1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PGTABLE_H 3 #define _ASM_X86_PGTABLE_H 4 5 #include <linux/mem_encrypt.h> 6 #include <asm/page.h> 7 #include <asm/pgtable_types.h> 8 9 /* 10 * Macro to mark a page protection value as UC- 11 */ 12 #define pgprot_noncached(prot) \ 13 ((boot_cpu_data.x86 > 3) \ 14 ? (__pgprot(pgprot_val(prot) | \ 15 cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS))) \ 16 : (prot)) 17 18 /* 19 * Macros to add or remove encryption attribute 20 */ 21 #define pgprot_encrypted(prot) __pgprot(__sme_set(pgprot_val(prot))) 22 #define pgprot_decrypted(prot) __pgprot(__sme_clr(pgprot_val(prot))) 23 24 #ifndef __ASSEMBLY__ 25 #include <asm/x86_init.h> 26 27 extern pgd_t early_top_pgt[PTRS_PER_PGD]; 28 int __init __early_make_pgtable(unsigned long address, pmdval_t pmd); 29 30 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd); 31 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, pgd_t *pgd, bool user); 32 void ptdump_walk_pgd_level_checkwx(void); 33 34 #ifdef CONFIG_DEBUG_WX 35 #define debug_checkwx() ptdump_walk_pgd_level_checkwx() 36 #else 37 #define debug_checkwx() do { } while (0) 38 #endif 39 40 /* 41 * ZERO_PAGE is a global shared page that is always zero: used 42 * for zero-mapped memory areas etc.. 43 */ 44 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 45 __visible; 46 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page)) 47 48 extern spinlock_t pgd_lock; 49 extern struct list_head pgd_list; 50 51 extern struct mm_struct *pgd_page_get_mm(struct page *page); 52 53 extern pmdval_t early_pmd_flags; 54 55 #ifdef CONFIG_PARAVIRT 56 #include <asm/paravirt.h> 57 #else /* !CONFIG_PARAVIRT */ 58 #define set_pte(ptep, pte) native_set_pte(ptep, pte) 59 #define set_pte_at(mm, addr, ptep, pte) native_set_pte_at(mm, addr, ptep, pte) 60 61 #define set_pte_atomic(ptep, pte) \ 62 native_set_pte_atomic(ptep, pte) 63 64 #define set_pmd(pmdp, pmd) native_set_pmd(pmdp, pmd) 65 66 #ifndef __PAGETABLE_P4D_FOLDED 67 #define set_pgd(pgdp, pgd) native_set_pgd(pgdp, pgd) 68 #define pgd_clear(pgd) (pgtable_l5_enabled ? native_pgd_clear(pgd) : 0) 69 #endif 70 71 #ifndef set_p4d 72 # define set_p4d(p4dp, p4d) native_set_p4d(p4dp, p4d) 73 #endif 74 75 #ifndef __PAGETABLE_PUD_FOLDED 76 #define p4d_clear(p4d) native_p4d_clear(p4d) 77 #endif 78 79 #ifndef set_pud 80 # define set_pud(pudp, pud) native_set_pud(pudp, pud) 81 #endif 82 83 #ifndef __PAGETABLE_PUD_FOLDED 84 #define pud_clear(pud) native_pud_clear(pud) 85 #endif 86 87 #define pte_clear(mm, addr, ptep) native_pte_clear(mm, addr, ptep) 88 #define pmd_clear(pmd) native_pmd_clear(pmd) 89 90 #define pgd_val(x) native_pgd_val(x) 91 #define __pgd(x) native_make_pgd(x) 92 93 #ifndef __PAGETABLE_P4D_FOLDED 94 #define p4d_val(x) native_p4d_val(x) 95 #define __p4d(x) native_make_p4d(x) 96 #endif 97 98 #ifndef __PAGETABLE_PUD_FOLDED 99 #define pud_val(x) native_pud_val(x) 100 #define __pud(x) native_make_pud(x) 101 #endif 102 103 #ifndef __PAGETABLE_PMD_FOLDED 104 #define pmd_val(x) native_pmd_val(x) 105 #define __pmd(x) native_make_pmd(x) 106 #endif 107 108 #define pte_val(x) native_pte_val(x) 109 #define __pte(x) native_make_pte(x) 110 111 #define arch_end_context_switch(prev) do {} while(0) 112 113 #endif /* CONFIG_PARAVIRT */ 114 115 /* 116 * The following only work if pte_present() is true. 117 * Undefined behaviour if not.. 118 */ 119 static inline int pte_dirty(pte_t pte) 120 { 121 return pte_flags(pte) & _PAGE_DIRTY; 122 } 123 124 125 static inline u32 read_pkru(void) 126 { 127 if (boot_cpu_has(X86_FEATURE_OSPKE)) 128 return __read_pkru(); 129 return 0; 130 } 131 132 static inline void write_pkru(u32 pkru) 133 { 134 if (boot_cpu_has(X86_FEATURE_OSPKE)) 135 __write_pkru(pkru); 136 } 137 138 static inline int pte_young(pte_t pte) 139 { 140 return pte_flags(pte) & _PAGE_ACCESSED; 141 } 142 143 static inline int pmd_dirty(pmd_t pmd) 144 { 145 return pmd_flags(pmd) & _PAGE_DIRTY; 146 } 147 148 static inline int pmd_young(pmd_t pmd) 149 { 150 return pmd_flags(pmd) & _PAGE_ACCESSED; 151 } 152 153 static inline int pud_dirty(pud_t pud) 154 { 155 return pud_flags(pud) & _PAGE_DIRTY; 156 } 157 158 static inline int pud_young(pud_t pud) 159 { 160 return pud_flags(pud) & _PAGE_ACCESSED; 161 } 162 163 static inline int pte_write(pte_t pte) 164 { 165 return pte_flags(pte) & _PAGE_RW; 166 } 167 168 static inline int pte_huge(pte_t pte) 169 { 170 return pte_flags(pte) & _PAGE_PSE; 171 } 172 173 static inline int pte_global(pte_t pte) 174 { 175 return pte_flags(pte) & _PAGE_GLOBAL; 176 } 177 178 static inline int pte_exec(pte_t pte) 179 { 180 return !(pte_flags(pte) & _PAGE_NX); 181 } 182 183 static inline int pte_special(pte_t pte) 184 { 185 return pte_flags(pte) & _PAGE_SPECIAL; 186 } 187 188 static inline unsigned long pte_pfn(pte_t pte) 189 { 190 return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT; 191 } 192 193 static inline unsigned long pmd_pfn(pmd_t pmd) 194 { 195 return (pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT; 196 } 197 198 static inline unsigned long pud_pfn(pud_t pud) 199 { 200 return (pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT; 201 } 202 203 static inline unsigned long p4d_pfn(p4d_t p4d) 204 { 205 return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT; 206 } 207 208 static inline unsigned long pgd_pfn(pgd_t pgd) 209 { 210 return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT; 211 } 212 213 static inline int p4d_large(p4d_t p4d) 214 { 215 /* No 512 GiB pages yet */ 216 return 0; 217 } 218 219 #define pte_page(pte) pfn_to_page(pte_pfn(pte)) 220 221 static inline int pmd_large(pmd_t pte) 222 { 223 return pmd_flags(pte) & _PAGE_PSE; 224 } 225 226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 227 static inline int pmd_trans_huge(pmd_t pmd) 228 { 229 return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE; 230 } 231 232 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 233 static inline int pud_trans_huge(pud_t pud) 234 { 235 return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE; 236 } 237 #endif 238 239 #define has_transparent_hugepage has_transparent_hugepage 240 static inline int has_transparent_hugepage(void) 241 { 242 return boot_cpu_has(X86_FEATURE_PSE); 243 } 244 245 #ifdef __HAVE_ARCH_PTE_DEVMAP 246 static inline int pmd_devmap(pmd_t pmd) 247 { 248 return !!(pmd_val(pmd) & _PAGE_DEVMAP); 249 } 250 251 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 252 static inline int pud_devmap(pud_t pud) 253 { 254 return !!(pud_val(pud) & _PAGE_DEVMAP); 255 } 256 #else 257 static inline int pud_devmap(pud_t pud) 258 { 259 return 0; 260 } 261 #endif 262 263 static inline int pgd_devmap(pgd_t pgd) 264 { 265 return 0; 266 } 267 #endif 268 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 269 270 static inline pte_t pte_set_flags(pte_t pte, pteval_t set) 271 { 272 pteval_t v = native_pte_val(pte); 273 274 return native_make_pte(v | set); 275 } 276 277 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear) 278 { 279 pteval_t v = native_pte_val(pte); 280 281 return native_make_pte(v & ~clear); 282 } 283 284 static inline pte_t pte_mkclean(pte_t pte) 285 { 286 return pte_clear_flags(pte, _PAGE_DIRTY); 287 } 288 289 static inline pte_t pte_mkold(pte_t pte) 290 { 291 return pte_clear_flags(pte, _PAGE_ACCESSED); 292 } 293 294 static inline pte_t pte_wrprotect(pte_t pte) 295 { 296 return pte_clear_flags(pte, _PAGE_RW); 297 } 298 299 static inline pte_t pte_mkexec(pte_t pte) 300 { 301 return pte_clear_flags(pte, _PAGE_NX); 302 } 303 304 static inline pte_t pte_mkdirty(pte_t pte) 305 { 306 return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY); 307 } 308 309 static inline pte_t pte_mkyoung(pte_t pte) 310 { 311 return pte_set_flags(pte, _PAGE_ACCESSED); 312 } 313 314 static inline pte_t pte_mkwrite(pte_t pte) 315 { 316 return pte_set_flags(pte, _PAGE_RW); 317 } 318 319 static inline pte_t pte_mkhuge(pte_t pte) 320 { 321 return pte_set_flags(pte, _PAGE_PSE); 322 } 323 324 static inline pte_t pte_clrhuge(pte_t pte) 325 { 326 return pte_clear_flags(pte, _PAGE_PSE); 327 } 328 329 static inline pte_t pte_mkglobal(pte_t pte) 330 { 331 return pte_set_flags(pte, _PAGE_GLOBAL); 332 } 333 334 static inline pte_t pte_clrglobal(pte_t pte) 335 { 336 return pte_clear_flags(pte, _PAGE_GLOBAL); 337 } 338 339 static inline pte_t pte_mkspecial(pte_t pte) 340 { 341 return pte_set_flags(pte, _PAGE_SPECIAL); 342 } 343 344 static inline pte_t pte_mkdevmap(pte_t pte) 345 { 346 return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP); 347 } 348 349 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set) 350 { 351 pmdval_t v = native_pmd_val(pmd); 352 353 return native_make_pmd(v | set); 354 } 355 356 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear) 357 { 358 pmdval_t v = native_pmd_val(pmd); 359 360 return native_make_pmd(v & ~clear); 361 } 362 363 static inline pmd_t pmd_mkold(pmd_t pmd) 364 { 365 return pmd_clear_flags(pmd, _PAGE_ACCESSED); 366 } 367 368 static inline pmd_t pmd_mkclean(pmd_t pmd) 369 { 370 return pmd_clear_flags(pmd, _PAGE_DIRTY); 371 } 372 373 static inline pmd_t pmd_wrprotect(pmd_t pmd) 374 { 375 return pmd_clear_flags(pmd, _PAGE_RW); 376 } 377 378 static inline pmd_t pmd_mkdirty(pmd_t pmd) 379 { 380 return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY); 381 } 382 383 static inline pmd_t pmd_mkdevmap(pmd_t pmd) 384 { 385 return pmd_set_flags(pmd, _PAGE_DEVMAP); 386 } 387 388 static inline pmd_t pmd_mkhuge(pmd_t pmd) 389 { 390 return pmd_set_flags(pmd, _PAGE_PSE); 391 } 392 393 static inline pmd_t pmd_mkyoung(pmd_t pmd) 394 { 395 return pmd_set_flags(pmd, _PAGE_ACCESSED); 396 } 397 398 static inline pmd_t pmd_mkwrite(pmd_t pmd) 399 { 400 return pmd_set_flags(pmd, _PAGE_RW); 401 } 402 403 static inline pmd_t pmd_mknotpresent(pmd_t pmd) 404 { 405 return pmd_clear_flags(pmd, _PAGE_PRESENT | _PAGE_PROTNONE); 406 } 407 408 static inline pud_t pud_set_flags(pud_t pud, pudval_t set) 409 { 410 pudval_t v = native_pud_val(pud); 411 412 return native_make_pud(v | set); 413 } 414 415 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear) 416 { 417 pudval_t v = native_pud_val(pud); 418 419 return native_make_pud(v & ~clear); 420 } 421 422 static inline pud_t pud_mkold(pud_t pud) 423 { 424 return pud_clear_flags(pud, _PAGE_ACCESSED); 425 } 426 427 static inline pud_t pud_mkclean(pud_t pud) 428 { 429 return pud_clear_flags(pud, _PAGE_DIRTY); 430 } 431 432 static inline pud_t pud_wrprotect(pud_t pud) 433 { 434 return pud_clear_flags(pud, _PAGE_RW); 435 } 436 437 static inline pud_t pud_mkdirty(pud_t pud) 438 { 439 return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY); 440 } 441 442 static inline pud_t pud_mkdevmap(pud_t pud) 443 { 444 return pud_set_flags(pud, _PAGE_DEVMAP); 445 } 446 447 static inline pud_t pud_mkhuge(pud_t pud) 448 { 449 return pud_set_flags(pud, _PAGE_PSE); 450 } 451 452 static inline pud_t pud_mkyoung(pud_t pud) 453 { 454 return pud_set_flags(pud, _PAGE_ACCESSED); 455 } 456 457 static inline pud_t pud_mkwrite(pud_t pud) 458 { 459 return pud_set_flags(pud, _PAGE_RW); 460 } 461 462 static inline pud_t pud_mknotpresent(pud_t pud) 463 { 464 return pud_clear_flags(pud, _PAGE_PRESENT | _PAGE_PROTNONE); 465 } 466 467 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY 468 static inline int pte_soft_dirty(pte_t pte) 469 { 470 return pte_flags(pte) & _PAGE_SOFT_DIRTY; 471 } 472 473 static inline int pmd_soft_dirty(pmd_t pmd) 474 { 475 return pmd_flags(pmd) & _PAGE_SOFT_DIRTY; 476 } 477 478 static inline int pud_soft_dirty(pud_t pud) 479 { 480 return pud_flags(pud) & _PAGE_SOFT_DIRTY; 481 } 482 483 static inline pte_t pte_mksoft_dirty(pte_t pte) 484 { 485 return pte_set_flags(pte, _PAGE_SOFT_DIRTY); 486 } 487 488 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd) 489 { 490 return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY); 491 } 492 493 static inline pud_t pud_mksoft_dirty(pud_t pud) 494 { 495 return pud_set_flags(pud, _PAGE_SOFT_DIRTY); 496 } 497 498 static inline pte_t pte_clear_soft_dirty(pte_t pte) 499 { 500 return pte_clear_flags(pte, _PAGE_SOFT_DIRTY); 501 } 502 503 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd) 504 { 505 return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY); 506 } 507 508 static inline pud_t pud_clear_soft_dirty(pud_t pud) 509 { 510 return pud_clear_flags(pud, _PAGE_SOFT_DIRTY); 511 } 512 513 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */ 514 515 /* 516 * Mask out unsupported bits in a present pgprot. Non-present pgprots 517 * can use those bits for other purposes, so leave them be. 518 */ 519 static inline pgprotval_t massage_pgprot(pgprot_t pgprot) 520 { 521 pgprotval_t protval = pgprot_val(pgprot); 522 523 if (protval & _PAGE_PRESENT) 524 protval &= __supported_pte_mask; 525 526 return protval; 527 } 528 529 static inline pgprotval_t check_pgprot(pgprot_t pgprot) 530 { 531 pgprotval_t massaged_val = massage_pgprot(pgprot); 532 533 /* mmdebug.h can not be included here because of dependencies */ 534 #ifdef CONFIG_DEBUG_VM 535 WARN_ONCE(pgprot_val(pgprot) != massaged_val, 536 "attempted to set unsupported pgprot: %016llx " 537 "bits: %016llx supported: %016llx\n", 538 (u64)pgprot_val(pgprot), 539 (u64)pgprot_val(pgprot) ^ massaged_val, 540 (u64)__supported_pte_mask); 541 #endif 542 543 return massaged_val; 544 } 545 546 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) 547 { 548 return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) | 549 check_pgprot(pgprot)); 550 } 551 552 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot) 553 { 554 return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) | 555 check_pgprot(pgprot)); 556 } 557 558 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot) 559 { 560 return __pud(((phys_addr_t)page_nr << PAGE_SHIFT) | 561 check_pgprot(pgprot)); 562 } 563 564 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) 565 { 566 pteval_t val = pte_val(pte); 567 568 /* 569 * Chop off the NX bit (if present), and add the NX portion of 570 * the newprot (if present): 571 */ 572 val &= _PAGE_CHG_MASK; 573 val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK; 574 575 return __pte(val); 576 } 577 578 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot) 579 { 580 pmdval_t val = pmd_val(pmd); 581 582 val &= _HPAGE_CHG_MASK; 583 val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK; 584 585 return __pmd(val); 586 } 587 588 /* mprotect needs to preserve PAT bits when updating vm_page_prot */ 589 #define pgprot_modify pgprot_modify 590 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot) 591 { 592 pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK; 593 pgprotval_t addbits = pgprot_val(newprot); 594 return __pgprot(preservebits | addbits); 595 } 596 597 #define pte_pgprot(x) __pgprot(pte_flags(x)) 598 #define pmd_pgprot(x) __pgprot(pmd_flags(x)) 599 #define pud_pgprot(x) __pgprot(pud_flags(x)) 600 #define p4d_pgprot(x) __pgprot(p4d_flags(x)) 601 602 #define canon_pgprot(p) __pgprot(massage_pgprot(p)) 603 604 static inline pgprot_t arch_filter_pgprot(pgprot_t prot) 605 { 606 return canon_pgprot(prot); 607 } 608 609 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size, 610 enum page_cache_mode pcm, 611 enum page_cache_mode new_pcm) 612 { 613 /* 614 * PAT type is always WB for untracked ranges, so no need to check. 615 */ 616 if (x86_platform.is_untracked_pat_range(paddr, paddr + size)) 617 return 1; 618 619 /* 620 * Certain new memtypes are not allowed with certain 621 * requested memtype: 622 * - request is uncached, return cannot be write-back 623 * - request is write-combine, return cannot be write-back 624 * - request is write-through, return cannot be write-back 625 * - request is write-through, return cannot be write-combine 626 */ 627 if ((pcm == _PAGE_CACHE_MODE_UC_MINUS && 628 new_pcm == _PAGE_CACHE_MODE_WB) || 629 (pcm == _PAGE_CACHE_MODE_WC && 630 new_pcm == _PAGE_CACHE_MODE_WB) || 631 (pcm == _PAGE_CACHE_MODE_WT && 632 new_pcm == _PAGE_CACHE_MODE_WB) || 633 (pcm == _PAGE_CACHE_MODE_WT && 634 new_pcm == _PAGE_CACHE_MODE_WC)) { 635 return 0; 636 } 637 638 return 1; 639 } 640 641 pmd_t *populate_extra_pmd(unsigned long vaddr); 642 pte_t *populate_extra_pte(unsigned long vaddr); 643 #endif /* __ASSEMBLY__ */ 644 645 #ifdef CONFIG_X86_32 646 # include <asm/pgtable_32.h> 647 #else 648 # include <asm/pgtable_64.h> 649 #endif 650 651 #ifndef __ASSEMBLY__ 652 #include <linux/mm_types.h> 653 #include <linux/mmdebug.h> 654 #include <linux/log2.h> 655 #include <asm/fixmap.h> 656 657 static inline int pte_none(pte_t pte) 658 { 659 return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK)); 660 } 661 662 #define __HAVE_ARCH_PTE_SAME 663 static inline int pte_same(pte_t a, pte_t b) 664 { 665 return a.pte == b.pte; 666 } 667 668 static inline int pte_present(pte_t a) 669 { 670 return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE); 671 } 672 673 #ifdef __HAVE_ARCH_PTE_DEVMAP 674 static inline int pte_devmap(pte_t a) 675 { 676 return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP; 677 } 678 #endif 679 680 #define pte_accessible pte_accessible 681 static inline bool pte_accessible(struct mm_struct *mm, pte_t a) 682 { 683 if (pte_flags(a) & _PAGE_PRESENT) 684 return true; 685 686 if ((pte_flags(a) & _PAGE_PROTNONE) && 687 mm_tlb_flush_pending(mm)) 688 return true; 689 690 return false; 691 } 692 693 static inline int pmd_present(pmd_t pmd) 694 { 695 /* 696 * Checking for _PAGE_PSE is needed too because 697 * split_huge_page will temporarily clear the present bit (but 698 * the _PAGE_PSE flag will remain set at all times while the 699 * _PAGE_PRESENT bit is clear). 700 */ 701 return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE); 702 } 703 704 #ifdef CONFIG_NUMA_BALANCING 705 /* 706 * These work without NUMA balancing but the kernel does not care. See the 707 * comment in include/asm-generic/pgtable.h 708 */ 709 static inline int pte_protnone(pte_t pte) 710 { 711 return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT)) 712 == _PAGE_PROTNONE; 713 } 714 715 static inline int pmd_protnone(pmd_t pmd) 716 { 717 return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT)) 718 == _PAGE_PROTNONE; 719 } 720 #endif /* CONFIG_NUMA_BALANCING */ 721 722 static inline int pmd_none(pmd_t pmd) 723 { 724 /* Only check low word on 32-bit platforms, since it might be 725 out of sync with upper half. */ 726 unsigned long val = native_pmd_val(pmd); 727 return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0; 728 } 729 730 static inline unsigned long pmd_page_vaddr(pmd_t pmd) 731 { 732 return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd)); 733 } 734 735 /* 736 * Currently stuck as a macro due to indirect forward reference to 737 * linux/mmzone.h's __section_mem_map_addr() definition: 738 */ 739 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd)) 740 741 /* 742 * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD] 743 * 744 * this macro returns the index of the entry in the pmd page which would 745 * control the given virtual address 746 */ 747 static inline unsigned long pmd_index(unsigned long address) 748 { 749 return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1); 750 } 751 752 /* 753 * Conversion functions: convert a page and protection to a page entry, 754 * and a page entry and page directory to the page they refer to. 755 * 756 * (Currently stuck as a macro because of indirect forward reference 757 * to linux/mm.h:page_to_nid()) 758 */ 759 #define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot)) 760 761 /* 762 * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE] 763 * 764 * this function returns the index of the entry in the pte page which would 765 * control the given virtual address 766 */ 767 static inline unsigned long pte_index(unsigned long address) 768 { 769 return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 770 } 771 772 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address) 773 { 774 return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address); 775 } 776 777 static inline int pmd_bad(pmd_t pmd) 778 { 779 return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE; 780 } 781 782 static inline unsigned long pages_to_mb(unsigned long npg) 783 { 784 return npg >> (20 - PAGE_SHIFT); 785 } 786 787 #if CONFIG_PGTABLE_LEVELS > 2 788 static inline int pud_none(pud_t pud) 789 { 790 return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0; 791 } 792 793 static inline int pud_present(pud_t pud) 794 { 795 return pud_flags(pud) & _PAGE_PRESENT; 796 } 797 798 static inline unsigned long pud_page_vaddr(pud_t pud) 799 { 800 return (unsigned long)__va(pud_val(pud) & pud_pfn_mask(pud)); 801 } 802 803 /* 804 * Currently stuck as a macro due to indirect forward reference to 805 * linux/mmzone.h's __section_mem_map_addr() definition: 806 */ 807 #define pud_page(pud) pfn_to_page(pud_pfn(pud)) 808 809 /* Find an entry in the second-level page table.. */ 810 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address) 811 { 812 return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address); 813 } 814 815 static inline int pud_large(pud_t pud) 816 { 817 return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) == 818 (_PAGE_PSE | _PAGE_PRESENT); 819 } 820 821 static inline int pud_bad(pud_t pud) 822 { 823 return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0; 824 } 825 #else 826 static inline int pud_large(pud_t pud) 827 { 828 return 0; 829 } 830 #endif /* CONFIG_PGTABLE_LEVELS > 2 */ 831 832 static inline unsigned long pud_index(unsigned long address) 833 { 834 return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1); 835 } 836 837 #if CONFIG_PGTABLE_LEVELS > 3 838 static inline int p4d_none(p4d_t p4d) 839 { 840 return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0; 841 } 842 843 static inline int p4d_present(p4d_t p4d) 844 { 845 return p4d_flags(p4d) & _PAGE_PRESENT; 846 } 847 848 static inline unsigned long p4d_page_vaddr(p4d_t p4d) 849 { 850 return (unsigned long)__va(p4d_val(p4d) & p4d_pfn_mask(p4d)); 851 } 852 853 /* 854 * Currently stuck as a macro due to indirect forward reference to 855 * linux/mmzone.h's __section_mem_map_addr() definition: 856 */ 857 #define p4d_page(p4d) pfn_to_page(p4d_pfn(p4d)) 858 859 /* Find an entry in the third-level page table.. */ 860 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address) 861 { 862 return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address); 863 } 864 865 static inline int p4d_bad(p4d_t p4d) 866 { 867 unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER; 868 869 if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION)) 870 ignore_flags |= _PAGE_NX; 871 872 return (p4d_flags(p4d) & ~ignore_flags) != 0; 873 } 874 #endif /* CONFIG_PGTABLE_LEVELS > 3 */ 875 876 static inline unsigned long p4d_index(unsigned long address) 877 { 878 return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1); 879 } 880 881 #if CONFIG_PGTABLE_LEVELS > 4 882 static inline int pgd_present(pgd_t pgd) 883 { 884 if (!pgtable_l5_enabled) 885 return 1; 886 return pgd_flags(pgd) & _PAGE_PRESENT; 887 } 888 889 static inline unsigned long pgd_page_vaddr(pgd_t pgd) 890 { 891 return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK); 892 } 893 894 /* 895 * Currently stuck as a macro due to indirect forward reference to 896 * linux/mmzone.h's __section_mem_map_addr() definition: 897 */ 898 #define pgd_page(pgd) pfn_to_page(pgd_pfn(pgd)) 899 900 /* to find an entry in a page-table-directory. */ 901 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address) 902 { 903 if (!pgtable_l5_enabled) 904 return (p4d_t *)pgd; 905 return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address); 906 } 907 908 static inline int pgd_bad(pgd_t pgd) 909 { 910 unsigned long ignore_flags = _PAGE_USER; 911 912 if (!pgtable_l5_enabled) 913 return 0; 914 915 if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION)) 916 ignore_flags |= _PAGE_NX; 917 918 return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE; 919 } 920 921 static inline int pgd_none(pgd_t pgd) 922 { 923 if (!pgtable_l5_enabled) 924 return 0; 925 /* 926 * There is no need to do a workaround for the KNL stray 927 * A/D bit erratum here. PGDs only point to page tables 928 * except on 32-bit non-PAE which is not supported on 929 * KNL. 930 */ 931 return !native_pgd_val(pgd); 932 } 933 #endif /* CONFIG_PGTABLE_LEVELS > 4 */ 934 935 #endif /* __ASSEMBLY__ */ 936 937 /* 938 * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD] 939 * 940 * this macro returns the index of the entry in the pgd page which would 941 * control the given virtual address 942 */ 943 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)) 944 945 /* 946 * pgd_offset() returns a (pgd_t *) 947 * pgd_index() is used get the offset into the pgd page's array of pgd_t's; 948 */ 949 #define pgd_offset_pgd(pgd, address) (pgd + pgd_index((address))) 950 /* 951 * a shortcut to get a pgd_t in a given mm 952 */ 953 #define pgd_offset(mm, address) pgd_offset_pgd((mm)->pgd, (address)) 954 /* 955 * a shortcut which implies the use of the kernel's pgd, instead 956 * of a process's 957 */ 958 #define pgd_offset_k(address) pgd_offset(&init_mm, (address)) 959 960 961 #define KERNEL_PGD_BOUNDARY pgd_index(PAGE_OFFSET) 962 #define KERNEL_PGD_PTRS (PTRS_PER_PGD - KERNEL_PGD_BOUNDARY) 963 964 #ifndef __ASSEMBLY__ 965 966 extern int direct_gbpages; 967 void init_mem_mapping(void); 968 void early_alloc_pgt_buf(void); 969 extern void memblock_find_dma_reserve(void); 970 971 #ifdef CONFIG_X86_64 972 /* Realmode trampoline initialization. */ 973 extern pgd_t trampoline_pgd_entry; 974 static inline void __meminit init_trampoline_default(void) 975 { 976 /* Default trampoline pgd value */ 977 trampoline_pgd_entry = init_top_pgt[pgd_index(__PAGE_OFFSET)]; 978 } 979 # ifdef CONFIG_RANDOMIZE_MEMORY 980 void __meminit init_trampoline(void); 981 # else 982 # define init_trampoline init_trampoline_default 983 # endif 984 #else 985 static inline void init_trampoline(void) { } 986 #endif 987 988 /* local pte updates need not use xchg for locking */ 989 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep) 990 { 991 pte_t res = *ptep; 992 993 /* Pure native function needs no input for mm, addr */ 994 native_pte_clear(NULL, 0, ptep); 995 return res; 996 } 997 998 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp) 999 { 1000 pmd_t res = *pmdp; 1001 1002 native_pmd_clear(pmdp); 1003 return res; 1004 } 1005 1006 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp) 1007 { 1008 pud_t res = *pudp; 1009 1010 native_pud_clear(pudp); 1011 return res; 1012 } 1013 1014 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr, 1015 pte_t *ptep , pte_t pte) 1016 { 1017 native_set_pte(ptep, pte); 1018 } 1019 1020 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr, 1021 pmd_t *pmdp, pmd_t pmd) 1022 { 1023 native_set_pmd(pmdp, pmd); 1024 } 1025 1026 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr, 1027 pud_t *pudp, pud_t pud) 1028 { 1029 native_set_pud(pudp, pud); 1030 } 1031 1032 /* 1033 * We only update the dirty/accessed state if we set 1034 * the dirty bit by hand in the kernel, since the hardware 1035 * will do the accessed bit for us, and we don't want to 1036 * race with other CPU's that might be updating the dirty 1037 * bit at the same time. 1038 */ 1039 struct vm_area_struct; 1040 1041 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS 1042 extern int ptep_set_access_flags(struct vm_area_struct *vma, 1043 unsigned long address, pte_t *ptep, 1044 pte_t entry, int dirty); 1045 1046 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG 1047 extern int ptep_test_and_clear_young(struct vm_area_struct *vma, 1048 unsigned long addr, pte_t *ptep); 1049 1050 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH 1051 extern int ptep_clear_flush_young(struct vm_area_struct *vma, 1052 unsigned long address, pte_t *ptep); 1053 1054 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR 1055 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, 1056 pte_t *ptep) 1057 { 1058 pte_t pte = native_ptep_get_and_clear(ptep); 1059 return pte; 1060 } 1061 1062 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL 1063 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, 1064 unsigned long addr, pte_t *ptep, 1065 int full) 1066 { 1067 pte_t pte; 1068 if (full) { 1069 /* 1070 * Full address destruction in progress; paravirt does not 1071 * care about updates and native needs no locking 1072 */ 1073 pte = native_local_ptep_get_and_clear(ptep); 1074 } else { 1075 pte = ptep_get_and_clear(mm, addr, ptep); 1076 } 1077 return pte; 1078 } 1079 1080 #define __HAVE_ARCH_PTEP_SET_WRPROTECT 1081 static inline void ptep_set_wrprotect(struct mm_struct *mm, 1082 unsigned long addr, pte_t *ptep) 1083 { 1084 clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte); 1085 } 1086 1087 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0) 1088 1089 #define mk_pmd(page, pgprot) pfn_pmd(page_to_pfn(page), (pgprot)) 1090 1091 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS 1092 extern int pmdp_set_access_flags(struct vm_area_struct *vma, 1093 unsigned long address, pmd_t *pmdp, 1094 pmd_t entry, int dirty); 1095 extern int pudp_set_access_flags(struct vm_area_struct *vma, 1096 unsigned long address, pud_t *pudp, 1097 pud_t entry, int dirty); 1098 1099 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG 1100 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma, 1101 unsigned long addr, pmd_t *pmdp); 1102 extern int pudp_test_and_clear_young(struct vm_area_struct *vma, 1103 unsigned long addr, pud_t *pudp); 1104 1105 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH 1106 extern int pmdp_clear_flush_young(struct vm_area_struct *vma, 1107 unsigned long address, pmd_t *pmdp); 1108 1109 1110 #define pmd_write pmd_write 1111 static inline int pmd_write(pmd_t pmd) 1112 { 1113 return pmd_flags(pmd) & _PAGE_RW; 1114 } 1115 1116 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR 1117 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr, 1118 pmd_t *pmdp) 1119 { 1120 return native_pmdp_get_and_clear(pmdp); 1121 } 1122 1123 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR 1124 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm, 1125 unsigned long addr, pud_t *pudp) 1126 { 1127 return native_pudp_get_and_clear(pudp); 1128 } 1129 1130 #define __HAVE_ARCH_PMDP_SET_WRPROTECT 1131 static inline void pmdp_set_wrprotect(struct mm_struct *mm, 1132 unsigned long addr, pmd_t *pmdp) 1133 { 1134 clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp); 1135 } 1136 1137 #define pud_write pud_write 1138 static inline int pud_write(pud_t pud) 1139 { 1140 return pud_flags(pud) & _PAGE_RW; 1141 } 1142 1143 #ifndef pmdp_establish 1144 #define pmdp_establish pmdp_establish 1145 static inline pmd_t pmdp_establish(struct vm_area_struct *vma, 1146 unsigned long address, pmd_t *pmdp, pmd_t pmd) 1147 { 1148 if (IS_ENABLED(CONFIG_SMP)) { 1149 return xchg(pmdp, pmd); 1150 } else { 1151 pmd_t old = *pmdp; 1152 *pmdp = pmd; 1153 return old; 1154 } 1155 } 1156 #endif 1157 1158 /* 1159 * clone_pgd_range(pgd_t *dst, pgd_t *src, int count); 1160 * 1161 * dst - pointer to pgd range anwhere on a pgd page 1162 * src - "" 1163 * count - the number of pgds to copy. 1164 * 1165 * dst and src can be on the same page, but the range must not overlap, 1166 * and must not cross a page boundary. 1167 */ 1168 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count) 1169 { 1170 memcpy(dst, src, count * sizeof(pgd_t)); 1171 #ifdef CONFIG_PAGE_TABLE_ISOLATION 1172 if (!static_cpu_has(X86_FEATURE_PTI)) 1173 return; 1174 /* Clone the user space pgd as well */ 1175 memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src), 1176 count * sizeof(pgd_t)); 1177 #endif 1178 } 1179 1180 #define PTE_SHIFT ilog2(PTRS_PER_PTE) 1181 static inline int page_level_shift(enum pg_level level) 1182 { 1183 return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT; 1184 } 1185 static inline unsigned long page_level_size(enum pg_level level) 1186 { 1187 return 1UL << page_level_shift(level); 1188 } 1189 static inline unsigned long page_level_mask(enum pg_level level) 1190 { 1191 return ~(page_level_size(level) - 1); 1192 } 1193 1194 /* 1195 * The x86 doesn't have any external MMU info: the kernel page 1196 * tables contain all the necessary information. 1197 */ 1198 static inline void update_mmu_cache(struct vm_area_struct *vma, 1199 unsigned long addr, pte_t *ptep) 1200 { 1201 } 1202 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma, 1203 unsigned long addr, pmd_t *pmd) 1204 { 1205 } 1206 static inline void update_mmu_cache_pud(struct vm_area_struct *vma, 1207 unsigned long addr, pud_t *pud) 1208 { 1209 } 1210 1211 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY 1212 static inline pte_t pte_swp_mksoft_dirty(pte_t pte) 1213 { 1214 return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY); 1215 } 1216 1217 static inline int pte_swp_soft_dirty(pte_t pte) 1218 { 1219 return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY; 1220 } 1221 1222 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte) 1223 { 1224 return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY); 1225 } 1226 1227 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1228 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd) 1229 { 1230 return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY); 1231 } 1232 1233 static inline int pmd_swp_soft_dirty(pmd_t pmd) 1234 { 1235 return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY; 1236 } 1237 1238 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd) 1239 { 1240 return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY); 1241 } 1242 #endif 1243 #endif 1244 1245 #define PKRU_AD_BIT 0x1 1246 #define PKRU_WD_BIT 0x2 1247 #define PKRU_BITS_PER_PKEY 2 1248 1249 static inline bool __pkru_allows_read(u32 pkru, u16 pkey) 1250 { 1251 int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY; 1252 return !(pkru & (PKRU_AD_BIT << pkru_pkey_bits)); 1253 } 1254 1255 static inline bool __pkru_allows_write(u32 pkru, u16 pkey) 1256 { 1257 int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY; 1258 /* 1259 * Access-disable disables writes too so we need to check 1260 * both bits here. 1261 */ 1262 return !(pkru & ((PKRU_AD_BIT|PKRU_WD_BIT) << pkru_pkey_bits)); 1263 } 1264 1265 static inline u16 pte_flags_pkey(unsigned long pte_flags) 1266 { 1267 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 1268 /* ifdef to avoid doing 59-bit shift on 32-bit values */ 1269 return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0; 1270 #else 1271 return 0; 1272 #endif 1273 } 1274 1275 static inline bool __pkru_allows_pkey(u16 pkey, bool write) 1276 { 1277 u32 pkru = read_pkru(); 1278 1279 if (!__pkru_allows_read(pkru, pkey)) 1280 return false; 1281 if (write && !__pkru_allows_write(pkru, pkey)) 1282 return false; 1283 1284 return true; 1285 } 1286 1287 /* 1288 * 'pteval' can come from a PTE, PMD or PUD. We only check 1289 * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the 1290 * same value on all 3 types. 1291 */ 1292 static inline bool __pte_access_permitted(unsigned long pteval, bool write) 1293 { 1294 unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER; 1295 1296 if (write) 1297 need_pte_bits |= _PAGE_RW; 1298 1299 if ((pteval & need_pte_bits) != need_pte_bits) 1300 return 0; 1301 1302 return __pkru_allows_pkey(pte_flags_pkey(pteval), write); 1303 } 1304 1305 #define pte_access_permitted pte_access_permitted 1306 static inline bool pte_access_permitted(pte_t pte, bool write) 1307 { 1308 return __pte_access_permitted(pte_val(pte), write); 1309 } 1310 1311 #define pmd_access_permitted pmd_access_permitted 1312 static inline bool pmd_access_permitted(pmd_t pmd, bool write) 1313 { 1314 return __pte_access_permitted(pmd_val(pmd), write); 1315 } 1316 1317 #define pud_access_permitted pud_access_permitted 1318 static inline bool pud_access_permitted(pud_t pud, bool write) 1319 { 1320 return __pte_access_permitted(pud_val(pud), write); 1321 } 1322 1323 #include <asm-generic/pgtable.h> 1324 #endif /* __ASSEMBLY__ */ 1325 1326 #endif /* _ASM_X86_PGTABLE_H */ 1327