xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 519a8a6c)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2012 ARM Ltd.
4  */
5 #ifndef __ASM_PGTABLE_H
6 #define __ASM_PGTABLE_H
7 
8 #include <asm/bug.h>
9 #include <asm/proc-fns.h>
10 
11 #include <asm/memory.h>
12 #include <asm/pgtable-hwdef.h>
13 #include <asm/pgtable-prot.h>
14 #include <asm/tlbflush.h>
15 
16 /*
17  * VMALLOC range.
18  *
19  * VMALLOC_START: beginning of the kernel vmalloc space
20  * VMALLOC_END: extends to the available space below vmemmap, PCI I/O space
21  *	and fixed mappings
22  */
23 #define VMALLOC_START		(MODULES_END)
24 #define VMALLOC_END		(- PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
25 
26 #define FIRST_USER_ADDRESS	0UL
27 
28 #ifndef __ASSEMBLY__
29 
30 #include <asm/cmpxchg.h>
31 #include <asm/fixmap.h>
32 #include <linux/mmdebug.h>
33 #include <linux/mm_types.h>
34 #include <linux/sched.h>
35 
36 extern struct page *vmemmap;
37 
38 extern void __pte_error(const char *file, int line, unsigned long val);
39 extern void __pmd_error(const char *file, int line, unsigned long val);
40 extern void __pud_error(const char *file, int line, unsigned long val);
41 extern void __pgd_error(const char *file, int line, unsigned long val);
42 
43 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
44 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
45 
46 /* Set stride and tlb_level in flush_*_tlb_range */
47 #define flush_pmd_tlb_range(vma, addr, end)	\
48 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
49 #define flush_pud_tlb_range(vma, addr, end)	\
50 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
51 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
52 
53 /*
54  * ZERO_PAGE is a global shared page that is always zero: used
55  * for zero-mapped memory areas etc..
56  */
57 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
58 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
59 
60 #define pte_ERROR(pte)		__pte_error(__FILE__, __LINE__, pte_val(pte))
61 
62 /*
63  * Macros to convert between a physical address and its placement in a
64  * page table entry, taking care of 52-bit addresses.
65  */
66 #ifdef CONFIG_ARM64_PA_BITS_52
67 #define __pte_to_phys(pte)	\
68 	((pte_val(pte) & PTE_ADDR_LOW) | ((pte_val(pte) & PTE_ADDR_HIGH) << 36))
69 #define __phys_to_pte_val(phys)	(((phys) | ((phys) >> 36)) & PTE_ADDR_MASK)
70 #else
71 #define __pte_to_phys(pte)	(pte_val(pte) & PTE_ADDR_MASK)
72 #define __phys_to_pte_val(phys)	(phys)
73 #endif
74 
75 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
76 #define pfn_pte(pfn,prot)	\
77 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
78 
79 #define pte_none(pte)		(!pte_val(pte))
80 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
81 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
82 
83 /*
84  * The following only work if pte_present(). Undefined behaviour otherwise.
85  */
86 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
87 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
88 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
89 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
90 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
91 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
92 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
93 
94 #define pte_cont_addr_end(addr, end)						\
95 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
96 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
97 })
98 
99 #define pmd_cont_addr_end(addr, end)						\
100 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
101 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
102 })
103 
104 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
105 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
106 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
107 
108 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
109 #define pte_valid_not_user(pte) \
110 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == PTE_VALID)
111 #define pte_valid_young(pte) \
112 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
113 #define pte_valid_user(pte) \
114 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
115 
116 /*
117  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
118  * so that we don't erroneously return false for pages that have been
119  * remapped as PROT_NONE but are yet to be flushed from the TLB.
120  */
121 #define pte_accessible(mm, pte)	\
122 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
123 
124 /*
125  * p??_access_permitted() is true for valid user mappings (subject to the
126  * write permission check). PROT_NONE mappings do not have the PTE_VALID bit
127  * set.
128  */
129 #define pte_access_permitted(pte, write) \
130 	(pte_valid_user(pte) && (!(write) || pte_write(pte)))
131 #define pmd_access_permitted(pmd, write) \
132 	(pte_access_permitted(pmd_pte(pmd), (write)))
133 #define pud_access_permitted(pud, write) \
134 	(pte_access_permitted(pud_pte(pud), (write)))
135 
136 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
137 {
138 	pte_val(pte) &= ~pgprot_val(prot);
139 	return pte;
140 }
141 
142 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
143 {
144 	pte_val(pte) |= pgprot_val(prot);
145 	return pte;
146 }
147 
148 static inline pte_t pte_wrprotect(pte_t pte)
149 {
150 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
151 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
152 	return pte;
153 }
154 
155 static inline pte_t pte_mkwrite(pte_t pte)
156 {
157 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
158 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
159 	return pte;
160 }
161 
162 static inline pte_t pte_mkclean(pte_t pte)
163 {
164 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
165 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
166 
167 	return pte;
168 }
169 
170 static inline pte_t pte_mkdirty(pte_t pte)
171 {
172 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
173 
174 	if (pte_write(pte))
175 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
176 
177 	return pte;
178 }
179 
180 static inline pte_t pte_mkold(pte_t pte)
181 {
182 	return clear_pte_bit(pte, __pgprot(PTE_AF));
183 }
184 
185 static inline pte_t pte_mkyoung(pte_t pte)
186 {
187 	return set_pte_bit(pte, __pgprot(PTE_AF));
188 }
189 
190 static inline pte_t pte_mkspecial(pte_t pte)
191 {
192 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
193 }
194 
195 static inline pte_t pte_mkcont(pte_t pte)
196 {
197 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
198 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
199 }
200 
201 static inline pte_t pte_mknoncont(pte_t pte)
202 {
203 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
204 }
205 
206 static inline pte_t pte_mkpresent(pte_t pte)
207 {
208 	return set_pte_bit(pte, __pgprot(PTE_VALID));
209 }
210 
211 static inline pmd_t pmd_mkcont(pmd_t pmd)
212 {
213 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
214 }
215 
216 static inline pte_t pte_mkdevmap(pte_t pte)
217 {
218 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
219 }
220 
221 static inline void set_pte(pte_t *ptep, pte_t pte)
222 {
223 	WRITE_ONCE(*ptep, pte);
224 
225 	/*
226 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
227 	 * or update_mmu_cache() have the necessary barriers.
228 	 */
229 	if (pte_valid_not_user(pte)) {
230 		dsb(ishst);
231 		isb();
232 	}
233 }
234 
235 extern void __sync_icache_dcache(pte_t pteval);
236 
237 /*
238  * PTE bits configuration in the presence of hardware Dirty Bit Management
239  * (PTE_WRITE == PTE_DBM):
240  *
241  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
242  *   0      0      |   1           0          0
243  *   0      1      |   1           1          0
244  *   1      0      |   1           0          1
245  *   1      1      |   0           1          x
246  *
247  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
248  * the page fault mechanism. Checking the dirty status of a pte becomes:
249  *
250  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
251  */
252 
253 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
254 					   pte_t pte)
255 {
256 	pte_t old_pte;
257 
258 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
259 		return;
260 
261 	old_pte = READ_ONCE(*ptep);
262 
263 	if (!pte_valid(old_pte) || !pte_valid(pte))
264 		return;
265 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
266 		return;
267 
268 	/*
269 	 * Check for potential race with hardware updates of the pte
270 	 * (ptep_set_access_flags safely changes valid ptes without going
271 	 * through an invalid entry).
272 	 */
273 	VM_WARN_ONCE(!pte_young(pte),
274 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
275 		     __func__, pte_val(old_pte), pte_val(pte));
276 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
277 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
278 		     __func__, pte_val(old_pte), pte_val(pte));
279 }
280 
281 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
282 			      pte_t *ptep, pte_t pte)
283 {
284 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
285 		__sync_icache_dcache(pte);
286 
287 	__check_racy_pte_update(mm, ptep, pte);
288 
289 	set_pte(ptep, pte);
290 }
291 
292 /*
293  * Huge pte definitions.
294  */
295 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
296 
297 /*
298  * Hugetlb definitions.
299  */
300 #define HUGE_MAX_HSTATE		4
301 #define HPAGE_SHIFT		PMD_SHIFT
302 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
303 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
304 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
305 
306 static inline pte_t pgd_pte(pgd_t pgd)
307 {
308 	return __pte(pgd_val(pgd));
309 }
310 
311 static inline pte_t p4d_pte(p4d_t p4d)
312 {
313 	return __pte(p4d_val(p4d));
314 }
315 
316 static inline pte_t pud_pte(pud_t pud)
317 {
318 	return __pte(pud_val(pud));
319 }
320 
321 static inline pud_t pte_pud(pte_t pte)
322 {
323 	return __pud(pte_val(pte));
324 }
325 
326 static inline pmd_t pud_pmd(pud_t pud)
327 {
328 	return __pmd(pud_val(pud));
329 }
330 
331 static inline pte_t pmd_pte(pmd_t pmd)
332 {
333 	return __pte(pmd_val(pmd));
334 }
335 
336 static inline pmd_t pte_pmd(pte_t pte)
337 {
338 	return __pmd(pte_val(pte));
339 }
340 
341 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
342 {
343 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
344 }
345 
346 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
347 {
348 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
349 }
350 
351 #ifdef CONFIG_NUMA_BALANCING
352 /*
353  * See the comment in include/linux/pgtable.h
354  */
355 static inline int pte_protnone(pte_t pte)
356 {
357 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
358 }
359 
360 static inline int pmd_protnone(pmd_t pmd)
361 {
362 	return pte_protnone(pmd_pte(pmd));
363 }
364 #endif
365 
366 /*
367  * THP definitions.
368  */
369 
370 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
371 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
372 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
373 
374 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
375 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
376 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
377 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
378 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
379 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
380 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
381 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
382 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
383 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
384 #define pmd_mkinvalid(pmd)	(__pmd(pmd_val(pmd) & ~PMD_SECT_VALID))
385 
386 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
387 
388 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
389 
390 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
391 
392 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
393 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
394 #endif
395 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
396 {
397 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
398 }
399 
400 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
401 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
402 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
403 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
404 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
405 
406 #define pud_young(pud)		pte_young(pud_pte(pud))
407 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
408 #define pud_write(pud)		pte_write(pud_pte(pud))
409 
410 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
411 
412 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
413 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
414 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
415 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
416 
417 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
418 
419 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
420 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
421 
422 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
423 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
424 
425 #define __pgprot_modify(prot,mask,bits) \
426 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
427 
428 #define pgprot_nx(prot) \
429 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
430 
431 /*
432  * Mark the prot value as uncacheable and unbufferable.
433  */
434 #define pgprot_noncached(prot) \
435 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
436 #define pgprot_writecombine(prot) \
437 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
438 #define pgprot_device(prot) \
439 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
440 /*
441  * DMA allocations for non-coherent devices use what the Arm architecture calls
442  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
443  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
444  * is intended for MMIO and thus forbids speculation, preserves access size,
445  * requires strict alignment and can also force write responses to come from the
446  * endpoint.
447  */
448 #define pgprot_dmacoherent(prot) \
449 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
450 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
451 
452 #define __HAVE_PHYS_MEM_ACCESS_PROT
453 struct file;
454 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
455 				     unsigned long size, pgprot_t vma_prot);
456 
457 #define pmd_none(pmd)		(!pmd_val(pmd))
458 
459 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
460 
461 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
462 				 PMD_TYPE_TABLE)
463 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
464 				 PMD_TYPE_SECT)
465 #define pmd_leaf(pmd)		pmd_sect(pmd)
466 
467 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
468 static inline bool pud_sect(pud_t pud) { return false; }
469 static inline bool pud_table(pud_t pud) { return true; }
470 #else
471 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
472 				 PUD_TYPE_SECT)
473 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
474 				 PUD_TYPE_TABLE)
475 #endif
476 
477 extern pgd_t init_pg_dir[PTRS_PER_PGD];
478 extern pgd_t init_pg_end[];
479 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
480 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
481 extern pgd_t idmap_pg_end[];
482 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
483 
484 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
485 
486 static inline bool in_swapper_pgdir(void *addr)
487 {
488 	return ((unsigned long)addr & PAGE_MASK) ==
489 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
490 }
491 
492 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
493 {
494 #ifdef __PAGETABLE_PMD_FOLDED
495 	if (in_swapper_pgdir(pmdp)) {
496 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
497 		return;
498 	}
499 #endif /* __PAGETABLE_PMD_FOLDED */
500 
501 	WRITE_ONCE(*pmdp, pmd);
502 
503 	if (pmd_valid(pmd)) {
504 		dsb(ishst);
505 		isb();
506 	}
507 }
508 
509 static inline void pmd_clear(pmd_t *pmdp)
510 {
511 	set_pmd(pmdp, __pmd(0));
512 }
513 
514 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
515 {
516 	return __pmd_to_phys(pmd);
517 }
518 
519 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
520 {
521 	return (unsigned long)__va(pmd_page_paddr(pmd));
522 }
523 
524 /* Find an entry in the third-level page table. */
525 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
526 
527 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
528 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
529 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
530 
531 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
532 
533 /* use ONLY for statically allocated translation tables */
534 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
535 
536 /*
537  * Conversion functions: convert a page and protection to a page entry,
538  * and a page entry and page directory to the page they refer to.
539  */
540 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
541 
542 #if CONFIG_PGTABLE_LEVELS > 2
543 
544 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
545 
546 #define pud_none(pud)		(!pud_val(pud))
547 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
548 #define pud_present(pud)	pte_present(pud_pte(pud))
549 #define pud_leaf(pud)		pud_sect(pud)
550 #define pud_valid(pud)		pte_valid(pud_pte(pud))
551 
552 static inline void set_pud(pud_t *pudp, pud_t pud)
553 {
554 #ifdef __PAGETABLE_PUD_FOLDED
555 	if (in_swapper_pgdir(pudp)) {
556 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
557 		return;
558 	}
559 #endif /* __PAGETABLE_PUD_FOLDED */
560 
561 	WRITE_ONCE(*pudp, pud);
562 
563 	if (pud_valid(pud)) {
564 		dsb(ishst);
565 		isb();
566 	}
567 }
568 
569 static inline void pud_clear(pud_t *pudp)
570 {
571 	set_pud(pudp, __pud(0));
572 }
573 
574 static inline phys_addr_t pud_page_paddr(pud_t pud)
575 {
576 	return __pud_to_phys(pud);
577 }
578 
579 static inline unsigned long pud_page_vaddr(pud_t pud)
580 {
581 	return (unsigned long)__va(pud_page_paddr(pud));
582 }
583 
584 /* Find an entry in the second-level page table. */
585 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
586 
587 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
588 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
589 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
590 
591 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
592 
593 /* use ONLY for statically allocated translation tables */
594 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
595 
596 #else
597 
598 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
599 
600 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
601 #define pmd_set_fixmap(addr)		NULL
602 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
603 #define pmd_clear_fixmap()
604 
605 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
606 
607 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
608 
609 #if CONFIG_PGTABLE_LEVELS > 3
610 
611 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
612 
613 #define p4d_none(p4d)		(!p4d_val(p4d))
614 #define p4d_bad(p4d)		(!(p4d_val(p4d) & 2))
615 #define p4d_present(p4d)	(p4d_val(p4d))
616 
617 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
618 {
619 	if (in_swapper_pgdir(p4dp)) {
620 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
621 		return;
622 	}
623 
624 	WRITE_ONCE(*p4dp, p4d);
625 	dsb(ishst);
626 	isb();
627 }
628 
629 static inline void p4d_clear(p4d_t *p4dp)
630 {
631 	set_p4d(p4dp, __p4d(0));
632 }
633 
634 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
635 {
636 	return __p4d_to_phys(p4d);
637 }
638 
639 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
640 {
641 	return (unsigned long)__va(p4d_page_paddr(p4d));
642 }
643 
644 /* Find an entry in the frst-level page table. */
645 #define pud_offset_phys(dir, addr)	(p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
646 
647 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
648 #define pud_set_fixmap_offset(p4d, addr)	pud_set_fixmap(pud_offset_phys(p4d, addr))
649 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
650 
651 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
652 
653 /* use ONLY for statically allocated translation tables */
654 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
655 
656 #else
657 
658 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
659 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
660 
661 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
662 #define pud_set_fixmap(addr)		NULL
663 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
664 #define pud_clear_fixmap()
665 
666 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
667 
668 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
669 
670 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
671 
672 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
673 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
674 
675 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
676 {
677 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
678 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP;
679 	/* preserve the hardware dirty information */
680 	if (pte_hw_dirty(pte))
681 		pte = pte_mkdirty(pte);
682 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
683 	return pte;
684 }
685 
686 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
687 {
688 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
689 }
690 
691 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
692 extern int ptep_set_access_flags(struct vm_area_struct *vma,
693 				 unsigned long address, pte_t *ptep,
694 				 pte_t entry, int dirty);
695 
696 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
697 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
698 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
699 					unsigned long address, pmd_t *pmdp,
700 					pmd_t entry, int dirty)
701 {
702 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
703 }
704 
705 static inline int pud_devmap(pud_t pud)
706 {
707 	return 0;
708 }
709 
710 static inline int pgd_devmap(pgd_t pgd)
711 {
712 	return 0;
713 }
714 #endif
715 
716 /*
717  * Atomic pte/pmd modifications.
718  */
719 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
720 static inline int __ptep_test_and_clear_young(pte_t *ptep)
721 {
722 	pte_t old_pte, pte;
723 
724 	pte = READ_ONCE(*ptep);
725 	do {
726 		old_pte = pte;
727 		pte = pte_mkold(pte);
728 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
729 					       pte_val(old_pte), pte_val(pte));
730 	} while (pte_val(pte) != pte_val(old_pte));
731 
732 	return pte_young(pte);
733 }
734 
735 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
736 					    unsigned long address,
737 					    pte_t *ptep)
738 {
739 	return __ptep_test_and_clear_young(ptep);
740 }
741 
742 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
743 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
744 					 unsigned long address, pte_t *ptep)
745 {
746 	int young = ptep_test_and_clear_young(vma, address, ptep);
747 
748 	if (young) {
749 		/*
750 		 * We can elide the trailing DSB here since the worst that can
751 		 * happen is that a CPU continues to use the young entry in its
752 		 * TLB and we mistakenly reclaim the associated page. The
753 		 * window for such an event is bounded by the next
754 		 * context-switch, which provides a DSB to complete the TLB
755 		 * invalidation.
756 		 */
757 		flush_tlb_page_nosync(vma, address);
758 	}
759 
760 	return young;
761 }
762 
763 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
764 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
765 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
766 					    unsigned long address,
767 					    pmd_t *pmdp)
768 {
769 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
770 }
771 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
772 
773 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
774 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
775 				       unsigned long address, pte_t *ptep)
776 {
777 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
778 }
779 
780 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
781 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
782 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
783 					    unsigned long address, pmd_t *pmdp)
784 {
785 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
786 }
787 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
788 
789 /*
790  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
791  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
792  */
793 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
794 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
795 {
796 	pte_t old_pte, pte;
797 
798 	pte = READ_ONCE(*ptep);
799 	do {
800 		old_pte = pte;
801 		/*
802 		 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
803 		 * clear), set the PTE_DIRTY bit.
804 		 */
805 		if (pte_hw_dirty(pte))
806 			pte = pte_mkdirty(pte);
807 		pte = pte_wrprotect(pte);
808 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
809 					       pte_val(old_pte), pte_val(pte));
810 	} while (pte_val(pte) != pte_val(old_pte));
811 }
812 
813 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
814 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
815 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
816 				      unsigned long address, pmd_t *pmdp)
817 {
818 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
819 }
820 
821 #define pmdp_establish pmdp_establish
822 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
823 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
824 {
825 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
826 }
827 #endif
828 
829 /*
830  * Encode and decode a swap entry:
831  *	bits 0-1:	present (must be zero)
832  *	bits 2-7:	swap type
833  *	bits 8-57:	swap offset
834  *	bit  58:	PTE_PROT_NONE (must be zero)
835  */
836 #define __SWP_TYPE_SHIFT	2
837 #define __SWP_TYPE_BITS		6
838 #define __SWP_OFFSET_BITS	50
839 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
840 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
841 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
842 
843 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
844 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
845 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
846 
847 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
848 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
849 
850 /*
851  * Ensure that there are not more swap files than can be encoded in the kernel
852  * PTEs.
853  */
854 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
855 
856 extern int kern_addr_valid(unsigned long addr);
857 
858 /*
859  * On AArch64, the cache coherency is handled via the set_pte_at() function.
860  */
861 static inline void update_mmu_cache(struct vm_area_struct *vma,
862 				    unsigned long addr, pte_t *ptep)
863 {
864 	/*
865 	 * We don't do anything here, so there's a very small chance of
866 	 * us retaking a user fault which we just fixed up. The alternative
867 	 * is doing a dsb(ishst), but that penalises the fastpath.
868 	 */
869 }
870 
871 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
872 
873 #ifdef CONFIG_ARM64_PA_BITS_52
874 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
875 #else
876 #define phys_to_ttbr(addr)	(addr)
877 #endif
878 
879 /*
880  * On arm64 without hardware Access Flag, copying from user will fail because
881  * the pte is old and cannot be marked young. So we always end up with zeroed
882  * page after fork() + CoW for pfn mappings. We don't always have a
883  * hardware-managed access flag on arm64.
884  */
885 static inline bool arch_faults_on_old_pte(void)
886 {
887 	WARN_ON(preemptible());
888 
889 	return !cpu_has_hw_af();
890 }
891 #define arch_faults_on_old_pte arch_faults_on_old_pte
892 
893 #endif /* !__ASSEMBLY__ */
894 
895 #endif /* __ASM_PGTABLE_H */
896