arch/x86/include/asm/pgtable.h

   1 #ifndef _ASM_X86_PGTABLE_H
   2 #define _ASM_X86_PGTABLE_H
   3
   4 #include <asm/page.h>
   5 #include <asm/e820.h>
   6
   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 #ifndef __ASSEMBLY__
  19 #include <asm/x86_init.h>
  20
  21 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd);
  22 void ptdump_walk_pgd_level_checkwx(void);
  23
  24 #ifdef CONFIG_DEBUG_WX
  25 #define debug_checkwx() ptdump_walk_pgd_level_checkwx()
  26 #else
  27 #define debug_checkwx() do { } while (0)
  28 #endif
  29
  30 /*
  31  * ZERO_PAGE is a global shared page that is always zero: used
  32  * for zero-mapped memory areas etc..
  33  */
  34 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
  35         __visible;
  36 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
  37
  38 extern spinlock_t pgd_lock;
  39 extern struct list_head pgd_list;
  40
  41 extern struct mm_struct *pgd_page_get_mm(struct page *page);
  42
  43 #ifdef CONFIG_PARAVIRT
  44 #include <asm/paravirt.h>
  45 #else  /* !CONFIG_PARAVIRT */
  46 #define set_pte(ptep, pte)              native_set_pte(ptep, pte)
  47 #define set_pte_at(mm, addr, ptep, pte) native_set_pte_at(mm, addr, ptep, pte)
  48 #define set_pmd_at(mm, addr, pmdp, pmd) native_set_pmd_at(mm, addr, pmdp, pmd)
  49 #define set_pud_at(mm, addr, pudp, pud) native_set_pud_at(mm, addr, pudp, pud)
  50
  51 #define set_pte_atomic(ptep, pte)                                       \
  52         native_set_pte_atomic(ptep, pte)
  53
  54 #define set_pmd(pmdp, pmd)              native_set_pmd(pmdp, pmd)
  55
  56 #ifndef __PAGETABLE_PUD_FOLDED
  57 #define set_pgd(pgdp, pgd)              native_set_pgd(pgdp, pgd)
  58 #define pgd_clear(pgd)                  native_pgd_clear(pgd)
  59 #endif
  60
  61 #ifndef set_pud
  62 # define set_pud(pudp, pud)             native_set_pud(pudp, pud)
  63 #endif
  64
  65 #ifndef __PAGETABLE_PUD_FOLDED
  66 #define pud_clear(pud)                  native_pud_clear(pud)
  67 #endif
  68
  69 #define pte_clear(mm, addr, ptep)       native_pte_clear(mm, addr, ptep)
  70 #define pmd_clear(pmd)                  native_pmd_clear(pmd)
  71
  72 #define pte_update(mm, addr, ptep)              do { } while (0)
  73
  74 #define pgd_val(x)      native_pgd_val(x)
  75 #define __pgd(x)        native_make_pgd(x)
  76
  77 #ifndef __PAGETABLE_PUD_FOLDED
  78 #define pud_val(x)      native_pud_val(x)
  79 #define __pud(x)        native_make_pud(x)
  80 #endif
  81
  82 #ifndef __PAGETABLE_PMD_FOLDED
  83 #define pmd_val(x)      native_pmd_val(x)
  84 #define __pmd(x)        native_make_pmd(x)
  85 #endif
  86
  87 #define pte_val(x)      native_pte_val(x)
  88 #define __pte(x)        native_make_pte(x)
  89
  90 #define arch_end_context_switch(prev)   do {} while(0)
  91
  92 #endif  /* CONFIG_PARAVIRT */
  93
  94 /*
  95  * The following only work if pte_present() is true.
  96  * Undefined behaviour if not..
  97  */
  98 static inline int pte_dirty(pte_t pte)
  99 {
 100         return pte_flags(pte) & _PAGE_DIRTY;
 101 }
 102
 103
 104 static inline u32 read_pkru(void)
 105 {
 106         if (boot_cpu_has(X86_FEATURE_OSPKE))
 107                 return __read_pkru();
 108         return 0;
 109 }
 110
 111 static inline void write_pkru(u32 pkru)
 112 {
 113         if (boot_cpu_has(X86_FEATURE_OSPKE))
 114                 __write_pkru(pkru);
 115 }
 116
 117 static inline int pte_young(pte_t pte)
 118 {
 119         return pte_flags(pte) & _PAGE_ACCESSED;
 120 }
 121
 122 static inline int pmd_dirty(pmd_t pmd)
 123 {
 124         return pmd_flags(pmd) & _PAGE_DIRTY;
 125 }
 126
 127 static inline int pmd_young(pmd_t pmd)
 128 {
 129         return pmd_flags(pmd) & _PAGE_ACCESSED;
 130 }
 131
 132 static inline int pud_dirty(pud_t pud)
 133 {
 134         return pud_flags(pud) & _PAGE_DIRTY;
 135 }
 136
 137 static inline int pud_young(pud_t pud)
 138 {
 139         return pud_flags(pud) & _PAGE_ACCESSED;
 140 }
 141
 142 static inline int pte_write(pte_t pte)
 143 {
 144         return pte_flags(pte) & _PAGE_RW;
 145 }
 146
 147 static inline int pte_huge(pte_t pte)
 148 {
 149         return pte_flags(pte) & _PAGE_PSE;
 150 }
 151
 152 static inline int pte_global(pte_t pte)
 153 {
 154         return pte_flags(pte) & _PAGE_GLOBAL;
 155 }
 156
 157 static inline int pte_exec(pte_t pte)
 158 {
 159         return !(pte_flags(pte) & _PAGE_NX);
 160 }
 161
 162 static inline int pte_special(pte_t pte)
 163 {
 164         return pte_flags(pte) & _PAGE_SPECIAL;
 165 }
 166
 167 static inline unsigned long pte_pfn(pte_t pte)
 168 {
 169         return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
 170 }
 171
 172 static inline unsigned long pmd_pfn(pmd_t pmd)
 173 {
 174         return (pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
 175 }
 176
 177 static inline unsigned long pud_pfn(pud_t pud)
 178 {
 179         return (pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT;
 180 }
 181
 182 #define pte_page(pte)   pfn_to_page(pte_pfn(pte))
 183
 184 static inline int pmd_large(pmd_t pte)
 185 {
 186         return pmd_flags(pte) & _PAGE_PSE;
 187 }
 188
 189 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 190 static inline int pmd_trans_huge(pmd_t pmd)
 191 {
 192         return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
 193 }
 194
 195 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 196 static inline int pud_trans_huge(pud_t pud)
 197 {
 198         return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
 199 }
 200 #endif
 201
 202 #define has_transparent_hugepage has_transparent_hugepage
 203 static inline int has_transparent_hugepage(void)
 204 {
 205         return boot_cpu_has(X86_FEATURE_PSE);
 206 }
 207
 208 #ifdef __HAVE_ARCH_PTE_DEVMAP
 209 static inline int pmd_devmap(pmd_t pmd)
 210 {
 211         return !!(pmd_val(pmd) & _PAGE_DEVMAP);
 212 }
 213
 214 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 215 static inline int pud_devmap(pud_t pud)
 216 {
 217         return !!(pud_val(pud) & _PAGE_DEVMAP);
 218 }
 219 #else
 220 static inline int pud_devmap(pud_t pud)
 221 {
 222         return 0;
 223 }
 224 #endif
 225 #endif
 226 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 227
 228 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
 229 {
 230         pteval_t v = native_pte_val(pte);
 231
 232         return native_make_pte(v | set);
 233 }
 234
 235 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
 236 {
 237         pteval_t v = native_pte_val(pte);
 238
 239         return native_make_pte(v & ~clear);
 240 }
 241
 242 static inline pte_t pte_mkclean(pte_t pte)
 243 {
 244         return pte_clear_flags(pte, _PAGE_DIRTY);
 245 }
 246
 247 static inline pte_t pte_mkold(pte_t pte)
 248 {
 249         return pte_clear_flags(pte, _PAGE_ACCESSED);
 250 }
 251
 252 static inline pte_t pte_wrprotect(pte_t pte)
 253 {
 254         return pte_clear_flags(pte, _PAGE_RW);
 255 }
 256
 257 static inline pte_t pte_mkexec(pte_t pte)
 258 {
 259         return pte_clear_flags(pte, _PAGE_NX);
 260 }
 261
 262 static inline pte_t pte_mkdirty(pte_t pte)
 263 {
 264         return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
 265 }
 266
 267 static inline pte_t pte_mkyoung(pte_t pte)
 268 {
 269         return pte_set_flags(pte, _PAGE_ACCESSED);
 270 }
 271
 272 static inline pte_t pte_mkwrite(pte_t pte)
 273 {
 274         return pte_set_flags(pte, _PAGE_RW);
 275 }
 276
 277 static inline pte_t pte_mkhuge(pte_t pte)
 278 {
 279         return pte_set_flags(pte, _PAGE_PSE);
 280 }
 281
 282 static inline pte_t pte_clrhuge(pte_t pte)
 283 {
 284         return pte_clear_flags(pte, _PAGE_PSE);
 285 }
 286
 287 static inline pte_t pte_mkglobal(pte_t pte)
 288 {
 289         return pte_set_flags(pte, _PAGE_GLOBAL);
 290 }
 291
 292 static inline pte_t pte_clrglobal(pte_t pte)
 293 {
 294         return pte_clear_flags(pte, _PAGE_GLOBAL);
 295 }
 296
 297 static inline pte_t pte_mkspecial(pte_t pte)
 298 {
 299         return pte_set_flags(pte, _PAGE_SPECIAL);
 300 }
 301
 302 static inline pte_t pte_mkdevmap(pte_t pte)
 303 {
 304         return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
 305 }
 306
 307 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
 308 {
 309         pmdval_t v = native_pmd_val(pmd);
 310
 311         return __pmd(v | set);
 312 }
 313
 314 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
 315 {
 316         pmdval_t v = native_pmd_val(pmd);
 317
 318         return __pmd(v & ~clear);
 319 }
 320
 321 static inline pmd_t pmd_mkold(pmd_t pmd)
 322 {
 323         return pmd_clear_flags(pmd, _PAGE_ACCESSED);
 324 }
 325
 326 static inline pmd_t pmd_mkclean(pmd_t pmd)
 327 {
 328         return pmd_clear_flags(pmd, _PAGE_DIRTY);
 329 }
 330
 331 static inline pmd_t pmd_wrprotect(pmd_t pmd)
 332 {
 333         return pmd_clear_flags(pmd, _PAGE_RW);
 334 }
 335
 336 static inline pmd_t pmd_mkdirty(pmd_t pmd)
 337 {
 338         return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
 339 }
 340
 341 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
 342 {
 343         return pmd_set_flags(pmd, _PAGE_DEVMAP);
 344 }
 345
 346 static inline pmd_t pmd_mkhuge(pmd_t pmd)
 347 {
 348         return pmd_set_flags(pmd, _PAGE_PSE);
 349 }
 350
 351 static inline pmd_t pmd_mkyoung(pmd_t pmd)
 352 {
 353         return pmd_set_flags(pmd, _PAGE_ACCESSED);
 354 }
 355
 356 static inline pmd_t pmd_mkwrite(pmd_t pmd)
 357 {
 358         return pmd_set_flags(pmd, _PAGE_RW);
 359 }
 360
 361 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
 362 {
 363         return pmd_clear_flags(pmd, _PAGE_PRESENT | _PAGE_PROTNONE);
 364 }
 365
 366 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
 367 {
 368         pudval_t v = native_pud_val(pud);
 369
 370         return __pud(v | set);
 371 }
 372
 373 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
 374 {
 375         pudval_t v = native_pud_val(pud);
 376
 377         return __pud(v & ~clear);
 378 }
 379
 380 static inline pud_t pud_mkold(pud_t pud)
 381 {
 382         return pud_clear_flags(pud, _PAGE_ACCESSED);
 383 }
 384
 385 static inline pud_t pud_mkclean(pud_t pud)
 386 {
 387         return pud_clear_flags(pud, _PAGE_DIRTY);
 388 }
 389
 390 static inline pud_t pud_wrprotect(pud_t pud)
 391 {
 392         return pud_clear_flags(pud, _PAGE_RW);
 393 }
 394
 395 static inline pud_t pud_mkdirty(pud_t pud)
 396 {
 397         return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
 398 }
 399
 400 static inline pud_t pud_mkdevmap(pud_t pud)
 401 {
 402         return pud_set_flags(pud, _PAGE_DEVMAP);
 403 }
 404
 405 static inline pud_t pud_mkhuge(pud_t pud)
 406 {
 407         return pud_set_flags(pud, _PAGE_PSE);
 408 }
 409
 410 static inline pud_t pud_mkyoung(pud_t pud)
 411 {
 412         return pud_set_flags(pud, _PAGE_ACCESSED);
 413 }
 414
 415 static inline pud_t pud_mkwrite(pud_t pud)
 416 {
 417         return pud_set_flags(pud, _PAGE_RW);
 418 }
 419
 420 static inline pud_t pud_mknotpresent(pud_t pud)
 421 {
 422         return pud_clear_flags(pud, _PAGE_PRESENT | _PAGE_PROTNONE);
 423 }
 424
 425 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
 426 static inline int pte_soft_dirty(pte_t pte)
 427 {
 428         return pte_flags(pte) & _PAGE_SOFT_DIRTY;
 429 }
 430
 431 static inline int pmd_soft_dirty(pmd_t pmd)
 432 {
 433         return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
 434 }
 435
 436 static inline int pud_soft_dirty(pud_t pud)
 437 {
 438         return pud_flags(pud) & _PAGE_SOFT_DIRTY;
 439 }
 440
 441 static inline pte_t pte_mksoft_dirty(pte_t pte)
 442 {
 443         return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
 444 }
 445
 446 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 447 {
 448         return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
 449 }
 450
 451 static inline pud_t pud_mksoft_dirty(pud_t pud)
 452 {
 453         return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
 454 }
 455
 456 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 457 {
 458         return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
 459 }
 460
 461 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 462 {
 463         return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
 464 }
 465
 466 static inline pud_t pud_clear_soft_dirty(pud_t pud)
 467 {
 468         return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
 469 }
 470
 471 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
 472
 473 /*
 474  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
 475  * can use those bits for other purposes, so leave them be.
 476  */
 477 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
 478 {
 479         pgprotval_t protval = pgprot_val(pgprot);
 480
 481         if (protval & _PAGE_PRESENT)
 482                 protval &= __supported_pte_mask;
 483
 484         return protval;
 485 }
 486
 487 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
 488 {
 489         return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) |
 490                      massage_pgprot(pgprot));
 491 }
 492
 493 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
 494 {
 495         return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) |
 496                      massage_pgprot(pgprot));
 497 }
 498
 499 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
 500 {
 501         return __pud(((phys_addr_t)page_nr << PAGE_SHIFT) |
 502                      massage_pgprot(pgprot));
 503 }
 504
 505 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 506 {
 507         pteval_t val = pte_val(pte);
 508
 509         /*
 510          * Chop off the NX bit (if present), and add the NX portion of
 511          * the newprot (if present):
 512          */
 513         val &= _PAGE_CHG_MASK;
 514         val |= massage_pgprot(newprot) & ~_PAGE_CHG_MASK;
 515
 516         return __pte(val);
 517 }
 518
 519 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
 520 {
 521         pmdval_t val = pmd_val(pmd);
 522
 523         val &= _HPAGE_CHG_MASK;
 524         val |= massage_pgprot(newprot) & ~_HPAGE_CHG_MASK;
 525
 526         return __pmd(val);
 527 }
 528
 529 /* mprotect needs to preserve PAT bits when updating vm_page_prot */
 530 #define pgprot_modify pgprot_modify
 531 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 532 {
 533         pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
 534         pgprotval_t addbits = pgprot_val(newprot);
 535         return __pgprot(preservebits | addbits);
 536 }
 537
 538 #define pte_pgprot(x) __pgprot(pte_flags(x))
 539 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
 540 #define pud_pgprot(x) __pgprot(pud_flags(x))
 541
 542 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
 543
 544 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
 545                                          enum page_cache_mode pcm,
 546                                          enum page_cache_mode new_pcm)
 547 {
 548         /*
 549          * PAT type is always WB for untracked ranges, so no need to check.
 550          */
 551         if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
 552                 return 1;
 553
 554         /*
 555          * Certain new memtypes are not allowed with certain
 556          * requested memtype:
 557          * - request is uncached, return cannot be write-back
 558          * - request is write-combine, return cannot be write-back
 559          * - request is write-through, return cannot be write-back
 560          * - request is write-through, return cannot be write-combine
 561          */
 562         if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
 563              new_pcm == _PAGE_CACHE_MODE_WB) ||
 564             (pcm == _PAGE_CACHE_MODE_WC &&
 565              new_pcm == _PAGE_CACHE_MODE_WB) ||
 566             (pcm == _PAGE_CACHE_MODE_WT &&
 567              new_pcm == _PAGE_CACHE_MODE_WB) ||
 568             (pcm == _PAGE_CACHE_MODE_WT &&
 569              new_pcm == _PAGE_CACHE_MODE_WC)) {
 570                 return 0;
 571         }
 572
 573         return 1;
 574 }
 575
 576 pmd_t *populate_extra_pmd(unsigned long vaddr);
 577 pte_t *populate_extra_pte(unsigned long vaddr);
 578 #endif  /* __ASSEMBLY__ */
 579
 580 #ifdef CONFIG_X86_32
 581 # include <asm/pgtable_32.h>
 582 #else
 583 # include <asm/pgtable_64.h>
 584 #endif
 585
 586 #ifndef __ASSEMBLY__
 587 #include <linux/mm_types.h>
 588 #include <linux/mmdebug.h>
 589 #include <linux/log2.h>
 590
 591 static inline int pte_none(pte_t pte)
 592 {
 593         return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
 594 }
 595
 596 #define __HAVE_ARCH_PTE_SAME
 597 static inline int pte_same(pte_t a, pte_t b)
 598 {
 599         return a.pte == b.pte;
 600 }
 601
 602 static inline int pte_present(pte_t a)
 603 {
 604         return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
 605 }
 606
 607 #ifdef __HAVE_ARCH_PTE_DEVMAP
 608 static inline int pte_devmap(pte_t a)
 609 {
 610         return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
 611 }
 612 #endif
 613
 614 #define pte_accessible pte_accessible
 615 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
 616 {
 617         if (pte_flags(a) & _PAGE_PRESENT)
 618                 return true;
 619
 620         if ((pte_flags(a) & _PAGE_PROTNONE) &&
 621                         mm_tlb_flush_pending(mm))
 622                 return true;
 623
 624         return false;
 625 }
 626
 627 static inline int pte_hidden(pte_t pte)
 628 {
 629         return pte_flags(pte) & _PAGE_HIDDEN;
 630 }
 631
 632 static inline int pmd_present(pmd_t pmd)
 633 {
 634         /*
 635          * Checking for _PAGE_PSE is needed too because
 636          * split_huge_page will temporarily clear the present bit (but
 637          * the _PAGE_PSE flag will remain set at all times while the
 638          * _PAGE_PRESENT bit is clear).
 639          */
 640         return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
 641 }
 642
 643 #ifdef CONFIG_NUMA_BALANCING
 644 /*
 645  * These work without NUMA balancing but the kernel does not care. See the
 646  * comment in include/asm-generic/pgtable.h
 647  */
 648 static inline int pte_protnone(pte_t pte)
 649 {
 650         return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
 651                 == _PAGE_PROTNONE;
 652 }
 653
 654 static inline int pmd_protnone(pmd_t pmd)
 655 {
 656         return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
 657                 == _PAGE_PROTNONE;
 658 }
 659 #endif /* CONFIG_NUMA_BALANCING */
 660
 661 static inline int pmd_none(pmd_t pmd)
 662 {
 663         /* Only check low word on 32-bit platforms, since it might be
 664            out of sync with upper half. */
 665         unsigned long val = native_pmd_val(pmd);
 666         return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
 667 }
 668
 669 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
 670 {
 671         return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
 672 }
 673
 674 /*
 675  * Currently stuck as a macro due to indirect forward reference to
 676  * linux/mmzone.h's __section_mem_map_addr() definition:
 677  */
 678 #define pmd_page(pmd)           \
 679         pfn_to_page((pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT)
 680
 681 /*
 682  * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
 683  *
 684  * this macro returns the index of the entry in the pmd page which would
 685  * control the given virtual address
 686  */
 687 static inline unsigned long pmd_index(unsigned long address)
 688 {
 689         return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
 690 }
 691
 692 /*
 693  * Conversion functions: convert a page and protection to a page entry,
 694  * and a page entry and page directory to the page they refer to.
 695  *
 696  * (Currently stuck as a macro because of indirect forward reference
 697  * to linux/mm.h:page_to_nid())
 698  */
 699 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
 700
 701 /*
 702  * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
 703  *
 704  * this function returns the index of the entry in the pte page which would
 705  * control the given virtual address
 706  */
 707 static inline unsigned long pte_index(unsigned long address)
 708 {
 709         return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
 710 }
 711
 712 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
 713 {
 714         return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
 715 }
 716
 717 static inline int pmd_bad(pmd_t pmd)
 718 {
 719         return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
 720 }
 721
 722 static inline unsigned long pages_to_mb(unsigned long npg)
 723 {
 724         return npg >> (20 - PAGE_SHIFT);
 725 }
 726
 727 #if CONFIG_PGTABLE_LEVELS > 2
 728 static inline int pud_none(pud_t pud)
 729 {
 730         return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
 731 }
 732
 733 static inline int pud_present(pud_t pud)
 734 {
 735         return pud_flags(pud) & _PAGE_PRESENT;
 736 }
 737
 738 static inline unsigned long pud_page_vaddr(pud_t pud)
 739 {
 740         return (unsigned long)__va(pud_val(pud) & pud_pfn_mask(pud));
 741 }
 742
 743 /*
 744  * Currently stuck as a macro due to indirect forward reference to
 745  * linux/mmzone.h's __section_mem_map_addr() definition:
 746  */
 747 #define pud_page(pud)           \
 748         pfn_to_page((pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT)
 749
 750 /* Find an entry in the second-level page table.. */
 751 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
 752 {
 753         return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
 754 }
 755
 756 static inline int pud_large(pud_t pud)
 757 {
 758         return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
 759                 (_PAGE_PSE | _PAGE_PRESENT);
 760 }
 761
 762 static inline int pud_bad(pud_t pud)
 763 {
 764         return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
 765 }
 766 #else
 767 static inline int pud_large(pud_t pud)
 768 {
 769         return 0;
 770 }
 771 #endif  /* CONFIG_PGTABLE_LEVELS > 2 */
 772
 773 #if CONFIG_PGTABLE_LEVELS > 3
 774 static inline int pgd_present(pgd_t pgd)
 775 {
 776         return pgd_flags(pgd) & _PAGE_PRESENT;
 777 }
 778
 779 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
 780 {
 781         return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
 782 }
 783
 784 /*
 785  * Currently stuck as a macro due to indirect forward reference to
 786  * linux/mmzone.h's __section_mem_map_addr() definition:
 787  */
 788 #define pgd_page(pgd)           pfn_to_page(pgd_val(pgd) >> PAGE_SHIFT)
 789
 790 /* to find an entry in a page-table-directory. */
 791 static inline unsigned long pud_index(unsigned long address)
 792 {
 793         return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
 794 }
 795
 796 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
 797 {
 798         return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(address);
 799 }
 800
 801 static inline int pgd_bad(pgd_t pgd)
 802 {
 803         return (pgd_flags(pgd) & ~_PAGE_USER) != _KERNPG_TABLE;
 804 }
 805
 806 static inline int pgd_none(pgd_t pgd)
 807 {
 808         /*
 809          * There is no need to do a workaround for the KNL stray
 810          * A/D bit erratum here.  PGDs only point to page tables
 811          * except on 32-bit non-PAE which is not supported on
 812          * KNL.
 813          */
 814         return !native_pgd_val(pgd);
 815 }
 816 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
 817
 818 #endif  /* __ASSEMBLY__ */
 819
 820 /*
 821  * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
 822  *
 823  * this macro returns the index of the entry in the pgd page which would
 824  * control the given virtual address
 825  */
 826 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
 827
 828 /*
 829  * pgd_offset() returns a (pgd_t *)
 830  * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
 831  */
 832 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index((address)))
 833 /*
 834  * a shortcut which implies the use of the kernel's pgd, instead
 835  * of a process's
 836  */
 837 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
 838
 839
 840 #define KERNEL_PGD_BOUNDARY     pgd_index(PAGE_OFFSET)
 841 #define KERNEL_PGD_PTRS         (PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
 842
 843 #ifndef __ASSEMBLY__
 844
 845 extern int direct_gbpages;
 846 void init_mem_mapping(void);
 847 void early_alloc_pgt_buf(void);
 848
 849 #ifdef CONFIG_X86_64
 850 /* Realmode trampoline initialization. */
 851 extern pgd_t trampoline_pgd_entry;
 852 static inline void __meminit init_trampoline_default(void)
 853 {
 854         /* Default trampoline pgd value */
 855         trampoline_pgd_entry = init_level4_pgt[pgd_index(__PAGE_OFFSET)];
 856 }
 857 # ifdef CONFIG_RANDOMIZE_MEMORY
 858 void __meminit init_trampoline(void);
 859 # else
 860 #  define init_trampoline init_trampoline_default
 861 # endif
 862 #else
 863 static inline void init_trampoline(void) { }
 864 #endif
 865
 866 /* local pte updates need not use xchg for locking */
 867 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
 868 {
 869         pte_t res = *ptep;
 870
 871         /* Pure native function needs no input for mm, addr */
 872         native_pte_clear(NULL, 0, ptep);
 873         return res;
 874 }
 875
 876 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
 877 {
 878         pmd_t res = *pmdp;
 879
 880         native_pmd_clear(pmdp);
 881         return res;
 882 }
 883
 884 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
 885 {
 886         pud_t res = *pudp;
 887
 888         native_pud_clear(pudp);
 889         return res;
 890 }
 891
 892 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
 893                                      pte_t *ptep , pte_t pte)
 894 {
 895         native_set_pte(ptep, pte);
 896 }
 897
 898 static inline void native_set_pmd_at(struct mm_struct *mm, unsigned long addr,
 899                                      pmd_t *pmdp , pmd_t pmd)
 900 {
 901         native_set_pmd(pmdp, pmd);
 902 }
 903
 904 static inline void native_set_pud_at(struct mm_struct *mm, unsigned long addr,
 905                                      pud_t *pudp, pud_t pud)
 906 {
 907         native_set_pud(pudp, pud);
 908 }
 909
 910 #ifndef CONFIG_PARAVIRT
 911 /*
 912  * Rules for using pte_update - it must be called after any PTE update which
 913  * has not been done using the set_pte / clear_pte interfaces.  It is used by
 914  * shadow mode hypervisors to resynchronize the shadow page tables.  Kernel PTE
 915  * updates should either be sets, clears, or set_pte_atomic for P->P
 916  * transitions, which means this hook should only be called for user PTEs.
 917  * This hook implies a P->P protection or access change has taken place, which
 918  * requires a subsequent TLB flush.
 919  */
 920 #define pte_update(mm, addr, ptep)              do { } while (0)
 921 #endif
 922
 923 /*
 924  * We only update the dirty/accessed state if we set
 925  * the dirty bit by hand in the kernel, since the hardware
 926  * will do the accessed bit for us, and we don't want to
 927  * race with other CPU's that might be updating the dirty
 928  * bit at the same time.
 929  */
 930 struct vm_area_struct;
 931
 932 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
 933 extern int ptep_set_access_flags(struct vm_area_struct *vma,
 934                                  unsigned long address, pte_t *ptep,
 935                                  pte_t entry, int dirty);
 936
 937 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
 938 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
 939                                      unsigned long addr, pte_t *ptep);
 940
 941 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
 942 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
 943                                   unsigned long address, pte_t *ptep);
 944
 945 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 946 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
 947                                        pte_t *ptep)
 948 {
 949         pte_t pte = native_ptep_get_and_clear(ptep);
 950         pte_update(mm, addr, ptep);
 951         return pte;
 952 }
 953
 954 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 955 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 956                                             unsigned long addr, pte_t *ptep,
 957                                             int full)
 958 {
 959         pte_t pte;
 960         if (full) {
 961                 /*
 962                  * Full address destruction in progress; paravirt does not
 963                  * care about updates and native needs no locking
 964                  */
 965                 pte = native_local_ptep_get_and_clear(ptep);
 966         } else {
 967                 pte = ptep_get_and_clear(mm, addr, ptep);
 968         }
 969         return pte;
 970 }
 971
 972 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
 973 static inline void ptep_set_wrprotect(struct mm_struct *mm,
 974                                       unsigned long addr, pte_t *ptep)
 975 {
 976         clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
 977         pte_update(mm, addr, ptep);
 978 }
 979
 980 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
 981
 982 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
 983
 984 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
 985 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
 986                                  unsigned long address, pmd_t *pmdp,
 987                                  pmd_t entry, int dirty);
 988 extern int pudp_set_access_flags(struct vm_area_struct *vma,
 989                                  unsigned long address, pud_t *pudp,
 990                                  pud_t entry, int dirty);
 991
 992 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
 993 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
 994                                      unsigned long addr, pmd_t *pmdp);
 995 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
 996                                      unsigned long addr, pud_t *pudp);
 997
 998 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
 999 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1000                                   unsigned long address, pmd_t *pmdp);
1001
1002
1003 #define __HAVE_ARCH_PMD_WRITE
1004 static inline int pmd_write(pmd_t pmd)
1005 {
1006         return pmd_flags(pmd) & _PAGE_RW;
1007 }
1008
1009 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1010 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1011                                        pmd_t *pmdp)
1012 {
1013         return native_pmdp_get_and_clear(pmdp);
1014 }
1015
1016 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1017 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1018                                         unsigned long addr, pud_t *pudp)
1019 {
1020         return native_pudp_get_and_clear(pudp);
1021 }
1022
1023 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1024 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1025                                       unsigned long addr, pmd_t *pmdp)
1026 {
1027         clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1028 }
1029
1030 /*
1031  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1032  *
1033  *  dst - pointer to pgd range anwhere on a pgd page
1034  *  src - ""
1035  *  count - the number of pgds to copy.
1036  *
1037  * dst and src can be on the same page, but the range must not overlap,
1038  * and must not cross a page boundary.
1039  */
1040 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1041 {
1042        memcpy(dst, src, count * sizeof(pgd_t));
1043 }
1044
1045 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1046 static inline int page_level_shift(enum pg_level level)
1047 {
1048         return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1049 }
1050 static inline unsigned long page_level_size(enum pg_level level)
1051 {
1052         return 1UL << page_level_shift(level);
1053 }
1054 static inline unsigned long page_level_mask(enum pg_level level)
1055 {
1056         return ~(page_level_size(level) - 1);
1057 }
1058
1059 /*
1060  * The x86 doesn't have any external MMU info: the kernel page
1061  * tables contain all the necessary information.
1062  */
1063 static inline void update_mmu_cache(struct vm_area_struct *vma,
1064                 unsigned long addr, pte_t *ptep)
1065 {
1066 }
1067 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1068                 unsigned long addr, pmd_t *pmd)
1069 {
1070 }
1071 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1072                 unsigned long addr, pud_t *pud)
1073 {
1074 }
1075
1076 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1077 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1078 {
1079         return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1080 }
1081
1082 static inline int pte_swp_soft_dirty(pte_t pte)
1083 {
1084         return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1085 }
1086
1087 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1088 {
1089         return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1090 }
1091 #endif
1092
1093 #define PKRU_AD_BIT 0x1
1094 #define PKRU_WD_BIT 0x2
1095 #define PKRU_BITS_PER_PKEY 2
1096
1097 static inline bool __pkru_allows_read(u32 pkru, u16 pkey)
1098 {
1099         int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1100         return !(pkru & (PKRU_AD_BIT << pkru_pkey_bits));
1101 }
1102
1103 static inline bool __pkru_allows_write(u32 pkru, u16 pkey)
1104 {
1105         int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1106         /*
1107          * Access-disable disables writes too so we need to check
1108          * both bits here.
1109          */
1110         return !(pkru & ((PKRU_AD_BIT|PKRU_WD_BIT) << pkru_pkey_bits));
1111 }
1112
1113 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1114 {
1115 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1116         /* ifdef to avoid doing 59-bit shift on 32-bit values */
1117         return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1118 #else
1119         return 0;
1120 #endif
1121 }
1122
1123 #include <asm-generic/pgtable.h>
1124 #endif  /* __ASSEMBLY__ */
1125
1126 #endif /* _ASM_X86_PGTABLE_H */