| 1 | // SPDX-License-Identifier: LGPL-2.1-or-later |
| 2 | /* |
| 3 | * Copyright (C) 2014-2021 Paul Cercueil <paul@crapouillou.net> |
| 4 | */ |
| 5 | |
| 6 | #include "blockcache.h" |
| 7 | #include "debug.h" |
| 8 | #include "disassembler.h" |
| 9 | #include "emitter.h" |
| 10 | #include "interpreter.h" |
| 11 | #include "lightrec-config.h" |
| 12 | #include "lightning-wrapper.h" |
| 13 | #include "lightrec.h" |
| 14 | #include "memmanager.h" |
| 15 | #include "reaper.h" |
| 16 | #include "recompiler.h" |
| 17 | #include "regcache.h" |
| 18 | #include "optimizer.h" |
| 19 | #include "tlsf/tlsf.h" |
| 20 | |
| 21 | #include <errno.h> |
| 22 | #include <inttypes.h> |
| 23 | #include <limits.h> |
| 24 | #if ENABLE_THREADED_COMPILER |
| 25 | #include <stdatomic.h> |
| 26 | #endif |
| 27 | #include <stdbool.h> |
| 28 | #include <stddef.h> |
| 29 | #include <string.h> |
| 30 | |
| 31 | #define GENMASK(h, l) \ |
| 32 | (((uintptr_t)-1 << (l)) & ((uintptr_t)-1 >> (__WORDSIZE - 1 - (h)))) |
| 33 | |
| 34 | static struct block * lightrec_precompile_block(struct lightrec_state *state, |
| 35 | u32 pc); |
| 36 | static bool lightrec_block_is_fully_tagged(const struct block *block); |
| 37 | |
| 38 | static void lightrec_mtc2(struct lightrec_state *state, u8 reg, u32 data); |
| 39 | static u32 lightrec_mfc2(struct lightrec_state *state, u8 reg); |
| 40 | |
| 41 | static void lightrec_default_sb(struct lightrec_state *state, u32 opcode, |
| 42 | void *host, u32 addr, u8 data) |
| 43 | { |
| 44 | *(u8 *)host = data; |
| 45 | |
| 46 | if (!state->invalidate_from_dma_only) |
| 47 | lightrec_invalidate(state, addr, 1); |
| 48 | } |
| 49 | |
| 50 | static void lightrec_default_sh(struct lightrec_state *state, u32 opcode, |
| 51 | void *host, u32 addr, u16 data) |
| 52 | { |
| 53 | *(u16 *)host = HTOLE16(data); |
| 54 | |
| 55 | if (!state->invalidate_from_dma_only) |
| 56 | lightrec_invalidate(state, addr, 2); |
| 57 | } |
| 58 | |
| 59 | static void lightrec_default_sw(struct lightrec_state *state, u32 opcode, |
| 60 | void *host, u32 addr, u32 data) |
| 61 | { |
| 62 | *(u32 *)host = HTOLE32(data); |
| 63 | |
| 64 | if (!state->invalidate_from_dma_only) |
| 65 | lightrec_invalidate(state, addr, 4); |
| 66 | } |
| 67 | |
| 68 | static u8 lightrec_default_lb(struct lightrec_state *state, |
| 69 | u32 opcode, void *host, u32 addr) |
| 70 | { |
| 71 | return *(u8 *)host; |
| 72 | } |
| 73 | |
| 74 | static u16 lightrec_default_lh(struct lightrec_state *state, |
| 75 | u32 opcode, void *host, u32 addr) |
| 76 | { |
| 77 | return LE16TOH(*(u16 *)host); |
| 78 | } |
| 79 | |
| 80 | static u32 lightrec_default_lw(struct lightrec_state *state, |
| 81 | u32 opcode, void *host, u32 addr) |
| 82 | { |
| 83 | return LE32TOH(*(u32 *)host); |
| 84 | } |
| 85 | |
| 86 | static const struct lightrec_mem_map_ops lightrec_default_ops = { |
| 87 | .sb = lightrec_default_sb, |
| 88 | .sh = lightrec_default_sh, |
| 89 | .sw = lightrec_default_sw, |
| 90 | .lb = lightrec_default_lb, |
| 91 | .lh = lightrec_default_lh, |
| 92 | .lw = lightrec_default_lw, |
| 93 | }; |
| 94 | |
| 95 | static void __segfault_cb(struct lightrec_state *state, u32 addr, |
| 96 | const struct block *block) |
| 97 | { |
| 98 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT); |
| 99 | pr_err("Segmentation fault in recompiled code: invalid " |
| 100 | "load/store at address 0x%08x\n", addr); |
| 101 | if (block) |
| 102 | pr_err("Was executing block PC 0x%08x\n", block->pc); |
| 103 | } |
| 104 | |
| 105 | static void lightrec_swl(struct lightrec_state *state, |
| 106 | const struct lightrec_mem_map_ops *ops, |
| 107 | u32 opcode, void *host, u32 addr, u32 data) |
| 108 | { |
| 109 | unsigned int shift = addr & 0x3; |
| 110 | unsigned int mask = GENMASK(31, (shift + 1) * 8); |
| 111 | u32 old_data; |
| 112 | |
| 113 | /* Align to 32 bits */ |
| 114 | addr &= ~3; |
| 115 | host = (void *)((uintptr_t)host & ~3); |
| 116 | |
| 117 | old_data = ops->lw(state, opcode, host, addr); |
| 118 | |
| 119 | data = (data >> ((3 - shift) * 8)) | (old_data & mask); |
| 120 | |
| 121 | ops->sw(state, opcode, host, addr, data); |
| 122 | } |
| 123 | |
| 124 | static void lightrec_swr(struct lightrec_state *state, |
| 125 | const struct lightrec_mem_map_ops *ops, |
| 126 | u32 opcode, void *host, u32 addr, u32 data) |
| 127 | { |
| 128 | unsigned int shift = addr & 0x3; |
| 129 | unsigned int mask = (1 << (shift * 8)) - 1; |
| 130 | u32 old_data; |
| 131 | |
| 132 | /* Align to 32 bits */ |
| 133 | addr &= ~3; |
| 134 | host = (void *)((uintptr_t)host & ~3); |
| 135 | |
| 136 | old_data = ops->lw(state, opcode, host, addr); |
| 137 | |
| 138 | data = (data << (shift * 8)) | (old_data & mask); |
| 139 | |
| 140 | ops->sw(state, opcode, host, addr, data); |
| 141 | } |
| 142 | |
| 143 | static void lightrec_swc2(struct lightrec_state *state, union code op, |
| 144 | const struct lightrec_mem_map_ops *ops, |
| 145 | void *host, u32 addr) |
| 146 | { |
| 147 | u32 data = lightrec_mfc2(state, op.i.rt); |
| 148 | |
| 149 | ops->sw(state, op.opcode, host, addr, data); |
| 150 | } |
| 151 | |
| 152 | static u32 lightrec_lwl(struct lightrec_state *state, |
| 153 | const struct lightrec_mem_map_ops *ops, |
| 154 | u32 opcode, void *host, u32 addr, u32 data) |
| 155 | { |
| 156 | unsigned int shift = addr & 0x3; |
| 157 | unsigned int mask = (1 << (24 - shift * 8)) - 1; |
| 158 | u32 old_data; |
| 159 | |
| 160 | /* Align to 32 bits */ |
| 161 | addr &= ~3; |
| 162 | host = (void *)((uintptr_t)host & ~3); |
| 163 | |
| 164 | old_data = ops->lw(state, opcode, host, addr); |
| 165 | |
| 166 | return (data & mask) | (old_data << (24 - shift * 8)); |
| 167 | } |
| 168 | |
| 169 | static u32 lightrec_lwr(struct lightrec_state *state, |
| 170 | const struct lightrec_mem_map_ops *ops, |
| 171 | u32 opcode, void *host, u32 addr, u32 data) |
| 172 | { |
| 173 | unsigned int shift = addr & 0x3; |
| 174 | unsigned int mask = GENMASK(31, 32 - shift * 8); |
| 175 | u32 old_data; |
| 176 | |
| 177 | /* Align to 32 bits */ |
| 178 | addr &= ~3; |
| 179 | host = (void *)((uintptr_t)host & ~3); |
| 180 | |
| 181 | old_data = ops->lw(state, opcode, host, addr); |
| 182 | |
| 183 | return (data & mask) | (old_data >> (shift * 8)); |
| 184 | } |
| 185 | |
| 186 | static void lightrec_lwc2(struct lightrec_state *state, union code op, |
| 187 | const struct lightrec_mem_map_ops *ops, |
| 188 | void *host, u32 addr) |
| 189 | { |
| 190 | u32 data = ops->lw(state, op.opcode, host, addr); |
| 191 | |
| 192 | lightrec_mtc2(state, op.i.rt, data); |
| 193 | } |
| 194 | |
| 195 | static void lightrec_invalidate_map(struct lightrec_state *state, |
| 196 | const struct lightrec_mem_map *map, u32 addr, u32 len) |
| 197 | { |
| 198 | if (map == &state->maps[PSX_MAP_KERNEL_USER_RAM]) { |
| 199 | memset(lut_address(state, lut_offset(addr)), 0, |
| 200 | ((len + 3) / 4) * lut_elm_size(state)); |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | enum psx_map |
| 205 | lightrec_get_map_idx(struct lightrec_state *state, u32 kaddr) |
| 206 | { |
| 207 | const struct lightrec_mem_map *map; |
| 208 | unsigned int i; |
| 209 | |
| 210 | for (i = 0; i < state->nb_maps; i++) { |
| 211 | map = &state->maps[i]; |
| 212 | |
| 213 | if (kaddr >= map->pc && kaddr < map->pc + map->length) |
| 214 | return (enum psx_map) i; |
| 215 | } |
| 216 | |
| 217 | return PSX_MAP_UNKNOWN; |
| 218 | } |
| 219 | |
| 220 | const struct lightrec_mem_map * |
| 221 | lightrec_get_map(struct lightrec_state *state, void **host, u32 kaddr) |
| 222 | { |
| 223 | const struct lightrec_mem_map *map; |
| 224 | enum psx_map idx; |
| 225 | u32 addr; |
| 226 | |
| 227 | idx = lightrec_get_map_idx(state, kaddr); |
| 228 | if (idx == PSX_MAP_UNKNOWN) |
| 229 | return NULL; |
| 230 | |
| 231 | map = &state->maps[idx]; |
| 232 | addr = kaddr - map->pc; |
| 233 | |
| 234 | while (map->mirror_of) |
| 235 | map = map->mirror_of; |
| 236 | |
| 237 | if (host) |
| 238 | *host = map->address + addr; |
| 239 | |
| 240 | return map; |
| 241 | } |
| 242 | |
| 243 | u32 lightrec_rw(struct lightrec_state *state, union code op, |
| 244 | u32 addr, u32 data, u16 *flags, struct block *block) |
| 245 | { |
| 246 | const struct lightrec_mem_map *map; |
| 247 | const struct lightrec_mem_map_ops *ops; |
| 248 | u32 opcode = op.opcode; |
| 249 | void *host; |
| 250 | |
| 251 | addr += (s16) op.i.imm; |
| 252 | |
| 253 | map = lightrec_get_map(state, &host, kunseg(addr)); |
| 254 | if (!map) { |
| 255 | __segfault_cb(state, addr, block); |
| 256 | return 0; |
| 257 | } |
| 258 | |
| 259 | if (unlikely(map->ops)) { |
| 260 | if (flags && !LIGHTREC_FLAGS_GET_IO_MODE(*flags)) |
| 261 | *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_HW); |
| 262 | |
| 263 | ops = map->ops; |
| 264 | } else { |
| 265 | if (flags && !LIGHTREC_FLAGS_GET_IO_MODE(*flags)) |
| 266 | *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_DIRECT); |
| 267 | |
| 268 | ops = &lightrec_default_ops; |
| 269 | } |
| 270 | |
| 271 | switch (op.i.op) { |
| 272 | case OP_SB: |
| 273 | ops->sb(state, opcode, host, addr, (u8) data); |
| 274 | return 0; |
| 275 | case OP_SH: |
| 276 | ops->sh(state, opcode, host, addr, (u16) data); |
| 277 | return 0; |
| 278 | case OP_SWL: |
| 279 | lightrec_swl(state, ops, opcode, host, addr, data); |
| 280 | return 0; |
| 281 | case OP_SWR: |
| 282 | lightrec_swr(state, ops, opcode, host, addr, data); |
| 283 | return 0; |
| 284 | case OP_SW: |
| 285 | ops->sw(state, opcode, host, addr, data); |
| 286 | return 0; |
| 287 | case OP_SWC2: |
| 288 | lightrec_swc2(state, op, ops, host, addr); |
| 289 | return 0; |
| 290 | case OP_LB: |
| 291 | return (s32) (s8) ops->lb(state, opcode, host, addr); |
| 292 | case OP_LBU: |
| 293 | return ops->lb(state, opcode, host, addr); |
| 294 | case OP_LH: |
| 295 | return (s32) (s16) ops->lh(state, opcode, host, addr); |
| 296 | case OP_LHU: |
| 297 | return ops->lh(state, opcode, host, addr); |
| 298 | case OP_LWC2: |
| 299 | lightrec_lwc2(state, op, ops, host, addr); |
| 300 | return 0; |
| 301 | case OP_LWL: |
| 302 | return lightrec_lwl(state, ops, opcode, host, addr, data); |
| 303 | case OP_LWR: |
| 304 | return lightrec_lwr(state, ops, opcode, host, addr, data); |
| 305 | case OP_LW: |
| 306 | default: |
| 307 | return ops->lw(state, opcode, host, addr); |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | static void lightrec_rw_helper(struct lightrec_state *state, |
| 312 | union code op, u16 *flags, |
| 313 | struct block *block) |
| 314 | { |
| 315 | u32 ret = lightrec_rw(state, op, state->regs.gpr[op.i.rs], |
| 316 | state->regs.gpr[op.i.rt], flags, block); |
| 317 | |
| 318 | switch (op.i.op) { |
| 319 | case OP_LB: |
| 320 | case OP_LBU: |
| 321 | case OP_LH: |
| 322 | case OP_LHU: |
| 323 | case OP_LWL: |
| 324 | case OP_LWR: |
| 325 | case OP_LW: |
| 326 | if (op.i.rt) |
| 327 | state->regs.gpr[op.i.rt] = ret; |
| 328 | fallthrough; |
| 329 | default: |
| 330 | break; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | static void lightrec_rw_cb(struct lightrec_state *state) |
| 335 | { |
| 336 | lightrec_rw_helper(state, (union code)state->c_wrapper_arg, NULL, NULL); |
| 337 | } |
| 338 | |
| 339 | static void lightrec_rw_generic_cb(struct lightrec_state *state) |
| 340 | { |
| 341 | struct block *block; |
| 342 | struct opcode *op; |
| 343 | bool was_tagged; |
| 344 | u32 arg = state->c_wrapper_arg; |
| 345 | u16 offset = (u16)arg; |
| 346 | |
| 347 | block = lightrec_find_block_from_lut(state->block_cache, |
| 348 | arg >> 16, state->next_pc); |
| 349 | if (unlikely(!block)) { |
| 350 | pr_err("rw_generic: No block found in LUT for PC 0x%x offset 0x%x\n", |
| 351 | state->next_pc, offset); |
| 352 | return; |
| 353 | } |
| 354 | |
| 355 | op = &block->opcode_list[offset]; |
| 356 | was_tagged = LIGHTREC_FLAGS_GET_IO_MODE(op->flags); |
| 357 | |
| 358 | lightrec_rw_helper(state, op->c, &op->flags, block); |
| 359 | |
| 360 | if (!was_tagged) { |
| 361 | pr_debug("Opcode of block at PC 0x%08x has been tagged - flag " |
| 362 | "for recompilation\n", block->pc); |
| 363 | |
| 364 | block->flags |= BLOCK_SHOULD_RECOMPILE; |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | static u32 clamp_s32(s32 val, s32 min, s32 max) |
| 369 | { |
| 370 | return val < min ? min : val > max ? max : val; |
| 371 | } |
| 372 | |
| 373 | static u16 load_u16(u32 *ptr) |
| 374 | { |
| 375 | return ((struct u16x2 *) ptr)->l; |
| 376 | } |
| 377 | |
| 378 | static void store_u16(u32 *ptr, u16 value) |
| 379 | { |
| 380 | ((struct u16x2 *) ptr)->l = value; |
| 381 | } |
| 382 | |
| 383 | static u32 lightrec_mfc2(struct lightrec_state *state, u8 reg) |
| 384 | { |
| 385 | s16 gteir1, gteir2, gteir3; |
| 386 | |
| 387 | switch (reg) { |
| 388 | case 1: |
| 389 | case 3: |
| 390 | case 5: |
| 391 | case 8: |
| 392 | case 9: |
| 393 | case 10: |
| 394 | case 11: |
| 395 | return (s32)(s16) load_u16(&state->regs.cp2d[reg]); |
| 396 | case 7: |
| 397 | case 16: |
| 398 | case 17: |
| 399 | case 18: |
| 400 | case 19: |
| 401 | return load_u16(&state->regs.cp2d[reg]); |
| 402 | case 28: |
| 403 | case 29: |
| 404 | gteir1 = (s16) load_u16(&state->regs.cp2d[9]); |
| 405 | gteir2 = (s16) load_u16(&state->regs.cp2d[10]); |
| 406 | gteir3 = (s16) load_u16(&state->regs.cp2d[11]); |
| 407 | |
| 408 | return clamp_s32(gteir1 >> 7, 0, 0x1f) << 0 | |
| 409 | clamp_s32(gteir2 >> 7, 0, 0x1f) << 5 | |
| 410 | clamp_s32(gteir3 >> 7, 0, 0x1f) << 10; |
| 411 | case 15: |
| 412 | reg = 14; |
| 413 | fallthrough; |
| 414 | default: |
| 415 | return state->regs.cp2d[reg]; |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | u32 lightrec_mfc(struct lightrec_state *state, union code op) |
| 420 | { |
| 421 | if (op.i.op == OP_CP0) |
| 422 | return state->regs.cp0[op.r.rd]; |
| 423 | else if (op.r.rs == OP_CP2_BASIC_MFC2) |
| 424 | return lightrec_mfc2(state, op.r.rd); |
| 425 | else |
| 426 | return state->regs.cp2c[op.r.rd]; |
| 427 | } |
| 428 | |
| 429 | static void lightrec_mtc0(struct lightrec_state *state, u8 reg, u32 data) |
| 430 | { |
| 431 | u32 status, oldstatus, cause; |
| 432 | |
| 433 | switch (reg) { |
| 434 | case 1: |
| 435 | case 4: |
| 436 | case 8: |
| 437 | case 14: |
| 438 | case 15: |
| 439 | /* Those registers are read-only */ |
| 440 | return; |
| 441 | default: |
| 442 | break; |
| 443 | } |
| 444 | |
| 445 | if (reg == 12) { |
| 446 | status = state->regs.cp0[12]; |
| 447 | oldstatus = status; |
| 448 | |
| 449 | if (status & ~data & BIT(16)) { |
| 450 | state->ops.enable_ram(state, true); |
| 451 | lightrec_invalidate_all(state); |
| 452 | } else if (~status & data & BIT(16)) { |
| 453 | state->ops.enable_ram(state, false); |
| 454 | } |
| 455 | } |
| 456 | |
| 457 | if (reg == 13) { |
| 458 | state->regs.cp0[13] &= ~0x300; |
| 459 | state->regs.cp0[13] |= data & 0x300; |
| 460 | } else { |
| 461 | state->regs.cp0[reg] = data; |
| 462 | } |
| 463 | |
| 464 | if (reg == 12 || reg == 13) { |
| 465 | cause = state->regs.cp0[13]; |
| 466 | status = state->regs.cp0[12]; |
| 467 | |
| 468 | /* Handle software interrupts */ |
| 469 | if (!!(status & cause & 0x300) & status) |
| 470 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_CHECK_INTERRUPT); |
| 471 | |
| 472 | /* Handle hardware interrupts */ |
| 473 | if (reg == 12 && !(~status & 0x401) && (~oldstatus & 0x401)) |
| 474 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_CHECK_INTERRUPT); |
| 475 | } |
| 476 | } |
| 477 | |
| 478 | static u32 count_leading_bits(s32 data) |
| 479 | { |
| 480 | u32 cnt = 33; |
| 481 | |
| 482 | #ifdef __has_builtin |
| 483 | #if __has_builtin(__builtin_clrsb) |
| 484 | return 1 + __builtin_clrsb(data); |
| 485 | #endif |
| 486 | #endif |
| 487 | |
| 488 | data = (data ^ (data >> 31)) << 1; |
| 489 | |
| 490 | do { |
| 491 | cnt -= 1; |
| 492 | data >>= 1; |
| 493 | } while (data); |
| 494 | |
| 495 | return cnt; |
| 496 | } |
| 497 | |
| 498 | static void lightrec_mtc2(struct lightrec_state *state, u8 reg, u32 data) |
| 499 | { |
| 500 | switch (reg) { |
| 501 | case 15: |
| 502 | state->regs.cp2d[12] = state->regs.cp2d[13]; |
| 503 | state->regs.cp2d[13] = state->regs.cp2d[14]; |
| 504 | state->regs.cp2d[14] = data; |
| 505 | break; |
| 506 | case 28: |
| 507 | state->regs.cp2d[9] = (data << 7) & 0xf80; |
| 508 | state->regs.cp2d[10] = (data << 2) & 0xf80; |
| 509 | state->regs.cp2d[11] = (data >> 3) & 0xf80; |
| 510 | break; |
| 511 | case 31: |
| 512 | return; |
| 513 | case 30: |
| 514 | state->regs.cp2d[31] = count_leading_bits((s32) data); |
| 515 | fallthrough; |
| 516 | default: |
| 517 | state->regs.cp2d[reg] = data; |
| 518 | break; |
| 519 | } |
| 520 | } |
| 521 | |
| 522 | static void lightrec_ctc2(struct lightrec_state *state, u8 reg, u32 data) |
| 523 | { |
| 524 | switch (reg) { |
| 525 | case 4: |
| 526 | case 12: |
| 527 | case 20: |
| 528 | case 26: |
| 529 | case 27: |
| 530 | case 29: |
| 531 | case 30: |
| 532 | store_u16(&state->regs.cp2c[reg], data); |
| 533 | break; |
| 534 | case 31: |
| 535 | data = (data & 0x7ffff000) | !!(data & 0x7f87e000) << 31; |
| 536 | fallthrough; |
| 537 | default: |
| 538 | state->regs.cp2c[reg] = data; |
| 539 | break; |
| 540 | } |
| 541 | } |
| 542 | |
| 543 | void lightrec_mtc(struct lightrec_state *state, union code op, u32 data) |
| 544 | { |
| 545 | if (op.i.op == OP_CP0) |
| 546 | lightrec_mtc0(state, op.r.rd, data); |
| 547 | else if (op.r.rs == OP_CP2_BASIC_CTC2) |
| 548 | lightrec_ctc2(state, op.r.rd, data); |
| 549 | else |
| 550 | lightrec_mtc2(state, op.r.rd, data); |
| 551 | } |
| 552 | |
| 553 | static void lightrec_mtc_cb(struct lightrec_state *state) |
| 554 | { |
| 555 | union code op = (union code) state->c_wrapper_arg; |
| 556 | |
| 557 | lightrec_mtc(state, op, state->regs.gpr[op.r.rt]); |
| 558 | } |
| 559 | |
| 560 | void lightrec_rfe(struct lightrec_state *state) |
| 561 | { |
| 562 | u32 status; |
| 563 | |
| 564 | /* Read CP0 Status register (r12) */ |
| 565 | status = state->regs.cp0[12]; |
| 566 | |
| 567 | /* Switch the bits */ |
| 568 | status = ((status & 0x3c) >> 2) | (status & ~0xf); |
| 569 | |
| 570 | /* Write it back */ |
| 571 | lightrec_mtc0(state, 12, status); |
| 572 | } |
| 573 | |
| 574 | void lightrec_cp(struct lightrec_state *state, union code op) |
| 575 | { |
| 576 | if (op.i.op == OP_CP0) { |
| 577 | pr_err("Invalid CP opcode to coprocessor #0\n"); |
| 578 | return; |
| 579 | } |
| 580 | |
| 581 | (*state->ops.cop2_op)(state, op.opcode); |
| 582 | } |
| 583 | |
| 584 | static void lightrec_cp_cb(struct lightrec_state *state) |
| 585 | { |
| 586 | lightrec_cp(state, (union code) state->c_wrapper_arg); |
| 587 | } |
| 588 | |
| 589 | static void lightrec_syscall_cb(struct lightrec_state *state) |
| 590 | { |
| 591 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SYSCALL); |
| 592 | } |
| 593 | |
| 594 | static void lightrec_break_cb(struct lightrec_state *state) |
| 595 | { |
| 596 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_BREAK); |
| 597 | } |
| 598 | |
| 599 | struct block * lightrec_get_block(struct lightrec_state *state, u32 pc) |
| 600 | { |
| 601 | struct block *block = lightrec_find_block(state->block_cache, pc); |
| 602 | |
| 603 | if (block && lightrec_block_is_outdated(state, block)) { |
| 604 | pr_debug("Block at PC 0x%08x is outdated!\n", block->pc); |
| 605 | |
| 606 | /* Make sure the recompiler isn't processing the block we'll |
| 607 | * destroy */ |
| 608 | if (ENABLE_THREADED_COMPILER) |
| 609 | lightrec_recompiler_remove(state->rec, block); |
| 610 | |
| 611 | lightrec_unregister_block(state->block_cache, block); |
| 612 | remove_from_code_lut(state->block_cache, block); |
| 613 | lightrec_free_block(state, block); |
| 614 | block = NULL; |
| 615 | } |
| 616 | |
| 617 | if (!block) { |
| 618 | block = lightrec_precompile_block(state, pc); |
| 619 | if (!block) { |
| 620 | pr_err("Unable to recompile block at PC 0x%x\n", pc); |
| 621 | lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT); |
| 622 | return NULL; |
| 623 | } |
| 624 | |
| 625 | lightrec_register_block(state->block_cache, block); |
| 626 | } |
| 627 | |
| 628 | return block; |
| 629 | } |
| 630 | |
| 631 | static void * get_next_block_func(struct lightrec_state *state, u32 pc) |
| 632 | { |
| 633 | struct block *block; |
| 634 | bool should_recompile; |
| 635 | void *func; |
| 636 | int err; |
| 637 | |
| 638 | for (;;) { |
| 639 | func = lut_read(state, lut_offset(pc)); |
| 640 | if (func && func != state->get_next_block) |
| 641 | break; |
| 642 | |
| 643 | block = lightrec_get_block(state, pc); |
| 644 | |
| 645 | if (unlikely(!block)) |
| 646 | break; |
| 647 | |
| 648 | if (OPT_REPLACE_MEMSET && (block->flags & BLOCK_IS_MEMSET)) { |
| 649 | func = state->memset_func; |
| 650 | break; |
| 651 | } |
| 652 | |
| 653 | should_recompile = block->flags & BLOCK_SHOULD_RECOMPILE && |
| 654 | !(block->flags & BLOCK_IS_DEAD); |
| 655 | |
| 656 | if (unlikely(should_recompile)) { |
| 657 | pr_debug("Block at PC 0x%08x should recompile\n", pc); |
| 658 | |
| 659 | lightrec_unregister(MEM_FOR_CODE, block->code_size); |
| 660 | |
| 661 | if (ENABLE_THREADED_COMPILER) { |
| 662 | lightrec_recompiler_add(state->rec, block); |
| 663 | } else { |
| 664 | err = lightrec_compile_block(state->cstate, block); |
| 665 | if (err) { |
| 666 | state->exit_flags = LIGHTREC_EXIT_NOMEM; |
| 667 | return NULL; |
| 668 | } |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | if (ENABLE_THREADED_COMPILER && likely(!should_recompile)) |
| 673 | func = lightrec_recompiler_run_first_pass(state, block, &pc); |
| 674 | else |
| 675 | func = block->function; |
| 676 | |
| 677 | if (likely(func)) |
| 678 | break; |
| 679 | |
| 680 | if (unlikely(block->flags & BLOCK_NEVER_COMPILE)) { |
| 681 | pc = lightrec_emulate_block(state, block, pc); |
| 682 | |
| 683 | } else if (!ENABLE_THREADED_COMPILER) { |
| 684 | /* Block wasn't compiled yet - run the interpreter */ |
| 685 | if (block->flags & BLOCK_FULLY_TAGGED) |
| 686 | pr_debug("Block fully tagged, skipping first pass\n"); |
| 687 | else if (ENABLE_FIRST_PASS && likely(!should_recompile)) |
| 688 | pc = lightrec_emulate_block(state, block, pc); |
| 689 | |
| 690 | /* Then compile it using the profiled data */ |
| 691 | err = lightrec_compile_block(state->cstate, block); |
| 692 | if (err) { |
| 693 | state->exit_flags = LIGHTREC_EXIT_NOMEM; |
| 694 | return NULL; |
| 695 | } |
| 696 | } else { |
| 697 | lightrec_recompiler_add(state->rec, block); |
| 698 | } |
| 699 | |
| 700 | if (state->exit_flags != LIGHTREC_EXIT_NORMAL || |
| 701 | state->current_cycle >= state->target_cycle) |
| 702 | break; |
| 703 | } |
| 704 | |
| 705 | state->next_pc = pc; |
| 706 | return func; |
| 707 | } |
| 708 | |
| 709 | static s32 c_function_wrapper(struct lightrec_state *state, s32 cycles_delta, |
| 710 | void (*f)(struct lightrec_state *)) |
| 711 | { |
| 712 | state->current_cycle = state->target_cycle - cycles_delta; |
| 713 | |
| 714 | (*f)(state); |
| 715 | |
| 716 | return state->target_cycle - state->current_cycle; |
| 717 | } |
| 718 | |
| 719 | static void * lightrec_alloc_code(struct lightrec_state *state, size_t size) |
| 720 | { |
| 721 | void *code; |
| 722 | |
| 723 | if (ENABLE_THREADED_COMPILER) |
| 724 | lightrec_code_alloc_lock(state); |
| 725 | |
| 726 | code = tlsf_malloc(state->tlsf, size); |
| 727 | |
| 728 | if (ENABLE_THREADED_COMPILER) |
| 729 | lightrec_code_alloc_unlock(state); |
| 730 | |
| 731 | return code; |
| 732 | } |
| 733 | |
| 734 | static void lightrec_realloc_code(struct lightrec_state *state, |
| 735 | void *ptr, size_t size) |
| 736 | { |
| 737 | /* NOTE: 'size' MUST be smaller than the size specified during |
| 738 | * the allocation. */ |
| 739 | |
| 740 | if (ENABLE_THREADED_COMPILER) |
| 741 | lightrec_code_alloc_lock(state); |
| 742 | |
| 743 | tlsf_realloc(state->tlsf, ptr, size); |
| 744 | |
| 745 | if (ENABLE_THREADED_COMPILER) |
| 746 | lightrec_code_alloc_unlock(state); |
| 747 | } |
| 748 | |
| 749 | static void lightrec_free_code(struct lightrec_state *state, void *ptr) |
| 750 | { |
| 751 | if (ENABLE_THREADED_COMPILER) |
| 752 | lightrec_code_alloc_lock(state); |
| 753 | |
| 754 | tlsf_free(state->tlsf, ptr); |
| 755 | |
| 756 | if (ENABLE_THREADED_COMPILER) |
| 757 | lightrec_code_alloc_unlock(state); |
| 758 | } |
| 759 | |
| 760 | static void * lightrec_emit_code(struct lightrec_state *state, |
| 761 | const struct block *block, |
| 762 | jit_state_t *_jit, unsigned int *size) |
| 763 | { |
| 764 | bool has_code_buffer = ENABLE_CODE_BUFFER && state->tlsf; |
| 765 | jit_word_t code_size, new_code_size; |
| 766 | void *code; |
| 767 | |
| 768 | jit_realize(); |
| 769 | |
| 770 | if (!ENABLE_DISASSEMBLER) |
| 771 | jit_set_data(NULL, 0, JIT_DISABLE_DATA | JIT_DISABLE_NOTE); |
| 772 | |
| 773 | if (has_code_buffer) { |
| 774 | jit_get_code(&code_size); |
| 775 | code = lightrec_alloc_code(state, (size_t) code_size); |
| 776 | |
| 777 | if (!code) { |
| 778 | if (ENABLE_THREADED_COMPILER) { |
| 779 | /* If we're using the threaded compiler, return |
| 780 | * an allocation error here. The threaded |
| 781 | * compiler will then empty its job queue and |
| 782 | * request a code flush using the reaper. */ |
| 783 | return NULL; |
| 784 | } |
| 785 | |
| 786 | /* Remove outdated blocks, and try again */ |
| 787 | lightrec_remove_outdated_blocks(state->block_cache, block); |
| 788 | |
| 789 | pr_debug("Re-try to alloc %zu bytes...\n", code_size); |
| 790 | |
| 791 | code = lightrec_alloc_code(state, code_size); |
| 792 | if (!code) { |
| 793 | pr_err("Could not alloc even after removing old blocks!\n"); |
| 794 | return NULL; |
| 795 | } |
| 796 | } |
| 797 | |
| 798 | jit_set_code(code, code_size); |
| 799 | } |
| 800 | |
| 801 | code = jit_emit(); |
| 802 | |
| 803 | jit_get_code(&new_code_size); |
| 804 | lightrec_register(MEM_FOR_CODE, new_code_size); |
| 805 | |
| 806 | if (has_code_buffer) { |
| 807 | lightrec_realloc_code(state, code, (size_t) new_code_size); |
| 808 | |
| 809 | pr_debug("Creating code block at address 0x%" PRIxPTR ", " |
| 810 | "code size: %" PRIuPTR " new: %" PRIuPTR "\n", |
| 811 | (uintptr_t) code, code_size, new_code_size); |
| 812 | } |
| 813 | |
| 814 | *size = (unsigned int) new_code_size; |
| 815 | |
| 816 | return code; |
| 817 | } |
| 818 | |
| 819 | static struct block * generate_wrapper(struct lightrec_state *state) |
| 820 | { |
| 821 | struct block *block; |
| 822 | jit_state_t *_jit; |
| 823 | unsigned int i; |
| 824 | int stack_ptr; |
| 825 | jit_node_t *to_tramp, *to_fn_epilog; |
| 826 | jit_node_t *addr[C_WRAPPERS_COUNT - 1]; |
| 827 | |
| 828 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 829 | if (!block) |
| 830 | goto err_no_mem; |
| 831 | |
| 832 | _jit = jit_new_state(); |
| 833 | if (!_jit) |
| 834 | goto err_free_block; |
| 835 | |
| 836 | jit_name("RW wrapper"); |
| 837 | jit_note(__FILE__, __LINE__); |
| 838 | |
| 839 | /* Wrapper entry point */ |
| 840 | jit_prolog(); |
| 841 | jit_tramp(256); |
| 842 | |
| 843 | /* Add entry points; separate them by opcodes that increment |
| 844 | * LIGHTREC_REG_STATE (since we cannot touch other registers). |
| 845 | * The difference will then tell us which C function to call. */ |
| 846 | for (i = C_WRAPPERS_COUNT - 1; i > 0; i--) { |
| 847 | jit_addi(LIGHTREC_REG_STATE, LIGHTREC_REG_STATE, __WORDSIZE / 8); |
| 848 | addr[i - 1] = jit_indirect(); |
| 849 | } |
| 850 | |
| 851 | jit_epilog(); |
| 852 | jit_prolog(); |
| 853 | |
| 854 | stack_ptr = jit_allocai(sizeof(uintptr_t) * NUM_TEMPS); |
| 855 | |
| 856 | /* Save all temporaries on stack */ |
| 857 | for (i = 0; i < NUM_TEMPS; i++) |
| 858 | jit_stxi(stack_ptr + i * sizeof(uintptr_t), JIT_FP, JIT_R(i)); |
| 859 | |
| 860 | /* Jump to the trampoline */ |
| 861 | to_tramp = jit_jmpi(); |
| 862 | |
| 863 | /* The trampoline will jump back here */ |
| 864 | to_fn_epilog = jit_label(); |
| 865 | |
| 866 | /* Restore temporaries from stack */ |
| 867 | for (i = 0; i < NUM_TEMPS; i++) |
| 868 | jit_ldxi(JIT_R(i), JIT_FP, stack_ptr + i * sizeof(uintptr_t)); |
| 869 | |
| 870 | jit_ret(); |
| 871 | jit_epilog(); |
| 872 | |
| 873 | /* Trampoline entry point. |
| 874 | * The sole purpose of the trampoline is to cheese Lightning not to |
| 875 | * save/restore the callee-saved register LIGHTREC_REG_CYCLE, since we |
| 876 | * do want to return to the caller with this register modified. */ |
| 877 | jit_prolog(); |
| 878 | jit_tramp(256); |
| 879 | jit_patch(to_tramp); |
| 880 | |
| 881 | /* Retrieve the wrapper function */ |
| 882 | jit_ldxi(JIT_R0, LIGHTREC_REG_STATE, |
| 883 | offsetof(struct lightrec_state, c_wrappers)); |
| 884 | |
| 885 | /* Restore LIGHTREC_REG_STATE to its correct value */ |
| 886 | jit_movi(LIGHTREC_REG_STATE, (uintptr_t) state); |
| 887 | |
| 888 | jit_prepare(); |
| 889 | jit_pushargr(LIGHTREC_REG_STATE); |
| 890 | jit_pushargr(LIGHTREC_REG_CYCLE); |
| 891 | jit_pushargr(JIT_R0); |
| 892 | jit_finishi(c_function_wrapper); |
| 893 | jit_retval_i(LIGHTREC_REG_CYCLE); |
| 894 | |
| 895 | jit_patch_at(jit_jmpi(), to_fn_epilog); |
| 896 | jit_epilog(); |
| 897 | |
| 898 | block->_jit = _jit; |
| 899 | block->opcode_list = NULL; |
| 900 | block->flags = 0; |
| 901 | block->nb_ops = 0; |
| 902 | |
| 903 | block->function = lightrec_emit_code(state, block, _jit, |
| 904 | &block->code_size); |
| 905 | if (!block->function) |
| 906 | goto err_free_block; |
| 907 | |
| 908 | state->wrappers_eps[C_WRAPPERS_COUNT - 1] = block->function; |
| 909 | |
| 910 | for (i = 0; i < C_WRAPPERS_COUNT - 1; i++) |
| 911 | state->wrappers_eps[i] = jit_address(addr[i]); |
| 912 | |
| 913 | if (ENABLE_DISASSEMBLER) { |
| 914 | pr_debug("Wrapper block:\n"); |
| 915 | jit_disassemble(); |
| 916 | } |
| 917 | |
| 918 | jit_clear_state(); |
| 919 | return block; |
| 920 | |
| 921 | err_free_block: |
| 922 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 923 | err_no_mem: |
| 924 | pr_err("Unable to compile wrapper: Out of memory\n"); |
| 925 | return NULL; |
| 926 | } |
| 927 | |
| 928 | static u32 lightrec_memset(struct lightrec_state *state) |
| 929 | { |
| 930 | u32 kunseg_pc = kunseg(state->regs.gpr[4]); |
| 931 | void *host; |
| 932 | const struct lightrec_mem_map *map = lightrec_get_map(state, &host, kunseg_pc); |
| 933 | u32 length = state->regs.gpr[5] * 4; |
| 934 | |
| 935 | if (!map) { |
| 936 | pr_err("Unable to find memory map for memset target address " |
| 937 | "0x%x\n", kunseg_pc); |
| 938 | return 0; |
| 939 | } |
| 940 | |
| 941 | pr_debug("Calling host memset, PC 0x%x (host address 0x%" PRIxPTR ") for %u bytes\n", |
| 942 | kunseg_pc, (uintptr_t)host, length); |
| 943 | memset(host, 0, length); |
| 944 | |
| 945 | if (!state->invalidate_from_dma_only) |
| 946 | lightrec_invalidate_map(state, map, kunseg_pc, length); |
| 947 | |
| 948 | /* Rough estimation of the number of cycles consumed */ |
| 949 | return 8 + 5 * (length + 3 / 4); |
| 950 | } |
| 951 | |
| 952 | static struct block * generate_dispatcher(struct lightrec_state *state) |
| 953 | { |
| 954 | struct block *block; |
| 955 | jit_state_t *_jit; |
| 956 | jit_node_t *to_end, *loop, *addr, *addr2, *addr3; |
| 957 | unsigned int i; |
| 958 | u32 offset; |
| 959 | |
| 960 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 961 | if (!block) |
| 962 | goto err_no_mem; |
| 963 | |
| 964 | _jit = jit_new_state(); |
| 965 | if (!_jit) |
| 966 | goto err_free_block; |
| 967 | |
| 968 | jit_name("dispatcher"); |
| 969 | jit_note(__FILE__, __LINE__); |
| 970 | |
| 971 | jit_prolog(); |
| 972 | jit_frame(256); |
| 973 | |
| 974 | jit_getarg(JIT_R0, jit_arg()); |
| 975 | jit_getarg_i(LIGHTREC_REG_CYCLE, jit_arg()); |
| 976 | |
| 977 | /* Force all callee-saved registers to be pushed on the stack */ |
| 978 | for (i = 0; i < NUM_REGS; i++) |
| 979 | jit_movr(JIT_V(i), JIT_V(i)); |
| 980 | |
| 981 | /* Pass lightrec_state structure to blocks, using the last callee-saved |
| 982 | * register that Lightning provides */ |
| 983 | jit_movi(LIGHTREC_REG_STATE, (intptr_t) state); |
| 984 | |
| 985 | loop = jit_label(); |
| 986 | |
| 987 | /* Call the block's code */ |
| 988 | jit_jmpr(JIT_R0); |
| 989 | |
| 990 | if (OPT_REPLACE_MEMSET) { |
| 991 | /* Blocks will jump here when they need to call |
| 992 | * lightrec_memset() */ |
| 993 | addr3 = jit_indirect(); |
| 994 | |
| 995 | jit_prepare(); |
| 996 | jit_pushargr(LIGHTREC_REG_STATE); |
| 997 | jit_finishi(lightrec_memset); |
| 998 | |
| 999 | jit_ldxi_ui(JIT_V0, LIGHTREC_REG_STATE, |
| 1000 | offsetof(struct lightrec_state, regs.gpr[31])); |
| 1001 | |
| 1002 | jit_retval(JIT_R0); |
| 1003 | jit_subr(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, JIT_R0); |
| 1004 | } |
| 1005 | |
| 1006 | /* The block will jump here, with the number of cycles remaining in |
| 1007 | * LIGHTREC_REG_CYCLE */ |
| 1008 | addr2 = jit_indirect(); |
| 1009 | |
| 1010 | /* Store back the next_pc to the lightrec_state structure */ |
| 1011 | offset = offsetof(struct lightrec_state, next_pc); |
| 1012 | jit_stxi_i(offset, LIGHTREC_REG_STATE, JIT_V0); |
| 1013 | |
| 1014 | /* Jump to end if state->target_cycle < state->current_cycle */ |
| 1015 | to_end = jit_blei(LIGHTREC_REG_CYCLE, 0); |
| 1016 | |
| 1017 | /* Convert next PC to KUNSEG and avoid mirrors */ |
| 1018 | jit_andi(JIT_R0, JIT_V0, 0x10000000 | (RAM_SIZE - 1)); |
| 1019 | jit_rshi_u(JIT_R1, JIT_R0, 28); |
| 1020 | jit_andi(JIT_R2, JIT_V0, BIOS_SIZE - 1); |
| 1021 | jit_addi(JIT_R2, JIT_R2, RAM_SIZE); |
| 1022 | jit_movnr(JIT_R0, JIT_R2, JIT_R1); |
| 1023 | |
| 1024 | /* If possible, use the code LUT */ |
| 1025 | if (!lut_is_32bit(state)) |
| 1026 | jit_lshi(JIT_R0, JIT_R0, 1); |
| 1027 | jit_addr(JIT_R0, JIT_R0, LIGHTREC_REG_STATE); |
| 1028 | |
| 1029 | offset = offsetof(struct lightrec_state, code_lut); |
| 1030 | if (lut_is_32bit(state)) |
| 1031 | jit_ldxi_ui(JIT_R0, JIT_R0, offset); |
| 1032 | else |
| 1033 | jit_ldxi(JIT_R0, JIT_R0, offset); |
| 1034 | |
| 1035 | /* If we get non-NULL, loop */ |
| 1036 | jit_patch_at(jit_bnei(JIT_R0, 0), loop); |
| 1037 | |
| 1038 | /* Slow path: call C function get_next_block_func() */ |
| 1039 | |
| 1040 | if (ENABLE_FIRST_PASS || OPT_DETECT_IMPOSSIBLE_BRANCHES) { |
| 1041 | /* We may call the interpreter - update state->current_cycle */ |
| 1042 | jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE, |
| 1043 | offsetof(struct lightrec_state, target_cycle)); |
| 1044 | jit_subr(JIT_R1, JIT_R2, LIGHTREC_REG_CYCLE); |
| 1045 | jit_stxi_i(offsetof(struct lightrec_state, current_cycle), |
| 1046 | LIGHTREC_REG_STATE, JIT_R1); |
| 1047 | } |
| 1048 | |
| 1049 | /* The code LUT will be set to this address when the block at the target |
| 1050 | * PC has been preprocessed but not yet compiled by the threaded |
| 1051 | * recompiler */ |
| 1052 | addr = jit_indirect(); |
| 1053 | |
| 1054 | /* Get the next block */ |
| 1055 | jit_prepare(); |
| 1056 | jit_pushargr(LIGHTREC_REG_STATE); |
| 1057 | jit_pushargr(JIT_V0); |
| 1058 | jit_finishi(&get_next_block_func); |
| 1059 | jit_retval(JIT_R0); |
| 1060 | |
| 1061 | if (ENABLE_FIRST_PASS || OPT_DETECT_IMPOSSIBLE_BRANCHES) { |
| 1062 | /* The interpreter may have updated state->current_cycle and |
| 1063 | * state->target_cycle - recalc the delta */ |
| 1064 | jit_ldxi_i(JIT_R1, LIGHTREC_REG_STATE, |
| 1065 | offsetof(struct lightrec_state, current_cycle)); |
| 1066 | jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE, |
| 1067 | offsetof(struct lightrec_state, target_cycle)); |
| 1068 | jit_subr(LIGHTREC_REG_CYCLE, JIT_R2, JIT_R1); |
| 1069 | } |
| 1070 | |
| 1071 | /* If we get non-NULL, loop */ |
| 1072 | jit_patch_at(jit_bnei(JIT_R0, 0), loop); |
| 1073 | |
| 1074 | /* When exiting, the recompiled code will jump to that address */ |
| 1075 | jit_note(__FILE__, __LINE__); |
| 1076 | jit_patch(to_end); |
| 1077 | |
| 1078 | jit_retr(LIGHTREC_REG_CYCLE); |
| 1079 | jit_epilog(); |
| 1080 | |
| 1081 | block->_jit = _jit; |
| 1082 | block->opcode_list = NULL; |
| 1083 | block->flags = 0; |
| 1084 | block->nb_ops = 0; |
| 1085 | |
| 1086 | block->function = lightrec_emit_code(state, block, _jit, |
| 1087 | &block->code_size); |
| 1088 | if (!block->function) |
| 1089 | goto err_free_block; |
| 1090 | |
| 1091 | state->eob_wrapper_func = jit_address(addr2); |
| 1092 | if (OPT_REPLACE_MEMSET) |
| 1093 | state->memset_func = jit_address(addr3); |
| 1094 | state->get_next_block = jit_address(addr); |
| 1095 | |
| 1096 | if (ENABLE_DISASSEMBLER) { |
| 1097 | pr_debug("Dispatcher block:\n"); |
| 1098 | jit_disassemble(); |
| 1099 | } |
| 1100 | |
| 1101 | /* We're done! */ |
| 1102 | jit_clear_state(); |
| 1103 | return block; |
| 1104 | |
| 1105 | err_free_block: |
| 1106 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 1107 | err_no_mem: |
| 1108 | pr_err("Unable to compile dispatcher: Out of memory\n"); |
| 1109 | return NULL; |
| 1110 | } |
| 1111 | |
| 1112 | union code lightrec_read_opcode(struct lightrec_state *state, u32 pc) |
| 1113 | { |
| 1114 | void *host = NULL; |
| 1115 | |
| 1116 | lightrec_get_map(state, &host, kunseg(pc)); |
| 1117 | |
| 1118 | const u32 *code = (u32 *)host; |
| 1119 | return (union code) LE32TOH(*code); |
| 1120 | } |
| 1121 | |
| 1122 | unsigned int lightrec_cycles_of_opcode(union code code) |
| 1123 | { |
| 1124 | return 2; |
| 1125 | } |
| 1126 | |
| 1127 | void lightrec_free_opcode_list(struct lightrec_state *state, struct block *block) |
| 1128 | { |
| 1129 | lightrec_free(state, MEM_FOR_IR, |
| 1130 | sizeof(*block->opcode_list) * block->nb_ops, |
| 1131 | block->opcode_list); |
| 1132 | } |
| 1133 | |
| 1134 | static unsigned int lightrec_get_mips_block_len(const u32 *src) |
| 1135 | { |
| 1136 | unsigned int i; |
| 1137 | union code c; |
| 1138 | |
| 1139 | for (i = 1; ; i++) { |
| 1140 | c.opcode = LE32TOH(*src++); |
| 1141 | |
| 1142 | if (is_syscall(c)) |
| 1143 | return i; |
| 1144 | |
| 1145 | if (is_unconditional_jump(c)) |
| 1146 | return i + 1; |
| 1147 | } |
| 1148 | } |
| 1149 | |
| 1150 | static struct opcode * lightrec_disassemble(struct lightrec_state *state, |
| 1151 | const u32 *src, unsigned int *len) |
| 1152 | { |
| 1153 | struct opcode *list; |
| 1154 | unsigned int i, length; |
| 1155 | |
| 1156 | length = lightrec_get_mips_block_len(src); |
| 1157 | |
| 1158 | list = lightrec_malloc(state, MEM_FOR_IR, sizeof(*list) * length); |
| 1159 | if (!list) { |
| 1160 | pr_err("Unable to allocate memory\n"); |
| 1161 | return NULL; |
| 1162 | } |
| 1163 | |
| 1164 | for (i = 0; i < length; i++) { |
| 1165 | list[i].opcode = LE32TOH(src[i]); |
| 1166 | list[i].flags = 0; |
| 1167 | } |
| 1168 | |
| 1169 | *len = length * sizeof(u32); |
| 1170 | |
| 1171 | return list; |
| 1172 | } |
| 1173 | |
| 1174 | static struct block * lightrec_precompile_block(struct lightrec_state *state, |
| 1175 | u32 pc) |
| 1176 | { |
| 1177 | struct opcode *list; |
| 1178 | struct block *block; |
| 1179 | void *host; |
| 1180 | const struct lightrec_mem_map *map = lightrec_get_map(state, &host, kunseg(pc)); |
| 1181 | const u32 *code = (u32 *) host; |
| 1182 | unsigned int length; |
| 1183 | bool fully_tagged; |
| 1184 | |
| 1185 | if (!map) |
| 1186 | return NULL; |
| 1187 | |
| 1188 | block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block)); |
| 1189 | if (!block) { |
| 1190 | pr_err("Unable to recompile block: Out of memory\n"); |
| 1191 | return NULL; |
| 1192 | } |
| 1193 | |
| 1194 | list = lightrec_disassemble(state, code, &length); |
| 1195 | if (!list) { |
| 1196 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 1197 | return NULL; |
| 1198 | } |
| 1199 | |
| 1200 | block->pc = pc; |
| 1201 | block->_jit = NULL; |
| 1202 | block->function = NULL; |
| 1203 | block->opcode_list = list; |
| 1204 | block->code = code; |
| 1205 | block->next = NULL; |
| 1206 | block->flags = 0; |
| 1207 | block->code_size = 0; |
| 1208 | block->precompile_date = state->current_cycle; |
| 1209 | #if ENABLE_THREADED_COMPILER |
| 1210 | block->op_list_freed = (atomic_flag)ATOMIC_FLAG_INIT; |
| 1211 | #endif |
| 1212 | block->nb_ops = length / sizeof(u32); |
| 1213 | |
| 1214 | lightrec_optimize(state, block); |
| 1215 | |
| 1216 | length = block->nb_ops * sizeof(u32); |
| 1217 | |
| 1218 | lightrec_register(MEM_FOR_MIPS_CODE, length); |
| 1219 | |
| 1220 | if (ENABLE_DISASSEMBLER) { |
| 1221 | pr_debug("Disassembled block at PC: 0x%08x\n", block->pc); |
| 1222 | lightrec_print_disassembly(block, code); |
| 1223 | } |
| 1224 | |
| 1225 | pr_debug("Block size: %hu opcodes\n", block->nb_ops); |
| 1226 | |
| 1227 | /* If the first opcode is an 'impossible' branch, never compile the |
| 1228 | * block */ |
| 1229 | if (should_emulate(block->opcode_list)) |
| 1230 | block->flags |= BLOCK_NEVER_COMPILE; |
| 1231 | |
| 1232 | fully_tagged = lightrec_block_is_fully_tagged(block); |
| 1233 | if (fully_tagged) |
| 1234 | block->flags |= BLOCK_FULLY_TAGGED; |
| 1235 | |
| 1236 | if (OPT_REPLACE_MEMSET && (block->flags & BLOCK_IS_MEMSET)) |
| 1237 | lut_write(state, lut_offset(pc), state->memset_func); |
| 1238 | |
| 1239 | block->hash = lightrec_calculate_block_hash(block); |
| 1240 | |
| 1241 | pr_debug("Recompile count: %u\n", state->nb_precompile++); |
| 1242 | |
| 1243 | return block; |
| 1244 | } |
| 1245 | |
| 1246 | static bool lightrec_block_is_fully_tagged(const struct block *block) |
| 1247 | { |
| 1248 | const struct opcode *op; |
| 1249 | unsigned int i; |
| 1250 | |
| 1251 | for (i = 0; i < block->nb_ops; i++) { |
| 1252 | op = &block->opcode_list[i]; |
| 1253 | |
| 1254 | /* Verify that all load/stores of the opcode list |
| 1255 | * Check all loads/stores of the opcode list and mark the |
| 1256 | * block as fully compiled if they all have been tagged. */ |
| 1257 | switch (op->c.i.op) { |
| 1258 | case OP_LB: |
| 1259 | case OP_LH: |
| 1260 | case OP_LWL: |
| 1261 | case OP_LW: |
| 1262 | case OP_LBU: |
| 1263 | case OP_LHU: |
| 1264 | case OP_LWR: |
| 1265 | case OP_SB: |
| 1266 | case OP_SH: |
| 1267 | case OP_SWL: |
| 1268 | case OP_SW: |
| 1269 | case OP_SWR: |
| 1270 | case OP_LWC2: |
| 1271 | case OP_SWC2: |
| 1272 | if (!LIGHTREC_FLAGS_GET_IO_MODE(op->flags)) |
| 1273 | return false; |
| 1274 | fallthrough; |
| 1275 | default: |
| 1276 | continue; |
| 1277 | } |
| 1278 | } |
| 1279 | |
| 1280 | return true; |
| 1281 | } |
| 1282 | |
| 1283 | static void lightrec_reap_block(struct lightrec_state *state, void *data) |
| 1284 | { |
| 1285 | struct block *block = data; |
| 1286 | |
| 1287 | pr_debug("Reap dead block at PC 0x%08x\n", block->pc); |
| 1288 | lightrec_unregister_block(state->block_cache, block); |
| 1289 | lightrec_free_block(state, block); |
| 1290 | } |
| 1291 | |
| 1292 | static void lightrec_reap_jit(struct lightrec_state *state, void *data) |
| 1293 | { |
| 1294 | _jit_destroy_state(data); |
| 1295 | } |
| 1296 | |
| 1297 | static void lightrec_free_function(struct lightrec_state *state, void *fn) |
| 1298 | { |
| 1299 | if (ENABLE_CODE_BUFFER && state->tlsf) { |
| 1300 | pr_debug("Freeing code block at 0x%" PRIxPTR "\n", (uintptr_t) fn); |
| 1301 | lightrec_free_code(state, fn); |
| 1302 | } |
| 1303 | } |
| 1304 | |
| 1305 | static void lightrec_reap_function(struct lightrec_state *state, void *data) |
| 1306 | { |
| 1307 | lightrec_free_function(state, data); |
| 1308 | } |
| 1309 | |
| 1310 | int lightrec_compile_block(struct lightrec_cstate *cstate, |
| 1311 | struct block *block) |
| 1312 | { |
| 1313 | struct lightrec_state *state = cstate->state; |
| 1314 | struct lightrec_branch_target *target; |
| 1315 | bool op_list_freed = false, fully_tagged = false; |
| 1316 | struct block *block2; |
| 1317 | struct opcode *elm; |
| 1318 | jit_state_t *_jit, *oldjit; |
| 1319 | jit_node_t *start_of_block; |
| 1320 | bool skip_next = false; |
| 1321 | void *old_fn, *new_fn; |
| 1322 | unsigned int i, j; |
| 1323 | u32 offset; |
| 1324 | |
| 1325 | fully_tagged = lightrec_block_is_fully_tagged(block); |
| 1326 | if (fully_tagged) |
| 1327 | block->flags |= BLOCK_FULLY_TAGGED; |
| 1328 | |
| 1329 | _jit = jit_new_state(); |
| 1330 | if (!_jit) |
| 1331 | return -ENOMEM; |
| 1332 | |
| 1333 | oldjit = block->_jit; |
| 1334 | old_fn = block->function; |
| 1335 | block->_jit = _jit; |
| 1336 | |
| 1337 | lightrec_regcache_reset(cstate->reg_cache); |
| 1338 | cstate->cycles = 0; |
| 1339 | cstate->nb_branches = 0; |
| 1340 | cstate->nb_local_branches = 0; |
| 1341 | cstate->nb_targets = 0; |
| 1342 | |
| 1343 | jit_prolog(); |
| 1344 | jit_tramp(256); |
| 1345 | |
| 1346 | start_of_block = jit_label(); |
| 1347 | |
| 1348 | for (i = 0; i < block->nb_ops; i++) { |
| 1349 | elm = &block->opcode_list[i]; |
| 1350 | |
| 1351 | if (skip_next) { |
| 1352 | skip_next = false; |
| 1353 | continue; |
| 1354 | } |
| 1355 | |
| 1356 | cstate->cycles += lightrec_cycles_of_opcode(elm->c); |
| 1357 | |
| 1358 | if (should_emulate(elm)) { |
| 1359 | pr_debug("Branch at offset 0x%x will be emulated\n", |
| 1360 | i << 2); |
| 1361 | |
| 1362 | lightrec_emit_eob(cstate, block, i, false); |
| 1363 | skip_next = !(elm->flags & LIGHTREC_NO_DS); |
| 1364 | } else { |
| 1365 | lightrec_rec_opcode(cstate, block, i); |
| 1366 | skip_next = has_delay_slot(elm->c) && |
| 1367 | !(elm->flags & LIGHTREC_NO_DS); |
| 1368 | #if _WIN32 |
| 1369 | /* FIXME: GNU Lightning on Windows seems to use our |
| 1370 | * mapped registers as temporaries. Until the actual bug |
| 1371 | * is found and fixed, unconditionally mark our |
| 1372 | * registers as live here. */ |
| 1373 | lightrec_regcache_mark_live(cstate->reg_cache, _jit); |
| 1374 | #endif |
| 1375 | } |
| 1376 | } |
| 1377 | |
| 1378 | for (i = 0; i < cstate->nb_branches; i++) |
| 1379 | jit_patch(cstate->branches[i]); |
| 1380 | |
| 1381 | for (i = 0; i < cstate->nb_local_branches; i++) { |
| 1382 | struct lightrec_branch *branch = &cstate->local_branches[i]; |
| 1383 | |
| 1384 | pr_debug("Patch local branch to offset 0x%x\n", |
| 1385 | branch->target << 2); |
| 1386 | |
| 1387 | if (branch->target == 0) { |
| 1388 | jit_patch_at(branch->branch, start_of_block); |
| 1389 | continue; |
| 1390 | } |
| 1391 | |
| 1392 | for (j = 0; j < cstate->nb_targets; j++) { |
| 1393 | if (cstate->targets[j].offset == branch->target) { |
| 1394 | jit_patch_at(branch->branch, |
| 1395 | cstate->targets[j].label); |
| 1396 | break; |
| 1397 | } |
| 1398 | } |
| 1399 | |
| 1400 | if (j == cstate->nb_targets) |
| 1401 | pr_err("Unable to find branch target\n"); |
| 1402 | } |
| 1403 | |
| 1404 | jit_ldxi(JIT_R0, LIGHTREC_REG_STATE, |
| 1405 | offsetof(struct lightrec_state, eob_wrapper_func)); |
| 1406 | |
| 1407 | jit_jmpr(JIT_R0); |
| 1408 | |
| 1409 | jit_ret(); |
| 1410 | jit_epilog(); |
| 1411 | |
| 1412 | new_fn = lightrec_emit_code(state, block, _jit, &block->code_size); |
| 1413 | if (!new_fn) { |
| 1414 | if (!ENABLE_THREADED_COMPILER) |
| 1415 | pr_err("Unable to compile block!\n"); |
| 1416 | block->_jit = oldjit; |
| 1417 | _jit_destroy_state(_jit); |
| 1418 | return -ENOMEM; |
| 1419 | } |
| 1420 | |
| 1421 | block->function = new_fn; |
| 1422 | block->flags &= ~BLOCK_SHOULD_RECOMPILE; |
| 1423 | |
| 1424 | /* Add compiled function to the LUT */ |
| 1425 | lut_write(state, lut_offset(block->pc), block->function); |
| 1426 | |
| 1427 | if (ENABLE_THREADED_COMPILER) { |
| 1428 | /* Since we might try to reap the same block multiple times, |
| 1429 | * we need the reaper to wait until everything has been |
| 1430 | * submitted, so that the duplicate entries can be dropped. */ |
| 1431 | lightrec_reaper_pause(state->reaper); |
| 1432 | } |
| 1433 | |
| 1434 | /* Detect old blocks that have been covered by the new one */ |
| 1435 | for (i = 0; i < cstate->nb_targets; i++) { |
| 1436 | target = &cstate->targets[i]; |
| 1437 | |
| 1438 | if (!target->offset) |
| 1439 | continue; |
| 1440 | |
| 1441 | offset = block->pc + target->offset * sizeof(u32); |
| 1442 | block2 = lightrec_find_block(state->block_cache, offset); |
| 1443 | if (block2) { |
| 1444 | /* No need to check if block2 is compilable - it must |
| 1445 | * be, otherwise block wouldn't be compilable either */ |
| 1446 | |
| 1447 | /* Set the "block dead" flag to prevent the dynarec from |
| 1448 | * recompiling this block */ |
| 1449 | block2->flags |= BLOCK_IS_DEAD; |
| 1450 | |
| 1451 | /* If block2 was pending for compilation, cancel it. |
| 1452 | * If it's being compiled right now, wait until it |
| 1453 | * finishes. */ |
| 1454 | if (ENABLE_THREADED_COMPILER) |
| 1455 | lightrec_recompiler_remove(state->rec, block2); |
| 1456 | } |
| 1457 | |
| 1458 | /* We know from now on that block2 (if present) isn't going to |
| 1459 | * be compiled. We can override the LUT entry with our new |
| 1460 | * block's entry point. */ |
| 1461 | offset = lut_offset(block->pc) + target->offset; |
| 1462 | lut_write(state, offset, jit_address(target->label)); |
| 1463 | |
| 1464 | if (block2) { |
| 1465 | pr_debug("Reap block 0x%08x as it's covered by block " |
| 1466 | "0x%08x\n", block2->pc, block->pc); |
| 1467 | |
| 1468 | /* Finally, reap the block. */ |
| 1469 | if (ENABLE_THREADED_COMPILER) { |
| 1470 | lightrec_reaper_add(state->reaper, |
| 1471 | lightrec_reap_block, |
| 1472 | block2); |
| 1473 | } else { |
| 1474 | lightrec_unregister_block(state->block_cache, block2); |
| 1475 | lightrec_free_block(state, block2); |
| 1476 | } |
| 1477 | } |
| 1478 | } |
| 1479 | |
| 1480 | if (ENABLE_THREADED_COMPILER) |
| 1481 | lightrec_reaper_continue(state->reaper); |
| 1482 | |
| 1483 | if (ENABLE_DISASSEMBLER) { |
| 1484 | pr_debug("Compiling block at PC: 0x%08x\n", block->pc); |
| 1485 | jit_disassemble(); |
| 1486 | } |
| 1487 | |
| 1488 | jit_clear_state(); |
| 1489 | |
| 1490 | #if ENABLE_THREADED_COMPILER |
| 1491 | if (fully_tagged) |
| 1492 | op_list_freed = atomic_flag_test_and_set(&block->op_list_freed); |
| 1493 | #endif |
| 1494 | if (fully_tagged && !op_list_freed) { |
| 1495 | pr_debug("Block PC 0x%08x is fully tagged" |
| 1496 | " - free opcode list\n", block->pc); |
| 1497 | lightrec_free_opcode_list(state, block); |
| 1498 | block->opcode_list = NULL; |
| 1499 | } |
| 1500 | |
| 1501 | if (oldjit) { |
| 1502 | pr_debug("Block 0x%08x recompiled, reaping old jit context.\n", |
| 1503 | block->pc); |
| 1504 | |
| 1505 | if (ENABLE_THREADED_COMPILER) { |
| 1506 | lightrec_reaper_add(state->reaper, |
| 1507 | lightrec_reap_jit, oldjit); |
| 1508 | lightrec_reaper_add(state->reaper, |
| 1509 | lightrec_reap_function, old_fn); |
| 1510 | } else { |
| 1511 | _jit_destroy_state(oldjit); |
| 1512 | lightrec_free_function(state, old_fn); |
| 1513 | } |
| 1514 | } |
| 1515 | |
| 1516 | return 0; |
| 1517 | } |
| 1518 | |
| 1519 | static void lightrec_print_info(struct lightrec_state *state) |
| 1520 | { |
| 1521 | if ((state->current_cycle & ~0xfffffff) != state->old_cycle_counter) { |
| 1522 | pr_info("Lightrec RAM usage: IR %u KiB, CODE %u KiB, " |
| 1523 | "MIPS %u KiB, TOTAL %u KiB, avg. IPI %f\n", |
| 1524 | lightrec_get_mem_usage(MEM_FOR_IR) / 1024, |
| 1525 | lightrec_get_mem_usage(MEM_FOR_CODE) / 1024, |
| 1526 | lightrec_get_mem_usage(MEM_FOR_MIPS_CODE) / 1024, |
| 1527 | lightrec_get_total_mem_usage() / 1024, |
| 1528 | lightrec_get_average_ipi()); |
| 1529 | state->old_cycle_counter = state->current_cycle & ~0xfffffff; |
| 1530 | } |
| 1531 | } |
| 1532 | |
| 1533 | u32 lightrec_execute(struct lightrec_state *state, u32 pc, u32 target_cycle) |
| 1534 | { |
| 1535 | s32 (*func)(void *, s32) = (void *)state->dispatcher->function; |
| 1536 | void *block_trace; |
| 1537 | s32 cycles_delta; |
| 1538 | |
| 1539 | state->exit_flags = LIGHTREC_EXIT_NORMAL; |
| 1540 | |
| 1541 | /* Handle the cycle counter overflowing */ |
| 1542 | if (unlikely(target_cycle < state->current_cycle)) |
| 1543 | target_cycle = UINT_MAX; |
| 1544 | |
| 1545 | state->target_cycle = target_cycle; |
| 1546 | state->next_pc = pc; |
| 1547 | |
| 1548 | block_trace = get_next_block_func(state, pc); |
| 1549 | if (block_trace) { |
| 1550 | cycles_delta = state->target_cycle - state->current_cycle; |
| 1551 | |
| 1552 | cycles_delta = (*func)(block_trace, cycles_delta); |
| 1553 | |
| 1554 | state->current_cycle = state->target_cycle - cycles_delta; |
| 1555 | } |
| 1556 | |
| 1557 | if (ENABLE_THREADED_COMPILER) |
| 1558 | lightrec_reaper_reap(state->reaper); |
| 1559 | |
| 1560 | if (LOG_LEVEL >= INFO_L) |
| 1561 | lightrec_print_info(state); |
| 1562 | |
| 1563 | return state->next_pc; |
| 1564 | } |
| 1565 | |
| 1566 | u32 lightrec_execute_one(struct lightrec_state *state, u32 pc) |
| 1567 | { |
| 1568 | return lightrec_execute(state, pc, state->current_cycle); |
| 1569 | } |
| 1570 | |
| 1571 | u32 lightrec_run_interpreter(struct lightrec_state *state, u32 pc) |
| 1572 | { |
| 1573 | struct block *block = lightrec_get_block(state, pc); |
| 1574 | if (!block) |
| 1575 | return 0; |
| 1576 | |
| 1577 | state->exit_flags = LIGHTREC_EXIT_NORMAL; |
| 1578 | |
| 1579 | pc = lightrec_emulate_block(state, block, pc); |
| 1580 | |
| 1581 | if (LOG_LEVEL >= INFO_L) |
| 1582 | lightrec_print_info(state); |
| 1583 | |
| 1584 | return pc; |
| 1585 | } |
| 1586 | |
| 1587 | void lightrec_free_block(struct lightrec_state *state, struct block *block) |
| 1588 | { |
| 1589 | lightrec_unregister(MEM_FOR_MIPS_CODE, block->nb_ops * sizeof(u32)); |
| 1590 | if (block->opcode_list) |
| 1591 | lightrec_free_opcode_list(state, block); |
| 1592 | if (block->_jit) |
| 1593 | _jit_destroy_state(block->_jit); |
| 1594 | if (block->function) { |
| 1595 | lightrec_free_function(state, block->function); |
| 1596 | lightrec_unregister(MEM_FOR_CODE, block->code_size); |
| 1597 | } |
| 1598 | lightrec_free(state, MEM_FOR_IR, sizeof(*block), block); |
| 1599 | } |
| 1600 | |
| 1601 | struct lightrec_cstate * lightrec_create_cstate(struct lightrec_state *state) |
| 1602 | { |
| 1603 | struct lightrec_cstate *cstate; |
| 1604 | |
| 1605 | cstate = lightrec_malloc(state, MEM_FOR_LIGHTREC, sizeof(*cstate)); |
| 1606 | if (!cstate) |
| 1607 | return NULL; |
| 1608 | |
| 1609 | cstate->reg_cache = lightrec_regcache_init(state); |
| 1610 | if (!cstate->reg_cache) { |
| 1611 | lightrec_free(state, MEM_FOR_LIGHTREC, sizeof(*cstate), cstate); |
| 1612 | return NULL; |
| 1613 | } |
| 1614 | |
| 1615 | cstate->state = state; |
| 1616 | |
| 1617 | return cstate; |
| 1618 | } |
| 1619 | |
| 1620 | void lightrec_free_cstate(struct lightrec_cstate *cstate) |
| 1621 | { |
| 1622 | lightrec_free_regcache(cstate->reg_cache); |
| 1623 | lightrec_free(cstate->state, MEM_FOR_LIGHTREC, sizeof(*cstate), cstate); |
| 1624 | } |
| 1625 | |
| 1626 | struct lightrec_state * lightrec_init(char *argv0, |
| 1627 | const struct lightrec_mem_map *map, |
| 1628 | size_t nb, |
| 1629 | const struct lightrec_ops *ops) |
| 1630 | { |
| 1631 | const struct lightrec_mem_map *codebuf_map = &map[PSX_MAP_CODE_BUFFER]; |
| 1632 | struct lightrec_state *state; |
| 1633 | uintptr_t addr; |
| 1634 | void *tlsf = NULL; |
| 1635 | bool with_32bit_lut = false; |
| 1636 | size_t lut_size; |
| 1637 | |
| 1638 | /* Sanity-check ops */ |
| 1639 | if (!ops || !ops->cop2_op || !ops->enable_ram) { |
| 1640 | pr_err("Missing callbacks in lightrec_ops structure\n"); |
| 1641 | return NULL; |
| 1642 | } |
| 1643 | |
| 1644 | if (ENABLE_CODE_BUFFER && nb > PSX_MAP_CODE_BUFFER |
| 1645 | && codebuf_map->address) { |
| 1646 | tlsf = tlsf_create_with_pool(codebuf_map->address, |
| 1647 | codebuf_map->length); |
| 1648 | if (!tlsf) { |
| 1649 | pr_err("Unable to initialize code buffer\n"); |
| 1650 | return NULL; |
| 1651 | } |
| 1652 | |
| 1653 | if (__WORDSIZE == 64) { |
| 1654 | addr = (uintptr_t) codebuf_map->address + codebuf_map->length - 1; |
| 1655 | with_32bit_lut = addr == (u32) addr; |
| 1656 | } |
| 1657 | } |
| 1658 | |
| 1659 | if (with_32bit_lut) |
| 1660 | lut_size = CODE_LUT_SIZE * 4; |
| 1661 | else |
| 1662 | lut_size = CODE_LUT_SIZE * sizeof(void *); |
| 1663 | |
| 1664 | init_jit(argv0); |
| 1665 | |
| 1666 | state = calloc(1, sizeof(*state) + lut_size); |
| 1667 | if (!state) |
| 1668 | goto err_finish_jit; |
| 1669 | |
| 1670 | lightrec_register(MEM_FOR_LIGHTREC, sizeof(*state) + lut_size); |
| 1671 | |
| 1672 | state->tlsf = tlsf; |
| 1673 | state->with_32bit_lut = with_32bit_lut; |
| 1674 | |
| 1675 | state->block_cache = lightrec_blockcache_init(state); |
| 1676 | if (!state->block_cache) |
| 1677 | goto err_free_state; |
| 1678 | |
| 1679 | if (ENABLE_THREADED_COMPILER) { |
| 1680 | state->rec = lightrec_recompiler_init(state); |
| 1681 | if (!state->rec) |
| 1682 | goto err_free_block_cache; |
| 1683 | |
| 1684 | state->reaper = lightrec_reaper_init(state); |
| 1685 | if (!state->reaper) |
| 1686 | goto err_free_recompiler; |
| 1687 | } else { |
| 1688 | state->cstate = lightrec_create_cstate(state); |
| 1689 | if (!state->cstate) |
| 1690 | goto err_free_block_cache; |
| 1691 | } |
| 1692 | |
| 1693 | state->nb_maps = nb; |
| 1694 | state->maps = map; |
| 1695 | |
| 1696 | memcpy(&state->ops, ops, sizeof(*ops)); |
| 1697 | |
| 1698 | state->dispatcher = generate_dispatcher(state); |
| 1699 | if (!state->dispatcher) |
| 1700 | goto err_free_reaper; |
| 1701 | |
| 1702 | state->c_wrapper_block = generate_wrapper(state); |
| 1703 | if (!state->c_wrapper_block) |
| 1704 | goto err_free_dispatcher; |
| 1705 | |
| 1706 | state->c_wrappers[C_WRAPPER_RW] = lightrec_rw_cb; |
| 1707 | state->c_wrappers[C_WRAPPER_RW_GENERIC] = lightrec_rw_generic_cb; |
| 1708 | state->c_wrappers[C_WRAPPER_MTC] = lightrec_mtc_cb; |
| 1709 | state->c_wrappers[C_WRAPPER_CP] = lightrec_cp_cb; |
| 1710 | state->c_wrappers[C_WRAPPER_SYSCALL] = lightrec_syscall_cb; |
| 1711 | state->c_wrappers[C_WRAPPER_BREAK] = lightrec_break_cb; |
| 1712 | |
| 1713 | map = &state->maps[PSX_MAP_BIOS]; |
| 1714 | state->offset_bios = (uintptr_t)map->address - map->pc; |
| 1715 | |
| 1716 | map = &state->maps[PSX_MAP_SCRATCH_PAD]; |
| 1717 | state->offset_scratch = (uintptr_t)map->address - map->pc; |
| 1718 | |
| 1719 | map = &state->maps[PSX_MAP_KERNEL_USER_RAM]; |
| 1720 | state->offset_ram = (uintptr_t)map->address - map->pc; |
| 1721 | |
| 1722 | if (state->maps[PSX_MAP_MIRROR1].address == map->address + 0x200000 && |
| 1723 | state->maps[PSX_MAP_MIRROR2].address == map->address + 0x400000 && |
| 1724 | state->maps[PSX_MAP_MIRROR3].address == map->address + 0x600000) |
| 1725 | state->mirrors_mapped = true; |
| 1726 | |
| 1727 | if (state->offset_bios == 0 && |
| 1728 | state->offset_scratch == 0 && |
| 1729 | state->offset_ram == 0 && |
| 1730 | state->mirrors_mapped) { |
| 1731 | pr_info("Memory map is perfect. Emitted code will be best.\n"); |
| 1732 | } else { |
| 1733 | pr_info("Memory map is sub-par. Emitted code will be slow.\n"); |
| 1734 | } |
| 1735 | |
| 1736 | if (state->with_32bit_lut) |
| 1737 | pr_info("Using 32-bit LUT\n"); |
| 1738 | |
| 1739 | return state; |
| 1740 | |
| 1741 | err_free_dispatcher: |
| 1742 | lightrec_free_block(state, state->dispatcher); |
| 1743 | err_free_reaper: |
| 1744 | if (ENABLE_THREADED_COMPILER) |
| 1745 | lightrec_reaper_destroy(state->reaper); |
| 1746 | err_free_recompiler: |
| 1747 | if (ENABLE_THREADED_COMPILER) |
| 1748 | lightrec_free_recompiler(state->rec); |
| 1749 | else |
| 1750 | lightrec_free_cstate(state->cstate); |
| 1751 | err_free_block_cache: |
| 1752 | lightrec_free_block_cache(state->block_cache); |
| 1753 | err_free_state: |
| 1754 | lightrec_unregister(MEM_FOR_LIGHTREC, sizeof(*state) + |
| 1755 | lut_elm_size(state) * CODE_LUT_SIZE); |
| 1756 | free(state); |
| 1757 | err_finish_jit: |
| 1758 | finish_jit(); |
| 1759 | if (ENABLE_CODE_BUFFER && tlsf) |
| 1760 | tlsf_destroy(tlsf); |
| 1761 | return NULL; |
| 1762 | } |
| 1763 | |
| 1764 | void lightrec_destroy(struct lightrec_state *state) |
| 1765 | { |
| 1766 | /* Force a print info on destroy*/ |
| 1767 | state->current_cycle = ~state->current_cycle; |
| 1768 | lightrec_print_info(state); |
| 1769 | |
| 1770 | if (ENABLE_THREADED_COMPILER) { |
| 1771 | lightrec_free_recompiler(state->rec); |
| 1772 | lightrec_reaper_destroy(state->reaper); |
| 1773 | } else { |
| 1774 | lightrec_free_cstate(state->cstate); |
| 1775 | } |
| 1776 | |
| 1777 | lightrec_free_block_cache(state->block_cache); |
| 1778 | lightrec_free_block(state, state->dispatcher); |
| 1779 | lightrec_free_block(state, state->c_wrapper_block); |
| 1780 | finish_jit(); |
| 1781 | if (ENABLE_CODE_BUFFER && state->tlsf) |
| 1782 | tlsf_destroy(state->tlsf); |
| 1783 | |
| 1784 | lightrec_unregister(MEM_FOR_LIGHTREC, sizeof(*state) + |
| 1785 | lut_elm_size(state) * CODE_LUT_SIZE); |
| 1786 | free(state); |
| 1787 | } |
| 1788 | |
| 1789 | void lightrec_invalidate(struct lightrec_state *state, u32 addr, u32 len) |
| 1790 | { |
| 1791 | u32 kaddr = kunseg(addr & ~0x3); |
| 1792 | enum psx_map idx = lightrec_get_map_idx(state, kaddr); |
| 1793 | |
| 1794 | switch (idx) { |
| 1795 | case PSX_MAP_MIRROR1: |
| 1796 | case PSX_MAP_MIRROR2: |
| 1797 | case PSX_MAP_MIRROR3: |
| 1798 | /* Handle mirrors */ |
| 1799 | kaddr &= RAM_SIZE - 1; |
| 1800 | fallthrough; |
| 1801 | case PSX_MAP_KERNEL_USER_RAM: |
| 1802 | break; |
| 1803 | default: |
| 1804 | return; |
| 1805 | } |
| 1806 | |
| 1807 | memset(lut_address(state, lut_offset(kaddr)), 0, |
| 1808 | ((len + 3) / 4) * lut_elm_size(state)); |
| 1809 | } |
| 1810 | |
| 1811 | void lightrec_invalidate_all(struct lightrec_state *state) |
| 1812 | { |
| 1813 | memset(state->code_lut, 0, lut_elm_size(state) * CODE_LUT_SIZE); |
| 1814 | } |
| 1815 | |
| 1816 | void lightrec_set_invalidate_mode(struct lightrec_state *state, bool dma_only) |
| 1817 | { |
| 1818 | if (state->invalidate_from_dma_only != dma_only) |
| 1819 | lightrec_invalidate_all(state); |
| 1820 | |
| 1821 | state->invalidate_from_dma_only = dma_only; |
| 1822 | } |
| 1823 | |
| 1824 | void lightrec_set_exit_flags(struct lightrec_state *state, u32 flags) |
| 1825 | { |
| 1826 | if (flags != LIGHTREC_EXIT_NORMAL) { |
| 1827 | state->exit_flags |= flags; |
| 1828 | state->target_cycle = state->current_cycle; |
| 1829 | } |
| 1830 | } |
| 1831 | |
| 1832 | u32 lightrec_exit_flags(struct lightrec_state *state) |
| 1833 | { |
| 1834 | return state->exit_flags; |
| 1835 | } |
| 1836 | |
| 1837 | u32 lightrec_current_cycle_count(const struct lightrec_state *state) |
| 1838 | { |
| 1839 | return state->current_cycle; |
| 1840 | } |
| 1841 | |
| 1842 | void lightrec_reset_cycle_count(struct lightrec_state *state, u32 cycles) |
| 1843 | { |
| 1844 | state->current_cycle = cycles; |
| 1845 | |
| 1846 | if (state->target_cycle < cycles) |
| 1847 | state->target_cycle = cycles; |
| 1848 | } |
| 1849 | |
| 1850 | void lightrec_set_target_cycle_count(struct lightrec_state *state, u32 cycles) |
| 1851 | { |
| 1852 | if (state->exit_flags == LIGHTREC_EXIT_NORMAL) { |
| 1853 | if (cycles < state->current_cycle) |
| 1854 | cycles = state->current_cycle; |
| 1855 | |
| 1856 | state->target_cycle = cycles; |
| 1857 | } |
| 1858 | } |
| 1859 | |
| 1860 | struct lightrec_registers * lightrec_get_registers(struct lightrec_state *state) |
| 1861 | { |
| 1862 | return &state->regs; |
| 1863 | } |