xref: /openbmc/qemu/tcg/sparc64/tcg-target.c.inc (revision 0cadc1eda1a3120c37c713ab6d6b7a02da0d2e6f)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25/* We only support generating code for 64-bit mode.  */
26#ifndef __arch64__
27#error "unsupported code generation mode"
28#endif
29
30#include "../tcg-pool.c.inc"
31
32#ifdef CONFIG_DEBUG_TCG
33static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
34    "%g0",
35    "%g1",
36    "%g2",
37    "%g3",
38    "%g4",
39    "%g5",
40    "%g6",
41    "%g7",
42    "%o0",
43    "%o1",
44    "%o2",
45    "%o3",
46    "%o4",
47    "%o5",
48    "%o6",
49    "%o7",
50    "%l0",
51    "%l1",
52    "%l2",
53    "%l3",
54    "%l4",
55    "%l5",
56    "%l6",
57    "%l7",
58    "%i0",
59    "%i1",
60    "%i2",
61    "%i3",
62    "%i4",
63    "%i5",
64    "%i6",
65    "%i7",
66};
67#endif
68
69#define TCG_CT_CONST_S11  0x100
70#define TCG_CT_CONST_S13  0x200
71#define TCG_CT_CONST_ZERO 0x400
72
73/*
74 * For softmmu, we need to avoid conflicts with the first 3
75 * argument registers to perform the tlb lookup, and to call
76 * the helper function.
77 */
78#ifdef CONFIG_SOFTMMU
79#define SOFTMMU_RESERVE_REGS MAKE_64BIT_MASK(TCG_REG_O0, 3)
80#else
81#define SOFTMMU_RESERVE_REGS 0
82#endif
83#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 32)
84#define ALL_QLDST_REGS       (ALL_GENERAL_REGS & ~SOFTMMU_RESERVE_REGS)
85
86/* Define some temporary registers.  T2 is used for constant generation.  */
87#define TCG_REG_T1  TCG_REG_G1
88#define TCG_REG_T2  TCG_REG_O7
89
90#ifndef CONFIG_SOFTMMU
91# define TCG_GUEST_BASE_REG TCG_REG_I5
92#endif
93
94#define TCG_REG_TB  TCG_REG_I1
95
96static const int tcg_target_reg_alloc_order[] = {
97    TCG_REG_L0,
98    TCG_REG_L1,
99    TCG_REG_L2,
100    TCG_REG_L3,
101    TCG_REG_L4,
102    TCG_REG_L5,
103    TCG_REG_L6,
104    TCG_REG_L7,
105
106    TCG_REG_I0,
107    TCG_REG_I1,
108    TCG_REG_I2,
109    TCG_REG_I3,
110    TCG_REG_I4,
111    TCG_REG_I5,
112
113    TCG_REG_G2,
114    TCG_REG_G3,
115    TCG_REG_G4,
116    TCG_REG_G5,
117
118    TCG_REG_O0,
119    TCG_REG_O1,
120    TCG_REG_O2,
121    TCG_REG_O3,
122    TCG_REG_O4,
123    TCG_REG_O5,
124};
125
126static const int tcg_target_call_iarg_regs[6] = {
127    TCG_REG_O0,
128    TCG_REG_O1,
129    TCG_REG_O2,
130    TCG_REG_O3,
131    TCG_REG_O4,
132    TCG_REG_O5,
133};
134
135static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
136{
137    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
138    tcg_debug_assert(slot >= 0 && slot <= 3);
139    return TCG_REG_O0 + slot;
140}
141
142#define INSN_OP(x)  ((x) << 30)
143#define INSN_OP2(x) ((x) << 22)
144#define INSN_OP3(x) ((x) << 19)
145#define INSN_OPF(x) ((x) << 5)
146#define INSN_RD(x)  ((x) << 25)
147#define INSN_RS1(x) ((x) << 14)
148#define INSN_RS2(x) (x)
149#define INSN_ASI(x) ((x) << 5)
150
151#define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff))
152#define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff))
153#define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff))
154#define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20))
155#define INSN_OFF19(x) (((x) >> 2) & 0x07ffff)
156#define INSN_COND(x) ((x) << 25)
157
158#define COND_N     0x0
159#define COND_E     0x1
160#define COND_LE    0x2
161#define COND_L     0x3
162#define COND_LEU   0x4
163#define COND_CS    0x5
164#define COND_NEG   0x6
165#define COND_VS    0x7
166#define COND_A     0x8
167#define COND_NE    0x9
168#define COND_G     0xa
169#define COND_GE    0xb
170#define COND_GU    0xc
171#define COND_CC    0xd
172#define COND_POS   0xe
173#define COND_VC    0xf
174#define BA         (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2))
175
176#define RCOND_Z    1
177#define RCOND_LEZ  2
178#define RCOND_LZ   3
179#define RCOND_NZ   5
180#define RCOND_GZ   6
181#define RCOND_GEZ  7
182
183#define MOVCC_ICC  (1 << 18)
184#define MOVCC_XCC  (1 << 18 | 1 << 12)
185
186#define BPCC_ICC   0
187#define BPCC_XCC   (2 << 20)
188#define BPCC_PT    (1 << 19)
189#define BPCC_PN    0
190#define BPCC_A     (1 << 29)
191
192#define BPR_PT     BPCC_PT
193
194#define ARITH_ADD  (INSN_OP(2) | INSN_OP3(0x00))
195#define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10))
196#define ARITH_AND  (INSN_OP(2) | INSN_OP3(0x01))
197#define ARITH_ANDCC (INSN_OP(2) | INSN_OP3(0x11))
198#define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05))
199#define ARITH_OR   (INSN_OP(2) | INSN_OP3(0x02))
200#define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12))
201#define ARITH_ORN  (INSN_OP(2) | INSN_OP3(0x06))
202#define ARITH_XOR  (INSN_OP(2) | INSN_OP3(0x03))
203#define ARITH_SUB  (INSN_OP(2) | INSN_OP3(0x04))
204#define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14))
205#define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08))
206#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
207#define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
208#define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
209#define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
210#define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f))
211#define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09))
212#define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d))
213#define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d))
214#define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
215#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
216
217#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
218#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
219
220#define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
221#define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
222#define SHIFT_SRA  (INSN_OP(2) | INSN_OP3(0x27))
223
224#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
225#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
226#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
227
228#define RDY        (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
229#define WRY        (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
230#define JMPL       (INSN_OP(2) | INSN_OP3(0x38))
231#define RETURN     (INSN_OP(2) | INSN_OP3(0x39))
232#define SAVE       (INSN_OP(2) | INSN_OP3(0x3c))
233#define RESTORE    (INSN_OP(2) | INSN_OP3(0x3d))
234#define SETHI      (INSN_OP(0) | INSN_OP2(0x4))
235#define CALL       INSN_OP(1)
236#define LDUB       (INSN_OP(3) | INSN_OP3(0x01))
237#define LDSB       (INSN_OP(3) | INSN_OP3(0x09))
238#define LDUH       (INSN_OP(3) | INSN_OP3(0x02))
239#define LDSH       (INSN_OP(3) | INSN_OP3(0x0a))
240#define LDUW       (INSN_OP(3) | INSN_OP3(0x00))
241#define LDSW       (INSN_OP(3) | INSN_OP3(0x08))
242#define LDX        (INSN_OP(3) | INSN_OP3(0x0b))
243#define STB        (INSN_OP(3) | INSN_OP3(0x05))
244#define STH        (INSN_OP(3) | INSN_OP3(0x06))
245#define STW        (INSN_OP(3) | INSN_OP3(0x04))
246#define STX        (INSN_OP(3) | INSN_OP3(0x0e))
247#define LDUBA      (INSN_OP(3) | INSN_OP3(0x11))
248#define LDSBA      (INSN_OP(3) | INSN_OP3(0x19))
249#define LDUHA      (INSN_OP(3) | INSN_OP3(0x12))
250#define LDSHA      (INSN_OP(3) | INSN_OP3(0x1a))
251#define LDUWA      (INSN_OP(3) | INSN_OP3(0x10))
252#define LDSWA      (INSN_OP(3) | INSN_OP3(0x18))
253#define LDXA       (INSN_OP(3) | INSN_OP3(0x1b))
254#define STBA       (INSN_OP(3) | INSN_OP3(0x15))
255#define STHA       (INSN_OP(3) | INSN_OP3(0x16))
256#define STWA       (INSN_OP(3) | INSN_OP3(0x14))
257#define STXA       (INSN_OP(3) | INSN_OP3(0x1e))
258
259#define MEMBAR     (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
260
261#define NOP        (SETHI | INSN_RD(TCG_REG_G0) | 0)
262
263#ifndef ASI_PRIMARY_LITTLE
264#define ASI_PRIMARY_LITTLE 0x88
265#endif
266
267#define LDUH_LE    (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
268#define LDSH_LE    (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
269#define LDUW_LE    (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
270#define LDSW_LE    (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
271#define LDX_LE     (LDXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
272
273#define STH_LE     (STHA  | INSN_ASI(ASI_PRIMARY_LITTLE))
274#define STW_LE     (STWA  | INSN_ASI(ASI_PRIMARY_LITTLE))
275#define STX_LE     (STXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
276
277#ifndef use_vis3_instructions
278bool use_vis3_instructions;
279#endif
280
281static bool check_fit_i64(int64_t val, unsigned int bits)
282{
283    return val == sextract64(val, 0, bits);
284}
285
286static bool check_fit_i32(int32_t val, unsigned int bits)
287{
288    return val == sextract32(val, 0, bits);
289}
290
291#define check_fit_tl    check_fit_i64
292#define check_fit_ptr   check_fit_i64
293
294static bool patch_reloc(tcg_insn_unit *src_rw, int type,
295                        intptr_t value, intptr_t addend)
296{
297    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
298    uint32_t insn = *src_rw;
299    intptr_t pcrel;
300
301    value += addend;
302    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx);
303
304    switch (type) {
305    case R_SPARC_WDISP16:
306        if (!check_fit_ptr(pcrel >> 2, 16)) {
307            return false;
308        }
309        insn &= ~INSN_OFF16(-1);
310        insn |= INSN_OFF16(pcrel);
311        break;
312    case R_SPARC_WDISP19:
313        if (!check_fit_ptr(pcrel >> 2, 19)) {
314            return false;
315        }
316        insn &= ~INSN_OFF19(-1);
317        insn |= INSN_OFF19(pcrel);
318        break;
319    case R_SPARC_13:
320        if (!check_fit_ptr(value, 13)) {
321            return false;
322        }
323        insn &= ~INSN_IMM13(-1);
324        insn |= INSN_IMM13(value);
325        break;
326    default:
327        g_assert_not_reached();
328    }
329
330    *src_rw = insn;
331    return true;
332}
333
334/* test if a constant matches the constraint */
335static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
336{
337    if (ct & TCG_CT_CONST) {
338        return 1;
339    }
340
341    if (type == TCG_TYPE_I32) {
342        val = (int32_t)val;
343    }
344
345    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
346        return 1;
347    } else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
348        return 1;
349    } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
350        return 1;
351    } else {
352        return 0;
353    }
354}
355
356static void tcg_out_nop(TCGContext *s)
357{
358    tcg_out32(s, NOP);
359}
360
361static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
362                          TCGReg rs2, int op)
363{
364    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
365}
366
367static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
368                           int32_t offset, int op)
369{
370    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
371}
372
373static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
374			   int32_t val2, int val2const, int op)
375{
376    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
377              | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
378}
379
380static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
381{
382    if (ret != arg) {
383        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
384    }
385    return true;
386}
387
388static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg)
389{
390    if (ret != arg) {
391        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
392    } else {
393        tcg_out_nop(s);
394    }
395}
396
397static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
398{
399    tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
400}
401
402static void tcg_out_movi_imm13(TCGContext *s, TCGReg ret, int32_t arg)
403{
404    tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
405}
406
407static void tcg_out_movi_imm32(TCGContext *s, TCGReg ret, int32_t arg)
408{
409    if (check_fit_i32(arg, 13)) {
410        /* A 13-bit constant sign-extended to 64-bits.  */
411        tcg_out_movi_imm13(s, ret, arg);
412    } else {
413        /* A 32-bit constant zero-extended to 64 bits.  */
414        tcg_out_sethi(s, ret, arg);
415        if (arg & 0x3ff) {
416            tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
417        }
418    }
419}
420
421static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
422                             tcg_target_long arg, bool in_prologue,
423                             TCGReg scratch)
424{
425    tcg_target_long hi, lo = (int32_t)arg;
426    tcg_target_long test, lsb;
427
428    /* A 32-bit constant, or 32-bit zero-extended to 64-bits.  */
429    if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
430        tcg_out_movi_imm32(s, ret, arg);
431        return;
432    }
433
434    /* A 13-bit constant sign-extended to 64-bits.  */
435    if (check_fit_tl(arg, 13)) {
436        tcg_out_movi_imm13(s, ret, arg);
437        return;
438    }
439
440    /* A 13-bit constant relative to the TB.  */
441    if (!in_prologue) {
442        test = tcg_tbrel_diff(s, (void *)arg);
443        if (check_fit_ptr(test, 13)) {
444            tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
445            return;
446        }
447    }
448
449    /* A 32-bit constant sign-extended to 64-bits.  */
450    if (arg == lo) {
451        tcg_out_sethi(s, ret, ~arg);
452        tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
453        return;
454    }
455
456    /* A 32-bit constant, shifted.  */
457    lsb = ctz64(arg);
458    test = (tcg_target_long)arg >> lsb;
459    if (lsb > 10 && test == extract64(test, 0, 21)) {
460        tcg_out_sethi(s, ret, test << 10);
461        tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
462        return;
463    } else if (test == (uint32_t)test || test == (int32_t)test) {
464        tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch);
465        tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
466        return;
467    }
468
469    /* Use the constant pool, if possible. */
470    if (!in_prologue) {
471        new_pool_label(s, arg, R_SPARC_13, s->code_ptr,
472                       tcg_tbrel_diff(s, NULL));
473        tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB));
474        return;
475    }
476
477    /* A 64-bit constant decomposed into 2 32-bit pieces.  */
478    if (check_fit_i32(lo, 13)) {
479        hi = (arg - lo) >> 32;
480        tcg_out_movi_imm32(s, ret, hi);
481        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
482        tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
483    } else {
484        hi = arg >> 32;
485        tcg_out_movi_imm32(s, ret, hi);
486        tcg_out_movi_imm32(s, scratch, lo);
487        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
488        tcg_out_arith(s, ret, ret, scratch, ARITH_OR);
489    }
490}
491
492static void tcg_out_movi(TCGContext *s, TCGType type,
493                         TCGReg ret, tcg_target_long arg)
494{
495    tcg_debug_assert(ret != TCG_REG_T2);
496    tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T2);
497}
498
499static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
500{
501    g_assert_not_reached();
502}
503
504static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
505{
506    g_assert_not_reached();
507}
508
509static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs)
510{
511    tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND);
512}
513
514static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs)
515{
516    tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL);
517    tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL);
518}
519
520static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs)
521{
522    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA);
523}
524
525static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs)
526{
527    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL);
528}
529
530static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
531{
532    tcg_out_ext32s(s, rd, rs);
533}
534
535static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
536{
537    tcg_out_ext32u(s, rd, rs);
538}
539
540static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs)
541{
542    tcg_out_mov(s, TCG_TYPE_I32, rd, rs);
543}
544
545static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
546{
547    return false;
548}
549
550static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
551                             tcg_target_long imm)
552{
553    /* This function is only used for passing structs by reference. */
554    g_assert_not_reached();
555}
556
557static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
558                            TCGReg a2, int op)
559{
560    tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
561}
562
563static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
564                         intptr_t offset, int op)
565{
566    if (check_fit_ptr(offset, 13)) {
567        tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
568                  INSN_IMM13(offset));
569    } else {
570        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
571        tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
572    }
573}
574
575static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
576                       TCGReg arg1, intptr_t arg2)
577{
578    tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
579}
580
581static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
582                       TCGReg arg1, intptr_t arg2)
583{
584    tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
585}
586
587static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
588                        TCGReg base, intptr_t ofs)
589{
590    if (val == 0) {
591        tcg_out_st(s, type, TCG_REG_G0, base, ofs);
592        return true;
593    }
594    return false;
595}
596
597static void tcg_out_sety(TCGContext *s, TCGReg rs)
598{
599    tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
600}
601
602static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1,
603                          int32_t val2, int val2const, int uns)
604{
605    /* Load Y with the sign/zero extension of RS1 to 64-bits.  */
606    if (uns) {
607        tcg_out_sety(s, TCG_REG_G0);
608    } else {
609        tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA);
610        tcg_out_sety(s, TCG_REG_T1);
611    }
612
613    tcg_out_arithc(s, rd, rs1, val2, val2const,
614                   uns ? ARITH_UDIV : ARITH_SDIV);
615}
616
617static const uint8_t tcg_cond_to_bcond[] = {
618    [TCG_COND_EQ] = COND_E,
619    [TCG_COND_NE] = COND_NE,
620    [TCG_COND_LT] = COND_L,
621    [TCG_COND_GE] = COND_GE,
622    [TCG_COND_LE] = COND_LE,
623    [TCG_COND_GT] = COND_G,
624    [TCG_COND_LTU] = COND_CS,
625    [TCG_COND_GEU] = COND_CC,
626    [TCG_COND_LEU] = COND_LEU,
627    [TCG_COND_GTU] = COND_GU,
628};
629
630static const uint8_t tcg_cond_to_rcond[] = {
631    [TCG_COND_EQ] = RCOND_Z,
632    [TCG_COND_NE] = RCOND_NZ,
633    [TCG_COND_LT] = RCOND_LZ,
634    [TCG_COND_GT] = RCOND_GZ,
635    [TCG_COND_LE] = RCOND_LEZ,
636    [TCG_COND_GE] = RCOND_GEZ
637};
638
639static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
640{
641    tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
642}
643
644static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
645{
646    int off19 = 0;
647
648    if (l->has_value) {
649        off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
650    } else {
651        tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
652    }
653    tcg_out_bpcc0(s, scond, flags, off19);
654}
655
656static void tcg_out_cmp(TCGContext *s, TCGReg c1, int32_t c2, int c2const)
657{
658    tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const, ARITH_SUBCC);
659}
660
661static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
662                               int32_t arg2, int const_arg2, TCGLabel *l)
663{
664    tcg_out_cmp(s, arg1, arg2, const_arg2);
665    tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
666    tcg_out_nop(s);
667}
668
669static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret,
670                          int32_t v1, int v1const)
671{
672    tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret)
673              | INSN_RS1(tcg_cond_to_bcond[cond])
674              | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
675}
676
677static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
678                                TCGReg c1, int32_t c2, int c2const,
679                                int32_t v1, int v1const)
680{
681    tcg_out_cmp(s, c1, c2, c2const);
682    tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const);
683}
684
685static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
686                               int32_t arg2, int const_arg2, TCGLabel *l)
687{
688    /* For 64-bit signed comparisons vs zero, we can avoid the compare.  */
689    if (arg2 == 0 && !is_unsigned_cond(cond)) {
690        int off16 = 0;
691
692        if (l->has_value) {
693            off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
694        } else {
695            tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
696        }
697        tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
698                  | INSN_COND(tcg_cond_to_rcond[cond]) | off16);
699    } else {
700        tcg_out_cmp(s, arg1, arg2, const_arg2);
701        tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
702    }
703    tcg_out_nop(s);
704}
705
706static void tcg_out_movr(TCGContext *s, TCGCond cond, TCGReg ret, TCGReg c1,
707                         int32_t v1, int v1const)
708{
709    tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1)
710              | (tcg_cond_to_rcond[cond] << 10)
711              | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
712}
713
714static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
715                                TCGReg c1, int32_t c2, int c2const,
716                                int32_t v1, int v1const)
717{
718    /* For 64-bit signed comparisons vs zero, we can avoid the compare.
719       Note that the immediate range is one bit smaller, so we must check
720       for that as well.  */
721    if (c2 == 0 && !is_unsigned_cond(cond)
722        && (!v1const || check_fit_i32(v1, 10))) {
723        tcg_out_movr(s, cond, ret, c1, v1, v1const);
724    } else {
725        tcg_out_cmp(s, c1, c2, c2const);
726        tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const);
727    }
728}
729
730static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
731                                TCGReg c1, int32_t c2, int c2const)
732{
733    /* For 32-bit comparisons, we can play games with ADDC/SUBC.  */
734    switch (cond) {
735    case TCG_COND_LTU:
736    case TCG_COND_GEU:
737        /* The result of the comparison is in the carry bit.  */
738        break;
739
740    case TCG_COND_EQ:
741    case TCG_COND_NE:
742        /* For equality, we can transform to inequality vs zero.  */
743        if (c2 != 0) {
744            tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
745            c2 = TCG_REG_T1;
746        } else {
747            c2 = c1;
748        }
749        c1 = TCG_REG_G0, c2const = 0;
750        cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
751	break;
752
753    case TCG_COND_GTU:
754    case TCG_COND_LEU:
755        /* If we don't need to load a constant into a register, we can
756           swap the operands on GTU/LEU.  There's no benefit to loading
757           the constant into a temporary register.  */
758        if (!c2const || c2 == 0) {
759            TCGReg t = c1;
760            c1 = c2;
761            c2 = t;
762            c2const = 0;
763            cond = tcg_swap_cond(cond);
764            break;
765        }
766        /* FALLTHRU */
767
768    default:
769        tcg_out_cmp(s, c1, c2, c2const);
770        tcg_out_movi_imm13(s, ret, 0);
771        tcg_out_movcc(s, cond, MOVCC_ICC, ret, 1, 1);
772        return;
773    }
774
775    tcg_out_cmp(s, c1, c2, c2const);
776    if (cond == TCG_COND_LTU) {
777        tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
778    } else {
779        tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
780    }
781}
782
783static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
784                                TCGReg c1, int32_t c2, int c2const)
785{
786    if (use_vis3_instructions) {
787        switch (cond) {
788        case TCG_COND_NE:
789            if (c2 != 0) {
790                break;
791            }
792            c2 = c1, c2const = 0, c1 = TCG_REG_G0;
793            /* FALLTHRU */
794        case TCG_COND_LTU:
795            tcg_out_cmp(s, c1, c2, c2const);
796            tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
797            return;
798        default:
799            break;
800        }
801    }
802
803    /* For 64-bit signed comparisons vs zero, we can avoid the compare
804       if the input does not overlap the output.  */
805    if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) {
806        tcg_out_movi_imm13(s, ret, 0);
807        tcg_out_movr(s, cond, ret, c1, 1, 1);
808    } else {
809        tcg_out_cmp(s, c1, c2, c2const);
810        tcg_out_movi_imm13(s, ret, 0);
811        tcg_out_movcc(s, cond, MOVCC_XCC, ret, 1, 1);
812    }
813}
814
815static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh,
816                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
817                                int32_t bh, int bhconst, int opl, int oph)
818{
819    TCGReg tmp = TCG_REG_T1;
820
821    /* Note that the low parts are fully consumed before tmp is set.  */
822    if (rl != ah && (bhconst || rl != bh)) {
823        tmp = rl;
824    }
825
826    tcg_out_arithc(s, tmp, al, bl, blconst, opl);
827    tcg_out_arithc(s, rh, ah, bh, bhconst, oph);
828    tcg_out_mov(s, TCG_TYPE_I32, rl, tmp);
829}
830
831static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
832                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
833                                int32_t bh, int bhconst, bool is_sub)
834{
835    TCGReg tmp = TCG_REG_T1;
836
837    /* Note that the low parts are fully consumed before tmp is set.  */
838    if (rl != ah && (bhconst || rl != bh)) {
839        tmp = rl;
840    }
841
842    tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
843
844    if (use_vis3_instructions && !is_sub) {
845        /* Note that ADDXC doesn't accept immediates.  */
846        if (bhconst && bh != 0) {
847           tcg_out_movi_imm13(s, TCG_REG_T2, bh);
848           bh = TCG_REG_T2;
849        }
850        tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC);
851    } else if (bh == TCG_REG_G0) {
852	/* If we have a zero, we can perform the operation in two insns,
853           with the arithmetic first, and a conditional move into place.  */
854	if (rh == ah) {
855            tcg_out_arithi(s, TCG_REG_T2, ah, 1,
856			   is_sub ? ARITH_SUB : ARITH_ADD);
857            tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0);
858	} else {
859            tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD);
860	    tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0);
861	}
862    } else {
863        /*
864         * Otherwise adjust BH as if there is carry into T2.
865         * Note that constant BH is constrained to 11 bits for the MOVCC,
866         * so the adjustment fits 12 bits.
867         */
868        if (bhconst) {
869            tcg_out_movi_imm13(s, TCG_REG_T2, bh + (is_sub ? -1 : 1));
870        } else {
871            tcg_out_arithi(s, TCG_REG_T2, bh, 1,
872                           is_sub ? ARITH_SUB : ARITH_ADD);
873        }
874        /* ... smoosh T2 back to original BH if carry is clear ... */
875        tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst);
876	/* ... and finally perform the arithmetic with the new operand.  */
877        tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD);
878    }
879
880    tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
881}
882
883static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest,
884                               bool in_prologue, bool tail_call)
885{
886    uintptr_t desti = (uintptr_t)dest;
887
888    /* Be careful not to clobber %o7 for a tail call. */
889    tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
890                     desti & ~0xfff, in_prologue,
891                     tail_call ? TCG_REG_G2 : TCG_REG_O7);
892    tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7,
893                   TCG_REG_T1, desti & 0xfff, JMPL);
894}
895
896static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
897                                 bool in_prologue)
898{
899    ptrdiff_t disp = tcg_pcrel_diff(s, dest);
900
901    if (disp == (int32_t)disp) {
902        tcg_out32(s, CALL | (uint32_t)disp >> 2);
903    } else {
904        tcg_out_jmpl_const(s, dest, in_prologue, false);
905    }
906}
907
908static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
909                         const TCGHelperInfo *info)
910{
911    tcg_out_call_nodelay(s, dest, false);
912    tcg_out_nop(s);
913}
914
915static void tcg_out_mb(TCGContext *s, TCGArg a0)
916{
917    /* Note that the TCG memory order constants mirror the Sparc MEMBAR.  */
918    tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
919}
920
921#ifdef CONFIG_SOFTMMU
922static const tcg_insn_unit *qemu_ld_trampoline[MO_SSIZE + 1];
923static const tcg_insn_unit *qemu_st_trampoline[MO_SIZE + 1];
924
925static void build_trampolines(TCGContext *s)
926{
927    int i;
928
929    for (i = 0; i < ARRAY_SIZE(qemu_ld_helpers); ++i) {
930        if (qemu_ld_helpers[i] == NULL) {
931            continue;
932        }
933
934        /* May as well align the trampoline.  */
935        while ((uintptr_t)s->code_ptr & 15) {
936            tcg_out_nop(s);
937        }
938        qemu_ld_trampoline[i] = tcg_splitwx_to_rx(s->code_ptr);
939
940        /* Set the retaddr operand.  */
941        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O3, TCG_REG_O7);
942        /* Tail call.  */
943        tcg_out_jmpl_const(s, qemu_ld_helpers[i], true, true);
944        /* delay slot -- set the env argument */
945        tcg_out_mov_delay(s, TCG_REG_O0, TCG_AREG0);
946    }
947
948    for (i = 0; i < ARRAY_SIZE(qemu_st_helpers); ++i) {
949        if (qemu_st_helpers[i] == NULL) {
950            continue;
951        }
952
953        /* May as well align the trampoline.  */
954        while ((uintptr_t)s->code_ptr & 15) {
955            tcg_out_nop(s);
956        }
957        qemu_st_trampoline[i] = tcg_splitwx_to_rx(s->code_ptr);
958
959        /* Set the retaddr operand.  */
960        tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O4, TCG_REG_O7);
961
962        /* Tail call.  */
963        tcg_out_jmpl_const(s, qemu_st_helpers[i], true, true);
964        /* delay slot -- set the env argument */
965        tcg_out_mov_delay(s, TCG_REG_O0, TCG_AREG0);
966    }
967}
968#else
969static const tcg_insn_unit *qemu_unalign_ld_trampoline;
970static const tcg_insn_unit *qemu_unalign_st_trampoline;
971
972static void build_trampolines(TCGContext *s)
973{
974    for (int ld = 0; ld < 2; ++ld) {
975        void *helper;
976
977        while ((uintptr_t)s->code_ptr & 15) {
978            tcg_out_nop(s);
979        }
980
981        if (ld) {
982            helper = helper_unaligned_ld;
983            qemu_unalign_ld_trampoline = tcg_splitwx_to_rx(s->code_ptr);
984        } else {
985            helper = helper_unaligned_st;
986            qemu_unalign_st_trampoline = tcg_splitwx_to_rx(s->code_ptr);
987        }
988
989        /* Tail call.  */
990        tcg_out_jmpl_const(s, helper, true, true);
991        /* delay slot -- set the env argument */
992        tcg_out_mov_delay(s, TCG_REG_O0, TCG_AREG0);
993    }
994}
995#endif
996
997/* Generate global QEMU prologue and epilogue code */
998static void tcg_target_qemu_prologue(TCGContext *s)
999{
1000    int tmp_buf_size, frame_size;
1001
1002    /*
1003     * The TCG temp buffer is at the top of the frame, immediately
1004     * below the frame pointer.  Use the logical (aligned) offset here;
1005     * the stack bias is applied in temp_allocate_frame().
1006     */
1007    tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
1008    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
1009
1010    /*
1011     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
1012     * otherwise the minimal frame usable by callees.
1013     */
1014    frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
1015    frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
1016    frame_size += TCG_TARGET_STACK_ALIGN - 1;
1017    frame_size &= -TCG_TARGET_STACK_ALIGN;
1018    tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
1019              INSN_IMM13(-frame_size));
1020
1021#ifndef CONFIG_SOFTMMU
1022    if (guest_base != 0) {
1023        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG,
1024                         guest_base, true, TCG_REG_T1);
1025        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1026    }
1027#endif
1028
1029    /* We choose TCG_REG_TB such that no move is required.  */
1030    QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
1031    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
1032
1033    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
1034    /* delay slot */
1035    tcg_out_nop(s);
1036
1037    /* Epilogue for goto_ptr.  */
1038    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1039    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1040    /* delay slot */
1041    tcg_out_movi_imm13(s, TCG_REG_O0, 0);
1042
1043    build_trampolines(s);
1044}
1045
1046static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1047{
1048    int i;
1049    for (i = 0; i < count; ++i) {
1050        p[i] = NOP;
1051    }
1052}
1053
1054#if defined(CONFIG_SOFTMMU)
1055
1056/* We expect to use a 13-bit negative offset from ENV.  */
1057QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1058QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 12));
1059
1060/* Perform the TLB load and compare.
1061
1062   Inputs:
1063   ADDRLO and ADDRHI contain the possible two parts of the address.
1064
1065   MEM_INDEX and S_BITS are the memory context and log2 size of the load.
1066
1067   WHICH is the offset into the CPUTLBEntry structure of the slot to read.
1068   This should be offsetof addr_read or addr_write.
1069
1070   The result of the TLB comparison is in %[ix]cc.  The sanitized address
1071   is in the returned register, maybe %o0.  The TLB addend is in %o1.  */
1072
1073static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addr, int mem_index,
1074                               MemOp opc, int which)
1075{
1076    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1077    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1078    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1079    const TCGReg r0 = TCG_REG_O0;
1080    const TCGReg r1 = TCG_REG_O1;
1081    const TCGReg r2 = TCG_REG_O2;
1082    unsigned s_bits = opc & MO_SIZE;
1083    unsigned a_bits = get_alignment_bits(opc);
1084    tcg_target_long compare_mask;
1085
1086    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
1087    tcg_out_ld(s, TCG_TYPE_PTR, r0, TCG_AREG0, mask_off);
1088    tcg_out_ld(s, TCG_TYPE_PTR, r1, TCG_AREG0, table_off);
1089
1090    /* Extract the page index, shifted into place for tlb index.  */
1091    tcg_out_arithi(s, r2, addr, TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS,
1092                   SHIFT_SRL);
1093    tcg_out_arith(s, r2, r2, r0, ARITH_AND);
1094
1095    /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2.  */
1096    tcg_out_arith(s, r2, r2, r1, ARITH_ADD);
1097
1098    /* Load the tlb comparator and the addend.  */
1099    tcg_out_ld(s, TCG_TYPE_TL, r0, r2, which);
1100    tcg_out_ld(s, TCG_TYPE_PTR, r1, r2, offsetof(CPUTLBEntry, addend));
1101
1102    /* Mask out the page offset, except for the required alignment.
1103       We don't support unaligned accesses.  */
1104    if (a_bits < s_bits) {
1105        a_bits = s_bits;
1106    }
1107    compare_mask = (tcg_target_ulong)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
1108    if (check_fit_tl(compare_mask, 13)) {
1109        tcg_out_arithi(s, r2, addr, compare_mask, ARITH_AND);
1110    } else {
1111        tcg_out_movi(s, TCG_TYPE_TL, r2, compare_mask);
1112        tcg_out_arith(s, r2, addr, r2, ARITH_AND);
1113    }
1114    tcg_out_cmp(s, r0, r2, 0);
1115
1116    /* If the guest address must be zero-extended, do so now.  */
1117    if (TARGET_LONG_BITS == 32) {
1118        tcg_out_ext32u(s, r0, addr);
1119        return r0;
1120    }
1121    return addr;
1122}
1123#endif /* CONFIG_SOFTMMU */
1124
1125static const int qemu_ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = {
1126    [MO_UB]   = LDUB,
1127    [MO_SB]   = LDSB,
1128    [MO_UB | MO_LE] = LDUB,
1129    [MO_SB | MO_LE] = LDSB,
1130
1131    [MO_BEUW] = LDUH,
1132    [MO_BESW] = LDSH,
1133    [MO_BEUL] = LDUW,
1134    [MO_BESL] = LDSW,
1135    [MO_BEUQ] = LDX,
1136    [MO_BESQ] = LDX,
1137
1138    [MO_LEUW] = LDUH_LE,
1139    [MO_LESW] = LDSH_LE,
1140    [MO_LEUL] = LDUW_LE,
1141    [MO_LESL] = LDSW_LE,
1142    [MO_LEUQ] = LDX_LE,
1143    [MO_LESQ] = LDX_LE,
1144};
1145
1146static const int qemu_st_opc[(MO_SIZE | MO_BSWAP) + 1] = {
1147    [MO_UB]   = STB,
1148
1149    [MO_BEUW] = STH,
1150    [MO_BEUL] = STW,
1151    [MO_BEUQ] = STX,
1152
1153    [MO_LEUW] = STH_LE,
1154    [MO_LEUL] = STW_LE,
1155    [MO_LEUQ] = STX_LE,
1156};
1157
1158static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
1159                            MemOpIdx oi, TCGType data_type)
1160{
1161    MemOp memop = get_memop(oi);
1162    tcg_insn_unit *label_ptr;
1163
1164#ifdef CONFIG_SOFTMMU
1165    unsigned memi = get_mmuidx(oi);
1166    TCGReg addrz;
1167    const tcg_insn_unit *func;
1168
1169    addrz = tcg_out_tlb_load(s, addr, memi, memop,
1170                             offsetof(CPUTLBEntry, addr_read));
1171
1172    /* The fast path is exactly one insn.  Thus we can perform the
1173       entire TLB Hit in the (annulled) delay slot of the branch
1174       over the TLB Miss case.  */
1175
1176    /* beq,a,pt %[xi]cc, label0 */
1177    label_ptr = s->code_ptr;
1178    tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT
1179                  | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0);
1180    /* delay slot */
1181    tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1,
1182                    qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]);
1183
1184    /* TLB Miss.  */
1185
1186    tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_O1, addrz);
1187
1188    /* We use the helpers to extend SB and SW data, leaving the case
1189       of SL needing explicit extending below.  */
1190    if ((memop & MO_SSIZE) == MO_SL) {
1191        func = qemu_ld_trampoline[MO_UL];
1192    } else {
1193        func = qemu_ld_trampoline[memop & MO_SSIZE];
1194    }
1195    tcg_debug_assert(func != NULL);
1196    tcg_out_call_nodelay(s, func, false);
1197    /* delay slot */
1198    tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_O2, oi);
1199
1200    /* We let the helper sign-extend SB and SW, but leave SL for here.  */
1201    if ((memop & MO_SSIZE) == MO_SL) {
1202        tcg_out_ext32s(s, data, TCG_REG_O0);
1203    } else {
1204        tcg_out_mov(s, TCG_TYPE_REG, data, TCG_REG_O0);
1205    }
1206
1207    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1208#else
1209    TCGReg index = (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0);
1210    unsigned a_bits = get_alignment_bits(memop);
1211    unsigned s_bits = memop & MO_SIZE;
1212    unsigned t_bits;
1213
1214    if (TARGET_LONG_BITS == 32) {
1215        tcg_out_ext32u(s, TCG_REG_T1, addr);
1216        addr = TCG_REG_T1;
1217    }
1218
1219    /*
1220     * Normal case: alignment equal to access size.
1221     */
1222    if (a_bits == s_bits) {
1223        tcg_out_ldst_rr(s, data, addr, index,
1224                        qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]);
1225        return;
1226    }
1227
1228    /*
1229     * Test for at least natural alignment, and assume most accesses
1230     * will be aligned -- perform a straight load in the delay slot.
1231     * This is required to preserve atomicity for aligned accesses.
1232     */
1233    t_bits = MAX(a_bits, s_bits);
1234    tcg_debug_assert(t_bits < 13);
1235    tcg_out_arithi(s, TCG_REG_G0, addr, (1u << t_bits) - 1, ARITH_ANDCC);
1236
1237    /* beq,a,pt %icc, label */
1238    label_ptr = s->code_ptr;
1239    tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT | BPCC_ICC, 0);
1240    /* delay slot */
1241    tcg_out_ldst_rr(s, data, addr, index,
1242                    qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]);
1243
1244    if (a_bits >= s_bits) {
1245        /*
1246         * Overalignment: A successful alignment test will perform the memory
1247         * operation in the delay slot, and failure need only invoke the
1248         * handler for SIGBUS.
1249         */
1250        tcg_out_call_nodelay(s, qemu_unalign_ld_trampoline, false);
1251        /* delay slot -- move to low part of argument reg */
1252        tcg_out_mov_delay(s, TCG_REG_O1, addr);
1253    } else {
1254        /* Underalignment: load by pieces of minimum alignment. */
1255        int ld_opc, a_size, s_size, i;
1256
1257        /*
1258         * Force full address into T1 early; avoids problems with
1259         * overlap between @addr and @data.
1260         */
1261        tcg_out_arith(s, TCG_REG_T1, addr, index, ARITH_ADD);
1262
1263        a_size = 1 << a_bits;
1264        s_size = 1 << s_bits;
1265        if ((memop & MO_BSWAP) == MO_BE) {
1266            ld_opc = qemu_ld_opc[a_bits | MO_BE | (memop & MO_SIGN)];
1267            tcg_out_ldst(s, data, TCG_REG_T1, 0, ld_opc);
1268            ld_opc = qemu_ld_opc[a_bits | MO_BE];
1269            for (i = a_size; i < s_size; i += a_size) {
1270                tcg_out_ldst(s, TCG_REG_T2, TCG_REG_T1, i, ld_opc);
1271                tcg_out_arithi(s, data, data, a_size, SHIFT_SLLX);
1272                tcg_out_arith(s, data, data, TCG_REG_T2, ARITH_OR);
1273            }
1274        } else if (a_bits == 0) {
1275            ld_opc = LDUB;
1276            tcg_out_ldst(s, data, TCG_REG_T1, 0, ld_opc);
1277            for (i = a_size; i < s_size; i += a_size) {
1278                if ((memop & MO_SIGN) && i == s_size - a_size) {
1279                    ld_opc = LDSB;
1280                }
1281                tcg_out_ldst(s, TCG_REG_T2, TCG_REG_T1, i, ld_opc);
1282                tcg_out_arithi(s, TCG_REG_T2, TCG_REG_T2, i * 8, SHIFT_SLLX);
1283                tcg_out_arith(s, data, data, TCG_REG_T2, ARITH_OR);
1284            }
1285        } else {
1286            ld_opc = qemu_ld_opc[a_bits | MO_LE];
1287            tcg_out_ldst_rr(s, data, TCG_REG_T1, TCG_REG_G0, ld_opc);
1288            for (i = a_size; i < s_size; i += a_size) {
1289                tcg_out_arithi(s, TCG_REG_T1, TCG_REG_T1, a_size, ARITH_ADD);
1290                if ((memop & MO_SIGN) && i == s_size - a_size) {
1291                    ld_opc = qemu_ld_opc[a_bits | MO_LE | MO_SIGN];
1292                }
1293                tcg_out_ldst_rr(s, TCG_REG_T2, TCG_REG_T1, TCG_REG_G0, ld_opc);
1294                tcg_out_arithi(s, TCG_REG_T2, TCG_REG_T2, i * 8, SHIFT_SLLX);
1295                tcg_out_arith(s, data, data, TCG_REG_T2, ARITH_OR);
1296            }
1297        }
1298    }
1299
1300    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1301#endif /* CONFIG_SOFTMMU */
1302}
1303
1304static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr,
1305                            MemOpIdx oi, TCGType data_type)
1306{
1307    MemOp memop = get_memop(oi);
1308    tcg_insn_unit *label_ptr;
1309
1310#ifdef CONFIG_SOFTMMU
1311    unsigned memi = get_mmuidx(oi);
1312    TCGReg addrz;
1313    const tcg_insn_unit *func;
1314
1315    addrz = tcg_out_tlb_load(s, addr, memi, memop,
1316                             offsetof(CPUTLBEntry, addr_write));
1317
1318    /* The fast path is exactly one insn.  Thus we can perform the entire
1319       TLB Hit in the (annulled) delay slot of the branch over TLB Miss.  */
1320    /* beq,a,pt %[xi]cc, label0 */
1321    label_ptr = s->code_ptr;
1322    tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT
1323                  | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0);
1324    /* delay slot */
1325    tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1,
1326                    qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]);
1327
1328    /* TLB Miss.  */
1329
1330    tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_O1, addrz);
1331    tcg_out_movext(s, (memop & MO_SIZE) == MO_64 ? TCG_TYPE_I64 : TCG_TYPE_I32,
1332                   TCG_REG_O2, data_type, memop & MO_SIZE, data);
1333
1334    func = qemu_st_trampoline[memop & MO_SIZE];
1335    tcg_debug_assert(func != NULL);
1336    tcg_out_call_nodelay(s, func, false);
1337    /* delay slot */
1338    tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_O3, oi);
1339
1340    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1341#else
1342    TCGReg index = (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0);
1343    unsigned a_bits = get_alignment_bits(memop);
1344    unsigned s_bits = memop & MO_SIZE;
1345    unsigned t_bits;
1346
1347    if (TARGET_LONG_BITS == 32) {
1348        tcg_out_ext32u(s, TCG_REG_T1, addr);
1349        addr = TCG_REG_T1;
1350    }
1351
1352    /*
1353     * Normal case: alignment equal to access size.
1354     */
1355    if (a_bits == s_bits) {
1356        tcg_out_ldst_rr(s, data, addr, index,
1357                        qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]);
1358        return;
1359    }
1360
1361    /*
1362     * Test for at least natural alignment, and assume most accesses
1363     * will be aligned -- perform a straight store in the delay slot.
1364     * This is required to preserve atomicity for aligned accesses.
1365     */
1366    t_bits = MAX(a_bits, s_bits);
1367    tcg_debug_assert(t_bits < 13);
1368    tcg_out_arithi(s, TCG_REG_G0, addr, (1u << t_bits) - 1, ARITH_ANDCC);
1369
1370    /* beq,a,pt %icc, label */
1371    label_ptr = s->code_ptr;
1372    tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT | BPCC_ICC, 0);
1373    /* delay slot */
1374    tcg_out_ldst_rr(s, data, addr, index,
1375                    qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]);
1376
1377    if (a_bits >= s_bits) {
1378        /*
1379         * Overalignment: A successful alignment test will perform the memory
1380         * operation in the delay slot, and failure need only invoke the
1381         * handler for SIGBUS.
1382         */
1383        tcg_out_call_nodelay(s, qemu_unalign_st_trampoline, false);
1384        /* delay slot -- move to low part of argument reg */
1385        tcg_out_mov_delay(s, TCG_REG_O1, addr);
1386    } else {
1387        /* Underalignment: store by pieces of minimum alignment. */
1388        int st_opc, a_size, s_size, i;
1389
1390        /*
1391         * Force full address into T1 early; avoids problems with
1392         * overlap between @addr and @data.
1393         */
1394        tcg_out_arith(s, TCG_REG_T1, addr, index, ARITH_ADD);
1395
1396        a_size = 1 << a_bits;
1397        s_size = 1 << s_bits;
1398        if ((memop & MO_BSWAP) == MO_BE) {
1399            st_opc = qemu_st_opc[a_bits | MO_BE];
1400            for (i = 0; i < s_size; i += a_size) {
1401                TCGReg d = data;
1402                int shift = (s_size - a_size - i) * 8;
1403                if (shift) {
1404                    d = TCG_REG_T2;
1405                    tcg_out_arithi(s, d, data, shift, SHIFT_SRLX);
1406                }
1407                tcg_out_ldst(s, d, TCG_REG_T1, i, st_opc);
1408            }
1409        } else if (a_bits == 0) {
1410            tcg_out_ldst(s, data, TCG_REG_T1, 0, STB);
1411            for (i = 1; i < s_size; i++) {
1412                tcg_out_arithi(s, TCG_REG_T2, data, i * 8, SHIFT_SRLX);
1413                tcg_out_ldst(s, TCG_REG_T2, TCG_REG_T1, i, STB);
1414            }
1415        } else {
1416            /* Note that ST*A with immediate asi must use indexed address. */
1417            st_opc = qemu_st_opc[a_bits + MO_LE];
1418            tcg_out_ldst_rr(s, data, TCG_REG_T1, TCG_REG_G0, st_opc);
1419            for (i = a_size; i < s_size; i += a_size) {
1420                tcg_out_arithi(s, TCG_REG_T2, data, i * 8, SHIFT_SRLX);
1421                tcg_out_arithi(s, TCG_REG_T1, TCG_REG_T1, a_size, ARITH_ADD);
1422                tcg_out_ldst_rr(s, TCG_REG_T2, TCG_REG_T1, TCG_REG_G0, st_opc);
1423            }
1424        }
1425    }
1426
1427    *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1428#endif /* CONFIG_SOFTMMU */
1429}
1430
1431static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1432{
1433    if (check_fit_ptr(a0, 13)) {
1434        tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1435        tcg_out_movi_imm13(s, TCG_REG_O0, a0);
1436        return;
1437    } else {
1438        intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0);
1439        if (check_fit_ptr(tb_diff, 13)) {
1440            tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1441            /* Note that TCG_REG_TB has been unwound to O1.  */
1442            tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1443            return;
1444        }
1445    }
1446    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1447    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1448    tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1449}
1450
1451static void tcg_out_goto_tb(TCGContext *s, int which)
1452{
1453    ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which));
1454
1455    /* Load link and indirect branch. */
1456    set_jmp_insn_offset(s, which);
1457    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off);
1458    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1459    /* delay slot */
1460    tcg_out_nop(s);
1461    set_jmp_reset_offset(s, which);
1462
1463    /*
1464     * For the unlinked path of goto_tb, we need to reset TCG_REG_TB
1465     * to the beginning of this TB.
1466     */
1467    off = -tcg_current_code_size(s);
1468    if (check_fit_i32(off, 13)) {
1469        tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD);
1470    } else {
1471        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off);
1472        tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1473    }
1474}
1475
1476void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1477                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1478{
1479}
1480
1481static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1482                       const TCGArg args[TCG_MAX_OP_ARGS],
1483                       const int const_args[TCG_MAX_OP_ARGS])
1484{
1485    TCGArg a0, a1, a2;
1486    int c, c2;
1487
1488    /* Hoist the loads of the most common arguments.  */
1489    a0 = args[0];
1490    a1 = args[1];
1491    a2 = args[2];
1492    c2 = const_args[2];
1493
1494    switch (opc) {
1495    case INDEX_op_goto_ptr:
1496        tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1497        tcg_out_mov_delay(s, TCG_REG_TB, a0);
1498        break;
1499    case INDEX_op_br:
1500        tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0));
1501        tcg_out_nop(s);
1502        break;
1503
1504#define OP_32_64(x)                             \
1505        glue(glue(case INDEX_op_, x), _i32):    \
1506        glue(glue(case INDEX_op_, x), _i64)
1507
1508    OP_32_64(ld8u):
1509        tcg_out_ldst(s, a0, a1, a2, LDUB);
1510        break;
1511    OP_32_64(ld8s):
1512        tcg_out_ldst(s, a0, a1, a2, LDSB);
1513        break;
1514    OP_32_64(ld16u):
1515        tcg_out_ldst(s, a0, a1, a2, LDUH);
1516        break;
1517    OP_32_64(ld16s):
1518        tcg_out_ldst(s, a0, a1, a2, LDSH);
1519        break;
1520    case INDEX_op_ld_i32:
1521    case INDEX_op_ld32u_i64:
1522        tcg_out_ldst(s, a0, a1, a2, LDUW);
1523        break;
1524    OP_32_64(st8):
1525        tcg_out_ldst(s, a0, a1, a2, STB);
1526        break;
1527    OP_32_64(st16):
1528        tcg_out_ldst(s, a0, a1, a2, STH);
1529        break;
1530    case INDEX_op_st_i32:
1531    case INDEX_op_st32_i64:
1532        tcg_out_ldst(s, a0, a1, a2, STW);
1533        break;
1534    OP_32_64(add):
1535        c = ARITH_ADD;
1536        goto gen_arith;
1537    OP_32_64(sub):
1538        c = ARITH_SUB;
1539        goto gen_arith;
1540    OP_32_64(and):
1541        c = ARITH_AND;
1542        goto gen_arith;
1543    OP_32_64(andc):
1544        c = ARITH_ANDN;
1545        goto gen_arith;
1546    OP_32_64(or):
1547        c = ARITH_OR;
1548        goto gen_arith;
1549    OP_32_64(orc):
1550        c = ARITH_ORN;
1551        goto gen_arith;
1552    OP_32_64(xor):
1553        c = ARITH_XOR;
1554        goto gen_arith;
1555    case INDEX_op_shl_i32:
1556        c = SHIFT_SLL;
1557    do_shift32:
1558        /* Limit immediate shift count lest we create an illegal insn.  */
1559        tcg_out_arithc(s, a0, a1, a2 & 31, c2, c);
1560        break;
1561    case INDEX_op_shr_i32:
1562        c = SHIFT_SRL;
1563        goto do_shift32;
1564    case INDEX_op_sar_i32:
1565        c = SHIFT_SRA;
1566        goto do_shift32;
1567    case INDEX_op_mul_i32:
1568        c = ARITH_UMUL;
1569        goto gen_arith;
1570
1571    OP_32_64(neg):
1572	c = ARITH_SUB;
1573	goto gen_arith1;
1574    OP_32_64(not):
1575	c = ARITH_ORN;
1576	goto gen_arith1;
1577
1578    case INDEX_op_div_i32:
1579        tcg_out_div32(s, a0, a1, a2, c2, 0);
1580        break;
1581    case INDEX_op_divu_i32:
1582        tcg_out_div32(s, a0, a1, a2, c2, 1);
1583        break;
1584
1585    case INDEX_op_brcond_i32:
1586        tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1587        break;
1588    case INDEX_op_setcond_i32:
1589        tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2);
1590        break;
1591    case INDEX_op_movcond_i32:
1592        tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1593        break;
1594
1595    case INDEX_op_add2_i32:
1596        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1597                            args[4], const_args[4], args[5], const_args[5],
1598                            ARITH_ADDCC, ARITH_ADDC);
1599        break;
1600    case INDEX_op_sub2_i32:
1601        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1602                            args[4], const_args[4], args[5], const_args[5],
1603                            ARITH_SUBCC, ARITH_SUBC);
1604        break;
1605    case INDEX_op_mulu2_i32:
1606        c = ARITH_UMUL;
1607        goto do_mul2;
1608    case INDEX_op_muls2_i32:
1609        c = ARITH_SMUL;
1610    do_mul2:
1611        /* The 32-bit multiply insns produce a full 64-bit result. */
1612        tcg_out_arithc(s, a0, a2, args[3], const_args[3], c);
1613        tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1614        break;
1615
1616    case INDEX_op_qemu_ld_i32:
1617        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32);
1618        break;
1619    case INDEX_op_qemu_ld_i64:
1620        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64);
1621        break;
1622    case INDEX_op_qemu_st_i32:
1623        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32);
1624        break;
1625    case INDEX_op_qemu_st_i64:
1626        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64);
1627        break;
1628
1629    case INDEX_op_ld32s_i64:
1630        tcg_out_ldst(s, a0, a1, a2, LDSW);
1631        break;
1632    case INDEX_op_ld_i64:
1633        tcg_out_ldst(s, a0, a1, a2, LDX);
1634        break;
1635    case INDEX_op_st_i64:
1636        tcg_out_ldst(s, a0, a1, a2, STX);
1637        break;
1638    case INDEX_op_shl_i64:
1639        c = SHIFT_SLLX;
1640    do_shift64:
1641        /* Limit immediate shift count lest we create an illegal insn.  */
1642        tcg_out_arithc(s, a0, a1, a2 & 63, c2, c);
1643        break;
1644    case INDEX_op_shr_i64:
1645        c = SHIFT_SRLX;
1646        goto do_shift64;
1647    case INDEX_op_sar_i64:
1648        c = SHIFT_SRAX;
1649        goto do_shift64;
1650    case INDEX_op_mul_i64:
1651        c = ARITH_MULX;
1652        goto gen_arith;
1653    case INDEX_op_div_i64:
1654        c = ARITH_SDIVX;
1655        goto gen_arith;
1656    case INDEX_op_divu_i64:
1657        c = ARITH_UDIVX;
1658        goto gen_arith;
1659    case INDEX_op_extrh_i64_i32:
1660        tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX);
1661        break;
1662
1663    case INDEX_op_brcond_i64:
1664        tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1665        break;
1666    case INDEX_op_setcond_i64:
1667        tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2);
1668        break;
1669    case INDEX_op_movcond_i64:
1670        tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1671        break;
1672    case INDEX_op_add2_i64:
1673        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1674                            const_args[4], args[5], const_args[5], false);
1675        break;
1676    case INDEX_op_sub2_i64:
1677        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1678                            const_args[4], args[5], const_args[5], true);
1679        break;
1680    case INDEX_op_muluh_i64:
1681        tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI);
1682        break;
1683
1684    gen_arith:
1685        tcg_out_arithc(s, a0, a1, a2, c2, c);
1686        break;
1687
1688    gen_arith1:
1689	tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c);
1690	break;
1691
1692    case INDEX_op_mb:
1693        tcg_out_mb(s, a0);
1694        break;
1695
1696    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
1697    case INDEX_op_mov_i64:
1698    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
1699    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
1700    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
1701    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
1702    case INDEX_op_ext8s_i64:
1703    case INDEX_op_ext8u_i32:
1704    case INDEX_op_ext8u_i64:
1705    case INDEX_op_ext16s_i32:
1706    case INDEX_op_ext16s_i64:
1707    case INDEX_op_ext16u_i32:
1708    case INDEX_op_ext16u_i64:
1709    case INDEX_op_ext32s_i64:
1710    case INDEX_op_ext32u_i64:
1711    case INDEX_op_ext_i32_i64:
1712    case INDEX_op_extu_i32_i64:
1713    case INDEX_op_extrl_i64_i32:
1714    default:
1715        g_assert_not_reached();
1716    }
1717}
1718
1719static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
1720{
1721    switch (op) {
1722    case INDEX_op_goto_ptr:
1723        return C_O0_I1(r);
1724
1725    case INDEX_op_ld8u_i32:
1726    case INDEX_op_ld8u_i64:
1727    case INDEX_op_ld8s_i32:
1728    case INDEX_op_ld8s_i64:
1729    case INDEX_op_ld16u_i32:
1730    case INDEX_op_ld16u_i64:
1731    case INDEX_op_ld16s_i32:
1732    case INDEX_op_ld16s_i64:
1733    case INDEX_op_ld_i32:
1734    case INDEX_op_ld32u_i64:
1735    case INDEX_op_ld32s_i64:
1736    case INDEX_op_ld_i64:
1737    case INDEX_op_neg_i32:
1738    case INDEX_op_neg_i64:
1739    case INDEX_op_not_i32:
1740    case INDEX_op_not_i64:
1741    case INDEX_op_ext32s_i64:
1742    case INDEX_op_ext32u_i64:
1743    case INDEX_op_ext_i32_i64:
1744    case INDEX_op_extu_i32_i64:
1745    case INDEX_op_extrl_i64_i32:
1746    case INDEX_op_extrh_i64_i32:
1747        return C_O1_I1(r, r);
1748
1749    case INDEX_op_st8_i32:
1750    case INDEX_op_st8_i64:
1751    case INDEX_op_st16_i32:
1752    case INDEX_op_st16_i64:
1753    case INDEX_op_st_i32:
1754    case INDEX_op_st32_i64:
1755    case INDEX_op_st_i64:
1756        return C_O0_I2(rZ, r);
1757
1758    case INDEX_op_add_i32:
1759    case INDEX_op_add_i64:
1760    case INDEX_op_mul_i32:
1761    case INDEX_op_mul_i64:
1762    case INDEX_op_div_i32:
1763    case INDEX_op_div_i64:
1764    case INDEX_op_divu_i32:
1765    case INDEX_op_divu_i64:
1766    case INDEX_op_sub_i32:
1767    case INDEX_op_sub_i64:
1768    case INDEX_op_and_i32:
1769    case INDEX_op_and_i64:
1770    case INDEX_op_andc_i32:
1771    case INDEX_op_andc_i64:
1772    case INDEX_op_or_i32:
1773    case INDEX_op_or_i64:
1774    case INDEX_op_orc_i32:
1775    case INDEX_op_orc_i64:
1776    case INDEX_op_xor_i32:
1777    case INDEX_op_xor_i64:
1778    case INDEX_op_shl_i32:
1779    case INDEX_op_shl_i64:
1780    case INDEX_op_shr_i32:
1781    case INDEX_op_shr_i64:
1782    case INDEX_op_sar_i32:
1783    case INDEX_op_sar_i64:
1784    case INDEX_op_setcond_i32:
1785    case INDEX_op_setcond_i64:
1786        return C_O1_I2(r, rZ, rJ);
1787
1788    case INDEX_op_brcond_i32:
1789    case INDEX_op_brcond_i64:
1790        return C_O0_I2(rZ, rJ);
1791    case INDEX_op_movcond_i32:
1792    case INDEX_op_movcond_i64:
1793        return C_O1_I4(r, rZ, rJ, rI, 0);
1794    case INDEX_op_add2_i32:
1795    case INDEX_op_add2_i64:
1796    case INDEX_op_sub2_i32:
1797    case INDEX_op_sub2_i64:
1798        return C_O2_I4(r, r, rZ, rZ, rJ, rJ);
1799    case INDEX_op_mulu2_i32:
1800    case INDEX_op_muls2_i32:
1801        return C_O2_I2(r, r, rZ, rJ);
1802    case INDEX_op_muluh_i64:
1803        return C_O1_I2(r, r, r);
1804
1805    case INDEX_op_qemu_ld_i32:
1806    case INDEX_op_qemu_ld_i64:
1807        return C_O1_I1(r, s);
1808    case INDEX_op_qemu_st_i32:
1809    case INDEX_op_qemu_st_i64:
1810        return C_O0_I2(sZ, s);
1811
1812    default:
1813        g_assert_not_reached();
1814    }
1815}
1816
1817static void tcg_target_init(TCGContext *s)
1818{
1819    /*
1820     * Only probe for the platform and capabilities if we haven't already
1821     * determined maximum values at compile time.
1822     */
1823#ifndef use_vis3_instructions
1824    {
1825        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
1826        use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
1827    }
1828#endif
1829
1830    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
1831    tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;
1832
1833    tcg_target_call_clobber_regs = 0;
1834    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
1835    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
1836    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
1837    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
1838    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
1839    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
1840    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
1841    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
1842    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
1843    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
1844    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
1845    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
1846    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
1847    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
1848    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
1849
1850    s->reserved_regs = 0;
1851    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
1852    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
1853    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
1854    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
1855    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
1856    tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
1857    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
1858    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
1859}
1860
1861#define ELF_HOST_MACHINE  EM_SPARCV9
1862
1863typedef struct {
1864    DebugFrameHeader h;
1865    uint8_t fde_def_cfa[4];
1866    uint8_t fde_win_save;
1867    uint8_t fde_ret_save[3];
1868} DebugFrame;
1869
1870static const DebugFrame debug_frame = {
1871    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
1872    .h.cie.id = -1,
1873    .h.cie.version = 1,
1874    .h.cie.code_align = 1,
1875    .h.cie.data_align = -sizeof(void *) & 0x7f,
1876    .h.cie.return_column = 15,            /* o7 */
1877
1878    /* Total FDE size does not include the "len" member.  */
1879    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1880
1881    .fde_def_cfa = {
1882        12, 30,                         /* DW_CFA_def_cfa i6, 2047 */
1883        (2047 & 0x7f) | 0x80, (2047 >> 7)
1884    },
1885    .fde_win_save = 0x2d,               /* DW_CFA_GNU_window_save */
1886    .fde_ret_save = { 9, 15, 31 },      /* DW_CFA_register o7, i7 */
1887};
1888
1889void tcg_register_jit(const void *buf, size_t buf_size)
1890{
1891    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
1892}
1893