1 /* 2 * PowerPC memory access emulation helpers for QEMU. 3 * 4 * Copyright (c) 2003-2007 Jocelyn Mayer 5 * 6 * This library is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * This library is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with this library; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include "qemu/osdep.h" 21 #include "cpu.h" 22 #include "exec/exec-all.h" 23 #include "qemu/host-utils.h" 24 #include "exec/helper-proto.h" 25 #include "helper_regs.h" 26 #include "exec/cpu_ldst.h" 27 #include "internal.h" 28 #include "qemu/atomic128.h" 29 30 /* #define DEBUG_OP */ 31 32 static inline bool needs_byteswap(const CPUPPCState *env) 33 { 34 #if TARGET_BIG_ENDIAN 35 return FIELD_EX64(env->msr, MSR, LE); 36 #else 37 return !FIELD_EX64(env->msr, MSR, LE); 38 #endif 39 } 40 41 /*****************************************************************************/ 42 /* Memory load and stores */ 43 44 static inline target_ulong addr_add(CPUPPCState *env, target_ulong addr, 45 target_long arg) 46 { 47 #if defined(TARGET_PPC64) 48 if (!msr_is_64bit(env, env->msr)) { 49 return (uint32_t)(addr + arg); 50 } else 51 #endif 52 { 53 return addr + arg; 54 } 55 } 56 57 static void *probe_contiguous(CPUPPCState *env, target_ulong addr, uint32_t nb, 58 MMUAccessType access_type, int mmu_idx, 59 uintptr_t raddr) 60 { 61 void *host1, *host2; 62 uint32_t nb_pg1, nb_pg2; 63 64 nb_pg1 = -(addr | TARGET_PAGE_MASK); 65 if (likely(nb <= nb_pg1)) { 66 /* The entire operation is on a single page. */ 67 return probe_access(env, addr, nb, access_type, mmu_idx, raddr); 68 } 69 70 /* The operation spans two pages. */ 71 nb_pg2 = nb - nb_pg1; 72 host1 = probe_access(env, addr, nb_pg1, access_type, mmu_idx, raddr); 73 addr = addr_add(env, addr, nb_pg1); 74 host2 = probe_access(env, addr, nb_pg2, access_type, mmu_idx, raddr); 75 76 /* If the two host pages are contiguous, optimize. */ 77 if (host2 == host1 + nb_pg1) { 78 return host1; 79 } 80 return NULL; 81 } 82 83 void helper_lmw(CPUPPCState *env, target_ulong addr, uint32_t reg) 84 { 85 uintptr_t raddr = GETPC(); 86 int mmu_idx = cpu_mmu_index(env, false); 87 void *host = probe_contiguous(env, addr, (32 - reg) * 4, 88 MMU_DATA_LOAD, mmu_idx, raddr); 89 90 if (likely(host)) { 91 /* Fast path -- the entire operation is in RAM at host. */ 92 for (; reg < 32; reg++) { 93 env->gpr[reg] = (uint32_t)ldl_be_p(host); 94 host += 4; 95 } 96 } else { 97 /* Slow path -- at least some of the operation requires i/o. */ 98 for (; reg < 32; reg++) { 99 env->gpr[reg] = cpu_ldl_mmuidx_ra(env, addr, mmu_idx, raddr); 100 addr = addr_add(env, addr, 4); 101 } 102 } 103 } 104 105 void helper_stmw(CPUPPCState *env, target_ulong addr, uint32_t reg) 106 { 107 uintptr_t raddr = GETPC(); 108 int mmu_idx = cpu_mmu_index(env, false); 109 void *host = probe_contiguous(env, addr, (32 - reg) * 4, 110 MMU_DATA_STORE, mmu_idx, raddr); 111 112 if (likely(host)) { 113 /* Fast path -- the entire operation is in RAM at host. */ 114 for (; reg < 32; reg++) { 115 stl_be_p(host, env->gpr[reg]); 116 host += 4; 117 } 118 } else { 119 /* Slow path -- at least some of the operation requires i/o. */ 120 for (; reg < 32; reg++) { 121 cpu_stl_mmuidx_ra(env, addr, env->gpr[reg], mmu_idx, raddr); 122 addr = addr_add(env, addr, 4); 123 } 124 } 125 } 126 127 static void do_lsw(CPUPPCState *env, target_ulong addr, uint32_t nb, 128 uint32_t reg, uintptr_t raddr) 129 { 130 int mmu_idx; 131 void *host; 132 uint32_t val; 133 134 if (unlikely(nb == 0)) { 135 return; 136 } 137 138 mmu_idx = cpu_mmu_index(env, false); 139 host = probe_contiguous(env, addr, nb, MMU_DATA_LOAD, mmu_idx, raddr); 140 141 if (likely(host)) { 142 /* Fast path -- the entire operation is in RAM at host. */ 143 for (; nb > 3; nb -= 4) { 144 env->gpr[reg] = (uint32_t)ldl_be_p(host); 145 reg = (reg + 1) % 32; 146 host += 4; 147 } 148 switch (nb) { 149 default: 150 return; 151 case 1: 152 val = ldub_p(host) << 24; 153 break; 154 case 2: 155 val = lduw_be_p(host) << 16; 156 break; 157 case 3: 158 val = (lduw_be_p(host) << 16) | (ldub_p(host + 2) << 8); 159 break; 160 } 161 } else { 162 /* Slow path -- at least some of the operation requires i/o. */ 163 for (; nb > 3; nb -= 4) { 164 env->gpr[reg] = cpu_ldl_mmuidx_ra(env, addr, mmu_idx, raddr); 165 reg = (reg + 1) % 32; 166 addr = addr_add(env, addr, 4); 167 } 168 switch (nb) { 169 default: 170 return; 171 case 1: 172 val = cpu_ldub_mmuidx_ra(env, addr, mmu_idx, raddr) << 24; 173 break; 174 case 2: 175 val = cpu_lduw_mmuidx_ra(env, addr, mmu_idx, raddr) << 16; 176 break; 177 case 3: 178 val = cpu_lduw_mmuidx_ra(env, addr, mmu_idx, raddr) << 16; 179 addr = addr_add(env, addr, 2); 180 val |= cpu_ldub_mmuidx_ra(env, addr, mmu_idx, raddr) << 8; 181 break; 182 } 183 } 184 env->gpr[reg] = val; 185 } 186 187 void helper_lsw(CPUPPCState *env, target_ulong addr, 188 uint32_t nb, uint32_t reg) 189 { 190 do_lsw(env, addr, nb, reg, GETPC()); 191 } 192 193 /* 194 * PPC32 specification says we must generate an exception if rA is in 195 * the range of registers to be loaded. In an other hand, IBM says 196 * this is valid, but rA won't be loaded. For now, I'll follow the 197 * spec... 198 */ 199 void helper_lswx(CPUPPCState *env, target_ulong addr, uint32_t reg, 200 uint32_t ra, uint32_t rb) 201 { 202 if (likely(xer_bc != 0)) { 203 int num_used_regs = DIV_ROUND_UP(xer_bc, 4); 204 if (unlikely((ra != 0 && lsw_reg_in_range(reg, num_used_regs, ra)) || 205 lsw_reg_in_range(reg, num_used_regs, rb))) { 206 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM, 207 POWERPC_EXCP_INVAL | 208 POWERPC_EXCP_INVAL_LSWX, GETPC()); 209 } else { 210 do_lsw(env, addr, xer_bc, reg, GETPC()); 211 } 212 } 213 } 214 215 void helper_stsw(CPUPPCState *env, target_ulong addr, uint32_t nb, 216 uint32_t reg) 217 { 218 uintptr_t raddr = GETPC(); 219 int mmu_idx; 220 void *host; 221 uint32_t val; 222 223 if (unlikely(nb == 0)) { 224 return; 225 } 226 227 mmu_idx = cpu_mmu_index(env, false); 228 host = probe_contiguous(env, addr, nb, MMU_DATA_STORE, mmu_idx, raddr); 229 230 if (likely(host)) { 231 /* Fast path -- the entire operation is in RAM at host. */ 232 for (; nb > 3; nb -= 4) { 233 stl_be_p(host, env->gpr[reg]); 234 reg = (reg + 1) % 32; 235 host += 4; 236 } 237 val = env->gpr[reg]; 238 switch (nb) { 239 case 1: 240 stb_p(host, val >> 24); 241 break; 242 case 2: 243 stw_be_p(host, val >> 16); 244 break; 245 case 3: 246 stw_be_p(host, val >> 16); 247 stb_p(host + 2, val >> 8); 248 break; 249 } 250 } else { 251 for (; nb > 3; nb -= 4) { 252 cpu_stl_mmuidx_ra(env, addr, env->gpr[reg], mmu_idx, raddr); 253 reg = (reg + 1) % 32; 254 addr = addr_add(env, addr, 4); 255 } 256 val = env->gpr[reg]; 257 switch (nb) { 258 case 1: 259 cpu_stb_mmuidx_ra(env, addr, val >> 24, mmu_idx, raddr); 260 break; 261 case 2: 262 cpu_stw_mmuidx_ra(env, addr, val >> 16, mmu_idx, raddr); 263 break; 264 case 3: 265 cpu_stw_mmuidx_ra(env, addr, val >> 16, mmu_idx, raddr); 266 addr = addr_add(env, addr, 2); 267 cpu_stb_mmuidx_ra(env, addr, val >> 8, mmu_idx, raddr); 268 break; 269 } 270 } 271 } 272 273 static void dcbz_common(CPUPPCState *env, target_ulong addr, 274 uint32_t opcode, bool epid, uintptr_t retaddr) 275 { 276 target_ulong mask, dcbz_size = env->dcache_line_size; 277 uint32_t i; 278 void *haddr; 279 int mmu_idx = epid ? PPC_TLB_EPID_STORE : cpu_mmu_index(env, false); 280 281 #if defined(TARGET_PPC64) 282 /* Check for dcbz vs dcbzl on 970 */ 283 if (env->excp_model == POWERPC_EXCP_970 && 284 !(opcode & 0x00200000) && ((env->spr[SPR_970_HID5] >> 7) & 0x3) == 1) { 285 dcbz_size = 32; 286 } 287 #endif 288 289 /* Align address */ 290 mask = ~(dcbz_size - 1); 291 addr &= mask; 292 293 /* Check reservation */ 294 if ((env->reserve_addr & mask) == addr) { 295 env->reserve_addr = (target_ulong)-1ULL; 296 } 297 298 /* Try fast path translate */ 299 haddr = probe_write(env, addr, dcbz_size, mmu_idx, retaddr); 300 if (haddr) { 301 memset(haddr, 0, dcbz_size); 302 } else { 303 /* Slow path */ 304 for (i = 0; i < dcbz_size; i += 8) { 305 cpu_stq_mmuidx_ra(env, addr + i, 0, mmu_idx, retaddr); 306 } 307 } 308 } 309 310 void helper_dcbz(CPUPPCState *env, target_ulong addr, uint32_t opcode) 311 { 312 dcbz_common(env, addr, opcode, false, GETPC()); 313 } 314 315 void helper_dcbzep(CPUPPCState *env, target_ulong addr, uint32_t opcode) 316 { 317 dcbz_common(env, addr, opcode, true, GETPC()); 318 } 319 320 void helper_icbi(CPUPPCState *env, target_ulong addr) 321 { 322 addr &= ~(env->dcache_line_size - 1); 323 /* 324 * Invalidate one cache line : 325 * PowerPC specification says this is to be treated like a load 326 * (not a fetch) by the MMU. To be sure it will be so, 327 * do the load "by hand". 328 */ 329 cpu_ldl_data_ra(env, addr, GETPC()); 330 } 331 332 void helper_icbiep(CPUPPCState *env, target_ulong addr) 333 { 334 #if !defined(CONFIG_USER_ONLY) 335 /* See comments above */ 336 addr &= ~(env->dcache_line_size - 1); 337 cpu_ldl_mmuidx_ra(env, addr, PPC_TLB_EPID_LOAD, GETPC()); 338 #endif 339 } 340 341 /* XXX: to be tested */ 342 target_ulong helper_lscbx(CPUPPCState *env, target_ulong addr, uint32_t reg, 343 uint32_t ra, uint32_t rb) 344 { 345 int i, c, d; 346 347 d = 24; 348 for (i = 0; i < xer_bc; i++) { 349 c = cpu_ldub_data_ra(env, addr, GETPC()); 350 addr = addr_add(env, addr, 1); 351 /* ra (if not 0) and rb are never modified */ 352 if (likely(reg != rb && (ra == 0 || reg != ra))) { 353 env->gpr[reg] = (env->gpr[reg] & ~(0xFF << d)) | (c << d); 354 } 355 if (unlikely(c == xer_cmp)) { 356 break; 357 } 358 if (likely(d != 0)) { 359 d -= 8; 360 } else { 361 d = 24; 362 reg++; 363 reg = reg & 0x1F; 364 } 365 } 366 return i; 367 } 368 369 /*****************************************************************************/ 370 /* Altivec extension helpers */ 371 #if HOST_BIG_ENDIAN 372 #define HI_IDX 0 373 #define LO_IDX 1 374 #else 375 #define HI_IDX 1 376 #define LO_IDX 0 377 #endif 378 379 /* 380 * We use MSR_LE to determine index ordering in a vector. However, 381 * byteswapping is not simply controlled by MSR_LE. We also need to 382 * take into account endianness of the target. This is done for the 383 * little-endian PPC64 user-mode target. 384 */ 385 386 #define LVE(name, access, swap, element) \ 387 void helper_##name(CPUPPCState *env, ppc_avr_t *r, \ 388 target_ulong addr) \ 389 { \ 390 size_t n_elems = ARRAY_SIZE(r->element); \ 391 int adjust = HI_IDX * (n_elems - 1); \ 392 int sh = sizeof(r->element[0]) >> 1; \ 393 int index = (addr & 0xf) >> sh; \ 394 if (FIELD_EX64(env->msr, MSR, LE)) { \ 395 index = n_elems - index - 1; \ 396 } \ 397 \ 398 if (needs_byteswap(env)) { \ 399 r->element[LO_IDX ? index : (adjust - index)] = \ 400 swap(access(env, addr, GETPC())); \ 401 } else { \ 402 r->element[LO_IDX ? index : (adjust - index)] = \ 403 access(env, addr, GETPC()); \ 404 } \ 405 } 406 #define I(x) (x) 407 LVE(lvebx, cpu_ldub_data_ra, I, u8) 408 LVE(lvehx, cpu_lduw_data_ra, bswap16, u16) 409 LVE(lvewx, cpu_ldl_data_ra, bswap32, u32) 410 #undef I 411 #undef LVE 412 413 #define STVE(name, access, swap, element) \ 414 void helper_##name(CPUPPCState *env, ppc_avr_t *r, \ 415 target_ulong addr) \ 416 { \ 417 size_t n_elems = ARRAY_SIZE(r->element); \ 418 int adjust = HI_IDX * (n_elems - 1); \ 419 int sh = sizeof(r->element[0]) >> 1; \ 420 int index = (addr & 0xf) >> sh; \ 421 if (FIELD_EX64(env->msr, MSR, LE)) { \ 422 index = n_elems - index - 1; \ 423 } \ 424 \ 425 if (needs_byteswap(env)) { \ 426 access(env, addr, swap(r->element[LO_IDX ? index : \ 427 (adjust - index)]), \ 428 GETPC()); \ 429 } else { \ 430 access(env, addr, r->element[LO_IDX ? index : \ 431 (adjust - index)], GETPC()); \ 432 } \ 433 } 434 #define I(x) (x) 435 STVE(stvebx, cpu_stb_data_ra, I, u8) 436 STVE(stvehx, cpu_stw_data_ra, bswap16, u16) 437 STVE(stvewx, cpu_stl_data_ra, bswap32, u32) 438 #undef I 439 #undef LVE 440 441 #ifdef TARGET_PPC64 442 #define GET_NB(rb) ((rb >> 56) & 0xFF) 443 444 #define VSX_LXVL(name, lj) \ 445 void helper_##name(CPUPPCState *env, target_ulong addr, \ 446 ppc_vsr_t *xt, target_ulong rb) \ 447 { \ 448 ppc_vsr_t t; \ 449 uint64_t nb = GET_NB(rb); \ 450 int i; \ 451 \ 452 t.s128 = int128_zero(); \ 453 if (nb) { \ 454 nb = (nb >= 16) ? 16 : nb; \ 455 if (FIELD_EX64(env->msr, MSR, LE) && !lj) { \ 456 for (i = 16; i > 16 - nb; i--) { \ 457 t.VsrB(i - 1) = cpu_ldub_data_ra(env, addr, GETPC()); \ 458 addr = addr_add(env, addr, 1); \ 459 } \ 460 } else { \ 461 for (i = 0; i < nb; i++) { \ 462 t.VsrB(i) = cpu_ldub_data_ra(env, addr, GETPC()); \ 463 addr = addr_add(env, addr, 1); \ 464 } \ 465 } \ 466 } \ 467 *xt = t; \ 468 } 469 470 VSX_LXVL(lxvl, 0) 471 VSX_LXVL(lxvll, 1) 472 #undef VSX_LXVL 473 474 #define VSX_STXVL(name, lj) \ 475 void helper_##name(CPUPPCState *env, target_ulong addr, \ 476 ppc_vsr_t *xt, target_ulong rb) \ 477 { \ 478 target_ulong nb = GET_NB(rb); \ 479 int i; \ 480 \ 481 if (!nb) { \ 482 return; \ 483 } \ 484 \ 485 nb = (nb >= 16) ? 16 : nb; \ 486 if (FIELD_EX64(env->msr, MSR, LE) && !lj) { \ 487 for (i = 16; i > 16 - nb; i--) { \ 488 cpu_stb_data_ra(env, addr, xt->VsrB(i - 1), GETPC()); \ 489 addr = addr_add(env, addr, 1); \ 490 } \ 491 } else { \ 492 for (i = 0; i < nb; i++) { \ 493 cpu_stb_data_ra(env, addr, xt->VsrB(i), GETPC()); \ 494 addr = addr_add(env, addr, 1); \ 495 } \ 496 } \ 497 } 498 499 VSX_STXVL(stxvl, 0) 500 VSX_STXVL(stxvll, 1) 501 #undef VSX_STXVL 502 #undef GET_NB 503 #endif /* TARGET_PPC64 */ 504 505 #undef HI_IDX 506 #undef LO_IDX 507 508 void helper_tbegin(CPUPPCState *env) 509 { 510 /* 511 * As a degenerate implementation, always fail tbegin. The reason 512 * given is "Nesting overflow". The "persistent" bit is set, 513 * providing a hint to the error handler to not retry. The TFIAR 514 * captures the address of the failure, which is this tbegin 515 * instruction. Instruction execution will continue with the next 516 * instruction in memory, which is precisely what we want. 517 */ 518 519 env->spr[SPR_TEXASR] = 520 (1ULL << TEXASR_FAILURE_PERSISTENT) | 521 (1ULL << TEXASR_NESTING_OVERFLOW) | 522 (FIELD_EX64_HV(env->msr) << TEXASR_PRIVILEGE_HV) | 523 (FIELD_EX64(env->msr, MSR, PR) << TEXASR_PRIVILEGE_PR) | 524 (1ULL << TEXASR_FAILURE_SUMMARY) | 525 (1ULL << TEXASR_TFIAR_EXACT); 526 env->spr[SPR_TFIAR] = env->nip | (FIELD_EX64_HV(env->msr) << 1) | 527 FIELD_EX64(env->msr, MSR, PR); 528 env->spr[SPR_TFHAR] = env->nip + 4; 529 env->crf[0] = 0xB; /* 0b1010 = transaction failure */ 530 } 531