• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
16 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
17 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
18 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
19 #endif
20 
21 /*
22  * On almost all architectures and configurations, 0 can be used as the
23  * upper ceiling to free_pgtables(): on many architectures it has the same
24  * effect as using TASK_SIZE.  However, there is one configuration which
25  * must impose a more careful limit, to avoid freeing kernel pgtables.
26  */
27 #ifndef USER_PGTABLES_CEILING
28 #define USER_PGTABLES_CEILING	0UL
29 #endif
30 
31 /*
32  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
33  *
34  * The pXx_index() functions return the index of the entry in the page
35  * table page which would control the given virtual address
36  *
37  * As these functions may be used by the same code for different levels of
38  * the page table folding, they are always available, regardless of
39  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
40  * because in such cases PTRS_PER_PxD equals 1.
41  */
42 
pte_index(unsigned long address)43 static inline unsigned long pte_index(unsigned long address)
44 {
45 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
46 }
47 
48 #ifndef pmd_index
pmd_index(unsigned long address)49 static inline unsigned long pmd_index(unsigned long address)
50 {
51 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
52 }
53 #define pmd_index pmd_index
54 #endif
55 
56 #ifndef pud_index
pud_index(unsigned long address)57 static inline unsigned long pud_index(unsigned long address)
58 {
59 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
60 }
61 #define pud_index pud_index
62 #endif
63 
64 #ifndef pgd_index
65 /* Must be a compile-time constant, so implement it as a macro */
66 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
67 #endif
68 
69 #ifndef pte_offset_kernel
pte_offset_kernel(pmd_t * pmd,unsigned long address)70 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
71 {
72 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
73 }
74 #define pte_offset_kernel pte_offset_kernel
75 #endif
76 
77 #if defined(CONFIG_HIGHPTE)
78 #define pte_offset_map(dir, address)				\
79 	((pte_t *)kmap_atomic(pmd_page(*(dir))) +		\
80 	 pte_index((address)))
81 #define pte_unmap(pte) kunmap_atomic((pte))
82 #else
83 #define pte_offset_map(dir, address)	pte_offset_kernel((dir), (address))
84 #define pte_unmap(pte) ((void)(pte))	/* NOP */
85 #endif
86 
87 /* Find an entry in the second-level page table.. */
88 #ifndef pmd_offset
pmd_offset(pud_t * pud,unsigned long address)89 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
90 {
91 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
92 }
93 #define pmd_offset pmd_offset
94 #endif
95 
96 #ifndef pud_offset
pud_offset(p4d_t * p4d,unsigned long address)97 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
98 {
99 	return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
100 }
101 #define pud_offset pud_offset
102 #endif
103 
pgd_offset_pgd(pgd_t * pgd,unsigned long address)104 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
105 {
106 	return (pgd + pgd_index(address));
107 };
108 
109 /*
110  * a shortcut to get a pgd_t in a given mm
111  */
112 #ifndef pgd_offset
113 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
114 #endif
115 
116 /*
117  * a shortcut which implies the use of the kernel's pgd, instead
118  * of a process's
119  */
120 #ifndef pgd_offset_k
121 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
122 #endif
123 
124 /*
125  * In many cases it is known that a virtual address is mapped at PMD or PTE
126  * level, so instead of traversing all the page table levels, we can get a
127  * pointer to the PMD entry in user or kernel page table or translate a virtual
128  * address to the pointer in the PTE in the kernel page tables with simple
129  * helpers.
130  */
pmd_off(struct mm_struct * mm,unsigned long va)131 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
132 {
133 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
134 }
135 
pmd_off_k(unsigned long va)136 static inline pmd_t *pmd_off_k(unsigned long va)
137 {
138 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
139 }
140 
virt_to_kpte(unsigned long vaddr)141 static inline pte_t *virt_to_kpte(unsigned long vaddr)
142 {
143 	pmd_t *pmd = pmd_off_k(vaddr);
144 
145 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
146 }
147 
148 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
149 extern int ptep_set_access_flags(struct vm_area_struct *vma,
150 				 unsigned long address, pte_t *ptep,
151 				 pte_t entry, int dirty);
152 #endif
153 
154 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
155 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
156 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
157 				 unsigned long address, pmd_t *pmdp,
158 				 pmd_t entry, int dirty);
159 extern int pudp_set_access_flags(struct vm_area_struct *vma,
160 				 unsigned long address, pud_t *pudp,
161 				 pud_t entry, int dirty);
162 #else
pmdp_set_access_flags(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,pmd_t entry,int dirty)163 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
164 					unsigned long address, pmd_t *pmdp,
165 					pmd_t entry, int dirty)
166 {
167 	BUILD_BUG();
168 	return 0;
169 }
pudp_set_access_flags(struct vm_area_struct * vma,unsigned long address,pud_t * pudp,pud_t entry,int dirty)170 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
171 					unsigned long address, pud_t *pudp,
172 					pud_t entry, int dirty)
173 {
174 	BUILD_BUG();
175 	return 0;
176 }
177 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
178 #endif
179 
180 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
ptep_test_and_clear_young(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)181 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
182 					    unsigned long address,
183 					    pte_t *ptep)
184 {
185 	pte_t pte = *ptep;
186 	int r = 1;
187 	if (!pte_young(pte))
188 		r = 0;
189 	else
190 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
191 	return r;
192 }
193 #endif
194 
195 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmdp_test_and_clear_young(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)197 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
198 					    unsigned long address,
199 					    pmd_t *pmdp)
200 {
201 	pmd_t pmd = *pmdp;
202 	int r = 1;
203 	if (!pmd_young(pmd))
204 		r = 0;
205 	else
206 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
207 	return r;
208 }
209 #else
pmdp_test_and_clear_young(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)210 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
211 					    unsigned long address,
212 					    pmd_t *pmdp)
213 {
214 	BUILD_BUG();
215 	return 0;
216 }
217 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
218 #endif
219 
220 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
221 int ptep_clear_flush_young(struct vm_area_struct *vma,
222 			   unsigned long address, pte_t *ptep);
223 #endif
224 
225 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
227 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
228 				  unsigned long address, pmd_t *pmdp);
229 #else
230 /*
231  * Despite relevant to THP only, this API is called from generic rmap code
232  * under PageTransHuge(), hence needs a dummy implementation for !THP
233  */
pmdp_clear_flush_young(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)234 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
235 					 unsigned long address, pmd_t *pmdp)
236 {
237 	BUILD_BUG();
238 	return 0;
239 }
240 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
241 #endif
242 
243 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
ptep_get_and_clear(struct mm_struct * mm,unsigned long address,pte_t * ptep)244 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
245 				       unsigned long address,
246 				       pte_t *ptep)
247 {
248 	pte_t pte = *ptep;
249 	pte_clear(mm, address, ptep);
250 	return pte;
251 }
252 #endif
253 
254 #ifndef __HAVE_ARCH_PTEP_GET
ptep_get(pte_t * ptep)255 static inline pte_t ptep_get(pte_t *ptep)
256 {
257 	return READ_ONCE(*ptep);
258 }
259 #endif
260 
261 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
262 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
pmdp_huge_get_and_clear(struct mm_struct * mm,unsigned long address,pmd_t * pmdp)263 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
264 					    unsigned long address,
265 					    pmd_t *pmdp)
266 {
267 	pmd_t pmd = *pmdp;
268 	pmd_clear(pmdp);
269 	return pmd;
270 }
271 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
272 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
pudp_huge_get_and_clear(struct mm_struct * mm,unsigned long address,pud_t * pudp)273 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
274 					    unsigned long address,
275 					    pud_t *pudp)
276 {
277 	pud_t pud = *pudp;
278 
279 	pud_clear(pudp);
280 	return pud;
281 }
282 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
283 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
284 
285 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
286 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
pmdp_huge_get_and_clear_full(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,int full)287 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
288 					    unsigned long address, pmd_t *pmdp,
289 					    int full)
290 {
291 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
292 }
293 #endif
294 
295 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
pudp_huge_get_and_clear_full(struct mm_struct * mm,unsigned long address,pud_t * pudp,int full)296 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
297 					    unsigned long address, pud_t *pudp,
298 					    int full)
299 {
300 	return pudp_huge_get_and_clear(mm, address, pudp);
301 }
302 #endif
303 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
304 
305 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
ptep_get_and_clear_full(struct mm_struct * mm,unsigned long address,pte_t * ptep,int full)306 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
307 					    unsigned long address, pte_t *ptep,
308 					    int full)
309 {
310 	pte_t pte;
311 	pte = ptep_get_and_clear(mm, address, ptep);
312 	return pte;
313 }
314 #endif
315 
316 
317 /*
318  * If two threads concurrently fault at the same page, the thread that
319  * won the race updates the PTE and its local TLB/Cache. The other thread
320  * gives up, simply does nothing, and continues; on architectures where
321  * software can update TLB,  local TLB can be updated here to avoid next page
322  * fault. This function updates TLB only, do nothing with cache or others.
323  * It is the difference with function update_mmu_cache.
324  */
325 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
update_mmu_tlb(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)326 static inline void update_mmu_tlb(struct vm_area_struct *vma,
327 				unsigned long address, pte_t *ptep)
328 {
329 }
330 #define __HAVE_ARCH_UPDATE_MMU_TLB
331 #endif
332 
333 /*
334  * Some architectures may be able to avoid expensive synchronization
335  * primitives when modifications are made to PTE's which are already
336  * not present, or in the process of an address space destruction.
337  */
338 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
pte_clear_not_present_full(struct mm_struct * mm,unsigned long address,pte_t * ptep,int full)339 static inline void pte_clear_not_present_full(struct mm_struct *mm,
340 					      unsigned long address,
341 					      pte_t *ptep,
342 					      int full)
343 {
344 	pte_clear(mm, address, ptep);
345 }
346 #endif
347 
348 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
349 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
350 			      unsigned long address,
351 			      pte_t *ptep);
352 #endif
353 
354 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
355 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
356 			      unsigned long address,
357 			      pmd_t *pmdp);
358 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
359 			      unsigned long address,
360 			      pud_t *pudp);
361 #endif
362 
363 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
364 struct mm_struct;
ptep_set_wrprotect(struct mm_struct * mm,unsigned long address,pte_t * ptep)365 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
366 {
367 	pte_t old_pte = *ptep;
368 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
369 }
370 #endif
371 
372 /*
373  * On some architectures hardware does not set page access bit when accessing
374  * memory page, it is responsibilty of software setting this bit. It brings
375  * out extra page fault penalty to track page access bit. For optimization page
376  * access bit can be set during all page fault flow on these arches.
377  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
378  * where software maintains page access bit.
379  */
380 #ifndef pte_sw_mkyoung
pte_sw_mkyoung(pte_t pte)381 static inline pte_t pte_sw_mkyoung(pte_t pte)
382 {
383 	return pte;
384 }
385 #define pte_sw_mkyoung	pte_sw_mkyoung
386 #endif
387 
388 #ifndef pte_savedwrite
389 #define pte_savedwrite pte_write
390 #endif
391 
392 #ifndef pte_mk_savedwrite
393 #define pte_mk_savedwrite pte_mkwrite
394 #endif
395 
396 #ifndef pte_clear_savedwrite
397 #define pte_clear_savedwrite pte_wrprotect
398 #endif
399 
400 #ifndef pmd_savedwrite
401 #define pmd_savedwrite pmd_write
402 #endif
403 
404 #ifndef pmd_mk_savedwrite
405 #define pmd_mk_savedwrite pmd_mkwrite
406 #endif
407 
408 #ifndef pmd_clear_savedwrite
409 #define pmd_clear_savedwrite pmd_wrprotect
410 #endif
411 
412 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
413 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmdp_set_wrprotect(struct mm_struct * mm,unsigned long address,pmd_t * pmdp)414 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
415 				      unsigned long address, pmd_t *pmdp)
416 {
417 	pmd_t old_pmd = *pmdp;
418 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
419 }
420 #else
pmdp_set_wrprotect(struct mm_struct * mm,unsigned long address,pmd_t * pmdp)421 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
422 				      unsigned long address, pmd_t *pmdp)
423 {
424 	BUILD_BUG();
425 }
426 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
427 #endif
428 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
429 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
pudp_set_wrprotect(struct mm_struct * mm,unsigned long address,pud_t * pudp)430 static inline void pudp_set_wrprotect(struct mm_struct *mm,
431 				      unsigned long address, pud_t *pudp)
432 {
433 	pud_t old_pud = *pudp;
434 
435 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
436 }
437 #else
pudp_set_wrprotect(struct mm_struct * mm,unsigned long address,pud_t * pudp)438 static inline void pudp_set_wrprotect(struct mm_struct *mm,
439 				      unsigned long address, pud_t *pudp)
440 {
441 	BUILD_BUG();
442 }
443 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
444 #endif
445 
446 #ifndef pmdp_collapse_flush
447 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
448 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
449 				 unsigned long address, pmd_t *pmdp);
450 #else
pmdp_collapse_flush(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)451 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
452 					unsigned long address,
453 					pmd_t *pmdp)
454 {
455 	BUILD_BUG();
456 	return *pmdp;
457 }
458 #define pmdp_collapse_flush pmdp_collapse_flush
459 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
460 #endif
461 
462 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
463 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
464 				       pgtable_t pgtable);
465 #endif
466 
467 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
468 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
469 #endif
470 
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /*
473  * This is an implementation of pmdp_establish() that is only suitable for an
474  * architecture that doesn't have hardware dirty/accessed bits. In this case we
475  * can't race with CPU which sets these bits and non-atomic aproach is fine.
476  */
generic_pmdp_establish(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,pmd_t pmd)477 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
478 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
479 {
480 	pmd_t old_pmd = *pmdp;
481 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
482 	return old_pmd;
483 }
484 #endif
485 
486 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
487 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
488 			    pmd_t *pmdp);
489 #endif
490 
491 #ifndef __HAVE_ARCH_PTE_SAME
pte_same(pte_t pte_a,pte_t pte_b)492 static inline int pte_same(pte_t pte_a, pte_t pte_b)
493 {
494 	return pte_val(pte_a) == pte_val(pte_b);
495 }
496 #endif
497 
498 #ifndef __HAVE_ARCH_PTE_UNUSED
499 /*
500  * Some architectures provide facilities to virtualization guests
501  * so that they can flag allocated pages as unused. This allows the
502  * host to transparently reclaim unused pages. This function returns
503  * whether the pte's page is unused.
504  */
pte_unused(pte_t pte)505 static inline int pte_unused(pte_t pte)
506 {
507 	return 0;
508 }
509 #endif
510 
511 #ifndef pte_access_permitted
512 #define pte_access_permitted(pte, write) \
513 	(pte_present(pte) && (!(write) || pte_write(pte)))
514 #endif
515 
516 #ifndef pmd_access_permitted
517 #define pmd_access_permitted(pmd, write) \
518 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
519 #endif
520 
521 #ifndef pud_access_permitted
522 #define pud_access_permitted(pud, write) \
523 	(pud_present(pud) && (!(write) || pud_write(pud)))
524 #endif
525 
526 #ifndef p4d_access_permitted
527 #define p4d_access_permitted(p4d, write) \
528 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
529 #endif
530 
531 #ifndef pgd_access_permitted
532 #define pgd_access_permitted(pgd, write) \
533 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
534 #endif
535 
536 #ifndef __HAVE_ARCH_PMD_SAME
pmd_same(pmd_t pmd_a,pmd_t pmd_b)537 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
538 {
539 	return pmd_val(pmd_a) == pmd_val(pmd_b);
540 }
541 
pud_same(pud_t pud_a,pud_t pud_b)542 static inline int pud_same(pud_t pud_a, pud_t pud_b)
543 {
544 	return pud_val(pud_a) == pud_val(pud_b);
545 }
546 #endif
547 
548 #ifndef __HAVE_ARCH_P4D_SAME
p4d_same(p4d_t p4d_a,p4d_t p4d_b)549 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
550 {
551 	return p4d_val(p4d_a) == p4d_val(p4d_b);
552 }
553 #endif
554 
555 #ifndef __HAVE_ARCH_PGD_SAME
pgd_same(pgd_t pgd_a,pgd_t pgd_b)556 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
557 {
558 	return pgd_val(pgd_a) == pgd_val(pgd_b);
559 }
560 #endif
561 
562 /*
563  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
564  * TLB flush will be required as a result of the "set". For example, use
565  * in scenarios where it is known ahead of time that the routine is
566  * setting non-present entries, or re-setting an existing entry to the
567  * same value. Otherwise, use the typical "set" helpers and flush the
568  * TLB.
569  */
570 #define set_pte_safe(ptep, pte) \
571 ({ \
572 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
573 	set_pte(ptep, pte); \
574 })
575 
576 #define set_pmd_safe(pmdp, pmd) \
577 ({ \
578 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
579 	set_pmd(pmdp, pmd); \
580 })
581 
582 #define set_pud_safe(pudp, pud) \
583 ({ \
584 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
585 	set_pud(pudp, pud); \
586 })
587 
588 #define set_p4d_safe(p4dp, p4d) \
589 ({ \
590 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
591 	set_p4d(p4dp, p4d); \
592 })
593 
594 #define set_pgd_safe(pgdp, pgd) \
595 ({ \
596 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
597 	set_pgd(pgdp, pgd); \
598 })
599 
600 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
601 /*
602  * Some architectures support metadata associated with a page. When a
603  * page is being swapped out, this metadata must be saved so it can be
604  * restored when the page is swapped back in. SPARC M7 and newer
605  * processors support an ADI (Application Data Integrity) tag for the
606  * page as metadata for the page. arch_do_swap_page() can restore this
607  * metadata when a page is swapped back in.
608  */
arch_do_swap_page(struct mm_struct * mm,struct vm_area_struct * vma,unsigned long addr,pte_t pte,pte_t oldpte)609 static inline void arch_do_swap_page(struct mm_struct *mm,
610 				     struct vm_area_struct *vma,
611 				     unsigned long addr,
612 				     pte_t pte, pte_t oldpte)
613 {
614 
615 }
616 #endif
617 
618 #ifndef __HAVE_ARCH_UNMAP_ONE
619 /*
620  * Some architectures support metadata associated with a page. When a
621  * page is being swapped out, this metadata must be saved so it can be
622  * restored when the page is swapped back in. SPARC M7 and newer
623  * processors support an ADI (Application Data Integrity) tag for the
624  * page as metadata for the page. arch_unmap_one() can save this
625  * metadata on a swap-out of a page.
626  */
arch_unmap_one(struct mm_struct * mm,struct vm_area_struct * vma,unsigned long addr,pte_t orig_pte)627 static inline int arch_unmap_one(struct mm_struct *mm,
628 				  struct vm_area_struct *vma,
629 				  unsigned long addr,
630 				  pte_t orig_pte)
631 {
632 	return 0;
633 }
634 #endif
635 
636 /*
637  * Allow architectures to preserve additional metadata associated with
638  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
639  * prototypes must be defined in the arch-specific asm/pgtable.h file.
640  */
641 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
arch_prepare_to_swap(struct page * page)642 static inline int arch_prepare_to_swap(struct page *page)
643 {
644 	return 0;
645 }
646 #endif
647 
648 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
arch_swap_invalidate_page(int type,pgoff_t offset)649 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
650 {
651 }
652 
arch_swap_invalidate_area(int type)653 static inline void arch_swap_invalidate_area(int type)
654 {
655 }
656 #endif
657 
658 #ifndef __HAVE_ARCH_SWAP_RESTORE
arch_swap_restore(swp_entry_t entry,struct page * page)659 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
660 {
661 }
662 #endif
663 
664 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
665 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
666 #endif
667 
668 #ifndef __HAVE_ARCH_MOVE_PTE
669 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
670 #endif
671 
672 #ifndef pte_accessible
673 # define pte_accessible(mm, pte)	((void)(pte), 1)
674 #endif
675 
676 #ifndef flush_tlb_fix_spurious_fault
677 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
678 #endif
679 
680 /*
681  * When walking page tables, get the address of the next boundary,
682  * or the end address of the range if that comes earlier.  Although no
683  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
684  */
685 
686 #define pgd_addr_end(addr, end)						\
687 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
688 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
689 })
690 
691 #ifndef p4d_addr_end
692 #define p4d_addr_end(addr, end)						\
693 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
694 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
695 })
696 #endif
697 
698 #ifndef pud_addr_end
699 #define pud_addr_end(addr, end)						\
700 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
701 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
702 })
703 #endif
704 
705 #ifndef pmd_addr_end
706 #define pmd_addr_end(addr, end)						\
707 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
708 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
709 })
710 #endif
711 
712 /*
713  * When walking page tables, we usually want to skip any p?d_none entries;
714  * and any p?d_bad entries - reporting the error before resetting to none.
715  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
716  */
717 void pgd_clear_bad(pgd_t *);
718 
719 #ifndef __PAGETABLE_P4D_FOLDED
720 void p4d_clear_bad(p4d_t *);
721 #else
722 #define p4d_clear_bad(p4d)        do { } while (0)
723 #endif
724 
725 #ifndef __PAGETABLE_PUD_FOLDED
726 void pud_clear_bad(pud_t *);
727 #else
728 #define pud_clear_bad(p4d)        do { } while (0)
729 #endif
730 
731 void pmd_clear_bad(pmd_t *);
732 
pgd_none_or_clear_bad(pgd_t * pgd)733 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
734 {
735 	if (pgd_none(*pgd))
736 		return 1;
737 	if (unlikely(pgd_bad(*pgd))) {
738 		pgd_clear_bad(pgd);
739 		return 1;
740 	}
741 	return 0;
742 }
743 
p4d_none_or_clear_bad(p4d_t * p4d)744 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
745 {
746 	if (p4d_none(*p4d))
747 		return 1;
748 	if (unlikely(p4d_bad(*p4d))) {
749 		p4d_clear_bad(p4d);
750 		return 1;
751 	}
752 	return 0;
753 }
754 
pud_none_or_clear_bad(pud_t * pud)755 static inline int pud_none_or_clear_bad(pud_t *pud)
756 {
757 	if (pud_none(*pud))
758 		return 1;
759 	if (unlikely(pud_bad(*pud))) {
760 		pud_clear_bad(pud);
761 		return 1;
762 	}
763 	return 0;
764 }
765 
pmd_none_or_clear_bad(pmd_t * pmd)766 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
767 {
768 	if (pmd_none(*pmd))
769 		return 1;
770 	if (unlikely(pmd_bad(*pmd))) {
771 		pmd_clear_bad(pmd);
772 		return 1;
773 	}
774 	return 0;
775 }
776 
__ptep_modify_prot_start(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)777 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
778 					     unsigned long addr,
779 					     pte_t *ptep)
780 {
781 	/*
782 	 * Get the current pte state, but zero it out to make it
783 	 * non-present, preventing the hardware from asynchronously
784 	 * updating it.
785 	 */
786 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
787 }
788 
__ptep_modify_prot_commit(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,pte_t pte)789 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
790 					     unsigned long addr,
791 					     pte_t *ptep, pte_t pte)
792 {
793 	/*
794 	 * The pte is non-present, so there's no hardware state to
795 	 * preserve.
796 	 */
797 	set_pte_at(vma->vm_mm, addr, ptep, pte);
798 }
799 
800 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
801 /*
802  * Start a pte protection read-modify-write transaction, which
803  * protects against asynchronous hardware modifications to the pte.
804  * The intention is not to prevent the hardware from making pte
805  * updates, but to prevent any updates it may make from being lost.
806  *
807  * This does not protect against other software modifications of the
808  * pte; the appropriate pte lock must be held over the transation.
809  *
810  * Note that this interface is intended to be batchable, meaning that
811  * ptep_modify_prot_commit may not actually update the pte, but merely
812  * queue the update to be done at some later time.  The update must be
813  * actually committed before the pte lock is released, however.
814  */
ptep_modify_prot_start(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)815 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
816 					   unsigned long addr,
817 					   pte_t *ptep)
818 {
819 	return __ptep_modify_prot_start(vma, addr, ptep);
820 }
821 
822 /*
823  * Commit an update to a pte, leaving any hardware-controlled bits in
824  * the PTE unmodified.
825  */
ptep_modify_prot_commit(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,pte_t old_pte,pte_t pte)826 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
827 					   unsigned long addr,
828 					   pte_t *ptep, pte_t old_pte, pte_t pte)
829 {
830 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
831 }
832 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
833 #endif /* CONFIG_MMU */
834 
835 /*
836  * No-op macros that just return the current protection value. Defined here
837  * because these macros can be used even if CONFIG_MMU is not defined.
838  */
839 
840 #ifndef pgprot_nx
841 #define pgprot_nx(prot)	(prot)
842 #endif
843 
844 #ifndef pgprot_noncached
845 #define pgprot_noncached(prot)	(prot)
846 #endif
847 
848 #ifndef pgprot_writecombine
849 #define pgprot_writecombine pgprot_noncached
850 #endif
851 
852 #ifndef pgprot_writethrough
853 #define pgprot_writethrough pgprot_noncached
854 #endif
855 
856 #ifndef pgprot_device
857 #define pgprot_device pgprot_noncached
858 #endif
859 
860 #ifndef pgprot_mhp
861 #define pgprot_mhp(prot)	(prot)
862 #endif
863 
864 #ifdef CONFIG_MMU
865 #ifndef pgprot_modify
866 #define pgprot_modify pgprot_modify
pgprot_modify(pgprot_t oldprot,pgprot_t newprot)867 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
868 {
869 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
870 		newprot = pgprot_noncached(newprot);
871 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
872 		newprot = pgprot_writecombine(newprot);
873 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
874 		newprot = pgprot_device(newprot);
875 	return newprot;
876 }
877 #endif
878 #endif /* CONFIG_MMU */
879 
880 #ifndef pgprot_encrypted
881 #define pgprot_encrypted(prot)	(prot)
882 #endif
883 
884 #ifndef pgprot_decrypted
885 #define pgprot_decrypted(prot)	(prot)
886 #endif
887 
888 /*
889  * A facility to provide lazy MMU batching.  This allows PTE updates and
890  * page invalidations to be delayed until a call to leave lazy MMU mode
891  * is issued.  Some architectures may benefit from doing this, and it is
892  * beneficial for both shadow and direct mode hypervisors, which may batch
893  * the PTE updates which happen during this window.  Note that using this
894  * interface requires that read hazards be removed from the code.  A read
895  * hazard could result in the direct mode hypervisor case, since the actual
896  * write to the page tables may not yet have taken place, so reads though
897  * a raw PTE pointer after it has been modified are not guaranteed to be
898  * up to date.  This mode can only be entered and left under the protection of
899  * the page table locks for all page tables which may be modified.  In the UP
900  * case, this is required so that preemption is disabled, and in the SMP case,
901  * it must synchronize the delayed page table writes properly on other CPUs.
902  */
903 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
904 #define arch_enter_lazy_mmu_mode()	do {} while (0)
905 #define arch_leave_lazy_mmu_mode()	do {} while (0)
906 #define arch_flush_lazy_mmu_mode()	do {} while (0)
907 #endif
908 
909 /*
910  * A facility to provide batching of the reload of page tables and
911  * other process state with the actual context switch code for
912  * paravirtualized guests.  By convention, only one of the batched
913  * update (lazy) modes (CPU, MMU) should be active at any given time,
914  * entry should never be nested, and entry and exits should always be
915  * paired.  This is for sanity of maintaining and reasoning about the
916  * kernel code.  In this case, the exit (end of the context switch) is
917  * in architecture-specific code, and so doesn't need a generic
918  * definition.
919  */
920 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
921 #define arch_start_context_switch(prev)	do {} while (0)
922 #endif
923 
924 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
925 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
pmd_swp_mksoft_dirty(pmd_t pmd)926 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
927 {
928 	return pmd;
929 }
930 
pmd_swp_soft_dirty(pmd_t pmd)931 static inline int pmd_swp_soft_dirty(pmd_t pmd)
932 {
933 	return 0;
934 }
935 
pmd_swp_clear_soft_dirty(pmd_t pmd)936 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
937 {
938 	return pmd;
939 }
940 #endif
941 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
pte_soft_dirty(pte_t pte)942 static inline int pte_soft_dirty(pte_t pte)
943 {
944 	return 0;
945 }
946 
pmd_soft_dirty(pmd_t pmd)947 static inline int pmd_soft_dirty(pmd_t pmd)
948 {
949 	return 0;
950 }
951 
pte_mksoft_dirty(pte_t pte)952 static inline pte_t pte_mksoft_dirty(pte_t pte)
953 {
954 	return pte;
955 }
956 
pmd_mksoft_dirty(pmd_t pmd)957 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
958 {
959 	return pmd;
960 }
961 
pte_clear_soft_dirty(pte_t pte)962 static inline pte_t pte_clear_soft_dirty(pte_t pte)
963 {
964 	return pte;
965 }
966 
pmd_clear_soft_dirty(pmd_t pmd)967 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
968 {
969 	return pmd;
970 }
971 
pte_swp_mksoft_dirty(pte_t pte)972 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
973 {
974 	return pte;
975 }
976 
pte_swp_soft_dirty(pte_t pte)977 static inline int pte_swp_soft_dirty(pte_t pte)
978 {
979 	return 0;
980 }
981 
pte_swp_clear_soft_dirty(pte_t pte)982 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
983 {
984 	return pte;
985 }
986 
pmd_swp_mksoft_dirty(pmd_t pmd)987 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
988 {
989 	return pmd;
990 }
991 
pmd_swp_soft_dirty(pmd_t pmd)992 static inline int pmd_swp_soft_dirty(pmd_t pmd)
993 {
994 	return 0;
995 }
996 
pmd_swp_clear_soft_dirty(pmd_t pmd)997 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
998 {
999 	return pmd;
1000 }
1001 #endif
1002 
1003 #ifndef __HAVE_PFNMAP_TRACKING
1004 /*
1005  * Interfaces that can be used by architecture code to keep track of
1006  * memory type of pfn mappings specified by the remap_pfn_range,
1007  * vmf_insert_pfn.
1008  */
1009 
1010 /*
1011  * track_pfn_remap is called when a _new_ pfn mapping is being established
1012  * by remap_pfn_range() for physical range indicated by pfn and size.
1013  */
track_pfn_remap(struct vm_area_struct * vma,pgprot_t * prot,unsigned long pfn,unsigned long addr,unsigned long size)1014 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1015 				  unsigned long pfn, unsigned long addr,
1016 				  unsigned long size)
1017 {
1018 	return 0;
1019 }
1020 
1021 /*
1022  * track_pfn_insert is called when a _new_ single pfn is established
1023  * by vmf_insert_pfn().
1024  */
track_pfn_insert(struct vm_area_struct * vma,pgprot_t * prot,pfn_t pfn)1025 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1026 				    pfn_t pfn)
1027 {
1028 }
1029 
1030 /*
1031  * track_pfn_copy is called when vma that is covering the pfnmap gets
1032  * copied through copy_page_range().
1033  */
track_pfn_copy(struct vm_area_struct * vma)1034 static inline int track_pfn_copy(struct vm_area_struct *vma)
1035 {
1036 	return 0;
1037 }
1038 
1039 /*
1040  * untrack_pfn is called while unmapping a pfnmap for a region.
1041  * untrack can be called for a specific region indicated by pfn and size or
1042  * can be for the entire vma (in which case pfn, size are zero).
1043  */
untrack_pfn(struct vm_area_struct * vma,unsigned long pfn,unsigned long size)1044 static inline void untrack_pfn(struct vm_area_struct *vma,
1045 			       unsigned long pfn, unsigned long size)
1046 {
1047 }
1048 
1049 /*
1050  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1051  */
untrack_pfn_moved(struct vm_area_struct * vma)1052 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1053 {
1054 }
1055 #else
1056 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1057 			   unsigned long pfn, unsigned long addr,
1058 			   unsigned long size);
1059 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1060 			     pfn_t pfn);
1061 extern int track_pfn_copy(struct vm_area_struct *vma);
1062 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1063 			unsigned long size);
1064 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1065 #endif
1066 
1067 #ifdef __HAVE_COLOR_ZERO_PAGE
is_zero_pfn(unsigned long pfn)1068 static inline int is_zero_pfn(unsigned long pfn)
1069 {
1070 	extern unsigned long zero_pfn;
1071 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1072 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1073 }
1074 
1075 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1076 
1077 #else
is_zero_pfn(unsigned long pfn)1078 static inline int is_zero_pfn(unsigned long pfn)
1079 {
1080 	extern unsigned long zero_pfn;
1081 	return pfn == zero_pfn;
1082 }
1083 
my_zero_pfn(unsigned long addr)1084 static inline unsigned long my_zero_pfn(unsigned long addr)
1085 {
1086 	extern unsigned long zero_pfn;
1087 	return zero_pfn;
1088 }
1089 #endif
1090 
1091 #ifdef CONFIG_MMU
1092 
1093 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
pmd_trans_huge(pmd_t pmd)1094 static inline int pmd_trans_huge(pmd_t pmd)
1095 {
1096 	return 0;
1097 }
1098 #ifndef pmd_write
pmd_write(pmd_t pmd)1099 static inline int pmd_write(pmd_t pmd)
1100 {
1101 	BUG();
1102 	return 0;
1103 }
1104 #endif /* pmd_write */
1105 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1106 
1107 #ifndef pud_write
pud_write(pud_t pud)1108 static inline int pud_write(pud_t pud)
1109 {
1110 	BUG();
1111 	return 0;
1112 }
1113 #endif /* pud_write */
1114 
1115 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
pmd_devmap(pmd_t pmd)1116 static inline int pmd_devmap(pmd_t pmd)
1117 {
1118 	return 0;
1119 }
pud_devmap(pud_t pud)1120 static inline int pud_devmap(pud_t pud)
1121 {
1122 	return 0;
1123 }
pgd_devmap(pgd_t pgd)1124 static inline int pgd_devmap(pgd_t pgd)
1125 {
1126 	return 0;
1127 }
1128 #endif
1129 
1130 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1131 	(defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1132 	 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
pud_trans_huge(pud_t pud)1133 static inline int pud_trans_huge(pud_t pud)
1134 {
1135 	return 0;
1136 }
1137 #endif
1138 
1139 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t * pud)1140 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1141 {
1142 	pud_t pudval = READ_ONCE(*pud);
1143 
1144 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1145 		return 1;
1146 	if (unlikely(pud_bad(pudval))) {
1147 		pud_clear_bad(pud);
1148 		return 1;
1149 	}
1150 	return 0;
1151 }
1152 
1153 /* See pmd_trans_unstable for discussion. */
pud_trans_unstable(pud_t * pud)1154 static inline int pud_trans_unstable(pud_t *pud)
1155 {
1156 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) &&			\
1157 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1158 	return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1159 #else
1160 	return 0;
1161 #endif
1162 }
1163 
1164 #ifndef pmd_read_atomic
pmd_read_atomic(pmd_t * pmdp)1165 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1166 {
1167 	/*
1168 	 * Depend on compiler for an atomic pmd read. NOTE: this is
1169 	 * only going to work, if the pmdval_t isn't larger than
1170 	 * an unsigned long.
1171 	 */
1172 	return *pmdp;
1173 }
1174 #endif
1175 
1176 #ifndef arch_needs_pgtable_deposit
1177 #define arch_needs_pgtable_deposit() (false)
1178 #endif
1179 /*
1180  * This function is meant to be used by sites walking pagetables with
1181  * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1182  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1183  * into a null pmd and the transhuge page fault can convert a null pmd
1184  * into an hugepmd or into a regular pmd (if the hugepage allocation
1185  * fails). While holding the mmap_lock in read mode the pmd becomes
1186  * stable and stops changing under us only if it's not null and not a
1187  * transhuge pmd. When those races occurs and this function makes a
1188  * difference vs the standard pmd_none_or_clear_bad, the result is
1189  * undefined so behaving like if the pmd was none is safe (because it
1190  * can return none anyway). The compiler level barrier() is critically
1191  * important to compute the two checks atomically on the same pmdval.
1192  *
1193  * For 32bit kernels with a 64bit large pmd_t this automatically takes
1194  * care of reading the pmd atomically to avoid SMP race conditions
1195  * against pmd_populate() when the mmap_lock is hold for reading by the
1196  * caller (a special atomic read not done by "gcc" as in the generic
1197  * version above, is also needed when THP is disabled because the page
1198  * fault can populate the pmd from under us).
1199  */
pmd_none_or_trans_huge_or_clear_bad(pmd_t * pmd)1200 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1201 {
1202 	pmd_t pmdval = pmd_read_atomic(pmd);
1203 	/*
1204 	 * The barrier will stabilize the pmdval in a register or on
1205 	 * the stack so that it will stop changing under the code.
1206 	 *
1207 	 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1208 	 * pmd_read_atomic is allowed to return a not atomic pmdval
1209 	 * (for example pointing to an hugepage that has never been
1210 	 * mapped in the pmd). The below checks will only care about
1211 	 * the low part of the pmd with 32bit PAE x86 anyway, with the
1212 	 * exception of pmd_none(). So the important thing is that if
1213 	 * the low part of the pmd is found null, the high part will
1214 	 * be also null or the pmd_none() check below would be
1215 	 * confused.
1216 	 */
1217 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1218 	barrier();
1219 #endif
1220 	/*
1221 	 * !pmd_present() checks for pmd migration entries
1222 	 *
1223 	 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1224 	 * But using that requires moving current function and pmd_trans_unstable()
1225 	 * to linux/swapops.h to resovle dependency, which is too much code move.
1226 	 *
1227 	 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1228 	 * because !pmd_present() pages can only be under migration not swapped
1229 	 * out.
1230 	 *
1231 	 * pmd_none() is preseved for future condition checks on pmd migration
1232 	 * entries and not confusing with this function name, although it is
1233 	 * redundant with !pmd_present().
1234 	 */
1235 	if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1236 		(IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1237 		return 1;
1238 	if (unlikely(pmd_bad(pmdval))) {
1239 		pmd_clear_bad(pmd);
1240 		return 1;
1241 	}
1242 	return 0;
1243 }
1244 
1245 /*
1246  * This is a noop if Transparent Hugepage Support is not built into
1247  * the kernel. Otherwise it is equivalent to
1248  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1249  * places that already verified the pmd is not none and they want to
1250  * walk ptes while holding the mmap sem in read mode (write mode don't
1251  * need this). If THP is not enabled, the pmd can't go away under the
1252  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1253  * run a pmd_trans_unstable before walking the ptes after
1254  * split_huge_pmd returns (because it may have run when the pmd become
1255  * null, but then a page fault can map in a THP and not a regular page).
1256  */
pmd_trans_unstable(pmd_t * pmd)1257 static inline int pmd_trans_unstable(pmd_t *pmd)
1258 {
1259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1260 	return pmd_none_or_trans_huge_or_clear_bad(pmd);
1261 #else
1262 	return 0;
1263 #endif
1264 }
1265 
1266 #ifndef CONFIG_NUMA_BALANCING
1267 /*
1268  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1269  * the only case the kernel cares is for NUMA balancing and is only ever set
1270  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1271  * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1272  * is the responsibility of the caller to distinguish between PROT_NONE
1273  * protections and NUMA hinting fault protections.
1274  */
pte_protnone(pte_t pte)1275 static inline int pte_protnone(pte_t pte)
1276 {
1277 	return 0;
1278 }
1279 
pmd_protnone(pmd_t pmd)1280 static inline int pmd_protnone(pmd_t pmd)
1281 {
1282 	return 0;
1283 }
1284 #endif /* CONFIG_NUMA_BALANCING */
1285 
1286 #endif /* CONFIG_MMU */
1287 
1288 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1289 
1290 #ifndef __PAGETABLE_P4D_FOLDED
1291 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1292 int p4d_clear_huge(p4d_t *p4d);
1293 #else
p4d_set_huge(p4d_t * p4d,phys_addr_t addr,pgprot_t prot)1294 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1295 {
1296 	return 0;
1297 }
p4d_clear_huge(p4d_t * p4d)1298 static inline int p4d_clear_huge(p4d_t *p4d)
1299 {
1300 	return 0;
1301 }
1302 #endif /* !__PAGETABLE_P4D_FOLDED */
1303 
1304 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1305 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1306 int pud_clear_huge(pud_t *pud);
1307 int pmd_clear_huge(pmd_t *pmd);
1308 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1309 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1310 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1311 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
p4d_set_huge(p4d_t * p4d,phys_addr_t addr,pgprot_t prot)1312 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1313 {
1314 	return 0;
1315 }
pud_set_huge(pud_t * pud,phys_addr_t addr,pgprot_t prot)1316 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1317 {
1318 	return 0;
1319 }
pmd_set_huge(pmd_t * pmd,phys_addr_t addr,pgprot_t prot)1320 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1321 {
1322 	return 0;
1323 }
p4d_clear_huge(p4d_t * p4d)1324 static inline int p4d_clear_huge(p4d_t *p4d)
1325 {
1326 	return 0;
1327 }
pud_clear_huge(pud_t * pud)1328 static inline int pud_clear_huge(pud_t *pud)
1329 {
1330 	return 0;
1331 }
pmd_clear_huge(pmd_t * pmd)1332 static inline int pmd_clear_huge(pmd_t *pmd)
1333 {
1334 	return 0;
1335 }
p4d_free_pud_page(p4d_t * p4d,unsigned long addr)1336 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1337 {
1338 	return 0;
1339 }
pud_free_pmd_page(pud_t * pud,unsigned long addr)1340 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1341 {
1342 	return 0;
1343 }
pmd_free_pte_page(pmd_t * pmd,unsigned long addr)1344 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1345 {
1346 	return 0;
1347 }
1348 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1349 
1350 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1351 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1352 /*
1353  * ARCHes with special requirements for evicting THP backing TLB entries can
1354  * implement this. Otherwise also, it can help optimize normal TLB flush in
1355  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1356  * entire TLB if flush span is greater than a threshold, which will
1357  * likely be true for a single huge page. Thus a single THP flush will
1358  * invalidate the entire TLB which is not desirable.
1359  * e.g. see arch/arc: flush_pmd_tlb_range
1360  */
1361 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1362 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1363 #else
1364 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1365 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1366 #endif
1367 #endif
1368 
1369 struct file;
1370 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1371 			unsigned long size, pgprot_t *vma_prot);
1372 
1373 #ifndef CONFIG_X86_ESPFIX64
init_espfix_bsp(void)1374 static inline void init_espfix_bsp(void) { }
1375 #endif
1376 
1377 extern void __init pgtable_cache_init(void);
1378 
1379 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
pfn_modify_allowed(unsigned long pfn,pgprot_t prot)1380 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1381 {
1382 	return true;
1383 }
1384 
arch_has_pfn_modify_check(void)1385 static inline bool arch_has_pfn_modify_check(void)
1386 {
1387 	return false;
1388 }
1389 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1390 
1391 /*
1392  * Architecture PAGE_KERNEL_* fallbacks
1393  *
1394  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1395  * because they really don't support them, or the port needs to be updated to
1396  * reflect the required functionality. Below are a set of relatively safe
1397  * fallbacks, as best effort, which we can count on in lieu of the architectures
1398  * not defining them on their own yet.
1399  */
1400 
1401 #ifndef PAGE_KERNEL_RO
1402 # define PAGE_KERNEL_RO PAGE_KERNEL
1403 #endif
1404 
1405 #ifndef PAGE_KERNEL_EXEC
1406 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1407 #endif
1408 
1409 /*
1410  * Page Table Modification bits for pgtbl_mod_mask.
1411  *
1412  * These are used by the p?d_alloc_track*() set of functions an in the generic
1413  * vmalloc/ioremap code to track at which page-table levels entries have been
1414  * modified. Based on that the code can better decide when vmalloc and ioremap
1415  * mapping changes need to be synchronized to other page-tables in the system.
1416  */
1417 #define		__PGTBL_PGD_MODIFIED	0
1418 #define		__PGTBL_P4D_MODIFIED	1
1419 #define		__PGTBL_PUD_MODIFIED	2
1420 #define		__PGTBL_PMD_MODIFIED	3
1421 #define		__PGTBL_PTE_MODIFIED	4
1422 
1423 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1424 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1425 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1426 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1427 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1428 
1429 /* Page-Table Modification Mask */
1430 typedef unsigned int pgtbl_mod_mask;
1431 
1432 #endif /* !__ASSEMBLY__ */
1433 
1434 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1435 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1436 /*
1437  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1438  * with physical address space extension, but falls back to
1439  * BITS_PER_LONG otherwise.
1440  */
1441 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1442 #else
1443 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1444 #endif
1445 #endif
1446 
1447 #ifndef has_transparent_hugepage
1448 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1449 #define has_transparent_hugepage() 1
1450 #else
1451 #define has_transparent_hugepage() 0
1452 #endif
1453 #endif
1454 
1455 /*
1456  * On some architectures it depends on the mm if the p4d/pud or pmd
1457  * layer of the page table hierarchy is folded or not.
1458  */
1459 #ifndef mm_p4d_folded
1460 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1461 #endif
1462 
1463 #ifndef mm_pud_folded
1464 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1465 #endif
1466 
1467 #ifndef mm_pmd_folded
1468 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1469 #endif
1470 
1471 #ifndef p4d_offset_lockless
1472 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1473 #endif
1474 #ifndef pud_offset_lockless
1475 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1476 #endif
1477 #ifndef pmd_offset_lockless
1478 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1479 #endif
1480 
1481 /*
1482  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1483  * This differs from p?d_huge() by the fact that they are always available (if
1484  * the architecture supports large pages at the appropriate level) even
1485  * if CONFIG_HUGETLB_PAGE is not defined.
1486  * Only meaningful when called on a valid entry.
1487  */
1488 #ifndef pgd_leaf
1489 #define pgd_leaf(x)	0
1490 #endif
1491 #ifndef p4d_leaf
1492 #define p4d_leaf(x)	0
1493 #endif
1494 #ifndef pud_leaf
1495 #define pud_leaf(x)	0
1496 #endif
1497 #ifndef pmd_leaf
1498 #define pmd_leaf(x)	0
1499 #endif
1500 
1501 #endif /* _LINUX_PGTABLE_H */
1502