include/asm-generic/pgtable.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _ASM_GENERIC_PGTABLE_H
   3 #define _ASM_GENERIC_PGTABLE_H
   4
   5 #include <linux/pfn.h>
   6
   7 #ifndef __ASSEMBLY__
   8 #ifdef CONFIG_MMU
   9
  10 #include <linux/mm_types.h>
  11 #include <linux/bug.h>
  12 #include <linux/errno.h>
  13
  14 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
  15         defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
  16 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
  17 #endif
  18
  19 /*
  20  * On almost all architectures and configurations, 0 can be used as the
  21  * upper ceiling to free_pgtables(): on many architectures it has the same
  22  * effect as using TASK_SIZE.  However, there is one configuration which
  23  * must impose a more careful limit, to avoid freeing kernel pgtables.
  24  */
  25 #ifndef USER_PGTABLES_CEILING
  26 #define USER_PGTABLES_CEILING   0UL
  27 #endif
  28
  29 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  30 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  31                                  unsigned long address, pte_t *ptep,
  32                                  pte_t entry, int dirty);
  33 #endif
  34
  35 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  36 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  37 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  38                                  unsigned long address, pmd_t *pmdp,
  39                                  pmd_t entry, int dirty);
  40 extern int pudp_set_access_flags(struct vm_area_struct *vma,
  41                                  unsigned long address, pud_t *pudp,
  42                                  pud_t entry, int dirty);
  43 #else
  44 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
  45                                         unsigned long address, pmd_t *pmdp,
  46                                         pmd_t entry, int dirty)
  47 {
  48         BUILD_BUG();
  49         return 0;
  50 }
  51 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
  52                                         unsigned long address, pud_t *pudp,
  53                                         pud_t entry, int dirty)
  54 {
  55         BUILD_BUG();
  56         return 0;
  57 }
  58 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  59 #endif
  60
  61 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  62 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  63                                             unsigned long address,
  64                                             pte_t *ptep)
  65 {
  66         pte_t pte = *ptep;
  67         int r = 1;
  68         if (!pte_young(pte))
  69                 r = 0;
  70         else
  71                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  72         return r;
  73 }
  74 #endif
  75
  76 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  77 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  78 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  79                                             unsigned long address,
  80                                             pmd_t *pmdp)
  81 {
  82         pmd_t pmd = *pmdp;
  83         int r = 1;
  84         if (!pmd_young(pmd))
  85                 r = 0;
  86         else
  87                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  88         return r;
  89 }
  90 #else
  91 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  92                                             unsigned long address,
  93                                             pmd_t *pmdp)
  94 {
  95         BUILD_BUG();
  96         return 0;
  97 }
  98 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  99 #endif
 100
 101 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
 102 int ptep_clear_flush_young(struct vm_area_struct *vma,
 103                            unsigned long address, pte_t *ptep);
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
 107 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 108 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
 109                                   unsigned long address, pmd_t *pmdp);
 110 #else
 111 /*
 112  * Despite relevant to THP only, this API is called from generic rmap code
 113  * under PageTransHuge(), hence needs a dummy implementation for !THP
 114  */
 115 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
 116                                          unsigned long address, pmd_t *pmdp)
 117 {
 118         BUILD_BUG();
 119         return 0;
 120 }
 121 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 122 #endif
 123
 124 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
 125 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
 126                                        unsigned long address,
 127                                        pte_t *ptep)
 128 {
 129         pte_t pte = *ptep;
 130         pte_clear(mm, address, ptep);
 131         return pte;
 132 }
 133 #endif
 134
 135 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 136 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
 137 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
 138                                             unsigned long address,
 139                                             pmd_t *pmdp)
 140 {
 141         pmd_t pmd = *pmdp;
 142         pmd_clear(pmdp);
 143         return pmd;
 144 }
 145 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
 146 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
 147 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
 148                                             unsigned long address,
 149                                             pud_t *pudp)
 150 {
 151         pud_t pud = *pudp;
 152
 153         pud_clear(pudp);
 154         return pud;
 155 }
 156 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
 157 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 158
 159 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 160 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
 161 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
 162                                             unsigned long address, pmd_t *pmdp,
 163                                             int full)
 164 {
 165         return pmdp_huge_get_and_clear(mm, address, pmdp);
 166 }
 167 #endif
 168
 169 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
 170 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
 171                                             unsigned long address, pud_t *pudp,
 172                                             int full)
 173 {
 174         return pudp_huge_get_and_clear(mm, address, pudp);
 175 }
 176 #endif
 177 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 178
 179 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 180 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 181                                             unsigned long address, pte_t *ptep,
 182                                             int full)
 183 {
 184         pte_t pte;
 185         pte = ptep_get_and_clear(mm, address, ptep);
 186         return pte;
 187 }
 188 #endif
 189
 190 /*
 191  * Some architectures may be able to avoid expensive synchronization
 192  * primitives when modifications are made to PTE's which are already
 193  * not present, or in the process of an address space destruction.
 194  */
 195 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 196 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 197                                               unsigned long address,
 198                                               pte_t *ptep,
 199                                               int full)
 200 {
 201         pte_clear(mm, address, ptep);
 202 }
 203 #endif
 204
 205 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 206 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 207                               unsigned long address,
 208                               pte_t *ptep);
 209 #endif
 210
 211 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
 212 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
 213                               unsigned long address,
 214                               pmd_t *pmdp);
 215 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
 216                               unsigned long address,
 217                               pud_t *pudp);
 218 #endif
 219
 220 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 221 struct mm_struct;
 222 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 223 {
 224         pte_t old_pte = *ptep;
 225         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 226 }
 227 #endif
 228
 229 #ifndef pte_savedwrite
 230 #define pte_savedwrite pte_write
 231 #endif
 232
 233 #ifndef pte_mk_savedwrite
 234 #define pte_mk_savedwrite pte_mkwrite
 235 #endif
 236
 237 #ifndef pte_clear_savedwrite
 238 #define pte_clear_savedwrite pte_wrprotect
 239 #endif
 240
 241 #ifndef pmd_savedwrite
 242 #define pmd_savedwrite pmd_write
 243 #endif
 244
 245 #ifndef pmd_mk_savedwrite
 246 #define pmd_mk_savedwrite pmd_mkwrite
 247 #endif
 248
 249 #ifndef pmd_clear_savedwrite
 250 #define pmd_clear_savedwrite pmd_wrprotect
 251 #endif
 252
 253 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 254 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 255 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 256                                       unsigned long address, pmd_t *pmdp)
 257 {
 258         pmd_t old_pmd = *pmdp;
 259         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 260 }
 261 #else
 262 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 263                                       unsigned long address, pmd_t *pmdp)
 264 {
 265         BUILD_BUG();
 266 }
 267 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 268 #endif
 269 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
 270 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 271 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 272                                       unsigned long address, pud_t *pudp)
 273 {
 274         pud_t old_pud = *pudp;
 275
 276         set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
 277 }
 278 #else
 279 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 280                                       unsigned long address, pud_t *pudp)
 281 {
 282         BUILD_BUG();
 283 }
 284 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 285 #endif
 286
 287 #ifndef pmdp_collapse_flush
 288 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 289 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 290                                  unsigned long address, pmd_t *pmdp);
 291 #else
 292 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 293                                         unsigned long address,
 294                                         pmd_t *pmdp)
 295 {
 296         BUILD_BUG();
 297         return *pmdp;
 298 }
 299 #define pmdp_collapse_flush pmdp_collapse_flush
 300 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 301 #endif
 302
 303 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 304 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 305                                        pgtable_t pgtable);
 306 #endif
 307
 308 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 309 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 310 #endif
 311
 312 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 313 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 314                             pmd_t *pmdp);
 315 #endif
 316
 317 #ifndef __HAVE_ARCH_PMDP_HUGE_SPLIT_PREPARE
 318 static inline void pmdp_huge_split_prepare(struct vm_area_struct *vma,
 319                                            unsigned long address, pmd_t *pmdp)
 320 {
 321
 322 }
 323 #endif
 324
 325 #ifndef __HAVE_ARCH_PTE_SAME
 326 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 327 {
 328         return pte_val(pte_a) == pte_val(pte_b);
 329 }
 330 #endif
 331
 332 #ifndef __HAVE_ARCH_PTE_UNUSED
 333 /*
 334  * Some architectures provide facilities to virtualization guests
 335  * so that they can flag allocated pages as unused. This allows the
 336  * host to transparently reclaim unused pages. This function returns
 337  * whether the pte's page is unused.
 338  */
 339 static inline int pte_unused(pte_t pte)
 340 {
 341         return 0;
 342 }
 343 #endif
 344
 345 #ifndef pte_access_permitted
 346 #define pte_access_permitted(pte, write) \
 347         (pte_present(pte) && (!(write) || pte_write(pte)))
 348 #endif
 349
 350 #ifndef pmd_access_permitted
 351 #define pmd_access_permitted(pmd, write) \
 352         (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
 353 #endif
 354
 355 #ifndef pud_access_permitted
 356 #define pud_access_permitted(pud, write) \
 357         (pud_present(pud) && (!(write) || pud_write(pud)))
 358 #endif
 359
 360 #ifndef p4d_access_permitted
 361 #define p4d_access_permitted(p4d, write) \
 362         (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
 363 #endif
 364
 365 #ifndef pgd_access_permitted
 366 #define pgd_access_permitted(pgd, write) \
 367         (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
 368 #endif
 369
 370 #ifndef __HAVE_ARCH_PMD_SAME
 371 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 372 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 373 {
 374         return pmd_val(pmd_a) == pmd_val(pmd_b);
 375 }
 376
 377 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 378 {
 379         return pud_val(pud_a) == pud_val(pud_b);
 380 }
 381 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 382 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 383 {
 384         BUILD_BUG();
 385         return 0;
 386 }
 387
 388 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 389 {
 390         BUILD_BUG();
 391         return 0;
 392 }
 393 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 394 #endif
 395
 396 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 397 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 398 #endif
 399
 400 #ifndef __HAVE_ARCH_MOVE_PTE
 401 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 402 #endif
 403
 404 #ifndef pte_accessible
 405 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 406 #endif
 407
 408 #ifndef flush_tlb_fix_spurious_fault
 409 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 410 #endif
 411
 412 #ifndef pgprot_noncached
 413 #define pgprot_noncached(prot)  (prot)
 414 #endif
 415
 416 #ifndef pgprot_writecombine
 417 #define pgprot_writecombine pgprot_noncached
 418 #endif
 419
 420 #ifndef pgprot_writethrough
 421 #define pgprot_writethrough pgprot_noncached
 422 #endif
 423
 424 #ifndef pgprot_device
 425 #define pgprot_device pgprot_noncached
 426 #endif
 427
 428 #ifndef pgprot_modify
 429 #define pgprot_modify pgprot_modify
 430 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 431 {
 432         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 433                 newprot = pgprot_noncached(newprot);
 434         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 435                 newprot = pgprot_writecombine(newprot);
 436         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 437                 newprot = pgprot_device(newprot);
 438         return newprot;
 439 }
 440 #endif
 441
 442 /*
 443  * When walking page tables, get the address of the next boundary,
 444  * or the end address of the range if that comes earlier.  Although no
 445  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 446  */
 447
 448 #define pgd_addr_end(addr, end)                                         \
 449 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 450         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 451 })
 452
 453 #ifndef p4d_addr_end
 454 #define p4d_addr_end(addr, end)                                         \
 455 ({      unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;      \
 456         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 457 })
 458 #endif
 459
 460 #ifndef pud_addr_end
 461 #define pud_addr_end(addr, end)                                         \
 462 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 463         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 464 })
 465 #endif
 466
 467 #ifndef pmd_addr_end
 468 #define pmd_addr_end(addr, end)                                         \
 469 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 470         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 471 })
 472 #endif
 473
 474 /*
 475  * When walking page tables, we usually want to skip any p?d_none entries;
 476  * and any p?d_bad entries - reporting the error before resetting to none.
 477  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 478  */
 479 void pgd_clear_bad(pgd_t *);
 480 void p4d_clear_bad(p4d_t *);
 481 void pud_clear_bad(pud_t *);
 482 void pmd_clear_bad(pmd_t *);
 483
 484 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 485 {
 486         if (pgd_none(*pgd))
 487                 return 1;
 488         if (unlikely(pgd_bad(*pgd))) {
 489                 pgd_clear_bad(pgd);
 490                 return 1;
 491         }
 492         return 0;
 493 }
 494
 495 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
 496 {
 497         if (p4d_none(*p4d))
 498                 return 1;
 499         if (unlikely(p4d_bad(*p4d))) {
 500                 p4d_clear_bad(p4d);
 501                 return 1;
 502         }
 503         return 0;
 504 }
 505
 506 static inline int pud_none_or_clear_bad(pud_t *pud)
 507 {
 508         if (pud_none(*pud))
 509                 return 1;
 510         if (unlikely(pud_bad(*pud))) {
 511                 pud_clear_bad(pud);
 512                 return 1;
 513         }
 514         return 0;
 515 }
 516
 517 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 518 {
 519         if (pmd_none(*pmd))
 520                 return 1;
 521         if (unlikely(pmd_bad(*pmd))) {
 522                 pmd_clear_bad(pmd);
 523                 return 1;
 524         }
 525         return 0;
 526 }
 527
 528 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 529                                              unsigned long addr,
 530                                              pte_t *ptep)
 531 {
 532         /*
 533          * Get the current pte state, but zero it out to make it
 534          * non-present, preventing the hardware from asynchronously
 535          * updating it.
 536          */
 537         return ptep_get_and_clear(mm, addr, ptep);
 538 }
 539
 540 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 541                                              unsigned long addr,
 542                                              pte_t *ptep, pte_t pte)
 543 {
 544         /*
 545          * The pte is non-present, so there's no hardware state to
 546          * preserve.
 547          */
 548         set_pte_at(mm, addr, ptep, pte);
 549 }
 550
 551 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 552 /*
 553  * Start a pte protection read-modify-write transaction, which
 554  * protects against asynchronous hardware modifications to the pte.
 555  * The intention is not to prevent the hardware from making pte
 556  * updates, but to prevent any updates it may make from being lost.
 557  *
 558  * This does not protect against other software modifications of the
 559  * pte; the appropriate pte lock must be held over the transation.
 560  *
 561  * Note that this interface is intended to be batchable, meaning that
 562  * ptep_modify_prot_commit may not actually update the pte, but merely
 563  * queue the update to be done at some later time.  The update must be
 564  * actually committed before the pte lock is released, however.
 565  */
 566 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 567                                            unsigned long addr,
 568                                            pte_t *ptep)
 569 {
 570         return __ptep_modify_prot_start(mm, addr, ptep);
 571 }
 572
 573 /*
 574  * Commit an update to a pte, leaving any hardware-controlled bits in
 575  * the PTE unmodified.
 576  */
 577 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 578                                            unsigned long addr,
 579                                            pte_t *ptep, pte_t pte)
 580 {
 581         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 582 }
 583 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 584 #endif /* CONFIG_MMU */
 585
 586 /*
 587  * No-op macros that just return the current protection value. Defined here
 588  * because these macros can be used used even if CONFIG_MMU is not defined.
 589  */
 590 #ifndef pgprot_encrypted
 591 #define pgprot_encrypted(prot)  (prot)
 592 #endif
 593
 594 #ifndef pgprot_decrypted
 595 #define pgprot_decrypted(prot)  (prot)
 596 #endif
 597
 598 /*
 599  * A facility to provide lazy MMU batching.  This allows PTE updates and
 600  * page invalidations to be delayed until a call to leave lazy MMU mode
 601  * is issued.  Some architectures may benefit from doing this, and it is
 602  * beneficial for both shadow and direct mode hypervisors, which may batch
 603  * the PTE updates which happen during this window.  Note that using this
 604  * interface requires that read hazards be removed from the code.  A read
 605  * hazard could result in the direct mode hypervisor case, since the actual
 606  * write to the page tables may not yet have taken place, so reads though
 607  * a raw PTE pointer after it has been modified are not guaranteed to be
 608  * up to date.  This mode can only be entered and left under the protection of
 609  * the page table locks for all page tables which may be modified.  In the UP
 610  * case, this is required so that preemption is disabled, and in the SMP case,
 611  * it must synchronize the delayed page table writes properly on other CPUs.
 612  */
 613 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 614 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 615 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 616 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 617 #endif
 618
 619 /*
 620  * A facility to provide batching of the reload of page tables and
 621  * other process state with the actual context switch code for
 622  * paravirtualized guests.  By convention, only one of the batched
 623  * update (lazy) modes (CPU, MMU) should be active at any given time,
 624  * entry should never be nested, and entry and exits should always be
 625  * paired.  This is for sanity of maintaining and reasoning about the
 626  * kernel code.  In this case, the exit (end of the context switch) is
 627  * in architecture-specific code, and so doesn't need a generic
 628  * definition.
 629  */
 630 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 631 #define arch_start_context_switch(prev) do {} while (0)
 632 #endif
 633
 634 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
 635 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
 636 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
 637 {
 638         return pmd;
 639 }
 640
 641 static inline int pmd_swp_soft_dirty(pmd_t pmd)
 642 {
 643         return 0;
 644 }
 645
 646 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
 647 {
 648         return pmd;
 649 }
 650 #endif
 651 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
 652 static inline int pte_soft_dirty(pte_t pte)
 653 {
 654         return 0;
 655 }
 656
 657 static inline int pmd_soft_dirty(pmd_t pmd)
 658 {
 659         return 0;
 660 }
 661
 662 static inline pte_t pte_mksoft_dirty(pte_t pte)
 663 {
 664         return pte;
 665 }
 666
 667 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 668 {
 669         return pmd;
 670 }
 671
 672 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 673 {
 674         return pte;
 675 }
 676
 677 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 678 {
 679         return pmd;
 680 }
 681
 682 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 683 {
 684         return pte;
 685 }
 686
 687 static inline int pte_swp_soft_dirty(pte_t pte)
 688 {
 689         return 0;
 690 }
 691
 692 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 693 {
 694         return pte;
 695 }
 696
 697 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
 698 {
 699         return pmd;
 700 }
 701
 702 static inline int pmd_swp_soft_dirty(pmd_t pmd)
 703 {
 704         return 0;
 705 }
 706
 707 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
 708 {
 709         return pmd;
 710 }
 711 #endif
 712
 713 #ifndef __HAVE_PFNMAP_TRACKING
 714 /*
 715  * Interfaces that can be used by architecture code to keep track of
 716  * memory type of pfn mappings specified by the remap_pfn_range,
 717  * vm_insert_pfn.
 718  */
 719
 720 /*
 721  * track_pfn_remap is called when a _new_ pfn mapping is being established
 722  * by remap_pfn_range() for physical range indicated by pfn and size.
 723  */
 724 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 725                                   unsigned long pfn, unsigned long addr,
 726                                   unsigned long size)
 727 {
 728         return 0;
 729 }
 730
 731 /*
 732  * track_pfn_insert is called when a _new_ single pfn is established
 733  * by vm_insert_pfn().
 734  */
 735 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 736                                     pfn_t pfn)
 737 {
 738 }
 739
 740 /*
 741  * track_pfn_copy is called when vma that is covering the pfnmap gets
 742  * copied through copy_page_range().
 743  */
 744 static inline int track_pfn_copy(struct vm_area_struct *vma)
 745 {
 746         return 0;
 747 }
 748
 749 /*
 750  * untrack_pfn is called while unmapping a pfnmap for a region.
 751  * untrack can be called for a specific region indicated by pfn and size or
 752  * can be for the entire vma (in which case pfn, size are zero).
 753  */
 754 static inline void untrack_pfn(struct vm_area_struct *vma,
 755                                unsigned long pfn, unsigned long size)
 756 {
 757 }
 758
 759 /*
 760  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
 761  */
 762 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
 763 {
 764 }
 765 #else
 766 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 767                            unsigned long pfn, unsigned long addr,
 768                            unsigned long size);
 769 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 770                              pfn_t pfn);
 771 extern int track_pfn_copy(struct vm_area_struct *vma);
 772 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 773                         unsigned long size);
 774 extern void untrack_pfn_moved(struct vm_area_struct *vma);
 775 #endif
 776
 777 #ifdef __HAVE_COLOR_ZERO_PAGE
 778 static inline int is_zero_pfn(unsigned long pfn)
 779 {
 780         extern unsigned long zero_pfn;
 781         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 782         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 783 }
 784
 785 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 786
 787 #else
 788 static inline int is_zero_pfn(unsigned long pfn)
 789 {
 790         extern unsigned long zero_pfn;
 791         return pfn == zero_pfn;
 792 }
 793
 794 static inline unsigned long my_zero_pfn(unsigned long addr)
 795 {
 796         extern unsigned long zero_pfn;
 797         return zero_pfn;
 798 }
 799 #endif
 800
 801 #ifdef CONFIG_MMU
 802
 803 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 804 static inline int pmd_trans_huge(pmd_t pmd)
 805 {
 806         return 0;
 807 }
 808 #ifndef __HAVE_ARCH_PMD_WRITE
 809 static inline int pmd_write(pmd_t pmd)
 810 {
 811         BUG();
 812         return 0;
 813 }
 814 #endif /* __HAVE_ARCH_PMD_WRITE */
 815 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 816
 817 #ifndef pud_write
 818 static inline int pud_write(pud_t pud)
 819 {
 820         BUG();
 821         return 0;
 822 }
 823 #endif /* pud_write */
 824
 825 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
 826         (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
 827          !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
 828 static inline int pud_trans_huge(pud_t pud)
 829 {
 830         return 0;
 831 }
 832 #endif
 833
 834 #ifndef pmd_read_atomic
 835 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 836 {
 837         /*
 838          * Depend on compiler for an atomic pmd read. NOTE: this is
 839          * only going to work, if the pmdval_t isn't larger than
 840          * an unsigned long.
 841          */
 842         return *pmdp;
 843 }
 844 #endif
 845
 846 #ifndef arch_needs_pgtable_deposit
 847 #define arch_needs_pgtable_deposit() (false)
 848 #endif
 849 /*
 850  * This function is meant to be used by sites walking pagetables with
 851  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 852  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 853  * into a null pmd and the transhuge page fault can convert a null pmd
 854  * into an hugepmd or into a regular pmd (if the hugepage allocation
 855  * fails). While holding the mmap_sem in read mode the pmd becomes
 856  * stable and stops changing under us only if it's not null and not a
 857  * transhuge pmd. When those races occurs and this function makes a
 858  * difference vs the standard pmd_none_or_clear_bad, the result is
 859  * undefined so behaving like if the pmd was none is safe (because it
 860  * can return none anyway). The compiler level barrier() is critically
 861  * important to compute the two checks atomically on the same pmdval.
 862  *
 863  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 864  * care of reading the pmd atomically to avoid SMP race conditions
 865  * against pmd_populate() when the mmap_sem is hold for reading by the
 866  * caller (a special atomic read not done by "gcc" as in the generic
 867  * version above, is also needed when THP is disabled because the page
 868  * fault can populate the pmd from under us).
 869  */
 870 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 871 {
 872         pmd_t pmdval = pmd_read_atomic(pmd);
 873         /*
 874          * The barrier will stabilize the pmdval in a register or on
 875          * the stack so that it will stop changing under the code.
 876          *
 877          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 878          * pmd_read_atomic is allowed to return a not atomic pmdval
 879          * (for example pointing to an hugepage that has never been
 880          * mapped in the pmd). The below checks will only care about
 881          * the low part of the pmd with 32bit PAE x86 anyway, with the
 882          * exception of pmd_none(). So the important thing is that if
 883          * the low part of the pmd is found null, the high part will
 884          * be also null or the pmd_none() check below would be
 885          * confused.
 886          */
 887 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 888         barrier();
 889 #endif
 890         /*
 891          * !pmd_present() checks for pmd migration entries
 892          *
 893          * The complete check uses is_pmd_migration_entry() in linux/swapops.h
 894          * But using that requires moving current function and pmd_trans_unstable()
 895          * to linux/swapops.h to resovle dependency, which is too much code move.
 896          *
 897          * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
 898          * because !pmd_present() pages can only be under migration not swapped
 899          * out.
 900          *
 901          * pmd_none() is preseved for future condition checks on pmd migration
 902          * entries and not confusing with this function name, although it is
 903          * redundant with !pmd_present().
 904          */
 905         if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
 906                 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
 907                 return 1;
 908         if (unlikely(pmd_bad(pmdval))) {
 909                 pmd_clear_bad(pmd);
 910                 return 1;
 911         }
 912         return 0;
 913 }
 914
 915 /*
 916  * This is a noop if Transparent Hugepage Support is not built into
 917  * the kernel. Otherwise it is equivalent to
 918  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 919  * places that already verified the pmd is not none and they want to
 920  * walk ptes while holding the mmap sem in read mode (write mode don't
 921  * need this). If THP is not enabled, the pmd can't go away under the
 922  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 923  * run a pmd_trans_unstable before walking the ptes after
 924  * split_huge_page_pmd returns (because it may have run when the pmd
 925  * become null, but then a page fault can map in a THP and not a
 926  * regular page).
 927  */
 928 static inline int pmd_trans_unstable(pmd_t *pmd)
 929 {
 930 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 931         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 932 #else
 933         return 0;
 934 #endif
 935 }
 936
 937 #ifndef CONFIG_NUMA_BALANCING
 938 /*
 939  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 940  * the only case the kernel cares is for NUMA balancing and is only ever set
 941  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 942  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 943  * is the responsibility of the caller to distinguish between PROT_NONE
 944  * protections and NUMA hinting fault protections.
 945  */
 946 static inline int pte_protnone(pte_t pte)
 947 {
 948         return 0;
 949 }
 950
 951 static inline int pmd_protnone(pmd_t pmd)
 952 {
 953         return 0;
 954 }
 955 #endif /* CONFIG_NUMA_BALANCING */
 956
 957 #endif /* CONFIG_MMU */
 958
 959 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 960
 961 #ifndef __PAGETABLE_P4D_FOLDED
 962 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
 963 int p4d_clear_huge(p4d_t *p4d);
 964 #else
 965 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 966 {
 967         return 0;
 968 }
 969 static inline int p4d_clear_huge(p4d_t *p4d)
 970 {
 971         return 0;
 972 }
 973 #endif /* !__PAGETABLE_P4D_FOLDED */
 974
 975 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 976 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 977 int pud_clear_huge(pud_t *pud);
 978 int pmd_clear_huge(pmd_t *pmd);
 979 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 980 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 981 {
 982         return 0;
 983 }
 984 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 985 {
 986         return 0;
 987 }
 988 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 989 {
 990         return 0;
 991 }
 992 static inline int p4d_clear_huge(p4d_t *p4d)
 993 {
 994         return 0;
 995 }
 996 static inline int pud_clear_huge(pud_t *pud)
 997 {
 998         return 0;
 999 }
1000 static inline int pmd_clear_huge(pmd_t *pmd)
1001 {
1002         return 0;
1003 }
1004 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1005
1006 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1007 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1008 /*
1009  * ARCHes with special requirements for evicting THP backing TLB entries can
1010  * implement this. Otherwise also, it can help optimize normal TLB flush in
1011  * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1012  * entire TLB TLB if flush span is greater than a threshold, which will
1013  * likely be true for a single huge page. Thus a single thp flush will
1014  * invalidate the entire TLB which is not desitable.
1015  * e.g. see arch/arc: flush_pmd_tlb_range
1016  */
1017 #define flush_pmd_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
1018 #define flush_pud_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
1019 #else
1020 #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
1021 #define flush_pud_tlb_range(vma, addr, end)     BUILD_BUG()
1022 #endif
1023 #endif
1024
1025 struct file;
1026 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1027                         unsigned long size, pgprot_t *vma_prot);
1028 #endif /* !__ASSEMBLY__ */
1029
1030 #ifndef io_remap_pfn_range
1031 #define io_remap_pfn_range remap_pfn_range
1032 #endif
1033
1034 #ifndef has_transparent_hugepage
1035 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1036 #define has_transparent_hugepage() 1
1037 #else
1038 #define has_transparent_hugepage() 0
1039 #endif
1040 #endif
1041
1042 #endif /* _ASM_GENERIC_PGTABLE_H */