xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision dc6a81c3)
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 /*
44  * ZERO_PAGE is a global shared page that is always zero: used
45  * for zero-mapped memory areas etc..
46  */
47 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
48 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
49 
50 #define pte_ERROR(pte)		__pte_error(__FILE__, __LINE__, pte_val(pte))
51 
52 /*
53  * Macros to convert between a physical address and its placement in a
54  * page table entry, taking care of 52-bit addresses.
55  */
56 #ifdef CONFIG_ARM64_PA_BITS_52
57 #define __pte_to_phys(pte)	\
58 	((pte_val(pte) & PTE_ADDR_LOW) | ((pte_val(pte) & PTE_ADDR_HIGH) << 36))
59 #define __phys_to_pte_val(phys)	(((phys) | ((phys) >> 36)) & PTE_ADDR_MASK)
60 #else
61 #define __pte_to_phys(pte)	(pte_val(pte) & PTE_ADDR_MASK)
62 #define __phys_to_pte_val(phys)	(phys)
63 #endif
64 
65 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
66 #define pfn_pte(pfn,prot)	\
67 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
68 
69 #define pte_none(pte)		(!pte_val(pte))
70 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
71 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
72 
73 /*
74  * The following only work if pte_present(). Undefined behaviour otherwise.
75  */
76 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
77 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
78 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
79 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
80 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
81 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
82 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
83 
84 #define pte_cont_addr_end(addr, end)						\
85 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
86 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
87 })
88 
89 #define pmd_cont_addr_end(addr, end)						\
90 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
91 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
92 })
93 
94 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
95 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
96 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
97 
98 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
99 #define pte_valid_not_user(pte) \
100 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == PTE_VALID)
101 #define pte_valid_young(pte) \
102 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
103 #define pte_valid_user(pte) \
104 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
105 
106 /*
107  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
108  * so that we don't erroneously return false for pages that have been
109  * remapped as PROT_NONE but are yet to be flushed from the TLB.
110  */
111 #define pte_accessible(mm, pte)	\
112 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
113 
114 /*
115  * p??_access_permitted() is true for valid user mappings (subject to the
116  * write permission check). PROT_NONE mappings do not have the PTE_VALID bit
117  * set.
118  */
119 #define pte_access_permitted(pte, write) \
120 	(pte_valid_user(pte) && (!(write) || pte_write(pte)))
121 #define pmd_access_permitted(pmd, write) \
122 	(pte_access_permitted(pmd_pte(pmd), (write)))
123 #define pud_access_permitted(pud, write) \
124 	(pte_access_permitted(pud_pte(pud), (write)))
125 
126 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
127 {
128 	pte_val(pte) &= ~pgprot_val(prot);
129 	return pte;
130 }
131 
132 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
133 {
134 	pte_val(pte) |= pgprot_val(prot);
135 	return pte;
136 }
137 
138 static inline pte_t pte_wrprotect(pte_t pte)
139 {
140 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
141 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
142 	return pte;
143 }
144 
145 static inline pte_t pte_mkwrite(pte_t pte)
146 {
147 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
148 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
149 	return pte;
150 }
151 
152 static inline pte_t pte_mkclean(pte_t pte)
153 {
154 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
155 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
156 
157 	return pte;
158 }
159 
160 static inline pte_t pte_mkdirty(pte_t pte)
161 {
162 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
163 
164 	if (pte_write(pte))
165 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
166 
167 	return pte;
168 }
169 
170 static inline pte_t pte_mkold(pte_t pte)
171 {
172 	return clear_pte_bit(pte, __pgprot(PTE_AF));
173 }
174 
175 static inline pte_t pte_mkyoung(pte_t pte)
176 {
177 	return set_pte_bit(pte, __pgprot(PTE_AF));
178 }
179 
180 static inline pte_t pte_mkspecial(pte_t pte)
181 {
182 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
183 }
184 
185 static inline pte_t pte_mkcont(pte_t pte)
186 {
187 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
188 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
189 }
190 
191 static inline pte_t pte_mknoncont(pte_t pte)
192 {
193 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
194 }
195 
196 static inline pte_t pte_mkpresent(pte_t pte)
197 {
198 	return set_pte_bit(pte, __pgprot(PTE_VALID));
199 }
200 
201 static inline pmd_t pmd_mkcont(pmd_t pmd)
202 {
203 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
204 }
205 
206 static inline pte_t pte_mkdevmap(pte_t pte)
207 {
208 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
209 }
210 
211 static inline void set_pte(pte_t *ptep, pte_t pte)
212 {
213 	WRITE_ONCE(*ptep, pte);
214 
215 	/*
216 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
217 	 * or update_mmu_cache() have the necessary barriers.
218 	 */
219 	if (pte_valid_not_user(pte)) {
220 		dsb(ishst);
221 		isb();
222 	}
223 }
224 
225 extern void __sync_icache_dcache(pte_t pteval);
226 
227 /*
228  * PTE bits configuration in the presence of hardware Dirty Bit Management
229  * (PTE_WRITE == PTE_DBM):
230  *
231  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
232  *   0      0      |   1           0          0
233  *   0      1      |   1           1          0
234  *   1      0      |   1           0          1
235  *   1      1      |   0           1          x
236  *
237  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
238  * the page fault mechanism. Checking the dirty status of a pte becomes:
239  *
240  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
241  */
242 
243 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
244 					   pte_t pte)
245 {
246 	pte_t old_pte;
247 
248 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
249 		return;
250 
251 	old_pte = READ_ONCE(*ptep);
252 
253 	if (!pte_valid(old_pte) || !pte_valid(pte))
254 		return;
255 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
256 		return;
257 
258 	/*
259 	 * Check for potential race with hardware updates of the pte
260 	 * (ptep_set_access_flags safely changes valid ptes without going
261 	 * through an invalid entry).
262 	 */
263 	VM_WARN_ONCE(!pte_young(pte),
264 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
265 		     __func__, pte_val(old_pte), pte_val(pte));
266 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
267 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
268 		     __func__, pte_val(old_pte), pte_val(pte));
269 }
270 
271 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
272 			      pte_t *ptep, pte_t pte)
273 {
274 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
275 		__sync_icache_dcache(pte);
276 
277 	__check_racy_pte_update(mm, ptep, pte);
278 
279 	set_pte(ptep, pte);
280 }
281 
282 /*
283  * Huge pte definitions.
284  */
285 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
286 
287 /*
288  * Hugetlb definitions.
289  */
290 #define HUGE_MAX_HSTATE		4
291 #define HPAGE_SHIFT		PMD_SHIFT
292 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
293 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
294 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
295 
296 static inline pte_t pgd_pte(pgd_t pgd)
297 {
298 	return __pte(pgd_val(pgd));
299 }
300 
301 static inline pte_t pud_pte(pud_t pud)
302 {
303 	return __pte(pud_val(pud));
304 }
305 
306 static inline pud_t pte_pud(pte_t pte)
307 {
308 	return __pud(pte_val(pte));
309 }
310 
311 static inline pmd_t pud_pmd(pud_t pud)
312 {
313 	return __pmd(pud_val(pud));
314 }
315 
316 static inline pte_t pmd_pte(pmd_t pmd)
317 {
318 	return __pte(pmd_val(pmd));
319 }
320 
321 static inline pmd_t pte_pmd(pte_t pte)
322 {
323 	return __pmd(pte_val(pte));
324 }
325 
326 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
327 {
328 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
329 }
330 
331 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
332 {
333 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
334 }
335 
336 #ifdef CONFIG_NUMA_BALANCING
337 /*
338  * See the comment in include/asm-generic/pgtable.h
339  */
340 static inline int pte_protnone(pte_t pte)
341 {
342 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
343 }
344 
345 static inline int pmd_protnone(pmd_t pmd)
346 {
347 	return pte_protnone(pmd_pte(pmd));
348 }
349 #endif
350 
351 /*
352  * THP definitions.
353  */
354 
355 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
356 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
357 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
358 
359 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
360 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
361 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
362 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
363 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
364 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
365 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
366 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
367 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
368 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
369 #define pmd_mknotpresent(pmd)	(__pmd(pmd_val(pmd) & ~PMD_SECT_VALID))
370 
371 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
372 
373 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
374 
375 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
376 
377 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
378 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
379 #endif
380 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
381 {
382 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
383 }
384 
385 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
386 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
387 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
388 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
389 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
390 
391 #define pud_young(pud)		pte_young(pud_pte(pud))
392 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
393 #define pud_write(pud)		pte_write(pud_pte(pud))
394 
395 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
396 
397 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
398 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
399 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
400 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
401 
402 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
403 
404 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
405 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
406 
407 #define __pgprot_modify(prot,mask,bits) \
408 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
409 
410 /*
411  * Mark the prot value as uncacheable and unbufferable.
412  */
413 #define pgprot_noncached(prot) \
414 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
415 #define pgprot_writecombine(prot) \
416 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
417 #define pgprot_device(prot) \
418 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
419 /*
420  * DMA allocations for non-coherent devices use what the Arm architecture calls
421  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
422  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
423  * is intended for MMIO and thus forbids speculation, preserves access size,
424  * requires strict alignment and can also force write responses to come from the
425  * endpoint.
426  */
427 #define pgprot_dmacoherent(prot) \
428 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
429 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
430 
431 #define __HAVE_PHYS_MEM_ACCESS_PROT
432 struct file;
433 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
434 				     unsigned long size, pgprot_t vma_prot);
435 
436 #define pmd_none(pmd)		(!pmd_val(pmd))
437 
438 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
439 
440 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
441 				 PMD_TYPE_TABLE)
442 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
443 				 PMD_TYPE_SECT)
444 #define pmd_leaf(pmd)		pmd_sect(pmd)
445 
446 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
447 static inline bool pud_sect(pud_t pud) { return false; }
448 static inline bool pud_table(pud_t pud) { return true; }
449 #else
450 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
451 				 PUD_TYPE_SECT)
452 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
453 				 PUD_TYPE_TABLE)
454 #endif
455 
456 extern pgd_t init_pg_dir[PTRS_PER_PGD];
457 extern pgd_t init_pg_end[];
458 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
459 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
460 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
461 
462 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
463 
464 static inline bool in_swapper_pgdir(void *addr)
465 {
466 	return ((unsigned long)addr & PAGE_MASK) ==
467 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
468 }
469 
470 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
471 {
472 #ifdef __PAGETABLE_PMD_FOLDED
473 	if (in_swapper_pgdir(pmdp)) {
474 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
475 		return;
476 	}
477 #endif /* __PAGETABLE_PMD_FOLDED */
478 
479 	WRITE_ONCE(*pmdp, pmd);
480 
481 	if (pmd_valid(pmd)) {
482 		dsb(ishst);
483 		isb();
484 	}
485 }
486 
487 static inline void pmd_clear(pmd_t *pmdp)
488 {
489 	set_pmd(pmdp, __pmd(0));
490 }
491 
492 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
493 {
494 	return __pmd_to_phys(pmd);
495 }
496 
497 static inline void pte_unmap(pte_t *pte) { }
498 
499 /* Find an entry in the third-level page table. */
500 #define pte_index(addr)		(((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
501 
502 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
503 #define pte_offset_kernel(dir,addr)	((pte_t *)__va(pte_offset_phys((dir), (addr))))
504 
505 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
506 
507 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
508 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
509 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
510 
511 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(__pmd_to_phys(pmd)))
512 
513 /* use ONLY for statically allocated translation tables */
514 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
515 
516 /*
517  * Conversion functions: convert a page and protection to a page entry,
518  * and a page entry and page directory to the page they refer to.
519  */
520 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
521 
522 #if CONFIG_PGTABLE_LEVELS > 2
523 
524 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
525 
526 #define pud_none(pud)		(!pud_val(pud))
527 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
528 #define pud_present(pud)	pte_present(pud_pte(pud))
529 #define pud_leaf(pud)		pud_sect(pud)
530 #define pud_valid(pud)		pte_valid(pud_pte(pud))
531 
532 static inline void set_pud(pud_t *pudp, pud_t pud)
533 {
534 #ifdef __PAGETABLE_PUD_FOLDED
535 	if (in_swapper_pgdir(pudp)) {
536 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
537 		return;
538 	}
539 #endif /* __PAGETABLE_PUD_FOLDED */
540 
541 	WRITE_ONCE(*pudp, pud);
542 
543 	if (pud_valid(pud)) {
544 		dsb(ishst);
545 		isb();
546 	}
547 }
548 
549 static inline void pud_clear(pud_t *pudp)
550 {
551 	set_pud(pudp, __pud(0));
552 }
553 
554 static inline phys_addr_t pud_page_paddr(pud_t pud)
555 {
556 	return __pud_to_phys(pud);
557 }
558 
559 /* Find an entry in the second-level page table. */
560 #define pmd_index(addr)		(((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
561 
562 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
563 #define pmd_offset(dir, addr)		((pmd_t *)__va(pmd_offset_phys((dir), (addr))))
564 
565 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
566 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
567 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
568 
569 #define pud_page(pud)		pfn_to_page(__phys_to_pfn(__pud_to_phys(pud)))
570 
571 /* use ONLY for statically allocated translation tables */
572 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
573 
574 #else
575 
576 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
577 
578 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
579 #define pmd_set_fixmap(addr)		NULL
580 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
581 #define pmd_clear_fixmap()
582 
583 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
584 
585 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
586 
587 #if CONFIG_PGTABLE_LEVELS > 3
588 
589 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
590 
591 #define pgd_none(pgd)		(!pgd_val(pgd))
592 #define pgd_bad(pgd)		(!(pgd_val(pgd) & 2))
593 #define pgd_present(pgd)	(pgd_val(pgd))
594 
595 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
596 {
597 	if (in_swapper_pgdir(pgdp)) {
598 		set_swapper_pgd(pgdp, pgd);
599 		return;
600 	}
601 
602 	WRITE_ONCE(*pgdp, pgd);
603 	dsb(ishst);
604 	isb();
605 }
606 
607 static inline void pgd_clear(pgd_t *pgdp)
608 {
609 	set_pgd(pgdp, __pgd(0));
610 }
611 
612 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
613 {
614 	return __pgd_to_phys(pgd);
615 }
616 
617 /* Find an entry in the frst-level page table. */
618 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
619 
620 #define pud_offset_phys(dir, addr)	(pgd_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
621 #define pud_offset(dir, addr)		((pud_t *)__va(pud_offset_phys((dir), (addr))))
622 
623 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
624 #define pud_set_fixmap_offset(pgd, addr)	pud_set_fixmap(pud_offset_phys(pgd, addr))
625 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
626 
627 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd)))
628 
629 /* use ONLY for statically allocated translation tables */
630 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
631 
632 #else
633 
634 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
635 
636 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
637 #define pud_set_fixmap(addr)		NULL
638 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
639 #define pud_clear_fixmap()
640 
641 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
642 
643 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
644 
645 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
646 
647 /* to find an entry in a page-table-directory */
648 #define pgd_index(addr)		(((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
649 
650 #define pgd_offset_raw(pgd, addr)	((pgd) + pgd_index(addr))
651 
652 #define pgd_offset(mm, addr)	(pgd_offset_raw((mm)->pgd, (addr)))
653 
654 /* to find an entry in a kernel page-table-directory */
655 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
656 
657 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
658 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
659 
660 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
661 {
662 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
663 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE;
664 	/* preserve the hardware dirty information */
665 	if (pte_hw_dirty(pte))
666 		pte = pte_mkdirty(pte);
667 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
668 	return pte;
669 }
670 
671 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
672 {
673 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
674 }
675 
676 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
677 extern int ptep_set_access_flags(struct vm_area_struct *vma,
678 				 unsigned long address, pte_t *ptep,
679 				 pte_t entry, int dirty);
680 
681 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
682 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
683 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
684 					unsigned long address, pmd_t *pmdp,
685 					pmd_t entry, int dirty)
686 {
687 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
688 }
689 
690 static inline int pud_devmap(pud_t pud)
691 {
692 	return 0;
693 }
694 
695 static inline int pgd_devmap(pgd_t pgd)
696 {
697 	return 0;
698 }
699 #endif
700 
701 /*
702  * Atomic pte/pmd modifications.
703  */
704 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
705 static inline int __ptep_test_and_clear_young(pte_t *ptep)
706 {
707 	pte_t old_pte, pte;
708 
709 	pte = READ_ONCE(*ptep);
710 	do {
711 		old_pte = pte;
712 		pte = pte_mkold(pte);
713 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
714 					       pte_val(old_pte), pte_val(pte));
715 	} while (pte_val(pte) != pte_val(old_pte));
716 
717 	return pte_young(pte);
718 }
719 
720 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
721 					    unsigned long address,
722 					    pte_t *ptep)
723 {
724 	return __ptep_test_and_clear_young(ptep);
725 }
726 
727 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
728 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
729 					 unsigned long address, pte_t *ptep)
730 {
731 	int young = ptep_test_and_clear_young(vma, address, ptep);
732 
733 	if (young) {
734 		/*
735 		 * We can elide the trailing DSB here since the worst that can
736 		 * happen is that a CPU continues to use the young entry in its
737 		 * TLB and we mistakenly reclaim the associated page. The
738 		 * window for such an event is bounded by the next
739 		 * context-switch, which provides a DSB to complete the TLB
740 		 * invalidation.
741 		 */
742 		flush_tlb_page_nosync(vma, address);
743 	}
744 
745 	return young;
746 }
747 
748 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
749 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
750 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
751 					    unsigned long address,
752 					    pmd_t *pmdp)
753 {
754 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
755 }
756 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
757 
758 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
759 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
760 				       unsigned long address, pte_t *ptep)
761 {
762 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
763 }
764 
765 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
766 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
767 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
768 					    unsigned long address, pmd_t *pmdp)
769 {
770 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
771 }
772 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
773 
774 /*
775  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
776  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
777  */
778 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
779 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
780 {
781 	pte_t old_pte, pte;
782 
783 	pte = READ_ONCE(*ptep);
784 	do {
785 		old_pte = pte;
786 		/*
787 		 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
788 		 * clear), set the PTE_DIRTY bit.
789 		 */
790 		if (pte_hw_dirty(pte))
791 			pte = pte_mkdirty(pte);
792 		pte = pte_wrprotect(pte);
793 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
794 					       pte_val(old_pte), pte_val(pte));
795 	} while (pte_val(pte) != pte_val(old_pte));
796 }
797 
798 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
799 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
800 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
801 				      unsigned long address, pmd_t *pmdp)
802 {
803 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
804 }
805 
806 #define pmdp_establish pmdp_establish
807 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
808 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
809 {
810 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
811 }
812 #endif
813 
814 /*
815  * Encode and decode a swap entry:
816  *	bits 0-1:	present (must be zero)
817  *	bits 2-7:	swap type
818  *	bits 8-57:	swap offset
819  *	bit  58:	PTE_PROT_NONE (must be zero)
820  */
821 #define __SWP_TYPE_SHIFT	2
822 #define __SWP_TYPE_BITS		6
823 #define __SWP_OFFSET_BITS	50
824 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
825 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
826 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
827 
828 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
829 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
830 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
831 
832 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
833 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
834 
835 /*
836  * Ensure that there are not more swap files than can be encoded in the kernel
837  * PTEs.
838  */
839 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
840 
841 extern int kern_addr_valid(unsigned long addr);
842 
843 #include <asm-generic/pgtable.h>
844 
845 /*
846  * On AArch64, the cache coherency is handled via the set_pte_at() function.
847  */
848 static inline void update_mmu_cache(struct vm_area_struct *vma,
849 				    unsigned long addr, pte_t *ptep)
850 {
851 	/*
852 	 * We don't do anything here, so there's a very small chance of
853 	 * us retaking a user fault which we just fixed up. The alternative
854 	 * is doing a dsb(ishst), but that penalises the fastpath.
855 	 */
856 }
857 
858 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
859 
860 #ifdef CONFIG_ARM64_PA_BITS_52
861 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
862 #else
863 #define phys_to_ttbr(addr)	(addr)
864 #endif
865 
866 /*
867  * On arm64 without hardware Access Flag, copying from user will fail because
868  * the pte is old and cannot be marked young. So we always end up with zeroed
869  * page after fork() + CoW for pfn mappings. We don't always have a
870  * hardware-managed access flag on arm64.
871  */
872 static inline bool arch_faults_on_old_pte(void)
873 {
874 	WARN_ON(preemptible());
875 
876 	return !cpu_has_hw_af();
877 }
878 #define arch_faults_on_old_pte arch_faults_on_old_pte
879 
880 #endif /* !__ASSEMBLY__ */
881 
882 #endif /* __ASM_PGTABLE_H */
883