]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Tiny Code Generator for QEMU | |
3 | * | |
4 | * Copyright (c) 2008 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
25 | #include "tcg-be-null.h" | |
26 | ||
27 | #ifdef CONFIG_DEBUG_TCG | |
28 | static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { | |
29 | "%g0", | |
30 | "%g1", | |
31 | "%g2", | |
32 | "%g3", | |
33 | "%g4", | |
34 | "%g5", | |
35 | "%g6", | |
36 | "%g7", | |
37 | "%o0", | |
38 | "%o1", | |
39 | "%o2", | |
40 | "%o3", | |
41 | "%o4", | |
42 | "%o5", | |
43 | "%o6", | |
44 | "%o7", | |
45 | "%l0", | |
46 | "%l1", | |
47 | "%l2", | |
48 | "%l3", | |
49 | "%l4", | |
50 | "%l5", | |
51 | "%l6", | |
52 | "%l7", | |
53 | "%i0", | |
54 | "%i1", | |
55 | "%i2", | |
56 | "%i3", | |
57 | "%i4", | |
58 | "%i5", | |
59 | "%i6", | |
60 | "%i7", | |
61 | }; | |
62 | #endif | |
63 | ||
64 | #ifdef __arch64__ | |
65 | # define SPARC64 1 | |
66 | #else | |
67 | # define SPARC64 0 | |
68 | #endif | |
69 | ||
70 | /* Note that sparcv8plus can only hold 64 bit quantities in %g and %o | |
71 | registers. These are saved manually by the kernel in full 64-bit | |
72 | slots. The %i and %l registers are saved by the register window | |
73 | mechanism, which only allocates space for 32 bits. Given that this | |
74 | window spill/fill can happen on any signal, we must consider the | |
75 | high bits of the %i and %l registers garbage at all times. */ | |
76 | #if SPARC64 | |
77 | # define ALL_64 0xffffffffu | |
78 | #else | |
79 | # define ALL_64 0xffffu | |
80 | #endif | |
81 | ||
82 | /* Define some temporary registers. T2 is used for constant generation. */ | |
83 | #define TCG_REG_T1 TCG_REG_G1 | |
84 | #define TCG_REG_T2 TCG_REG_O7 | |
85 | ||
86 | #ifndef CONFIG_SOFTMMU | |
87 | # define TCG_GUEST_BASE_REG TCG_REG_I5 | |
88 | #endif | |
89 | ||
90 | static const int tcg_target_reg_alloc_order[] = { | |
91 | TCG_REG_L0, | |
92 | TCG_REG_L1, | |
93 | TCG_REG_L2, | |
94 | TCG_REG_L3, | |
95 | TCG_REG_L4, | |
96 | TCG_REG_L5, | |
97 | TCG_REG_L6, | |
98 | TCG_REG_L7, | |
99 | ||
100 | TCG_REG_I0, | |
101 | TCG_REG_I1, | |
102 | TCG_REG_I2, | |
103 | TCG_REG_I3, | |
104 | TCG_REG_I4, | |
105 | TCG_REG_I5, | |
106 | ||
107 | TCG_REG_G2, | |
108 | TCG_REG_G3, | |
109 | TCG_REG_G4, | |
110 | TCG_REG_G5, | |
111 | ||
112 | TCG_REG_O0, | |
113 | TCG_REG_O1, | |
114 | TCG_REG_O2, | |
115 | TCG_REG_O3, | |
116 | TCG_REG_O4, | |
117 | TCG_REG_O5, | |
118 | }; | |
119 | ||
120 | static const int tcg_target_call_iarg_regs[6] = { | |
121 | TCG_REG_O0, | |
122 | TCG_REG_O1, | |
123 | TCG_REG_O2, | |
124 | TCG_REG_O3, | |
125 | TCG_REG_O4, | |
126 | TCG_REG_O5, | |
127 | }; | |
128 | ||
129 | static const int tcg_target_call_oarg_regs[] = { | |
130 | TCG_REG_O0, | |
131 | TCG_REG_O1, | |
132 | TCG_REG_O2, | |
133 | TCG_REG_O3, | |
134 | }; | |
135 | ||
136 | #define INSN_OP(x) ((x) << 30) | |
137 | #define INSN_OP2(x) ((x) << 22) | |
138 | #define INSN_OP3(x) ((x) << 19) | |
139 | #define INSN_OPF(x) ((x) << 5) | |
140 | #define INSN_RD(x) ((x) << 25) | |
141 | #define INSN_RS1(x) ((x) << 14) | |
142 | #define INSN_RS2(x) (x) | |
143 | #define INSN_ASI(x) ((x) << 5) | |
144 | ||
145 | #define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff)) | |
146 | #define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff)) | |
147 | #define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff)) | |
148 | #define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20)) | |
149 | #define INSN_OFF19(x) (((x) >> 2) & 0x07ffff) | |
150 | #define INSN_COND(x) ((x) << 25) | |
151 | ||
152 | #define COND_N 0x0 | |
153 | #define COND_E 0x1 | |
154 | #define COND_LE 0x2 | |
155 | #define COND_L 0x3 | |
156 | #define COND_LEU 0x4 | |
157 | #define COND_CS 0x5 | |
158 | #define COND_NEG 0x6 | |
159 | #define COND_VS 0x7 | |
160 | #define COND_A 0x8 | |
161 | #define COND_NE 0x9 | |
162 | #define COND_G 0xa | |
163 | #define COND_GE 0xb | |
164 | #define COND_GU 0xc | |
165 | #define COND_CC 0xd | |
166 | #define COND_POS 0xe | |
167 | #define COND_VC 0xf | |
168 | #define BA (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2)) | |
169 | ||
170 | #define RCOND_Z 1 | |
171 | #define RCOND_LEZ 2 | |
172 | #define RCOND_LZ 3 | |
173 | #define RCOND_NZ 5 | |
174 | #define RCOND_GZ 6 | |
175 | #define RCOND_GEZ 7 | |
176 | ||
177 | #define MOVCC_ICC (1 << 18) | |
178 | #define MOVCC_XCC (1 << 18 | 1 << 12) | |
179 | ||
180 | #define BPCC_ICC 0 | |
181 | #define BPCC_XCC (2 << 20) | |
182 | #define BPCC_PT (1 << 19) | |
183 | #define BPCC_PN 0 | |
184 | #define BPCC_A (1 << 29) | |
185 | ||
186 | #define BPR_PT BPCC_PT | |
187 | ||
188 | #define ARITH_ADD (INSN_OP(2) | INSN_OP3(0x00)) | |
189 | #define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10)) | |
190 | #define ARITH_AND (INSN_OP(2) | INSN_OP3(0x01)) | |
191 | #define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05)) | |
192 | #define ARITH_OR (INSN_OP(2) | INSN_OP3(0x02)) | |
193 | #define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12)) | |
194 | #define ARITH_ORN (INSN_OP(2) | INSN_OP3(0x06)) | |
195 | #define ARITH_XOR (INSN_OP(2) | INSN_OP3(0x03)) | |
196 | #define ARITH_SUB (INSN_OP(2) | INSN_OP3(0x04)) | |
197 | #define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14)) | |
198 | #define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08)) | |
199 | #define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c)) | |
200 | #define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a)) | |
201 | #define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b)) | |
202 | #define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e)) | |
203 | #define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f)) | |
204 | #define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09)) | |
205 | #define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d)) | |
206 | #define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d)) | |
207 | #define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c)) | |
208 | #define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f)) | |
209 | ||
210 | #define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11)) | |
211 | #define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16)) | |
212 | ||
213 | #define SHIFT_SLL (INSN_OP(2) | INSN_OP3(0x25)) | |
214 | #define SHIFT_SRL (INSN_OP(2) | INSN_OP3(0x26)) | |
215 | #define SHIFT_SRA (INSN_OP(2) | INSN_OP3(0x27)) | |
216 | ||
217 | #define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12)) | |
218 | #define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12)) | |
219 | #define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12)) | |
220 | ||
221 | #define RDY (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0)) | |
222 | #define WRY (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0)) | |
223 | #define JMPL (INSN_OP(2) | INSN_OP3(0x38)) | |
224 | #define RETURN (INSN_OP(2) | INSN_OP3(0x39)) | |
225 | #define SAVE (INSN_OP(2) | INSN_OP3(0x3c)) | |
226 | #define RESTORE (INSN_OP(2) | INSN_OP3(0x3d)) | |
227 | #define SETHI (INSN_OP(0) | INSN_OP2(0x4)) | |
228 | #define CALL INSN_OP(1) | |
229 | #define LDUB (INSN_OP(3) | INSN_OP3(0x01)) | |
230 | #define LDSB (INSN_OP(3) | INSN_OP3(0x09)) | |
231 | #define LDUH (INSN_OP(3) | INSN_OP3(0x02)) | |
232 | #define LDSH (INSN_OP(3) | INSN_OP3(0x0a)) | |
233 | #define LDUW (INSN_OP(3) | INSN_OP3(0x00)) | |
234 | #define LDSW (INSN_OP(3) | INSN_OP3(0x08)) | |
235 | #define LDX (INSN_OP(3) | INSN_OP3(0x0b)) | |
236 | #define STB (INSN_OP(3) | INSN_OP3(0x05)) | |
237 | #define STH (INSN_OP(3) | INSN_OP3(0x06)) | |
238 | #define STW (INSN_OP(3) | INSN_OP3(0x04)) | |
239 | #define STX (INSN_OP(3) | INSN_OP3(0x0e)) | |
240 | #define LDUBA (INSN_OP(3) | INSN_OP3(0x11)) | |
241 | #define LDSBA (INSN_OP(3) | INSN_OP3(0x19)) | |
242 | #define LDUHA (INSN_OP(3) | INSN_OP3(0x12)) | |
243 | #define LDSHA (INSN_OP(3) | INSN_OP3(0x1a)) | |
244 | #define LDUWA (INSN_OP(3) | INSN_OP3(0x10)) | |
245 | #define LDSWA (INSN_OP(3) | INSN_OP3(0x18)) | |
246 | #define LDXA (INSN_OP(3) | INSN_OP3(0x1b)) | |
247 | #define STBA (INSN_OP(3) | INSN_OP3(0x15)) | |
248 | #define STHA (INSN_OP(3) | INSN_OP3(0x16)) | |
249 | #define STWA (INSN_OP(3) | INSN_OP3(0x14)) | |
250 | #define STXA (INSN_OP(3) | INSN_OP3(0x1e)) | |
251 | ||
252 | #ifndef ASI_PRIMARY_LITTLE | |
253 | #define ASI_PRIMARY_LITTLE 0x88 | |
254 | #endif | |
255 | ||
256 | #define LDUH_LE (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
257 | #define LDSH_LE (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
258 | #define LDUW_LE (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
259 | #define LDSW_LE (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
260 | #define LDX_LE (LDXA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
261 | ||
262 | #define STH_LE (STHA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
263 | #define STW_LE (STWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
264 | #define STX_LE (STXA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
265 | ||
266 | #ifndef use_vis3_instructions | |
267 | bool use_vis3_instructions; | |
268 | #endif | |
269 | ||
270 | static inline int check_fit_i64(int64_t val, unsigned int bits) | |
271 | { | |
272 | return val == sextract64(val, 0, bits); | |
273 | } | |
274 | ||
275 | static inline int check_fit_i32(int32_t val, unsigned int bits) | |
276 | { | |
277 | return val == sextract32(val, 0, bits); | |
278 | } | |
279 | ||
280 | #define check_fit_tl check_fit_i64 | |
281 | #if SPARC64 | |
282 | # define check_fit_ptr check_fit_i64 | |
283 | #else | |
284 | # define check_fit_ptr check_fit_i32 | |
285 | #endif | |
286 | ||
287 | static void patch_reloc(tcg_insn_unit *code_ptr, int type, | |
288 | intptr_t value, intptr_t addend) | |
289 | { | |
290 | uint32_t insn; | |
291 | ||
292 | tcg_debug_assert(addend == 0); | |
293 | value = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr); | |
294 | ||
295 | switch (type) { | |
296 | case R_SPARC_WDISP16: | |
297 | if (!check_fit_ptr(value >> 2, 16)) { | |
298 | tcg_abort(); | |
299 | } | |
300 | insn = *code_ptr; | |
301 | insn &= ~INSN_OFF16(-1); | |
302 | insn |= INSN_OFF16(value); | |
303 | *code_ptr = insn; | |
304 | break; | |
305 | case R_SPARC_WDISP19: | |
306 | if (!check_fit_ptr(value >> 2, 19)) { | |
307 | tcg_abort(); | |
308 | } | |
309 | insn = *code_ptr; | |
310 | insn &= ~INSN_OFF19(-1); | |
311 | insn |= INSN_OFF19(value); | |
312 | *code_ptr = insn; | |
313 | break; | |
314 | default: | |
315 | tcg_abort(); | |
316 | } | |
317 | } | |
318 | ||
319 | /* parse target specific constraints */ | |
320 | static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str) | |
321 | { | |
322 | const char *ct_str; | |
323 | ||
324 | ct_str = *pct_str; | |
325 | switch (ct_str[0]) { | |
326 | case 'r': | |
327 | ct->ct |= TCG_CT_REG; | |
328 | tcg_regset_set32(ct->u.regs, 0, 0xffffffff); | |
329 | break; | |
330 | case 'R': | |
331 | ct->ct |= TCG_CT_REG; | |
332 | tcg_regset_set32(ct->u.regs, 0, ALL_64); | |
333 | break; | |
334 | case 'A': /* qemu_ld/st address constraint */ | |
335 | ct->ct |= TCG_CT_REG; | |
336 | tcg_regset_set32(ct->u.regs, 0, | |
337 | TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff); | |
338 | reserve_helpers: | |
339 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O0); | |
340 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O1); | |
341 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O2); | |
342 | break; | |
343 | case 's': /* qemu_st data 32-bit constraint */ | |
344 | ct->ct |= TCG_CT_REG; | |
345 | tcg_regset_set32(ct->u.regs, 0, 0xffffffff); | |
346 | goto reserve_helpers; | |
347 | case 'S': /* qemu_st data 64-bit constraint */ | |
348 | ct->ct |= TCG_CT_REG; | |
349 | tcg_regset_set32(ct->u.regs, 0, ALL_64); | |
350 | goto reserve_helpers; | |
351 | case 'I': | |
352 | ct->ct |= TCG_CT_CONST_S11; | |
353 | break; | |
354 | case 'J': | |
355 | ct->ct |= TCG_CT_CONST_S13; | |
356 | break; | |
357 | case 'Z': | |
358 | ct->ct |= TCG_CT_CONST_ZERO; | |
359 | break; | |
360 | default: | |
361 | return -1; | |
362 | } | |
363 | ct_str++; | |
364 | *pct_str = ct_str; | |
365 | return 0; | |
366 | } | |
367 | ||
368 | /* test if a constant matches the constraint */ | |
369 | static inline int tcg_target_const_match(tcg_target_long val, TCGType type, | |
370 | const TCGArgConstraint *arg_ct) | |
371 | { | |
372 | int ct = arg_ct->ct; | |
373 | ||
374 | if (ct & TCG_CT_CONST) { | |
375 | return 1; | |
376 | } | |
377 | ||
378 | if (type == TCG_TYPE_I32) { | |
379 | val = (int32_t)val; | |
380 | } | |
381 | ||
382 | if ((ct & TCG_CT_CONST_ZERO) && val == 0) { | |
383 | return 1; | |
384 | } else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) { | |
385 | return 1; | |
386 | } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) { | |
387 | return 1; | |
388 | } else { | |
389 | return 0; | |
390 | } | |
391 | } | |
392 | ||
393 | static inline void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1, | |
394 | TCGReg rs2, int op) | |
395 | { | |
396 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2)); | |
397 | } | |
398 | ||
399 | static inline void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1, | |
400 | int32_t offset, int op) | |
401 | { | |
402 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset)); | |
403 | } | |
404 | ||
405 | static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1, | |
406 | int32_t val2, int val2const, int op) | |
407 | { | |
408 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | |
409 | | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2))); | |
410 | } | |
411 | ||
412 | static inline void tcg_out_mov(TCGContext *s, TCGType type, | |
413 | TCGReg ret, TCGReg arg) | |
414 | { | |
415 | if (ret != arg) { | |
416 | tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR); | |
417 | } | |
418 | } | |
419 | ||
420 | static inline void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg) | |
421 | { | |
422 | tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10)); | |
423 | } | |
424 | ||
425 | static inline void tcg_out_movi_imm13(TCGContext *s, TCGReg ret, int32_t arg) | |
426 | { | |
427 | tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR); | |
428 | } | |
429 | ||
430 | static void tcg_out_movi(TCGContext *s, TCGType type, | |
431 | TCGReg ret, tcg_target_long arg) | |
432 | { | |
433 | tcg_target_long hi, lo = (int32_t)arg; | |
434 | ||
435 | /* Make sure we test 32-bit constants for imm13 properly. */ | |
436 | if (type == TCG_TYPE_I32) { | |
437 | arg = lo; | |
438 | } | |
439 | ||
440 | /* A 13-bit constant sign-extended to 64-bits. */ | |
441 | if (check_fit_tl(arg, 13)) { | |
442 | tcg_out_movi_imm13(s, ret, arg); | |
443 | return; | |
444 | } | |
445 | ||
446 | /* A 32-bit constant, or 32-bit zero-extended to 64-bits. */ | |
447 | if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) { | |
448 | tcg_out_sethi(s, ret, arg); | |
449 | if (arg & 0x3ff) { | |
450 | tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR); | |
451 | } | |
452 | return; | |
453 | } | |
454 | ||
455 | /* A 32-bit constant sign-extended to 64-bits. */ | |
456 | if (arg == lo) { | |
457 | tcg_out_sethi(s, ret, ~arg); | |
458 | tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR); | |
459 | return; | |
460 | } | |
461 | ||
462 | /* A 64-bit constant decomposed into 2 32-bit pieces. */ | |
463 | if (check_fit_i32(lo, 13)) { | |
464 | hi = (arg - lo) >> 32; | |
465 | tcg_out_movi(s, TCG_TYPE_I32, ret, hi); | |
466 | tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); | |
467 | tcg_out_arithi(s, ret, ret, lo, ARITH_ADD); | |
468 | } else { | |
469 | hi = arg >> 32; | |
470 | tcg_out_movi(s, TCG_TYPE_I32, ret, hi); | |
471 | tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_T2, lo); | |
472 | tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); | |
473 | tcg_out_arith(s, ret, ret, TCG_REG_T2, ARITH_OR); | |
474 | } | |
475 | } | |
476 | ||
477 | static inline void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1, | |
478 | TCGReg a2, int op) | |
479 | { | |
480 | tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2)); | |
481 | } | |
482 | ||
483 | static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr, | |
484 | intptr_t offset, int op) | |
485 | { | |
486 | if (check_fit_ptr(offset, 13)) { | |
487 | tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) | | |
488 | INSN_IMM13(offset)); | |
489 | } else { | |
490 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset); | |
491 | tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op); | |
492 | } | |
493 | } | |
494 | ||
495 | static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, | |
496 | TCGReg arg1, intptr_t arg2) | |
497 | { | |
498 | tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX)); | |
499 | } | |
500 | ||
501 | static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, | |
502 | TCGReg arg1, intptr_t arg2) | |
503 | { | |
504 | tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX)); | |
505 | } | |
506 | ||
507 | static void tcg_out_ld_ptr(TCGContext *s, TCGReg ret, uintptr_t arg) | |
508 | { | |
509 | tcg_out_movi(s, TCG_TYPE_PTR, ret, arg & ~0x3ff); | |
510 | tcg_out_ld(s, TCG_TYPE_PTR, ret, ret, arg & 0x3ff); | |
511 | } | |
512 | ||
513 | static inline void tcg_out_sety(TCGContext *s, TCGReg rs) | |
514 | { | |
515 | tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs)); | |
516 | } | |
517 | ||
518 | static inline void tcg_out_rdy(TCGContext *s, TCGReg rd) | |
519 | { | |
520 | tcg_out32(s, RDY | INSN_RD(rd)); | |
521 | } | |
522 | ||
523 | static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1, | |
524 | int32_t val2, int val2const, int uns) | |
525 | { | |
526 | /* Load Y with the sign/zero extension of RS1 to 64-bits. */ | |
527 | if (uns) { | |
528 | tcg_out_sety(s, TCG_REG_G0); | |
529 | } else { | |
530 | tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA); | |
531 | tcg_out_sety(s, TCG_REG_T1); | |
532 | } | |
533 | ||
534 | tcg_out_arithc(s, rd, rs1, val2, val2const, | |
535 | uns ? ARITH_UDIV : ARITH_SDIV); | |
536 | } | |
537 | ||
538 | static inline void tcg_out_nop(TCGContext *s) | |
539 | { | |
540 | tcg_out_sethi(s, TCG_REG_G0, 0); | |
541 | } | |
542 | ||
543 | static const uint8_t tcg_cond_to_bcond[] = { | |
544 | [TCG_COND_EQ] = COND_E, | |
545 | [TCG_COND_NE] = COND_NE, | |
546 | [TCG_COND_LT] = COND_L, | |
547 | [TCG_COND_GE] = COND_GE, | |
548 | [TCG_COND_LE] = COND_LE, | |
549 | [TCG_COND_GT] = COND_G, | |
550 | [TCG_COND_LTU] = COND_CS, | |
551 | [TCG_COND_GEU] = COND_CC, | |
552 | [TCG_COND_LEU] = COND_LEU, | |
553 | [TCG_COND_GTU] = COND_GU, | |
554 | }; | |
555 | ||
556 | static const uint8_t tcg_cond_to_rcond[] = { | |
557 | [TCG_COND_EQ] = RCOND_Z, | |
558 | [TCG_COND_NE] = RCOND_NZ, | |
559 | [TCG_COND_LT] = RCOND_LZ, | |
560 | [TCG_COND_GT] = RCOND_GZ, | |
561 | [TCG_COND_LE] = RCOND_LEZ, | |
562 | [TCG_COND_GE] = RCOND_GEZ | |
563 | }; | |
564 | ||
565 | static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19) | |
566 | { | |
567 | tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19); | |
568 | } | |
569 | ||
570 | static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l) | |
571 | { | |
572 | int off19; | |
573 | ||
574 | if (l->has_value) { | |
575 | off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr)); | |
576 | } else { | |
577 | /* Make sure to preserve destinations during retranslation. */ | |
578 | off19 = *s->code_ptr & INSN_OFF19(-1); | |
579 | tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0); | |
580 | } | |
581 | tcg_out_bpcc0(s, scond, flags, off19); | |
582 | } | |
583 | ||
584 | static void tcg_out_cmp(TCGContext *s, TCGReg c1, int32_t c2, int c2const) | |
585 | { | |
586 | tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const, ARITH_SUBCC); | |
587 | } | |
588 | ||
589 | static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1, | |
590 | int32_t arg2, int const_arg2, TCGLabel *l) | |
591 | { | |
592 | tcg_out_cmp(s, arg1, arg2, const_arg2); | |
593 | tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l); | |
594 | tcg_out_nop(s); | |
595 | } | |
596 | ||
597 | static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret, | |
598 | int32_t v1, int v1const) | |
599 | { | |
600 | tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret) | |
601 | | INSN_RS1(tcg_cond_to_bcond[cond]) | |
602 | | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1))); | |
603 | } | |
604 | ||
605 | static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, | |
606 | TCGReg c1, int32_t c2, int c2const, | |
607 | int32_t v1, int v1const) | |
608 | { | |
609 | tcg_out_cmp(s, c1, c2, c2const); | |
610 | tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const); | |
611 | } | |
612 | ||
613 | static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1, | |
614 | int32_t arg2, int const_arg2, TCGLabel *l) | |
615 | { | |
616 | /* For 64-bit signed comparisons vs zero, we can avoid the compare. */ | |
617 | if (arg2 == 0 && !is_unsigned_cond(cond)) { | |
618 | int off16; | |
619 | ||
620 | if (l->has_value) { | |
621 | off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr)); | |
622 | } else { | |
623 | /* Make sure to preserve destinations during retranslation. */ | |
624 | off16 = *s->code_ptr & INSN_OFF16(-1); | |
625 | tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0); | |
626 | } | |
627 | tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1) | |
628 | | INSN_COND(tcg_cond_to_rcond[cond]) | off16); | |
629 | } else { | |
630 | tcg_out_cmp(s, arg1, arg2, const_arg2); | |
631 | tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l); | |
632 | } | |
633 | tcg_out_nop(s); | |
634 | } | |
635 | ||
636 | static void tcg_out_movr(TCGContext *s, TCGCond cond, TCGReg ret, TCGReg c1, | |
637 | int32_t v1, int v1const) | |
638 | { | |
639 | tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | |
640 | | (tcg_cond_to_rcond[cond] << 10) | |
641 | | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1))); | |
642 | } | |
643 | ||
644 | static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, | |
645 | TCGReg c1, int32_t c2, int c2const, | |
646 | int32_t v1, int v1const) | |
647 | { | |
648 | /* For 64-bit signed comparisons vs zero, we can avoid the compare. | |
649 | Note that the immediate range is one bit smaller, so we must check | |
650 | for that as well. */ | |
651 | if (c2 == 0 && !is_unsigned_cond(cond) | |
652 | && (!v1const || check_fit_i32(v1, 10))) { | |
653 | tcg_out_movr(s, cond, ret, c1, v1, v1const); | |
654 | } else { | |
655 | tcg_out_cmp(s, c1, c2, c2const); | |
656 | tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const); | |
657 | } | |
658 | } | |
659 | ||
660 | static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, | |
661 | TCGReg c1, int32_t c2, int c2const) | |
662 | { | |
663 | /* For 32-bit comparisons, we can play games with ADDC/SUBC. */ | |
664 | switch (cond) { | |
665 | case TCG_COND_LTU: | |
666 | case TCG_COND_GEU: | |
667 | /* The result of the comparison is in the carry bit. */ | |
668 | break; | |
669 | ||
670 | case TCG_COND_EQ: | |
671 | case TCG_COND_NE: | |
672 | /* For equality, we can transform to inequality vs zero. */ | |
673 | if (c2 != 0) { | |
674 | tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR); | |
675 | c2 = TCG_REG_T1; | |
676 | } else { | |
677 | c2 = c1; | |
678 | } | |
679 | c1 = TCG_REG_G0, c2const = 0; | |
680 | cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU); | |
681 | break; | |
682 | ||
683 | case TCG_COND_GTU: | |
684 | case TCG_COND_LEU: | |
685 | /* If we don't need to load a constant into a register, we can | |
686 | swap the operands on GTU/LEU. There's no benefit to loading | |
687 | the constant into a temporary register. */ | |
688 | if (!c2const || c2 == 0) { | |
689 | TCGReg t = c1; | |
690 | c1 = c2; | |
691 | c2 = t; | |
692 | c2const = 0; | |
693 | cond = tcg_swap_cond(cond); | |
694 | break; | |
695 | } | |
696 | /* FALLTHRU */ | |
697 | ||
698 | default: | |
699 | tcg_out_cmp(s, c1, c2, c2const); | |
700 | tcg_out_movi_imm13(s, ret, 0); | |
701 | tcg_out_movcc(s, cond, MOVCC_ICC, ret, 1, 1); | |
702 | return; | |
703 | } | |
704 | ||
705 | tcg_out_cmp(s, c1, c2, c2const); | |
706 | if (cond == TCG_COND_LTU) { | |
707 | tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC); | |
708 | } else { | |
709 | tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC); | |
710 | } | |
711 | } | |
712 | ||
713 | static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, | |
714 | TCGReg c1, int32_t c2, int c2const) | |
715 | { | |
716 | if (use_vis3_instructions) { | |
717 | switch (cond) { | |
718 | case TCG_COND_NE: | |
719 | if (c2 != 0) { | |
720 | break; | |
721 | } | |
722 | c2 = c1, c2const = 0, c1 = TCG_REG_G0; | |
723 | /* FALLTHRU */ | |
724 | case TCG_COND_LTU: | |
725 | tcg_out_cmp(s, c1, c2, c2const); | |
726 | tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC); | |
727 | return; | |
728 | default: | |
729 | break; | |
730 | } | |
731 | } | |
732 | ||
733 | /* For 64-bit signed comparisons vs zero, we can avoid the compare | |
734 | if the input does not overlap the output. */ | |
735 | if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) { | |
736 | tcg_out_movi_imm13(s, ret, 0); | |
737 | tcg_out_movr(s, cond, ret, c1, 1, 1); | |
738 | } else { | |
739 | tcg_out_cmp(s, c1, c2, c2const); | |
740 | tcg_out_movi_imm13(s, ret, 0); | |
741 | tcg_out_movcc(s, cond, MOVCC_XCC, ret, 1, 1); | |
742 | } | |
743 | } | |
744 | ||
745 | static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh, | |
746 | TCGReg al, TCGReg ah, int32_t bl, int blconst, | |
747 | int32_t bh, int bhconst, int opl, int oph) | |
748 | { | |
749 | TCGReg tmp = TCG_REG_T1; | |
750 | ||
751 | /* Note that the low parts are fully consumed before tmp is set. */ | |
752 | if (rl != ah && (bhconst || rl != bh)) { | |
753 | tmp = rl; | |
754 | } | |
755 | ||
756 | tcg_out_arithc(s, tmp, al, bl, blconst, opl); | |
757 | tcg_out_arithc(s, rh, ah, bh, bhconst, oph); | |
758 | tcg_out_mov(s, TCG_TYPE_I32, rl, tmp); | |
759 | } | |
760 | ||
761 | static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh, | |
762 | TCGReg al, TCGReg ah, int32_t bl, int blconst, | |
763 | int32_t bh, int bhconst, bool is_sub) | |
764 | { | |
765 | TCGReg tmp = TCG_REG_T1; | |
766 | ||
767 | /* Note that the low parts are fully consumed before tmp is set. */ | |
768 | if (rl != ah && (bhconst || rl != bh)) { | |
769 | tmp = rl; | |
770 | } | |
771 | ||
772 | tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC); | |
773 | ||
774 | if (use_vis3_instructions && !is_sub) { | |
775 | /* Note that ADDXC doesn't accept immediates. */ | |
776 | if (bhconst && bh != 0) { | |
777 | tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh); | |
778 | bh = TCG_REG_T2; | |
779 | } | |
780 | tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC); | |
781 | } else if (bh == TCG_REG_G0) { | |
782 | /* If we have a zero, we can perform the operation in two insns, | |
783 | with the arithmetic first, and a conditional move into place. */ | |
784 | if (rh == ah) { | |
785 | tcg_out_arithi(s, TCG_REG_T2, ah, 1, | |
786 | is_sub ? ARITH_SUB : ARITH_ADD); | |
787 | tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0); | |
788 | } else { | |
789 | tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD); | |
790 | tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0); | |
791 | } | |
792 | } else { | |
793 | /* Otherwise adjust BH as if there is carry into T2 ... */ | |
794 | if (bhconst) { | |
795 | tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh + (is_sub ? -1 : 1)); | |
796 | } else { | |
797 | tcg_out_arithi(s, TCG_REG_T2, bh, 1, | |
798 | is_sub ? ARITH_SUB : ARITH_ADD); | |
799 | } | |
800 | /* ... smoosh T2 back to original BH if carry is clear ... */ | |
801 | tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst); | |
802 | /* ... and finally perform the arithmetic with the new operand. */ | |
803 | tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD); | |
804 | } | |
805 | ||
806 | tcg_out_mov(s, TCG_TYPE_I64, rl, tmp); | |
807 | } | |
808 | ||
809 | static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest) | |
810 | { | |
811 | ptrdiff_t disp = tcg_pcrel_diff(s, dest); | |
812 | ||
813 | if (disp == (int32_t)disp) { | |
814 | tcg_out32(s, CALL | (uint32_t)disp >> 2); | |
815 | } else { | |
816 | uintptr_t desti = (uintptr_t)dest; | |
817 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, desti & ~0xfff); | |
818 | tcg_out_arithi(s, TCG_REG_O7, TCG_REG_T1, desti & 0xfff, JMPL); | |
819 | } | |
820 | } | |
821 | ||
822 | static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest) | |
823 | { | |
824 | tcg_out_call_nodelay(s, dest); | |
825 | tcg_out_nop(s); | |
826 | } | |
827 | ||
828 | #ifdef CONFIG_SOFTMMU | |
829 | static tcg_insn_unit *qemu_ld_trampoline[16]; | |
830 | static tcg_insn_unit *qemu_st_trampoline[16]; | |
831 | ||
832 | static void build_trampolines(TCGContext *s) | |
833 | { | |
834 | static void * const qemu_ld_helpers[16] = { | |
835 | [MO_UB] = helper_ret_ldub_mmu, | |
836 | [MO_SB] = helper_ret_ldsb_mmu, | |
837 | [MO_LEUW] = helper_le_lduw_mmu, | |
838 | [MO_LESW] = helper_le_ldsw_mmu, | |
839 | [MO_LEUL] = helper_le_ldul_mmu, | |
840 | [MO_LEQ] = helper_le_ldq_mmu, | |
841 | [MO_BEUW] = helper_be_lduw_mmu, | |
842 | [MO_BESW] = helper_be_ldsw_mmu, | |
843 | [MO_BEUL] = helper_be_ldul_mmu, | |
844 | [MO_BEQ] = helper_be_ldq_mmu, | |
845 | }; | |
846 | static void * const qemu_st_helpers[16] = { | |
847 | [MO_UB] = helper_ret_stb_mmu, | |
848 | [MO_LEUW] = helper_le_stw_mmu, | |
849 | [MO_LEUL] = helper_le_stl_mmu, | |
850 | [MO_LEQ] = helper_le_stq_mmu, | |
851 | [MO_BEUW] = helper_be_stw_mmu, | |
852 | [MO_BEUL] = helper_be_stl_mmu, | |
853 | [MO_BEQ] = helper_be_stq_mmu, | |
854 | }; | |
855 | ||
856 | int i; | |
857 | TCGReg ra; | |
858 | ||
859 | for (i = 0; i < 16; ++i) { | |
860 | if (qemu_ld_helpers[i] == NULL) { | |
861 | continue; | |
862 | } | |
863 | ||
864 | /* May as well align the trampoline. */ | |
865 | while ((uintptr_t)s->code_ptr & 15) { | |
866 | tcg_out_nop(s); | |
867 | } | |
868 | qemu_ld_trampoline[i] = s->code_ptr; | |
869 | ||
870 | if (SPARC64 || TARGET_LONG_BITS == 32) { | |
871 | ra = TCG_REG_O3; | |
872 | } else { | |
873 | /* Install the high part of the address. */ | |
874 | tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O2, 32, SHIFT_SRLX); | |
875 | ra = TCG_REG_O4; | |
876 | } | |
877 | ||
878 | /* Set the retaddr operand. */ | |
879 | tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7); | |
880 | /* Set the env operand. */ | |
881 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0); | |
882 | /* Tail call. */ | |
883 | tcg_out_call_nodelay(s, qemu_ld_helpers[i]); | |
884 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra); | |
885 | } | |
886 | ||
887 | for (i = 0; i < 16; ++i) { | |
888 | if (qemu_st_helpers[i] == NULL) { | |
889 | continue; | |
890 | } | |
891 | ||
892 | /* May as well align the trampoline. */ | |
893 | while ((uintptr_t)s->code_ptr & 15) { | |
894 | tcg_out_nop(s); | |
895 | } | |
896 | qemu_st_trampoline[i] = s->code_ptr; | |
897 | ||
898 | if (SPARC64) { | |
899 | ra = TCG_REG_O4; | |
900 | } else { | |
901 | ra = TCG_REG_O1; | |
902 | if (TARGET_LONG_BITS == 64) { | |
903 | /* Install the high part of the address. */ | |
904 | tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX); | |
905 | ra += 2; | |
906 | } else { | |
907 | ra += 1; | |
908 | } | |
909 | if ((i & MO_SIZE) == MO_64) { | |
910 | /* Install the high part of the data. */ | |
911 | tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX); | |
912 | ra += 2; | |
913 | } else { | |
914 | ra += 1; | |
915 | } | |
916 | /* Skip the oi argument. */ | |
917 | ra += 1; | |
918 | } | |
919 | ||
920 | /* Set the retaddr operand. */ | |
921 | if (ra >= TCG_REG_O6) { | |
922 | tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_O7, TCG_REG_CALL_STACK, | |
923 | TCG_TARGET_CALL_STACK_OFFSET); | |
924 | ra = TCG_REG_G1; | |
925 | } | |
926 | tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7); | |
927 | /* Set the env operand. */ | |
928 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0); | |
929 | /* Tail call. */ | |
930 | tcg_out_call_nodelay(s, qemu_st_helpers[i]); | |
931 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra); | |
932 | } | |
933 | } | |
934 | #endif | |
935 | ||
936 | /* Generate global QEMU prologue and epilogue code */ | |
937 | static void tcg_target_qemu_prologue(TCGContext *s) | |
938 | { | |
939 | int tmp_buf_size, frame_size; | |
940 | ||
941 | /* The TCG temp buffer is at the top of the frame, immediately | |
942 | below the frame pointer. */ | |
943 | tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long); | |
944 | tcg_set_frame(s, TCG_REG_I6, TCG_TARGET_STACK_BIAS - tmp_buf_size, | |
945 | tmp_buf_size); | |
946 | ||
947 | /* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is | |
948 | otherwise the minimal frame usable by callees. */ | |
949 | frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS; | |
950 | frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size; | |
951 | frame_size += TCG_TARGET_STACK_ALIGN - 1; | |
952 | frame_size &= -TCG_TARGET_STACK_ALIGN; | |
953 | tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) | | |
954 | INSN_IMM13(-frame_size)); | |
955 | ||
956 | #ifndef CONFIG_SOFTMMU | |
957 | if (guest_base != 0) { | |
958 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base); | |
959 | tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); | |
960 | } | |
961 | #endif | |
962 | ||
963 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL); | |
964 | /* delay slot */ | |
965 | tcg_out_nop(s); | |
966 | ||
967 | /* No epilogue required. We issue ret + restore directly in the TB. */ | |
968 | ||
969 | #ifdef CONFIG_SOFTMMU | |
970 | build_trampolines(s); | |
971 | #endif | |
972 | } | |
973 | ||
974 | #if defined(CONFIG_SOFTMMU) | |
975 | /* Perform the TLB load and compare. | |
976 | ||
977 | Inputs: | |
978 | ADDRLO and ADDRHI contain the possible two parts of the address. | |
979 | ||
980 | MEM_INDEX and S_BITS are the memory context and log2 size of the load. | |
981 | ||
982 | WHICH is the offset into the CPUTLBEntry structure of the slot to read. | |
983 | This should be offsetof addr_read or addr_write. | |
984 | ||
985 | The result of the TLB comparison is in %[ix]cc. The sanitized address | |
986 | is in the returned register, maybe %o0. The TLB addend is in %o1. */ | |
987 | ||
988 | static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addr, int mem_index, | |
989 | TCGMemOp s_bits, int which) | |
990 | { | |
991 | const TCGReg r0 = TCG_REG_O0; | |
992 | const TCGReg r1 = TCG_REG_O1; | |
993 | const TCGReg r2 = TCG_REG_O2; | |
994 | int tlb_ofs; | |
995 | ||
996 | /* Shift the page number down. */ | |
997 | tcg_out_arithi(s, r1, addr, TARGET_PAGE_BITS, SHIFT_SRL); | |
998 | ||
999 | /* Mask out the page offset, except for the required alignment. */ | |
1000 | tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_T1, | |
1001 | TARGET_PAGE_MASK | ((1 << s_bits) - 1)); | |
1002 | ||
1003 | /* Mask the tlb index. */ | |
1004 | tcg_out_arithi(s, r1, r1, CPU_TLB_SIZE - 1, ARITH_AND); | |
1005 | ||
1006 | /* Mask page, part 2. */ | |
1007 | tcg_out_arith(s, r0, addr, TCG_REG_T1, ARITH_AND); | |
1008 | ||
1009 | /* Shift the tlb index into place. */ | |
1010 | tcg_out_arithi(s, r1, r1, CPU_TLB_ENTRY_BITS, SHIFT_SLL); | |
1011 | ||
1012 | /* Relative to the current ENV. */ | |
1013 | tcg_out_arith(s, r1, TCG_AREG0, r1, ARITH_ADD); | |
1014 | ||
1015 | /* Find a base address that can load both tlb comparator and addend. */ | |
1016 | tlb_ofs = offsetof(CPUArchState, tlb_table[mem_index][0]); | |
1017 | if (!check_fit_ptr(tlb_ofs + sizeof(CPUTLBEntry), 13)) { | |
1018 | if (tlb_ofs & ~0x3ff) { | |
1019 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, tlb_ofs & ~0x3ff); | |
1020 | tcg_out_arith(s, r1, r1, TCG_REG_T1, ARITH_ADD); | |
1021 | } | |
1022 | tlb_ofs &= 0x3ff; | |
1023 | } | |
1024 | ||
1025 | /* Load the tlb comparator and the addend. */ | |
1026 | tcg_out_ld(s, TCG_TYPE_TL, r2, r1, tlb_ofs + which); | |
1027 | tcg_out_ld(s, TCG_TYPE_PTR, r1, r1, tlb_ofs+offsetof(CPUTLBEntry, addend)); | |
1028 | ||
1029 | /* subcc arg0, arg2, %g0 */ | |
1030 | tcg_out_cmp(s, r0, r2, 0); | |
1031 | ||
1032 | /* If the guest address must be zero-extended, do so now. */ | |
1033 | if (SPARC64 && TARGET_LONG_BITS == 32) { | |
1034 | tcg_out_arithi(s, r0, addr, 0, SHIFT_SRL); | |
1035 | return r0; | |
1036 | } | |
1037 | return addr; | |
1038 | } | |
1039 | #endif /* CONFIG_SOFTMMU */ | |
1040 | ||
1041 | static const int qemu_ld_opc[16] = { | |
1042 | [MO_UB] = LDUB, | |
1043 | [MO_SB] = LDSB, | |
1044 | ||
1045 | [MO_BEUW] = LDUH, | |
1046 | [MO_BESW] = LDSH, | |
1047 | [MO_BEUL] = LDUW, | |
1048 | [MO_BESL] = LDSW, | |
1049 | [MO_BEQ] = LDX, | |
1050 | ||
1051 | [MO_LEUW] = LDUH_LE, | |
1052 | [MO_LESW] = LDSH_LE, | |
1053 | [MO_LEUL] = LDUW_LE, | |
1054 | [MO_LESL] = LDSW_LE, | |
1055 | [MO_LEQ] = LDX_LE, | |
1056 | }; | |
1057 | ||
1058 | static const int qemu_st_opc[16] = { | |
1059 | [MO_UB] = STB, | |
1060 | ||
1061 | [MO_BEUW] = STH, | |
1062 | [MO_BEUL] = STW, | |
1063 | [MO_BEQ] = STX, | |
1064 | ||
1065 | [MO_LEUW] = STH_LE, | |
1066 | [MO_LEUL] = STW_LE, | |
1067 | [MO_LEQ] = STX_LE, | |
1068 | }; | |
1069 | ||
1070 | static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr, | |
1071 | TCGMemOpIdx oi, bool is_64) | |
1072 | { | |
1073 | TCGMemOp memop = get_memop(oi); | |
1074 | #ifdef CONFIG_SOFTMMU | |
1075 | unsigned memi = get_mmuidx(oi); | |
1076 | TCGReg addrz, param; | |
1077 | tcg_insn_unit *func; | |
1078 | tcg_insn_unit *label_ptr; | |
1079 | ||
1080 | addrz = tcg_out_tlb_load(s, addr, memi, memop & MO_SIZE, | |
1081 | offsetof(CPUTLBEntry, addr_read)); | |
1082 | ||
1083 | /* The fast path is exactly one insn. Thus we can perform the | |
1084 | entire TLB Hit in the (annulled) delay slot of the branch | |
1085 | over the TLB Miss case. */ | |
1086 | ||
1087 | /* beq,a,pt %[xi]cc, label0 */ | |
1088 | label_ptr = s->code_ptr; | |
1089 | tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT | |
1090 | | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0); | |
1091 | /* delay slot */ | |
1092 | tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1, | |
1093 | qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]); | |
1094 | ||
1095 | /* TLB Miss. */ | |
1096 | ||
1097 | param = TCG_REG_O1; | |
1098 | if (!SPARC64 && TARGET_LONG_BITS == 64) { | |
1099 | /* Skip the high-part; we'll perform the extract in the trampoline. */ | |
1100 | param++; | |
1101 | } | |
1102 | tcg_out_mov(s, TCG_TYPE_REG, param++, addr); | |
1103 | ||
1104 | /* We use the helpers to extend SB and SW data, leaving the case | |
1105 | of SL needing explicit extending below. */ | |
1106 | if ((memop & MO_SSIZE) == MO_SL) { | |
1107 | func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SIZE)]; | |
1108 | } else { | |
1109 | func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SSIZE)]; | |
1110 | } | |
1111 | tcg_debug_assert(func != NULL); | |
1112 | tcg_out_call_nodelay(s, func); | |
1113 | /* delay slot */ | |
1114 | tcg_out_movi(s, TCG_TYPE_I32, param, oi); | |
1115 | ||
1116 | /* Recall that all of the helpers return 64-bit results. | |
1117 | Which complicates things for sparcv8plus. */ | |
1118 | if (SPARC64) { | |
1119 | /* We let the helper sign-extend SB and SW, but leave SL for here. */ | |
1120 | if (is_64 && (memop & MO_SSIZE) == MO_SL) { | |
1121 | tcg_out_arithi(s, data, TCG_REG_O0, 0, SHIFT_SRA); | |
1122 | } else { | |
1123 | tcg_out_mov(s, TCG_TYPE_REG, data, TCG_REG_O0); | |
1124 | } | |
1125 | } else { | |
1126 | if ((memop & MO_SIZE) == MO_64) { | |
1127 | tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, 32, SHIFT_SLLX); | |
1128 | tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O1, 0, SHIFT_SRL); | |
1129 | tcg_out_arith(s, data, TCG_REG_O0, TCG_REG_O1, ARITH_OR); | |
1130 | } else if (is_64) { | |
1131 | /* Re-extend from 32-bit rather than reassembling when we | |
1132 | know the high register must be an extension. */ | |
1133 | tcg_out_arithi(s, data, TCG_REG_O1, 0, | |
1134 | memop & MO_SIGN ? SHIFT_SRA : SHIFT_SRL); | |
1135 | } else { | |
1136 | tcg_out_mov(s, TCG_TYPE_I32, data, TCG_REG_O1); | |
1137 | } | |
1138 | } | |
1139 | ||
1140 | *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr)); | |
1141 | #else | |
1142 | if (SPARC64 && TARGET_LONG_BITS == 32) { | |
1143 | tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL); | |
1144 | addr = TCG_REG_T1; | |
1145 | } | |
1146 | tcg_out_ldst_rr(s, data, addr, | |
1147 | (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0), | |
1148 | qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]); | |
1149 | #endif /* CONFIG_SOFTMMU */ | |
1150 | } | |
1151 | ||
1152 | static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr, | |
1153 | TCGMemOpIdx oi) | |
1154 | { | |
1155 | TCGMemOp memop = get_memop(oi); | |
1156 | #ifdef CONFIG_SOFTMMU | |
1157 | unsigned memi = get_mmuidx(oi); | |
1158 | TCGReg addrz, param; | |
1159 | tcg_insn_unit *func; | |
1160 | tcg_insn_unit *label_ptr; | |
1161 | ||
1162 | addrz = tcg_out_tlb_load(s, addr, memi, memop & MO_SIZE, | |
1163 | offsetof(CPUTLBEntry, addr_write)); | |
1164 | ||
1165 | /* The fast path is exactly one insn. Thus we can perform the entire | |
1166 | TLB Hit in the (annulled) delay slot of the branch over TLB Miss. */ | |
1167 | /* beq,a,pt %[xi]cc, label0 */ | |
1168 | label_ptr = s->code_ptr; | |
1169 | tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT | |
1170 | | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0); | |
1171 | /* delay slot */ | |
1172 | tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1, | |
1173 | qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]); | |
1174 | ||
1175 | /* TLB Miss. */ | |
1176 | ||
1177 | param = TCG_REG_O1; | |
1178 | if (!SPARC64 && TARGET_LONG_BITS == 64) { | |
1179 | /* Skip the high-part; we'll perform the extract in the trampoline. */ | |
1180 | param++; | |
1181 | } | |
1182 | tcg_out_mov(s, TCG_TYPE_REG, param++, addr); | |
1183 | if (!SPARC64 && (memop & MO_SIZE) == MO_64) { | |
1184 | /* Skip the high-part; we'll perform the extract in the trampoline. */ | |
1185 | param++; | |
1186 | } | |
1187 | tcg_out_mov(s, TCG_TYPE_REG, param++, data); | |
1188 | ||
1189 | func = qemu_st_trampoline[memop & (MO_BSWAP | MO_SIZE)]; | |
1190 | tcg_debug_assert(func != NULL); | |
1191 | tcg_out_call_nodelay(s, func); | |
1192 | /* delay slot */ | |
1193 | tcg_out_movi(s, TCG_TYPE_I32, param, oi); | |
1194 | ||
1195 | *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr)); | |
1196 | #else | |
1197 | if (SPARC64 && TARGET_LONG_BITS == 32) { | |
1198 | tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL); | |
1199 | addr = TCG_REG_T1; | |
1200 | } | |
1201 | tcg_out_ldst_rr(s, data, addr, | |
1202 | (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0), | |
1203 | qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]); | |
1204 | #endif /* CONFIG_SOFTMMU */ | |
1205 | } | |
1206 | ||
1207 | static void tcg_out_op(TCGContext *s, TCGOpcode opc, | |
1208 | const TCGArg args[TCG_MAX_OP_ARGS], | |
1209 | const int const_args[TCG_MAX_OP_ARGS]) | |
1210 | { | |
1211 | TCGArg a0, a1, a2; | |
1212 | int c, c2; | |
1213 | ||
1214 | /* Hoist the loads of the most common arguments. */ | |
1215 | a0 = args[0]; | |
1216 | a1 = args[1]; | |
1217 | a2 = args[2]; | |
1218 | c2 = const_args[2]; | |
1219 | ||
1220 | switch (opc) { | |
1221 | case INDEX_op_exit_tb: | |
1222 | if (check_fit_ptr(a0, 13)) { | |
1223 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); | |
1224 | tcg_out_movi_imm13(s, TCG_REG_O0, a0); | |
1225 | } else { | |
1226 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff); | |
1227 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); | |
1228 | tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR); | |
1229 | } | |
1230 | break; | |
1231 | case INDEX_op_goto_tb: | |
1232 | if (s->tb_jmp_offset) { | |
1233 | /* direct jump method */ | |
1234 | s->tb_jmp_offset[a0] = tcg_current_code_size(s); | |
1235 | /* Make sure to preserve links during retranslation. */ | |
1236 | tcg_out32(s, CALL | (*s->code_ptr & ~INSN_OP(-1))); | |
1237 | } else { | |
1238 | /* indirect jump method */ | |
1239 | tcg_out_ld_ptr(s, TCG_REG_T1, (uintptr_t)(s->tb_next + a0)); | |
1240 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_T1, 0, JMPL); | |
1241 | } | |
1242 | tcg_out_nop(s); | |
1243 | s->tb_next_offset[a0] = tcg_current_code_size(s); | |
1244 | break; | |
1245 | case INDEX_op_br: | |
1246 | tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0)); | |
1247 | tcg_out_nop(s); | |
1248 | break; | |
1249 | ||
1250 | #define OP_32_64(x) \ | |
1251 | glue(glue(case INDEX_op_, x), _i32): \ | |
1252 | glue(glue(case INDEX_op_, x), _i64) | |
1253 | ||
1254 | OP_32_64(ld8u): | |
1255 | tcg_out_ldst(s, a0, a1, a2, LDUB); | |
1256 | break; | |
1257 | OP_32_64(ld8s): | |
1258 | tcg_out_ldst(s, a0, a1, a2, LDSB); | |
1259 | break; | |
1260 | OP_32_64(ld16u): | |
1261 | tcg_out_ldst(s, a0, a1, a2, LDUH); | |
1262 | break; | |
1263 | OP_32_64(ld16s): | |
1264 | tcg_out_ldst(s, a0, a1, a2, LDSH); | |
1265 | break; | |
1266 | case INDEX_op_ld_i32: | |
1267 | case INDEX_op_ld32u_i64: | |
1268 | tcg_out_ldst(s, a0, a1, a2, LDUW); | |
1269 | break; | |
1270 | OP_32_64(st8): | |
1271 | tcg_out_ldst(s, a0, a1, a2, STB); | |
1272 | break; | |
1273 | OP_32_64(st16): | |
1274 | tcg_out_ldst(s, a0, a1, a2, STH); | |
1275 | break; | |
1276 | case INDEX_op_st_i32: | |
1277 | case INDEX_op_st32_i64: | |
1278 | tcg_out_ldst(s, a0, a1, a2, STW); | |
1279 | break; | |
1280 | OP_32_64(add): | |
1281 | c = ARITH_ADD; | |
1282 | goto gen_arith; | |
1283 | OP_32_64(sub): | |
1284 | c = ARITH_SUB; | |
1285 | goto gen_arith; | |
1286 | OP_32_64(and): | |
1287 | c = ARITH_AND; | |
1288 | goto gen_arith; | |
1289 | OP_32_64(andc): | |
1290 | c = ARITH_ANDN; | |
1291 | goto gen_arith; | |
1292 | OP_32_64(or): | |
1293 | c = ARITH_OR; | |
1294 | goto gen_arith; | |
1295 | OP_32_64(orc): | |
1296 | c = ARITH_ORN; | |
1297 | goto gen_arith; | |
1298 | OP_32_64(xor): | |
1299 | c = ARITH_XOR; | |
1300 | goto gen_arith; | |
1301 | case INDEX_op_shl_i32: | |
1302 | c = SHIFT_SLL; | |
1303 | do_shift32: | |
1304 | /* Limit immediate shift count lest we create an illegal insn. */ | |
1305 | tcg_out_arithc(s, a0, a1, a2 & 31, c2, c); | |
1306 | break; | |
1307 | case INDEX_op_shr_i32: | |
1308 | c = SHIFT_SRL; | |
1309 | goto do_shift32; | |
1310 | case INDEX_op_sar_i32: | |
1311 | c = SHIFT_SRA; | |
1312 | goto do_shift32; | |
1313 | case INDEX_op_mul_i32: | |
1314 | c = ARITH_UMUL; | |
1315 | goto gen_arith; | |
1316 | ||
1317 | OP_32_64(neg): | |
1318 | c = ARITH_SUB; | |
1319 | goto gen_arith1; | |
1320 | OP_32_64(not): | |
1321 | c = ARITH_ORN; | |
1322 | goto gen_arith1; | |
1323 | ||
1324 | case INDEX_op_div_i32: | |
1325 | tcg_out_div32(s, a0, a1, a2, c2, 0); | |
1326 | break; | |
1327 | case INDEX_op_divu_i32: | |
1328 | tcg_out_div32(s, a0, a1, a2, c2, 1); | |
1329 | break; | |
1330 | ||
1331 | case INDEX_op_brcond_i32: | |
1332 | tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3])); | |
1333 | break; | |
1334 | case INDEX_op_setcond_i32: | |
1335 | tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2); | |
1336 | break; | |
1337 | case INDEX_op_movcond_i32: | |
1338 | tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); | |
1339 | break; | |
1340 | ||
1341 | case INDEX_op_add2_i32: | |
1342 | tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], | |
1343 | args[4], const_args[4], args[5], const_args[5], | |
1344 | ARITH_ADDCC, ARITH_ADDC); | |
1345 | break; | |
1346 | case INDEX_op_sub2_i32: | |
1347 | tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], | |
1348 | args[4], const_args[4], args[5], const_args[5], | |
1349 | ARITH_SUBCC, ARITH_SUBC); | |
1350 | break; | |
1351 | case INDEX_op_mulu2_i32: | |
1352 | c = ARITH_UMUL; | |
1353 | goto do_mul2; | |
1354 | case INDEX_op_muls2_i32: | |
1355 | c = ARITH_SMUL; | |
1356 | do_mul2: | |
1357 | /* The 32-bit multiply insns produce a full 64-bit result. If the | |
1358 | destination register can hold it, we can avoid the slower RDY. */ | |
1359 | tcg_out_arithc(s, a0, a2, args[3], const_args[3], c); | |
1360 | if (SPARC64 || a0 <= TCG_REG_O7) { | |
1361 | tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX); | |
1362 | } else { | |
1363 | tcg_out_rdy(s, a1); | |
1364 | } | |
1365 | break; | |
1366 | ||
1367 | case INDEX_op_qemu_ld_i32: | |
1368 | tcg_out_qemu_ld(s, a0, a1, a2, false); | |
1369 | break; | |
1370 | case INDEX_op_qemu_ld_i64: | |
1371 | tcg_out_qemu_ld(s, a0, a1, a2, true); | |
1372 | break; | |
1373 | case INDEX_op_qemu_st_i32: | |
1374 | case INDEX_op_qemu_st_i64: | |
1375 | tcg_out_qemu_st(s, a0, a1, a2); | |
1376 | break; | |
1377 | ||
1378 | case INDEX_op_ld32s_i64: | |
1379 | tcg_out_ldst(s, a0, a1, a2, LDSW); | |
1380 | break; | |
1381 | case INDEX_op_ld_i64: | |
1382 | tcg_out_ldst(s, a0, a1, a2, LDX); | |
1383 | break; | |
1384 | case INDEX_op_st_i64: | |
1385 | tcg_out_ldst(s, a0, a1, a2, STX); | |
1386 | break; | |
1387 | case INDEX_op_shl_i64: | |
1388 | c = SHIFT_SLLX; | |
1389 | do_shift64: | |
1390 | /* Limit immediate shift count lest we create an illegal insn. */ | |
1391 | tcg_out_arithc(s, a0, a1, a2 & 63, c2, c); | |
1392 | break; | |
1393 | case INDEX_op_shr_i64: | |
1394 | c = SHIFT_SRLX; | |
1395 | goto do_shift64; | |
1396 | case INDEX_op_sar_i64: | |
1397 | c = SHIFT_SRAX; | |
1398 | goto do_shift64; | |
1399 | case INDEX_op_mul_i64: | |
1400 | c = ARITH_MULX; | |
1401 | goto gen_arith; | |
1402 | case INDEX_op_div_i64: | |
1403 | c = ARITH_SDIVX; | |
1404 | goto gen_arith; | |
1405 | case INDEX_op_divu_i64: | |
1406 | c = ARITH_UDIVX; | |
1407 | goto gen_arith; | |
1408 | case INDEX_op_ext_i32_i64: | |
1409 | case INDEX_op_ext32s_i64: | |
1410 | tcg_out_arithi(s, a0, a1, 0, SHIFT_SRA); | |
1411 | break; | |
1412 | case INDEX_op_extu_i32_i64: | |
1413 | case INDEX_op_ext32u_i64: | |
1414 | tcg_out_arithi(s, a0, a1, 0, SHIFT_SRL); | |
1415 | break; | |
1416 | case INDEX_op_extrl_i64_i32: | |
1417 | tcg_out_mov(s, TCG_TYPE_I32, a0, a1); | |
1418 | break; | |
1419 | case INDEX_op_extrh_i64_i32: | |
1420 | tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX); | |
1421 | break; | |
1422 | ||
1423 | case INDEX_op_brcond_i64: | |
1424 | tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3])); | |
1425 | break; | |
1426 | case INDEX_op_setcond_i64: | |
1427 | tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2); | |
1428 | break; | |
1429 | case INDEX_op_movcond_i64: | |
1430 | tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); | |
1431 | break; | |
1432 | case INDEX_op_add2_i64: | |
1433 | tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], | |
1434 | const_args[4], args[5], const_args[5], false); | |
1435 | break; | |
1436 | case INDEX_op_sub2_i64: | |
1437 | tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], | |
1438 | const_args[4], args[5], const_args[5], true); | |
1439 | break; | |
1440 | case INDEX_op_muluh_i64: | |
1441 | tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI); | |
1442 | break; | |
1443 | ||
1444 | gen_arith: | |
1445 | tcg_out_arithc(s, a0, a1, a2, c2, c); | |
1446 | break; | |
1447 | ||
1448 | gen_arith1: | |
1449 | tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c); | |
1450 | break; | |
1451 | ||
1452 | case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ | |
1453 | case INDEX_op_mov_i64: | |
1454 | case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */ | |
1455 | case INDEX_op_movi_i64: | |
1456 | case INDEX_op_call: /* Always emitted via tcg_out_call. */ | |
1457 | default: | |
1458 | tcg_abort(); | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | static const TCGTargetOpDef sparc_op_defs[] = { | |
1463 | { INDEX_op_exit_tb, { } }, | |
1464 | { INDEX_op_goto_tb, { } }, | |
1465 | { INDEX_op_br, { } }, | |
1466 | ||
1467 | { INDEX_op_ld8u_i32, { "r", "r" } }, | |
1468 | { INDEX_op_ld8s_i32, { "r", "r" } }, | |
1469 | { INDEX_op_ld16u_i32, { "r", "r" } }, | |
1470 | { INDEX_op_ld16s_i32, { "r", "r" } }, | |
1471 | { INDEX_op_ld_i32, { "r", "r" } }, | |
1472 | { INDEX_op_st8_i32, { "rZ", "r" } }, | |
1473 | { INDEX_op_st16_i32, { "rZ", "r" } }, | |
1474 | { INDEX_op_st_i32, { "rZ", "r" } }, | |
1475 | ||
1476 | { INDEX_op_add_i32, { "r", "rZ", "rJ" } }, | |
1477 | { INDEX_op_mul_i32, { "r", "rZ", "rJ" } }, | |
1478 | { INDEX_op_div_i32, { "r", "rZ", "rJ" } }, | |
1479 | { INDEX_op_divu_i32, { "r", "rZ", "rJ" } }, | |
1480 | { INDEX_op_sub_i32, { "r", "rZ", "rJ" } }, | |
1481 | { INDEX_op_and_i32, { "r", "rZ", "rJ" } }, | |
1482 | { INDEX_op_andc_i32, { "r", "rZ", "rJ" } }, | |
1483 | { INDEX_op_or_i32, { "r", "rZ", "rJ" } }, | |
1484 | { INDEX_op_orc_i32, { "r", "rZ", "rJ" } }, | |
1485 | { INDEX_op_xor_i32, { "r", "rZ", "rJ" } }, | |
1486 | ||
1487 | { INDEX_op_shl_i32, { "r", "rZ", "rJ" } }, | |
1488 | { INDEX_op_shr_i32, { "r", "rZ", "rJ" } }, | |
1489 | { INDEX_op_sar_i32, { "r", "rZ", "rJ" } }, | |
1490 | ||
1491 | { INDEX_op_neg_i32, { "r", "rJ" } }, | |
1492 | { INDEX_op_not_i32, { "r", "rJ" } }, | |
1493 | ||
1494 | { INDEX_op_brcond_i32, { "rZ", "rJ" } }, | |
1495 | { INDEX_op_setcond_i32, { "r", "rZ", "rJ" } }, | |
1496 | { INDEX_op_movcond_i32, { "r", "rZ", "rJ", "rI", "0" } }, | |
1497 | ||
1498 | { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rJ", "rJ" } }, | |
1499 | { INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rJ", "rJ" } }, | |
1500 | { INDEX_op_mulu2_i32, { "r", "r", "rZ", "rJ" } }, | |
1501 | { INDEX_op_muls2_i32, { "r", "r", "rZ", "rJ" } }, | |
1502 | ||
1503 | { INDEX_op_ld8u_i64, { "R", "r" } }, | |
1504 | { INDEX_op_ld8s_i64, { "R", "r" } }, | |
1505 | { INDEX_op_ld16u_i64, { "R", "r" } }, | |
1506 | { INDEX_op_ld16s_i64, { "R", "r" } }, | |
1507 | { INDEX_op_ld32u_i64, { "R", "r" } }, | |
1508 | { INDEX_op_ld32s_i64, { "R", "r" } }, | |
1509 | { INDEX_op_ld_i64, { "R", "r" } }, | |
1510 | { INDEX_op_st8_i64, { "RZ", "r" } }, | |
1511 | { INDEX_op_st16_i64, { "RZ", "r" } }, | |
1512 | { INDEX_op_st32_i64, { "RZ", "r" } }, | |
1513 | { INDEX_op_st_i64, { "RZ", "r" } }, | |
1514 | ||
1515 | { INDEX_op_add_i64, { "R", "RZ", "RJ" } }, | |
1516 | { INDEX_op_mul_i64, { "R", "RZ", "RJ" } }, | |
1517 | { INDEX_op_div_i64, { "R", "RZ", "RJ" } }, | |
1518 | { INDEX_op_divu_i64, { "R", "RZ", "RJ" } }, | |
1519 | { INDEX_op_sub_i64, { "R", "RZ", "RJ" } }, | |
1520 | { INDEX_op_and_i64, { "R", "RZ", "RJ" } }, | |
1521 | { INDEX_op_andc_i64, { "R", "RZ", "RJ" } }, | |
1522 | { INDEX_op_or_i64, { "R", "RZ", "RJ" } }, | |
1523 | { INDEX_op_orc_i64, { "R", "RZ", "RJ" } }, | |
1524 | { INDEX_op_xor_i64, { "R", "RZ", "RJ" } }, | |
1525 | ||
1526 | { INDEX_op_shl_i64, { "R", "RZ", "RJ" } }, | |
1527 | { INDEX_op_shr_i64, { "R", "RZ", "RJ" } }, | |
1528 | { INDEX_op_sar_i64, { "R", "RZ", "RJ" } }, | |
1529 | ||
1530 | { INDEX_op_neg_i64, { "R", "RJ" } }, | |
1531 | { INDEX_op_not_i64, { "R", "RJ" } }, | |
1532 | ||
1533 | { INDEX_op_ext32s_i64, { "R", "R" } }, | |
1534 | { INDEX_op_ext32u_i64, { "R", "R" } }, | |
1535 | { INDEX_op_ext_i32_i64, { "R", "r" } }, | |
1536 | { INDEX_op_extu_i32_i64, { "R", "r" } }, | |
1537 | { INDEX_op_extrl_i64_i32, { "r", "R" } }, | |
1538 | { INDEX_op_extrh_i64_i32, { "r", "R" } }, | |
1539 | ||
1540 | { INDEX_op_brcond_i64, { "RZ", "RJ" } }, | |
1541 | { INDEX_op_setcond_i64, { "R", "RZ", "RJ" } }, | |
1542 | { INDEX_op_movcond_i64, { "R", "RZ", "RJ", "RI", "0" } }, | |
1543 | ||
1544 | { INDEX_op_add2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } }, | |
1545 | { INDEX_op_sub2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } }, | |
1546 | { INDEX_op_muluh_i64, { "R", "RZ", "RZ" } }, | |
1547 | ||
1548 | { INDEX_op_qemu_ld_i32, { "r", "A" } }, | |
1549 | { INDEX_op_qemu_ld_i64, { "R", "A" } }, | |
1550 | { INDEX_op_qemu_st_i32, { "sZ", "A" } }, | |
1551 | { INDEX_op_qemu_st_i64, { "SZ", "A" } }, | |
1552 | ||
1553 | { -1 }, | |
1554 | }; | |
1555 | ||
1556 | static void tcg_target_init(TCGContext *s) | |
1557 | { | |
1558 | /* Only probe for the platform and capabilities if we havn't already | |
1559 | determined maximum values at compile time. */ | |
1560 | #ifndef use_vis3_instructions | |
1561 | { | |
1562 | unsigned long hwcap = qemu_getauxval(AT_HWCAP); | |
1563 | use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0; | |
1564 | } | |
1565 | #endif | |
1566 | ||
1567 | tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffffffff); | |
1568 | tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, ALL_64); | |
1569 | ||
1570 | tcg_regset_set32(tcg_target_call_clobber_regs, 0, | |
1571 | (1 << TCG_REG_G1) | | |
1572 | (1 << TCG_REG_G2) | | |
1573 | (1 << TCG_REG_G3) | | |
1574 | (1 << TCG_REG_G4) | | |
1575 | (1 << TCG_REG_G5) | | |
1576 | (1 << TCG_REG_G6) | | |
1577 | (1 << TCG_REG_G7) | | |
1578 | (1 << TCG_REG_O0) | | |
1579 | (1 << TCG_REG_O1) | | |
1580 | (1 << TCG_REG_O2) | | |
1581 | (1 << TCG_REG_O3) | | |
1582 | (1 << TCG_REG_O4) | | |
1583 | (1 << TCG_REG_O5) | | |
1584 | (1 << TCG_REG_O7)); | |
1585 | ||
1586 | tcg_regset_clear(s->reserved_regs); | |
1587 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */ | |
1588 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */ | |
1589 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */ | |
1590 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */ | |
1591 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */ | |
1592 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */ | |
1593 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */ | |
1594 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */ | |
1595 | ||
1596 | tcg_add_target_add_op_defs(sparc_op_defs); | |
1597 | } | |
1598 | ||
1599 | #if SPARC64 | |
1600 | # define ELF_HOST_MACHINE EM_SPARCV9 | |
1601 | #else | |
1602 | # define ELF_HOST_MACHINE EM_SPARC32PLUS | |
1603 | # define ELF_HOST_FLAGS EF_SPARC_32PLUS | |
1604 | #endif | |
1605 | ||
1606 | typedef struct { | |
1607 | DebugFrameHeader h; | |
1608 | uint8_t fde_def_cfa[SPARC64 ? 4 : 2]; | |
1609 | uint8_t fde_win_save; | |
1610 | uint8_t fde_ret_save[3]; | |
1611 | } DebugFrame; | |
1612 | ||
1613 | static const DebugFrame debug_frame = { | |
1614 | .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */ | |
1615 | .h.cie.id = -1, | |
1616 | .h.cie.version = 1, | |
1617 | .h.cie.code_align = 1, | |
1618 | .h.cie.data_align = -sizeof(void *) & 0x7f, | |
1619 | .h.cie.return_column = 15, /* o7 */ | |
1620 | ||
1621 | /* Total FDE size does not include the "len" member. */ | |
1622 | .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset), | |
1623 | ||
1624 | .fde_def_cfa = { | |
1625 | #if SPARC64 | |
1626 | 12, 30, /* DW_CFA_def_cfa i6, 2047 */ | |
1627 | (2047 & 0x7f) | 0x80, (2047 >> 7) | |
1628 | #else | |
1629 | 13, 30 /* DW_CFA_def_cfa_register i6 */ | |
1630 | #endif | |
1631 | }, | |
1632 | .fde_win_save = 0x2d, /* DW_CFA_GNU_window_save */ | |
1633 | .fde_ret_save = { 9, 15, 31 }, /* DW_CFA_register o7, i7 */ | |
1634 | }; | |
1635 | ||
1636 | void tcg_register_jit(void *buf, size_t buf_size) | |
1637 | { | |
1638 | tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); | |
1639 | } | |
1640 | ||
1641 | void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr) | |
1642 | { | |
1643 | uint32_t *ptr = (uint32_t *)jmp_addr; | |
1644 | uintptr_t disp = addr - jmp_addr; | |
1645 | ||
1646 | /* We can reach the entire address space for 32-bit. For 64-bit | |
1647 | the code_gen_buffer can't be larger than 2GB. */ | |
1648 | tcg_debug_assert(disp == (int32_t)disp); | |
1649 | ||
1650 | atomic_set(ptr, deposit32(CALL, 0, 30, disp >> 2)); | |
1651 | flush_icache_range(jmp_addr, jmp_addr + 4); | |
1652 | } |