| 1 | /* |
| 2 | * Copyright (C) 2014-2020 Paul Cercueil <paul@crapouillou.net> |
| 3 | * |
| 4 | * This library is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU Lesser General Public |
| 6 | * License as published by the Free Software Foundation; either |
| 7 | * version 2.1 of the License, or (at your option) any later version. |
| 8 | * |
| 9 | * This library is distributed in the hope that it will be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 12 | * Lesser General Public License for more details. |
| 13 | */ |
| 14 | |
| 15 | #include "blockcache.h" |
| 16 | #include "config.h" |
| 17 | #include "debug.h" |
| 18 | #include "disassembler.h" |
| 19 | #include "emitter.h" |
| 20 | #include "interpreter.h" |
| 21 | #include "lightrec.h" |
| 22 | #include "memmanager.h" |
| 23 | #include "reaper.h" |
| 24 | #include "recompiler.h" |
| 25 | #include "regcache.h" |
| 26 | #include "optimizer.h" |
| 27 | |
| 28 | #include <errno.h> |
| 29 | #include <lightning.h> |
| 30 | #include <limits.h> |
| 31 | #if ENABLE_THREADED_COMPILER |
| 32 | #include <stdatomic.h> |
| 33 | #endif |
| 34 | #include <stdbool.h> |
| 35 | #include <stddef.h> |
| 36 | #include <string.h> |
| 37 | #if ENABLE_TINYMM |
| 38 | #include <tinymm.h> |
| 39 | #endif |
| 40 | |
| 41 | #define GENMASK(h, l) \ |
| 42 | (((uintptr_t)-1 << (l)) & ((uintptr_t)-1 >> (__WORDSIZE - 1 - (h)))) |
| 43 | |
| 44 | static struct block * lightrec_precompile_block(struct lightrec_state *state, |
| 45 | u32 pc); |
| 46 | |
| 47 | static void lightrec_default_sb(struct lightrec_state *state, u32 opcode, |
| 48 | void *host, u32 addr, u8 data) |
| 49 | { |
| 50 | *(u8 *)host = data; |
| 51 | |
| 52 | if (!state->invalidate_from_dma_only) |
| 53 | lightrec_invalidate(state, addr, 1); |
| 54 | } |
| 55 | |
| 56 | static void lightrec_default_sh(struct lightrec_state *state, u32 opcode, |
| 57 | void *host, u32 addr, u16 data) |
| 58 | { |
| 59 | *(u16 *)host = HTOLE16(data); |
| 60 | |
| 61 | if (!state->invalidate_from_dma_only) |
| 62 | lightrec_invalidate(state, addr, 2); |
| 63 | } |
| 64 | |
| 65 | static void lightrec_default_sw(struct lightrec_state *state, u32 opcode, |
| 66 | void *host, u32 addr, u32 data) |
| 67 | { |
| 68 | *(u32 *)host = HTOLE32(data); |
| 69 | |
| 70 | if (!state->invalidate_from_dma_only) |
| 71 | lightrec_invalidate(state, addr, 4); |
| 72 | } |
| 73 | |
| 74 | static u8 lightrec_default_lb(struct lightrec_state *state, |
| 75 | u32 opcode, void *host, u32 addr) |
| 76 | { |
| 77 | return *(u8 *)host; |
| 78 | } |
| 79 | |
| 80 | static u16 lightrec_default_lh(struct lightrec_state *state, |
| 81 | u32 opcode, void *host, u32 addr) |
| 82 | { |
| 83 | return LE16TOH(*(u16 *)host); |
| 84 | } |
| 85 | |
| 86 | static u32 lightrec_default_lw(struct lightrec_state *state, |
| 87 | u32 opcode, void *host, u32 addr) |
| 88 | { |
| 89 | return LE32TOH(*(u32 *)host); |
| 90 | } |
| 91 | |
| 92 | static const struct lightrec_mem_map_ops lightrec_default_ops = { |
| 93 | .sb = lightrec_default_sb, |
| 94 | .sh = lightrec_default_sh, |
| 95 | .sw = lightrec_default_sw, |
| 96 | .lb = lightrec_default_lb, |
| 97 | .lh = lightrec_default_lh, |
| 98 | .lw = lightrec_default_lw, |
| 99 | }; |
| 100 | |
| 101 | static void __segfault_cb(struct lightrec_state *state, u32 addr) |
| 102 | { |
| 103 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT); |
| 104 | pr_err("Segmentation fault in recompiled code: invalid " |
| 105 | "load/store at address 0x%08x\n", addr); |
| 106 | } |
| 107 | |
| 108 | static void lightrec_swl(struct lightrec_state *state, |
| 109 | const struct lightrec_mem_map_ops *ops, |
| 110 | u32 opcode, void *host, u32 addr, u32 data) |
| 111 | { |
| 112 | unsigned int shift = addr & 0x3; |
| 113 | unsigned int mask = GENMASK(31, (shift + 1) * 8); |
| 114 | u32 old_data; |
| 115 | |
| 116 | /* Align to 32 bits */ |
| 117 | addr &= ~3; |
| 118 | host = (void *)((uintptr_t)host & ~3); |
| 119 | |
| 120 | old_data = ops->lw(state, opcode, host, addr); |
| 121 | |
| 122 | data = (data >> ((3 - shift) * 8)) | (old_data & mask); |
| 123 | |
| 124 | ops->sw(state, opcode, host, addr, data); |
| 125 | } |
| 126 | |
| 127 | static void lightrec_swr(struct lightrec_state *state, |
| 128 | const struct lightrec_mem_map_ops *ops, |
| 129 | u32 opcode, void *host, u32 addr, u32 data) |
| 130 | { |
| 131 | unsigned int shift = addr & 0x3; |
| 132 | unsigned int mask = (1 << (shift * 8)) - 1; |
| 133 | u32 old_data; |
| 134 | |
| 135 | /* Align to 32 bits */ |
| 136 | addr &= ~3; |
| 137 | host = (void *)((uintptr_t)host & ~3); |
| 138 | |
| 139 | old_data = ops->lw(state, opcode, host, addr); |
| 140 | |
| 141 | data = (data << (shift * 8)) | (old_data & mask); |
| 142 | |
| 143 | ops->sw(state, opcode, host, addr, data); |
| 144 | } |
| 145 | |
| 146 | static void lightrec_swc2(struct lightrec_state *state, union code op, |
| 147 | const struct lightrec_mem_map_ops *ops, |
| 148 | void *host, u32 addr) |
| 149 | { |
| 150 | u32 data = state->ops.cop2_ops.mfc(state, op.opcode, op.i.rt); |
| 151 | |
| 152 | ops->sw(state, op.opcode, host, addr, data); |
| 153 | } |
| 154 | |
| 155 | static u32 lightrec_lwl(struct lightrec_state *state, |
| 156 | const struct lightrec_mem_map_ops *ops, |
| 157 | u32 opcode, void *host, u32 addr, u32 data) |
| 158 | { |
| 159 | unsigned int shift = addr & 0x3; |
| 160 | unsigned int mask = (1 << (24 - shift * 8)) - 1; |
| 161 | u32 old_data; |
| 162 | |
| 163 | /* Align to 32 bits */ |
| 164 | addr &= ~3; |
| 165 | host = (void *)((uintptr_t)host & ~3); |
| 166 | |
| 167 | old_data = ops->lw(state, opcode, host, addr); |
| 168 | |
| 169 | return (data & mask) | (old_data << (24 - shift * 8)); |
| 170 | } |
| 171 | |
| 172 | static u32 lightrec_lwr(struct lightrec_state *state, |
| 173 | const struct lightrec_mem_map_ops *ops, |
| 174 | u32 opcode, void *host, u32 addr, u32 data) |
| 175 | { |
| 176 | unsigned int shift = addr & 0x3; |
| 177 | unsigned int mask = GENMASK(31, 32 - shift * 8); |
| 178 | u32 old_data; |
| 179 | |
| 180 | /* Align to 32 bits */ |
| 181 | addr &= ~3; |
| 182 | host = (void *)((uintptr_t)host & ~3); |
| 183 | |
| 184 | old_data = ops->lw(state, opcode, host, addr); |
| 185 | |
| 186 | return (data & mask) | (old_data >> (shift * 8)); |
| 187 | } |
| 188 | |
| 189 | static void lightrec_lwc2(struct lightrec_state *state, union code op, |
| 190 | const struct lightrec_mem_map_ops *ops, |
| 191 | void *host, u32 addr) |
| 192 | { |
| 193 | u32 data = ops->lw(state, op.opcode, host, addr); |
| 194 | |
| 195 | state->ops.cop2_ops.mtc(state, op.opcode, op.i.rt, data); |
| 196 | } |
| 197 | |
| 198 | static void lightrec_invalidate_map(struct lightrec_state *state, |
| 199 | const struct lightrec_mem_map *map, u32 addr) |
| 200 | { |
| 201 | if (map == &state->maps[PSX_MAP_KERNEL_USER_RAM]) |
| 202 | state->code_lut[lut_offset(addr)] = NULL; |
| 203 | } |
| 204 | |
| 205 | static const struct lightrec_mem_map * |
| 206 | lightrec_get_map(struct lightrec_state *state, u32 kaddr) |
| 207 | { |
| 208 | unsigned int i; |
| 209 | |
| 210 | for (i = 0; i < state->nb_maps; i++) { |
| 211 | const struct lightrec_mem_map *map = &state->maps[i]; |
| 212 | |
| 213 | if (kaddr >= map->pc && kaddr < map->pc + map->length) |
| 214 | return map; |
| 215 | } |
| 216 | |
| 217 | return NULL; |
| 218 | } |
| 219 | |
| 220 | u32 lightrec_rw(struct lightrec_state *state, union code op, |
| 221 | u32 addr, u32 data, u16 *flags) |
| 222 | { |
| 223 | const struct lightrec_mem_map *map; |
| 224 | const struct lightrec_mem_map_ops *ops; |
| 225 | u32 kaddr, pc, opcode = op.opcode; |
| 226 | void *host; |
| 227 | |
| 228 | addr += (s16) op.i.imm; |
| 229 | kaddr = kunseg(addr); |
| 230 | |
| 231 | map = lightrec_get_map(state, kaddr); |
| 232 | if (!map) { |
| 233 | __segfault_cb(state, addr); |
| 234 | return 0; |
| 235 | } |
| 236 | |
| 237 | pc = map->pc; |
| 238 | |
| 239 | while (map->mirror_of) |
| 240 | map = map->mirror_of; |
| 241 | |
| 242 | host = (void *)((uintptr_t)map->address + kaddr - pc); |
| 243 | |
| 244 | if (unlikely(map->ops)) { |
| 245 | if (flags) |
| 246 | *flags |= LIGHTREC_HW_IO; |
| 247 | |
| 248 | ops = map->ops; |
| 249 | } else { |
| 250 | if (flags) |
| 251 | *flags |= LIGHTREC_DIRECT_IO; |
| 252 | |
| 253 | ops = &lightrec_default_ops; |
| 254 | } |
| 255 | |
| 256 | switch (op.i.op) { |
| 257 | case OP_SB: |
| 258 | ops->sb(state, opcode, host, addr, (u8) data); |
| 259 | return 0; |
| 260 | case OP_SH: |
| 261 | ops->sh(state, opcode, host, addr, (u16) data); |
| 262 | return 0; |
| 263 | case OP_SWL: |
| 264 | lightrec_swl(state, ops, opcode, host, addr, data); |
| 265 | return 0; |
| 266 | case OP_SWR: |
| 267 | lightrec_swr(state, ops, opcode, host, addr, data); |
| 268 | return 0; |
| 269 | case OP_SW: |
| 270 | ops->sw(state, opcode, host, addr, data); |
| 271 | return 0; |
| 272 | case OP_SWC2: |
| 273 | lightrec_swc2(state, op, ops, host, addr); |
| 274 | return 0; |
| 275 | case OP_LB: |
| 276 | return (s32) (s8) ops->lb(state, opcode, host, addr); |
| 277 | case OP_LBU: |
| 278 | return ops->lb(state, opcode, host, addr); |
| 279 | case OP_LH: |
| 280 | return (s32) (s16) ops->lh(state, opcode, host, addr); |
| 281 | case OP_LHU: |
| 282 | return ops->lh(state, opcode, host, addr); |
| 283 | case OP_LWC2: |
| 284 | lightrec_lwc2(state, op, ops, host, addr); |
| 285 | return 0; |
| 286 | case OP_LWL: |
| 287 | return lightrec_lwl(state, ops, opcode, host, addr, data); |
| 288 | case OP_LWR: |
| 289 | return lightrec_lwr(state, ops, opcode, host, addr, data); |
| 290 | case OP_LW: |
| 291 | default: |
| 292 | return ops->lw(state, opcode, host, addr); |
| 293 | } |
| 294 | } |
| 295 | |
| 296 | static void lightrec_rw_helper(struct lightrec_state *state, |
| 297 | union code op, u16 *flags) |
| 298 | { |
| 299 | u32 ret = lightrec_rw(state, op, |
| 300 | state->native_reg_cache[op.i.rs], |
| 301 | state->native_reg_cache[op.i.rt], flags); |
| 302 | |
| 303 | switch (op.i.op) { |
| 304 | case OP_LB: |
| 305 | case OP_LBU: |
| 306 | case OP_LH: |
| 307 | case OP_LHU: |
| 308 | case OP_LWL: |
| 309 | case OP_LWR: |
| 310 | case OP_LW: |
| 311 | if (op.i.rt) |
| 312 | state->native_reg_cache[op.i.rt] = ret; |
| 313 | default: /* fall-through */ |
| 314 | break; |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | static void lightrec_rw_cb(struct lightrec_state *state, union code op) |
| 319 | { |
| 320 | lightrec_rw_helper(state, op, NULL); |
| 321 | } |
| 322 | |
| 323 | static void lightrec_rw_generic_cb(struct lightrec_state *state, |
| 324 | struct opcode *op, struct block *block) |
| 325 | { |
| 326 | bool was_tagged = op->flags & (LIGHTREC_HW_IO | LIGHTREC_DIRECT_IO); |
| 327 | |
| 328 | lightrec_rw_helper(state, op->c, &op->flags); |
| 329 | |
| 330 | if (!was_tagged) { |
| 331 | pr_debug("Opcode of block at PC 0x%08x offset 0x%x has been " |
| 332 | "tagged - flag for recompilation\n", |
| 333 | block->pc, op->offset << 2); |
| 334 | |
| 335 | block->flags |= BLOCK_SHOULD_RECOMPILE; |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | u32 lightrec_mfc(struct lightrec_state *state, union code op) |
| 340 | { |
| 341 | bool is_cfc = (op.i.op == OP_CP0 && op.r.rs == OP_CP0_CFC0) || |
| 342 | (op.i.op == OP_CP2 && op.r.rs == OP_CP2_BASIC_CFC2); |
| 343 | u32 (*func)(struct lightrec_state *, u32, u8); |
| 344 | const struct lightrec_cop_ops *ops; |
| 345 | |
| 346 | if (op.i.op == OP_CP0) |
| 347 | ops = &state->ops.cop0_ops; |
| 348 | else |
| 349 | ops = &state->ops.cop2_ops; |
| 350 | |
| 351 | if (is_cfc) |
| 352 | func = ops->cfc; |
| 353 | else |
| 354 | func = ops->mfc; |
| 355 | |
| 356 | return (*func)(state, op.opcode, op.r.rd); |
| 357 | } |
| 358 | |
| 359 | static void lightrec_mfc_cb(struct lightrec_state *state, union code op) |
| 360 | { |
| 361 | u32 rt = lightrec_mfc(state, op); |
| 362 | |
| 363 | if (op.r.rt) |
| 364 | state->native_reg_cache[op.r.rt] = rt; |
| 365 | } |
| 366 | |
| 367 | void lightrec_mtc(struct lightrec_state *state, union code op, u32 data) |
| 368 | { |
| 369 | bool is_ctc = (op.i.op == OP_CP0 && op.r.rs == OP_CP0_CTC0) || |
| 370 | (op.i.op == OP_CP2 && op.r.rs == OP_CP2_BASIC_CTC2); |
| 371 | void (*func)(struct lightrec_state *, u32, u8, u32); |
| 372 | const struct lightrec_cop_ops *ops; |
| 373 | |
| 374 | if (op.i.op == OP_CP0) |
| 375 | ops = &state->ops.cop0_ops; |
| 376 | else |
| 377 | ops = &state->ops.cop2_ops; |
| 378 | |
| 379 | if (is_ctc) |
| 380 | func = ops->ctc; |
| 381 | else |
| 382 | func = ops->mtc; |
| 383 | |
| 384 | (*func)(state, op.opcode, op.r.rd, data); |
| 385 | } |
| 386 | |
| 387 | static void lightrec_mtc_cb(struct lightrec_state *state, union code op) |
| 388 | { |
| 389 | lightrec_mtc(state, op, state->native_reg_cache[op.r.rt]); |
| 390 | } |
| 391 | |
| 392 | static void lightrec_rfe_cb(struct lightrec_state *state, union code op) |
| 393 | { |
| 394 | u32 status; |
| 395 | |
| 396 | /* Read CP0 Status register (r12) */ |
| 397 | status = state->ops.cop0_ops.mfc(state, op.opcode, 12); |
| 398 | |
| 399 | /* Switch the bits */ |
| 400 | status = ((status & 0x3c) >> 2) | (status & ~0xf); |
| 401 | |
| 402 | /* Write it back */ |
| 403 | state->ops.cop0_ops.ctc(state, op.opcode, 12, status); |
| 404 | } |
| 405 | |
| 406 | static void lightrec_cp_cb(struct lightrec_state *state, union code op) |
| 407 | { |
| 408 | void (*func)(struct lightrec_state *, u32); |
| 409 | |
| 410 | if ((op.opcode >> 25) & 1) |
| 411 | func = state->ops.cop2_ops.op; |
| 412 | else |
| 413 | func = state->ops.cop0_ops.op; |
| 414 | |
| 415 | (*func)(state, op.opcode); |
| 416 | } |
| 417 | |
| 418 | static void lightrec_syscall_cb(struct lightrec_state *state, union code op) |
| 419 | { |
| 420 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SYSCALL); |
| 421 | } |
| 422 | |
| 423 | static void lightrec_break_cb(struct lightrec_state *state, union code op) |
| 424 | { |
| 425 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_BREAK); |
| 426 | } |
| 427 | |
| 428 | struct block * lightrec_get_block(struct lightrec_state *state, u32 pc) |
| 429 | { |
| 430 | struct block *block = lightrec_find_block(state->block_cache, pc); |
| 431 | |
| 432 | if (block && lightrec_block_is_outdated(block)) { |
| 433 | pr_debug("Block at PC 0x%08x is outdated!\n", block->pc); |
| 434 | |
| 435 | /* Make sure the recompiler isn't processing the block we'll |
| 436 | * destroy */ |
| 437 | if (ENABLE_THREADED_COMPILER) |
| 438 | lightrec_recompiler_remove(state->rec, block); |
| 439 | |
| 440 | lightrec_unregister_block(state->block_cache, block); |
| 441 | remove_from_code_lut(state->block_cache, block); |
| 442 | lightrec_free_block(block); |
| 443 | block = NULL; |
| 444 | } |
| 445 | |
| 446 | if (!block) { |
| 447 | block = lightrec_precompile_block(state, pc); |
| 448 | if (!block) { |
| 449 | pr_err("Unable to recompile block at PC 0x%x\n", pc); |
| 450 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT); |
| 451 | return NULL; |
| 452 | } |
| 453 | |
| 454 | lightrec_register_block(state->block_cache, block); |
| 455 | } |
| 456 | |
| 457 | return block; |
| 458 | } |
| 459 | |
| 460 | static void * get_next_block_func(struct lightrec_state *state, u32 pc) |
| 461 | { |
| 462 | struct block *block; |
| 463 | bool should_recompile; |
| 464 | void *func; |
| 465 | |
| 466 | for (;;) { |
| 467 | func = state->code_lut[lut_offset(pc)]; |
| 468 | if (func && func != state->get_next_block) |
| 469 | return func; |
| 470 | |
| 471 | block = lightrec_get_block(state, pc); |
| 472 | |
| 473 | if (unlikely(!block)) |
| 474 | return NULL; |
| 475 | |
| 476 | should_recompile = block->flags & BLOCK_SHOULD_RECOMPILE && |
| 477 | !(block->flags & BLOCK_IS_DEAD); |
| 478 | |
| 479 | if (unlikely(should_recompile)) { |
| 480 | pr_debug("Block at PC 0x%08x should recompile\n", pc); |
| 481 | |
| 482 | lightrec_unregister(MEM_FOR_CODE, block->code_size); |
| 483 | |
| 484 | if (ENABLE_THREADED_COMPILER) |
| 485 | lightrec_recompiler_add(state->rec, block); |
| 486 | else |
| 487 | lightrec_compile_block(block); |
| 488 | } |
| 489 | |
| 490 | if (ENABLE_THREADED_COMPILER && likely(!should_recompile)) |
| 491 | func = lightrec_recompiler_run_first_pass(block, &pc); |
| 492 | else |
| 493 | func = block->function; |
| 494 | |
| 495 | if (likely(func)) |
| 496 | return func; |
| 497 | |
| 498 | /* Block wasn't compiled yet - run the interpreter */ |
| 499 | if (!ENABLE_THREADED_COMPILER && |
| 500 | ((ENABLE_FIRST_PASS && likely(!should_recompile)) || |
| 501 | unlikely(block->flags & BLOCK_NEVER_COMPILE))) |
| 502 | pc = lightrec_emulate_block(block, pc); |
| 503 | |
| 504 | if (likely(!(block->flags & BLOCK_NEVER_COMPILE))) { |
| 505 | /* Then compile it using the profiled data */ |
| 506 | if (ENABLE_THREADED_COMPILER) |
| 507 | lightrec_recompiler_add(state->rec, block); |
| 508 | else |
| 509 | lightrec_compile_block(block); |
| 510 | } |
| 511 | |
| 512 | if (state->exit_flags != LIGHTREC_EXIT_NORMAL || |
| 513 | state->current_cycle >= state->target_cycle) { |
| 514 | state->next_pc = pc; |
| 515 | return NULL; |
| 516 | } |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | static s32 c_generic_function_wrapper(struct lightrec_state *state, |
| 521 | s32 cycles_delta, |
| 522 | void (*f)(struct lightrec_state *, |
| 523 | struct opcode *, |
| 524 | struct block *), |
| 525 | struct opcode *op, struct block *block) |
| 526 | { |
| 527 | state->current_cycle = state->target_cycle - cycles_delta; |
| 528 | |
| 529 | (*f)(state, op, block); |
| 530 | |
| 531 | return state->target_cycle - state->current_cycle; |
| 532 | } |
| 533 | |
| 534 | static s32 c_function_wrapper(struct lightrec_state *state, s32 cycles_delta, |
| 535 | void (*f)(struct lightrec_state *, union code), |
| 536 | union code op) |
| 537 | { |
| 538 | state->current_cycle = state->target_cycle - cycles_delta; |
| 539 | |
| 540 | (*f)(state, op); |
| 541 | |
| 542 | return state->target_cycle - state->current_cycle; |
| 543 | } |
| 544 | |
| 545 | static struct block * generate_wrapper(struct lightrec_state *state, |
| 546 | void *f, bool generic) |
| 547 | { |
| 548 | struct block *block; |
| 549 | jit_state_t *_jit; |
| 550 | unsigned int i; |
| 551 | int stack_ptr; |
| 552 | jit_word_t code_size; |
| 553 | jit_node_t *to_tramp, *to_fn_epilog; |
| 554 | |
| 555 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 556 | if (!block) |
| 557 | goto err_no_mem; |
| 558 | |
| 559 | _jit = jit_new_state(); |
| 560 | if (!_jit) |
| 561 | goto err_free_block; |
| 562 | |
| 563 | jit_name("RW wrapper"); |
| 564 | jit_note(__FILE__, __LINE__); |
| 565 | |
| 566 | /* Wrapper entry point */ |
| 567 | jit_prolog(); |
| 568 | |
| 569 | stack_ptr = jit_allocai(sizeof(uintptr_t) * NUM_TEMPS); |
| 570 | |
| 571 | for (i = 0; i < NUM_TEMPS; i++) |
| 572 | jit_stxi(stack_ptr + i * sizeof(uintptr_t), JIT_FP, JIT_R(i)); |
| 573 | |
| 574 | /* Jump to the trampoline */ |
| 575 | to_tramp = jit_jmpi(); |
| 576 | |
| 577 | /* The trampoline will jump back here */ |
| 578 | to_fn_epilog = jit_label(); |
| 579 | |
| 580 | for (i = 0; i < NUM_TEMPS; i++) |
| 581 | jit_ldxi(JIT_R(i), JIT_FP, stack_ptr + i * sizeof(uintptr_t)); |
| 582 | |
| 583 | jit_ret(); |
| 584 | jit_epilog(); |
| 585 | |
| 586 | /* Trampoline entry point. |
| 587 | * The sole purpose of the trampoline is to cheese Lightning not to |
| 588 | * save/restore the callee-saved register LIGHTREC_REG_CYCLE, since we |
| 589 | * do want to return to the caller with this register modified. */ |
| 590 | jit_prolog(); |
| 591 | jit_tramp(256); |
| 592 | jit_patch(to_tramp); |
| 593 | |
| 594 | jit_prepare(); |
| 595 | jit_pushargr(LIGHTREC_REG_STATE); |
| 596 | jit_pushargr(LIGHTREC_REG_CYCLE); |
| 597 | jit_pushargi((uintptr_t)f); |
| 598 | jit_pushargr(JIT_R0); |
| 599 | if (generic) { |
| 600 | jit_pushargr(JIT_R1); |
| 601 | jit_finishi(c_generic_function_wrapper); |
| 602 | } else { |
| 603 | jit_finishi(c_function_wrapper); |
| 604 | } |
| 605 | |
| 606 | #if __WORDSIZE == 64 |
| 607 | jit_retval_i(LIGHTREC_REG_CYCLE); |
| 608 | #else |
| 609 | jit_retval(LIGHTREC_REG_CYCLE); |
| 610 | #endif |
| 611 | |
| 612 | jit_patch_at(jit_jmpi(), to_fn_epilog); |
| 613 | jit_epilog(); |
| 614 | |
| 615 | block->state = state; |
| 616 | block->_jit = _jit; |
| 617 | block->function = jit_emit(); |
| 618 | block->opcode_list = NULL; |
| 619 | block->flags = 0; |
| 620 | block->nb_ops = 0; |
| 621 | |
| 622 | jit_get_code(&code_size); |
| 623 | lightrec_register(MEM_FOR_CODE, code_size); |
| 624 | |
| 625 | block->code_size = code_size; |
| 626 | |
| 627 | if (ENABLE_DISASSEMBLER) { |
| 628 | pr_debug("Wrapper block:\n"); |
| 629 | jit_disassemble(); |
| 630 | } |
| 631 | |
| 632 | jit_clear_state(); |
| 633 | return block; |
| 634 | |
| 635 | err_free_block: |
| 636 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 637 | err_no_mem: |
| 638 | pr_err("Unable to compile wrapper: Out of memory\n"); |
| 639 | return NULL; |
| 640 | } |
| 641 | |
| 642 | static struct block * generate_dispatcher(struct lightrec_state *state) |
| 643 | { |
| 644 | struct block *block; |
| 645 | jit_state_t *_jit; |
| 646 | jit_node_t *to_end, *to_end2, *to_c, *loop, *addr, *addr2; |
| 647 | unsigned int i; |
| 648 | u32 offset, ram_len; |
| 649 | jit_word_t code_size; |
| 650 | |
| 651 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 652 | if (!block) |
| 653 | goto err_no_mem; |
| 654 | |
| 655 | _jit = jit_new_state(); |
| 656 | if (!_jit) |
| 657 | goto err_free_block; |
| 658 | |
| 659 | jit_name("dispatcher"); |
| 660 | jit_note(__FILE__, __LINE__); |
| 661 | |
| 662 | jit_prolog(); |
| 663 | jit_frame(256); |
| 664 | |
| 665 | jit_getarg(JIT_R0, jit_arg()); |
| 666 | #if __WORDSIZE == 64 |
| 667 | jit_getarg_i(LIGHTREC_REG_CYCLE, jit_arg()); |
| 668 | #else |
| 669 | jit_getarg(LIGHTREC_REG_CYCLE, jit_arg()); |
| 670 | #endif |
| 671 | |
| 672 | /* Force all callee-saved registers to be pushed on the stack */ |
| 673 | for (i = 0; i < NUM_REGS; i++) |
| 674 | jit_movr(JIT_V(i), JIT_V(i)); |
| 675 | |
| 676 | /* Pass lightrec_state structure to blocks, using the last callee-saved |
| 677 | * register that Lightning provides */ |
| 678 | jit_movi(LIGHTREC_REG_STATE, (intptr_t) state); |
| 679 | |
| 680 | loop = jit_label(); |
| 681 | |
| 682 | /* Call the block's code */ |
| 683 | jit_jmpr(JIT_R0); |
| 684 | |
| 685 | /* The block will jump here, with the number of cycles remaining in |
| 686 | * LIGHTREC_REG_CYCLE */ |
| 687 | addr2 = jit_indirect(); |
| 688 | |
| 689 | /* Jump to end if state->target_cycle < state->current_cycle */ |
| 690 | to_end = jit_blei(LIGHTREC_REG_CYCLE, 0); |
| 691 | |
| 692 | /* Convert next PC to KUNSEG and avoid mirrors */ |
| 693 | ram_len = state->maps[PSX_MAP_KERNEL_USER_RAM].length; |
| 694 | jit_andi(JIT_R0, JIT_V0, 0x10000000 | (ram_len - 1)); |
| 695 | to_c = jit_bgei(JIT_R0, ram_len); |
| 696 | |
| 697 | /* Fast path: code is running from RAM, use the code LUT */ |
| 698 | #if __WORDSIZE == 64 |
| 699 | jit_lshi(JIT_R0, JIT_R0, 1); |
| 700 | #endif |
| 701 | jit_addr(JIT_R0, JIT_R0, LIGHTREC_REG_STATE); |
| 702 | jit_ldxi(JIT_R0, JIT_R0, offsetof(struct lightrec_state, code_lut)); |
| 703 | |
| 704 | /* If we get non-NULL, loop */ |
| 705 | jit_patch_at(jit_bnei(JIT_R0, 0), loop); |
| 706 | |
| 707 | /* Slow path: call C function get_next_block_func() */ |
| 708 | jit_patch(to_c); |
| 709 | |
| 710 | if (ENABLE_FIRST_PASS) { |
| 711 | /* We may call the interpreter - update state->current_cycle */ |
| 712 | jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE, |
| 713 | offsetof(struct lightrec_state, target_cycle)); |
| 714 | jit_subr(JIT_R1, JIT_R2, LIGHTREC_REG_CYCLE); |
| 715 | jit_stxi_i(offsetof(struct lightrec_state, current_cycle), |
| 716 | LIGHTREC_REG_STATE, JIT_R1); |
| 717 | } |
| 718 | |
| 719 | /* The code LUT will be set to this address when the block at the target |
| 720 | * PC has been preprocessed but not yet compiled by the threaded |
| 721 | * recompiler */ |
| 722 | addr = jit_indirect(); |
| 723 | |
| 724 | /* Get the next block */ |
| 725 | jit_prepare(); |
| 726 | jit_pushargr(LIGHTREC_REG_STATE); |
| 727 | jit_pushargr(JIT_V0); |
| 728 | jit_finishi(&get_next_block_func); |
| 729 | jit_retval(JIT_R0); |
| 730 | |
| 731 | if (ENABLE_FIRST_PASS) { |
| 732 | /* The interpreter may have updated state->current_cycle and |
| 733 | * state->target_cycle - recalc the delta */ |
| 734 | jit_ldxi_i(JIT_R1, LIGHTREC_REG_STATE, |
| 735 | offsetof(struct lightrec_state, current_cycle)); |
| 736 | jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE, |
| 737 | offsetof(struct lightrec_state, target_cycle)); |
| 738 | jit_subr(LIGHTREC_REG_CYCLE, JIT_R2, JIT_R1); |
| 739 | } |
| 740 | |
| 741 | /* If we get non-NULL, loop */ |
| 742 | jit_patch_at(jit_bnei(JIT_R0, 0), loop); |
| 743 | |
| 744 | to_end2 = jit_jmpi(); |
| 745 | |
| 746 | /* When exiting, the recompiled code will jump to that address */ |
| 747 | jit_note(__FILE__, __LINE__); |
| 748 | jit_patch(to_end); |
| 749 | |
| 750 | /* Store back the next_pc to the lightrec_state structure */ |
| 751 | offset = offsetof(struct lightrec_state, next_pc); |
| 752 | jit_stxi_i(offset, LIGHTREC_REG_STATE, JIT_V0); |
| 753 | |
| 754 | jit_patch(to_end2); |
| 755 | |
| 756 | jit_retr(LIGHTREC_REG_CYCLE); |
| 757 | jit_epilog(); |
| 758 | |
| 759 | block->state = state; |
| 760 | block->_jit = _jit; |
| 761 | block->function = jit_emit(); |
| 762 | block->opcode_list = NULL; |
| 763 | block->flags = 0; |
| 764 | block->nb_ops = 0; |
| 765 | |
| 766 | jit_get_code(&code_size); |
| 767 | lightrec_register(MEM_FOR_CODE, code_size); |
| 768 | |
| 769 | block->code_size = code_size; |
| 770 | |
| 771 | state->eob_wrapper_func = jit_address(addr2); |
| 772 | state->get_next_block = jit_address(addr); |
| 773 | |
| 774 | if (ENABLE_DISASSEMBLER) { |
| 775 | pr_debug("Dispatcher block:\n"); |
| 776 | jit_disassemble(); |
| 777 | } |
| 778 | |
| 779 | /* We're done! */ |
| 780 | jit_clear_state(); |
| 781 | return block; |
| 782 | |
| 783 | err_free_block: |
| 784 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 785 | err_no_mem: |
| 786 | pr_err("Unable to compile dispatcher: Out of memory\n"); |
| 787 | return NULL; |
| 788 | } |
| 789 | |
| 790 | union code lightrec_read_opcode(struct lightrec_state *state, u32 pc) |
| 791 | { |
| 792 | u32 addr, kunseg_pc = kunseg(pc); |
| 793 | const u32 *code; |
| 794 | const struct lightrec_mem_map *map = lightrec_get_map(state, kunseg_pc); |
| 795 | |
| 796 | addr = kunseg_pc - map->pc; |
| 797 | |
| 798 | while (map->mirror_of) |
| 799 | map = map->mirror_of; |
| 800 | |
| 801 | code = map->address + addr; |
| 802 | |
| 803 | return (union code) *code; |
| 804 | } |
| 805 | |
| 806 | static struct block * lightrec_precompile_block(struct lightrec_state *state, |
| 807 | u32 pc) |
| 808 | { |
| 809 | struct opcode *list; |
| 810 | struct block *block; |
| 811 | const u32 *code; |
| 812 | u32 addr, kunseg_pc = kunseg(pc); |
| 813 | const struct lightrec_mem_map *map = lightrec_get_map(state, kunseg_pc); |
| 814 | unsigned int length; |
| 815 | |
| 816 | if (!map) |
| 817 | return NULL; |
| 818 | |
| 819 | addr = kunseg_pc - map->pc; |
| 820 | |
| 821 | while (map->mirror_of) |
| 822 | map = map->mirror_of; |
| 823 | |
| 824 | code = map->address + addr; |
| 825 | |
| 826 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 827 | if (!block) { |
| 828 | pr_err("Unable to recompile block: Out of memory\n"); |
| 829 | return NULL; |
| 830 | } |
| 831 | |
| 832 | list = lightrec_disassemble(state, code, &length); |
| 833 | if (!list) { |
| 834 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 835 | return NULL; |
| 836 | } |
| 837 | |
| 838 | block->pc = pc; |
| 839 | block->state = state; |
| 840 | block->_jit = NULL; |
| 841 | block->function = NULL; |
| 842 | block->opcode_list = list; |
| 843 | block->map = map; |
| 844 | block->next = NULL; |
| 845 | block->flags = 0; |
| 846 | block->code_size = 0; |
| 847 | #if ENABLE_THREADED_COMPILER |
| 848 | block->op_list_freed = (atomic_flag)ATOMIC_FLAG_INIT; |
| 849 | #endif |
| 850 | block->nb_ops = length / sizeof(u32); |
| 851 | |
| 852 | lightrec_optimize(block); |
| 853 | |
| 854 | length = block->nb_ops * sizeof(u32); |
| 855 | |
| 856 | lightrec_register(MEM_FOR_MIPS_CODE, length); |
| 857 | |
| 858 | if (ENABLE_DISASSEMBLER) { |
| 859 | pr_debug("Disassembled block at PC: 0x%x\n", block->pc); |
| 860 | lightrec_print_disassembly(block, code, length); |
| 861 | } |
| 862 | |
| 863 | pr_debug("Block size: %lu opcodes\n", block->nb_ops); |
| 864 | |
| 865 | /* If the first opcode is an 'impossible' branch, never compile the |
| 866 | * block */ |
| 867 | if (list->flags & LIGHTREC_EMULATE_BRANCH) |
| 868 | block->flags |= BLOCK_NEVER_COMPILE; |
| 869 | |
| 870 | block->hash = lightrec_calculate_block_hash(block); |
| 871 | |
| 872 | pr_debug("Recompile count: %u\n", state->nb_precompile++); |
| 873 | |
| 874 | return block; |
| 875 | } |
| 876 | |
| 877 | static bool lightrec_block_is_fully_tagged(struct block *block) |
| 878 | { |
| 879 | struct opcode *op; |
| 880 | |
| 881 | for (op = block->opcode_list; op; op = op->next) { |
| 882 | /* Verify that all load/stores of the opcode list |
| 883 | * Check all loads/stores of the opcode list and mark the |
| 884 | * block as fully compiled if they all have been tagged. */ |
| 885 | switch (op->c.i.op) { |
| 886 | case OP_LB: |
| 887 | case OP_LH: |
| 888 | case OP_LWL: |
| 889 | case OP_LW: |
| 890 | case OP_LBU: |
| 891 | case OP_LHU: |
| 892 | case OP_LWR: |
| 893 | case OP_SB: |
| 894 | case OP_SH: |
| 895 | case OP_SWL: |
| 896 | case OP_SW: |
| 897 | case OP_SWR: |
| 898 | case OP_LWC2: |
| 899 | case OP_SWC2: |
| 900 | if (!(op->flags & (LIGHTREC_DIRECT_IO | |
| 901 | LIGHTREC_HW_IO))) |
| 902 | return false; |
| 903 | default: /* fall-through */ |
| 904 | continue; |
| 905 | } |
| 906 | } |
| 907 | |
| 908 | return true; |
| 909 | } |
| 910 | |
| 911 | static void lightrec_reap_block(void *data) |
| 912 | { |
| 913 | struct block *block = data; |
| 914 | |
| 915 | pr_debug("Reap dead block at PC 0x%08x\n", block->pc); |
| 916 | lightrec_free_block(block); |
| 917 | } |
| 918 | |
| 919 | static void lightrec_reap_jit(void *data) |
| 920 | { |
| 921 | _jit_destroy_state(data); |
| 922 | } |
| 923 | |
| 924 | int lightrec_compile_block(struct block *block) |
| 925 | { |
| 926 | struct lightrec_state *state = block->state; |
| 927 | struct lightrec_branch_target *target; |
| 928 | bool op_list_freed = false, fully_tagged = false; |
| 929 | struct block *block2; |
| 930 | struct opcode *elm; |
| 931 | jit_state_t *_jit, *oldjit; |
| 932 | jit_node_t *start_of_block; |
| 933 | bool skip_next = false; |
| 934 | jit_word_t code_size; |
| 935 | unsigned int i, j; |
| 936 | u32 next_pc, offset; |
| 937 | |
| 938 | fully_tagged = lightrec_block_is_fully_tagged(block); |
| 939 | if (fully_tagged) |
| 940 | block->flags |= BLOCK_FULLY_TAGGED; |
| 941 | |
| 942 | _jit = jit_new_state(); |
| 943 | if (!_jit) |
| 944 | return -ENOMEM; |
| 945 | |
| 946 | oldjit = block->_jit; |
| 947 | block->_jit = _jit; |
| 948 | |
| 949 | lightrec_regcache_reset(state->reg_cache); |
| 950 | state->cycles = 0; |
| 951 | state->nb_branches = 0; |
| 952 | state->nb_local_branches = 0; |
| 953 | state->nb_targets = 0; |
| 954 | |
| 955 | jit_prolog(); |
| 956 | jit_tramp(256); |
| 957 | |
| 958 | start_of_block = jit_label(); |
| 959 | |
| 960 | for (elm = block->opcode_list; elm; elm = elm->next) { |
| 961 | next_pc = block->pc + elm->offset * sizeof(u32); |
| 962 | |
| 963 | if (skip_next) { |
| 964 | skip_next = false; |
| 965 | continue; |
| 966 | } |
| 967 | |
| 968 | state->cycles += lightrec_cycles_of_opcode(elm->c); |
| 969 | |
| 970 | if (elm->flags & LIGHTREC_EMULATE_BRANCH) { |
| 971 | pr_debug("Branch at offset 0x%x will be emulated\n", |
| 972 | elm->offset << 2); |
| 973 | lightrec_emit_eob(block, elm, next_pc); |
| 974 | skip_next = !(elm->flags & LIGHTREC_NO_DS); |
| 975 | } else if (elm->opcode) { |
| 976 | lightrec_rec_opcode(block, elm, next_pc); |
| 977 | skip_next = has_delay_slot(elm->c) && |
| 978 | !(elm->flags & LIGHTREC_NO_DS); |
| 979 | #if _WIN32 |
| 980 | /* FIXME: GNU Lightning on Windows seems to use our |
| 981 | * mapped registers as temporaries. Until the actual bug |
| 982 | * is found and fixed, unconditionally mark our |
| 983 | * registers as live here. */ |
| 984 | lightrec_regcache_mark_live(state->reg_cache, _jit); |
| 985 | #endif |
| 986 | } |
| 987 | } |
| 988 | |
| 989 | for (i = 0; i < state->nb_branches; i++) |
| 990 | jit_patch(state->branches[i]); |
| 991 | |
| 992 | for (i = 0; i < state->nb_local_branches; i++) { |
| 993 | struct lightrec_branch *branch = &state->local_branches[i]; |
| 994 | |
| 995 | pr_debug("Patch local branch to offset 0x%x\n", |
| 996 | branch->target << 2); |
| 997 | |
| 998 | if (branch->target == 0) { |
| 999 | jit_patch_at(branch->branch, start_of_block); |
| 1000 | continue; |
| 1001 | } |
| 1002 | |
| 1003 | for (j = 0; j < state->nb_targets; j++) { |
| 1004 | if (state->targets[j].offset == branch->target) { |
| 1005 | jit_patch_at(branch->branch, |
| 1006 | state->targets[j].label); |
| 1007 | break; |
| 1008 | } |
| 1009 | } |
| 1010 | |
| 1011 | if (j == state->nb_targets) |
| 1012 | pr_err("Unable to find branch target\n"); |
| 1013 | } |
| 1014 | |
| 1015 | jit_ldxi(JIT_R0, LIGHTREC_REG_STATE, |
| 1016 | offsetof(struct lightrec_state, eob_wrapper_func)); |
| 1017 | |
| 1018 | jit_jmpr(JIT_R0); |
| 1019 | |
| 1020 | jit_ret(); |
| 1021 | jit_epilog(); |
| 1022 | |
| 1023 | block->function = jit_emit(); |
| 1024 | block->flags &= ~BLOCK_SHOULD_RECOMPILE; |
| 1025 | |
| 1026 | /* Add compiled function to the LUT */ |
| 1027 | state->code_lut[lut_offset(block->pc)] = block->function; |
| 1028 | |
| 1029 | /* Fill code LUT with the block's entry points */ |
| 1030 | for (i = 0; i < state->nb_targets; i++) { |
| 1031 | target = &state->targets[i]; |
| 1032 | |
| 1033 | if (target->offset) { |
| 1034 | offset = lut_offset(block->pc) + target->offset; |
| 1035 | state->code_lut[offset] = jit_address(target->label); |
| 1036 | } |
| 1037 | } |
| 1038 | |
| 1039 | /* Detect old blocks that have been covered by the new one */ |
| 1040 | for (i = 0; i < state->nb_targets; i++) { |
| 1041 | target = &state->targets[i]; |
| 1042 | |
| 1043 | if (!target->offset) |
| 1044 | continue; |
| 1045 | |
| 1046 | offset = block->pc + target->offset * sizeof(u32); |
| 1047 | block2 = lightrec_find_block(state->block_cache, offset); |
| 1048 | if (block2) { |
| 1049 | /* No need to check if block2 is compilable - it must |
| 1050 | * be, otherwise block wouldn't be compilable either */ |
| 1051 | |
| 1052 | block2->flags |= BLOCK_IS_DEAD; |
| 1053 | |
| 1054 | pr_debug("Reap block 0x%08x as it's covered by block " |
| 1055 | "0x%08x\n", block2->pc, block->pc); |
| 1056 | |
| 1057 | lightrec_unregister_block(state->block_cache, block2); |
| 1058 | |
| 1059 | if (ENABLE_THREADED_COMPILER) { |
| 1060 | lightrec_recompiler_remove(state->rec, block2); |
| 1061 | lightrec_reaper_add(state->reaper, |
| 1062 | lightrec_reap_block, |
| 1063 | block2); |
| 1064 | } else { |
| 1065 | lightrec_free_block(block2); |
| 1066 | } |
| 1067 | } |
| 1068 | } |
| 1069 | |
| 1070 | jit_get_code(&code_size); |
| 1071 | lightrec_register(MEM_FOR_CODE, code_size); |
| 1072 | |
| 1073 | block->code_size = code_size; |
| 1074 | |
| 1075 | if (ENABLE_DISASSEMBLER) { |
| 1076 | pr_debug("Compiling block at PC: 0x%x\n", block->pc); |
| 1077 | jit_disassemble(); |
| 1078 | } |
| 1079 | |
| 1080 | jit_clear_state(); |
| 1081 | |
| 1082 | #if ENABLE_THREADED_COMPILER |
| 1083 | if (fully_tagged) |
| 1084 | op_list_freed = atomic_flag_test_and_set(&block->op_list_freed); |
| 1085 | #endif |
| 1086 | if (fully_tagged && !op_list_freed) { |
| 1087 | pr_debug("Block PC 0x%08x is fully tagged" |
| 1088 | " - free opcode list\n", block->pc); |
| 1089 | lightrec_free_opcode_list(state, block->opcode_list); |
| 1090 | block->opcode_list = NULL; |
| 1091 | } |
| 1092 | |
| 1093 | if (oldjit) { |
| 1094 | pr_debug("Block 0x%08x recompiled, reaping old jit context.\n", |
| 1095 | block->pc); |
| 1096 | |
| 1097 | if (ENABLE_THREADED_COMPILER) |
| 1098 | lightrec_reaper_add(state->reaper, |
| 1099 | lightrec_reap_jit, oldjit); |
| 1100 | else |
| 1101 | _jit_destroy_state(oldjit); |
| 1102 | } |
| 1103 | |
| 1104 | return 0; |
| 1105 | } |
| 1106 | |
| 1107 | u32 lightrec_execute(struct lightrec_state *state, u32 pc, u32 target_cycle) |
| 1108 | { |
| 1109 | s32 (*func)(void *, s32) = (void *)state->dispatcher->function; |
| 1110 | void *block_trace; |
| 1111 | s32 cycles_delta; |
| 1112 | |
| 1113 | state->exit_flags = LIGHTREC_EXIT_NORMAL; |
| 1114 | |
| 1115 | /* Handle the cycle counter overflowing */ |
| 1116 | if (unlikely(target_cycle < state->current_cycle)) |
| 1117 | target_cycle = UINT_MAX; |
| 1118 | |
| 1119 | state->target_cycle = target_cycle; |
| 1120 | |
| 1121 | block_trace = get_next_block_func(state, pc); |
| 1122 | if (block_trace) { |
| 1123 | cycles_delta = state->target_cycle - state->current_cycle; |
| 1124 | |
| 1125 | cycles_delta = (*func)(block_trace, cycles_delta); |
| 1126 | |
| 1127 | state->current_cycle = state->target_cycle - cycles_delta; |
| 1128 | } |
| 1129 | |
| 1130 | if (ENABLE_THREADED_COMPILER) |
| 1131 | lightrec_reaper_reap(state->reaper); |
| 1132 | |
| 1133 | return state->next_pc; |
| 1134 | } |
| 1135 | |
| 1136 | u32 lightrec_execute_one(struct lightrec_state *state, u32 pc) |
| 1137 | { |
| 1138 | return lightrec_execute(state, pc, state->current_cycle); |
| 1139 | } |
| 1140 | |
| 1141 | u32 lightrec_run_interpreter(struct lightrec_state *state, u32 pc) |
| 1142 | { |
| 1143 | struct block *block = lightrec_get_block(state, pc); |
| 1144 | if (!block) |
| 1145 | return 0; |
| 1146 | |
| 1147 | state->exit_flags = LIGHTREC_EXIT_NORMAL; |
| 1148 | |
| 1149 | return lightrec_emulate_block(block, pc); |
| 1150 | } |
| 1151 | |
| 1152 | void lightrec_free_block(struct block *block) |
| 1153 | { |
| 1154 | lightrec_unregister(MEM_FOR_MIPS_CODE, block->nb_ops * sizeof(u32)); |
| 1155 | if (block->opcode_list) |
| 1156 | lightrec_free_opcode_list(block->state, block->opcode_list); |
| 1157 | if (block->_jit) |
| 1158 | _jit_destroy_state(block->_jit); |
| 1159 | lightrec_unregister(MEM_FOR_CODE, block->code_size); |
| 1160 | lightrec_free(block->state, MEM_FOR_IR, sizeof(*block), block); |
| 1161 | } |
| 1162 | |
| 1163 | struct lightrec_state * lightrec_init(char *argv0, |
| 1164 | const struct lightrec_mem_map *map, |
| 1165 | size_t nb, |
| 1166 | const struct lightrec_ops *ops) |
| 1167 | { |
| 1168 | struct lightrec_state *state; |
| 1169 | |
| 1170 | /* Sanity-check ops */ |
| 1171 | if (!ops || |
| 1172 | !ops->cop0_ops.mfc || !ops->cop0_ops.cfc || !ops->cop0_ops.mtc || |
| 1173 | !ops->cop0_ops.ctc || !ops->cop0_ops.op || |
| 1174 | !ops->cop2_ops.mfc || !ops->cop2_ops.cfc || !ops->cop2_ops.mtc || |
| 1175 | !ops->cop2_ops.ctc || !ops->cop2_ops.op) { |
| 1176 | pr_err("Missing callbacks in lightrec_ops structure\n"); |
| 1177 | return NULL; |
| 1178 | } |
| 1179 | |
| 1180 | init_jit(argv0); |
| 1181 | |
| 1182 | state = calloc(1, sizeof(*state) + |
| 1183 | sizeof(*state->code_lut) * CODE_LUT_SIZE); |
| 1184 | if (!state) |
| 1185 | goto err_finish_jit; |
| 1186 | |
| 1187 | lightrec_register(MEM_FOR_LIGHTREC, sizeof(*state) + |
| 1188 | sizeof(*state->code_lut) * CODE_LUT_SIZE); |
| 1189 | |
| 1190 | #if ENABLE_TINYMM |
| 1191 | state->tinymm = tinymm_init(malloc, free, 4096); |
| 1192 | if (!state->tinymm) |
| 1193 | goto err_free_state; |
| 1194 | #endif |
| 1195 | |
| 1196 | state->block_cache = lightrec_blockcache_init(state); |
| 1197 | if (!state->block_cache) |
| 1198 | goto err_free_tinymm; |
| 1199 | |
| 1200 | state->reg_cache = lightrec_regcache_init(state); |
| 1201 | if (!state->reg_cache) |
| 1202 | goto err_free_block_cache; |
| 1203 | |
| 1204 | if (ENABLE_THREADED_COMPILER) { |
| 1205 | state->rec = lightrec_recompiler_init(state); |
| 1206 | if (!state->rec) |
| 1207 | goto err_free_reg_cache; |
| 1208 | |
| 1209 | state->reaper = lightrec_reaper_init(state); |
| 1210 | if (!state->reaper) |
| 1211 | goto err_free_recompiler; |
| 1212 | } |
| 1213 | |
| 1214 | state->nb_maps = nb; |
| 1215 | state->maps = map; |
| 1216 | |
| 1217 | memcpy(&state->ops, ops, sizeof(*ops)); |
| 1218 | |
| 1219 | state->dispatcher = generate_dispatcher(state); |
| 1220 | if (!state->dispatcher) |
| 1221 | goto err_free_reaper; |
| 1222 | |
| 1223 | state->rw_generic_wrapper = generate_wrapper(state, |
| 1224 | lightrec_rw_generic_cb, |
| 1225 | true); |
| 1226 | if (!state->rw_generic_wrapper) |
| 1227 | goto err_free_dispatcher; |
| 1228 | |
| 1229 | state->rw_wrapper = generate_wrapper(state, lightrec_rw_cb, false); |
| 1230 | if (!state->rw_wrapper) |
| 1231 | goto err_free_generic_rw_wrapper; |
| 1232 | |
| 1233 | state->mfc_wrapper = generate_wrapper(state, lightrec_mfc_cb, false); |
| 1234 | if (!state->mfc_wrapper) |
| 1235 | goto err_free_rw_wrapper; |
| 1236 | |
| 1237 | state->mtc_wrapper = generate_wrapper(state, lightrec_mtc_cb, false); |
| 1238 | if (!state->mtc_wrapper) |
| 1239 | goto err_free_mfc_wrapper; |
| 1240 | |
| 1241 | state->rfe_wrapper = generate_wrapper(state, lightrec_rfe_cb, false); |
| 1242 | if (!state->rfe_wrapper) |
| 1243 | goto err_free_mtc_wrapper; |
| 1244 | |
| 1245 | state->cp_wrapper = generate_wrapper(state, lightrec_cp_cb, false); |
| 1246 | if (!state->cp_wrapper) |
| 1247 | goto err_free_rfe_wrapper; |
| 1248 | |
| 1249 | state->syscall_wrapper = generate_wrapper(state, lightrec_syscall_cb, |
| 1250 | false); |
| 1251 | if (!state->syscall_wrapper) |
| 1252 | goto err_free_cp_wrapper; |
| 1253 | |
| 1254 | state->break_wrapper = generate_wrapper(state, lightrec_break_cb, |
| 1255 | false); |
| 1256 | if (!state->break_wrapper) |
| 1257 | goto err_free_syscall_wrapper; |
| 1258 | |
| 1259 | state->rw_generic_func = state->rw_generic_wrapper->function; |
| 1260 | state->rw_func = state->rw_wrapper->function; |
| 1261 | state->mfc_func = state->mfc_wrapper->function; |
| 1262 | state->mtc_func = state->mtc_wrapper->function; |
| 1263 | state->rfe_func = state->rfe_wrapper->function; |
| 1264 | state->cp_func = state->cp_wrapper->function; |
| 1265 | state->syscall_func = state->syscall_wrapper->function; |
| 1266 | state->break_func = state->break_wrapper->function; |
| 1267 | |
| 1268 | map = &state->maps[PSX_MAP_BIOS]; |
| 1269 | state->offset_bios = (uintptr_t)map->address - map->pc; |
| 1270 | |
| 1271 | map = &state->maps[PSX_MAP_SCRATCH_PAD]; |
| 1272 | state->offset_scratch = (uintptr_t)map->address - map->pc; |
| 1273 | |
| 1274 | map = &state->maps[PSX_MAP_KERNEL_USER_RAM]; |
| 1275 | state->offset_ram = (uintptr_t)map->address - map->pc; |
| 1276 | |
| 1277 | if (state->maps[PSX_MAP_MIRROR1].address == map->address + 0x200000 && |
| 1278 | state->maps[PSX_MAP_MIRROR2].address == map->address + 0x400000 && |
| 1279 | state->maps[PSX_MAP_MIRROR3].address == map->address + 0x600000) |
| 1280 | state->mirrors_mapped = true; |
| 1281 | |
| 1282 | return state; |
| 1283 | |
| 1284 | err_free_syscall_wrapper: |
| 1285 | lightrec_free_block(state->syscall_wrapper); |
| 1286 | err_free_cp_wrapper: |
| 1287 | lightrec_free_block(state->cp_wrapper); |
| 1288 | err_free_rfe_wrapper: |
| 1289 | lightrec_free_block(state->rfe_wrapper); |
| 1290 | err_free_mtc_wrapper: |
| 1291 | lightrec_free_block(state->mtc_wrapper); |
| 1292 | err_free_mfc_wrapper: |
| 1293 | lightrec_free_block(state->mfc_wrapper); |
| 1294 | err_free_rw_wrapper: |
| 1295 | lightrec_free_block(state->rw_wrapper); |
| 1296 | err_free_generic_rw_wrapper: |
| 1297 | lightrec_free_block(state->rw_generic_wrapper); |
| 1298 | err_free_dispatcher: |
| 1299 | lightrec_free_block(state->dispatcher); |
| 1300 | err_free_reaper: |
| 1301 | if (ENABLE_THREADED_COMPILER) |
| 1302 | lightrec_reaper_destroy(state->reaper); |
| 1303 | err_free_recompiler: |
| 1304 | if (ENABLE_THREADED_COMPILER) |
| 1305 | lightrec_free_recompiler(state->rec); |
| 1306 | err_free_reg_cache: |
| 1307 | lightrec_free_regcache(state->reg_cache); |
| 1308 | err_free_block_cache: |
| 1309 | lightrec_free_block_cache(state->block_cache); |
| 1310 | err_free_tinymm: |
| 1311 | #if ENABLE_TINYMM |
| 1312 | tinymm_shutdown(state->tinymm); |
| 1313 | err_free_state: |
| 1314 | #endif |
| 1315 | lightrec_unregister(MEM_FOR_LIGHTREC, sizeof(*state) + |
| 1316 | sizeof(*state->code_lut) * CODE_LUT_SIZE); |
| 1317 | free(state); |
| 1318 | err_finish_jit: |
| 1319 | finish_jit(); |
| 1320 | return NULL; |
| 1321 | } |
| 1322 | |
| 1323 | void lightrec_destroy(struct lightrec_state *state) |
| 1324 | { |
| 1325 | if (ENABLE_THREADED_COMPILER) { |
| 1326 | lightrec_free_recompiler(state->rec); |
| 1327 | lightrec_reaper_destroy(state->reaper); |
| 1328 | } |
| 1329 | |
| 1330 | lightrec_free_regcache(state->reg_cache); |
| 1331 | lightrec_free_block_cache(state->block_cache); |
| 1332 | lightrec_free_block(state->dispatcher); |
| 1333 | lightrec_free_block(state->rw_generic_wrapper); |
| 1334 | lightrec_free_block(state->rw_wrapper); |
| 1335 | lightrec_free_block(state->mfc_wrapper); |
| 1336 | lightrec_free_block(state->mtc_wrapper); |
| 1337 | lightrec_free_block(state->rfe_wrapper); |
| 1338 | lightrec_free_block(state->cp_wrapper); |
| 1339 | lightrec_free_block(state->syscall_wrapper); |
| 1340 | lightrec_free_block(state->break_wrapper); |
| 1341 | finish_jit(); |
| 1342 | |
| 1343 | #if ENABLE_TINYMM |
| 1344 | tinymm_shutdown(state->tinymm); |
| 1345 | #endif |
| 1346 | lightrec_unregister(MEM_FOR_LIGHTREC, sizeof(*state) + |
| 1347 | sizeof(*state->code_lut) * CODE_LUT_SIZE); |
| 1348 | free(state); |
| 1349 | } |
| 1350 | |
| 1351 | void lightrec_invalidate(struct lightrec_state *state, u32 addr, u32 len) |
| 1352 | { |
| 1353 | u32 kaddr = kunseg(addr & ~0x3); |
| 1354 | const struct lightrec_mem_map *map = lightrec_get_map(state, kaddr); |
| 1355 | |
| 1356 | if (map) { |
| 1357 | while (map->mirror_of) |
| 1358 | map = map->mirror_of; |
| 1359 | |
| 1360 | if (map != &state->maps[PSX_MAP_KERNEL_USER_RAM]) |
| 1361 | return; |
| 1362 | |
| 1363 | /* Handle mirrors */ |
| 1364 | kaddr &= (state->maps[PSX_MAP_KERNEL_USER_RAM].length - 1); |
| 1365 | |
| 1366 | for (; len > 4; len -= 4, kaddr += 4) |
| 1367 | lightrec_invalidate_map(state, map, kaddr); |
| 1368 | |
| 1369 | lightrec_invalidate_map(state, map, kaddr); |
| 1370 | } |
| 1371 | } |
| 1372 | |
| 1373 | void lightrec_invalidate_all(struct lightrec_state *state) |
| 1374 | { |
| 1375 | memset(state->code_lut, 0, sizeof(*state->code_lut) * CODE_LUT_SIZE); |
| 1376 | } |
| 1377 | |
| 1378 | void lightrec_set_invalidate_mode(struct lightrec_state *state, bool dma_only) |
| 1379 | { |
| 1380 | if (state->invalidate_from_dma_only != dma_only) |
| 1381 | lightrec_invalidate_all(state); |
| 1382 | |
| 1383 | state->invalidate_from_dma_only = dma_only; |
| 1384 | } |
| 1385 | |
| 1386 | void lightrec_set_exit_flags(struct lightrec_state *state, u32 flags) |
| 1387 | { |
| 1388 | if (flags != LIGHTREC_EXIT_NORMAL) { |
| 1389 | state->exit_flags |= flags; |
| 1390 | state->target_cycle = state->current_cycle; |
| 1391 | } |
| 1392 | } |
| 1393 | |
| 1394 | u32 lightrec_exit_flags(struct lightrec_state *state) |
| 1395 | { |
| 1396 | return state->exit_flags; |
| 1397 | } |
| 1398 | |
| 1399 | void lightrec_dump_registers(struct lightrec_state *state, u32 regs[34]) |
| 1400 | { |
| 1401 | memcpy(regs, state->native_reg_cache, sizeof(state->native_reg_cache)); |
| 1402 | } |
| 1403 | |
| 1404 | void lightrec_restore_registers(struct lightrec_state *state, u32 regs[34]) |
| 1405 | { |
| 1406 | memcpy(state->native_reg_cache, regs, sizeof(state->native_reg_cache)); |
| 1407 | } |
| 1408 | |
| 1409 | u32 lightrec_current_cycle_count(const struct lightrec_state *state) |
| 1410 | { |
| 1411 | return state->current_cycle; |
| 1412 | } |
| 1413 | |
| 1414 | void lightrec_reset_cycle_count(struct lightrec_state *state, u32 cycles) |
| 1415 | { |
| 1416 | state->current_cycle = cycles; |
| 1417 | |
| 1418 | if (state->target_cycle < cycles) |
| 1419 | state->target_cycle = cycles; |
| 1420 | } |
| 1421 | |
| 1422 | void lightrec_set_target_cycle_count(struct lightrec_state *state, u32 cycles) |
| 1423 | { |
| 1424 | if (state->exit_flags == LIGHTREC_EXIT_NORMAL) { |
| 1425 | if (cycles < state->current_cycle) |
| 1426 | cycles = state->current_cycle; |
| 1427 | |
| 1428 | state->target_cycle = cycles; |
| 1429 | } |
| 1430 | } |