYPE_I64, data, data);
1414 break;
1415 case MO_SW:
1416 tcg_out_insn(s, RXY, LGH, data, base, index, disp);
1417 break;
1419 case MO_UL | MO_BSWAP:
1420 /* swapped unsigned int load with upper bits zeroed */
1421 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1422 tgen_ext32u(s, data, data);
1423 break;
1424 case MO_UL:
1425 tcg_out_insn(s, RXY, LLGF, data, base, index, disp);
1426 break;
1428 case MO_SL | MO_BSWAP:
1429 /* swapped sign-extended int load */
1430 tcg_out_insn(s, RXY, LRV, data, base, index, disp);
1431 tgen_ext32s(s, data, data);
1432 break;
1433 case MO_SL:
1434 tcg_out_insn(s, RXY, LGF, data, base, index, disp);
1435 break;
1437 case MO_Q | MO_BSWAP:
1438 tcg_out_insn(s, RXY, LRVG, data, base, index, disp);
1439 break;
1440 case MO_Q:
1441 tcg_out_insn(s, RXY, LG, data, base, index, disp);
1442 break;
1444 default:
1445 tcg_abort();
1449 static void tcg_out_qemu_st_direct(TCGContext *s, TCGMemOp opc, TCGReg data,
1450 TCGReg base, TCGReg index, int disp)
1452 switch (opc) {
1453 case MO_UB:
1454 if (disp >= 0 && disp < 0x1000) {
1455 tcg_out_insn(s, RX, STC, data, base, index, disp);
1456 } else {
1457 tcg_out_insn(s, RXY, STCY, data, base, index, disp);
1459 break;
1461 case MO_UW | MO_BSWAP:
1462 tcg_out_insn(s, RXY, STRVH, data, base, index, disp);
1463 break;
1464 case MO_UW:
1465 if (disp >= 0 && disp < 0x1000) {
1466 tcg_out_insn(s, RX, STH, data, base, index, disp);
1467 } else {
1468 tcg_out_insn(s, RXY, STHY, data, base, index, disp);
1470 break;
1472 case MO_UL | MO_BSWAP:
1473 tcg_out_insn(s, RXY, STRV, data, base, index, disp);
1474 break;
1475 case MO_UL:
1476 if (disp >= 0 && disp < 0x1000) {
1477 tcg_out_insn(s, RX, ST, data, base, index, disp);
1478 } else {
1479 tcg_out_insn(s, RXY, STY, data, base, index, disp);
1481 break;
1483 case MO_Q | MO_BSWAP:
1484 tcg_out_insn(s, RXY, STRVG, data, base, index, disp);
1485 break;
1486 case MO_Q:
1487 tcg_out_insn(s, RXY, STG, data, base, index, disp);
1488 break;
1490 default:
1491 tcg_abort();
1495 #if defined(CONFIG_SOFTMMU)
1496 /* We're expecting to use a 20-bit signed offset on the tlb memory ops.
1497 Using the offset of the second entry in the last tlb table ensures
1498 that we can index all of the elements of the first entry. */
1499 QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1][1])
1500 > 0x7ffff);
1502 /* Load and compare a TLB entry, leaving the flags set. Loads the TLB
1503 addend into R2. Returns a register with the santitized guest address. */
1504 static TCGReg tcg_out_tlb_read(TCGContext* s, TCGReg addr_reg, TCGMemOp opc,
1505 int mem_index, bool is_ld)
1507 TCGMemOp s_bits = opc & MO_SIZE;
1508 uint64_t tlb_mask = TARGET_PAGE_MASK | ((1 << s_bits) - 1);
1509 int ofs;
1511 if (facilities & FACILITY_GEN_INST_EXT) {
1512 tcg_out_risbg(s, TCG_REG_R2, addr_reg,
1513 64 - CPU_TLB_BITS - CPU_TLB_ENTRY_BITS,
1514 63 - CPU_TLB_ENTRY_BITS,
1515 64 + CPU_TLB_ENTRY_BITS - TARGET_PAGE_BITS, 1);
1516 tgen_andi_risbg(s, TCG_REG_R3, addr_reg, tlb_mask);
1517 } else {
1518 tcg_out_sh64(s, RSY_SRLG, TCG_REG_R2, addr_reg, TCG_REG_NONE,
1519 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1520 tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_R3, addr_reg);
1521 tgen_andi(s, TCG_TYPE_I64, TCG_REG_R2,
1522 (CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS);
1523 tgen_andi(s, TCG_TYPE_TL, TCG_REG_R3, tlb_mask);
1526 if (is_ld) {
1527 ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addr_read);
1528 } else {
1529 ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addr_write);
1531 if (TARGET_LONG_BITS == 32) {
1532 tcg_out_mem(s, RX_C, RXY_CY, TCG_REG_R3, TCG_REG_R2, TCG_AREG0, ofs);
1533 } else {
1534 tcg_out_mem(s, 0, RXY_CG, TCG_REG_R3, TCG_REG_R2, TCG_AREG0, ofs);
1537 ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
1538 tcg_out_mem(s, 0, RXY_LG, TCG_REG_R2, TCG_REG_R2, TCG_AREG0, ofs);
1540 if (TARGET_LONG_BITS == 32) {
1541 tgen_ext32u(s, TCG_REG_R3, addr_reg);
1542 return TCG_REG_R3;
1544 return addr_reg;
1547 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOp opc,
1548 TCGReg data, TCGReg addr, int mem_index,
1549 tcg_insn_unit *raddr, tcg_insn_unit *label_ptr)
1551 TCGLabelQemuLdst *label = new_ldst_label(s);
1553 label->is_ld = is_ld;
1554 label->opc = opc;
1555 label->datalo_reg = data;
1556 label->addrlo_reg = addr;
1557 label->mem_index = mem_index;
1558 label->raddr = raddr;
1559 label->label_ptr[0] = label_ptr;
1562 static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1564 TCGReg addr_reg = lb->addrlo_reg;
1565 TCGReg data_reg = lb->datalo_reg;
1566 TCGMemOp opc = lb->opc;
1568 patch_reloc(lb->label_ptr[0], R_390_PC16DBL, (intptr_t)s->code_ptr, -2);
1570 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_AREG0);
1571 if (TARGET_LONG_BITS == 64) {
1572 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R3, addr_reg);
1574 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R4, lb->mem_index);
1575 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R5, (uintptr_t)lb->raddr);
1576 tcg_out_call(s, qemu_ld_helpers[opc]);
1577 tcg_out_mov(s, TCG_TYPE_I64, data_reg, TCG_REG_R2);
1579 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1582 static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1584 TCGReg addr_reg = lb->addrlo_reg;
1585 TCGReg data_reg = lb->datalo_reg;
1586 TCGMemOp opc = lb->opc;
1588 patch_reloc(lb->label_ptr[0], R_390_PC16DBL, (intptr_t)s->code_ptr, -2);
1590 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_R2, TCG_AREG0);
1591 if (TARGET_LONG_BITS == 64) {
1592 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R3, addr_reg);
1594 switch (opc & MO_SIZE) {
1595 case MO_UB:
1596 tgen_ext8u(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1597 break;
1598 case MO_UW:
1599 tgen_ext16u(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1600 break;
1601 case MO_UL:
1602 tgen_ext32u(s, TCG_REG_R4, data_reg);
1603 break;
1604 case MO_Q:
1605 tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_R4, data_reg);
1606 break;
1607 default:
1608 tcg_abort();
1610 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_R5, lb->mem_index);
1611 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R6, (uintptr_t)lb->raddr);
1612 tcg_out_call(s, qemu_st_helpers[opc]);
1614 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1616 #else
1617 static void tcg_prepare_user_ldst(TCGContext *s, TCGReg *addr_reg,
1618 TCGReg *index_reg, tcg_target_long *disp)
1620 if (TARGET_LONG_BITS == 32) {
1621 tgen_ext32u(s, TCG_TMP0, *addr_reg);
1622 *addr_reg = TCG_TMP0;
1624 if (GUEST_BASE < 0x80000) {
1625 *index_reg = TCG_REG_NONE;
1626 *disp = GUEST_BASE;
1627 } else {
1628 *index_reg = TCG_GUEST_BASE_REG;
1629 *disp = 0;
1632 #endif /* CONFIG_SOFTMMU */
1634 static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1635 TCGMemOp opc, int mem_index)
1637 #ifdef CONFIG_SOFTMMU
1638 tcg_insn_unit *label_ptr;
1639 TCGReg base_reg;
1641 base_reg = tcg_out_tlb_read(s, addr_reg, opc, mem_index, 1);
1643 label_ptr = s->code_ptr + 1;
1644 tcg_out_insn(s, RI, BRC, S390_CC_NE, 0);
1646 tcg_out_qemu_ld_direct(s, opc, data_reg, base_reg, TCG_REG_R2, 0);
1648 add_qemu_ldst_label(s, 1, opc, data_reg, addr_reg, mem_index,
1649 s->code_ptr, label_ptr);
1650 #else
1651 TCGReg index_reg;
1652 tcg_target_long disp;
1654 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1655 tcg_out_qemu_ld_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1656 #endif
1659 static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1660 TCGMemOp opc, int mem_index)
1662 #ifdef CONFIG_SOFTMMU
1663 tcg_insn_unit *label_ptr;
1664 TCGReg base_reg;
1666 base_reg = tcg_out_tlb_read(s, addr_reg, opc, mem_index, 0);
1668 label_ptr = s->code_ptr + 1;
1669 tcg_out_insn(s, RI, BRC, S390_CC_NE, 0);
1671 tcg_out_qemu_st_direct(s, opc, data_reg, base_reg, TCG_REG_R2, 0);
1673 add_qemu_ldst_label(s, 0, opc, data_reg, addr_reg, mem_index,
1674 s->code_ptr, label_ptr);
1675 #else
1676 TCGReg index_reg;
1677 tcg_target_long disp;
1679 tcg_prepare_user_ldst(s, &addr_reg, &index_reg, &disp);
1680 tcg_out_qemu_st_direct(s, opc, data_reg, addr_reg, index_reg, disp);
1681 #endif
1684 # define OP_32_64(x) \
1685 case glue(glue(INDEX_op_,x),_i32): \
1686 case glue(glue(INDEX_op_,x),_i64)
1688 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1689 const TCGArg *args, const int *const_args)
1691 S390Opcode op;
1692 TCGArg a0, a1, a2;
1694 switch (opc) {
1695 case INDEX_op_exit_tb:
1696 /* return value */
1697 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, args[0]);
1698 tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
1699 break;
1701 case INDEX_op_goto_tb:
1702 if (s->tb_jmp_offset) {
1703 tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
1704 s->tb_jmp_offset[args[0]] = tcg_current_code_size(s);
1705 s->code_ptr += 2;
1706 } else {
1707 /* load address stored at s->tb_next + args[0] */
1708 tcg_out_ld_abs(s, TCG_TYPE_PTR, TCG_TMP0, s->tb_next + args[0]);
1709 /* and go there */
1710 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_TMP0);
1712 s->tb_next_offset[args[0]] = tcg_current_code_size(s);
1713 break;
1715 OP_32_64(ld8u):
1716 /* ??? LLC (RXY format) is only present with the extended-immediate
1717 facility, whereas LLGC is always present. */
1718 tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
1719 break;
1721 OP_32_64(ld8s):
1722 /* ??? LB is no smaller than LGB, so no point to using it. */
1723 tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
1724 break;
1726 OP_32_64(ld16u):
1727 /* ??? LLH (RXY format) is only present with the extended-immediate
1728 facility, whereas LLGH is always present. */
1729 tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
1730 break;
1732 case INDEX_op_ld16s_i32:
1733 tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
1734 break;
1736 case INDEX_op_ld_i32:
1737 tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1738 break;
1740 OP_32_64(st8):
1741 tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
1742 TCG_REG_NONE, args[2]);
1743 break;
1745 OP_32_64(st16):
1746 tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
1747 TCG_REG_NONE, args[2]);
1748 break;
1750 case INDEX_op_st_i32:
1751 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1752 break;
1754 case INDEX_op_add_i32:
1755 a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1756 if (const_args[2]) {
1757 do_addi_32:
1758 if (a0 == a1) {
1759 if (a2 == (int16_t)a2) {
1760 tcg_out_insn(s, RI, AHI, a0, a2);
1761 break;
1763 if (facilities & FACILITY_EXT_IMM) {
1764 tcg_out_insn(s, RIL, AFI, a0, a2);
1765 break;
1768 tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
1769 } else if (a0 == a1) {
1770 tcg_out_insn(s, RR, AR, a0, a2);
1771 } else {
1772 tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
1774 break;
1775 case INDEX_op_sub_i32:
1776 a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1777 if (const_args[2]) {
1778 a2 = -a2;
1779 goto do_addi_32;
1781 tcg_out_insn(s, RR, SR, args[0], args[2]);
1782 break;
1784 case INDEX_op_and_i32:
1785 if (const_args[2]) {
1786 tgen_andi(s, TCG_TYPE_I32, args[0], args[2]);
1787 } else {
1788 tcg_out_insn(s, RR, NR, args[0], args[2]);
1790 break;
1791 case INDEX_op_or_i32:
1792 if (const_args[2]) {
1793 tgen64_ori(s, args[0], args[2] & 0xffffffff);
1794 } else {
1795 tcg_out_insn(s, RR, OR, args[0], args[2]);
1797 break;
1798 case INDEX_op_xor_i32:
1799 if (const_args[2]) {
1800 tgen64_xori(s, args[0], args[2] & 0xffffffff);
1801 } else {
1802 tcg_out_insn(s, RR, XR, args[0], args[2]);
1804 break;
1806 case INDEX_op_neg_i32:
1807 tcg_out_insn(s, RR, LCR, args[0], args[1]);
1808 break;
1810 case INDEX_op_mul_i32:
1811 if (const_args[2]) {
1812 if ((int32_t)args[2] == (int16_t)args[2]) {
1813 tcg_out_insn(s, RI, MHI, args[0], args[2]);
1814 } else {
1815 tcg_out_insn(s, RIL, MSFI, args[0], args[2]);
1817 } else {
1818 tcg_out_insn(s, RRE, MSR, args[0], args[2]);
1820 break;
1822 case INDEX_op_div2_i32:
1823 tcg_out_insn(s, RR, DR, TCG_REG_R2, args[4]);
1824 break;
1825 case INDEX_op_divu2_i32:
1826 tcg_out_insn(s, RRE, DLR, TCG_REG_R2, args[4]);
1827 break;
1829 case INDEX_op_shl_i32:
1830 op = RS_SLL;
1831 do_shift32:
1832 if (const_args[2]) {
1833 tcg_out_sh32(s, op, args[0], TCG_REG_NONE, args[2]);
1834 } else {
1835 tcg_out_sh32(s, op, args[0], args[2], 0);
1837 break;
1838 case INDEX_op_shr_i32:
1839 op = RS_SRL;
1840 goto do_shift32;
1841 case INDEX_op_sar_i32:
1842 op = RS_SRA;
1843 goto do_shift32;
1845 case INDEX_op_rotl_i32:
1846 /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol. */
1847 if (const_args[2]) {
1848 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
1849 } else {
1850 tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
1852 break;
1853 case INDEX_op_rotr_i32:
1854 if (const_args[2]) {
1855 tcg_out_sh64(s, RSY_RLL, args[0], args[1],
1856 TCG_REG_NONE, (32 - args[2]) & 31);
1857 } else {
1858 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
1859 tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
1861 break;
1863 case INDEX_op_ext8s_i32:
1864 tgen_ext8s(s, TCG_TYPE_I32, args[0], args[1]);
1865 break;
1866 case INDEX_op_ext16s_i32:
1867 tgen_ext16s(s, TCG_TYPE_I32, args[0], args[1]);
1868 break;
1869 case INDEX_op_ext8u_i32:
1870 tgen_ext8u(s, TCG_TYPE_I32, args[0], args[1]);
1871 break;
1872 case INDEX_op_ext16u_i32:
1873 tgen_ext16u(s, TCG_TYPE_I32, args[0], args[1]);
1874 break;
1876 OP_32_64(bswap16):
1877 /* The TCG bswap definition requires bits 0-47 already be zero.
1878 Thus we don't need the G-type insns to implement bswap16_i64. */
1879 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
1880 tcg_out_sh32(s, RS_SRL, args[0], TCG_REG_NONE, 16);
1881 break;
1882 OP_32_64(bswap32):
1883 tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
1884 break;
1886 case INDEX_op_add2_i32:
1887 if (const_args[4]) {
1888 tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
1889 } else {
1890 tcg_out_insn(s, RR, ALR, args[0], args[4]);
1892 tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
1893 break;
1894 case INDEX_op_sub2_i32:
1895 if (const_args[4]) {
1896 tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
1897 } else {
1898 tcg_out_insn(s, RR, SLR, args[0], args[4]);
1900 tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
1901 break;
1903 case INDEX_op_br:
1904 tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
1905 break;
1907 case INDEX_op_brcond_i32:
1908 tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
1909 args[1], const_args[1], arg_label(args[3]));
1910 break;
1911 case INDEX_op_setcond_i32:
1912 tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
1913 args[2], const_args[2]);
1914 break;
1915 case INDEX_op_movcond_i32:
1916 tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
1917 args[2], const_args[2], args[3]);
1918 break;
1920 case INDEX_op_qemu_ld_i32:
1921 /* ??? Technically we can use a non-extending instruction. */
1922 case INDEX_op_qemu_ld_i64:
1923 tcg_out_qemu_ld(s, args[0], args[1], args[2], args[3]);
1924 break;
1925 case INDEX_op_qemu_st_i32:
1926 case INDEX_op_qemu_st_i64:
1927 tcg_out_qemu_st(s, args[0], args[1], args[2], args[3]);
1928 break;
1930 case INDEX_op_ld16s_i64:
1931 tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
1932 break;
1933 case INDEX_op_ld32u_i64:
1934 tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
1935 break;
1936 case INDEX_op_ld32s_i64:
1937 tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
1938 break;
1939 case INDEX_op_ld_i64:
1940 tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
1941 break;
1943 case INDEX_op_st32_i64:
1944 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1945 break;
1946 case INDEX_op_st_i64:
1947 tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
1948 break;
1950 case INDEX_op_add_i64:
1951 a0 = args[0], a1 = args[1], a2 = args[2];
1952 if (const_args[2]) {
1953 do_addi_64:
1954 if (a0 == a1) {
1955 if (a2 == (int16_t)a2) {
1956 tcg_out_insn(s, RI, AGHI, a0, a2);
1957 break;
1959 if (facilities & FACILITY_EXT_IMM) {
1960 if (a2 == (int32_t)a2) {
1961 tcg_out_insn(s, RIL, AGFI, a0, a2);
1962 break;
1963 } else if (a2 == (uint32_t)a2) {
1964 tcg_out_insn(s, RIL, ALGFI, a0, a2);
1965 break;
1966 } else if (-a2 == (uint32_t)-a2) {
1967 tcg_out_insn(s, RIL, SLGFI, a0, -a2);
1968 break;
1972 tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
1973 } else if (a0 == a1) {
1974 tcg_out_insn(s, RRE, AGR, a0, a2);
1975 } else {
1976 tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
1978 break;
1979 case INDEX_op_sub_i64:
1980 a0 = args[0], a1 = args[1], a2 = args[2];
1981 if (const_args[2]) {
1982 a2 = -a2;
1983 goto do_addi_64;
1984 } else {
1985 tcg_out_insn(s, RRE, SGR, args[0], args[2]);
1987 break;
1989 case INDEX_op_and_i64:
1990 if (const_args[2]) {
1991 tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
1992 } else {
1993 tcg_out_insn(s, RRE, NGR, args[0], args[2]);
1995 break;
1996 case INDEX_op_or_i64:
1997 if (const_args[2]) {
1998 tgen64_ori(s, args[0], args[2]);
1999 } else {
2000 tcg_out_insn(s, RRE, OGR, args[0], args[2]);
2002 break;
2003 case INDEX_op_xor_i64:
2004 if (const_args[2]) {
2005 tgen64_xori(s, args[0], args[2]);
2006 } else {
2007 tcg_out_insn(s, RRE, XGR, args[0], args[2]);
2009 break;
2011 case INDEX_op_neg_i64:
2012 tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2013 break;
2014 case INDEX_op_bswap64_i64:
2015 tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2016 break;
2018 case INDEX_op_mul_i64:
2019 if (const_args[2]) {
2020 if (args[2] == (int16_t)args[2]) {
2021 tcg_out_insn(s, RI, MGHI, args[0], args[2]);
2022 } else {
2023 tcg_out_insn(s, RIL, MSGFI, args[0], args[2]);
2025 } else {
2026 tcg_out_insn(s, RRE, MSGR, args[0], args[2]);
2028 break;
2030 case INDEX_op_div2_i64:
2031 /* ??? We get an unnecessary sign-extension of the dividend
2032 into R3 with this definition, but as we do in fact always
2033 produce both quotient and remainder using INDEX_op_div_i64
2034 instead requires jumping through even more hoops. */
2035 tcg_out_insn(s, RRE, DSGR, TCG_REG_R2, args[4]);
2036 break;
2037 case INDEX_op_divu2_i64:
2038 tcg_out_insn(s, RRE, DLGR, TCG_REG_R2, args[4]);
2039 break;
2040 case INDEX_op_mulu2_i64:
2041 tcg_out_insn(s, RRE, MLGR, TCG_REG_R2, args[3]);
2042 break;
2044 case INDEX_op_shl_i64:
2045 op = RSY_SLLG;
2046 do_shift64:
2047 if (const_args[2]) {
2048 tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2049 } else {
2050 tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2052 break;
2053 case INDEX_op_shr_i64:
2054 op = RSY_SRLG;
2055 goto do_shift64;
2056 case INDEX_op_sar_i64:
2057 op = RSY_SRAG;
2058 goto do_shift64;
2060 case INDEX_op_rotl_i64:
2061 if (const_args[2]) {
2062 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2063 TCG_REG_NONE, args[2]);
2064 } else {
2065 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2067 break;
2068 case INDEX_op_rotr_i64:
2069 if (const_args[2]) {
2070 tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2071 TCG_REG_NONE, (64 - args[2]) & 63);
2072 } else {
2073 /* We can use the smaller 32-bit negate because only the
2074 low 6 bits are examined for the rotate. */
2075 tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2076 tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2078 break;
2080 case INDEX_op_ext8s_i64:
2081 tgen_ext8s(s, TCG_TYPE_I64, args[0], args[1]);
2082 break;
2083 case INDEX_op_ext16s_i64:
2084 tgen_ext16s(s, TCG_TYPE_I64, args[0], args[1]);
2085 break;
2086 case INDEX_op_ext32s_i64:
2087 tgen_ext32s(s, args[0], args[1]);
2088 break;
2089 case INDEX_op_ext8u_i64:
2090 tgen_ext8u(s, TCG_TYPE_I64, args[0], args[1]);
2091 break;
2092 case INDEX_op_ext16u_i64:
2093 tgen_ext16u(s, TCG_TYPE_I64, args[0], args[1]);
2094 break;
2095 case INDEX_op_ext32u_i64:
2096 tgen_ext32u(s, args[0], args[1]);
2097 break;
2099 case INDEX_op_add2_i64:
2100 if (const_args[4]) {
2101 if ((int64_t)args[4] >= 0) {
2102 tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2103 } else {
2104 tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2106 } else {
2107 tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2109 tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2110 break;
2111 case INDEX_op_sub2_i64:
2112 if (const_args[4]) {
2113 if ((int64_t)args[4] >= 0) {
2114 tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2115 } else {
2116 tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2118 } else {
2119 tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2121 tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2122 break;
2124 case INDEX_op_brcond_i64:
2125 tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2126 args[1], const_args[1], arg_label(args[3]));
2127 break;
2128 case INDEX_op_setcond_i64:
2129 tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2130 args[2], const_args[2]);
2131 break;
2132 case INDEX_op_movcond_i64:
2133 tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2134 args[2], const_args[2], args[3]);
2135 break;
2137 OP_32_64(deposit):
2138 tgen_deposit(s, args[0], args[2], args[3], args[4]);
2139 break;
2141 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2142 case INDEX_op_mov_i64:
2143 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2144 case INDEX_op_movi_i64:
2145 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2146 default:
2147 tcg_abort();
2151 static const TCGTargetOpDef s390_op_defs[] = {
2152 { INDEX_op_exit_tb, { } },
2153 { INDEX_op_goto_tb, { } },
2154 { INDEX_op_br, { } },
2156 { INDEX_op_ld8u_i32, { "r", "r" } },
2157 { INDEX_op_ld8s_i32, { "r", "r" } },
2158 { INDEX_op_ld16u_i32, { "r", "r" } },
2159 { INDEX_op_ld16s_i32, { "r", "r" } },
2160 { INDEX_op_ld_i32, { "r", "r" } },
2161 { INDEX_op_st8_i32, { "r", "r" } },
2162 { INDEX_op_st16_i32, { "r", "r" } },
2163 { INDEX_op_st_i32, { "r", "r" } },
2165 { INDEX_op_add_i32, { "r", "r", "ri" } },
2166 { INDEX_op_sub_i32, { "r", "0", "ri" } },
2167 { INDEX_op_mul_i32, { "r", "0", "rK" } },
2169 { INDEX_op_div2_i32, { "b", "a", "0", "1", "r" } },
2170 { INDEX_op_divu2_i32, { "b", "a", "0", "1", "r" } },
2172 { INDEX_op_and_i32, { "r", "0", "ri" } },
2173 { INDEX_op_or_i32, { "r", "0", "rO" } },
2174 { INDEX_op_xor_i32, { "r", "0", "rX" } },
2176 { INDEX_op_neg_i32, { "r", "r" } },
2178 { INDEX_op_shl_i32, { "r", "0", "Ri" } },
2179 { INDEX_op_shr_i32, { "r", "0", "Ri" } },
2180 { INDEX_op_sar_i32, { "r", "0", "Ri" } },
2182 { INDEX_op_rotl_i32, { "r", "r", "Ri" } },
2183 { INDEX_op_rotr_i32, { "r", "r", "Ri" } },
2185 { INDEX_op_ext8s_i32, { "r", "r" } },
2186 { INDEX_op_ext8u_i32, { "r", "r" } },
2187 { INDEX_op_ext16s_i32, { "r", "r" } },
2188 { INDEX_op_ext16u_i32, { "r", "r" } },
2190 { INDEX_op_bswap16_i32, { "r", "r" } },
2191 { INDEX_op_bswap32_i32, { "r", "r" } },
2193 { INDEX_op_add2_i32, { "r", "r", "0", "1", "rA", "r" } },
2194 { INDEX_op_sub2_i32, { "r", "r", "0", "1", "rA", "r" } },
2196 { INDEX_op_brcond_i32, { "r", "rC" } },
2197 { INDEX_op_setcond_i32, { "r", "r", "rC" } },
2198 { INDEX_op_movcond_i32, { "r", "r", "rC", "r", "0" } },
2199 { INDEX_op_deposit_i32, { "r", "0", "r" } },
2201 { INDEX_op_qemu_ld_i32, { "r", "L" } },
2202 { INDEX_op_qemu_ld_i64, { "r", "L" } },
2203 { INDEX_op_qemu_st_i32, { "L", "L" } },
2204 { INDEX_op_qemu_st_i64, { "L", "L" } },
2206 { INDEX_op_ld8u_i64, { "r", "r" } },
2207 { INDEX_op_ld8s_i64, { "r", "r" } },
2208 { INDEX_op_ld16u_i64, { "r", "r" } },
2209 { INDEX_op_ld16s_i64, { "r", "r" } },
2210 { INDEX_op_ld32u_i64, { "r", "r" } },
2211 { INDEX_op_ld32s_i64, { "r", "r" } },
2212 { INDEX_op_ld_i64, { "r", "r" } },
2214 { INDEX_op_st8_i64, { "r", "r" } },
2215 { INDEX_op_st16_i64, { "r", "r" } },
2216 { INDEX_op_st32_i64, { "r", "r" } },
2217 { INDEX_op_st_i64, { "r", "r" } },
2219 { INDEX_op_add_i64, { "r", "r", "ri" } },
2220 { INDEX_op_sub_i64, { "r", "0", "ri" } },
2221 { INDEX_op_mul_i64, { "r", "0", "rK" } },
2223 { INDEX_op_div2_i64, { "b", "a", "0", "1", "r" } },
2224 { INDEX_op_divu2_i64, { "b", "a", "0", "1", "r" } },
2225 { INDEX_op_mulu2_i64, { "b", "a", "0", "r" } },
2227 { INDEX_op_and_i64, { "r", "0", "ri" } },
2228 { INDEX_op_or_i64, { "r", "0", "rO" } },
2229 { INDEX_op_xor_i64, { "r", "0", "rX" } },
2231 { INDEX_op_neg_i64, { "r", "r" } },
2233 { INDEX_op_shl_i64, { "r", "r", "Ri" } },
2234 { INDEX_op_shr_i64, { "r", "r", "Ri" } },
2235 { INDEX_op_sar_i64, { "r", "r", "Ri" } },
2237 { INDEX_op_rotl_i64, { "r", "r", "Ri" } },
2238 { INDEX_op_rotr_i64, { "r", "r", "Ri" } },
2240 { INDEX_op_ext8s_i64, { "r", "r" } },
2241 { INDEX_op_ext8u_i64, { "r", "r" } },
2242 { INDEX_op_ext16s_i64, { "r", "r" } },
2243 { INDEX_op_ext16u_i64, { "r", "r" } },
2244 { INDEX_op_ext32s_i64, { "r", "r" } },
2245 { INDEX_op_ext32u_i64, { "r", "r" } },
2247 { INDEX_op_bswap16_i64, { "r", "r" } },
2248 { INDEX_op_bswap32_i64, { "r", "r" } },
2249 { INDEX_op_bswap64_i64, { "r", "r" } },
2251 { INDEX_op_add2_i64, { "r", "r", "0", "1", "rA", "r" } },
2252 { INDEX_op_sub2_i64, { "r", "r", "0", "1", "rA", "r" } },
2254 { INDEX_op_brcond_i64, { "r", "rC" } },
2255 { INDEX_op_setcond_i64, { "r", "r", "rC" } },
2256 { INDEX_op_movcond_i64, { "r", "r", "rC", "r", "0" } },
2257 { INDEX_op_deposit_i64, { "r", "0", "r" } },
2259 { -1 },
2262 static void query_facilities(void)
2264 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2266 /* Is STORE FACILITY LIST EXTENDED available? Honestly, I believe this
2267 is present on all 64-bit systems, but let's check for it anyway. */
2268 if (hwcap & HWCAP_S390_STFLE) {
2269 register int r0 __asm__("0");
2270 register void *r1 __asm__("1");
2272 /* stfle 0(%r1) */
2273 r1 = &facilities;
2274 asm volatile(".word 0xb2b0,0x1000"
2275 : "=r"(r0) : "0"(0), "r"(r1) : "memory", "cc");
2279 static void tcg_target_init(TCGContext *s)
2281 query_facilities();
2283 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffff);
2284 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, 0xffff);
2286 tcg_regset_clear(tcg_target_call_clobber_regs);
2287 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2288 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2289 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2290 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2291 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
2292 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
2293 /* The r6 register is technically call-saved, but it's also a parameter
2294 register, so it can get killed by setup for the qemu_st helper. */
2295 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
2296 /* The return register can be considered call-clobbered. */
2297 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2299 tcg_regset_clear(s->reserved_regs);
2300 tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
2301 /* XXX many insns can't be used with R0, so we better avoid it for now */
2302 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
2303 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2305 tcg_add_target_add_op_defs(s390_op_defs);
2308 #define FRAME_SIZE ((int)(TCG_TARGET_CALL_STACK_OFFSET \
2309 + TCG_STATIC_CALL_ARGS_SIZE \
2310 + CPU_TEMP_BUF_NLONGS * sizeof(long)))
2312 static void tcg_target_qemu_prologue(TCGContext *s)
2314 /* stmg %r6,%r15,48(%r15) (save registers) */
2315 tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
2317 /* aghi %r15,-frame_size */
2318 tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
2320 tcg_set_frame(s, TCG_REG_CALL_STACK,
2321 TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
2322 CPU_TEMP_BUF_NLONGS * sizeof(long));
2324 if (GUEST_BASE >= 0x80000) {
2325 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, GUEST_BASE);
2326 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2329 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2330 /* br %r3 (go to TB) */
2331 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
2333 tb_ret_addr = s->code_ptr;
2335 /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
2336 tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
2337 FRAME_SIZE + 48);
2339 /* br %r14 (return) */
2340 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
2343 typedef struct {
2344 DebugFrameHeader h;
2345 uint8_t fde_def_cfa[4];
2346 uint8_t fde_reg_ofs[18];
2347 } DebugFrame;
2349 /* We're expecting a 2 byte uleb128 encoded value. */
2350 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2352 #define ELF_HOST_MACHINE EM_S390
2354 static const DebugFrame debug_frame = {
2355 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2356 .h.cie.id = -1,
2357 .h.cie.version = 1,
2358 .h.cie.code_align = 1,
2359 .h.cie.data_align = 8, /* sleb128 8 */
2360 .h.cie.return_column = TCG_REG_R14,
2362 /* Total FDE size does not include the "len" member. */
2363 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2365 .fde_def_cfa = {
2366 12, TCG_REG_CALL_STACK, /* DW_CFA_def_cfa %r15, ... */
2367 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2368 (FRAME_SIZE >> 7)
2370 .fde_reg_ofs = {
2371 0x86, 6, /* DW_CFA_offset, %r6, 48 */
2372 0x87, 7, /* DW_CFA_offset, %r7, 56 */
2373 0x88, 8, /* DW_CFA_offset, %r8, 64 */
2374 0x89, 9, /* DW_CFA_offset, %r92, 72 */
2375 0x8a, 10, /* DW_CFA_offset, %r10, 80 */
2376 0x8b, 11, /* DW_CFA_offset, %r11, 88 */
2377 0x8c, 12, /* DW_CFA_offset, %r12, 96 */
2378 0x8d, 13, /* DW_CFA_offset, %r13, 104 */
2379 0x8e, 14, /* DW_CFA_offset, %r14, 112 */
2383 void tcg_register_jit(void *buf, size_t buf_size)
2385 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));