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 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 313 /*
 314  * This is an implementation of pmdp_establish() that is only suitable for an
 315  * architecture that doesn't have hardware dirty/accessed bits. In this case we
 316  * can't race with CPU which sets these bits and non-atomic aproach is fine.
 317  */
 318 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
 319                 unsigned long address, pmd_t *pmdp, pmd_t pmd)
 320 {
 321         pmd_t old_pmd = *pmdp;
 322         set_pmd_at(vma->vm_mm, address, pmdp, pmd);
 323         return old_pmd;
 324 }
 325 #endif
 326
 327 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 328 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 329                             pmd_t *pmdp);
 330 #endif
 331
 332 #ifndef __HAVE_ARCH_PTE_SAME
 333 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 334 {
 335         return pte_val(pte_a) == pte_val(pte_b);
 336 }
 337 #endif
 338
 339 #ifndef __HAVE_ARCH_PTE_UNUSED
 340 /*
 341  * Some architectures provide facilities to virtualization guests
 342  * so that they can flag allocated pages as unused. This allows the
 343  * host to transparently reclaim unused pages. This function returns
 344  * whether the pte's page is unused.
 345  */
 346 static inline int pte_unused(pte_t pte)
 347 {
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifndef pte_access_permitted
 353 #define pte_access_permitted(pte, write) \
 354         (pte_present(pte) && (!(write) || pte_write(pte)))
 355 #endif
 356
 357 #ifndef pmd_access_permitted
 358 #define pmd_access_permitted(pmd, write) \
 359         (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
 360 #endif
 361
 362 #ifndef pud_access_permitted
 363 #define pud_access_permitted(pud, write) \
 364         (pud_present(pud) && (!(write) || pud_write(pud)))
 365 #endif
 366
 367 #ifndef p4d_access_permitted
 368 #define p4d_access_permitted(p4d, write) \
 369         (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
 370 #endif
 371
 372 #ifndef pgd_access_permitted
 373 #define pgd_access_permitted(pgd, write) \
 374         (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
 375 #endif
 376
 377 #ifndef __HAVE_ARCH_PMD_SAME
 378 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 379 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 380 {
 381         return pmd_val(pmd_a) == pmd_val(pmd_b);
 382 }
 383
 384 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 385 {
 386         return pud_val(pud_a) == pud_val(pud_b);
 387 }
 388 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 389 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 390 {
 391         BUILD_BUG();
 392         return 0;
 393 }
 394
 395 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 396 {
 397         BUILD_BUG();
 398         return 0;
 399 }
 400 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 401 #endif
 402
 403 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 404 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 405 #endif
 406
 407 #ifndef __HAVE_ARCH_MOVE_PTE
 408 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 409 #endif
 410
 411 #ifndef pte_accessible
 412 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 413 #endif
 414
 415 #ifndef flush_tlb_fix_spurious_fault
 416 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 417 #endif
 418
 419 #ifndef pgprot_noncached
 420 #define pgprot_noncached(prot)  (prot)
 421 #endif
 422
 423 #ifndef pgprot_writecombine
 424 #define pgprot_writecombine pgprot_noncached
 425 #endif
 426
 427 #ifndef pgprot_writethrough
 428 #define pgprot_writethrough pgprot_noncached
 429 #endif
 430
 431 #ifndef pgprot_device
 432 #define pgprot_device pgprot_noncached
 433 #endif
 434
 435 #ifndef pgprot_modify
 436 #define pgprot_modify pgprot_modify
 437 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 438 {
 439         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 440                 newprot = pgprot_noncached(newprot);
 441         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 442                 newprot = pgprot_writecombine(newprot);
 443         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 444                 newprot = pgprot_device(newprot);
 445         return newprot;
 446 }
 447 #endif
 448
 449 /*
 450  * When walking page tables, get the address of the next boundary,
 451  * or the end address of the range if that comes earlier.  Although no
 452  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 453  */
 454
 455 #define pgd_addr_end(addr, end)                                         \
 456 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 457         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 458 })
 459
 460 #ifndef p4d_addr_end
 461 #define p4d_addr_end(addr, end)                                         \
 462 ({      unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;      \
 463         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 464 })
 465 #endif
 466
 467 #ifndef pud_addr_end
 468 #define pud_addr_end(addr, end)                                         \
 469 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 470         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 471 })
 472 #endif
 473
 474 #ifndef pmd_addr_end
 475 #define pmd_addr_end(addr, end)                                         \
 476 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 477         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 478 })
 479 #endif
 480
 481 /*
 482  * When walking page tables, we usually want to skip any p?d_none entries;
 483  * and any p?d_bad entries - reporting the error before resetting to none.
 484  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 485  */
 486 void pgd_clear_bad(pgd_t *);
 487 void p4d_clear_bad(p4d_t *);
 488 void pud_clear_bad(pud_t *);
 489 void pmd_clear_bad(pmd_t *);
 490
 491 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 492 {
 493         if (pgd_none(*pgd))
 494                 return 1;
 495         if (unlikely(pgd_bad(*pgd))) {
 496                 pgd_clear_bad(pgd);
 497                 return 1;
 498         }
 499         return 0;
 500 }
 501
 502 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
 503 {
 504         if (p4d_none(*p4d))
 505                 return 1;
 506         if (unlikely(p4d_bad(*p4d))) {
 507                 p4d_clear_bad(p4d);
 508                 return 1;
 509         }
 510         return 0;
 511 }
 512
 513 static inline int pud_none_or_clear_bad(pud_t *pud)
 514 {
 515         if (pud_none(*pud))
 516                 return 1;
 517         if (unlikely(pud_bad(*pud))) {
 518                 pud_clear_bad(pud);
 519                 return 1;
 520         }
 521         return 0;
 522 }
 523
 524 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 525 {
 526         if (pmd_none(*pmd))
 527                 return 1;
 528         if (unlikely(pmd_bad(*pmd))) {
 529                 pmd_clear_bad(pmd);
 530                 return 1;
 531         }
 532         return 0;
 533 }
 534
 535 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 536                                              unsigned long addr,
 537                                              pte_t *ptep)
 538 {
 539         /*
 540          * Get the current pte state, but zero it out to make it
 541          * non-present, preventing the hardware from asynchronously
 542          * updating it.
 543          */
 544         return ptep_get_and_clear(mm, addr, ptep);
 545 }
 546
 547 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 548                                              unsigned long addr,
 549                                              pte_t *ptep, pte_t pte)
 550 {
 551         /*
 552          * The pte is non-present, so there's no hardware state to
 553          * preserve.
 554          */
 555         set_pte_at(mm, addr, ptep, pte);
 556 }
 557
 558 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 559 /*
 560  * Start a pte protection read-modify-write transaction, which
 561  * protects against asynchronous hardware modifications to the pte.
 562  * The intention is not to prevent the hardware from making pte
 563  * updates, but to prevent any updates it may make from being lost.
 564  *
 565  * This does not protect against other software modifications of the
 566  * pte; the appropriate pte lock must be held over the transation.
 567  *
 568  * Note that this interface is intended to be batchable, meaning that
 569  * ptep_modify_prot_commit may not actually update the pte, but merely
 570  * queue the update to be done at some later time.  The update must be
 571  * actually committed before the pte lock is released, however.
 572  */
 573 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 574                                            unsigned long addr,
 575                                            pte_t *ptep)
 576 {
 577         return __ptep_modify_prot_start(mm, addr, ptep);
 578 }
 579
 580 /*
 581  * Commit an update to a pte, leaving any hardware-controlled bits in
 582  * the PTE unmodified.
 583  */
 584 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 585                                            unsigned long addr,
 586                                            pte_t *ptep, pte_t pte)
 587 {
 588         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 589 }
 590 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 591 #endif /* CONFIG_MMU */
 592
 593 /*
 594  * No-op macros that just return the current protection value. Defined here
 595  * because these macros can be used used even if CONFIG_MMU is not defined.
 596  */
 597 #ifndef pgprot_encrypted
 598 #define pgprot_encrypted(prot)  (prot)
 599 #endif
 600
 601 #ifndef pgprot_decrypted
 602 #define pgprot_decrypted(prot)  (prot)
 603 #endif
 604
 605 /*
 606  * A facility to provide lazy MMU batching.  This allows PTE updates and
 607  * page invalidations to be delayed until a call to leave lazy MMU mode
 608  * is issued.  Some architectures may benefit from doing this, and it is
 609  * beneficial for both shadow and direct mode hypervisors, which may batch
 610  * the PTE updates which happen during this window.  Note that using this
 611  * interface requires that read hazards be removed from the code.  A read
 612  * hazard could result in the direct mode hypervisor case, since the actual
 613  * write to the page tables may not yet have taken place, so reads though
 614  * a raw PTE pointer after it has been modified are not guaranteed to be
 615  * up to date.  This mode can only be entered and left under the protection of
 616  * the page table locks for all page tables which may be modified.  In the UP
 617  * case, this is required so that preemption is disabled, and in the SMP case,
 618  * it must synchronize the delayed page table writes properly on other CPUs.
 619  */
 620 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 621 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 622 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 623 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 624 #endif
 625
 626 /*
 627  * A facility to provide batching of the reload of page tables and
 628  * other process state with the actual context switch code for
 629  * paravirtualized guests.  By convention, only one of the batched
 630  * update (lazy) modes (CPU, MMU) should be active at any given time,
 631  * entry should never be nested, and entry and exits should always be
 632  * paired.  This is for sanity of maintaining and reasoning about the
 633  * kernel code.  In this case, the exit (end of the context switch) is
 634  * in architecture-specific code, and so doesn't need a generic
 635  * definition.
 636  */
 637 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 638 #define arch_start_context_switch(prev) do {} while (0)
 639 #endif
 640
 641 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
 642 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
 643 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
 644 {
 645         return pmd;
 646 }
 647
 648 static inline int pmd_swp_soft_dirty(pmd_t pmd)
 649 {
 650         return 0;
 651 }
 652
 653 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
 654 {
 655         return pmd;
 656 }
 657 #endif
 658 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
 659 static inline int pte_soft_dirty(pte_t pte)
 660 {
 661         return 0;
 662 }
 663
 664 static inline int pmd_soft_dirty(pmd_t pmd)
 665 {
 666         return 0;
 667 }
 668
 669 static inline pte_t pte_mksoft_dirty(pte_t pte)
 670 {
 671         return pte;
 672 }
 673
 674 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 675 {
 676         return pmd;
 677 }
 678
 679 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 680 {
 681         return pte;
 682 }
 683
 684 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 685 {
 686         return pmd;
 687 }
 688
 689 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 690 {
 691         return pte;
 692 }
 693
 694 static inline int pte_swp_soft_dirty(pte_t pte)
 695 {
 696         return 0;
 697 }
 698
 699 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 700 {
 701         return pte;
 702 }
 703
 704 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
 705 {
 706         return pmd;
 707 }
 708
 709 static inline int pmd_swp_soft_dirty(pmd_t pmd)
 710 {
 711         return 0;
 712 }
 713
 714 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
 715 {
 716         return pmd;
 717 }
 718 #endif
 719
 720 #ifndef __HAVE_PFNMAP_TRACKING
 721 /*
 722  * Interfaces that can be used by architecture code to keep track of
 723  * memory type of pfn mappings specified by the remap_pfn_range,
 724  * vm_insert_pfn.
 725  */
 726
 727 /*
 728  * track_pfn_remap is called when a _new_ pfn mapping is being established
 729  * by remap_pfn_range() for physical range indicated by pfn and size.
 730  */
 731 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 732                                   unsigned long pfn, unsigned long addr,
 733                                   unsigned long size)
 734 {
 735         return 0;
 736 }
 737
 738 /*
 739  * track_pfn_insert is called when a _new_ single pfn is established
 740  * by vm_insert_pfn().
 741  */
 742 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 743                                     pfn_t pfn)
 744 {
 745 }
 746
 747 /*
 748  * track_pfn_copy is called when vma that is covering the pfnmap gets
 749  * copied through copy_page_range().
 750  */
 751 static inline int track_pfn_copy(struct vm_area_struct *vma)
 752 {
 753         return 0;
 754 }
 755
 756 /*
 757  * untrack_pfn is called while unmapping a pfnmap for a region.
 758  * untrack can be called for a specific region indicated by pfn and size or
 759  * can be for the entire vma (in which case pfn, size are zero).
 760  */
 761 static inline void untrack_pfn(struct vm_area_struct *vma,
 762                                unsigned long pfn, unsigned long size)
 763 {
 764 }
 765
 766 /*
 767  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
 768  */
 769 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
 770 {
 771 }
 772 #else
 773 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 774                            unsigned long pfn, unsigned long addr,
 775                            unsigned long size);
 776 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 777                              pfn_t pfn);
 778 extern int track_pfn_copy(struct vm_area_struct *vma);
 779 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 780                         unsigned long size);
 781 extern void untrack_pfn_moved(struct vm_area_struct *vma);
 782 #endif
 783
 784 #ifdef __HAVE_COLOR_ZERO_PAGE
 785 static inline int is_zero_pfn(unsigned long pfn)
 786 {
 787         extern unsigned long zero_pfn;
 788         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 789         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 790 }
 791
 792 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 793
 794 #else
 795 static inline int is_zero_pfn(unsigned long pfn)
 796 {
 797         extern unsigned long zero_pfn;
 798         return pfn == zero_pfn;
 799 }
 800
 801 static inline unsigned long my_zero_pfn(unsigned long addr)
 802 {
 803         extern unsigned long zero_pfn;
 804         return zero_pfn;
 805 }
 806 #endif
 807
 808 #ifdef CONFIG_MMU
 809
 810 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 811 static inline int pmd_trans_huge(pmd_t pmd)
 812 {
 813         return 0;
 814 }
 815 #ifndef pmd_write
 816 static inline int pmd_write(pmd_t pmd)
 817 {
 818         BUG();
 819         return 0;
 820 }
 821 #endif /* pmd_write */
 822 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 823
 824 #ifndef pud_write
 825 static inline int pud_write(pud_t pud)
 826 {
 827         BUG();
 828         return 0;
 829 }
 830 #endif /* pud_write */
 831
 832 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
 833         (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
 834          !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
 835 static inline int pud_trans_huge(pud_t pud)
 836 {
 837         return 0;
 838 }
 839 #endif
 840
 841 #ifndef pmd_read_atomic
 842 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 843 {
 844         /*
 845          * Depend on compiler for an atomic pmd read. NOTE: this is
 846          * only going to work, if the pmdval_t isn't larger than
 847          * an unsigned long.
 848          */
 849         return *pmdp;
 850 }
 851 #endif
 852
 853 #ifndef arch_needs_pgtable_deposit
 854 #define arch_needs_pgtable_deposit() (false)
 855 #endif
 856 /*
 857  * This function is meant to be used by sites walking pagetables with
 858  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 859  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 860  * into a null pmd and the transhuge page fault can convert a null pmd
 861  * into an hugepmd or into a regular pmd (if the hugepage allocation
 862  * fails). While holding the mmap_sem in read mode the pmd becomes
 863  * stable and stops changing under us only if it's not null and not a
 864  * transhuge pmd. When those races occurs and this function makes a
 865  * difference vs the standard pmd_none_or_clear_bad, the result is
 866  * undefined so behaving like if the pmd was none is safe (because it
 867  * can return none anyway). The compiler level barrier() is critically
 868  * important to compute the two checks atomically on the same pmdval.
 869  *
 870  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 871  * care of reading the pmd atomically to avoid SMP race conditions
 872  * against pmd_populate() when the mmap_sem is hold for reading by the
 873  * caller (a special atomic read not done by "gcc" as in the generic
 874  * version above, is also needed when THP is disabled because the page
 875  * fault can populate the pmd from under us).
 876  */
 877 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 878 {
 879         pmd_t pmdval = pmd_read_atomic(pmd);
 880         /*
 881          * The barrier will stabilize the pmdval in a register or on
 882          * the stack so that it will stop changing under the code.
 883          *
 884          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 885          * pmd_read_atomic is allowed to return a not atomic pmdval
 886          * (for example pointing to an hugepage that has never been
 887          * mapped in the pmd). The below checks will only care about
 888          * the low part of the pmd with 32bit PAE x86 anyway, with the
 889          * exception of pmd_none(). So the important thing is that if
 890          * the low part of the pmd is found null, the high part will
 891          * be also null or the pmd_none() check below would be
 892          * confused.
 893          */
 894 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 895         barrier();
 896 #endif
 897         /*
 898          * !pmd_present() checks for pmd migration entries
 899          *
 900          * The complete check uses is_pmd_migration_entry() in linux/swapops.h
 901          * But using that requires moving current function and pmd_trans_unstable()
 902          * to linux/swapops.h to resovle dependency, which is too much code move.
 903          *
 904          * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
 905          * because !pmd_present() pages can only be under migration not swapped
 906          * out.
 907          *
 908          * pmd_none() is preseved for future condition checks on pmd migration
 909          * entries and not confusing with this function name, although it is
 910          * redundant with !pmd_present().
 911          */
 912         if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
 913                 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
 914                 return 1;
 915         if (unlikely(pmd_bad(pmdval))) {
 916                 pmd_clear_bad(pmd);
 917                 return 1;
 918         }
 919         return 0;
 920 }
 921
 922 /*
 923  * This is a noop if Transparent Hugepage Support is not built into
 924  * the kernel. Otherwise it is equivalent to
 925  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 926  * places that already verified the pmd is not none and they want to
 927  * walk ptes while holding the mmap sem in read mode (write mode don't
 928  * need this). If THP is not enabled, the pmd can't go away under the
 929  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 930  * run a pmd_trans_unstable before walking the ptes after
 931  * split_huge_page_pmd returns (because it may have run when the pmd
 932  * become null, but then a page fault can map in a THP and not a
 933  * regular page).
 934  */
 935 static inline int pmd_trans_unstable(pmd_t *pmd)
 936 {
 937 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 938         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 939 #else
 940         return 0;
 941 #endif
 942 }
 943
 944 #ifndef CONFIG_NUMA_BALANCING
 945 /*
 946  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 947  * the only case the kernel cares is for NUMA balancing and is only ever set
 948  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 949  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 950  * is the responsibility of the caller to distinguish between PROT_NONE
 951  * protections and NUMA hinting fault protections.
 952  */
 953 static inline int pte_protnone(pte_t pte)
 954 {
 955         return 0;
 956 }
 957
 958 static inline int pmd_protnone(pmd_t pmd)
 959 {
 960         return 0;
 961 }
 962 #endif /* CONFIG_NUMA_BALANCING */
 963
 964 #endif /* CONFIG_MMU */
 965
 966 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 967
 968 #ifndef __PAGETABLE_P4D_FOLDED
 969 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
 970 int p4d_clear_huge(p4d_t *p4d);
 971 #else
 972 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 973 {
 974         return 0;
 975 }
 976 static inline int p4d_clear_huge(p4d_t *p4d)
 977 {
 978         return 0;
 979 }
 980 #endif /* !__PAGETABLE_P4D_FOLDED */
 981
 982 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 983 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 984 int pud_clear_huge(pud_t *pud);
 985 int pmd_clear_huge(pmd_t *pmd);
 986 int pud_free_pmd_page(pud_t *pud);
 987 int pmd_free_pte_page(pmd_t *pmd);
 988 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 989 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 990 {
 991         return 0;
 992 }
 993 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 994 {
 995         return 0;
 996 }
 997 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 998 {
 999         return 0;
1000 }
1001 static inline int p4d_clear_huge(p4d_t *p4d)
1002 {
1003         return 0;
1004 }
1005 static inline int pud_clear_huge(pud_t *pud)
1006 {
1007         return 0;
1008 }
1009 static inline int pmd_clear_huge(pmd_t *pmd)
1010 {
1011         return 0;
1012 }
1013 static inline int pud_free_pmd_page(pud_t *pud)
1014 {
1015         return 0;
1016 }
1017 static inline int pmd_free_pte_page(pmd_t *pmd)
1018 {
1019         return 0;
1020 }
1021 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1022
1023 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1024 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1025 /*
1026  * ARCHes with special requirements for evicting THP backing TLB entries can
1027  * implement this. Otherwise also, it can help optimize normal TLB flush in
1028  * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1029  * entire TLB TLB if flush span is greater than a threshold, which will
1030  * likely be true for a single huge page. Thus a single thp flush will
1031  * invalidate the entire TLB which is not desitable.
1032  * e.g. see arch/arc: flush_pmd_tlb_range
1033  */
1034 #define flush_pmd_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
1035 #define flush_pud_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
1036 #else
1037 #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
1038 #define flush_pud_tlb_range(vma, addr, end)     BUILD_BUG()
1039 #endif
1040 #endif
1041
1042 struct file;
1043 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1044                         unsigned long size, pgprot_t *vma_prot);
1045
1046 #ifndef CONFIG_X86_ESPFIX64
1047 static inline void init_espfix_bsp(void) { }
1048 #endif
1049
1050 #endif /* !__ASSEMBLY__ */
1051
1052 #ifndef io_remap_pfn_range
1053 #define io_remap_pfn_range remap_pfn_range
1054 #endif
1055
1056 #ifndef has_transparent_hugepage
1057 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1058 #define has_transparent_hugepage() 1
1059 #else
1060 #define has_transparent_hugepage() 0
1061 #endif
1062 #endif
1063
1064 #endif /* _ASM_GENERIC_PGTABLE_H */