xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 6724ed7f)
1 /*
2  * Copyright (C) 2012 ARM Ltd.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15  */
16 #ifndef __ASM_PGTABLE_H
17 #define __ASM_PGTABLE_H
18 
19 #include <asm/bug.h>
20 #include <asm/proc-fns.h>
21 
22 #include <asm/memory.h>
23 #include <asm/pgtable-hwdef.h>
24 #include <asm/pgtable-prot.h>
25 
26 /*
27  * VMALLOC range.
28  *
29  * VMALLOC_START: beginning of the kernel vmalloc space
30  * VMALLOC_END: extends to the available space below vmmemmap, PCI I/O space
31  *	and fixed mappings
32  */
33 #define VMALLOC_START		(MODULES_END)
34 #define VMALLOC_END		(PAGE_OFFSET - PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
35 
36 #define vmemmap			((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
37 
38 #define FIRST_USER_ADDRESS	0UL
39 
40 #ifndef __ASSEMBLY__
41 
42 #include <asm/cmpxchg.h>
43 #include <asm/fixmap.h>
44 #include <linux/mmdebug.h>
45 #include <linux/mm_types.h>
46 #include <linux/sched.h>
47 
48 extern void __pte_error(const char *file, int line, unsigned long val);
49 extern void __pmd_error(const char *file, int line, unsigned long val);
50 extern void __pud_error(const char *file, int line, unsigned long val);
51 extern void __pgd_error(const char *file, int line, unsigned long val);
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 #define pte_pfn(pte)		((pte_val(pte) & PHYS_MASK) >> PAGE_SHIFT)
63 
64 #define pfn_pte(pfn,prot)	(__pte(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
65 
66 #define pte_none(pte)		(!pte_val(pte))
67 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
68 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
69 
70 /*
71  * The following only work if pte_present(). Undefined behaviour otherwise.
72  */
73 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
74 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
75 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
76 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
77 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
78 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
79 
80 #define pte_cont_addr_end(addr, end)						\
81 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
82 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
83 })
84 
85 #define pmd_cont_addr_end(addr, end)						\
86 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
87 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
88 })
89 
90 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
91 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
92 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
93 
94 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
95 /*
96  * Execute-only user mappings do not have the PTE_USER bit set. All valid
97  * kernel mappings have the PTE_UXN bit set.
98  */
99 #define pte_valid_not_user(pte) \
100 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
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) other than user execute-only which do not have the
117  * PTE_USER bit set. PROT_NONE mappings do not have the PTE_VALID bit 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 void set_pte(pte_t *ptep, pte_t pte)
207 {
208 	*ptep = pte;
209 
210 	/*
211 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
212 	 * or update_mmu_cache() have the necessary barriers.
213 	 */
214 	if (pte_valid_not_user(pte)) {
215 		dsb(ishst);
216 		isb();
217 	}
218 }
219 
220 extern void __sync_icache_dcache(pte_t pteval, unsigned long addr);
221 
222 /*
223  * PTE bits configuration in the presence of hardware Dirty Bit Management
224  * (PTE_WRITE == PTE_DBM):
225  *
226  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
227  *   0      0      |   1           0          0
228  *   0      1      |   1           1          0
229  *   1      0      |   1           0          1
230  *   1      1      |   0           1          x
231  *
232  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
233  * the page fault mechanism. Checking the dirty status of a pte becomes:
234  *
235  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
236  */
237 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
238 			      pte_t *ptep, pte_t pte)
239 {
240 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
241 		__sync_icache_dcache(pte, addr);
242 
243 	/*
244 	 * If the existing pte is valid, check for potential race with
245 	 * hardware updates of the pte (ptep_set_access_flags safely changes
246 	 * valid ptes without going through an invalid entry).
247 	 */
248 	if (IS_ENABLED(CONFIG_DEBUG_VM) && pte_valid(*ptep) && pte_valid(pte) &&
249 	   (mm == current->active_mm || atomic_read(&mm->mm_users) > 1)) {
250 		VM_WARN_ONCE(!pte_young(pte),
251 			     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
252 			     __func__, pte_val(*ptep), pte_val(pte));
253 		VM_WARN_ONCE(pte_write(*ptep) && !pte_dirty(pte),
254 			     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
255 			     __func__, pte_val(*ptep), pte_val(pte));
256 	}
257 
258 	set_pte(ptep, pte);
259 }
260 
261 #define __HAVE_ARCH_PTE_SAME
262 static inline int pte_same(pte_t pte_a, pte_t pte_b)
263 {
264 	pteval_t lhs, rhs;
265 
266 	lhs = pte_val(pte_a);
267 	rhs = pte_val(pte_b);
268 
269 	if (pte_present(pte_a))
270 		lhs &= ~PTE_RDONLY;
271 
272 	if (pte_present(pte_b))
273 		rhs &= ~PTE_RDONLY;
274 
275 	return (lhs == rhs);
276 }
277 
278 /*
279  * Huge pte definitions.
280  */
281 #define pte_huge(pte)		(!(pte_val(pte) & PTE_TABLE_BIT))
282 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
283 
284 /*
285  * Hugetlb definitions.
286  */
287 #define HUGE_MAX_HSTATE		4
288 #define HPAGE_SHIFT		PMD_SHIFT
289 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
290 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
291 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
292 
293 #define __HAVE_ARCH_PTE_SPECIAL
294 
295 static inline pte_t pud_pte(pud_t pud)
296 {
297 	return __pte(pud_val(pud));
298 }
299 
300 static inline pmd_t pud_pmd(pud_t pud)
301 {
302 	return __pmd(pud_val(pud));
303 }
304 
305 static inline pte_t pmd_pte(pmd_t pmd)
306 {
307 	return __pte(pmd_val(pmd));
308 }
309 
310 static inline pmd_t pte_pmd(pte_t pte)
311 {
312 	return __pmd(pte_val(pte));
313 }
314 
315 static inline pgprot_t mk_sect_prot(pgprot_t prot)
316 {
317 	return __pgprot(pgprot_val(prot) & ~PTE_TABLE_BIT);
318 }
319 
320 #ifdef CONFIG_NUMA_BALANCING
321 /*
322  * See the comment in include/asm-generic/pgtable.h
323  */
324 static inline int pte_protnone(pte_t pte)
325 {
326 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
327 }
328 
329 static inline int pmd_protnone(pmd_t pmd)
330 {
331 	return pte_protnone(pmd_pte(pmd));
332 }
333 #endif
334 
335 /*
336  * THP definitions.
337  */
338 
339 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
340 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
341 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
342 
343 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
344 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
345 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
346 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
347 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
348 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
349 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
350 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
351 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
352 #define pmd_mknotpresent(pmd)	(__pmd(pmd_val(pmd) & ~PMD_SECT_VALID))
353 
354 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
355 
356 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
357 
358 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
359 
360 #define pmd_pfn(pmd)		(((pmd_val(pmd) & PMD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
361 #define pfn_pmd(pfn,prot)	(__pmd(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
362 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
363 
364 #define pud_write(pud)		pte_write(pud_pte(pud))
365 #define pud_pfn(pud)		(((pud_val(pud) & PUD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
366 
367 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
368 
369 #define __pgprot_modify(prot,mask,bits) \
370 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
371 
372 /*
373  * Mark the prot value as uncacheable and unbufferable.
374  */
375 #define pgprot_noncached(prot) \
376 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
377 #define pgprot_writecombine(prot) \
378 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
379 #define pgprot_device(prot) \
380 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
381 #define __HAVE_PHYS_MEM_ACCESS_PROT
382 struct file;
383 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
384 				     unsigned long size, pgprot_t vma_prot);
385 
386 #define pmd_none(pmd)		(!pmd_val(pmd))
387 
388 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
389 
390 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
391 				 PMD_TYPE_TABLE)
392 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
393 				 PMD_TYPE_SECT)
394 
395 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
396 #define pud_sect(pud)		(0)
397 #define pud_table(pud)		(1)
398 #else
399 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
400 				 PUD_TYPE_SECT)
401 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
402 				 PUD_TYPE_TABLE)
403 #endif
404 
405 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
406 {
407 	*pmdp = pmd;
408 	dsb(ishst);
409 	isb();
410 }
411 
412 static inline void pmd_clear(pmd_t *pmdp)
413 {
414 	set_pmd(pmdp, __pmd(0));
415 }
416 
417 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
418 {
419 	return pmd_val(pmd) & PHYS_MASK & (s32)PAGE_MASK;
420 }
421 
422 /* Find an entry in the third-level page table. */
423 #define pte_index(addr)		(((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
424 
425 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
426 #define pte_offset_kernel(dir,addr)	((pte_t *)__va(pte_offset_phys((dir), (addr))))
427 
428 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
429 #define pte_offset_map_nested(dir,addr)	pte_offset_kernel((dir), (addr))
430 #define pte_unmap(pte)			do { } while (0)
431 #define pte_unmap_nested(pte)		do { } while (0)
432 
433 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
434 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
435 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
436 
437 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(pmd_val(pmd) & PHYS_MASK))
438 
439 /* use ONLY for statically allocated translation tables */
440 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
441 
442 /*
443  * Conversion functions: convert a page and protection to a page entry,
444  * and a page entry and page directory to the page they refer to.
445  */
446 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
447 
448 #if CONFIG_PGTABLE_LEVELS > 2
449 
450 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
451 
452 #define pud_none(pud)		(!pud_val(pud))
453 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
454 #define pud_present(pud)	pte_present(pud_pte(pud))
455 
456 static inline void set_pud(pud_t *pudp, pud_t pud)
457 {
458 	*pudp = pud;
459 	dsb(ishst);
460 	isb();
461 }
462 
463 static inline void pud_clear(pud_t *pudp)
464 {
465 	set_pud(pudp, __pud(0));
466 }
467 
468 static inline phys_addr_t pud_page_paddr(pud_t pud)
469 {
470 	return pud_val(pud) & PHYS_MASK & (s32)PAGE_MASK;
471 }
472 
473 /* Find an entry in the second-level page table. */
474 #define pmd_index(addr)		(((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
475 
476 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(*(dir)) + pmd_index(addr) * sizeof(pmd_t))
477 #define pmd_offset(dir, addr)		((pmd_t *)__va(pmd_offset_phys((dir), (addr))))
478 
479 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
480 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
481 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
482 
483 #define pud_page(pud)		pfn_to_page(__phys_to_pfn(pud_val(pud) & PHYS_MASK))
484 
485 /* use ONLY for statically allocated translation tables */
486 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
487 
488 #else
489 
490 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
491 
492 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
493 #define pmd_set_fixmap(addr)		NULL
494 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
495 #define pmd_clear_fixmap()
496 
497 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
498 
499 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
500 
501 #if CONFIG_PGTABLE_LEVELS > 3
502 
503 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
504 
505 #define pgd_none(pgd)		(!pgd_val(pgd))
506 #define pgd_bad(pgd)		(!(pgd_val(pgd) & 2))
507 #define pgd_present(pgd)	(pgd_val(pgd))
508 
509 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
510 {
511 	*pgdp = pgd;
512 	dsb(ishst);
513 }
514 
515 static inline void pgd_clear(pgd_t *pgdp)
516 {
517 	set_pgd(pgdp, __pgd(0));
518 }
519 
520 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
521 {
522 	return pgd_val(pgd) & PHYS_MASK & (s32)PAGE_MASK;
523 }
524 
525 /* Find an entry in the frst-level page table. */
526 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
527 
528 #define pud_offset_phys(dir, addr)	(pgd_page_paddr(*(dir)) + pud_index(addr) * sizeof(pud_t))
529 #define pud_offset(dir, addr)		((pud_t *)__va(pud_offset_phys((dir), (addr))))
530 
531 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
532 #define pud_set_fixmap_offset(pgd, addr)	pud_set_fixmap(pud_offset_phys(pgd, addr))
533 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
534 
535 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(pgd_val(pgd) & PHYS_MASK))
536 
537 /* use ONLY for statically allocated translation tables */
538 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
539 
540 #else
541 
542 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
543 
544 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
545 #define pud_set_fixmap(addr)		NULL
546 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
547 #define pud_clear_fixmap()
548 
549 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
550 
551 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
552 
553 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
554 
555 /* to find an entry in a page-table-directory */
556 #define pgd_index(addr)		(((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
557 
558 #define pgd_offset_raw(pgd, addr)	((pgd) + pgd_index(addr))
559 
560 #define pgd_offset(mm, addr)	(pgd_offset_raw((mm)->pgd, (addr)))
561 
562 /* to find an entry in a kernel page-table-directory */
563 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
564 
565 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
566 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
567 
568 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
569 {
570 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
571 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE;
572 	/* preserve the hardware dirty information */
573 	if (pte_hw_dirty(pte))
574 		pte = pte_mkdirty(pte);
575 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
576 	return pte;
577 }
578 
579 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
580 {
581 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
582 }
583 
584 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
585 extern int ptep_set_access_flags(struct vm_area_struct *vma,
586 				 unsigned long address, pte_t *ptep,
587 				 pte_t entry, int dirty);
588 
589 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
590 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
591 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
592 					unsigned long address, pmd_t *pmdp,
593 					pmd_t entry, int dirty)
594 {
595 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
596 }
597 #endif
598 
599 /*
600  * Atomic pte/pmd modifications.
601  */
602 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
603 static inline int __ptep_test_and_clear_young(pte_t *ptep)
604 {
605 	pte_t old_pte, pte;
606 
607 	pte = READ_ONCE(*ptep);
608 	do {
609 		old_pte = pte;
610 		pte = pte_mkold(pte);
611 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
612 					       pte_val(old_pte), pte_val(pte));
613 	} while (pte_val(pte) != pte_val(old_pte));
614 
615 	return pte_young(pte);
616 }
617 
618 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
619 					    unsigned long address,
620 					    pte_t *ptep)
621 {
622 	return __ptep_test_and_clear_young(ptep);
623 }
624 
625 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
626 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
627 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
628 					    unsigned long address,
629 					    pmd_t *pmdp)
630 {
631 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
632 }
633 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
634 
635 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
636 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
637 				       unsigned long address, pte_t *ptep)
638 {
639 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
640 }
641 
642 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
643 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
644 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
645 					    unsigned long address, pmd_t *pmdp)
646 {
647 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
648 }
649 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
650 
651 /*
652  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
653  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
654  */
655 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
656 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
657 {
658 	pte_t old_pte, pte;
659 
660 	pte = READ_ONCE(*ptep);
661 	do {
662 		old_pte = pte;
663 		/*
664 		 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
665 		 * clear), set the PTE_DIRTY bit.
666 		 */
667 		if (pte_hw_dirty(pte))
668 			pte = pte_mkdirty(pte);
669 		pte = pte_wrprotect(pte);
670 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
671 					       pte_val(old_pte), pte_val(pte));
672 	} while (pte_val(pte) != pte_val(old_pte));
673 }
674 
675 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
676 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
677 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
678 				      unsigned long address, pmd_t *pmdp)
679 {
680 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
681 }
682 #endif
683 
684 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
685 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
686 
687 /*
688  * Encode and decode a swap entry:
689  *	bits 0-1:	present (must be zero)
690  *	bits 2-7:	swap type
691  *	bits 8-57:	swap offset
692  *	bit  58:	PTE_PROT_NONE (must be zero)
693  */
694 #define __SWP_TYPE_SHIFT	2
695 #define __SWP_TYPE_BITS		6
696 #define __SWP_OFFSET_BITS	50
697 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
698 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
699 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
700 
701 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
702 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
703 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
704 
705 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
706 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
707 
708 /*
709  * Ensure that there are not more swap files than can be encoded in the kernel
710  * PTEs.
711  */
712 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
713 
714 extern int kern_addr_valid(unsigned long addr);
715 
716 #include <asm-generic/pgtable.h>
717 
718 void pgd_cache_init(void);
719 #define pgtable_cache_init	pgd_cache_init
720 
721 /*
722  * On AArch64, the cache coherency is handled via the set_pte_at() function.
723  */
724 static inline void update_mmu_cache(struct vm_area_struct *vma,
725 				    unsigned long addr, pte_t *ptep)
726 {
727 	/*
728 	 * We don't do anything here, so there's a very small chance of
729 	 * us retaking a user fault which we just fixed up. The alternative
730 	 * is doing a dsb(ishst), but that penalises the fastpath.
731 	 */
732 }
733 
734 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
735 
736 #define kc_vaddr_to_offset(v)	((v) & ~VA_START)
737 #define kc_offset_to_vaddr(o)	((o) | VA_START)
738 
739 #endif /* !__ASSEMBLY__ */
740 
741 #endif /* __ASM_PGTABLE_H */
742