Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : #ifndef _LINUX_PGTABLE_H
3 : #define _LINUX_PGTABLE_H
4 :
5 : #include <linux/pfn.h>
6 : #include <asm/pgtable.h>
7 :
8 : #ifndef __ASSEMBLY__
9 : #ifdef CONFIG_MMU
10 :
11 : #include <linux/mm_types.h>
12 : #include <linux/bug.h>
13 : #include <linux/errno.h>
14 : #include <asm-generic/pgtable_uffd.h>
15 : #include <linux/page_table_check.h>
16 :
17 : #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
18 : defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
19 : #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
20 : #endif
21 :
22 : /*
23 : * On almost all architectures and configurations, 0 can be used as the
24 : * upper ceiling to free_pgtables(): on many architectures it has the same
25 : * effect as using TASK_SIZE. However, there is one configuration which
26 : * must impose a more careful limit, to avoid freeing kernel pgtables.
27 : */
28 : #ifndef USER_PGTABLES_CEILING
29 : #define USER_PGTABLES_CEILING 0UL
30 : #endif
31 :
32 : /*
33 : * This defines the first usable user address. Platforms
34 : * can override its value with custom FIRST_USER_ADDRESS
35 : * defined in their respective <asm/pgtable.h>.
36 : */
37 : #ifndef FIRST_USER_ADDRESS
38 : #define FIRST_USER_ADDRESS 0UL
39 : #endif
40 :
41 : /*
42 : * This defines the generic helper for accessing PMD page
43 : * table page. Although platforms can still override this
44 : * via their respective <asm/pgtable.h>.
45 : */
46 : #ifndef pmd_pgtable
47 : #define pmd_pgtable(pmd) pmd_page(pmd)
48 : #endif
49 :
50 : /*
51 : * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
52 : *
53 : * The pXx_index() functions return the index of the entry in the page
54 : * table page which would control the given virtual address
55 : *
56 : * As these functions may be used by the same code for different levels of
57 : * the page table folding, they are always available, regardless of
58 : * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
59 : * because in such cases PTRS_PER_PxD equals 1.
60 : */
61 :
62 : static inline unsigned long pte_index(unsigned long address)
63 : {
64 : return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
65 : }
66 : #define pte_index pte_index
67 :
68 : #ifndef pmd_index
69 : static inline unsigned long pmd_index(unsigned long address)
70 : {
71 : return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
72 : }
73 : #define pmd_index pmd_index
74 : #endif
75 :
76 : #ifndef pud_index
77 : static inline unsigned long pud_index(unsigned long address)
78 : {
79 : return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
80 : }
81 : #define pud_index pud_index
82 : #endif
83 :
84 : #ifndef pgd_index
85 : /* Must be a compile-time constant, so implement it as a macro */
86 : #define pgd_index(a) (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
87 : #endif
88 :
89 : #ifndef pte_offset_kernel
90 : static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
91 : {
92 : return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
93 : }
94 : #define pte_offset_kernel pte_offset_kernel
95 : #endif
96 :
97 : #ifdef CONFIG_HIGHPTE
98 : #define __pte_map(pmd, address) \
99 : ((pte_t *)kmap_local_page(pmd_page(*(pmd))) + pte_index((address)))
100 : #define pte_unmap(pte) do { \
101 : kunmap_local((pte)); \
102 : /* rcu_read_unlock() to be added later */ \
103 : } while (0)
104 : #else
105 : static inline pte_t *__pte_map(pmd_t *pmd, unsigned long address)
106 : {
107 : return pte_offset_kernel(pmd, address);
108 : }
109 : static inline void pte_unmap(pte_t *pte)
110 : {
111 : /* rcu_read_unlock() to be added later */
112 : }
113 : #endif
114 :
115 : /* Find an entry in the second-level page table.. */
116 : #ifndef pmd_offset
117 : static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
118 : {
119 : return pud_pgtable(*pud) + pmd_index(address);
120 : }
121 : #define pmd_offset pmd_offset
122 : #endif
123 :
124 : #ifndef pud_offset
125 : static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
126 : {
127 : return p4d_pgtable(*p4d) + pud_index(address);
128 : }
129 : #define pud_offset pud_offset
130 : #endif
131 :
132 : static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
133 : {
134 : return (pgd + pgd_index(address));
135 : };
136 :
137 : /*
138 : * a shortcut to get a pgd_t in a given mm
139 : */
140 : #ifndef pgd_offset
141 : #define pgd_offset(mm, address) pgd_offset_pgd((mm)->pgd, (address))
142 : #endif
143 :
144 : /*
145 : * a shortcut which implies the use of the kernel's pgd, instead
146 : * of a process's
147 : */
148 : #ifndef pgd_offset_k
149 : #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
150 : #endif
151 :
152 : /*
153 : * In many cases it is known that a virtual address is mapped at PMD or PTE
154 : * level, so instead of traversing all the page table levels, we can get a
155 : * pointer to the PMD entry in user or kernel page table or translate a virtual
156 : * address to the pointer in the PTE in the kernel page tables with simple
157 : * helpers.
158 : */
159 : static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
160 : {
161 : return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
162 : }
163 :
164 : static inline pmd_t *pmd_off_k(unsigned long va)
165 : {
166 : return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
167 : }
168 :
169 : static inline pte_t *virt_to_kpte(unsigned long vaddr)
170 : {
171 : pmd_t *pmd = pmd_off_k(vaddr);
172 :
173 : return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
174 : }
175 :
176 : #ifndef pmd_young
177 : static inline int pmd_young(pmd_t pmd)
178 : {
179 : return 0;
180 : }
181 : #endif
182 :
183 : #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
184 : extern int ptep_set_access_flags(struct vm_area_struct *vma,
185 : unsigned long address, pte_t *ptep,
186 : pte_t entry, int dirty);
187 : #endif
188 :
189 : #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
190 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
191 : extern int pmdp_set_access_flags(struct vm_area_struct *vma,
192 : unsigned long address, pmd_t *pmdp,
193 : pmd_t entry, int dirty);
194 : extern int pudp_set_access_flags(struct vm_area_struct *vma,
195 : unsigned long address, pud_t *pudp,
196 : pud_t entry, int dirty);
197 : #else
198 : static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
199 : unsigned long address, pmd_t *pmdp,
200 : pmd_t entry, int dirty)
201 : {
202 : BUILD_BUG();
203 : return 0;
204 : }
205 : static inline int pudp_set_access_flags(struct vm_area_struct *vma,
206 : unsigned long address, pud_t *pudp,
207 : pud_t entry, int dirty)
208 : {
209 : BUILD_BUG();
210 : return 0;
211 : }
212 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
213 : #endif
214 :
215 : #ifndef ptep_get
216 : static inline pte_t ptep_get(pte_t *ptep)
217 : {
218 0 : return READ_ONCE(*ptep);
219 : }
220 : #endif
221 :
222 : #ifndef pmdp_get
223 : static inline pmd_t pmdp_get(pmd_t *pmdp)
224 : {
225 : return READ_ONCE(*pmdp);
226 : }
227 : #endif
228 :
229 : #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
230 : static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
231 : unsigned long address,
232 : pte_t *ptep)
233 : {
234 : pte_t pte = ptep_get(ptep);
235 : int r = 1;
236 : if (!pte_young(pte))
237 : r = 0;
238 : else
239 : set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
240 : return r;
241 : }
242 : #endif
243 :
244 : #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
245 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
246 : static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
247 : unsigned long address,
248 : pmd_t *pmdp)
249 : {
250 : pmd_t pmd = *pmdp;
251 : int r = 1;
252 : if (!pmd_young(pmd))
253 : r = 0;
254 : else
255 : set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
256 : return r;
257 : }
258 : #else
259 : static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
260 : unsigned long address,
261 : pmd_t *pmdp)
262 : {
263 : BUILD_BUG();
264 : return 0;
265 : }
266 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
267 : #endif
268 :
269 : #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
270 : int ptep_clear_flush_young(struct vm_area_struct *vma,
271 : unsigned long address, pte_t *ptep);
272 : #endif
273 :
274 : #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
275 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
276 : extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
277 : unsigned long address, pmd_t *pmdp);
278 : #else
279 : /*
280 : * Despite relevant to THP only, this API is called from generic rmap code
281 : * under PageTransHuge(), hence needs a dummy implementation for !THP
282 : */
283 : static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
284 : unsigned long address, pmd_t *pmdp)
285 : {
286 : BUILD_BUG();
287 : return 0;
288 : }
289 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
290 : #endif
291 :
292 : #ifndef arch_has_hw_nonleaf_pmd_young
293 : /*
294 : * Return whether the accessed bit in non-leaf PMD entries is supported on the
295 : * local CPU.
296 : */
297 : static inline bool arch_has_hw_nonleaf_pmd_young(void)
298 : {
299 : return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
300 : }
301 : #endif
302 :
303 : #ifndef arch_has_hw_pte_young
304 : /*
305 : * Return whether the accessed bit is supported on the local CPU.
306 : *
307 : * This stub assumes accessing through an old PTE triggers a page fault.
308 : * Architectures that automatically set the access bit should overwrite it.
309 : */
310 : static inline bool arch_has_hw_pte_young(void)
311 : {
312 : return false;
313 : }
314 : #endif
315 :
316 : #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
317 : static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
318 : unsigned long address,
319 : pte_t *ptep)
320 : {
321 : pte_t pte = ptep_get(ptep);
322 : pte_clear(mm, address, ptep);
323 : page_table_check_pte_clear(mm, address, pte);
324 : return pte;
325 : }
326 : #endif
327 :
328 : static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
329 : pte_t *ptep)
330 : {
331 : ptep_get_and_clear(mm, addr, ptep);
332 : }
333 :
334 : #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
335 : /*
336 : * For walking the pagetables without holding any locks. Some architectures
337 : * (eg x86-32 PAE) cannot load the entries atomically without using expensive
338 : * instructions. We are guaranteed that a PTE will only either go from not
339 : * present to present, or present to not present -- it will not switch to a
340 : * completely different present page without a TLB flush inbetween; which we
341 : * are blocking by holding interrupts off.
342 : *
343 : * Setting ptes from not present to present goes:
344 : *
345 : * ptep->pte_high = h;
346 : * smp_wmb();
347 : * ptep->pte_low = l;
348 : *
349 : * And present to not present goes:
350 : *
351 : * ptep->pte_low = 0;
352 : * smp_wmb();
353 : * ptep->pte_high = 0;
354 : *
355 : * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
356 : * We load pte_high *after* loading pte_low, which ensures we don't see an older
357 : * value of pte_high. *Then* we recheck pte_low, which ensures that we haven't
358 : * picked up a changed pte high. We might have gotten rubbish values from
359 : * pte_low and pte_high, but we are guaranteed that pte_low will not have the
360 : * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
361 : * operates on present ptes we're safe.
362 : */
363 : static inline pte_t ptep_get_lockless(pte_t *ptep)
364 : {
365 : pte_t pte;
366 :
367 : do {
368 : pte.pte_low = ptep->pte_low;
369 : smp_rmb();
370 : pte.pte_high = ptep->pte_high;
371 : smp_rmb();
372 : } while (unlikely(pte.pte_low != ptep->pte_low));
373 :
374 : return pte;
375 : }
376 : #define ptep_get_lockless ptep_get_lockless
377 :
378 : #if CONFIG_PGTABLE_LEVELS > 2
379 : static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
380 : {
381 : pmd_t pmd;
382 :
383 : do {
384 : pmd.pmd_low = pmdp->pmd_low;
385 : smp_rmb();
386 : pmd.pmd_high = pmdp->pmd_high;
387 : smp_rmb();
388 : } while (unlikely(pmd.pmd_low != pmdp->pmd_low));
389 :
390 : return pmd;
391 : }
392 : #define pmdp_get_lockless pmdp_get_lockless
393 : #endif /* CONFIG_PGTABLE_LEVELS > 2 */
394 : #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
395 :
396 : /*
397 : * We require that the PTE can be read atomically.
398 : */
399 : #ifndef ptep_get_lockless
400 : static inline pte_t ptep_get_lockless(pte_t *ptep)
401 : {
402 : return ptep_get(ptep);
403 : }
404 : #endif
405 :
406 : #ifndef pmdp_get_lockless
407 : static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
408 : {
409 : return pmdp_get(pmdp);
410 : }
411 : #endif
412 :
413 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
414 : #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
415 : static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
416 : unsigned long address,
417 : pmd_t *pmdp)
418 : {
419 : pmd_t pmd = *pmdp;
420 :
421 : pmd_clear(pmdp);
422 : page_table_check_pmd_clear(mm, address, pmd);
423 :
424 : return pmd;
425 : }
426 : #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
427 : #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
428 : static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
429 : unsigned long address,
430 : pud_t *pudp)
431 : {
432 : pud_t pud = *pudp;
433 :
434 : pud_clear(pudp);
435 : page_table_check_pud_clear(mm, address, pud);
436 :
437 : return pud;
438 : }
439 : #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
440 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
441 :
442 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
443 : #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
444 : static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
445 : unsigned long address, pmd_t *pmdp,
446 : int full)
447 : {
448 : return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
449 : }
450 : #endif
451 :
452 : #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
453 : static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
454 : unsigned long address, pud_t *pudp,
455 : int full)
456 : {
457 : return pudp_huge_get_and_clear(mm, address, pudp);
458 : }
459 : #endif
460 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
461 :
462 : #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
463 : static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
464 : unsigned long address, pte_t *ptep,
465 : int full)
466 : {
467 : return ptep_get_and_clear(mm, address, ptep);
468 : }
469 : #endif
470 :
471 :
472 : /*
473 : * If two threads concurrently fault at the same page, the thread that
474 : * won the race updates the PTE and its local TLB/Cache. The other thread
475 : * gives up, simply does nothing, and continues; on architectures where
476 : * software can update TLB, local TLB can be updated here to avoid next page
477 : * fault. This function updates TLB only, do nothing with cache or others.
478 : * It is the difference with function update_mmu_cache.
479 : */
480 : #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
481 : static inline void update_mmu_tlb(struct vm_area_struct *vma,
482 : unsigned long address, pte_t *ptep)
483 : {
484 : }
485 : #define __HAVE_ARCH_UPDATE_MMU_TLB
486 : #endif
487 :
488 : /*
489 : * Some architectures may be able to avoid expensive synchronization
490 : * primitives when modifications are made to PTE's which are already
491 : * not present, or in the process of an address space destruction.
492 : */
493 : #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
494 : static inline void pte_clear_not_present_full(struct mm_struct *mm,
495 : unsigned long address,
496 : pte_t *ptep,
497 : int full)
498 : {
499 : pte_clear(mm, address, ptep);
500 : }
501 : #endif
502 :
503 : #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
504 : extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
505 : unsigned long address,
506 : pte_t *ptep);
507 : #endif
508 :
509 : #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
510 : extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
511 : unsigned long address,
512 : pmd_t *pmdp);
513 : extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
514 : unsigned long address,
515 : pud_t *pudp);
516 : #endif
517 :
518 : #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
519 : struct mm_struct;
520 : static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
521 : {
522 : pte_t old_pte = ptep_get(ptep);
523 : set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
524 : }
525 : #endif
526 :
527 : /*
528 : * On some architectures hardware does not set page access bit when accessing
529 : * memory page, it is responsibility of software setting this bit. It brings
530 : * out extra page fault penalty to track page access bit. For optimization page
531 : * access bit can be set during all page fault flow on these arches.
532 : * To be differentiate with macro pte_mkyoung, this macro is used on platforms
533 : * where software maintains page access bit.
534 : */
535 : #ifndef pte_sw_mkyoung
536 : static inline pte_t pte_sw_mkyoung(pte_t pte)
537 : {
538 : return pte;
539 : }
540 : #define pte_sw_mkyoung pte_sw_mkyoung
541 : #endif
542 :
543 : #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
544 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
545 : static inline void pmdp_set_wrprotect(struct mm_struct *mm,
546 : unsigned long address, pmd_t *pmdp)
547 : {
548 : pmd_t old_pmd = *pmdp;
549 : set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
550 : }
551 : #else
552 : static inline void pmdp_set_wrprotect(struct mm_struct *mm,
553 : unsigned long address, pmd_t *pmdp)
554 : {
555 : BUILD_BUG();
556 : }
557 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
558 : #endif
559 : #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
560 : #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
561 : static inline void pudp_set_wrprotect(struct mm_struct *mm,
562 : unsigned long address, pud_t *pudp)
563 : {
564 : pud_t old_pud = *pudp;
565 :
566 : set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
567 : }
568 : #else
569 : static inline void pudp_set_wrprotect(struct mm_struct *mm,
570 : unsigned long address, pud_t *pudp)
571 : {
572 : BUILD_BUG();
573 : }
574 : #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
575 : #endif
576 :
577 : #ifndef pmdp_collapse_flush
578 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
579 : extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
580 : unsigned long address, pmd_t *pmdp);
581 : #else
582 : static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
583 : unsigned long address,
584 : pmd_t *pmdp)
585 : {
586 : BUILD_BUG();
587 : return *pmdp;
588 : }
589 : #define pmdp_collapse_flush pmdp_collapse_flush
590 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
591 : #endif
592 :
593 : #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
594 : extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
595 : pgtable_t pgtable);
596 : #endif
597 :
598 : #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
599 : extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
600 : #endif
601 :
602 : #ifndef arch_needs_pgtable_deposit
603 : #define arch_needs_pgtable_deposit() (false)
604 : #endif
605 :
606 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
607 : /*
608 : * This is an implementation of pmdp_establish() that is only suitable for an
609 : * architecture that doesn't have hardware dirty/accessed bits. In this case we
610 : * can't race with CPU which sets these bits and non-atomic approach is fine.
611 : */
612 : static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
613 : unsigned long address, pmd_t *pmdp, pmd_t pmd)
614 : {
615 : pmd_t old_pmd = *pmdp;
616 : set_pmd_at(vma->vm_mm, address, pmdp, pmd);
617 : return old_pmd;
618 : }
619 : #endif
620 :
621 : #ifndef __HAVE_ARCH_PMDP_INVALIDATE
622 : extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
623 : pmd_t *pmdp);
624 : #endif
625 :
626 : #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
627 :
628 : /*
629 : * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
630 : * hugepage mapping in the page tables. This function is similar to
631 : * pmdp_invalidate(), but should only be used if the access and dirty bits would
632 : * not be cleared by the software in the new PMD value. The function ensures
633 : * that hardware changes of the access and dirty bits updates would not be lost.
634 : *
635 : * Doing so can allow in certain architectures to avoid a TLB flush in most
636 : * cases. Yet, another TLB flush might be necessary later if the PMD update
637 : * itself requires such flush (e.g., if protection was set to be stricter). Yet,
638 : * even when a TLB flush is needed because of the update, the caller may be able
639 : * to batch these TLB flushing operations, so fewer TLB flush operations are
640 : * needed.
641 : */
642 : extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
643 : unsigned long address, pmd_t *pmdp);
644 : #endif
645 :
646 : #ifndef __HAVE_ARCH_PTE_SAME
647 : static inline int pte_same(pte_t pte_a, pte_t pte_b)
648 : {
649 : return pte_val(pte_a) == pte_val(pte_b);
650 : }
651 : #endif
652 :
653 : #ifndef __HAVE_ARCH_PTE_UNUSED
654 : /*
655 : * Some architectures provide facilities to virtualization guests
656 : * so that they can flag allocated pages as unused. This allows the
657 : * host to transparently reclaim unused pages. This function returns
658 : * whether the pte's page is unused.
659 : */
660 : static inline int pte_unused(pte_t pte)
661 : {
662 : return 0;
663 : }
664 : #endif
665 :
666 : #ifndef pte_access_permitted
667 : #define pte_access_permitted(pte, write) \
668 : (pte_present(pte) && (!(write) || pte_write(pte)))
669 : #endif
670 :
671 : #ifndef pmd_access_permitted
672 : #define pmd_access_permitted(pmd, write) \
673 : (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
674 : #endif
675 :
676 : #ifndef pud_access_permitted
677 : #define pud_access_permitted(pud, write) \
678 : (pud_present(pud) && (!(write) || pud_write(pud)))
679 : #endif
680 :
681 : #ifndef p4d_access_permitted
682 : #define p4d_access_permitted(p4d, write) \
683 : (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
684 : #endif
685 :
686 : #ifndef pgd_access_permitted
687 : #define pgd_access_permitted(pgd, write) \
688 : (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
689 : #endif
690 :
691 : #ifndef __HAVE_ARCH_PMD_SAME
692 : static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
693 : {
694 : return pmd_val(pmd_a) == pmd_val(pmd_b);
695 : }
696 :
697 : static inline int pud_same(pud_t pud_a, pud_t pud_b)
698 : {
699 : return pud_val(pud_a) == pud_val(pud_b);
700 : }
701 : #endif
702 :
703 : #ifndef __HAVE_ARCH_P4D_SAME
704 : static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
705 : {
706 : return p4d_val(p4d_a) == p4d_val(p4d_b);
707 : }
708 : #endif
709 :
710 : #ifndef __HAVE_ARCH_PGD_SAME
711 : static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
712 : {
713 : return pgd_val(pgd_a) == pgd_val(pgd_b);
714 : }
715 : #endif
716 :
717 : /*
718 : * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
719 : * TLB flush will be required as a result of the "set". For example, use
720 : * in scenarios where it is known ahead of time that the routine is
721 : * setting non-present entries, or re-setting an existing entry to the
722 : * same value. Otherwise, use the typical "set" helpers and flush the
723 : * TLB.
724 : */
725 : #define set_pte_safe(ptep, pte) \
726 : ({ \
727 : WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
728 : set_pte(ptep, pte); \
729 : })
730 :
731 : #define set_pmd_safe(pmdp, pmd) \
732 : ({ \
733 : WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
734 : set_pmd(pmdp, pmd); \
735 : })
736 :
737 : #define set_pud_safe(pudp, pud) \
738 : ({ \
739 : WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
740 : set_pud(pudp, pud); \
741 : })
742 :
743 : #define set_p4d_safe(p4dp, p4d) \
744 : ({ \
745 : WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
746 : set_p4d(p4dp, p4d); \
747 : })
748 :
749 : #define set_pgd_safe(pgdp, pgd) \
750 : ({ \
751 : WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
752 : set_pgd(pgdp, pgd); \
753 : })
754 :
755 : #ifndef __HAVE_ARCH_DO_SWAP_PAGE
756 : /*
757 : * Some architectures support metadata associated with a page. When a
758 : * page is being swapped out, this metadata must be saved so it can be
759 : * restored when the page is swapped back in. SPARC M7 and newer
760 : * processors support an ADI (Application Data Integrity) tag for the
761 : * page as metadata for the page. arch_do_swap_page() can restore this
762 : * metadata when a page is swapped back in.
763 : */
764 : static inline void arch_do_swap_page(struct mm_struct *mm,
765 : struct vm_area_struct *vma,
766 : unsigned long addr,
767 : pte_t pte, pte_t oldpte)
768 : {
769 :
770 : }
771 : #endif
772 :
773 : #ifndef __HAVE_ARCH_UNMAP_ONE
774 : /*
775 : * Some architectures support metadata associated with a page. When a
776 : * page is being swapped out, this metadata must be saved so it can be
777 : * restored when the page is swapped back in. SPARC M7 and newer
778 : * processors support an ADI (Application Data Integrity) tag for the
779 : * page as metadata for the page. arch_unmap_one() can save this
780 : * metadata on a swap-out of a page.
781 : */
782 : static inline int arch_unmap_one(struct mm_struct *mm,
783 : struct vm_area_struct *vma,
784 : unsigned long addr,
785 : pte_t orig_pte)
786 : {
787 : return 0;
788 : }
789 : #endif
790 :
791 : /*
792 : * Allow architectures to preserve additional metadata associated with
793 : * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
794 : * prototypes must be defined in the arch-specific asm/pgtable.h file.
795 : */
796 : #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
797 : static inline int arch_prepare_to_swap(struct page *page)
798 : {
799 : return 0;
800 : }
801 : #endif
802 :
803 : #ifndef __HAVE_ARCH_SWAP_INVALIDATE
804 : static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
805 : {
806 : }
807 :
808 : static inline void arch_swap_invalidate_area(int type)
809 : {
810 : }
811 : #endif
812 :
813 : #ifndef __HAVE_ARCH_SWAP_RESTORE
814 : static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
815 : {
816 : }
817 : #endif
818 :
819 : #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
820 : #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
821 : #endif
822 :
823 : #ifndef __HAVE_ARCH_MOVE_PTE
824 : #define move_pte(pte, prot, old_addr, new_addr) (pte)
825 : #endif
826 :
827 : #ifndef pte_accessible
828 : # define pte_accessible(mm, pte) ((void)(pte), 1)
829 : #endif
830 :
831 : #ifndef flush_tlb_fix_spurious_fault
832 : #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
833 : #endif
834 :
835 : /*
836 : * When walking page tables, get the address of the next boundary,
837 : * or the end address of the range if that comes earlier. Although no
838 : * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
839 : */
840 :
841 : #define pgd_addr_end(addr, end) \
842 : ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
843 : (__boundary - 1 < (end) - 1)? __boundary: (end); \
844 : })
845 :
846 : #ifndef p4d_addr_end
847 : #define p4d_addr_end(addr, end) \
848 : ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
849 : (__boundary - 1 < (end) - 1)? __boundary: (end); \
850 : })
851 : #endif
852 :
853 : #ifndef pud_addr_end
854 : #define pud_addr_end(addr, end) \
855 : ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
856 : (__boundary - 1 < (end) - 1)? __boundary: (end); \
857 : })
858 : #endif
859 :
860 : #ifndef pmd_addr_end
861 : #define pmd_addr_end(addr, end) \
862 : ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
863 : (__boundary - 1 < (end) - 1)? __boundary: (end); \
864 : })
865 : #endif
866 :
867 : /*
868 : * When walking page tables, we usually want to skip any p?d_none entries;
869 : * and any p?d_bad entries - reporting the error before resetting to none.
870 : * Do the tests inline, but report and clear the bad entry in mm/memory.c.
871 : */
872 : void pgd_clear_bad(pgd_t *);
873 :
874 : #ifndef __PAGETABLE_P4D_FOLDED
875 : void p4d_clear_bad(p4d_t *);
876 : #else
877 : #define p4d_clear_bad(p4d) do { } while (0)
878 : #endif
879 :
880 : #ifndef __PAGETABLE_PUD_FOLDED
881 : void pud_clear_bad(pud_t *);
882 : #else
883 : #define pud_clear_bad(p4d) do { } while (0)
884 : #endif
885 :
886 : void pmd_clear_bad(pmd_t *);
887 :
888 : static inline int pgd_none_or_clear_bad(pgd_t *pgd)
889 : {
890 : if (pgd_none(*pgd))
891 : return 1;
892 : if (unlikely(pgd_bad(*pgd))) {
893 : pgd_clear_bad(pgd);
894 : return 1;
895 : }
896 : return 0;
897 : }
898 :
899 : static inline int p4d_none_or_clear_bad(p4d_t *p4d)
900 : {
901 : if (p4d_none(*p4d))
902 : return 1;
903 : if (unlikely(p4d_bad(*p4d))) {
904 : p4d_clear_bad(p4d);
905 : return 1;
906 : }
907 : return 0;
908 : }
909 :
910 : static inline int pud_none_or_clear_bad(pud_t *pud)
911 : {
912 : if (pud_none(*pud))
913 : return 1;
914 : if (unlikely(pud_bad(*pud))) {
915 : pud_clear_bad(pud);
916 : return 1;
917 : }
918 : return 0;
919 : }
920 :
921 : static inline int pmd_none_or_clear_bad(pmd_t *pmd)
922 : {
923 : if (pmd_none(*pmd))
924 : return 1;
925 : if (unlikely(pmd_bad(*pmd))) {
926 : pmd_clear_bad(pmd);
927 : return 1;
928 : }
929 : return 0;
930 : }
931 :
932 : static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
933 : unsigned long addr,
934 : pte_t *ptep)
935 : {
936 : /*
937 : * Get the current pte state, but zero it out to make it
938 : * non-present, preventing the hardware from asynchronously
939 : * updating it.
940 : */
941 : return ptep_get_and_clear(vma->vm_mm, addr, ptep);
942 : }
943 :
944 : static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
945 : unsigned long addr,
946 : pte_t *ptep, pte_t pte)
947 : {
948 : /*
949 : * The pte is non-present, so there's no hardware state to
950 : * preserve.
951 : */
952 : set_pte_at(vma->vm_mm, addr, ptep, pte);
953 : }
954 :
955 : #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
956 : /*
957 : * Start a pte protection read-modify-write transaction, which
958 : * protects against asynchronous hardware modifications to the pte.
959 : * The intention is not to prevent the hardware from making pte
960 : * updates, but to prevent any updates it may make from being lost.
961 : *
962 : * This does not protect against other software modifications of the
963 : * pte; the appropriate pte lock must be held over the transaction.
964 : *
965 : * Note that this interface is intended to be batchable, meaning that
966 : * ptep_modify_prot_commit may not actually update the pte, but merely
967 : * queue the update to be done at some later time. The update must be
968 : * actually committed before the pte lock is released, however.
969 : */
970 : static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
971 : unsigned long addr,
972 : pte_t *ptep)
973 : {
974 : return __ptep_modify_prot_start(vma, addr, ptep);
975 : }
976 :
977 : /*
978 : * Commit an update to a pte, leaving any hardware-controlled bits in
979 : * the PTE unmodified.
980 : */
981 : static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
982 : unsigned long addr,
983 : pte_t *ptep, pte_t old_pte, pte_t pte)
984 : {
985 : __ptep_modify_prot_commit(vma, addr, ptep, pte);
986 : }
987 : #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
988 : #endif /* CONFIG_MMU */
989 :
990 : /*
991 : * No-op macros that just return the current protection value. Defined here
992 : * because these macros can be used even if CONFIG_MMU is not defined.
993 : */
994 :
995 : #ifndef pgprot_nx
996 : #define pgprot_nx(prot) (prot)
997 : #endif
998 :
999 : #ifndef pgprot_noncached
1000 : #define pgprot_noncached(prot) (prot)
1001 : #endif
1002 :
1003 : #ifndef pgprot_writecombine
1004 : #define pgprot_writecombine pgprot_noncached
1005 : #endif
1006 :
1007 : #ifndef pgprot_writethrough
1008 : #define pgprot_writethrough pgprot_noncached
1009 : #endif
1010 :
1011 : #ifndef pgprot_device
1012 : #define pgprot_device pgprot_noncached
1013 : #endif
1014 :
1015 : #ifndef pgprot_mhp
1016 : #define pgprot_mhp(prot) (prot)
1017 : #endif
1018 :
1019 : #ifdef CONFIG_MMU
1020 : #ifndef pgprot_modify
1021 : #define pgprot_modify pgprot_modify
1022 : static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1023 : {
1024 : if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1025 : newprot = pgprot_noncached(newprot);
1026 : if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1027 : newprot = pgprot_writecombine(newprot);
1028 : if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1029 : newprot = pgprot_device(newprot);
1030 : return newprot;
1031 : }
1032 : #endif
1033 : #endif /* CONFIG_MMU */
1034 :
1035 : #ifndef pgprot_encrypted
1036 : #define pgprot_encrypted(prot) (prot)
1037 : #endif
1038 :
1039 : #ifndef pgprot_decrypted
1040 : #define pgprot_decrypted(prot) (prot)
1041 : #endif
1042 :
1043 : /*
1044 : * A facility to provide lazy MMU batching. This allows PTE updates and
1045 : * page invalidations to be delayed until a call to leave lazy MMU mode
1046 : * is issued. Some architectures may benefit from doing this, and it is
1047 : * beneficial for both shadow and direct mode hypervisors, which may batch
1048 : * the PTE updates which happen during this window. Note that using this
1049 : * interface requires that read hazards be removed from the code. A read
1050 : * hazard could result in the direct mode hypervisor case, since the actual
1051 : * write to the page tables may not yet have taken place, so reads though
1052 : * a raw PTE pointer after it has been modified are not guaranteed to be
1053 : * up to date. This mode can only be entered and left under the protection of
1054 : * the page table locks for all page tables which may be modified. In the UP
1055 : * case, this is required so that preemption is disabled, and in the SMP case,
1056 : * it must synchronize the delayed page table writes properly on other CPUs.
1057 : */
1058 : #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1059 : #define arch_enter_lazy_mmu_mode() do {} while (0)
1060 : #define arch_leave_lazy_mmu_mode() do {} while (0)
1061 : #define arch_flush_lazy_mmu_mode() do {} while (0)
1062 : #endif
1063 :
1064 : /*
1065 : * A facility to provide batching of the reload of page tables and
1066 : * other process state with the actual context switch code for
1067 : * paravirtualized guests. By convention, only one of the batched
1068 : * update (lazy) modes (CPU, MMU) should be active at any given time,
1069 : * entry should never be nested, and entry and exits should always be
1070 : * paired. This is for sanity of maintaining and reasoning about the
1071 : * kernel code. In this case, the exit (end of the context switch) is
1072 : * in architecture-specific code, and so doesn't need a generic
1073 : * definition.
1074 : */
1075 : #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1076 : #define arch_start_context_switch(prev) do {} while (0)
1077 : #endif
1078 :
1079 : #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1080 : #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1081 : static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1082 : {
1083 : return pmd;
1084 : }
1085 :
1086 : static inline int pmd_swp_soft_dirty(pmd_t pmd)
1087 : {
1088 : return 0;
1089 : }
1090 :
1091 : static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1092 : {
1093 : return pmd;
1094 : }
1095 : #endif
1096 : #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1097 : static inline int pte_soft_dirty(pte_t pte)
1098 : {
1099 : return 0;
1100 : }
1101 :
1102 : static inline int pmd_soft_dirty(pmd_t pmd)
1103 : {
1104 : return 0;
1105 : }
1106 :
1107 : static inline pte_t pte_mksoft_dirty(pte_t pte)
1108 : {
1109 : return pte;
1110 : }
1111 :
1112 : static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1113 : {
1114 : return pmd;
1115 : }
1116 :
1117 : static inline pte_t pte_clear_soft_dirty(pte_t pte)
1118 : {
1119 : return pte;
1120 : }
1121 :
1122 : static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1123 : {
1124 : return pmd;
1125 : }
1126 :
1127 : static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1128 : {
1129 : return pte;
1130 : }
1131 :
1132 : static inline int pte_swp_soft_dirty(pte_t pte)
1133 : {
1134 : return 0;
1135 : }
1136 :
1137 : static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1138 : {
1139 : return pte;
1140 : }
1141 :
1142 : static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1143 : {
1144 : return pmd;
1145 : }
1146 :
1147 : static inline int pmd_swp_soft_dirty(pmd_t pmd)
1148 : {
1149 : return 0;
1150 : }
1151 :
1152 : static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1153 : {
1154 : return pmd;
1155 : }
1156 : #endif
1157 :
1158 : #ifndef __HAVE_PFNMAP_TRACKING
1159 : /*
1160 : * Interfaces that can be used by architecture code to keep track of
1161 : * memory type of pfn mappings specified by the remap_pfn_range,
1162 : * vmf_insert_pfn.
1163 : */
1164 :
1165 : /*
1166 : * track_pfn_remap is called when a _new_ pfn mapping is being established
1167 : * by remap_pfn_range() for physical range indicated by pfn and size.
1168 : */
1169 : static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1170 : unsigned long pfn, unsigned long addr,
1171 : unsigned long size)
1172 : {
1173 : return 0;
1174 : }
1175 :
1176 : /*
1177 : * track_pfn_insert is called when a _new_ single pfn is established
1178 : * by vmf_insert_pfn().
1179 : */
1180 : static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1181 : pfn_t pfn)
1182 : {
1183 : }
1184 :
1185 : /*
1186 : * track_pfn_copy is called when vma that is covering the pfnmap gets
1187 : * copied through copy_page_range().
1188 : */
1189 : static inline int track_pfn_copy(struct vm_area_struct *vma)
1190 : {
1191 : return 0;
1192 : }
1193 :
1194 : /*
1195 : * untrack_pfn is called while unmapping a pfnmap for a region.
1196 : * untrack can be called for a specific region indicated by pfn and size or
1197 : * can be for the entire vma (in which case pfn, size are zero).
1198 : */
1199 : static inline void untrack_pfn(struct vm_area_struct *vma,
1200 : unsigned long pfn, unsigned long size,
1201 : bool mm_wr_locked)
1202 : {
1203 : }
1204 :
1205 : /*
1206 : * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1207 : * or fails to copy pgtable during duplicate vm area.
1208 : */
1209 : static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1210 : {
1211 : }
1212 : #else
1213 : extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1214 : unsigned long pfn, unsigned long addr,
1215 : unsigned long size);
1216 : extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1217 : pfn_t pfn);
1218 : extern int track_pfn_copy(struct vm_area_struct *vma);
1219 : extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1220 : unsigned long size, bool mm_wr_locked);
1221 : extern void untrack_pfn_clear(struct vm_area_struct *vma);
1222 : #endif
1223 :
1224 : #ifdef CONFIG_MMU
1225 : #ifdef __HAVE_COLOR_ZERO_PAGE
1226 : static inline int is_zero_pfn(unsigned long pfn)
1227 : {
1228 : extern unsigned long zero_pfn;
1229 : unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1230 : return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1231 : }
1232 :
1233 : #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
1234 :
1235 : #else
1236 : static inline int is_zero_pfn(unsigned long pfn)
1237 : {
1238 : extern unsigned long zero_pfn;
1239 : return pfn == zero_pfn;
1240 : }
1241 :
1242 : static inline unsigned long my_zero_pfn(unsigned long addr)
1243 : {
1244 : extern unsigned long zero_pfn;
1245 : return zero_pfn;
1246 : }
1247 : #endif
1248 : #else
1249 : static inline int is_zero_pfn(unsigned long pfn)
1250 : {
1251 : return 0;
1252 : }
1253 :
1254 : static inline unsigned long my_zero_pfn(unsigned long addr)
1255 : {
1256 : return 0;
1257 : }
1258 : #endif /* CONFIG_MMU */
1259 :
1260 : #ifdef CONFIG_MMU
1261 :
1262 : #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1263 : static inline int pmd_trans_huge(pmd_t pmd)
1264 : {
1265 : return 0;
1266 : }
1267 : #ifndef pmd_write
1268 : static inline int pmd_write(pmd_t pmd)
1269 : {
1270 : BUG();
1271 : return 0;
1272 : }
1273 : #endif /* pmd_write */
1274 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1275 :
1276 : #ifndef pud_write
1277 : static inline int pud_write(pud_t pud)
1278 : {
1279 : BUG();
1280 : return 0;
1281 : }
1282 : #endif /* pud_write */
1283 :
1284 : #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1285 : static inline int pmd_devmap(pmd_t pmd)
1286 : {
1287 : return 0;
1288 : }
1289 : static inline int pud_devmap(pud_t pud)
1290 : {
1291 : return 0;
1292 : }
1293 : static inline int pgd_devmap(pgd_t pgd)
1294 : {
1295 : return 0;
1296 : }
1297 : #endif
1298 :
1299 : #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1300 : !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1301 : static inline int pud_trans_huge(pud_t pud)
1302 : {
1303 : return 0;
1304 : }
1305 : #endif
1306 :
1307 : static inline int pud_trans_unstable(pud_t *pud)
1308 : {
1309 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1310 : defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1311 : pud_t pudval = READ_ONCE(*pud);
1312 :
1313 : if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1314 : return 1;
1315 : if (unlikely(pud_bad(pudval))) {
1316 : pud_clear_bad(pud);
1317 : return 1;
1318 : }
1319 : #endif
1320 : return 0;
1321 : }
1322 :
1323 : #ifndef CONFIG_NUMA_BALANCING
1324 : /*
1325 : * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1326 : * the only case the kernel cares is for NUMA balancing and is only ever set
1327 : * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1328 : * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1329 : * is the responsibility of the caller to distinguish between PROT_NONE
1330 : * protections and NUMA hinting fault protections.
1331 : */
1332 : static inline int pte_protnone(pte_t pte)
1333 : {
1334 : return 0;
1335 : }
1336 :
1337 : static inline int pmd_protnone(pmd_t pmd)
1338 : {
1339 : return 0;
1340 : }
1341 : #endif /* CONFIG_NUMA_BALANCING */
1342 :
1343 : #endif /* CONFIG_MMU */
1344 :
1345 : #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1346 :
1347 : #ifndef __PAGETABLE_P4D_FOLDED
1348 : int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1349 : void p4d_clear_huge(p4d_t *p4d);
1350 : #else
1351 : static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1352 : {
1353 : return 0;
1354 : }
1355 : static inline void p4d_clear_huge(p4d_t *p4d) { }
1356 : #endif /* !__PAGETABLE_P4D_FOLDED */
1357 :
1358 : int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1359 : int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1360 : int pud_clear_huge(pud_t *pud);
1361 : int pmd_clear_huge(pmd_t *pmd);
1362 : int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1363 : int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1364 : int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1365 : #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1366 : static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1367 : {
1368 : return 0;
1369 : }
1370 : static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1371 : {
1372 : return 0;
1373 : }
1374 : static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1375 : {
1376 : return 0;
1377 : }
1378 : static inline void p4d_clear_huge(p4d_t *p4d) { }
1379 : static inline int pud_clear_huge(pud_t *pud)
1380 : {
1381 : return 0;
1382 : }
1383 : static inline int pmd_clear_huge(pmd_t *pmd)
1384 : {
1385 : return 0;
1386 : }
1387 : static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1388 : {
1389 : return 0;
1390 : }
1391 : static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1392 : {
1393 : return 0;
1394 : }
1395 : static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1396 : {
1397 : return 0;
1398 : }
1399 : #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1400 :
1401 : #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1402 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1403 : /*
1404 : * ARCHes with special requirements for evicting THP backing TLB entries can
1405 : * implement this. Otherwise also, it can help optimize normal TLB flush in
1406 : * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1407 : * entire TLB if flush span is greater than a threshold, which will
1408 : * likely be true for a single huge page. Thus a single THP flush will
1409 : * invalidate the entire TLB which is not desirable.
1410 : * e.g. see arch/arc: flush_pmd_tlb_range
1411 : */
1412 : #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1413 : #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1414 : #else
1415 : #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1416 : #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1417 : #endif
1418 : #endif
1419 :
1420 : struct file;
1421 : int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1422 : unsigned long size, pgprot_t *vma_prot);
1423 :
1424 : #ifndef CONFIG_X86_ESPFIX64
1425 : static inline void init_espfix_bsp(void) { }
1426 : #endif
1427 :
1428 : extern void __init pgtable_cache_init(void);
1429 :
1430 : #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1431 : static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1432 : {
1433 : return true;
1434 : }
1435 :
1436 : static inline bool arch_has_pfn_modify_check(void)
1437 : {
1438 : return false;
1439 : }
1440 : #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1441 :
1442 : /*
1443 : * Architecture PAGE_KERNEL_* fallbacks
1444 : *
1445 : * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1446 : * because they really don't support them, or the port needs to be updated to
1447 : * reflect the required functionality. Below are a set of relatively safe
1448 : * fallbacks, as best effort, which we can count on in lieu of the architectures
1449 : * not defining them on their own yet.
1450 : */
1451 :
1452 : #ifndef PAGE_KERNEL_RO
1453 : # define PAGE_KERNEL_RO PAGE_KERNEL
1454 : #endif
1455 :
1456 : #ifndef PAGE_KERNEL_EXEC
1457 : # define PAGE_KERNEL_EXEC PAGE_KERNEL
1458 : #endif
1459 :
1460 : /*
1461 : * Page Table Modification bits for pgtbl_mod_mask.
1462 : *
1463 : * These are used by the p?d_alloc_track*() set of functions an in the generic
1464 : * vmalloc/ioremap code to track at which page-table levels entries have been
1465 : * modified. Based on that the code can better decide when vmalloc and ioremap
1466 : * mapping changes need to be synchronized to other page-tables in the system.
1467 : */
1468 : #define __PGTBL_PGD_MODIFIED 0
1469 : #define __PGTBL_P4D_MODIFIED 1
1470 : #define __PGTBL_PUD_MODIFIED 2
1471 : #define __PGTBL_PMD_MODIFIED 3
1472 : #define __PGTBL_PTE_MODIFIED 4
1473 :
1474 : #define PGTBL_PGD_MODIFIED BIT(__PGTBL_PGD_MODIFIED)
1475 : #define PGTBL_P4D_MODIFIED BIT(__PGTBL_P4D_MODIFIED)
1476 : #define PGTBL_PUD_MODIFIED BIT(__PGTBL_PUD_MODIFIED)
1477 : #define PGTBL_PMD_MODIFIED BIT(__PGTBL_PMD_MODIFIED)
1478 : #define PGTBL_PTE_MODIFIED BIT(__PGTBL_PTE_MODIFIED)
1479 :
1480 : /* Page-Table Modification Mask */
1481 : typedef unsigned int pgtbl_mod_mask;
1482 :
1483 : #endif /* !__ASSEMBLY__ */
1484 :
1485 : #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1486 : #ifdef CONFIG_PHYS_ADDR_T_64BIT
1487 : /*
1488 : * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1489 : * with physical address space extension, but falls back to
1490 : * BITS_PER_LONG otherwise.
1491 : */
1492 : #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1493 : #else
1494 : #define MAX_POSSIBLE_PHYSMEM_BITS 32
1495 : #endif
1496 : #endif
1497 :
1498 : #ifndef has_transparent_hugepage
1499 : #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1500 : #endif
1501 :
1502 : /*
1503 : * On some architectures it depends on the mm if the p4d/pud or pmd
1504 : * layer of the page table hierarchy is folded or not.
1505 : */
1506 : #ifndef mm_p4d_folded
1507 : #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1508 : #endif
1509 :
1510 : #ifndef mm_pud_folded
1511 : #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1512 : #endif
1513 :
1514 : #ifndef mm_pmd_folded
1515 : #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1516 : #endif
1517 :
1518 : #ifndef p4d_offset_lockless
1519 : #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1520 : #endif
1521 : #ifndef pud_offset_lockless
1522 : #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1523 : #endif
1524 : #ifndef pmd_offset_lockless
1525 : #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1526 : #endif
1527 :
1528 : /*
1529 : * p?d_leaf() - true if this entry is a final mapping to a physical address.
1530 : * This differs from p?d_huge() by the fact that they are always available (if
1531 : * the architecture supports large pages at the appropriate level) even
1532 : * if CONFIG_HUGETLB_PAGE is not defined.
1533 : * Only meaningful when called on a valid entry.
1534 : */
1535 : #ifndef pgd_leaf
1536 : #define pgd_leaf(x) 0
1537 : #endif
1538 : #ifndef p4d_leaf
1539 : #define p4d_leaf(x) 0
1540 : #endif
1541 : #ifndef pud_leaf
1542 : #define pud_leaf(x) 0
1543 : #endif
1544 : #ifndef pmd_leaf
1545 : #define pmd_leaf(x) 0
1546 : #endif
1547 :
1548 : #ifndef pgd_leaf_size
1549 : #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1550 : #endif
1551 : #ifndef p4d_leaf_size
1552 : #define p4d_leaf_size(x) P4D_SIZE
1553 : #endif
1554 : #ifndef pud_leaf_size
1555 : #define pud_leaf_size(x) PUD_SIZE
1556 : #endif
1557 : #ifndef pmd_leaf_size
1558 : #define pmd_leaf_size(x) PMD_SIZE
1559 : #endif
1560 : #ifndef pte_leaf_size
1561 : #define pte_leaf_size(x) PAGE_SIZE
1562 : #endif
1563 :
1564 : /*
1565 : * Some architectures have MMUs that are configurable or selectable at boot
1566 : * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1567 : * helps to have a static maximum value.
1568 : */
1569 :
1570 : #ifndef MAX_PTRS_PER_PTE
1571 : #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1572 : #endif
1573 :
1574 : #ifndef MAX_PTRS_PER_PMD
1575 : #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1576 : #endif
1577 :
1578 : #ifndef MAX_PTRS_PER_PUD
1579 : #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1580 : #endif
1581 :
1582 : #ifndef MAX_PTRS_PER_P4D
1583 : #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1584 : #endif
1585 :
1586 : /* description of effects of mapping type and prot in current implementation.
1587 : * this is due to the limited x86 page protection hardware. The expected
1588 : * behavior is in parens:
1589 : *
1590 : * map_type prot
1591 : * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
1592 : * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
1593 : * w: (no) no w: (no) no w: (yes) yes w: (no) no
1594 : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
1595 : *
1596 : * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
1597 : * w: (no) no w: (no) no w: (copy) copy w: (no) no
1598 : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
1599 : *
1600 : * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1601 : * MAP_PRIVATE (with Enhanced PAN supported):
1602 : * r: (no) no
1603 : * w: (no) no
1604 : * x: (yes) yes
1605 : */
1606 : #define DECLARE_VM_GET_PAGE_PROT \
1607 : pgprot_t vm_get_page_prot(unsigned long vm_flags) \
1608 : { \
1609 : return protection_map[vm_flags & \
1610 : (VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)]; \
1611 : } \
1612 : EXPORT_SYMBOL(vm_get_page_prot);
1613 :
1614 : #endif /* _LINUX_PGTABLE_H */
|
