xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 22d55f02)
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 vmmemmap, PCI I/O space
21  *	and fixed mappings
22  */
23 #define VMALLOC_START		(MODULES_END)
24 #define VMALLOC_END		(PAGE_OFFSET - PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
25 
26 #define vmemmap			((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
27 
28 #define FIRST_USER_ADDRESS	0UL
29 
30 #ifndef __ASSEMBLY__
31 
32 #include <asm/cmpxchg.h>
33 #include <asm/fixmap.h>
34 #include <linux/mmdebug.h>
35 #include <linux/mm_types.h>
36 #include <linux/sched.h>
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 
83 #define pte_cont_addr_end(addr, end)						\
84 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
85 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
86 })
87 
88 #define pmd_cont_addr_end(addr, end)						\
89 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
90 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
91 })
92 
93 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
94 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
95 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
96 
97 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
98 /*
99  * Execute-only user mappings do not have the PTE_USER bit set. All valid
100  * kernel mappings have the PTE_UXN bit set.
101  */
102 #define pte_valid_not_user(pte) \
103 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
104 #define pte_valid_young(pte) \
105 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
106 #define pte_valid_user(pte) \
107 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
108 
109 /*
110  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
111  * so that we don't erroneously return false for pages that have been
112  * remapped as PROT_NONE but are yet to be flushed from the TLB.
113  */
114 #define pte_accessible(mm, pte)	\
115 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
116 
117 /*
118  * p??_access_permitted() is true for valid user mappings (subject to the
119  * write permission check) other than user execute-only which do not have the
120  * PTE_USER bit set. PROT_NONE mappings do not have the PTE_VALID bit set.
121  */
122 #define pte_access_permitted(pte, write) \
123 	(pte_valid_user(pte) && (!(write) || pte_write(pte)))
124 #define pmd_access_permitted(pmd, write) \
125 	(pte_access_permitted(pmd_pte(pmd), (write)))
126 #define pud_access_permitted(pud, write) \
127 	(pte_access_permitted(pud_pte(pud), (write)))
128 
129 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
130 {
131 	pte_val(pte) &= ~pgprot_val(prot);
132 	return pte;
133 }
134 
135 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
136 {
137 	pte_val(pte) |= pgprot_val(prot);
138 	return pte;
139 }
140 
141 static inline pte_t pte_wrprotect(pte_t pte)
142 {
143 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
144 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
145 	return pte;
146 }
147 
148 static inline pte_t pte_mkwrite(pte_t pte)
149 {
150 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
151 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
152 	return pte;
153 }
154 
155 static inline pte_t pte_mkclean(pte_t pte)
156 {
157 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
158 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
159 
160 	return pte;
161 }
162 
163 static inline pte_t pte_mkdirty(pte_t pte)
164 {
165 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
166 
167 	if (pte_write(pte))
168 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
169 
170 	return pte;
171 }
172 
173 static inline pte_t pte_mkold(pte_t pte)
174 {
175 	return clear_pte_bit(pte, __pgprot(PTE_AF));
176 }
177 
178 static inline pte_t pte_mkyoung(pte_t pte)
179 {
180 	return set_pte_bit(pte, __pgprot(PTE_AF));
181 }
182 
183 static inline pte_t pte_mkspecial(pte_t pte)
184 {
185 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
186 }
187 
188 static inline pte_t pte_mkcont(pte_t pte)
189 {
190 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
191 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
192 }
193 
194 static inline pte_t pte_mknoncont(pte_t pte)
195 {
196 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
197 }
198 
199 static inline pte_t pte_mkpresent(pte_t pte)
200 {
201 	return set_pte_bit(pte, __pgprot(PTE_VALID));
202 }
203 
204 static inline pmd_t pmd_mkcont(pmd_t pmd)
205 {
206 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
207 }
208 
209 static inline void set_pte(pte_t *ptep, pte_t pte)
210 {
211 	WRITE_ONCE(*ptep, pte);
212 
213 	/*
214 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
215 	 * or update_mmu_cache() have the necessary barriers.
216 	 */
217 	if (pte_valid_not_user(pte))
218 		dsb(ishst);
219 }
220 
221 extern void __sync_icache_dcache(pte_t pteval);
222 
223 /*
224  * PTE bits configuration in the presence of hardware Dirty Bit Management
225  * (PTE_WRITE == PTE_DBM):
226  *
227  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
228  *   0      0      |   1           0          0
229  *   0      1      |   1           1          0
230  *   1      0      |   1           0          1
231  *   1      1      |   0           1          x
232  *
233  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
234  * the page fault mechanism. Checking the dirty status of a pte becomes:
235  *
236  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
237  */
238 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
239 			      pte_t *ptep, pte_t pte)
240 {
241 	pte_t old_pte;
242 
243 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
244 		__sync_icache_dcache(pte);
245 
246 	/*
247 	 * If the existing pte is valid, check for potential race with
248 	 * hardware updates of the pte (ptep_set_access_flags safely changes
249 	 * valid ptes without going through an invalid entry).
250 	 */
251 	old_pte = READ_ONCE(*ptep);
252 	if (IS_ENABLED(CONFIG_DEBUG_VM) && pte_valid(old_pte) && pte_valid(pte) &&
253 	   (mm == current->active_mm || atomic_read(&mm->mm_users) > 1)) {
254 		VM_WARN_ONCE(!pte_young(pte),
255 			     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
256 			     __func__, pte_val(old_pte), pte_val(pte));
257 		VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
258 			     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
259 			     __func__, pte_val(old_pte), pte_val(pte));
260 	}
261 
262 	set_pte(ptep, pte);
263 }
264 
265 #define __HAVE_ARCH_PTE_SAME
266 static inline int pte_same(pte_t pte_a, pte_t pte_b)
267 {
268 	pteval_t lhs, rhs;
269 
270 	lhs = pte_val(pte_a);
271 	rhs = pte_val(pte_b);
272 
273 	if (pte_present(pte_a))
274 		lhs &= ~PTE_RDONLY;
275 
276 	if (pte_present(pte_b))
277 		rhs &= ~PTE_RDONLY;
278 
279 	return (lhs == rhs);
280 }
281 
282 /*
283  * Huge pte definitions.
284  */
285 #define pte_huge(pte)		(!(pte_val(pte) & PTE_TABLE_BIT))
286 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
287 
288 /*
289  * Hugetlb definitions.
290  */
291 #define HUGE_MAX_HSTATE		4
292 #define HPAGE_SHIFT		PMD_SHIFT
293 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
294 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
295 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
296 
297 static inline pte_t pgd_pte(pgd_t pgd)
298 {
299 	return __pte(pgd_val(pgd));
300 }
301 
302 static inline pte_t pud_pte(pud_t pud)
303 {
304 	return __pte(pud_val(pud));
305 }
306 
307 static inline pud_t pte_pud(pte_t pte)
308 {
309 	return __pud(pte_val(pte));
310 }
311 
312 static inline pmd_t pud_pmd(pud_t pud)
313 {
314 	return __pmd(pud_val(pud));
315 }
316 
317 static inline pte_t pmd_pte(pmd_t pmd)
318 {
319 	return __pte(pmd_val(pmd));
320 }
321 
322 static inline pmd_t pte_pmd(pte_t pte)
323 {
324 	return __pmd(pte_val(pte));
325 }
326 
327 static inline pgprot_t mk_sect_prot(pgprot_t prot)
328 {
329 	return __pgprot(pgprot_val(prot) & ~PTE_TABLE_BIT);
330 }
331 
332 #ifdef CONFIG_NUMA_BALANCING
333 /*
334  * See the comment in include/asm-generic/pgtable.h
335  */
336 static inline int pte_protnone(pte_t pte)
337 {
338 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
339 }
340 
341 static inline int pmd_protnone(pmd_t pmd)
342 {
343 	return pte_protnone(pmd_pte(pmd));
344 }
345 #endif
346 
347 /*
348  * THP definitions.
349  */
350 
351 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
352 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
353 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
354 
355 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
356 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
357 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
358 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
359 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
360 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
361 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
362 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
363 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
364 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
365 #define pmd_mknotpresent(pmd)	(__pmd(pmd_val(pmd) & ~PMD_SECT_VALID))
366 
367 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
368 
369 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
370 
371 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
372 
373 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
374 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
375 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
376 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
377 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
378 
379 #define pud_young(pud)		pte_young(pud_pte(pud))
380 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
381 #define pud_write(pud)		pte_write(pud_pte(pud))
382 
383 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
384 
385 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
386 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
387 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
388 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
389 
390 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
391 
392 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
393 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
394 
395 #define __pgprot_modify(prot,mask,bits) \
396 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
397 
398 /*
399  * Mark the prot value as uncacheable and unbufferable.
400  */
401 #define pgprot_noncached(prot) \
402 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
403 #define pgprot_writecombine(prot) \
404 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
405 #define pgprot_device(prot) \
406 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
407 #define __HAVE_PHYS_MEM_ACCESS_PROT
408 struct file;
409 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
410 				     unsigned long size, pgprot_t vma_prot);
411 
412 #define pmd_none(pmd)		(!pmd_val(pmd))
413 
414 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
415 
416 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
417 				 PMD_TYPE_TABLE)
418 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
419 				 PMD_TYPE_SECT)
420 
421 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
422 #define pud_sect(pud)		(0)
423 #define pud_table(pud)		(1)
424 #else
425 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
426 				 PUD_TYPE_SECT)
427 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
428 				 PUD_TYPE_TABLE)
429 #endif
430 
431 extern pgd_t init_pg_dir[PTRS_PER_PGD];
432 extern pgd_t init_pg_end[];
433 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
434 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
435 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
436 
437 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
438 
439 static inline bool in_swapper_pgdir(void *addr)
440 {
441 	return ((unsigned long)addr & PAGE_MASK) ==
442 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
443 }
444 
445 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
446 {
447 #ifdef __PAGETABLE_PMD_FOLDED
448 	if (in_swapper_pgdir(pmdp)) {
449 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
450 		return;
451 	}
452 #endif /* __PAGETABLE_PMD_FOLDED */
453 
454 	WRITE_ONCE(*pmdp, pmd);
455 
456 	if (pmd_valid(pmd))
457 		dsb(ishst);
458 }
459 
460 static inline void pmd_clear(pmd_t *pmdp)
461 {
462 	set_pmd(pmdp, __pmd(0));
463 }
464 
465 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
466 {
467 	return __pmd_to_phys(pmd);
468 }
469 
470 static inline void pte_unmap(pte_t *pte) { }
471 
472 /* Find an entry in the third-level page table. */
473 #define pte_index(addr)		(((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
474 
475 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
476 #define pte_offset_kernel(dir,addr)	((pte_t *)__va(pte_offset_phys((dir), (addr))))
477 
478 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
479 
480 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
481 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
482 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
483 
484 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(__pmd_to_phys(pmd)))
485 
486 /* use ONLY for statically allocated translation tables */
487 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
488 
489 /*
490  * Conversion functions: convert a page and protection to a page entry,
491  * and a page entry and page directory to the page they refer to.
492  */
493 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
494 
495 #if CONFIG_PGTABLE_LEVELS > 2
496 
497 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
498 
499 #define pud_none(pud)		(!pud_val(pud))
500 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
501 #define pud_present(pud)	pte_present(pud_pte(pud))
502 #define pud_valid(pud)		pte_valid(pud_pte(pud))
503 
504 static inline void set_pud(pud_t *pudp, pud_t pud)
505 {
506 #ifdef __PAGETABLE_PUD_FOLDED
507 	if (in_swapper_pgdir(pudp)) {
508 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
509 		return;
510 	}
511 #endif /* __PAGETABLE_PUD_FOLDED */
512 
513 	WRITE_ONCE(*pudp, pud);
514 
515 	if (pud_valid(pud))
516 		dsb(ishst);
517 }
518 
519 static inline void pud_clear(pud_t *pudp)
520 {
521 	set_pud(pudp, __pud(0));
522 }
523 
524 static inline phys_addr_t pud_page_paddr(pud_t pud)
525 {
526 	return __pud_to_phys(pud);
527 }
528 
529 /* Find an entry in the second-level page table. */
530 #define pmd_index(addr)		(((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
531 
532 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
533 #define pmd_offset(dir, addr)		((pmd_t *)__va(pmd_offset_phys((dir), (addr))))
534 
535 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
536 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
537 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
538 
539 #define pud_page(pud)		pfn_to_page(__phys_to_pfn(__pud_to_phys(pud)))
540 
541 /* use ONLY for statically allocated translation tables */
542 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
543 
544 #else
545 
546 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
547 
548 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
549 #define pmd_set_fixmap(addr)		NULL
550 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
551 #define pmd_clear_fixmap()
552 
553 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
554 
555 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
556 
557 #if CONFIG_PGTABLE_LEVELS > 3
558 
559 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
560 
561 #define pgd_none(pgd)		(!pgd_val(pgd))
562 #define pgd_bad(pgd)		(!(pgd_val(pgd) & 2))
563 #define pgd_present(pgd)	(pgd_val(pgd))
564 
565 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
566 {
567 	if (in_swapper_pgdir(pgdp)) {
568 		set_swapper_pgd(pgdp, pgd);
569 		return;
570 	}
571 
572 	WRITE_ONCE(*pgdp, pgd);
573 	dsb(ishst);
574 }
575 
576 static inline void pgd_clear(pgd_t *pgdp)
577 {
578 	set_pgd(pgdp, __pgd(0));
579 }
580 
581 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
582 {
583 	return __pgd_to_phys(pgd);
584 }
585 
586 /* Find an entry in the frst-level page table. */
587 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
588 
589 #define pud_offset_phys(dir, addr)	(pgd_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
590 #define pud_offset(dir, addr)		((pud_t *)__va(pud_offset_phys((dir), (addr))))
591 
592 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
593 #define pud_set_fixmap_offset(pgd, addr)	pud_set_fixmap(pud_offset_phys(pgd, addr))
594 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
595 
596 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd)))
597 
598 /* use ONLY for statically allocated translation tables */
599 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
600 
601 #else
602 
603 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
604 
605 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
606 #define pud_set_fixmap(addr)		NULL
607 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
608 #define pud_clear_fixmap()
609 
610 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
611 
612 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
613 
614 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
615 
616 /* to find an entry in a page-table-directory */
617 #define pgd_index(addr)		(((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
618 
619 #define pgd_offset_raw(pgd, addr)	((pgd) + pgd_index(addr))
620 
621 #define pgd_offset(mm, addr)	(pgd_offset_raw((mm)->pgd, (addr)))
622 
623 /* to find an entry in a kernel page-table-directory */
624 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
625 
626 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
627 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
628 
629 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
630 {
631 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
632 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE;
633 	/* preserve the hardware dirty information */
634 	if (pte_hw_dirty(pte))
635 		pte = pte_mkdirty(pte);
636 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
637 	return pte;
638 }
639 
640 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
641 {
642 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
643 }
644 
645 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
646 extern int ptep_set_access_flags(struct vm_area_struct *vma,
647 				 unsigned long address, pte_t *ptep,
648 				 pte_t entry, int dirty);
649 
650 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
651 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
652 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
653 					unsigned long address, pmd_t *pmdp,
654 					pmd_t entry, int dirty)
655 {
656 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
657 }
658 #endif
659 
660 /*
661  * Atomic pte/pmd modifications.
662  */
663 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
664 static inline int __ptep_test_and_clear_young(pte_t *ptep)
665 {
666 	pte_t old_pte, pte;
667 
668 	pte = READ_ONCE(*ptep);
669 	do {
670 		old_pte = pte;
671 		pte = pte_mkold(pte);
672 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
673 					       pte_val(old_pte), pte_val(pte));
674 	} while (pte_val(pte) != pte_val(old_pte));
675 
676 	return pte_young(pte);
677 }
678 
679 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
680 					    unsigned long address,
681 					    pte_t *ptep)
682 {
683 	return __ptep_test_and_clear_young(ptep);
684 }
685 
686 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
687 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
688 					 unsigned long address, pte_t *ptep)
689 {
690 	int young = ptep_test_and_clear_young(vma, address, ptep);
691 
692 	if (young) {
693 		/*
694 		 * We can elide the trailing DSB here since the worst that can
695 		 * happen is that a CPU continues to use the young entry in its
696 		 * TLB and we mistakenly reclaim the associated page. The
697 		 * window for such an event is bounded by the next
698 		 * context-switch, which provides a DSB to complete the TLB
699 		 * invalidation.
700 		 */
701 		flush_tlb_page_nosync(vma, address);
702 	}
703 
704 	return young;
705 }
706 
707 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
708 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
709 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
710 					    unsigned long address,
711 					    pmd_t *pmdp)
712 {
713 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
714 }
715 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
716 
717 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
718 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
719 				       unsigned long address, pte_t *ptep)
720 {
721 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
722 }
723 
724 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
725 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
726 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
727 					    unsigned long address, pmd_t *pmdp)
728 {
729 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
730 }
731 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
732 
733 /*
734  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
735  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
736  */
737 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
738 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
739 {
740 	pte_t old_pte, pte;
741 
742 	pte = READ_ONCE(*ptep);
743 	do {
744 		old_pte = pte;
745 		/*
746 		 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
747 		 * clear), set the PTE_DIRTY bit.
748 		 */
749 		if (pte_hw_dirty(pte))
750 			pte = pte_mkdirty(pte);
751 		pte = pte_wrprotect(pte);
752 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
753 					       pte_val(old_pte), pte_val(pte));
754 	} while (pte_val(pte) != pte_val(old_pte));
755 }
756 
757 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
758 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
759 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
760 				      unsigned long address, pmd_t *pmdp)
761 {
762 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
763 }
764 
765 #define pmdp_establish pmdp_establish
766 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
767 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
768 {
769 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
770 }
771 #endif
772 
773 /*
774  * Encode and decode a swap entry:
775  *	bits 0-1:	present (must be zero)
776  *	bits 2-7:	swap type
777  *	bits 8-57:	swap offset
778  *	bit  58:	PTE_PROT_NONE (must be zero)
779  */
780 #define __SWP_TYPE_SHIFT	2
781 #define __SWP_TYPE_BITS		6
782 #define __SWP_OFFSET_BITS	50
783 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
784 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
785 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
786 
787 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
788 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
789 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
790 
791 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
792 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
793 
794 /*
795  * Ensure that there are not more swap files than can be encoded in the kernel
796  * PTEs.
797  */
798 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
799 
800 extern int kern_addr_valid(unsigned long addr);
801 
802 #include <asm-generic/pgtable.h>
803 
804 static inline void pgtable_cache_init(void) { }
805 
806 /*
807  * On AArch64, the cache coherency is handled via the set_pte_at() function.
808  */
809 static inline void update_mmu_cache(struct vm_area_struct *vma,
810 				    unsigned long addr, pte_t *ptep)
811 {
812 	/*
813 	 * We don't do anything here, so there's a very small chance of
814 	 * us retaking a user fault which we just fixed up. The alternative
815 	 * is doing a dsb(ishst), but that penalises the fastpath.
816 	 */
817 }
818 
819 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
820 
821 #define kc_vaddr_to_offset(v)	((v) & ~VA_START)
822 #define kc_offset_to_vaddr(o)	((o) | VA_START)
823 
824 #ifdef CONFIG_ARM64_PA_BITS_52
825 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
826 #else
827 #define phys_to_ttbr(addr)	(addr)
828 #endif
829 
830 #endif /* !__ASSEMBLY__ */
831 
832 #endif /* __ASM_PGTABLE_H */
833