1 /* 2 * x86 exception helpers - sysemu code 3 * 4 * Copyright (c) 2003 Fabrice Bellard 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 "tcg/helper-tcg.h" 24 25 typedef struct TranslateParams { 26 target_ulong addr; 27 target_ulong cr3; 28 int pg_mode; 29 int mmu_idx; 30 int ptw_idx; 31 MMUAccessType access_type; 32 } TranslateParams; 33 34 typedef struct TranslateResult { 35 hwaddr paddr; 36 int prot; 37 int page_size; 38 } TranslateResult; 39 40 typedef enum TranslateFaultStage2 { 41 S2_NONE, 42 S2_GPA, 43 S2_GPT, 44 } TranslateFaultStage2; 45 46 typedef struct TranslateFault { 47 int exception_index; 48 int error_code; 49 target_ulong cr2; 50 TranslateFaultStage2 stage2; 51 } TranslateFault; 52 53 typedef struct PTETranslate { 54 CPUX86State *env; 55 TranslateFault *err; 56 int ptw_idx; 57 void *haddr; 58 hwaddr gaddr; 59 } PTETranslate; 60 61 static bool ptw_translate(PTETranslate *inout, hwaddr addr) 62 { 63 CPUTLBEntryFull *full; 64 int flags; 65 66 inout->gaddr = addr; 67 flags = probe_access_full(inout->env, addr, MMU_DATA_STORE, 68 inout->ptw_idx, true, &inout->haddr, &full, 0); 69 70 if (unlikely(flags & TLB_INVALID_MASK)) { 71 TranslateFault *err = inout->err; 72 73 assert(inout->ptw_idx == MMU_NESTED_IDX); 74 err->exception_index = 0; /* unused */ 75 err->error_code = inout->env->error_code; 76 err->cr2 = addr; 77 err->stage2 = S2_GPT; 78 return false; 79 } 80 return true; 81 } 82 83 static inline uint32_t ptw_ldl(const PTETranslate *in) 84 { 85 if (likely(in->haddr)) { 86 return ldl_p(in->haddr); 87 } 88 return cpu_ldl_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); 89 } 90 91 static inline uint64_t ptw_ldq(const PTETranslate *in) 92 { 93 if (likely(in->haddr)) { 94 return ldq_p(in->haddr); 95 } 96 return cpu_ldq_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); 97 } 98 99 /* 100 * Note that we can use a 32-bit cmpxchg for all page table entries, 101 * even 64-bit ones, because PG_PRESENT_MASK, PG_ACCESSED_MASK and 102 * PG_DIRTY_MASK are all in the low 32 bits. 103 */ 104 static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) 105 { 106 uint32_t cmp; 107 108 /* Does x86 really perform a rmw cycle on mmio for ptw? */ 109 start_exclusive(); 110 cmp = cpu_ldl_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); 111 if (cmp == old) { 112 cpu_stl_mmuidx_ra(in->env, in->gaddr, new, in->ptw_idx, 0); 113 } 114 end_exclusive(); 115 return cmp == old; 116 } 117 118 static inline bool ptw_setl(const PTETranslate *in, uint32_t old, uint32_t set) 119 { 120 if (set & ~old) { 121 uint32_t new = old | set; 122 if (likely(in->haddr)) { 123 old = cpu_to_le32(old); 124 new = cpu_to_le32(new); 125 return qatomic_cmpxchg((uint32_t *)in->haddr, old, new) == old; 126 } 127 return ptw_setl_slow(in, old, new); 128 } 129 return true; 130 } 131 132 static bool mmu_translate(CPUX86State *env, const TranslateParams *in, 133 TranslateResult *out, TranslateFault *err) 134 { 135 const int32_t a20_mask = x86_get_a20_mask(env); 136 const target_ulong addr = in->addr; 137 const int pg_mode = in->pg_mode; 138 const bool is_user = (in->mmu_idx == MMU_USER_IDX); 139 const MMUAccessType access_type = in->access_type; 140 uint64_t ptep, pte, rsvd_mask; 141 PTETranslate pte_trans = { 142 .env = env, 143 .err = err, 144 .ptw_idx = in->ptw_idx, 145 }; 146 hwaddr pte_addr, paddr; 147 uint32_t pkr; 148 int page_size; 149 150 restart_all: 151 rsvd_mask = ~MAKE_64BIT_MASK(0, env_archcpu(env)->phys_bits); 152 rsvd_mask &= PG_ADDRESS_MASK; 153 if (!(pg_mode & PG_MODE_NXE)) { 154 rsvd_mask |= PG_NX_MASK; 155 } 156 157 if (pg_mode & PG_MODE_PAE) { 158 #ifdef TARGET_X86_64 159 if (pg_mode & PG_MODE_LMA) { 160 if (pg_mode & PG_MODE_LA57) { 161 /* 162 * Page table level 5 163 */ 164 pte_addr = ((in->cr3 & ~0xfff) + 165 (((addr >> 48) & 0x1ff) << 3)) & a20_mask; 166 if (!ptw_translate(&pte_trans, pte_addr)) { 167 return false; 168 } 169 restart_5: 170 pte = ptw_ldq(&pte_trans); 171 if (!(pte & PG_PRESENT_MASK)) { 172 goto do_fault; 173 } 174 if (pte & (rsvd_mask | PG_PSE_MASK)) { 175 goto do_fault_rsvd; 176 } 177 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 178 goto restart_5; 179 } 180 ptep = pte ^ PG_NX_MASK; 181 } else { 182 pte = in->cr3; 183 ptep = PG_NX_MASK | PG_USER_MASK | PG_RW_MASK; 184 } 185 186 /* 187 * Page table level 4 188 */ 189 pte_addr = ((pte & PG_ADDRESS_MASK) + 190 (((addr >> 39) & 0x1ff) << 3)) & a20_mask; 191 if (!ptw_translate(&pte_trans, pte_addr)) { 192 return false; 193 } 194 restart_4: 195 pte = ptw_ldq(&pte_trans); 196 if (!(pte & PG_PRESENT_MASK)) { 197 goto do_fault; 198 } 199 if (pte & (rsvd_mask | PG_PSE_MASK)) { 200 goto do_fault_rsvd; 201 } 202 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 203 goto restart_4; 204 } 205 ptep &= pte ^ PG_NX_MASK; 206 207 /* 208 * Page table level 3 209 */ 210 pte_addr = ((pte & PG_ADDRESS_MASK) + 211 (((addr >> 30) & 0x1ff) << 3)) & a20_mask; 212 if (!ptw_translate(&pte_trans, pte_addr)) { 213 return false; 214 } 215 restart_3_lma: 216 pte = ptw_ldq(&pte_trans); 217 if (!(pte & PG_PRESENT_MASK)) { 218 goto do_fault; 219 } 220 if (pte & rsvd_mask) { 221 goto do_fault_rsvd; 222 } 223 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 224 goto restart_3_lma; 225 } 226 ptep &= pte ^ PG_NX_MASK; 227 if (pte & PG_PSE_MASK) { 228 /* 1 GB page */ 229 page_size = 1024 * 1024 * 1024; 230 goto do_check_protect; 231 } 232 } else 233 #endif 234 { 235 /* 236 * Page table level 3 237 */ 238 pte_addr = ((in->cr3 & ~0x1f) + ((addr >> 27) & 0x18)) & a20_mask; 239 if (!ptw_translate(&pte_trans, pte_addr)) { 240 return false; 241 } 242 rsvd_mask |= PG_HI_USER_MASK; 243 restart_3_nolma: 244 pte = ptw_ldq(&pte_trans); 245 if (!(pte & PG_PRESENT_MASK)) { 246 goto do_fault; 247 } 248 if (pte & (rsvd_mask | PG_NX_MASK)) { 249 goto do_fault_rsvd; 250 } 251 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 252 goto restart_3_nolma; 253 } 254 ptep = PG_NX_MASK | PG_USER_MASK | PG_RW_MASK; 255 } 256 257 /* 258 * Page table level 2 259 */ 260 pte_addr = ((pte & PG_ADDRESS_MASK) + 261 (((addr >> 21) & 0x1ff) << 3)) & a20_mask; 262 if (!ptw_translate(&pte_trans, pte_addr)) { 263 return false; 264 } 265 restart_2_pae: 266 pte = ptw_ldq(&pte_trans); 267 if (!(pte & PG_PRESENT_MASK)) { 268 goto do_fault; 269 } 270 if (pte & rsvd_mask) { 271 goto do_fault_rsvd; 272 } 273 if (pte & PG_PSE_MASK) { 274 /* 2 MB page */ 275 page_size = 2048 * 1024; 276 ptep &= pte ^ PG_NX_MASK; 277 goto do_check_protect; 278 } 279 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 280 goto restart_2_pae; 281 } 282 ptep &= pte ^ PG_NX_MASK; 283 284 /* 285 * Page table level 1 286 */ 287 pte_addr = ((pte & PG_ADDRESS_MASK) + 288 (((addr >> 12) & 0x1ff) << 3)) & a20_mask; 289 if (!ptw_translate(&pte_trans, pte_addr)) { 290 return false; 291 } 292 pte = ptw_ldq(&pte_trans); 293 if (!(pte & PG_PRESENT_MASK)) { 294 goto do_fault; 295 } 296 if (pte & rsvd_mask) { 297 goto do_fault_rsvd; 298 } 299 /* combine pde and pte nx, user and rw protections */ 300 ptep &= pte ^ PG_NX_MASK; 301 page_size = 4096; 302 } else { 303 /* 304 * Page table level 2 305 */ 306 pte_addr = ((in->cr3 & ~0xfff) + ((addr >> 20) & 0xffc)) & a20_mask; 307 if (!ptw_translate(&pte_trans, pte_addr)) { 308 return false; 309 } 310 restart_2_nopae: 311 pte = ptw_ldl(&pte_trans); 312 if (!(pte & PG_PRESENT_MASK)) { 313 goto do_fault; 314 } 315 ptep = pte | PG_NX_MASK; 316 317 /* if PSE bit is set, then we use a 4MB page */ 318 if ((pte & PG_PSE_MASK) && (pg_mode & PG_MODE_PSE)) { 319 page_size = 4096 * 1024; 320 /* 321 * Bits 20-13 provide bits 39-32 of the address, bit 21 is reserved. 322 * Leave bits 20-13 in place for setting accessed/dirty bits below. 323 */ 324 pte = (uint32_t)pte | ((pte & 0x1fe000LL) << (32 - 13)); 325 rsvd_mask = 0x200000; 326 goto do_check_protect_pse36; 327 } 328 if (!ptw_setl(&pte_trans, pte, PG_ACCESSED_MASK)) { 329 goto restart_2_nopae; 330 } 331 332 /* 333 * Page table level 1 334 */ 335 pte_addr = ((pte & ~0xfffu) + ((addr >> 10) & 0xffc)) & a20_mask; 336 if (!ptw_translate(&pte_trans, pte_addr)) { 337 return false; 338 } 339 pte = ptw_ldl(&pte_trans); 340 if (!(pte & PG_PRESENT_MASK)) { 341 goto do_fault; 342 } 343 /* combine pde and pte user and rw protections */ 344 ptep &= pte | PG_NX_MASK; 345 page_size = 4096; 346 rsvd_mask = 0; 347 } 348 349 do_check_protect: 350 rsvd_mask |= (page_size - 1) & PG_ADDRESS_MASK & ~PG_PSE_PAT_MASK; 351 do_check_protect_pse36: 352 if (pte & rsvd_mask) { 353 goto do_fault_rsvd; 354 } 355 ptep ^= PG_NX_MASK; 356 357 /* can the page can be put in the TLB? prot will tell us */ 358 if (is_user && !(ptep & PG_USER_MASK)) { 359 goto do_fault_protect; 360 } 361 362 int prot = 0; 363 if (in->mmu_idx != MMU_KSMAP_IDX || !(ptep & PG_USER_MASK)) { 364 prot |= PAGE_READ; 365 if ((ptep & PG_RW_MASK) || !(is_user || (pg_mode & PG_MODE_WP))) { 366 prot |= PAGE_WRITE; 367 } 368 } 369 if (!(ptep & PG_NX_MASK) && 370 (is_user || 371 !((pg_mode & PG_MODE_SMEP) && (ptep & PG_USER_MASK)))) { 372 prot |= PAGE_EXEC; 373 } 374 375 if (ptep & PG_USER_MASK) { 376 pkr = pg_mode & PG_MODE_PKE ? env->pkru : 0; 377 } else { 378 pkr = pg_mode & PG_MODE_PKS ? env->pkrs : 0; 379 } 380 if (pkr) { 381 uint32_t pk = (pte & PG_PKRU_MASK) >> PG_PKRU_BIT; 382 uint32_t pkr_ad = (pkr >> pk * 2) & 1; 383 uint32_t pkr_wd = (pkr >> pk * 2) & 2; 384 uint32_t pkr_prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC; 385 386 if (pkr_ad) { 387 pkr_prot &= ~(PAGE_READ | PAGE_WRITE); 388 } else if (pkr_wd && (is_user || (pg_mode & PG_MODE_WP))) { 389 pkr_prot &= ~PAGE_WRITE; 390 } 391 if ((pkr_prot & (1 << access_type)) == 0) { 392 goto do_fault_pk_protect; 393 } 394 prot &= pkr_prot; 395 } 396 397 if ((prot & (1 << access_type)) == 0) { 398 goto do_fault_protect; 399 } 400 401 /* yes, it can! */ 402 { 403 uint32_t set = PG_ACCESSED_MASK; 404 if (access_type == MMU_DATA_STORE) { 405 set |= PG_DIRTY_MASK; 406 } else if (!(pte & PG_DIRTY_MASK)) { 407 /* 408 * Only set write access if already dirty... 409 * otherwise wait for dirty access. 410 */ 411 prot &= ~PAGE_WRITE; 412 } 413 if (!ptw_setl(&pte_trans, pte, set)) { 414 /* 415 * We can arrive here from any of 3 levels and 2 formats. 416 * The only safe thing is to restart the entire lookup. 417 */ 418 goto restart_all; 419 } 420 } 421 422 /* align to page_size */ 423 paddr = (pte & a20_mask & PG_ADDRESS_MASK & ~(page_size - 1)) 424 | (addr & (page_size - 1)); 425 426 if (in->ptw_idx == MMU_NESTED_IDX) { 427 CPUTLBEntryFull *full; 428 int flags, nested_page_size; 429 430 flags = probe_access_full(env, paddr, access_type, 431 MMU_NESTED_IDX, true, 432 &pte_trans.haddr, &full, 0); 433 if (unlikely(flags & TLB_INVALID_MASK)) { 434 err->exception_index = 0; /* unused */ 435 err->error_code = env->error_code; 436 err->cr2 = paddr; 437 err->stage2 = S2_GPA; 438 return false; 439 } 440 441 /* Merge stage1 & stage2 protection bits. */ 442 prot &= full->prot; 443 444 /* Re-verify resulting protection. */ 445 if ((prot & (1 << access_type)) == 0) { 446 goto do_fault_protect; 447 } 448 449 /* Merge stage1 & stage2 addresses to final physical address. */ 450 nested_page_size = 1 << full->lg_page_size; 451 paddr = (full->phys_addr & ~(nested_page_size - 1)) 452 | (paddr & (nested_page_size - 1)); 453 454 /* 455 * Use the larger of stage1 & stage2 page sizes, so that 456 * invalidation works. 457 */ 458 if (nested_page_size > page_size) { 459 page_size = nested_page_size; 460 } 461 } 462 463 out->paddr = paddr; 464 out->prot = prot; 465 out->page_size = page_size; 466 return true; 467 468 int error_code; 469 do_fault_rsvd: 470 error_code = PG_ERROR_RSVD_MASK; 471 goto do_fault_cont; 472 do_fault_protect: 473 error_code = PG_ERROR_P_MASK; 474 goto do_fault_cont; 475 do_fault_pk_protect: 476 assert(access_type != MMU_INST_FETCH); 477 error_code = PG_ERROR_PK_MASK | PG_ERROR_P_MASK; 478 goto do_fault_cont; 479 do_fault: 480 error_code = 0; 481 do_fault_cont: 482 if (is_user) { 483 error_code |= PG_ERROR_U_MASK; 484 } 485 switch (access_type) { 486 case MMU_DATA_LOAD: 487 break; 488 case MMU_DATA_STORE: 489 error_code |= PG_ERROR_W_MASK; 490 break; 491 case MMU_INST_FETCH: 492 if (pg_mode & (PG_MODE_NXE | PG_MODE_SMEP)) { 493 error_code |= PG_ERROR_I_D_MASK; 494 } 495 break; 496 } 497 err->exception_index = EXCP0E_PAGE; 498 err->error_code = error_code; 499 err->cr2 = addr; 500 err->stage2 = S2_NONE; 501 return false; 502 } 503 504 static G_NORETURN void raise_stage2(CPUX86State *env, TranslateFault *err, 505 uintptr_t retaddr) 506 { 507 uint64_t exit_info_1 = err->error_code; 508 509 switch (err->stage2) { 510 case S2_GPT: 511 exit_info_1 |= SVM_NPTEXIT_GPT; 512 break; 513 case S2_GPA: 514 exit_info_1 |= SVM_NPTEXIT_GPA; 515 break; 516 default: 517 g_assert_not_reached(); 518 } 519 520 x86_stq_phys(env_cpu(env), 521 env->vm_vmcb + offsetof(struct vmcb, control.exit_info_2), 522 err->cr2); 523 cpu_vmexit(env, SVM_EXIT_NPF, exit_info_1, retaddr); 524 } 525 526 static bool get_physical_address(CPUX86State *env, vaddr addr, 527 MMUAccessType access_type, int mmu_idx, 528 TranslateResult *out, TranslateFault *err) 529 { 530 TranslateParams in; 531 bool use_stage2 = env->hflags2 & HF2_NPT_MASK; 532 533 in.addr = addr; 534 in.access_type = access_type; 535 536 switch (mmu_idx) { 537 case MMU_PHYS_IDX: 538 break; 539 540 case MMU_NESTED_IDX: 541 if (likely(use_stage2)) { 542 in.cr3 = env->nested_cr3; 543 in.pg_mode = env->nested_pg_mode; 544 in.mmu_idx = MMU_USER_IDX; 545 in.ptw_idx = MMU_PHYS_IDX; 546 547 if (!mmu_translate(env, &in, out, err)) { 548 err->stage2 = S2_GPA; 549 return false; 550 } 551 return true; 552 } 553 break; 554 555 default: 556 in.cr3 = env->cr[3]; 557 in.mmu_idx = mmu_idx; 558 in.ptw_idx = use_stage2 ? MMU_NESTED_IDX : MMU_PHYS_IDX; 559 in.pg_mode = get_pg_mode(env); 560 561 if (likely(in.pg_mode)) { 562 if (in.pg_mode & PG_MODE_LMA) { 563 /* test virtual address sign extension */ 564 int shift = in.pg_mode & PG_MODE_LA57 ? 56 : 47; 565 int64_t sext = (int64_t)addr >> shift; 566 if (sext != 0 && sext != -1) { 567 err->exception_index = EXCP0D_GPF; 568 err->error_code = 0; 569 err->cr2 = addr; 570 return false; 571 } 572 } 573 return mmu_translate(env, &in, out, err); 574 } 575 break; 576 } 577 578 /* Translation disabled. */ 579 out->paddr = addr & x86_get_a20_mask(env); 580 #ifdef TARGET_X86_64 581 if (!(env->hflags & HF_LMA_MASK)) { 582 /* Without long mode we can only address 32bits in real mode */ 583 out->paddr = (uint32_t)out->paddr; 584 } 585 #endif 586 out->prot = PAGE_READ | PAGE_WRITE | PAGE_EXEC; 587 out->page_size = TARGET_PAGE_SIZE; 588 return true; 589 } 590 591 bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size, 592 MMUAccessType access_type, int mmu_idx, 593 bool probe, uintptr_t retaddr) 594 { 595 CPUX86State *env = cs->env_ptr; 596 TranslateResult out; 597 TranslateFault err; 598 599 if (get_physical_address(env, addr, access_type, mmu_idx, &out, &err)) { 600 /* 601 * Even if 4MB pages, we map only one 4KB page in the cache to 602 * avoid filling it too fast. 603 */ 604 assert(out.prot & (1 << access_type)); 605 tlb_set_page_with_attrs(cs, addr & TARGET_PAGE_MASK, 606 out.paddr & TARGET_PAGE_MASK, 607 cpu_get_mem_attrs(env), 608 out.prot, mmu_idx, out.page_size); 609 return true; 610 } 611 612 if (probe) { 613 /* This will be used if recursing for stage2 translation. */ 614 env->error_code = err.error_code; 615 return false; 616 } 617 618 if (err.stage2 != S2_NONE) { 619 raise_stage2(env, &err, retaddr); 620 } 621 622 if (env->intercept_exceptions & (1 << err.exception_index)) { 623 /* cr2 is not modified in case of exceptions */ 624 x86_stq_phys(cs, env->vm_vmcb + 625 offsetof(struct vmcb, control.exit_info_2), 626 err.cr2); 627 } else { 628 env->cr[2] = err.cr2; 629 } 630 raise_exception_err_ra(env, err.exception_index, err.error_code, retaddr); 631 } 632 633 G_NORETURN void x86_cpu_do_unaligned_access(CPUState *cs, vaddr vaddr, 634 MMUAccessType access_type, 635 int mmu_idx, uintptr_t retaddr) 636 { 637 X86CPU *cpu = X86_CPU(cs); 638 handle_unaligned_access(&cpu->env, vaddr, access_type, retaddr); 639 } 640