1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * Copyright (C) 2012 ARM Ltd.
5 #ifndef __ASM_PGTABLE_H
6 #define __ASM_PGTABLE_H
9 #include <asm/proc-fns.h>
11 #include <asm/memory.h>
13 #include <asm/pgtable-hwdef.h>
14 #include <asm/pgtable-prot.h>
15 #include <asm/tlbflush.h>
20 * VMALLOC_START: beginning of the kernel vmalloc space
21 * VMALLOC_END: extends to the available space below vmemmap, PCI I/O space
24 #define VMALLOC_START (MODULES_END)
25 #define VMALLOC_END (VMEMMAP_START - SZ_256M)
27 #define vmemmap ((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
31 #include <asm/cmpxchg.h>
32 #include <asm/fixmap.h>
33 #include <linux/mmdebug.h>
34 #include <linux/mm_types.h>
35 #include <linux/sched.h>
36 #include <linux/page_table_check.h>
38 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
39 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
41 /* Set stride and tlb_level in flush_*_tlb_range */
42 #define flush_pmd_tlb_range(vma, addr, end) \
43 __flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
44 #define flush_pud_tlb_range(vma, addr, end) \
45 __flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
46 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
49 * Outside of a few very special situations (e.g. hibernation), we always
50 * use broadcast TLB invalidation instructions, therefore a spurious page
51 * fault on one CPU which has been handled concurrently by another CPU
52 * does not need to perform additional invalidation.
54 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
57 * ZERO_PAGE is a global shared page that is always zero: used
58 * for zero-mapped memory areas etc..
60 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
61 #define ZERO_PAGE(vaddr) phys_to_page(__pa_symbol(empty_zero_page))
63 #define pte_ERROR(e) \
64 pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
67 * Macros to convert between a physical address and its placement in a
68 * page table entry, taking care of 52-bit addresses.
70 #ifdef CONFIG_ARM64_PA_BITS_52
71 static inline phys_addr_t __pte_to_phys(pte_t pte)
73 return (pte_val(pte) & PTE_ADDR_LOW) |
74 ((pte_val(pte) & PTE_ADDR_HIGH) << 36);
76 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
78 return (phys | (phys >> 36)) & PTE_ADDR_MASK;
81 #define __pte_to_phys(pte) (pte_val(pte) & PTE_ADDR_MASK)
82 #define __phys_to_pte_val(phys) (phys)
85 #define pte_pfn(pte) (__pte_to_phys(pte) >> PAGE_SHIFT)
86 #define pfn_pte(pfn,prot) \
87 __pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
89 #define pte_none(pte) (!pte_val(pte))
90 #define pte_clear(mm,addr,ptep) set_pte(ptep, __pte(0))
91 #define pte_page(pte) (pfn_to_page(pte_pfn(pte)))
94 * The following only work if pte_present(). Undefined behaviour otherwise.
96 #define pte_present(pte) (!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
97 #define pte_young(pte) (!!(pte_val(pte) & PTE_AF))
98 #define pte_special(pte) (!!(pte_val(pte) & PTE_SPECIAL))
99 #define pte_write(pte) (!!(pte_val(pte) & PTE_WRITE))
100 #define pte_user(pte) (!!(pte_val(pte) & PTE_USER))
101 #define pte_user_exec(pte) (!(pte_val(pte) & PTE_UXN))
102 #define pte_cont(pte) (!!(pte_val(pte) & PTE_CONT))
103 #define pte_devmap(pte) (!!(pte_val(pte) & PTE_DEVMAP))
104 #define pte_tagged(pte) ((pte_val(pte) & PTE_ATTRINDX_MASK) == \
105 PTE_ATTRINDX(MT_NORMAL_TAGGED))
107 #define pte_cont_addr_end(addr, end) \
108 ({ unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK; \
109 (__boundary - 1 < (end) - 1) ? __boundary : (end); \
112 #define pmd_cont_addr_end(addr, end) \
113 ({ unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK; \
114 (__boundary - 1 < (end) - 1) ? __boundary : (end); \
117 #define pte_hw_dirty(pte) (pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
118 #define pte_sw_dirty(pte) (!!(pte_val(pte) & PTE_DIRTY))
119 #define pte_dirty(pte) (pte_sw_dirty(pte) || pte_hw_dirty(pte))
121 #define pte_valid(pte) (!!(pte_val(pte) & PTE_VALID))
123 * Execute-only user mappings do not have the PTE_USER bit set. All valid
124 * kernel mappings have the PTE_UXN bit set.
126 #define pte_valid_not_user(pte) \
127 ((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
129 * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
130 * so that we don't erroneously return false for pages that have been
131 * remapped as PROT_NONE but are yet to be flushed from the TLB.
132 * Note that we can't make any assumptions based on the state of the access
133 * flag, since ptep_clear_flush_young() elides a DSB when invalidating the
136 #define pte_accessible(mm, pte) \
137 (mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
140 * p??_access_permitted() is true for valid user mappings (PTE_USER
141 * bit set, subject to the write permission check). For execute-only
142 * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
143 * not set) must return false. PROT_NONE mappings do not have the
146 #define pte_access_permitted(pte, write) \
147 (((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
148 #define pmd_access_permitted(pmd, write) \
149 (pte_access_permitted(pmd_pte(pmd), (write)))
150 #define pud_access_permitted(pud, write) \
151 (pte_access_permitted(pud_pte(pud), (write)))
153 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
155 pte_val(pte) &= ~pgprot_val(prot);
159 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
161 pte_val(pte) |= pgprot_val(prot);
165 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
167 pmd_val(pmd) &= ~pgprot_val(prot);
171 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
173 pmd_val(pmd) |= pgprot_val(prot);
177 static inline pte_t pte_mkwrite(pte_t pte)
179 pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
180 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
184 static inline pte_t pte_mkclean(pte_t pte)
186 pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
187 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
192 static inline pte_t pte_mkdirty(pte_t pte)
194 pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
197 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
202 static inline pte_t pte_wrprotect(pte_t pte)
205 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
206 * clear), set the PTE_DIRTY bit.
208 if (pte_hw_dirty(pte))
209 pte = pte_mkdirty(pte);
211 pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
212 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
216 static inline pte_t pte_mkold(pte_t pte)
218 return clear_pte_bit(pte, __pgprot(PTE_AF));
221 static inline pte_t pte_mkyoung(pte_t pte)
223 return set_pte_bit(pte, __pgprot(PTE_AF));
226 static inline pte_t pte_mkspecial(pte_t pte)
228 return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
231 static inline pte_t pte_mkcont(pte_t pte)
233 pte = set_pte_bit(pte, __pgprot(PTE_CONT));
234 return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
237 static inline pte_t pte_mknoncont(pte_t pte)
239 return clear_pte_bit(pte, __pgprot(PTE_CONT));
242 static inline pte_t pte_mkpresent(pte_t pte)
244 return set_pte_bit(pte, __pgprot(PTE_VALID));
247 static inline pmd_t pmd_mkcont(pmd_t pmd)
249 return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
252 static inline pte_t pte_mkdevmap(pte_t pte)
254 return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
257 static inline void set_pte(pte_t *ptep, pte_t pte)
259 WRITE_ONCE(*ptep, pte);
262 * Only if the new pte is valid and kernel, otherwise TLB maintenance
263 * or update_mmu_cache() have the necessary barriers.
265 if (pte_valid_not_user(pte)) {
271 extern void __sync_icache_dcache(pte_t pteval);
274 * PTE bits configuration in the presence of hardware Dirty Bit Management
275 * (PTE_WRITE == PTE_DBM):
277 * Dirty Writable | PTE_RDONLY PTE_WRITE PTE_DIRTY (sw)
283 * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
284 * the page fault mechanism. Checking the dirty status of a pte becomes:
286 * PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
289 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
294 if (!IS_ENABLED(CONFIG_DEBUG_VM))
297 old_pte = READ_ONCE(*ptep);
299 if (!pte_valid(old_pte) || !pte_valid(pte))
301 if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
305 * Check for potential race with hardware updates of the pte
306 * (ptep_set_access_flags safely changes valid ptes without going
307 * through an invalid entry).
309 VM_WARN_ONCE(!pte_young(pte),
310 "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
311 __func__, pte_val(old_pte), pte_val(pte));
312 VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
313 "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
314 __func__, pte_val(old_pte), pte_val(pte));
317 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
318 pte_t *ptep, pte_t pte)
320 if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
321 __sync_icache_dcache(pte);
324 * If the PTE would provide user space access to the tags associated
325 * with it then ensure that the MTE tags are synchronised. Although
326 * pte_access_permitted() returns false for exec only mappings, they
327 * don't expose tags (instruction fetches don't check tags).
329 if (system_supports_mte() && pte_access_permitted(pte, false) &&
331 pte_t old_pte = READ_ONCE(*ptep);
333 * We only need to synchronise if the new PTE has tags enabled
334 * or if swapping in (in which case another mapping may have
335 * set tags in the past even if this PTE isn't tagged).
336 * (!pte_none() && !pte_present()) is an open coded version of
339 if (pte_tagged(pte) || (!pte_none(old_pte) && !pte_present(old_pte)))
340 mte_sync_tags(old_pte, pte);
343 __check_racy_pte_update(mm, ptep, pte);
348 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
349 pte_t *ptep, pte_t pte)
351 page_table_check_pte_set(mm, addr, ptep, pte);
352 return __set_pte_at(mm, addr, ptep, pte);
356 * Huge pte definitions.
358 #define pte_mkhuge(pte) (__pte(pte_val(pte) & ~PTE_TABLE_BIT))
361 * Hugetlb definitions.
363 #define HUGE_MAX_HSTATE 4
364 #define HPAGE_SHIFT PMD_SHIFT
365 #define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
366 #define HPAGE_MASK (~(HPAGE_SIZE - 1))
367 #define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
369 static inline pte_t pgd_pte(pgd_t pgd)
371 return __pte(pgd_val(pgd));
374 static inline pte_t p4d_pte(p4d_t p4d)
376 return __pte(p4d_val(p4d));
379 static inline pte_t pud_pte(pud_t pud)
381 return __pte(pud_val(pud));
384 static inline pud_t pte_pud(pte_t pte)
386 return __pud(pte_val(pte));
389 static inline pmd_t pud_pmd(pud_t pud)
391 return __pmd(pud_val(pud));
394 static inline pte_t pmd_pte(pmd_t pmd)
396 return __pte(pmd_val(pmd));
399 static inline pmd_t pte_pmd(pte_t pte)
401 return __pmd(pte_val(pte));
404 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
406 return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
409 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
411 return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
414 #define __HAVE_ARCH_PTE_SWP_EXCLUSIVE
415 static inline pte_t pte_swp_mkexclusive(pte_t pte)
417 return set_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
420 static inline int pte_swp_exclusive(pte_t pte)
422 return pte_val(pte) & PTE_SWP_EXCLUSIVE;
425 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
427 return clear_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
430 #ifdef CONFIG_NUMA_BALANCING
432 * See the comment in include/linux/pgtable.h
434 static inline int pte_protnone(pte_t pte)
436 return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
439 static inline int pmd_protnone(pmd_t pmd)
441 return pte_protnone(pmd_pte(pmd));
445 #define pmd_present_invalid(pmd) (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
447 static inline int pmd_present(pmd_t pmd)
449 return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
456 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
457 static inline int pmd_trans_huge(pmd_t pmd)
459 return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
461 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
463 #define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd))
464 #define pmd_young(pmd) pte_young(pmd_pte(pmd))
465 #define pmd_valid(pmd) pte_valid(pmd_pte(pmd))
466 #define pmd_user(pmd) pte_user(pmd_pte(pmd))
467 #define pmd_user_exec(pmd) pte_user_exec(pmd_pte(pmd))
468 #define pmd_cont(pmd) pte_cont(pmd_pte(pmd))
469 #define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
470 #define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
471 #define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
472 #define pmd_mkclean(pmd) pte_pmd(pte_mkclean(pmd_pte(pmd)))
473 #define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
474 #define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
476 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
478 pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
479 pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
484 #define pmd_thp_or_huge(pmd) (pmd_huge(pmd) || pmd_trans_huge(pmd))
486 #define pmd_write(pmd) pte_write(pmd_pte(pmd))
488 #define pmd_mkhuge(pmd) (__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
490 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
491 #define pmd_devmap(pmd) pte_devmap(pmd_pte(pmd))
493 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
495 return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
498 #define __pmd_to_phys(pmd) __pte_to_phys(pmd_pte(pmd))
499 #define __phys_to_pmd_val(phys) __phys_to_pte_val(phys)
500 #define pmd_pfn(pmd) ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
501 #define pfn_pmd(pfn,prot) __pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
502 #define mk_pmd(page,prot) pfn_pmd(page_to_pfn(page),prot)
504 #define pud_young(pud) pte_young(pud_pte(pud))
505 #define pud_mkyoung(pud) pte_pud(pte_mkyoung(pud_pte(pud)))
506 #define pud_write(pud) pte_write(pud_pte(pud))
508 #define pud_mkhuge(pud) (__pud(pud_val(pud) & ~PUD_TABLE_BIT))
510 #define __pud_to_phys(pud) __pte_to_phys(pud_pte(pud))
511 #define __phys_to_pud_val(phys) __phys_to_pte_val(phys)
512 #define pud_pfn(pud) ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
513 #define pfn_pud(pfn,prot) __pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
515 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
516 pmd_t *pmdp, pmd_t pmd)
518 page_table_check_pmd_set(mm, addr, pmdp, pmd);
519 return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd));
522 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
523 pud_t *pudp, pud_t pud)
525 page_table_check_pud_set(mm, addr, pudp, pud);
526 return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud));
529 #define __p4d_to_phys(p4d) __pte_to_phys(p4d_pte(p4d))
530 #define __phys_to_p4d_val(phys) __phys_to_pte_val(phys)
532 #define __pgd_to_phys(pgd) __pte_to_phys(pgd_pte(pgd))
533 #define __phys_to_pgd_val(phys) __phys_to_pte_val(phys)
535 #define __pgprot_modify(prot,mask,bits) \
536 __pgprot((pgprot_val(prot) & ~(mask)) | (bits))
538 #define pgprot_nx(prot) \
539 __pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
542 * Mark the prot value as uncacheable and unbufferable.
544 #define pgprot_noncached(prot) \
545 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
546 #define pgprot_writecombine(prot) \
547 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
548 #define pgprot_device(prot) \
549 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
550 #define pgprot_tagged(prot) \
551 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
552 #define pgprot_mhp pgprot_tagged
554 * DMA allocations for non-coherent devices use what the Arm architecture calls
555 * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
556 * and merging of writes. This is different from "Device-nGnR[nE]" memory which
557 * is intended for MMIO and thus forbids speculation, preserves access size,
558 * requires strict alignment and can also force write responses to come from the
561 #define pgprot_dmacoherent(prot) \
562 __pgprot_modify(prot, PTE_ATTRINDX_MASK, \
563 PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
565 #define __HAVE_PHYS_MEM_ACCESS_PROT
567 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
568 unsigned long size, pgprot_t vma_prot);
570 #define pmd_none(pmd) (!pmd_val(pmd))
572 #define pmd_table(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \
574 #define pmd_sect(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \
576 #define pmd_leaf(pmd) (pmd_present(pmd) && !pmd_table(pmd))
577 #define pmd_bad(pmd) (!pmd_table(pmd))
579 #define pmd_leaf_size(pmd) (pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
580 #define pte_leaf_size(pte) (pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
582 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
583 static inline bool pud_sect(pud_t pud) { return false; }
584 static inline bool pud_table(pud_t pud) { return true; }
586 #define pud_sect(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \
588 #define pud_table(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \
592 extern pgd_t init_pg_dir[PTRS_PER_PGD];
593 extern pgd_t init_pg_end[];
594 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
595 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
596 extern pgd_t idmap_pg_end[];
597 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
598 extern pgd_t reserved_pg_dir[PTRS_PER_PGD];
600 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
602 static inline bool in_swapper_pgdir(void *addr)
604 return ((unsigned long)addr & PAGE_MASK) ==
605 ((unsigned long)swapper_pg_dir & PAGE_MASK);
608 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
610 #ifdef __PAGETABLE_PMD_FOLDED
611 if (in_swapper_pgdir(pmdp)) {
612 set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
615 #endif /* __PAGETABLE_PMD_FOLDED */
617 WRITE_ONCE(*pmdp, pmd);
619 if (pmd_valid(pmd)) {
625 static inline void pmd_clear(pmd_t *pmdp)
627 set_pmd(pmdp, __pmd(0));
630 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
632 return __pmd_to_phys(pmd);
635 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
637 return (unsigned long)__va(pmd_page_paddr(pmd));
640 /* Find an entry in the third-level page table. */
641 #define pte_offset_phys(dir,addr) (pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
643 #define pte_set_fixmap(addr) ((pte_t *)set_fixmap_offset(FIX_PTE, addr))
644 #define pte_set_fixmap_offset(pmd, addr) pte_set_fixmap(pte_offset_phys(pmd, addr))
645 #define pte_clear_fixmap() clear_fixmap(FIX_PTE)
647 #define pmd_page(pmd) phys_to_page(__pmd_to_phys(pmd))
649 /* use ONLY for statically allocated translation tables */
650 #define pte_offset_kimg(dir,addr) ((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
653 * Conversion functions: convert a page and protection to a page entry,
654 * and a page entry and page directory to the page they refer to.
656 #define mk_pte(page,prot) pfn_pte(page_to_pfn(page),prot)
658 #if CONFIG_PGTABLE_LEVELS > 2
660 #define pmd_ERROR(e) \
661 pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
663 #define pud_none(pud) (!pud_val(pud))
664 #define pud_bad(pud) (!pud_table(pud))
665 #define pud_present(pud) pte_present(pud_pte(pud))
666 #define pud_leaf(pud) (pud_present(pud) && !pud_table(pud))
667 #define pud_valid(pud) pte_valid(pud_pte(pud))
668 #define pud_user(pud) pte_user(pud_pte(pud))
671 static inline void set_pud(pud_t *pudp, pud_t pud)
673 #ifdef __PAGETABLE_PUD_FOLDED
674 if (in_swapper_pgdir(pudp)) {
675 set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
678 #endif /* __PAGETABLE_PUD_FOLDED */
680 WRITE_ONCE(*pudp, pud);
682 if (pud_valid(pud)) {
688 static inline void pud_clear(pud_t *pudp)
690 set_pud(pudp, __pud(0));
693 static inline phys_addr_t pud_page_paddr(pud_t pud)
695 return __pud_to_phys(pud);
698 static inline pmd_t *pud_pgtable(pud_t pud)
700 return (pmd_t *)__va(pud_page_paddr(pud));
703 /* Find an entry in the second-level page table. */
704 #define pmd_offset_phys(dir, addr) (pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
706 #define pmd_set_fixmap(addr) ((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
707 #define pmd_set_fixmap_offset(pud, addr) pmd_set_fixmap(pmd_offset_phys(pud, addr))
708 #define pmd_clear_fixmap() clear_fixmap(FIX_PMD)
710 #define pud_page(pud) phys_to_page(__pud_to_phys(pud))
712 /* use ONLY for statically allocated translation tables */
713 #define pmd_offset_kimg(dir,addr) ((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
717 #define pud_page_paddr(pud) ({ BUILD_BUG(); 0; })
719 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
720 #define pmd_set_fixmap(addr) NULL
721 #define pmd_set_fixmap_offset(pudp, addr) ((pmd_t *)pudp)
722 #define pmd_clear_fixmap()
724 #define pmd_offset_kimg(dir,addr) ((pmd_t *)dir)
726 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
728 #if CONFIG_PGTABLE_LEVELS > 3
730 #define pud_ERROR(e) \
731 pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
733 #define p4d_none(p4d) (!p4d_val(p4d))
734 #define p4d_bad(p4d) (!(p4d_val(p4d) & 2))
735 #define p4d_present(p4d) (p4d_val(p4d))
737 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
739 if (in_swapper_pgdir(p4dp)) {
740 set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
744 WRITE_ONCE(*p4dp, p4d);
749 static inline void p4d_clear(p4d_t *p4dp)
751 set_p4d(p4dp, __p4d(0));
754 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
756 return __p4d_to_phys(p4d);
759 static inline pud_t *p4d_pgtable(p4d_t p4d)
761 return (pud_t *)__va(p4d_page_paddr(p4d));
764 /* Find an entry in the first-level page table. */
765 #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
767 #define pud_set_fixmap(addr) ((pud_t *)set_fixmap_offset(FIX_PUD, addr))
768 #define pud_set_fixmap_offset(p4d, addr) pud_set_fixmap(pud_offset_phys(p4d, addr))
769 #define pud_clear_fixmap() clear_fixmap(FIX_PUD)
771 #define p4d_page(p4d) pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
773 /* use ONLY for statically allocated translation tables */
774 #define pud_offset_kimg(dir,addr) ((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
778 #define p4d_page_paddr(p4d) ({ BUILD_BUG(); 0;})
779 #define pgd_page_paddr(pgd) ({ BUILD_BUG(); 0;})
781 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
782 #define pud_set_fixmap(addr) NULL
783 #define pud_set_fixmap_offset(pgdp, addr) ((pud_t *)pgdp)
784 #define pud_clear_fixmap()
786 #define pud_offset_kimg(dir,addr) ((pud_t *)dir)
788 #endif /* CONFIG_PGTABLE_LEVELS > 3 */
790 #define pgd_ERROR(e) \
791 pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
793 #define pgd_set_fixmap(addr) ((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
794 #define pgd_clear_fixmap() clear_fixmap(FIX_PGD)
796 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
799 * Normal and Normal-Tagged are two different memory types and indices
800 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
802 const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
803 PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
805 /* preserve the hardware dirty information */
806 if (pte_hw_dirty(pte))
807 pte = pte_mkdirty(pte);
808 pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
812 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
814 return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
817 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
818 extern int ptep_set_access_flags(struct vm_area_struct *vma,
819 unsigned long address, pte_t *ptep,
820 pte_t entry, int dirty);
822 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
823 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
824 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
825 unsigned long address, pmd_t *pmdp,
826 pmd_t entry, int dirty)
828 return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
831 static inline int pud_devmap(pud_t pud)
836 static inline int pgd_devmap(pgd_t pgd)
842 #ifdef CONFIG_PAGE_TABLE_CHECK
843 static inline bool pte_user_accessible_page(pte_t pte)
845 return pte_present(pte) && (pte_user(pte) || pte_user_exec(pte));
848 static inline bool pmd_user_accessible_page(pmd_t pmd)
850 return pmd_present(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
853 static inline bool pud_user_accessible_page(pud_t pud)
855 return pud_present(pud) && pud_user(pud);
860 * Atomic pte/pmd modifications.
862 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
863 static inline int __ptep_test_and_clear_young(pte_t *ptep)
867 pte = READ_ONCE(*ptep);
870 pte = pte_mkold(pte);
871 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
872 pte_val(old_pte), pte_val(pte));
873 } while (pte_val(pte) != pte_val(old_pte));
875 return pte_young(pte);
878 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
879 unsigned long address,
882 return __ptep_test_and_clear_young(ptep);
885 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
886 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
887 unsigned long address, pte_t *ptep)
889 int young = ptep_test_and_clear_young(vma, address, ptep);
893 * We can elide the trailing DSB here since the worst that can
894 * happen is that a CPU continues to use the young entry in its
895 * TLB and we mistakenly reclaim the associated page. The
896 * window for such an event is bounded by the next
897 * context-switch, which provides a DSB to complete the TLB
900 flush_tlb_page_nosync(vma, address);
906 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
907 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
908 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
909 unsigned long address,
912 return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
914 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
916 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
917 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
918 unsigned long address, pte_t *ptep)
920 pte_t pte = __pte(xchg_relaxed(&pte_val(*ptep), 0));
922 page_table_check_pte_clear(mm, address, pte);
927 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
928 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
929 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
930 unsigned long address, pmd_t *pmdp)
932 pmd_t pmd = __pmd(xchg_relaxed(&pmd_val(*pmdp), 0));
934 page_table_check_pmd_clear(mm, address, pmd);
938 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
941 * ptep_set_wrprotect - mark read-only while trasferring potential hardware
942 * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
944 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
945 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
949 pte = READ_ONCE(*ptep);
952 pte = pte_wrprotect(pte);
953 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
954 pte_val(old_pte), pte_val(pte));
955 } while (pte_val(pte) != pte_val(old_pte));
958 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
959 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
960 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
961 unsigned long address, pmd_t *pmdp)
963 ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
966 #define pmdp_establish pmdp_establish
967 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
968 unsigned long address, pmd_t *pmdp, pmd_t pmd)
970 page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
971 return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
976 * Encode and decode a swap entry:
977 * bits 0-1: present (must be zero)
978 * bits 2: remember PG_anon_exclusive
979 * bits 3-7: swap type
980 * bits 8-57: swap offset
981 * bit 58: PTE_PROT_NONE (must be zero)
983 #define __SWP_TYPE_SHIFT 3
984 #define __SWP_TYPE_BITS 5
985 #define __SWP_OFFSET_BITS 50
986 #define __SWP_TYPE_MASK ((1 << __SWP_TYPE_BITS) - 1)
987 #define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
988 #define __SWP_OFFSET_MASK ((1UL << __SWP_OFFSET_BITS) - 1)
990 #define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
991 #define __swp_offset(x) (((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
992 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
994 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
995 #define __swp_entry_to_pte(swp) ((pte_t) { (swp).val })
997 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
998 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
999 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
1000 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1003 * Ensure that there are not more swap files than can be encoded in the kernel
1006 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1008 extern int kern_addr_valid(unsigned long addr);
1010 #ifdef CONFIG_ARM64_MTE
1012 #define __HAVE_ARCH_PREPARE_TO_SWAP
1013 static inline int arch_prepare_to_swap(struct page *page)
1015 if (system_supports_mte())
1016 return mte_save_tags(page);
1020 #define __HAVE_ARCH_SWAP_INVALIDATE
1021 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1023 if (system_supports_mte())
1024 mte_invalidate_tags(type, offset);
1027 static inline void arch_swap_invalidate_area(int type)
1029 if (system_supports_mte())
1030 mte_invalidate_tags_area(type);
1033 #define __HAVE_ARCH_SWAP_RESTORE
1034 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1036 if (system_supports_mte() && mte_restore_tags(entry, &folio->page))
1037 set_bit(PG_mte_tagged, &folio->flags);
1040 #endif /* CONFIG_ARM64_MTE */
1043 * On AArch64, the cache coherency is handled via the set_pte_at() function.
1045 static inline void update_mmu_cache(struct vm_area_struct *vma,
1046 unsigned long addr, pte_t *ptep)
1049 * We don't do anything here, so there's a very small chance of
1050 * us retaking a user fault which we just fixed up. The alternative
1051 * is doing a dsb(ishst), but that penalises the fastpath.
1055 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
1057 #ifdef CONFIG_ARM64_PA_BITS_52
1058 #define phys_to_ttbr(addr) (((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
1060 #define phys_to_ttbr(addr) (addr)
1064 * On arm64 without hardware Access Flag, copying from user will fail because
1065 * the pte is old and cannot be marked young. So we always end up with zeroed
1066 * page after fork() + CoW for pfn mappings. We don't always have a
1067 * hardware-managed access flag on arm64.
1069 static inline bool arch_faults_on_old_pte(void)
1071 /* The register read below requires a stable CPU to make any sense */
1074 return !cpu_has_hw_af();
1076 #define arch_faults_on_old_pte arch_faults_on_old_pte
1079 * Experimentally, it's cheap to set the access flag in hardware and we
1080 * benefit from prefaulting mappings as 'old' to start with.
1082 static inline bool arch_wants_old_prefaulted_pte(void)
1084 return !arch_faults_on_old_pte();
1086 #define arch_wants_old_prefaulted_pte arch_wants_old_prefaulted_pte
1088 static inline bool pud_sect_supported(void)
1090 return PAGE_SIZE == SZ_4K;
1094 #endif /* !__ASSEMBLY__ */
1096 #endif /* __ASM_PGTABLE_H */