xref: /openbmc/qemu/tcg/arm/tcg-target.c.inc (revision c00506aa)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Andrzej Zaborowski
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "elf.h"
26#include "../tcg-pool.c.inc"
27
28int arm_arch = __ARM_ARCH;
29
30#ifndef use_idiv_instructions
31bool use_idiv_instructions;
32#endif
33
34/* ??? Ought to think about changing CONFIG_SOFTMMU to always defined.  */
35#ifdef CONFIG_SOFTMMU
36# define USING_SOFTMMU 1
37#else
38# define USING_SOFTMMU 0
39#endif
40
41#ifdef CONFIG_DEBUG_TCG
42static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
43    "%r0",
44    "%r1",
45    "%r2",
46    "%r3",
47    "%r4",
48    "%r5",
49    "%r6",
50    "%r7",
51    "%r8",
52    "%r9",
53    "%r10",
54    "%r11",
55    "%r12",
56    "%r13",
57    "%r14",
58    "%pc",
59};
60#endif
61
62static const int tcg_target_reg_alloc_order[] = {
63    TCG_REG_R4,
64    TCG_REG_R5,
65    TCG_REG_R6,
66    TCG_REG_R7,
67    TCG_REG_R8,
68    TCG_REG_R9,
69    TCG_REG_R10,
70    TCG_REG_R11,
71    TCG_REG_R13,
72    TCG_REG_R0,
73    TCG_REG_R1,
74    TCG_REG_R2,
75    TCG_REG_R3,
76    TCG_REG_R12,
77    TCG_REG_R14,
78};
79
80static const int tcg_target_call_iarg_regs[4] = {
81    TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3
82};
83static const int tcg_target_call_oarg_regs[2] = {
84    TCG_REG_R0, TCG_REG_R1
85};
86
87#define TCG_REG_TMP  TCG_REG_R12
88
89enum arm_cond_code_e {
90    COND_EQ = 0x0,
91    COND_NE = 0x1,
92    COND_CS = 0x2,	/* Unsigned greater or equal */
93    COND_CC = 0x3,	/* Unsigned less than */
94    COND_MI = 0x4,	/* Negative */
95    COND_PL = 0x5,	/* Zero or greater */
96    COND_VS = 0x6,	/* Overflow */
97    COND_VC = 0x7,	/* No overflow */
98    COND_HI = 0x8,	/* Unsigned greater than */
99    COND_LS = 0x9,	/* Unsigned less or equal */
100    COND_GE = 0xa,
101    COND_LT = 0xb,
102    COND_GT = 0xc,
103    COND_LE = 0xd,
104    COND_AL = 0xe,
105};
106
107#define TO_CPSR (1 << 20)
108
109#define SHIFT_IMM_LSL(im)	(((im) << 7) | 0x00)
110#define SHIFT_IMM_LSR(im)	(((im) << 7) | 0x20)
111#define SHIFT_IMM_ASR(im)	(((im) << 7) | 0x40)
112#define SHIFT_IMM_ROR(im)	(((im) << 7) | 0x60)
113#define SHIFT_REG_LSL(rs)	(((rs) << 8) | 0x10)
114#define SHIFT_REG_LSR(rs)	(((rs) << 8) | 0x30)
115#define SHIFT_REG_ASR(rs)	(((rs) << 8) | 0x50)
116#define SHIFT_REG_ROR(rs)	(((rs) << 8) | 0x70)
117
118typedef enum {
119    ARITH_AND = 0x0 << 21,
120    ARITH_EOR = 0x1 << 21,
121    ARITH_SUB = 0x2 << 21,
122    ARITH_RSB = 0x3 << 21,
123    ARITH_ADD = 0x4 << 21,
124    ARITH_ADC = 0x5 << 21,
125    ARITH_SBC = 0x6 << 21,
126    ARITH_RSC = 0x7 << 21,
127    ARITH_TST = 0x8 << 21 | TO_CPSR,
128    ARITH_CMP = 0xa << 21 | TO_CPSR,
129    ARITH_CMN = 0xb << 21 | TO_CPSR,
130    ARITH_ORR = 0xc << 21,
131    ARITH_MOV = 0xd << 21,
132    ARITH_BIC = 0xe << 21,
133    ARITH_MVN = 0xf << 21,
134
135    INSN_CLZ       = 0x016f0f10,
136    INSN_RBIT      = 0x06ff0f30,
137
138    INSN_LDR_IMM   = 0x04100000,
139    INSN_LDR_REG   = 0x06100000,
140    INSN_STR_IMM   = 0x04000000,
141    INSN_STR_REG   = 0x06000000,
142
143    INSN_LDRH_IMM  = 0x005000b0,
144    INSN_LDRH_REG  = 0x001000b0,
145    INSN_LDRSH_IMM = 0x005000f0,
146    INSN_LDRSH_REG = 0x001000f0,
147    INSN_STRH_IMM  = 0x004000b0,
148    INSN_STRH_REG  = 0x000000b0,
149
150    INSN_LDRB_IMM  = 0x04500000,
151    INSN_LDRB_REG  = 0x06500000,
152    INSN_LDRSB_IMM = 0x005000d0,
153    INSN_LDRSB_REG = 0x001000d0,
154    INSN_STRB_IMM  = 0x04400000,
155    INSN_STRB_REG  = 0x06400000,
156
157    INSN_LDRD_IMM  = 0x004000d0,
158    INSN_LDRD_REG  = 0x000000d0,
159    INSN_STRD_IMM  = 0x004000f0,
160    INSN_STRD_REG  = 0x000000f0,
161
162    INSN_DMB_ISH   = 0xf57ff05b,
163    INSN_DMB_MCR   = 0xee070fba,
164
165    /* Architected nop introduced in v6k.  */
166    /* ??? This is an MSR (imm) 0,0,0 insn.  Anyone know if this
167       also Just So Happened to do nothing on pre-v6k so that we
168       don't need to conditionalize it?  */
169    INSN_NOP_v6k   = 0xe320f000,
170    /* Otherwise the assembler uses mov r0,r0 */
171    INSN_NOP_v4    = (COND_AL << 28) | ARITH_MOV,
172} ARMInsn;
173
174#define INSN_NOP   (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
175
176static const uint8_t tcg_cond_to_arm_cond[] = {
177    [TCG_COND_EQ] = COND_EQ,
178    [TCG_COND_NE] = COND_NE,
179    [TCG_COND_LT] = COND_LT,
180    [TCG_COND_GE] = COND_GE,
181    [TCG_COND_LE] = COND_LE,
182    [TCG_COND_GT] = COND_GT,
183    /* unsigned */
184    [TCG_COND_LTU] = COND_CC,
185    [TCG_COND_GEU] = COND_CS,
186    [TCG_COND_LEU] = COND_LS,
187    [TCG_COND_GTU] = COND_HI,
188};
189
190static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
191{
192    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
193    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) >> 2;
194
195    if (offset == sextract32(offset, 0, 24)) {
196        *src_rw = deposit32(*src_rw, 0, 24, offset);
197        return true;
198    }
199    return false;
200}
201
202static bool reloc_pc13(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
203{
204    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
205    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
206
207    if (offset >= -0xfff && offset <= 0xfff) {
208        tcg_insn_unit insn = *src_rw;
209        bool u = (offset >= 0);
210        if (!u) {
211            offset = -offset;
212        }
213        insn = deposit32(insn, 23, 1, u);
214        insn = deposit32(insn, 0, 12, offset);
215        *src_rw = insn;
216        return true;
217    }
218    return false;
219}
220
221static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
222                        intptr_t value, intptr_t addend)
223{
224    tcg_debug_assert(addend == 0);
225
226    if (type == R_ARM_PC24) {
227        return reloc_pc24(code_ptr, (const tcg_insn_unit *)value);
228    } else if (type == R_ARM_PC13) {
229        return reloc_pc13(code_ptr, (const tcg_insn_unit *)value);
230    } else {
231        g_assert_not_reached();
232    }
233}
234
235#define TCG_CT_CONST_ARM  0x100
236#define TCG_CT_CONST_INV  0x200
237#define TCG_CT_CONST_NEG  0x400
238#define TCG_CT_CONST_ZERO 0x800
239
240/* parse target specific constraints */
241static const char *target_parse_constraint(TCGArgConstraint *ct,
242                                           const char *ct_str, TCGType type)
243{
244    switch (*ct_str++) {
245    case 'I':
246        ct->ct |= TCG_CT_CONST_ARM;
247        break;
248    case 'K':
249        ct->ct |= TCG_CT_CONST_INV;
250        break;
251    case 'N': /* The gcc constraint letter is L, already used here.  */
252        ct->ct |= TCG_CT_CONST_NEG;
253        break;
254    case 'Z':
255        ct->ct |= TCG_CT_CONST_ZERO;
256        break;
257
258    case 'r':
259        ct->regs = 0xffff;
260        break;
261
262    /* qemu_ld address */
263    case 'l':
264        ct->regs = 0xffff;
265#ifdef CONFIG_SOFTMMU
266        /* r0-r2,lr will be overwritten when reading the tlb entry,
267           so don't use these. */
268        tcg_regset_reset_reg(ct->regs, TCG_REG_R0);
269        tcg_regset_reset_reg(ct->regs, TCG_REG_R1);
270        tcg_regset_reset_reg(ct->regs, TCG_REG_R2);
271        tcg_regset_reset_reg(ct->regs, TCG_REG_R3);
272        tcg_regset_reset_reg(ct->regs, TCG_REG_R14);
273#endif
274        break;
275
276    /* qemu_st address & data */
277    case 's':
278        ct->regs = 0xffff;
279        /* r0-r2 will be overwritten when reading the tlb entry (softmmu only)
280           and r0-r1 doing the byte swapping, so don't use these. */
281        tcg_regset_reset_reg(ct->regs, TCG_REG_R0);
282        tcg_regset_reset_reg(ct->regs, TCG_REG_R1);
283#if defined(CONFIG_SOFTMMU)
284        /* Avoid clashes with registers being used for helper args */
285        tcg_regset_reset_reg(ct->regs, TCG_REG_R2);
286#if TARGET_LONG_BITS == 64
287        /* Avoid clashes with registers being used for helper args */
288        tcg_regset_reset_reg(ct->regs, TCG_REG_R3);
289#endif
290        tcg_regset_reset_reg(ct->regs, TCG_REG_R14);
291#endif
292        break;
293
294    default:
295        return NULL;
296    }
297    return ct_str;
298}
299
300static inline uint32_t rotl(uint32_t val, int n)
301{
302  return (val << n) | (val >> (32 - n));
303}
304
305/* ARM immediates for ALU instructions are made of an unsigned 8-bit
306   right-rotated by an even amount between 0 and 30. */
307static inline int encode_imm(uint32_t imm)
308{
309    int shift;
310
311    /* simple case, only lower bits */
312    if ((imm & ~0xff) == 0)
313        return 0;
314    /* then try a simple even shift */
315    shift = ctz32(imm) & ~1;
316    if (((imm >> shift) & ~0xff) == 0)
317        return 32 - shift;
318    /* now try harder with rotations */
319    if ((rotl(imm, 2) & ~0xff) == 0)
320        return 2;
321    if ((rotl(imm, 4) & ~0xff) == 0)
322        return 4;
323    if ((rotl(imm, 6) & ~0xff) == 0)
324        return 6;
325    /* imm can't be encoded */
326    return -1;
327}
328
329static inline int check_fit_imm(uint32_t imm)
330{
331    return encode_imm(imm) >= 0;
332}
333
334/* Test if a constant matches the constraint.
335 * TODO: define constraints for:
336 *
337 * ldr/str offset:   between -0xfff and 0xfff
338 * ldrh/strh offset: between -0xff and 0xff
339 * mov operand2:     values represented with x << (2 * y), x < 0x100
340 * add, sub, eor...: ditto
341 */
342static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
343                                         const TCGArgConstraint *arg_ct)
344{
345    int ct;
346    ct = arg_ct->ct;
347    if (ct & TCG_CT_CONST) {
348        return 1;
349    } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
350        return 1;
351    } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
352        return 1;
353    } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
354        return 1;
355    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
356        return 1;
357    } else {
358        return 0;
359    }
360}
361
362static inline void tcg_out_b(TCGContext *s, int cond, int32_t offset)
363{
364    tcg_out32(s, (cond << 28) | 0x0a000000 |
365                    (((offset - 8) >> 2) & 0x00ffffff));
366}
367
368static inline void tcg_out_bl(TCGContext *s, int cond, int32_t offset)
369{
370    tcg_out32(s, (cond << 28) | 0x0b000000 |
371                    (((offset - 8) >> 2) & 0x00ffffff));
372}
373
374static inline void tcg_out_blx(TCGContext *s, int cond, int rn)
375{
376    tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
377}
378
379static inline void tcg_out_blx_imm(TCGContext *s, int32_t offset)
380{
381    tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
382                (((offset - 8) >> 2) & 0x00ffffff));
383}
384
385static inline void tcg_out_dat_reg(TCGContext *s,
386                int cond, int opc, int rd, int rn, int rm, int shift)
387{
388    tcg_out32(s, (cond << 28) | (0 << 25) | opc |
389                    (rn << 16) | (rd << 12) | shift | rm);
390}
391
392static inline void tcg_out_nop(TCGContext *s)
393{
394    tcg_out32(s, INSN_NOP);
395}
396
397static inline void tcg_out_mov_reg(TCGContext *s, int cond, int rd, int rm)
398{
399    /* Simple reg-reg move, optimising out the 'do nothing' case */
400    if (rd != rm) {
401        tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
402    }
403}
404
405static inline void tcg_out_bx(TCGContext *s, int cond, TCGReg rn)
406{
407    /* Unless the C portion of QEMU is compiled as thumb, we don't
408       actually need true BX semantics; merely a branch to an address
409       held in a register.  */
410    if (use_armv5t_instructions) {
411        tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
412    } else {
413        tcg_out_mov_reg(s, cond, TCG_REG_PC, rn);
414    }
415}
416
417static inline void tcg_out_dat_imm(TCGContext *s,
418                int cond, int opc, int rd, int rn, int im)
419{
420    tcg_out32(s, (cond << 28) | (1 << 25) | opc |
421                    (rn << 16) | (rd << 12) | im);
422}
423
424/* Note that this routine is used for both LDR and LDRH formats, so we do
425   not wish to include an immediate shift at this point.  */
426static void tcg_out_memop_r(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
427                            TCGReg rn, TCGReg rm, bool u, bool p, bool w)
428{
429    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
430              | (w << 21) | (rn << 16) | (rt << 12) | rm);
431}
432
433static void tcg_out_memop_8(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
434                            TCGReg rn, int imm8, bool p, bool w)
435{
436    bool u = 1;
437    if (imm8 < 0) {
438        imm8 = -imm8;
439        u = 0;
440    }
441    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
442              (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
443}
444
445static void tcg_out_memop_12(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
446                             TCGReg rn, int imm12, bool p, bool w)
447{
448    bool u = 1;
449    if (imm12 < 0) {
450        imm12 = -imm12;
451        u = 0;
452    }
453    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
454              (rn << 16) | (rt << 12) | imm12);
455}
456
457static inline void tcg_out_ld32_12(TCGContext *s, int cond, TCGReg rt,
458                                   TCGReg rn, int imm12)
459{
460    tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
461}
462
463static inline void tcg_out_st32_12(TCGContext *s, int cond, TCGReg rt,
464                                   TCGReg rn, int imm12)
465{
466    tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
467}
468
469static inline void tcg_out_ld32_r(TCGContext *s, int cond, TCGReg rt,
470                                  TCGReg rn, TCGReg rm)
471{
472    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
473}
474
475static inline void tcg_out_st32_r(TCGContext *s, int cond, TCGReg rt,
476                                  TCGReg rn, TCGReg rm)
477{
478    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
479}
480
481static inline void tcg_out_ldrd_8(TCGContext *s, int cond, TCGReg rt,
482                                   TCGReg rn, int imm8)
483{
484    tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
485}
486
487static inline void tcg_out_ldrd_r(TCGContext *s, int cond, TCGReg rt,
488                                  TCGReg rn, TCGReg rm)
489{
490    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
491}
492
493static inline void tcg_out_ldrd_rwb(TCGContext *s, int cond, TCGReg rt,
494                                    TCGReg rn, TCGReg rm)
495{
496    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 1);
497}
498
499static inline void tcg_out_strd_8(TCGContext *s, int cond, TCGReg rt,
500                                   TCGReg rn, int imm8)
501{
502    tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
503}
504
505static inline void tcg_out_strd_r(TCGContext *s, int cond, TCGReg rt,
506                                  TCGReg rn, TCGReg rm)
507{
508    tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
509}
510
511/* Register pre-increment with base writeback.  */
512static inline void tcg_out_ld32_rwb(TCGContext *s, int cond, TCGReg rt,
513                                    TCGReg rn, TCGReg rm)
514{
515    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
516}
517
518static inline void tcg_out_st32_rwb(TCGContext *s, int cond, TCGReg rt,
519                                    TCGReg rn, TCGReg rm)
520{
521    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
522}
523
524static inline void tcg_out_ld16u_8(TCGContext *s, int cond, TCGReg rt,
525                                   TCGReg rn, int imm8)
526{
527    tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
528}
529
530static inline void tcg_out_st16_8(TCGContext *s, int cond, TCGReg rt,
531                                  TCGReg rn, int imm8)
532{
533    tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
534}
535
536static inline void tcg_out_ld16u_r(TCGContext *s, int cond, TCGReg rt,
537                                   TCGReg rn, TCGReg rm)
538{
539    tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
540}
541
542static inline void tcg_out_st16_r(TCGContext *s, int cond, TCGReg rt,
543                                  TCGReg rn, TCGReg rm)
544{
545    tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
546}
547
548static inline void tcg_out_ld16s_8(TCGContext *s, int cond, TCGReg rt,
549                                   TCGReg rn, int imm8)
550{
551    tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
552}
553
554static inline void tcg_out_ld16s_r(TCGContext *s, int cond, TCGReg rt,
555                                   TCGReg rn, TCGReg rm)
556{
557    tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
558}
559
560static inline void tcg_out_ld8_12(TCGContext *s, int cond, TCGReg rt,
561                                  TCGReg rn, int imm12)
562{
563    tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
564}
565
566static inline void tcg_out_st8_12(TCGContext *s, int cond, TCGReg rt,
567                                  TCGReg rn, int imm12)
568{
569    tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
570}
571
572static inline void tcg_out_ld8_r(TCGContext *s, int cond, TCGReg rt,
573                                 TCGReg rn, TCGReg rm)
574{
575    tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
576}
577
578static inline void tcg_out_st8_r(TCGContext *s, int cond, TCGReg rt,
579                                 TCGReg rn, TCGReg rm)
580{
581    tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
582}
583
584static inline void tcg_out_ld8s_8(TCGContext *s, int cond, TCGReg rt,
585                                  TCGReg rn, int imm8)
586{
587    tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
588}
589
590static inline void tcg_out_ld8s_r(TCGContext *s, int cond, TCGReg rt,
591                                  TCGReg rn, TCGReg rm)
592{
593    tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
594}
595
596static void tcg_out_movi_pool(TCGContext *s, int cond, int rd, uint32_t arg)
597{
598    new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
599    tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
600}
601
602static void tcg_out_movi32(TCGContext *s, int cond, int rd, uint32_t arg)
603{
604    int rot, diff, opc, sh1, sh2;
605    uint32_t tt0, tt1, tt2;
606
607    /* Check a single MOV/MVN before anything else.  */
608    rot = encode_imm(arg);
609    if (rot >= 0) {
610        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0,
611                        rotl(arg, rot) | (rot << 7));
612        return;
613    }
614    rot = encode_imm(~arg);
615    if (rot >= 0) {
616        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0,
617                        rotl(~arg, rot) | (rot << 7));
618        return;
619    }
620
621    /* Check for a pc-relative address.  This will usually be the TB,
622       or within the TB, which is immediately before the code block.  */
623    diff = tcg_pcrel_diff(s, (void *)arg) - 8;
624    if (diff >= 0) {
625        rot = encode_imm(diff);
626        if (rot >= 0) {
627            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC,
628                            rotl(diff, rot) | (rot << 7));
629            return;
630        }
631    } else {
632        rot = encode_imm(-diff);
633        if (rot >= 0) {
634            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC,
635                            rotl(-diff, rot) | (rot << 7));
636            return;
637        }
638    }
639
640    /* Use movw + movt.  */
641    if (use_armv7_instructions) {
642        /* movw */
643        tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
644                  | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
645        if (arg & 0xffff0000) {
646            /* movt */
647            tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
648                      | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
649        }
650        return;
651    }
652
653    /* Look for sequences of two insns.  If we have lots of 1's, we can
654       shorten the sequence by beginning with mvn and then clearing
655       higher bits with eor.  */
656    tt0 = arg;
657    opc = ARITH_MOV;
658    if (ctpop32(arg) > 16) {
659        tt0 = ~arg;
660        opc = ARITH_MVN;
661    }
662    sh1 = ctz32(tt0) & ~1;
663    tt1 = tt0 & ~(0xff << sh1);
664    sh2 = ctz32(tt1) & ~1;
665    tt2 = tt1 & ~(0xff << sh2);
666    if (tt2 == 0) {
667        rot = ((32 - sh1) << 7) & 0xf00;
668        tcg_out_dat_imm(s, cond, opc, rd,  0, ((tt0 >> sh1) & 0xff) | rot);
669        rot = ((32 - sh2) << 7) & 0xf00;
670        tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
671                        ((tt0 >> sh2) & 0xff) | rot);
672        return;
673    }
674
675    /* Otherwise, drop it into the constant pool.  */
676    tcg_out_movi_pool(s, cond, rd, arg);
677}
678
679static inline void tcg_out_dat_rI(TCGContext *s, int cond, int opc, TCGArg dst,
680                                  TCGArg lhs, TCGArg rhs, int rhs_is_const)
681{
682    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
683     * rhs must satisfy the "rI" constraint.
684     */
685    if (rhs_is_const) {
686        int rot = encode_imm(rhs);
687        tcg_debug_assert(rot >= 0);
688        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
689    } else {
690        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
691    }
692}
693
694static void tcg_out_dat_rIK(TCGContext *s, int cond, int opc, int opinv,
695                            TCGReg dst, TCGReg lhs, TCGArg rhs,
696                            bool rhs_is_const)
697{
698    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
699     * rhs must satisfy the "rIK" constraint.
700     */
701    if (rhs_is_const) {
702        int rot = encode_imm(rhs);
703        if (rot < 0) {
704            rhs = ~rhs;
705            rot = encode_imm(rhs);
706            tcg_debug_assert(rot >= 0);
707            opc = opinv;
708        }
709        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
710    } else {
711        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
712    }
713}
714
715static void tcg_out_dat_rIN(TCGContext *s, int cond, int opc, int opneg,
716                            TCGArg dst, TCGArg lhs, TCGArg rhs,
717                            bool rhs_is_const)
718{
719    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
720     * rhs must satisfy the "rIN" constraint.
721     */
722    if (rhs_is_const) {
723        int rot = encode_imm(rhs);
724        if (rot < 0) {
725            rhs = -rhs;
726            rot = encode_imm(rhs);
727            tcg_debug_assert(rot >= 0);
728            opc = opneg;
729        }
730        tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
731    } else {
732        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
733    }
734}
735
736static inline void tcg_out_mul32(TCGContext *s, int cond, TCGReg rd,
737                                 TCGReg rn, TCGReg rm)
738{
739    /* if ArchVersion() < 6 && d == n then UNPREDICTABLE;  */
740    if (!use_armv6_instructions && rd == rn) {
741        if (rd == rm) {
742            /* rd == rn == rm; copy an input to tmp first.  */
743            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
744            rm = rn = TCG_REG_TMP;
745        } else {
746            rn = rm;
747            rm = rd;
748        }
749    }
750    /* mul */
751    tcg_out32(s, (cond << 28) | 0x90 | (rd << 16) | (rm << 8) | rn);
752}
753
754static inline void tcg_out_umull32(TCGContext *s, int cond, TCGReg rd0,
755                                   TCGReg rd1, TCGReg rn, TCGReg rm)
756{
757    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
758    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
759        if (rd0 == rm || rd1 == rm) {
760            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
761            rn = TCG_REG_TMP;
762        } else {
763            TCGReg t = rn;
764            rn = rm;
765            rm = t;
766        }
767    }
768    /* umull */
769    tcg_out32(s, (cond << 28) | 0x00800090 |
770              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
771}
772
773static inline void tcg_out_smull32(TCGContext *s, int cond, TCGReg rd0,
774                                   TCGReg rd1, TCGReg rn, TCGReg rm)
775{
776    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
777    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
778        if (rd0 == rm || rd1 == rm) {
779            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
780            rn = TCG_REG_TMP;
781        } else {
782            TCGReg t = rn;
783            rn = rm;
784            rm = t;
785        }
786    }
787    /* smull */
788    tcg_out32(s, (cond << 28) | 0x00c00090 |
789              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
790}
791
792static inline void tcg_out_sdiv(TCGContext *s, int cond, int rd, int rn, int rm)
793{
794    tcg_out32(s, 0x0710f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
795}
796
797static inline void tcg_out_udiv(TCGContext *s, int cond, int rd, int rn, int rm)
798{
799    tcg_out32(s, 0x0730f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
800}
801
802static inline void tcg_out_ext8s(TCGContext *s, int cond,
803                                 int rd, int rn)
804{
805    if (use_armv6_instructions) {
806        /* sxtb */
807        tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
808    } else {
809        tcg_out_dat_reg(s, cond, ARITH_MOV,
810                        rd, 0, rn, SHIFT_IMM_LSL(24));
811        tcg_out_dat_reg(s, cond, ARITH_MOV,
812                        rd, 0, rd, SHIFT_IMM_ASR(24));
813    }
814}
815
816static inline void tcg_out_ext8u(TCGContext *s, int cond,
817                                 int rd, int rn)
818{
819    tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
820}
821
822static inline void tcg_out_ext16s(TCGContext *s, int cond,
823                                  int rd, int rn)
824{
825    if (use_armv6_instructions) {
826        /* sxth */
827        tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
828    } else {
829        tcg_out_dat_reg(s, cond, ARITH_MOV,
830                        rd, 0, rn, SHIFT_IMM_LSL(16));
831        tcg_out_dat_reg(s, cond, ARITH_MOV,
832                        rd, 0, rd, SHIFT_IMM_ASR(16));
833    }
834}
835
836static inline void tcg_out_ext16u(TCGContext *s, int cond,
837                                  int rd, int rn)
838{
839    if (use_armv6_instructions) {
840        /* uxth */
841        tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
842    } else {
843        tcg_out_dat_reg(s, cond, ARITH_MOV,
844                        rd, 0, rn, SHIFT_IMM_LSL(16));
845        tcg_out_dat_reg(s, cond, ARITH_MOV,
846                        rd, 0, rd, SHIFT_IMM_LSR(16));
847    }
848}
849
850static inline void tcg_out_bswap16s(TCGContext *s, int cond, int rd, int rn)
851{
852    if (use_armv6_instructions) {
853        /* revsh */
854        tcg_out32(s, 0x06ff0fb0 | (cond << 28) | (rd << 12) | rn);
855    } else {
856        tcg_out_dat_reg(s, cond, ARITH_MOV,
857                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
858        tcg_out_dat_reg(s, cond, ARITH_MOV,
859                        TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_ASR(16));
860        tcg_out_dat_reg(s, cond, ARITH_ORR,
861                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
862    }
863}
864
865static inline void tcg_out_bswap16(TCGContext *s, int cond, int rd, int rn)
866{
867    if (use_armv6_instructions) {
868        /* rev16 */
869        tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
870    } else {
871        tcg_out_dat_reg(s, cond, ARITH_MOV,
872                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
873        tcg_out_dat_reg(s, cond, ARITH_MOV,
874                        TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_LSR(16));
875        tcg_out_dat_reg(s, cond, ARITH_ORR,
876                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
877    }
878}
879
880/* swap the two low bytes assuming that the two high input bytes and the
881   two high output bit can hold any value. */
882static inline void tcg_out_bswap16st(TCGContext *s, int cond, int rd, int rn)
883{
884    if (use_armv6_instructions) {
885        /* rev16 */
886        tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
887    } else {
888        tcg_out_dat_reg(s, cond, ARITH_MOV,
889                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSR(8));
890        tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_TMP, TCG_REG_TMP, 0xff);
891        tcg_out_dat_reg(s, cond, ARITH_ORR,
892                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSL(8));
893    }
894}
895
896static inline void tcg_out_bswap32(TCGContext *s, int cond, int rd, int rn)
897{
898    if (use_armv6_instructions) {
899        /* rev */
900        tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
901    } else {
902        tcg_out_dat_reg(s, cond, ARITH_EOR,
903                        TCG_REG_TMP, rn, rn, SHIFT_IMM_ROR(16));
904        tcg_out_dat_imm(s, cond, ARITH_BIC,
905                        TCG_REG_TMP, TCG_REG_TMP, 0xff | 0x800);
906        tcg_out_dat_reg(s, cond, ARITH_MOV,
907                        rd, 0, rn, SHIFT_IMM_ROR(8));
908        tcg_out_dat_reg(s, cond, ARITH_EOR,
909                        rd, rd, TCG_REG_TMP, SHIFT_IMM_LSR(8));
910    }
911}
912
913static inline void tcg_out_deposit(TCGContext *s, int cond, TCGReg rd,
914                                   TCGArg a1, int ofs, int len, bool const_a1)
915{
916    if (const_a1) {
917        /* bfi becomes bfc with rn == 15.  */
918        a1 = 15;
919    }
920    /* bfi/bfc */
921    tcg_out32(s, 0x07c00010 | (cond << 28) | (rd << 12) | a1
922              | (ofs << 7) | ((ofs + len - 1) << 16));
923}
924
925static inline void tcg_out_extract(TCGContext *s, int cond, TCGReg rd,
926                                   TCGArg a1, int ofs, int len)
927{
928    /* ubfx */
929    tcg_out32(s, 0x07e00050 | (cond << 28) | (rd << 12) | a1
930              | (ofs << 7) | ((len - 1) << 16));
931}
932
933static inline void tcg_out_sextract(TCGContext *s, int cond, TCGReg rd,
934                                    TCGArg a1, int ofs, int len)
935{
936    /* sbfx */
937    tcg_out32(s, 0x07a00050 | (cond << 28) | (rd << 12) | a1
938              | (ofs << 7) | ((len - 1) << 16));
939}
940
941static inline void tcg_out_ld32u(TCGContext *s, int cond,
942                int rd, int rn, int32_t offset)
943{
944    if (offset > 0xfff || offset < -0xfff) {
945        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
946        tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
947    } else
948        tcg_out_ld32_12(s, cond, rd, rn, offset);
949}
950
951static inline void tcg_out_st32(TCGContext *s, int cond,
952                int rd, int rn, int32_t offset)
953{
954    if (offset > 0xfff || offset < -0xfff) {
955        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
956        tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
957    } else
958        tcg_out_st32_12(s, cond, rd, rn, offset);
959}
960
961static inline void tcg_out_ld16u(TCGContext *s, int cond,
962                int rd, int rn, int32_t offset)
963{
964    if (offset > 0xff || offset < -0xff) {
965        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
966        tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_TMP);
967    } else
968        tcg_out_ld16u_8(s, cond, rd, rn, offset);
969}
970
971static inline void tcg_out_ld16s(TCGContext *s, int cond,
972                int rd, int rn, int32_t offset)
973{
974    if (offset > 0xff || offset < -0xff) {
975        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
976        tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_TMP);
977    } else
978        tcg_out_ld16s_8(s, cond, rd, rn, offset);
979}
980
981static inline void tcg_out_st16(TCGContext *s, int cond,
982                int rd, int rn, int32_t offset)
983{
984    if (offset > 0xff || offset < -0xff) {
985        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
986        tcg_out_st16_r(s, cond, rd, rn, TCG_REG_TMP);
987    } else
988        tcg_out_st16_8(s, cond, rd, rn, offset);
989}
990
991static inline void tcg_out_ld8u(TCGContext *s, int cond,
992                int rd, int rn, int32_t offset)
993{
994    if (offset > 0xfff || offset < -0xfff) {
995        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
996        tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_TMP);
997    } else
998        tcg_out_ld8_12(s, cond, rd, rn, offset);
999}
1000
1001static inline void tcg_out_ld8s(TCGContext *s, int cond,
1002                int rd, int rn, int32_t offset)
1003{
1004    if (offset > 0xff || offset < -0xff) {
1005        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1006        tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_TMP);
1007    } else
1008        tcg_out_ld8s_8(s, cond, rd, rn, offset);
1009}
1010
1011static inline void tcg_out_st8(TCGContext *s, int cond,
1012                int rd, int rn, int32_t offset)
1013{
1014    if (offset > 0xfff || offset < -0xfff) {
1015        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1016        tcg_out_st8_r(s, cond, rd, rn, TCG_REG_TMP);
1017    } else
1018        tcg_out_st8_12(s, cond, rd, rn, offset);
1019}
1020
1021/* The _goto case is normally between TBs within the same code buffer, and
1022 * with the code buffer limited to 16MB we wouldn't need the long case.
1023 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1024 */
1025static void tcg_out_goto(TCGContext *s, int cond, const tcg_insn_unit *addr)
1026{
1027    intptr_t addri = (intptr_t)addr;
1028    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1029
1030    if ((addri & 1) == 0 && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1031        tcg_out_b(s, cond, disp);
1032        return;
1033    }
1034    tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1035}
1036
1037/* The call case is mostly used for helpers - so it's not unreasonable
1038 * for them to be beyond branch range */
1039static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr)
1040{
1041    intptr_t addri = (intptr_t)addr;
1042    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1043
1044    if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1045        if (addri & 1) {
1046            /* Use BLX if the target is in Thumb mode */
1047            if (!use_armv5t_instructions) {
1048                tcg_abort();
1049            }
1050            tcg_out_blx_imm(s, disp);
1051        } else {
1052            tcg_out_bl(s, COND_AL, disp);
1053        }
1054    } else if (use_armv7_instructions) {
1055        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1056        tcg_out_blx(s, COND_AL, TCG_REG_TMP);
1057    } else {
1058        /* ??? Know that movi_pool emits exactly 1 insn.  */
1059        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 0);
1060        tcg_out_movi_pool(s, COND_AL, TCG_REG_PC, addri);
1061    }
1062}
1063
1064static inline void tcg_out_goto_label(TCGContext *s, int cond, TCGLabel *l)
1065{
1066    if (l->has_value) {
1067        tcg_out_goto(s, cond, l->u.value_ptr);
1068    } else {
1069        tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1070        tcg_out_b(s, cond, 0);
1071    }
1072}
1073
1074static inline void tcg_out_mb(TCGContext *s, TCGArg a0)
1075{
1076    if (use_armv7_instructions) {
1077        tcg_out32(s, INSN_DMB_ISH);
1078    } else if (use_armv6_instructions) {
1079        tcg_out32(s, INSN_DMB_MCR);
1080    }
1081}
1082
1083static TCGCond tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1084                            const int *const_args)
1085{
1086    TCGReg al = args[0];
1087    TCGReg ah = args[1];
1088    TCGArg bl = args[2];
1089    TCGArg bh = args[3];
1090    TCGCond cond = args[4];
1091    int const_bl = const_args[2];
1092    int const_bh = const_args[3];
1093
1094    switch (cond) {
1095    case TCG_COND_EQ:
1096    case TCG_COND_NE:
1097    case TCG_COND_LTU:
1098    case TCG_COND_LEU:
1099    case TCG_COND_GTU:
1100    case TCG_COND_GEU:
1101        /* We perform a conditional comparision.  If the high half is
1102           equal, then overwrite the flags with the comparison of the
1103           low half.  The resulting flags cover the whole.  */
1104        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, ah, bh, const_bh);
1105        tcg_out_dat_rI(s, COND_EQ, ARITH_CMP, 0, al, bl, const_bl);
1106        return cond;
1107
1108    case TCG_COND_LT:
1109    case TCG_COND_GE:
1110        /* We perform a double-word subtraction and examine the result.
1111           We do not actually need the result of the subtract, so the
1112           low part "subtract" is a compare.  For the high half we have
1113           no choice but to compute into a temporary.  */
1114        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, al, bl, const_bl);
1115        tcg_out_dat_rI(s, COND_AL, ARITH_SBC | TO_CPSR,
1116                       TCG_REG_TMP, ah, bh, const_bh);
1117        return cond;
1118
1119    case TCG_COND_LE:
1120    case TCG_COND_GT:
1121        /* Similar, but with swapped arguments, via reversed subtract.  */
1122        tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR,
1123                       TCG_REG_TMP, al, bl, const_bl);
1124        tcg_out_dat_rI(s, COND_AL, ARITH_RSC | TO_CPSR,
1125                       TCG_REG_TMP, ah, bh, const_bh);
1126        return tcg_swap_cond(cond);
1127
1128    default:
1129        g_assert_not_reached();
1130    }
1131}
1132
1133#ifdef CONFIG_SOFTMMU
1134#include "../tcg-ldst.c.inc"
1135
1136/* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
1137 *                                     int mmu_idx, uintptr_t ra)
1138 */
1139static void * const qemu_ld_helpers[16] = {
1140    [MO_UB]   = helper_ret_ldub_mmu,
1141    [MO_SB]   = helper_ret_ldsb_mmu,
1142
1143    [MO_LEUW] = helper_le_lduw_mmu,
1144    [MO_LEUL] = helper_le_ldul_mmu,
1145    [MO_LEQ]  = helper_le_ldq_mmu,
1146    [MO_LESW] = helper_le_ldsw_mmu,
1147    [MO_LESL] = helper_le_ldul_mmu,
1148
1149    [MO_BEUW] = helper_be_lduw_mmu,
1150    [MO_BEUL] = helper_be_ldul_mmu,
1151    [MO_BEQ]  = helper_be_ldq_mmu,
1152    [MO_BESW] = helper_be_ldsw_mmu,
1153    [MO_BESL] = helper_be_ldul_mmu,
1154};
1155
1156/* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
1157 *                                     uintxx_t val, int mmu_idx, uintptr_t ra)
1158 */
1159static void * const qemu_st_helpers[16] = {
1160    [MO_UB]   = helper_ret_stb_mmu,
1161    [MO_LEUW] = helper_le_stw_mmu,
1162    [MO_LEUL] = helper_le_stl_mmu,
1163    [MO_LEQ]  = helper_le_stq_mmu,
1164    [MO_BEUW] = helper_be_stw_mmu,
1165    [MO_BEUL] = helper_be_stl_mmu,
1166    [MO_BEQ]  = helper_be_stq_mmu,
1167};
1168
1169/* Helper routines for marshalling helper function arguments into
1170 * the correct registers and stack.
1171 * argreg is where we want to put this argument, arg is the argument itself.
1172 * Return value is the updated argreg ready for the next call.
1173 * Note that argreg 0..3 is real registers, 4+ on stack.
1174 *
1175 * We provide routines for arguments which are: immediate, 32 bit
1176 * value in register, 16 and 8 bit values in register (which must be zero
1177 * extended before use) and 64 bit value in a lo:hi register pair.
1178 */
1179#define DEFINE_TCG_OUT_ARG(NAME, ARGTYPE, MOV_ARG, EXT_ARG)                \
1180static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGTYPE arg)              \
1181{                                                                          \
1182    if (argreg < 4) {                                                      \
1183        MOV_ARG(s, COND_AL, argreg, arg);                                  \
1184    } else {                                                               \
1185        int ofs = (argreg - 4) * 4;                                        \
1186        EXT_ARG;                                                           \
1187        tcg_debug_assert(ofs + 4 <= TCG_STATIC_CALL_ARGS_SIZE);            \
1188        tcg_out_st32_12(s, COND_AL, arg, TCG_REG_CALL_STACK, ofs);         \
1189    }                                                                      \
1190    return argreg + 1;                                                     \
1191}
1192
1193DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t, tcg_out_movi32,
1194    (tcg_out_movi32(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1195DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg, tcg_out_ext8u,
1196    (tcg_out_ext8u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1197DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg, tcg_out_ext16u,
1198    (tcg_out_ext16u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1199DEFINE_TCG_OUT_ARG(tcg_out_arg_reg32, TCGReg, tcg_out_mov_reg, )
1200
1201static TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1202                                TCGReg arglo, TCGReg arghi)
1203{
1204    /* 64 bit arguments must go in even/odd register pairs
1205     * and in 8-aligned stack slots.
1206     */
1207    if (argreg & 1) {
1208        argreg++;
1209    }
1210    if (use_armv6_instructions && argreg >= 4
1211        && (arglo & 1) == 0 && arghi == arglo + 1) {
1212        tcg_out_strd_8(s, COND_AL, arglo,
1213                       TCG_REG_CALL_STACK, (argreg - 4) * 4);
1214        return argreg + 2;
1215    } else {
1216        argreg = tcg_out_arg_reg32(s, argreg, arglo);
1217        argreg = tcg_out_arg_reg32(s, argreg, arghi);
1218        return argreg;
1219    }
1220}
1221
1222#define TLB_SHIFT	(CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1223
1224/* We expect to use an 9-bit sign-magnitude negative offset from ENV.  */
1225QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1226QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -256);
1227
1228/* These offsets are built into the LDRD below.  */
1229QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1230QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 4);
1231
1232/* Load and compare a TLB entry, leaving the flags set.  Returns the register
1233   containing the addend of the tlb entry.  Clobbers R0, R1, R2, TMP.  */
1234
1235static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
1236                               MemOp opc, int mem_index, bool is_load)
1237{
1238    int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read)
1239                   : offsetof(CPUTLBEntry, addr_write));
1240    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1241    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1242    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1243    unsigned s_bits = opc & MO_SIZE;
1244    unsigned a_bits = get_alignment_bits(opc);
1245
1246    /*
1247     * We don't support inline unaligned acceses, but we can easily
1248     * support overalignment checks.
1249     */
1250    if (a_bits < s_bits) {
1251        a_bits = s_bits;
1252    }
1253
1254    /* Load env_tlb(env)->f[mmu_idx].{mask,table} into {r0,r1}.  */
1255    if (use_armv6_instructions) {
1256        tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_AREG0, fast_off);
1257    } else {
1258        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R0, TCG_AREG0, mask_off);
1259        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R1, TCG_AREG0, table_off);
1260    }
1261
1262    /* Extract the tlb index from the address into R0.  */
1263    tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_R0, TCG_REG_R0, addrlo,
1264                    SHIFT_IMM_LSR(TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS));
1265
1266    /*
1267     * Add the tlb_table pointer, creating the CPUTLBEntry address in R1.
1268     * Load the tlb comparator into R2/R3 and the fast path addend into R1.
1269     */
1270    if (cmp_off == 0) {
1271        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1272            tcg_out_ldrd_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1273        } else {
1274            tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1275        }
1276    } else {
1277        tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1278                        TCG_REG_R1, TCG_REG_R1, TCG_REG_R0, 0);
1279        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1280            tcg_out_ldrd_8(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1281        } else {
1282            tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1283        }
1284    }
1285    if (!use_armv6_instructions && TARGET_LONG_BITS == 64) {
1286        tcg_out_ld32_12(s, COND_AL, TCG_REG_R3, TCG_REG_R1, cmp_off + 4);
1287    }
1288
1289    /* Load the tlb addend.  */
1290    tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R1,
1291                    offsetof(CPUTLBEntry, addend));
1292
1293    /*
1294     * Check alignment, check comparators.
1295     * Do this in no more than 3 insns.  Use MOVW for v7, if possible,
1296     * to reduce the number of sequential conditional instructions.
1297     * Almost all guests have at least 4k pages, which means that we need
1298     * to clear at least 9 bits even for an 8-byte memory, which means it
1299     * isn't worth checking for an immediate operand for BIC.
1300     */
1301    if (use_armv7_instructions && TARGET_PAGE_BITS <= 16) {
1302        tcg_target_ulong mask = ~(TARGET_PAGE_MASK | ((1 << a_bits) - 1));
1303
1304        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, mask);
1305        tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1306                        addrlo, TCG_REG_TMP, 0);
1307        tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R2, TCG_REG_TMP, 0);
1308    } else {
1309        if (a_bits) {
1310            tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addrlo,
1311                            (1 << a_bits) - 1);
1312        }
1313        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0, addrlo,
1314                        SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1315        tcg_out_dat_reg(s, (a_bits ? COND_EQ : COND_AL), ARITH_CMP,
1316                        0, TCG_REG_R2, TCG_REG_TMP,
1317                        SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1318    }
1319
1320    if (TARGET_LONG_BITS == 64) {
1321        tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0, TCG_REG_R3, addrhi, 0);
1322    }
1323
1324    return TCG_REG_R1;
1325}
1326
1327/* Record the context of a call to the out of line helper code for the slow
1328   path for a load or store, so that we can later generate the correct
1329   helper code.  */
1330static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1331                                TCGReg datalo, TCGReg datahi, TCGReg addrlo,
1332                                TCGReg addrhi, tcg_insn_unit *raddr,
1333                                tcg_insn_unit *label_ptr)
1334{
1335    TCGLabelQemuLdst *label = new_ldst_label(s);
1336
1337    label->is_ld = is_ld;
1338    label->oi = oi;
1339    label->datalo_reg = datalo;
1340    label->datahi_reg = datahi;
1341    label->addrlo_reg = addrlo;
1342    label->addrhi_reg = addrhi;
1343    label->raddr = tcg_splitwx_to_rx(raddr);
1344    label->label_ptr[0] = label_ptr;
1345}
1346
1347static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1348{
1349    TCGReg argreg, datalo, datahi;
1350    TCGMemOpIdx oi = lb->oi;
1351    MemOp opc = get_memop(oi);
1352    void *func;
1353
1354    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1355        return false;
1356    }
1357
1358    argreg = tcg_out_arg_reg32(s, TCG_REG_R0, TCG_AREG0);
1359    if (TARGET_LONG_BITS == 64) {
1360        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1361    } else {
1362        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1363    }
1364    argreg = tcg_out_arg_imm32(s, argreg, oi);
1365    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1366
1367    /* For armv6 we can use the canonical unsigned helpers and minimize
1368       icache usage.  For pre-armv6, use the signed helpers since we do
1369       not have a single insn sign-extend.  */
1370    if (use_armv6_instructions) {
1371        func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)];
1372    } else {
1373        func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)];
1374        if (opc & MO_SIGN) {
1375            opc = MO_UL;
1376        }
1377    }
1378    tcg_out_call(s, func);
1379
1380    datalo = lb->datalo_reg;
1381    datahi = lb->datahi_reg;
1382    switch (opc & MO_SSIZE) {
1383    case MO_SB:
1384        tcg_out_ext8s(s, COND_AL, datalo, TCG_REG_R0);
1385        break;
1386    case MO_SW:
1387        tcg_out_ext16s(s, COND_AL, datalo, TCG_REG_R0);
1388        break;
1389    default:
1390        tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1391        break;
1392    case MO_Q:
1393        if (datalo != TCG_REG_R1) {
1394            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1395            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1396        } else if (datahi != TCG_REG_R0) {
1397            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1398            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1399        } else {
1400            tcg_out_mov_reg(s, COND_AL, TCG_REG_TMP, TCG_REG_R0);
1401            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1402            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_TMP);
1403        }
1404        break;
1405    }
1406
1407    tcg_out_goto(s, COND_AL, lb->raddr);
1408    return true;
1409}
1410
1411static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1412{
1413    TCGReg argreg, datalo, datahi;
1414    TCGMemOpIdx oi = lb->oi;
1415    MemOp opc = get_memop(oi);
1416
1417    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1418        return false;
1419    }
1420
1421    argreg = TCG_REG_R0;
1422    argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1423    if (TARGET_LONG_BITS == 64) {
1424        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1425    } else {
1426        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1427    }
1428
1429    datalo = lb->datalo_reg;
1430    datahi = lb->datahi_reg;
1431    switch (opc & MO_SIZE) {
1432    case MO_8:
1433        argreg = tcg_out_arg_reg8(s, argreg, datalo);
1434        break;
1435    case MO_16:
1436        argreg = tcg_out_arg_reg16(s, argreg, datalo);
1437        break;
1438    case MO_32:
1439    default:
1440        argreg = tcg_out_arg_reg32(s, argreg, datalo);
1441        break;
1442    case MO_64:
1443        argreg = tcg_out_arg_reg64(s, argreg, datalo, datahi);
1444        break;
1445    }
1446
1447    argreg = tcg_out_arg_imm32(s, argreg, oi);
1448    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1449
1450    /* Tail-call to the helper, which will return to the fast path.  */
1451    tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1452    return true;
1453}
1454#endif /* SOFTMMU */
1455
1456static inline void tcg_out_qemu_ld_index(TCGContext *s, MemOp opc,
1457                                         TCGReg datalo, TCGReg datahi,
1458                                         TCGReg addrlo, TCGReg addend)
1459{
1460    MemOp bswap = opc & MO_BSWAP;
1461
1462    switch (opc & MO_SSIZE) {
1463    case MO_UB:
1464        tcg_out_ld8_r(s, COND_AL, datalo, addrlo, addend);
1465        break;
1466    case MO_SB:
1467        tcg_out_ld8s_r(s, COND_AL, datalo, addrlo, addend);
1468        break;
1469    case MO_UW:
1470        tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1471        if (bswap) {
1472            tcg_out_bswap16(s, COND_AL, datalo, datalo);
1473        }
1474        break;
1475    case MO_SW:
1476        if (bswap) {
1477            tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1478            tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1479        } else {
1480            tcg_out_ld16s_r(s, COND_AL, datalo, addrlo, addend);
1481        }
1482        break;
1483    case MO_UL:
1484    default:
1485        tcg_out_ld32_r(s, COND_AL, datalo, addrlo, addend);
1486        if (bswap) {
1487            tcg_out_bswap32(s, COND_AL, datalo, datalo);
1488        }
1489        break;
1490    case MO_Q:
1491        {
1492            TCGReg dl = (bswap ? datahi : datalo);
1493            TCGReg dh = (bswap ? datalo : datahi);
1494
1495            /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1496            if (USING_SOFTMMU && use_armv6_instructions
1497                && (dl & 1) == 0 && dh == dl + 1) {
1498                tcg_out_ldrd_r(s, COND_AL, dl, addrlo, addend);
1499            } else if (dl != addend) {
1500                tcg_out_ld32_rwb(s, COND_AL, dl, addend, addrlo);
1501                tcg_out_ld32_12(s, COND_AL, dh, addend, 4);
1502            } else {
1503                tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_TMP,
1504                                addend, addrlo, SHIFT_IMM_LSL(0));
1505                tcg_out_ld32_12(s, COND_AL, dl, TCG_REG_TMP, 0);
1506                tcg_out_ld32_12(s, COND_AL, dh, TCG_REG_TMP, 4);
1507            }
1508            if (bswap) {
1509                tcg_out_bswap32(s, COND_AL, dl, dl);
1510                tcg_out_bswap32(s, COND_AL, dh, dh);
1511            }
1512        }
1513        break;
1514    }
1515}
1516
1517static inline void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc,
1518                                          TCGReg datalo, TCGReg datahi,
1519                                          TCGReg addrlo)
1520{
1521    MemOp bswap = opc & MO_BSWAP;
1522
1523    switch (opc & MO_SSIZE) {
1524    case MO_UB:
1525        tcg_out_ld8_12(s, COND_AL, datalo, addrlo, 0);
1526        break;
1527    case MO_SB:
1528        tcg_out_ld8s_8(s, COND_AL, datalo, addrlo, 0);
1529        break;
1530    case MO_UW:
1531        tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1532        if (bswap) {
1533            tcg_out_bswap16(s, COND_AL, datalo, datalo);
1534        }
1535        break;
1536    case MO_SW:
1537        if (bswap) {
1538            tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1539            tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1540        } else {
1541            tcg_out_ld16s_8(s, COND_AL, datalo, addrlo, 0);
1542        }
1543        break;
1544    case MO_UL:
1545    default:
1546        tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1547        if (bswap) {
1548            tcg_out_bswap32(s, COND_AL, datalo, datalo);
1549        }
1550        break;
1551    case MO_Q:
1552        {
1553            TCGReg dl = (bswap ? datahi : datalo);
1554            TCGReg dh = (bswap ? datalo : datahi);
1555
1556            /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1557            if (USING_SOFTMMU && use_armv6_instructions
1558                && (dl & 1) == 0 && dh == dl + 1) {
1559                tcg_out_ldrd_8(s, COND_AL, dl, addrlo, 0);
1560            } else if (dl == addrlo) {
1561                tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1562                tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1563            } else {
1564                tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1565                tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1566            }
1567            if (bswap) {
1568                tcg_out_bswap32(s, COND_AL, dl, dl);
1569                tcg_out_bswap32(s, COND_AL, dh, dh);
1570            }
1571        }
1572        break;
1573    }
1574}
1575
1576static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is64)
1577{
1578    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1579    TCGMemOpIdx oi;
1580    MemOp opc;
1581#ifdef CONFIG_SOFTMMU
1582    int mem_index;
1583    TCGReg addend;
1584    tcg_insn_unit *label_ptr;
1585#endif
1586
1587    datalo = *args++;
1588    datahi = (is64 ? *args++ : 0);
1589    addrlo = *args++;
1590    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1591    oi = *args++;
1592    opc = get_memop(oi);
1593
1594#ifdef CONFIG_SOFTMMU
1595    mem_index = get_mmuidx(oi);
1596    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 1);
1597
1598    /* This a conditional BL only to load a pointer within this opcode into LR
1599       for the slow path.  We will not be using the value for a tail call.  */
1600    label_ptr = s->code_ptr;
1601    tcg_out_bl(s, COND_NE, 0);
1602
1603    tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, addend);
1604
1605    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1606                        s->code_ptr, label_ptr);
1607#else /* !CONFIG_SOFTMMU */
1608    if (guest_base) {
1609        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1610        tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, TCG_REG_TMP);
1611    } else {
1612        tcg_out_qemu_ld_direct(s, opc, datalo, datahi, addrlo);
1613    }
1614#endif
1615}
1616
1617static inline void tcg_out_qemu_st_index(TCGContext *s, int cond, MemOp opc,
1618                                         TCGReg datalo, TCGReg datahi,
1619                                         TCGReg addrlo, TCGReg addend)
1620{
1621    MemOp bswap = opc & MO_BSWAP;
1622
1623    switch (opc & MO_SIZE) {
1624    case MO_8:
1625        tcg_out_st8_r(s, cond, datalo, addrlo, addend);
1626        break;
1627    case MO_16:
1628        if (bswap) {
1629            tcg_out_bswap16st(s, cond, TCG_REG_R0, datalo);
1630            tcg_out_st16_r(s, cond, TCG_REG_R0, addrlo, addend);
1631        } else {
1632            tcg_out_st16_r(s, cond, datalo, addrlo, addend);
1633        }
1634        break;
1635    case MO_32:
1636    default:
1637        if (bswap) {
1638            tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1639            tcg_out_st32_r(s, cond, TCG_REG_R0, addrlo, addend);
1640        } else {
1641            tcg_out_st32_r(s, cond, datalo, addrlo, addend);
1642        }
1643        break;
1644    case MO_64:
1645        /* Avoid strd for user-only emulation, to handle unaligned.  */
1646        if (bswap) {
1647            tcg_out_bswap32(s, cond, TCG_REG_R0, datahi);
1648            tcg_out_st32_rwb(s, cond, TCG_REG_R0, addend, addrlo);
1649            tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1650            tcg_out_st32_12(s, cond, TCG_REG_R0, addend, 4);
1651        } else if (USING_SOFTMMU && use_armv6_instructions
1652                   && (datalo & 1) == 0 && datahi == datalo + 1) {
1653            tcg_out_strd_r(s, cond, datalo, addrlo, addend);
1654        } else {
1655            tcg_out_st32_rwb(s, cond, datalo, addend, addrlo);
1656            tcg_out_st32_12(s, cond, datahi, addend, 4);
1657        }
1658        break;
1659    }
1660}
1661
1662static inline void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc,
1663                                          TCGReg datalo, TCGReg datahi,
1664                                          TCGReg addrlo)
1665{
1666    MemOp bswap = opc & MO_BSWAP;
1667
1668    switch (opc & MO_SIZE) {
1669    case MO_8:
1670        tcg_out_st8_12(s, COND_AL, datalo, addrlo, 0);
1671        break;
1672    case MO_16:
1673        if (bswap) {
1674            tcg_out_bswap16st(s, COND_AL, TCG_REG_R0, datalo);
1675            tcg_out_st16_8(s, COND_AL, TCG_REG_R0, addrlo, 0);
1676        } else {
1677            tcg_out_st16_8(s, COND_AL, datalo, addrlo, 0);
1678        }
1679        break;
1680    case MO_32:
1681    default:
1682        if (bswap) {
1683            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1684            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1685        } else {
1686            tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1687        }
1688        break;
1689    case MO_64:
1690        /* Avoid strd for user-only emulation, to handle unaligned.  */
1691        if (bswap) {
1692            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datahi);
1693            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1694            tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1695            tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 4);
1696        } else if (USING_SOFTMMU && use_armv6_instructions
1697                   && (datalo & 1) == 0 && datahi == datalo + 1) {
1698            tcg_out_strd_8(s, COND_AL, datalo, addrlo, 0);
1699        } else {
1700            tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1701            tcg_out_st32_12(s, COND_AL, datahi, addrlo, 4);
1702        }
1703        break;
1704    }
1705}
1706
1707static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is64)
1708{
1709    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1710    TCGMemOpIdx oi;
1711    MemOp opc;
1712#ifdef CONFIG_SOFTMMU
1713    int mem_index;
1714    TCGReg addend;
1715    tcg_insn_unit *label_ptr;
1716#endif
1717
1718    datalo = *args++;
1719    datahi = (is64 ? *args++ : 0);
1720    addrlo = *args++;
1721    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1722    oi = *args++;
1723    opc = get_memop(oi);
1724
1725#ifdef CONFIG_SOFTMMU
1726    mem_index = get_mmuidx(oi);
1727    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 0);
1728
1729    tcg_out_qemu_st_index(s, COND_EQ, opc, datalo, datahi, addrlo, addend);
1730
1731    /* The conditional call must come last, as we're going to return here.  */
1732    label_ptr = s->code_ptr;
1733    tcg_out_bl(s, COND_NE, 0);
1734
1735    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1736                        s->code_ptr, label_ptr);
1737#else /* !CONFIG_SOFTMMU */
1738    if (guest_base) {
1739        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1740        tcg_out_qemu_st_index(s, COND_AL, opc, datalo,
1741                              datahi, addrlo, TCG_REG_TMP);
1742    } else {
1743        tcg_out_qemu_st_direct(s, opc, datalo, datahi, addrlo);
1744    }
1745#endif
1746}
1747
1748static void tcg_out_epilogue(TCGContext *s);
1749
1750static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1751                const TCGArg *args, const int *const_args)
1752{
1753    TCGArg a0, a1, a2, a3, a4, a5;
1754    int c;
1755
1756    switch (opc) {
1757    case INDEX_op_exit_tb:
1758        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, args[0]);
1759        tcg_out_epilogue(s);
1760        break;
1761    case INDEX_op_goto_tb:
1762        {
1763            /* Indirect jump method */
1764            intptr_t ptr, dif, dil;
1765            TCGReg base = TCG_REG_PC;
1766
1767            tcg_debug_assert(s->tb_jmp_insn_offset == 0);
1768            ptr = (intptr_t)tcg_splitwx_to_rx(s->tb_jmp_target_addr + args[0]);
1769            dif = tcg_pcrel_diff(s, (void *)ptr) - 8;
1770            dil = sextract32(dif, 0, 12);
1771            if (dif != dil) {
1772                /* The TB is close, but outside the 12 bits addressable by
1773                   the load.  We can extend this to 20 bits with a sub of a
1774                   shifted immediate from pc.  In the vastly unlikely event
1775                   the code requires more than 1MB, we'll use 2 insns and
1776                   be no worse off.  */
1777                base = TCG_REG_R0;
1778                tcg_out_movi32(s, COND_AL, base, ptr - dil);
1779            }
1780            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, base, dil);
1781            set_jmp_reset_offset(s, args[0]);
1782        }
1783        break;
1784    case INDEX_op_goto_ptr:
1785        tcg_out_bx(s, COND_AL, args[0]);
1786        break;
1787    case INDEX_op_br:
1788        tcg_out_goto_label(s, COND_AL, arg_label(args[0]));
1789        break;
1790
1791    case INDEX_op_ld8u_i32:
1792        tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
1793        break;
1794    case INDEX_op_ld8s_i32:
1795        tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
1796        break;
1797    case INDEX_op_ld16u_i32:
1798        tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
1799        break;
1800    case INDEX_op_ld16s_i32:
1801        tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
1802        break;
1803    case INDEX_op_ld_i32:
1804        tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
1805        break;
1806    case INDEX_op_st8_i32:
1807        tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
1808        break;
1809    case INDEX_op_st16_i32:
1810        tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
1811        break;
1812    case INDEX_op_st_i32:
1813        tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
1814        break;
1815
1816    case INDEX_op_movcond_i32:
1817        /* Constraints mean that v2 is always in the same register as dest,
1818         * so we only need to do "if condition passed, move v1 to dest".
1819         */
1820        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1821                        args[1], args[2], const_args[2]);
1822        tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[args[5]], ARITH_MOV,
1823                        ARITH_MVN, args[0], 0, args[3], const_args[3]);
1824        break;
1825    case INDEX_op_add_i32:
1826        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1827                        args[0], args[1], args[2], const_args[2]);
1828        break;
1829    case INDEX_op_sub_i32:
1830        if (const_args[1]) {
1831            if (const_args[2]) {
1832                tcg_out_movi32(s, COND_AL, args[0], args[1] - args[2]);
1833            } else {
1834                tcg_out_dat_rI(s, COND_AL, ARITH_RSB,
1835                               args[0], args[2], args[1], 1);
1836            }
1837        } else {
1838            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB, ARITH_ADD,
1839                            args[0], args[1], args[2], const_args[2]);
1840        }
1841        break;
1842    case INDEX_op_and_i32:
1843        tcg_out_dat_rIK(s, COND_AL, ARITH_AND, ARITH_BIC,
1844                        args[0], args[1], args[2], const_args[2]);
1845        break;
1846    case INDEX_op_andc_i32:
1847        tcg_out_dat_rIK(s, COND_AL, ARITH_BIC, ARITH_AND,
1848                        args[0], args[1], args[2], const_args[2]);
1849        break;
1850    case INDEX_op_or_i32:
1851        c = ARITH_ORR;
1852        goto gen_arith;
1853    case INDEX_op_xor_i32:
1854        c = ARITH_EOR;
1855        /* Fall through.  */
1856    gen_arith:
1857        tcg_out_dat_rI(s, COND_AL, c, args[0], args[1], args[2], const_args[2]);
1858        break;
1859    case INDEX_op_add2_i32:
1860        a0 = args[0], a1 = args[1], a2 = args[2];
1861        a3 = args[3], a4 = args[4], a5 = args[5];
1862        if (a0 == a3 || (a0 == a5 && !const_args[5])) {
1863            a0 = TCG_REG_TMP;
1864        }
1865        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
1866                        a0, a2, a4, const_args[4]);
1867        tcg_out_dat_rIK(s, COND_AL, ARITH_ADC, ARITH_SBC,
1868                        a1, a3, a5, const_args[5]);
1869        tcg_out_mov_reg(s, COND_AL, args[0], a0);
1870        break;
1871    case INDEX_op_sub2_i32:
1872        a0 = args[0], a1 = args[1], a2 = args[2];
1873        a3 = args[3], a4 = args[4], a5 = args[5];
1874        if ((a0 == a3 && !const_args[3]) || (a0 == a5 && !const_args[5])) {
1875            a0 = TCG_REG_TMP;
1876        }
1877        if (const_args[2]) {
1878            if (const_args[4]) {
1879                tcg_out_movi32(s, COND_AL, a0, a4);
1880                a4 = a0;
1881            }
1882            tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR, a0, a4, a2, 1);
1883        } else {
1884            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB | TO_CPSR,
1885                            ARITH_ADD | TO_CPSR, a0, a2, a4, const_args[4]);
1886        }
1887        if (const_args[3]) {
1888            if (const_args[5]) {
1889                tcg_out_movi32(s, COND_AL, a1, a5);
1890                a5 = a1;
1891            }
1892            tcg_out_dat_rI(s, COND_AL, ARITH_RSC, a1, a5, a3, 1);
1893        } else {
1894            tcg_out_dat_rIK(s, COND_AL, ARITH_SBC, ARITH_ADC,
1895                            a1, a3, a5, const_args[5]);
1896        }
1897        tcg_out_mov_reg(s, COND_AL, args[0], a0);
1898        break;
1899    case INDEX_op_neg_i32:
1900        tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
1901        break;
1902    case INDEX_op_not_i32:
1903        tcg_out_dat_reg(s, COND_AL,
1904                        ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
1905        break;
1906    case INDEX_op_mul_i32:
1907        tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
1908        break;
1909    case INDEX_op_mulu2_i32:
1910        tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1911        break;
1912    case INDEX_op_muls2_i32:
1913        tcg_out_smull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1914        break;
1915    /* XXX: Perhaps args[2] & 0x1f is wrong */
1916    case INDEX_op_shl_i32:
1917        c = const_args[2] ?
1918                SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
1919        goto gen_shift32;
1920    case INDEX_op_shr_i32:
1921        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
1922                SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
1923        goto gen_shift32;
1924    case INDEX_op_sar_i32:
1925        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
1926                SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
1927        goto gen_shift32;
1928    case INDEX_op_rotr_i32:
1929        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
1930                SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
1931        /* Fall through.  */
1932    gen_shift32:
1933        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
1934        break;
1935
1936    case INDEX_op_rotl_i32:
1937        if (const_args[2]) {
1938            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1939                            ((0x20 - args[2]) & 0x1f) ?
1940                            SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
1941                            SHIFT_IMM_LSL(0));
1942        } else {
1943            tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_TMP, args[2], 0x20);
1944            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1945                            SHIFT_REG_ROR(TCG_REG_TMP));
1946        }
1947        break;
1948
1949    case INDEX_op_ctz_i32:
1950        tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, args[1], 0);
1951        a1 = TCG_REG_TMP;
1952        goto do_clz;
1953
1954    case INDEX_op_clz_i32:
1955        a1 = args[1];
1956    do_clz:
1957        a0 = args[0];
1958        a2 = args[2];
1959        c = const_args[2];
1960        if (c && a2 == 32) {
1961            tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
1962            break;
1963        }
1964        tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1965        tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1966        if (c || a0 != a2) {
1967            tcg_out_dat_rIK(s, COND_EQ, ARITH_MOV, ARITH_MVN, a0, 0, a2, c);
1968        }
1969        break;
1970
1971    case INDEX_op_brcond_i32:
1972        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1973                       args[0], args[1], const_args[1]);
1974        tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]],
1975                           arg_label(args[3]));
1976        break;
1977    case INDEX_op_setcond_i32:
1978        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1979                        args[1], args[2], const_args[2]);
1980        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
1981                        ARITH_MOV, args[0], 0, 1);
1982        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
1983                        ARITH_MOV, args[0], 0, 0);
1984        break;
1985
1986    case INDEX_op_brcond2_i32:
1987        c = tcg_out_cmp2(s, args, const_args);
1988        tcg_out_goto_label(s, tcg_cond_to_arm_cond[c], arg_label(args[5]));
1989        break;
1990    case INDEX_op_setcond2_i32:
1991        c = tcg_out_cmp2(s, args + 1, const_args + 1);
1992        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[c], ARITH_MOV, args[0], 0, 1);
1993        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(c)],
1994                        ARITH_MOV, args[0], 0, 0);
1995        break;
1996
1997    case INDEX_op_qemu_ld_i32:
1998        tcg_out_qemu_ld(s, args, 0);
1999        break;
2000    case INDEX_op_qemu_ld_i64:
2001        tcg_out_qemu_ld(s, args, 1);
2002        break;
2003    case INDEX_op_qemu_st_i32:
2004        tcg_out_qemu_st(s, args, 0);
2005        break;
2006    case INDEX_op_qemu_st_i64:
2007        tcg_out_qemu_st(s, args, 1);
2008        break;
2009
2010    case INDEX_op_bswap16_i32:
2011        tcg_out_bswap16(s, COND_AL, args[0], args[1]);
2012        break;
2013    case INDEX_op_bswap32_i32:
2014        tcg_out_bswap32(s, COND_AL, args[0], args[1]);
2015        break;
2016
2017    case INDEX_op_ext8s_i32:
2018        tcg_out_ext8s(s, COND_AL, args[0], args[1]);
2019        break;
2020    case INDEX_op_ext16s_i32:
2021        tcg_out_ext16s(s, COND_AL, args[0], args[1]);
2022        break;
2023    case INDEX_op_ext16u_i32:
2024        tcg_out_ext16u(s, COND_AL, args[0], args[1]);
2025        break;
2026
2027    case INDEX_op_deposit_i32:
2028        tcg_out_deposit(s, COND_AL, args[0], args[2],
2029                        args[3], args[4], const_args[2]);
2030        break;
2031    case INDEX_op_extract_i32:
2032        tcg_out_extract(s, COND_AL, args[0], args[1], args[2], args[3]);
2033        break;
2034    case INDEX_op_sextract_i32:
2035        tcg_out_sextract(s, COND_AL, args[0], args[1], args[2], args[3]);
2036        break;
2037    case INDEX_op_extract2_i32:
2038        /* ??? These optimization vs zero should be generic.  */
2039        /* ??? But we can't substitute 2 for 1 in the opcode stream yet.  */
2040        if (const_args[1]) {
2041            if (const_args[2]) {
2042                tcg_out_movi(s, TCG_TYPE_REG, args[0], 0);
2043            } else {
2044                tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2045                                args[2], SHIFT_IMM_LSL(32 - args[3]));
2046            }
2047        } else if (const_args[2]) {
2048            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2049                            args[1], SHIFT_IMM_LSR(args[3]));
2050        } else {
2051            /* We can do extract2 in 2 insns, vs the 3 required otherwise.  */
2052            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0,
2053                            args[2], SHIFT_IMM_LSL(32 - args[3]));
2054            tcg_out_dat_reg(s, COND_AL, ARITH_ORR, args[0], TCG_REG_TMP,
2055                            args[1], SHIFT_IMM_LSR(args[3]));
2056        }
2057        break;
2058
2059    case INDEX_op_div_i32:
2060        tcg_out_sdiv(s, COND_AL, args[0], args[1], args[2]);
2061        break;
2062    case INDEX_op_divu_i32:
2063        tcg_out_udiv(s, COND_AL, args[0], args[1], args[2]);
2064        break;
2065
2066    case INDEX_op_mb:
2067        tcg_out_mb(s, args[0]);
2068        break;
2069
2070    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2071    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2072    default:
2073        tcg_abort();
2074    }
2075}
2076
2077static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
2078{
2079    static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
2080    static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
2081    static const TCGTargetOpDef s_s = { .args_ct_str = { "s", "s" } };
2082    static const TCGTargetOpDef r_l = { .args_ct_str = { "r", "l" } };
2083    static const TCGTargetOpDef r_r_r = { .args_ct_str = { "r", "r", "r" } };
2084    static const TCGTargetOpDef r_r_l = { .args_ct_str = { "r", "r", "l" } };
2085    static const TCGTargetOpDef r_l_l = { .args_ct_str = { "r", "l", "l" } };
2086    static const TCGTargetOpDef s_s_s = { .args_ct_str = { "s", "s", "s" } };
2087    static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
2088    static const TCGTargetOpDef r_r_rI = { .args_ct_str = { "r", "r", "rI" } };
2089    static const TCGTargetOpDef r_r_rIN
2090        = { .args_ct_str = { "r", "r", "rIN" } };
2091    static const TCGTargetOpDef r_r_rIK
2092        = { .args_ct_str = { "r", "r", "rIK" } };
2093    static const TCGTargetOpDef r_r_r_r
2094        = { .args_ct_str = { "r", "r", "r", "r" } };
2095    static const TCGTargetOpDef r_r_l_l
2096        = { .args_ct_str = { "r", "r", "l", "l" } };
2097    static const TCGTargetOpDef s_s_s_s
2098        = { .args_ct_str = { "s", "s", "s", "s" } };
2099    static const TCGTargetOpDef br
2100        = { .args_ct_str = { "r", "rIN" } };
2101    static const TCGTargetOpDef ext2
2102        = { .args_ct_str = { "r", "rZ", "rZ" } };
2103    static const TCGTargetOpDef dep
2104        = { .args_ct_str = { "r", "0", "rZ" } };
2105    static const TCGTargetOpDef movc
2106        = { .args_ct_str = { "r", "r", "rIN", "rIK", "0" } };
2107    static const TCGTargetOpDef add2
2108        = { .args_ct_str = { "r", "r", "r", "r", "rIN", "rIK" } };
2109    static const TCGTargetOpDef sub2
2110        = { .args_ct_str = { "r", "r", "rI", "rI", "rIN", "rIK" } };
2111    static const TCGTargetOpDef br2
2112        = { .args_ct_str = { "r", "r", "rI", "rI" } };
2113    static const TCGTargetOpDef setc2
2114        = { .args_ct_str = { "r", "r", "r", "rI", "rI" } };
2115
2116    switch (op) {
2117    case INDEX_op_goto_ptr:
2118        return &r;
2119
2120    case INDEX_op_ld8u_i32:
2121    case INDEX_op_ld8s_i32:
2122    case INDEX_op_ld16u_i32:
2123    case INDEX_op_ld16s_i32:
2124    case INDEX_op_ld_i32:
2125    case INDEX_op_st8_i32:
2126    case INDEX_op_st16_i32:
2127    case INDEX_op_st_i32:
2128    case INDEX_op_neg_i32:
2129    case INDEX_op_not_i32:
2130    case INDEX_op_bswap16_i32:
2131    case INDEX_op_bswap32_i32:
2132    case INDEX_op_ext8s_i32:
2133    case INDEX_op_ext16s_i32:
2134    case INDEX_op_ext16u_i32:
2135    case INDEX_op_extract_i32:
2136    case INDEX_op_sextract_i32:
2137        return &r_r;
2138
2139    case INDEX_op_add_i32:
2140    case INDEX_op_sub_i32:
2141    case INDEX_op_setcond_i32:
2142        return &r_r_rIN;
2143    case INDEX_op_and_i32:
2144    case INDEX_op_andc_i32:
2145    case INDEX_op_clz_i32:
2146    case INDEX_op_ctz_i32:
2147        return &r_r_rIK;
2148    case INDEX_op_mul_i32:
2149    case INDEX_op_div_i32:
2150    case INDEX_op_divu_i32:
2151        return &r_r_r;
2152    case INDEX_op_mulu2_i32:
2153    case INDEX_op_muls2_i32:
2154        return &r_r_r_r;
2155    case INDEX_op_or_i32:
2156    case INDEX_op_xor_i32:
2157        return &r_r_rI;
2158    case INDEX_op_shl_i32:
2159    case INDEX_op_shr_i32:
2160    case INDEX_op_sar_i32:
2161    case INDEX_op_rotl_i32:
2162    case INDEX_op_rotr_i32:
2163        return &r_r_ri;
2164
2165    case INDEX_op_brcond_i32:
2166        return &br;
2167    case INDEX_op_deposit_i32:
2168        return &dep;
2169    case INDEX_op_extract2_i32:
2170        return &ext2;
2171    case INDEX_op_movcond_i32:
2172        return &movc;
2173    case INDEX_op_add2_i32:
2174        return &add2;
2175    case INDEX_op_sub2_i32:
2176        return &sub2;
2177    case INDEX_op_brcond2_i32:
2178        return &br2;
2179    case INDEX_op_setcond2_i32:
2180        return &setc2;
2181
2182    case INDEX_op_qemu_ld_i32:
2183        return TARGET_LONG_BITS == 32 ? &r_l : &r_l_l;
2184    case INDEX_op_qemu_ld_i64:
2185        return TARGET_LONG_BITS == 32 ? &r_r_l : &r_r_l_l;
2186    case INDEX_op_qemu_st_i32:
2187        return TARGET_LONG_BITS == 32 ? &s_s : &s_s_s;
2188    case INDEX_op_qemu_st_i64:
2189        return TARGET_LONG_BITS == 32 ? &s_s_s : &s_s_s_s;
2190
2191    default:
2192        return NULL;
2193    }
2194}
2195
2196static void tcg_target_init(TCGContext *s)
2197{
2198    /* Only probe for the platform and capabilities if we havn't already
2199       determined maximum values at compile time.  */
2200#ifndef use_idiv_instructions
2201    {
2202        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2203        use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2204    }
2205#endif
2206    if (__ARM_ARCH < 7) {
2207        const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2208        if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2209            arm_arch = pl[1] - '0';
2210        }
2211    }
2212
2213    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
2214
2215    tcg_target_call_clobber_regs = 0;
2216    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2217    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2218    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2219    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2220    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2221    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2222
2223    s->reserved_regs = 0;
2224    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2225    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2226    tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2227}
2228
2229static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2230                              TCGReg arg1, intptr_t arg2)
2231{
2232    tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2233}
2234
2235static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2236                              TCGReg arg1, intptr_t arg2)
2237{
2238    tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2239}
2240
2241static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2242                               TCGReg base, intptr_t ofs)
2243{
2244    return false;
2245}
2246
2247static inline bool tcg_out_mov(TCGContext *s, TCGType type,
2248                               TCGReg ret, TCGReg arg)
2249{
2250    tcg_out_mov_reg(s, COND_AL, ret, arg);
2251    return true;
2252}
2253
2254static inline void tcg_out_movi(TCGContext *s, TCGType type,
2255                                TCGReg ret, tcg_target_long arg)
2256{
2257    tcg_out_movi32(s, COND_AL, ret, arg);
2258}
2259
2260static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2261{
2262    int i;
2263    for (i = 0; i < count; ++i) {
2264        p[i] = INSN_NOP;
2265    }
2266}
2267
2268/* Compute frame size via macros, to share between tcg_target_qemu_prologue
2269   and tcg_register_jit.  */
2270
2271#define PUSH_SIZE  ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
2272
2273#define FRAME_SIZE \
2274    ((PUSH_SIZE \
2275      + TCG_STATIC_CALL_ARGS_SIZE \
2276      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
2277      + TCG_TARGET_STACK_ALIGN - 1) \
2278     & -TCG_TARGET_STACK_ALIGN)
2279
2280#define STACK_ADDEND  (FRAME_SIZE - PUSH_SIZE)
2281
2282static void tcg_target_qemu_prologue(TCGContext *s)
2283{
2284    /* Calling convention requires us to save r4-r11 and lr.  */
2285    /* stmdb sp!, { r4 - r11, lr } */
2286    tcg_out32(s, (COND_AL << 28) | 0x092d4ff0);
2287
2288    /* Reserve callee argument and tcg temp space.  */
2289    tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
2290                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
2291    tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
2292                  CPU_TEMP_BUF_NLONGS * sizeof(long));
2293
2294    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2295
2296    tcg_out_bx(s, COND_AL, tcg_target_call_iarg_regs[1]);
2297
2298    /*
2299     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
2300     * and fall through to the rest of the epilogue.
2301     */
2302    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2303    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
2304    tcg_out_epilogue(s);
2305}
2306
2307static void tcg_out_epilogue(TCGContext *s)
2308{
2309    /* Release local stack frame.  */
2310    tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
2311                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
2312
2313    /* ldmia sp!, { r4 - r11, pc } */
2314    tcg_out32(s, (COND_AL << 28) | 0x08bd8ff0);
2315}
2316
2317typedef struct {
2318    DebugFrameHeader h;
2319    uint8_t fde_def_cfa[4];
2320    uint8_t fde_reg_ofs[18];
2321} DebugFrame;
2322
2323#define ELF_HOST_MACHINE EM_ARM
2324
2325/* We're expecting a 2 byte uleb128 encoded value.  */
2326QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2327
2328static const DebugFrame debug_frame = {
2329    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2330    .h.cie.id = -1,
2331    .h.cie.version = 1,
2332    .h.cie.code_align = 1,
2333    .h.cie.data_align = 0x7c,             /* sleb128 -4 */
2334    .h.cie.return_column = 14,
2335
2336    /* Total FDE size does not include the "len" member.  */
2337    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2338
2339    .fde_def_cfa = {
2340        12, 13,                         /* DW_CFA_def_cfa sp, ... */
2341        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
2342        (FRAME_SIZE >> 7)
2343    },
2344    .fde_reg_ofs = {
2345        /* The following must match the stmdb in the prologue.  */
2346        0x8e, 1,                        /* DW_CFA_offset, lr, -4 */
2347        0x8b, 2,                        /* DW_CFA_offset, r11, -8 */
2348        0x8a, 3,                        /* DW_CFA_offset, r10, -12 */
2349        0x89, 4,                        /* DW_CFA_offset, r9, -16 */
2350        0x88, 5,                        /* DW_CFA_offset, r8, -20 */
2351        0x87, 6,                        /* DW_CFA_offset, r7, -24 */
2352        0x86, 7,                        /* DW_CFA_offset, r6, -28 */
2353        0x85, 8,                        /* DW_CFA_offset, r5, -32 */
2354        0x84, 9,                        /* DW_CFA_offset, r4, -36 */
2355    }
2356};
2357
2358void tcg_register_jit(const void *buf, size_t buf_size)
2359{
2360    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2361}
2362