pgtable.h 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830
  1. #ifndef _ASM_GENERIC_PGTABLE_H
  2. #define _ASM_GENERIC_PGTABLE_H
  3. #ifndef __ASSEMBLY__
  4. #ifdef CONFIG_MMU
  5. #include <linux/mm_types.h>
  6. #include <linux/bug.h>
  7. /*
  8. * On almost all architectures and configurations, 0 can be used as the
  9. * upper ceiling to free_pgtables(): on many architectures it has the same
  10. * effect as using TASK_SIZE. However, there is one configuration which
  11. * must impose a more careful limit, to avoid freeing kernel pgtables.
  12. */
  13. #ifndef USER_PGTABLES_CEILING
  14. #define USER_PGTABLES_CEILING 0UL
  15. #endif
  16. #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  17. extern int ptep_set_access_flags(struct vm_area_struct *vma,
  18. unsigned long address, pte_t *ptep,
  19. pte_t entry, int dirty);
  20. #endif
  21. #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  22. extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  23. unsigned long address, pmd_t *pmdp,
  24. pmd_t entry, int dirty);
  25. #endif
  26. #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  27. static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  28. unsigned long address,
  29. pte_t *ptep)
  30. {
  31. pte_t pte = *ptep;
  32. int r = 1;
  33. if (!pte_young(pte))
  34. r = 0;
  35. else
  36. set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  37. return r;
  38. }
  39. #endif
  40. #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  41. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  42. static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  43. unsigned long address,
  44. pmd_t *pmdp)
  45. {
  46. pmd_t pmd = *pmdp;
  47. int r = 1;
  48. if (!pmd_young(pmd))
  49. r = 0;
  50. else
  51. set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  52. return r;
  53. }
  54. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  55. static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  56. unsigned long address,
  57. pmd_t *pmdp)
  58. {
  59. BUG();
  60. return 0;
  61. }
  62. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  63. #endif
  64. #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  65. int ptep_clear_flush_young(struct vm_area_struct *vma,
  66. unsigned long address, pte_t *ptep);
  67. #endif
  68. #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  69. int pmdp_clear_flush_young(struct vm_area_struct *vma,
  70. unsigned long address, pmd_t *pmdp);
  71. #endif
  72. #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  73. static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  74. unsigned long address,
  75. pte_t *ptep)
  76. {
  77. pte_t pte = *ptep;
  78. pte_clear(mm, address, ptep);
  79. return pte;
  80. }
  81. #endif
  82. #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
  83. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  84. static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
  85. unsigned long address,
  86. pmd_t *pmdp)
  87. {
  88. pmd_t pmd = *pmdp;
  89. pmd_clear(pmdp);
  90. return pmd;
  91. }
  92. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  93. #endif
  94. #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
  95. static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
  96. unsigned long address, pte_t *ptep,
  97. int full)
  98. {
  99. pte_t pte;
  100. pte = ptep_get_and_clear(mm, address, ptep);
  101. return pte;
  102. }
  103. #endif
  104. /*
  105. * Some architectures may be able to avoid expensive synchronization
  106. * primitives when modifications are made to PTE's which are already
  107. * not present, or in the process of an address space destruction.
  108. */
  109. #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
  110. static inline void pte_clear_not_present_full(struct mm_struct *mm,
  111. unsigned long address,
  112. pte_t *ptep,
  113. int full)
  114. {
  115. pte_clear(mm, address, ptep);
  116. }
  117. #endif
  118. #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
  119. extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
  120. unsigned long address,
  121. pte_t *ptep);
  122. #endif
  123. #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
  124. extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
  125. unsigned long address,
  126. pmd_t *pmdp);
  127. #endif
  128. #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
  129. struct mm_struct;
  130. static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
  131. {
  132. pte_t old_pte = *ptep;
  133. set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
  134. }
  135. #endif
  136. #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
  137. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  138. static inline void pmdp_set_wrprotect(struct mm_struct *mm,
  139. unsigned long address, pmd_t *pmdp)
  140. {
  141. pmd_t old_pmd = *pmdp;
  142. set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
  143. }
  144. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  145. static inline void pmdp_set_wrprotect(struct mm_struct *mm,
  146. unsigned long address, pmd_t *pmdp)
  147. {
  148. BUG();
  149. }
  150. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  151. #endif
  152. #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
  153. extern void pmdp_splitting_flush(struct vm_area_struct *vma,
  154. unsigned long address, pmd_t *pmdp);
  155. #endif
  156. #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
  157. extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
  158. pgtable_t pgtable);
  159. #endif
  160. #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
  161. extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
  162. #endif
  163. #ifndef __HAVE_ARCH_PMDP_INVALIDATE
  164. extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
  165. pmd_t *pmdp);
  166. #endif
  167. #ifndef __HAVE_ARCH_PTE_SAME
  168. static inline int pte_same(pte_t pte_a, pte_t pte_b)
  169. {
  170. return pte_val(pte_a) == pte_val(pte_b);
  171. }
  172. #endif
  173. #ifndef __HAVE_ARCH_PTE_UNUSED
  174. /*
  175. * Some architectures provide facilities to virtualization guests
  176. * so that they can flag allocated pages as unused. This allows the
  177. * host to transparently reclaim unused pages. This function returns
  178. * whether the pte's page is unused.
  179. */
  180. static inline int pte_unused(pte_t pte)
  181. {
  182. return 0;
  183. }
  184. #endif
  185. #ifndef __HAVE_ARCH_PMD_SAME
  186. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  187. static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
  188. {
  189. return pmd_val(pmd_a) == pmd_val(pmd_b);
  190. }
  191. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  192. static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
  193. {
  194. BUG();
  195. return 0;
  196. }
  197. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  198. #endif
  199. #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
  200. #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
  201. #endif
  202. #ifndef __HAVE_ARCH_MOVE_PTE
  203. #define move_pte(pte, prot, old_addr, new_addr) (pte)
  204. #endif
  205. #ifndef pte_accessible
  206. # define pte_accessible(mm, pte) ((void)(pte), 1)
  207. #endif
  208. #ifndef pte_present_nonuma
  209. #define pte_present_nonuma(pte) pte_present(pte)
  210. #endif
  211. #ifndef flush_tlb_fix_spurious_fault
  212. #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
  213. #endif
  214. #ifndef pgprot_noncached
  215. #define pgprot_noncached(prot) (prot)
  216. #endif
  217. #ifndef pgprot_writecombine
  218. #define pgprot_writecombine pgprot_noncached
  219. #endif
  220. /*
  221. * When walking page tables, get the address of the next boundary,
  222. * or the end address of the range if that comes earlier. Although no
  223. * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
  224. */
  225. #define pgd_addr_end(addr, end) \
  226. ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
  227. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  228. })
  229. #ifndef pud_addr_end
  230. #define pud_addr_end(addr, end) \
  231. ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
  232. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  233. })
  234. #endif
  235. #ifndef pmd_addr_end
  236. #define pmd_addr_end(addr, end) \
  237. ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
  238. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  239. })
  240. #endif
  241. /*
  242. * When walking page tables, we usually want to skip any p?d_none entries;
  243. * and any p?d_bad entries - reporting the error before resetting to none.
  244. * Do the tests inline, but report and clear the bad entry in mm/memory.c.
  245. */
  246. void pgd_clear_bad(pgd_t *);
  247. void pud_clear_bad(pud_t *);
  248. void pmd_clear_bad(pmd_t *);
  249. static inline int pgd_none_or_clear_bad(pgd_t *pgd)
  250. {
  251. if (pgd_none(*pgd))
  252. return 1;
  253. if (unlikely(pgd_bad(*pgd))) {
  254. pgd_clear_bad(pgd);
  255. return 1;
  256. }
  257. return 0;
  258. }
  259. static inline int pud_none_or_clear_bad(pud_t *pud)
  260. {
  261. if (pud_none(*pud))
  262. return 1;
  263. if (unlikely(pud_bad(*pud))) {
  264. pud_clear_bad(pud);
  265. return 1;
  266. }
  267. return 0;
  268. }
  269. static inline int pmd_none_or_clear_bad(pmd_t *pmd)
  270. {
  271. if (pmd_none(*pmd))
  272. return 1;
  273. if (unlikely(pmd_bad(*pmd))) {
  274. pmd_clear_bad(pmd);
  275. return 1;
  276. }
  277. return 0;
  278. }
  279. static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
  280. unsigned long addr,
  281. pte_t *ptep)
  282. {
  283. /*
  284. * Get the current pte state, but zero it out to make it
  285. * non-present, preventing the hardware from asynchronously
  286. * updating it.
  287. */
  288. return ptep_get_and_clear(mm, addr, ptep);
  289. }
  290. static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
  291. unsigned long addr,
  292. pte_t *ptep, pte_t pte)
  293. {
  294. /*
  295. * The pte is non-present, so there's no hardware state to
  296. * preserve.
  297. */
  298. set_pte_at(mm, addr, ptep, pte);
  299. }
  300. #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
  301. /*
  302. * Start a pte protection read-modify-write transaction, which
  303. * protects against asynchronous hardware modifications to the pte.
  304. * The intention is not to prevent the hardware from making pte
  305. * updates, but to prevent any updates it may make from being lost.
  306. *
  307. * This does not protect against other software modifications of the
  308. * pte; the appropriate pte lock must be held over the transation.
  309. *
  310. * Note that this interface is intended to be batchable, meaning that
  311. * ptep_modify_prot_commit may not actually update the pte, but merely
  312. * queue the update to be done at some later time. The update must be
  313. * actually committed before the pte lock is released, however.
  314. */
  315. static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
  316. unsigned long addr,
  317. pte_t *ptep)
  318. {
  319. return __ptep_modify_prot_start(mm, addr, ptep);
  320. }
  321. /*
  322. * Commit an update to a pte, leaving any hardware-controlled bits in
  323. * the PTE unmodified.
  324. */
  325. static inline void ptep_modify_prot_commit(struct mm_struct *mm,
  326. unsigned long addr,
  327. pte_t *ptep, pte_t pte)
  328. {
  329. __ptep_modify_prot_commit(mm, addr, ptep, pte);
  330. }
  331. #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
  332. #endif /* CONFIG_MMU */
  333. /*
  334. * A facility to provide lazy MMU batching. This allows PTE updates and
  335. * page invalidations to be delayed until a call to leave lazy MMU mode
  336. * is issued. Some architectures may benefit from doing this, and it is
  337. * beneficial for both shadow and direct mode hypervisors, which may batch
  338. * the PTE updates which happen during this window. Note that using this
  339. * interface requires that read hazards be removed from the code. A read
  340. * hazard could result in the direct mode hypervisor case, since the actual
  341. * write to the page tables may not yet have taken place, so reads though
  342. * a raw PTE pointer after it has been modified are not guaranteed to be
  343. * up to date. This mode can only be entered and left under the protection of
  344. * the page table locks for all page tables which may be modified. In the UP
  345. * case, this is required so that preemption is disabled, and in the SMP case,
  346. * it must synchronize the delayed page table writes properly on other CPUs.
  347. */
  348. #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
  349. #define arch_enter_lazy_mmu_mode() do {} while (0)
  350. #define arch_leave_lazy_mmu_mode() do {} while (0)
  351. #define arch_flush_lazy_mmu_mode() do {} while (0)
  352. #endif
  353. /*
  354. * A facility to provide batching of the reload of page tables and
  355. * other process state with the actual context switch code for
  356. * paravirtualized guests. By convention, only one of the batched
  357. * update (lazy) modes (CPU, MMU) should be active at any given time,
  358. * entry should never be nested, and entry and exits should always be
  359. * paired. This is for sanity of maintaining and reasoning about the
  360. * kernel code. In this case, the exit (end of the context switch) is
  361. * in architecture-specific code, and so doesn't need a generic
  362. * definition.
  363. */
  364. #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
  365. #define arch_start_context_switch(prev) do {} while (0)
  366. #endif
  367. #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
  368. static inline int pte_soft_dirty(pte_t pte)
  369. {
  370. return 0;
  371. }
  372. static inline int pmd_soft_dirty(pmd_t pmd)
  373. {
  374. return 0;
  375. }
  376. static inline pte_t pte_mksoft_dirty(pte_t pte)
  377. {
  378. return pte;
  379. }
  380. static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
  381. {
  382. return pmd;
  383. }
  384. static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
  385. {
  386. return pte;
  387. }
  388. static inline int pte_swp_soft_dirty(pte_t pte)
  389. {
  390. return 0;
  391. }
  392. static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
  393. {
  394. return pte;
  395. }
  396. static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
  397. {
  398. return pte;
  399. }
  400. static inline pte_t pte_file_mksoft_dirty(pte_t pte)
  401. {
  402. return pte;
  403. }
  404. static inline int pte_file_soft_dirty(pte_t pte)
  405. {
  406. return 0;
  407. }
  408. #endif
  409. #ifndef __HAVE_PFNMAP_TRACKING
  410. /*
  411. * Interfaces that can be used by architecture code to keep track of
  412. * memory type of pfn mappings specified by the remap_pfn_range,
  413. * vm_insert_pfn.
  414. */
  415. /*
  416. * track_pfn_remap is called when a _new_ pfn mapping is being established
  417. * by remap_pfn_range() for physical range indicated by pfn and size.
  418. */
  419. static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
  420. unsigned long pfn, unsigned long addr,
  421. unsigned long size)
  422. {
  423. return 0;
  424. }
  425. /*
  426. * track_pfn_insert is called when a _new_ single pfn is established
  427. * by vm_insert_pfn().
  428. */
  429. static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
  430. unsigned long pfn)
  431. {
  432. return 0;
  433. }
  434. /*
  435. * track_pfn_copy is called when vma that is covering the pfnmap gets
  436. * copied through copy_page_range().
  437. */
  438. static inline int track_pfn_copy(struct vm_area_struct *vma)
  439. {
  440. return 0;
  441. }
  442. /*
  443. * untrack_pfn_vma is called while unmapping a pfnmap for a region.
  444. * untrack can be called for a specific region indicated by pfn and size or
  445. * can be for the entire vma (in which case pfn, size are zero).
  446. */
  447. static inline void untrack_pfn(struct vm_area_struct *vma,
  448. unsigned long pfn, unsigned long size)
  449. {
  450. }
  451. #else
  452. extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
  453. unsigned long pfn, unsigned long addr,
  454. unsigned long size);
  455. extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
  456. unsigned long pfn);
  457. extern int track_pfn_copy(struct vm_area_struct *vma);
  458. extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
  459. unsigned long size);
  460. #endif
  461. #ifdef __HAVE_COLOR_ZERO_PAGE
  462. static inline int is_zero_pfn(unsigned long pfn)
  463. {
  464. extern unsigned long zero_pfn;
  465. unsigned long offset_from_zero_pfn = pfn - zero_pfn;
  466. return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
  467. }
  468. #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
  469. #else
  470. static inline int is_zero_pfn(unsigned long pfn)
  471. {
  472. extern unsigned long zero_pfn;
  473. return pfn == zero_pfn;
  474. }
  475. static inline unsigned long my_zero_pfn(unsigned long addr)
  476. {
  477. extern unsigned long zero_pfn;
  478. return zero_pfn;
  479. }
  480. #endif
  481. #ifdef CONFIG_MMU
  482. #ifndef CONFIG_TRANSPARENT_HUGEPAGE
  483. static inline int pmd_trans_huge(pmd_t pmd)
  484. {
  485. return 0;
  486. }
  487. static inline int pmd_trans_splitting(pmd_t pmd)
  488. {
  489. return 0;
  490. }
  491. #ifndef __HAVE_ARCH_PMD_WRITE
  492. static inline int pmd_write(pmd_t pmd)
  493. {
  494. BUG();
  495. return 0;
  496. }
  497. #endif /* __HAVE_ARCH_PMD_WRITE */
  498. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  499. #ifndef pmd_read_atomic
  500. static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
  501. {
  502. /*
  503. * Depend on compiler for an atomic pmd read. NOTE: this is
  504. * only going to work, if the pmdval_t isn't larger than
  505. * an unsigned long.
  506. */
  507. return *pmdp;
  508. }
  509. #endif
  510. #ifndef pmd_move_must_withdraw
  511. static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
  512. spinlock_t *old_pmd_ptl)
  513. {
  514. /*
  515. * With split pmd lock we also need to move preallocated
  516. * PTE page table if new_pmd is on different PMD page table.
  517. */
  518. return new_pmd_ptl != old_pmd_ptl;
  519. }
  520. #endif
  521. /*
  522. * This function is meant to be used by sites walking pagetables with
  523. * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
  524. * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
  525. * into a null pmd and the transhuge page fault can convert a null pmd
  526. * into an hugepmd or into a regular pmd (if the hugepage allocation
  527. * fails). While holding the mmap_sem in read mode the pmd becomes
  528. * stable and stops changing under us only if it's not null and not a
  529. * transhuge pmd. When those races occurs and this function makes a
  530. * difference vs the standard pmd_none_or_clear_bad, the result is
  531. * undefined so behaving like if the pmd was none is safe (because it
  532. * can return none anyway). The compiler level barrier() is critically
  533. * important to compute the two checks atomically on the same pmdval.
  534. *
  535. * For 32bit kernels with a 64bit large pmd_t this automatically takes
  536. * care of reading the pmd atomically to avoid SMP race conditions
  537. * against pmd_populate() when the mmap_sem is hold for reading by the
  538. * caller (a special atomic read not done by "gcc" as in the generic
  539. * version above, is also needed when THP is disabled because the page
  540. * fault can populate the pmd from under us).
  541. */
  542. static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
  543. {
  544. pmd_t pmdval = pmd_read_atomic(pmd);
  545. /*
  546. * The barrier will stabilize the pmdval in a register or on
  547. * the stack so that it will stop changing under the code.
  548. *
  549. * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
  550. * pmd_read_atomic is allowed to return a not atomic pmdval
  551. * (for example pointing to an hugepage that has never been
  552. * mapped in the pmd). The below checks will only care about
  553. * the low part of the pmd with 32bit PAE x86 anyway, with the
  554. * exception of pmd_none(). So the important thing is that if
  555. * the low part of the pmd is found null, the high part will
  556. * be also null or the pmd_none() check below would be
  557. * confused.
  558. */
  559. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  560. barrier();
  561. #endif
  562. if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
  563. return 1;
  564. if (unlikely(pmd_bad(pmdval))) {
  565. pmd_clear_bad(pmd);
  566. return 1;
  567. }
  568. return 0;
  569. }
  570. /*
  571. * This is a noop if Transparent Hugepage Support is not built into
  572. * the kernel. Otherwise it is equivalent to
  573. * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
  574. * places that already verified the pmd is not none and they want to
  575. * walk ptes while holding the mmap sem in read mode (write mode don't
  576. * need this). If THP is not enabled, the pmd can't go away under the
  577. * code even if MADV_DONTNEED runs, but if THP is enabled we need to
  578. * run a pmd_trans_unstable before walking the ptes after
  579. * split_huge_page_pmd returns (because it may have run when the pmd
  580. * become null, but then a page fault can map in a THP and not a
  581. * regular page).
  582. */
  583. static inline int pmd_trans_unstable(pmd_t *pmd)
  584. {
  585. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  586. return pmd_none_or_trans_huge_or_clear_bad(pmd);
  587. #else
  588. return 0;
  589. #endif
  590. }
  591. #ifdef CONFIG_NUMA_BALANCING
  592. #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
  593. /*
  594. * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
  595. * same bit too). It's set only when _PAGE_PRESET is not set and it's
  596. * never set if _PAGE_PRESENT is set.
  597. *
  598. * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
  599. * fault triggers on those regions if pte/pmd_numa returns true
  600. * (because _PAGE_PRESENT is not set).
  601. */
  602. #ifndef pte_numa
  603. static inline int pte_numa(pte_t pte)
  604. {
  605. return (pte_flags(pte) &
  606. (_PAGE_NUMA|_PAGE_PROTNONE|_PAGE_PRESENT)) == _PAGE_NUMA;
  607. }
  608. #endif
  609. #ifndef pmd_numa
  610. static inline int pmd_numa(pmd_t pmd)
  611. {
  612. return (pmd_flags(pmd) &
  613. (_PAGE_NUMA|_PAGE_PROTNONE|_PAGE_PRESENT)) == _PAGE_NUMA;
  614. }
  615. #endif
  616. /*
  617. * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
  618. * because they're called by the NUMA hinting minor page fault. If we
  619. * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
  620. * would be forced to set it later while filling the TLB after we
  621. * return to userland. That would trigger a second write to memory
  622. * that we optimize away by setting _PAGE_ACCESSED here.
  623. */
  624. #ifndef pte_mknonnuma
  625. static inline pte_t pte_mknonnuma(pte_t pte)
  626. {
  627. pteval_t val = pte_val(pte);
  628. val &= ~_PAGE_NUMA;
  629. val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
  630. return __pte(val);
  631. }
  632. #endif
  633. #ifndef pmd_mknonnuma
  634. static inline pmd_t pmd_mknonnuma(pmd_t pmd)
  635. {
  636. pmdval_t val = pmd_val(pmd);
  637. val &= ~_PAGE_NUMA;
  638. val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
  639. return __pmd(val);
  640. }
  641. #endif
  642. #ifndef pte_mknuma
  643. static inline pte_t pte_mknuma(pte_t pte)
  644. {
  645. pteval_t val = pte_val(pte);
  646. val &= ~_PAGE_PRESENT;
  647. val |= _PAGE_NUMA;
  648. return __pte(val);
  649. }
  650. #endif
  651. #ifndef ptep_set_numa
  652. static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
  653. pte_t *ptep)
  654. {
  655. pte_t ptent = *ptep;
  656. ptent = pte_mknuma(ptent);
  657. set_pte_at(mm, addr, ptep, ptent);
  658. return;
  659. }
  660. #endif
  661. #ifndef pmd_mknuma
  662. static inline pmd_t pmd_mknuma(pmd_t pmd)
  663. {
  664. pmdval_t val = pmd_val(pmd);
  665. val &= ~_PAGE_PRESENT;
  666. val |= _PAGE_NUMA;
  667. return __pmd(val);
  668. }
  669. #endif
  670. #ifndef pmdp_set_numa
  671. static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
  672. pmd_t *pmdp)
  673. {
  674. pmd_t pmd = *pmdp;
  675. pmd = pmd_mknuma(pmd);
  676. set_pmd_at(mm, addr, pmdp, pmd);
  677. return;
  678. }
  679. #endif
  680. #else
  681. extern int pte_numa(pte_t pte);
  682. extern int pmd_numa(pmd_t pmd);
  683. extern pte_t pte_mknonnuma(pte_t pte);
  684. extern pmd_t pmd_mknonnuma(pmd_t pmd);
  685. extern pte_t pte_mknuma(pte_t pte);
  686. extern pmd_t pmd_mknuma(pmd_t pmd);
  687. extern void ptep_set_numa(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
  688. extern void pmdp_set_numa(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp);
  689. #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
  690. #else
  691. static inline int pmd_numa(pmd_t pmd)
  692. {
  693. return 0;
  694. }
  695. static inline int pte_numa(pte_t pte)
  696. {
  697. return 0;
  698. }
  699. static inline pte_t pte_mknonnuma(pte_t pte)
  700. {
  701. return pte;
  702. }
  703. static inline pmd_t pmd_mknonnuma(pmd_t pmd)
  704. {
  705. return pmd;
  706. }
  707. static inline pte_t pte_mknuma(pte_t pte)
  708. {
  709. return pte;
  710. }
  711. static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
  712. pte_t *ptep)
  713. {
  714. return;
  715. }
  716. static inline pmd_t pmd_mknuma(pmd_t pmd)
  717. {
  718. return pmd;
  719. }
  720. static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
  721. pmd_t *pmdp)
  722. {
  723. return ;
  724. }
  725. #endif /* CONFIG_NUMA_BALANCING */
  726. #endif /* CONFIG_MMU */
  727. #endif /* !__ASSEMBLY__ */
  728. #ifndef io_remap_pfn_range
  729. #define io_remap_pfn_range remap_pfn_range
  730. #endif
  731. #endif /* _ASM_GENERIC_PGTABLE_H */