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))
29 #define FIRST_USER_ADDRESS 0UL
33 #include <asm/cmpxchg.h>
34 #include <asm/fixmap.h>
35 #include <linux/mmdebug.h>
36 #include <linux/mm_types.h>
37 #include <linux/sched.h>
39 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
40 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
42 /* Set stride and tlb_level in flush_*_tlb_range */
43 #define flush_pmd_tlb_range(vma, addr, end) \
44 __flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
45 #define flush_pud_tlb_range(vma, addr, end) \
46 __flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
47 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
50 * Outside of a few very special situations (e.g. hibernation), we always
51 * use broadcast TLB invalidation instructions, therefore a spurious page
52 * fault on one CPU which has been handled concurrently by another CPU
53 * does not need to perform additional invalidation.
55 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
58 * ZERO_PAGE is a global shared page that is always zero: used
59 * for zero-mapped memory areas etc..
61 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
62 #define ZERO_PAGE(vaddr) phys_to_page(__pa_symbol(empty_zero_page))
64 #define pte_ERROR(e) \
65 pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
68 * Macros to convert between a physical address and its placement in a
69 * page table entry, taking care of 52-bit addresses.
71 #ifdef CONFIG_ARM64_PA_BITS_52
72 #define __pte_to_phys(pte) \
73 ((pte_val(pte) & PTE_ADDR_LOW) | ((pte_val(pte) & PTE_ADDR_HIGH) << 36))
74 #define __phys_to_pte_val(phys) (((phys) | ((phys) >> 36)) & PTE_ADDR_MASK)
76 #define __pte_to_phys(pte) (pte_val(pte) & PTE_ADDR_MASK)
77 #define __phys_to_pte_val(phys) (phys)
80 #define pte_pfn(pte) (__pte_to_phys(pte) >> PAGE_SHIFT)
81 #define pfn_pte(pfn,prot) \
82 __pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
84 #define pte_none(pte) (!pte_val(pte))
85 #define pte_clear(mm,addr,ptep) set_pte(ptep, __pte(0))
86 #define pte_page(pte) (pfn_to_page(pte_pfn(pte)))
89 * The following only work if pte_present(). Undefined behaviour otherwise.
91 #define pte_present(pte) (!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
92 #define pte_young(pte) (!!(pte_val(pte) & PTE_AF))
93 #define pte_special(pte) (!!(pte_val(pte) & PTE_SPECIAL))
94 #define pte_write(pte) (!!(pte_val(pte) & PTE_WRITE))
95 #define pte_user_exec(pte) (!(pte_val(pte) & PTE_UXN))
96 #define pte_cont(pte) (!!(pte_val(pte) & PTE_CONT))
97 #define pte_devmap(pte) (!!(pte_val(pte) & PTE_DEVMAP))
98 #define pte_tagged(pte) ((pte_val(pte) & PTE_ATTRINDX_MASK) == \
99 PTE_ATTRINDX(MT_NORMAL_TAGGED))
101 #define pte_cont_addr_end(addr, end) \
102 ({ unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK; \
103 (__boundary - 1 < (end) - 1) ? __boundary : (end); \
106 #define pmd_cont_addr_end(addr, end) \
107 ({ unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK; \
108 (__boundary - 1 < (end) - 1) ? __boundary : (end); \
111 #define pte_hw_dirty(pte) (pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
112 #define pte_sw_dirty(pte) (!!(pte_val(pte) & PTE_DIRTY))
113 #define pte_dirty(pte) (pte_sw_dirty(pte) || pte_hw_dirty(pte))
115 #define pte_valid(pte) (!!(pte_val(pte) & PTE_VALID))
117 * Execute-only user mappings do not have the PTE_USER bit set. All valid
118 * kernel mappings have the PTE_UXN bit set.
120 #define pte_valid_not_user(pte) \
121 ((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
123 * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
124 * so that we don't erroneously return false for pages that have been
125 * remapped as PROT_NONE but are yet to be flushed from the TLB.
126 * Note that we can't make any assumptions based on the state of the access
127 * flag, since ptep_clear_flush_young() elides a DSB when invalidating the
130 #define pte_accessible(mm, pte) \
131 (mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
134 * p??_access_permitted() is true for valid user mappings (PTE_USER
135 * bit set, subject to the write permission check). For execute-only
136 * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
137 * not set) must return false. PROT_NONE mappings do not have the
140 #define pte_access_permitted(pte, write) \
141 (((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
142 #define pmd_access_permitted(pmd, write) \
143 (pte_access_permitted(pmd_pte(pmd), (write)))
144 #define pud_access_permitted(pud, write) \
145 (pte_access_permitted(pud_pte(pud), (write)))
147 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
149 pte_val(pte) &= ~pgprot_val(prot);
153 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
155 pte_val(pte) |= pgprot_val(prot);
159 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
161 pmd_val(pmd) &= ~pgprot_val(prot);
165 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
167 pmd_val(pmd) |= pgprot_val(prot);
171 static inline pte_t pte_mkwrite(pte_t pte)
173 pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
174 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
178 static inline pte_t pte_mkclean(pte_t pte)
180 pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
181 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
186 static inline pte_t pte_mkdirty(pte_t pte)
188 pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
191 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
196 static inline pte_t pte_wrprotect(pte_t pte)
199 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
200 * clear), set the PTE_DIRTY bit.
202 if (pte_hw_dirty(pte))
203 pte = pte_mkdirty(pte);
205 pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
206 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
210 static inline pte_t pte_mkold(pte_t pte)
212 return clear_pte_bit(pte, __pgprot(PTE_AF));
215 static inline pte_t pte_mkyoung(pte_t pte)
217 return set_pte_bit(pte, __pgprot(PTE_AF));
220 static inline pte_t pte_mkspecial(pte_t pte)
222 return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
225 static inline pte_t pte_mkcont(pte_t pte)
227 pte = set_pte_bit(pte, __pgprot(PTE_CONT));
228 return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
231 static inline pte_t pte_mknoncont(pte_t pte)
233 return clear_pte_bit(pte, __pgprot(PTE_CONT));
236 static inline pte_t pte_mkpresent(pte_t pte)
238 return set_pte_bit(pte, __pgprot(PTE_VALID));
241 static inline pmd_t pmd_mkcont(pmd_t pmd)
243 return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
246 static inline pte_t pte_mkdevmap(pte_t pte)
248 return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
251 static inline void set_pte(pte_t *ptep, pte_t pte)
253 WRITE_ONCE(*ptep, pte);
256 * Only if the new pte is valid and kernel, otherwise TLB maintenance
257 * or update_mmu_cache() have the necessary barriers.
259 if (pte_valid_not_user(pte)) {
265 extern void __sync_icache_dcache(pte_t pteval);
268 * PTE bits configuration in the presence of hardware Dirty Bit Management
269 * (PTE_WRITE == PTE_DBM):
271 * Dirty Writable | PTE_RDONLY PTE_WRITE PTE_DIRTY (sw)
277 * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
278 * the page fault mechanism. Checking the dirty status of a pte becomes:
280 * PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
283 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
288 if (!IS_ENABLED(CONFIG_DEBUG_VM))
291 old_pte = READ_ONCE(*ptep);
293 if (!pte_valid(old_pte) || !pte_valid(pte))
295 if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
299 * Check for potential race with hardware updates of the pte
300 * (ptep_set_access_flags safely changes valid ptes without going
301 * through an invalid entry).
303 VM_WARN_ONCE(!pte_young(pte),
304 "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
305 __func__, pte_val(old_pte), pte_val(pte));
306 VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
307 "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
308 __func__, pte_val(old_pte), pte_val(pte));
311 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
312 pte_t *ptep, pte_t pte)
314 if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
315 __sync_icache_dcache(pte);
318 * If the PTE would provide user space access to the tags associated
319 * with it then ensure that the MTE tags are synchronised. Although
320 * pte_access_permitted() returns false for exec only mappings, they
321 * don't expose tags (instruction fetches don't check tags).
323 if (system_supports_mte() && pte_access_permitted(pte, false) &&
325 pte_t old_pte = READ_ONCE(*ptep);
327 * We only need to synchronise if the new PTE has tags enabled
328 * or if swapping in (in which case another mapping may have
329 * set tags in the past even if this PTE isn't tagged).
330 * (!pte_none() && !pte_present()) is an open coded version of
333 if (pte_tagged(pte) || (!pte_none(old_pte) && !pte_present(old_pte)))
334 mte_sync_tags(old_pte, pte);
337 __check_racy_pte_update(mm, ptep, pte);
343 * Huge pte definitions.
345 #define pte_mkhuge(pte) (__pte(pte_val(pte) & ~PTE_TABLE_BIT))
348 * Hugetlb definitions.
350 #define HUGE_MAX_HSTATE 4
351 #define HPAGE_SHIFT PMD_SHIFT
352 #define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
353 #define HPAGE_MASK (~(HPAGE_SIZE - 1))
354 #define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
356 static inline pte_t pgd_pte(pgd_t pgd)
358 return __pte(pgd_val(pgd));
361 static inline pte_t p4d_pte(p4d_t p4d)
363 return __pte(p4d_val(p4d));
366 static inline pte_t pud_pte(pud_t pud)
368 return __pte(pud_val(pud));
371 static inline pud_t pte_pud(pte_t pte)
373 return __pud(pte_val(pte));
376 static inline pmd_t pud_pmd(pud_t pud)
378 return __pmd(pud_val(pud));
381 static inline pte_t pmd_pte(pmd_t pmd)
383 return __pte(pmd_val(pmd));
386 static inline pmd_t pte_pmd(pte_t pte)
388 return __pmd(pte_val(pte));
391 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
393 return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
396 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
398 return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
401 #ifdef CONFIG_NUMA_BALANCING
403 * See the comment in include/linux/pgtable.h
405 static inline int pte_protnone(pte_t pte)
407 return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
410 static inline int pmd_protnone(pmd_t pmd)
412 return pte_protnone(pmd_pte(pmd));
416 #define pmd_present_invalid(pmd) (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
418 static inline int pmd_present(pmd_t pmd)
420 return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
427 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
428 static inline int pmd_trans_huge(pmd_t pmd)
430 return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
432 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
434 #define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd))
435 #define pmd_young(pmd) pte_young(pmd_pte(pmd))
436 #define pmd_valid(pmd) pte_valid(pmd_pte(pmd))
437 #define pmd_cont(pmd) pte_cont(pmd_pte(pmd))
438 #define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
439 #define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
440 #define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
441 #define pmd_mkclean(pmd) pte_pmd(pte_mkclean(pmd_pte(pmd)))
442 #define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
443 #define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
445 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
447 pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
448 pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
453 #define pmd_thp_or_huge(pmd) (pmd_huge(pmd) || pmd_trans_huge(pmd))
455 #define pmd_write(pmd) pte_write(pmd_pte(pmd))
457 #define pmd_mkhuge(pmd) (__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
459 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
460 #define pmd_devmap(pmd) pte_devmap(pmd_pte(pmd))
462 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
464 return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
467 #define __pmd_to_phys(pmd) __pte_to_phys(pmd_pte(pmd))
468 #define __phys_to_pmd_val(phys) __phys_to_pte_val(phys)
469 #define pmd_pfn(pmd) ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
470 #define pfn_pmd(pfn,prot) __pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
471 #define mk_pmd(page,prot) pfn_pmd(page_to_pfn(page),prot)
473 #define pud_young(pud) pte_young(pud_pte(pud))
474 #define pud_mkyoung(pud) pte_pud(pte_mkyoung(pud_pte(pud)))
475 #define pud_write(pud) pte_write(pud_pte(pud))
477 #define pud_mkhuge(pud) (__pud(pud_val(pud) & ~PUD_TABLE_BIT))
479 #define __pud_to_phys(pud) __pte_to_phys(pud_pte(pud))
480 #define __phys_to_pud_val(phys) __phys_to_pte_val(phys)
481 #define pud_pfn(pud) ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
482 #define pfn_pud(pfn,prot) __pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
484 #define set_pmd_at(mm, addr, pmdp, pmd) set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
485 #define set_pud_at(mm, addr, pudp, pud) set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud))
487 #define __p4d_to_phys(p4d) __pte_to_phys(p4d_pte(p4d))
488 #define __phys_to_p4d_val(phys) __phys_to_pte_val(phys)
490 #define __pgd_to_phys(pgd) __pte_to_phys(pgd_pte(pgd))
491 #define __phys_to_pgd_val(phys) __phys_to_pte_val(phys)
493 #define __pgprot_modify(prot,mask,bits) \
494 __pgprot((pgprot_val(prot) & ~(mask)) | (bits))
496 #define pgprot_nx(prot) \
497 __pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
500 * Mark the prot value as uncacheable and unbufferable.
502 #define pgprot_noncached(prot) \
503 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
504 #define pgprot_writecombine(prot) \
505 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
506 #define pgprot_device(prot) \
507 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
508 #define pgprot_tagged(prot) \
509 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
510 #define pgprot_mhp pgprot_tagged
512 * DMA allocations for non-coherent devices use what the Arm architecture calls
513 * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
514 * and merging of writes. This is different from "Device-nGnR[nE]" memory which
515 * is intended for MMIO and thus forbids speculation, preserves access size,
516 * requires strict alignment and can also force write responses to come from the
519 #define pgprot_dmacoherent(prot) \
520 __pgprot_modify(prot, PTE_ATTRINDX_MASK, \
521 PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
523 #define __HAVE_PHYS_MEM_ACCESS_PROT
525 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
526 unsigned long size, pgprot_t vma_prot);
528 #define pmd_none(pmd) (!pmd_val(pmd))
530 #define pmd_table(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \
532 #define pmd_sect(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \
534 #define pmd_leaf(pmd) pmd_sect(pmd)
535 #define pmd_bad(pmd) (!pmd_table(pmd))
537 #define pmd_leaf_size(pmd) (pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
538 #define pte_leaf_size(pte) (pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
540 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
541 static inline bool pud_sect(pud_t pud) { return false; }
542 static inline bool pud_table(pud_t pud) { return true; }
544 #define pud_sect(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \
546 #define pud_table(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \
550 extern pgd_t init_pg_dir[PTRS_PER_PGD];
551 extern pgd_t init_pg_end[];
552 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
553 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
554 extern pgd_t idmap_pg_end[];
555 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
556 extern pgd_t reserved_pg_dir[PTRS_PER_PGD];
558 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
560 static inline bool in_swapper_pgdir(void *addr)
562 return ((unsigned long)addr & PAGE_MASK) ==
563 ((unsigned long)swapper_pg_dir & PAGE_MASK);
566 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
568 #ifdef __PAGETABLE_PMD_FOLDED
569 if (in_swapper_pgdir(pmdp)) {
570 set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
573 #endif /* __PAGETABLE_PMD_FOLDED */
575 WRITE_ONCE(*pmdp, pmd);
577 if (pmd_valid(pmd)) {
583 static inline void pmd_clear(pmd_t *pmdp)
585 set_pmd(pmdp, __pmd(0));
588 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
590 return __pmd_to_phys(pmd);
593 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
595 return (unsigned long)__va(pmd_page_paddr(pmd));
598 /* Find an entry in the third-level page table. */
599 #define pte_offset_phys(dir,addr) (pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
601 #define pte_set_fixmap(addr) ((pte_t *)set_fixmap_offset(FIX_PTE, addr))
602 #define pte_set_fixmap_offset(pmd, addr) pte_set_fixmap(pte_offset_phys(pmd, addr))
603 #define pte_clear_fixmap() clear_fixmap(FIX_PTE)
605 #define pmd_page(pmd) phys_to_page(__pmd_to_phys(pmd))
607 /* use ONLY for statically allocated translation tables */
608 #define pte_offset_kimg(dir,addr) ((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
611 * Conversion functions: convert a page and protection to a page entry,
612 * and a page entry and page directory to the page they refer to.
614 #define mk_pte(page,prot) pfn_pte(page_to_pfn(page),prot)
616 #if CONFIG_PGTABLE_LEVELS > 2
618 #define pmd_ERROR(e) \
619 pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
621 #define pud_none(pud) (!pud_val(pud))
622 #define pud_bad(pud) (!pud_table(pud))
623 #define pud_present(pud) pte_present(pud_pte(pud))
624 #define pud_leaf(pud) pud_sect(pud)
625 #define pud_valid(pud) pte_valid(pud_pte(pud))
627 static inline void set_pud(pud_t *pudp, pud_t pud)
629 #ifdef __PAGETABLE_PUD_FOLDED
630 if (in_swapper_pgdir(pudp)) {
631 set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
634 #endif /* __PAGETABLE_PUD_FOLDED */
636 WRITE_ONCE(*pudp, pud);
638 if (pud_valid(pud)) {
644 static inline void pud_clear(pud_t *pudp)
646 set_pud(pudp, __pud(0));
649 static inline phys_addr_t pud_page_paddr(pud_t pud)
651 return __pud_to_phys(pud);
654 static inline unsigned long pud_page_vaddr(pud_t pud)
656 return (unsigned long)__va(pud_page_paddr(pud));
659 /* Find an entry in the second-level page table. */
660 #define pmd_offset_phys(dir, addr) (pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
662 #define pmd_set_fixmap(addr) ((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
663 #define pmd_set_fixmap_offset(pud, addr) pmd_set_fixmap(pmd_offset_phys(pud, addr))
664 #define pmd_clear_fixmap() clear_fixmap(FIX_PMD)
666 #define pud_page(pud) phys_to_page(__pud_to_phys(pud))
668 /* use ONLY for statically allocated translation tables */
669 #define pmd_offset_kimg(dir,addr) ((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
673 #define pud_page_paddr(pud) ({ BUILD_BUG(); 0; })
675 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
676 #define pmd_set_fixmap(addr) NULL
677 #define pmd_set_fixmap_offset(pudp, addr) ((pmd_t *)pudp)
678 #define pmd_clear_fixmap()
680 #define pmd_offset_kimg(dir,addr) ((pmd_t *)dir)
682 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
684 #if CONFIG_PGTABLE_LEVELS > 3
686 #define pud_ERROR(e) \
687 pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
689 #define p4d_none(p4d) (!p4d_val(p4d))
690 #define p4d_bad(p4d) (!(p4d_val(p4d) & 2))
691 #define p4d_present(p4d) (p4d_val(p4d))
693 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
695 if (in_swapper_pgdir(p4dp)) {
696 set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
700 WRITE_ONCE(*p4dp, p4d);
705 static inline void p4d_clear(p4d_t *p4dp)
707 set_p4d(p4dp, __p4d(0));
710 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
712 return __p4d_to_phys(p4d);
715 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
717 return (unsigned long)__va(p4d_page_paddr(p4d));
720 /* Find an entry in the frst-level page table. */
721 #define pud_offset_phys(dir, addr) (p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
723 #define pud_set_fixmap(addr) ((pud_t *)set_fixmap_offset(FIX_PUD, addr))
724 #define pud_set_fixmap_offset(p4d, addr) pud_set_fixmap(pud_offset_phys(p4d, addr))
725 #define pud_clear_fixmap() clear_fixmap(FIX_PUD)
727 #define p4d_page(p4d) pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
729 /* use ONLY for statically allocated translation tables */
730 #define pud_offset_kimg(dir,addr) ((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
734 #define p4d_page_paddr(p4d) ({ BUILD_BUG(); 0;})
735 #define pgd_page_paddr(pgd) ({ BUILD_BUG(); 0;})
737 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
738 #define pud_set_fixmap(addr) NULL
739 #define pud_set_fixmap_offset(pgdp, addr) ((pud_t *)pgdp)
740 #define pud_clear_fixmap()
742 #define pud_offset_kimg(dir,addr) ((pud_t *)dir)
744 #endif /* CONFIG_PGTABLE_LEVELS > 3 */
746 #define pgd_ERROR(e) \
747 pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
749 #define pgd_set_fixmap(addr) ((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
750 #define pgd_clear_fixmap() clear_fixmap(FIX_PGD)
752 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
755 * Normal and Normal-Tagged are two different memory types and indices
756 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
758 const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
759 PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
761 /* preserve the hardware dirty information */
762 if (pte_hw_dirty(pte))
763 pte = pte_mkdirty(pte);
764 pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
768 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
770 return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
773 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
774 extern int ptep_set_access_flags(struct vm_area_struct *vma,
775 unsigned long address, pte_t *ptep,
776 pte_t entry, int dirty);
778 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
779 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
780 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
781 unsigned long address, pmd_t *pmdp,
782 pmd_t entry, int dirty)
784 return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
787 static inline int pud_devmap(pud_t pud)
792 static inline int pgd_devmap(pgd_t pgd)
799 * Atomic pte/pmd modifications.
801 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
802 static inline int __ptep_test_and_clear_young(pte_t *ptep)
806 pte = READ_ONCE(*ptep);
809 pte = pte_mkold(pte);
810 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
811 pte_val(old_pte), pte_val(pte));
812 } while (pte_val(pte) != pte_val(old_pte));
814 return pte_young(pte);
817 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
818 unsigned long address,
821 return __ptep_test_and_clear_young(ptep);
824 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
825 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
826 unsigned long address, pte_t *ptep)
828 int young = ptep_test_and_clear_young(vma, address, ptep);
832 * We can elide the trailing DSB here since the worst that can
833 * happen is that a CPU continues to use the young entry in its
834 * TLB and we mistakenly reclaim the associated page. The
835 * window for such an event is bounded by the next
836 * context-switch, which provides a DSB to complete the TLB
839 flush_tlb_page_nosync(vma, address);
845 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
846 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
847 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
848 unsigned long address,
851 return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
853 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
855 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
856 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
857 unsigned long address, pte_t *ptep)
859 return __pte(xchg_relaxed(&pte_val(*ptep), 0));
862 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
863 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
864 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
865 unsigned long address, pmd_t *pmdp)
867 return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
869 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
872 * ptep_set_wrprotect - mark read-only while trasferring potential hardware
873 * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
875 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
876 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
880 pte = READ_ONCE(*ptep);
883 pte = pte_wrprotect(pte);
884 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
885 pte_val(old_pte), pte_val(pte));
886 } while (pte_val(pte) != pte_val(old_pte));
889 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
890 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
891 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
892 unsigned long address, pmd_t *pmdp)
894 ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
897 #define pmdp_establish pmdp_establish
898 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
899 unsigned long address, pmd_t *pmdp, pmd_t pmd)
901 return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
906 * Encode and decode a swap entry:
907 * bits 0-1: present (must be zero)
908 * bits 2-7: swap type
909 * bits 8-57: swap offset
910 * bit 58: PTE_PROT_NONE (must be zero)
912 #define __SWP_TYPE_SHIFT 2
913 #define __SWP_TYPE_BITS 6
914 #define __SWP_OFFSET_BITS 50
915 #define __SWP_TYPE_MASK ((1 << __SWP_TYPE_BITS) - 1)
916 #define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
917 #define __SWP_OFFSET_MASK ((1UL << __SWP_OFFSET_BITS) - 1)
919 #define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
920 #define __swp_offset(x) (((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
921 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
923 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
924 #define __swp_entry_to_pte(swp) ((pte_t) { (swp).val })
926 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
927 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
928 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
929 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
932 * Ensure that there are not more swap files than can be encoded in the kernel
935 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
937 extern int kern_addr_valid(unsigned long addr);
939 #ifdef CONFIG_ARM64_MTE
941 #define __HAVE_ARCH_PREPARE_TO_SWAP
942 static inline int arch_prepare_to_swap(struct page *page)
944 if (system_supports_mte())
945 return mte_save_tags(page);
949 #define __HAVE_ARCH_SWAP_INVALIDATE
950 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
952 if (system_supports_mte())
953 mte_invalidate_tags(type, offset);
956 static inline void arch_swap_invalidate_area(int type)
958 if (system_supports_mte())
959 mte_invalidate_tags_area(type);
962 #define __HAVE_ARCH_SWAP_RESTORE
963 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
965 if (system_supports_mte() && mte_restore_tags(entry, page))
966 set_bit(PG_mte_tagged, &page->flags);
969 #endif /* CONFIG_ARM64_MTE */
972 * On AArch64, the cache coherency is handled via the set_pte_at() function.
974 static inline void update_mmu_cache(struct vm_area_struct *vma,
975 unsigned long addr, pte_t *ptep)
978 * We don't do anything here, so there's a very small chance of
979 * us retaking a user fault which we just fixed up. The alternative
980 * is doing a dsb(ishst), but that penalises the fastpath.
984 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
986 #ifdef CONFIG_ARM64_PA_BITS_52
987 #define phys_to_ttbr(addr) (((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
989 #define phys_to_ttbr(addr) (addr)
993 * On arm64 without hardware Access Flag, copying from user will fail because
994 * the pte is old and cannot be marked young. So we always end up with zeroed
995 * page after fork() + CoW for pfn mappings. We don't always have a
996 * hardware-managed access flag on arm64.
998 static inline bool arch_faults_on_old_pte(void)
1000 WARN_ON(preemptible());
1002 return !cpu_has_hw_af();
1004 #define arch_faults_on_old_pte arch_faults_on_old_pte
1007 * Experimentally, it's cheap to set the access flag in hardware and we
1008 * benefit from prefaulting mappings as 'old' to start with.
1010 static inline bool arch_wants_old_prefaulted_pte(void)
1012 return !arch_faults_on_old_pte();
1014 #define arch_wants_old_prefaulted_pte arch_wants_old_prefaulted_pte
1016 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
1018 if (cpus_have_const_cap(ARM64_HAS_EPAN))
1021 if (pgprot_val(prot) != pgprot_val(PAGE_EXECONLY))
1024 return PAGE_READONLY_EXEC;
1028 #endif /* !__ASSEMBLY__ */
1030 #endif /* __ASM_PGTABLE_H */