xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 2596e07a)
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 
25 /*
26  * Software defined PTE bits definition.
27  */
28 #define PTE_VALID		(_AT(pteval_t, 1) << 0)
29 #define PTE_WRITE		(PTE_DBM)		 /* same as DBM (51) */
30 #define PTE_DIRTY		(_AT(pteval_t, 1) << 55)
31 #define PTE_SPECIAL		(_AT(pteval_t, 1) << 56)
32 #define PTE_PROT_NONE		(_AT(pteval_t, 1) << 58) /* only when !PTE_VALID */
33 
34 /*
35  * VMALLOC and SPARSEMEM_VMEMMAP ranges.
36  *
37  * VMEMAP_SIZE: allows the whole VA space to be covered by a struct page array
38  *	(rounded up to PUD_SIZE).
39  * VMALLOC_START: beginning of the kernel VA space
40  * VMALLOC_END: extends to the available space below vmmemmap, PCI I/O space,
41  *	fixed mappings and modules
42  */
43 #define VMEMMAP_SIZE		ALIGN((1UL << (VA_BITS - PAGE_SHIFT)) * sizeof(struct page), PUD_SIZE)
44 
45 #ifndef CONFIG_KASAN
46 #define VMALLOC_START		(VA_START)
47 #else
48 #include <asm/kasan.h>
49 #define VMALLOC_START		(KASAN_SHADOW_END + SZ_64K)
50 #endif
51 
52 #define VMALLOC_END		(PAGE_OFFSET - PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
53 
54 #define vmemmap			((struct page *)(VMALLOC_END + SZ_64K))
55 
56 #define FIRST_USER_ADDRESS	0UL
57 
58 #ifndef __ASSEMBLY__
59 
60 #include <linux/mmdebug.h>
61 
62 extern void __pte_error(const char *file, int line, unsigned long val);
63 extern void __pmd_error(const char *file, int line, unsigned long val);
64 extern void __pud_error(const char *file, int line, unsigned long val);
65 extern void __pgd_error(const char *file, int line, unsigned long val);
66 
67 #define PROT_DEFAULT		(PTE_TYPE_PAGE | PTE_AF | PTE_SHARED)
68 #define PROT_SECT_DEFAULT	(PMD_TYPE_SECT | PMD_SECT_AF | PMD_SECT_S)
69 
70 #define PROT_DEVICE_nGnRnE	(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_ATTRINDX(MT_DEVICE_nGnRnE))
71 #define PROT_DEVICE_nGnRE	(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_ATTRINDX(MT_DEVICE_nGnRE))
72 #define PROT_NORMAL_NC		(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_ATTRINDX(MT_NORMAL_NC))
73 #define PROT_NORMAL_WT		(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_ATTRINDX(MT_NORMAL_WT))
74 #define PROT_NORMAL		(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_ATTRINDX(MT_NORMAL))
75 
76 #define PROT_SECT_DEVICE_nGnRE	(PROT_SECT_DEFAULT | PMD_SECT_PXN | PMD_SECT_UXN | PMD_ATTRINDX(MT_DEVICE_nGnRE))
77 #define PROT_SECT_NORMAL	(PROT_SECT_DEFAULT | PMD_SECT_PXN | PMD_SECT_UXN | PMD_ATTRINDX(MT_NORMAL))
78 #define PROT_SECT_NORMAL_EXEC	(PROT_SECT_DEFAULT | PMD_SECT_UXN | PMD_ATTRINDX(MT_NORMAL))
79 
80 #define _PAGE_DEFAULT		(PROT_DEFAULT | PTE_ATTRINDX(MT_NORMAL))
81 
82 #define PAGE_KERNEL		__pgprot(_PAGE_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE)
83 #define PAGE_KERNEL_RO		__pgprot(_PAGE_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_RDONLY)
84 #define PAGE_KERNEL_ROX		__pgprot(_PAGE_DEFAULT | PTE_UXN | PTE_DIRTY | PTE_RDONLY)
85 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_DEFAULT | PTE_UXN | PTE_DIRTY | PTE_WRITE)
86 #define PAGE_KERNEL_EXEC_CONT	__pgprot(_PAGE_DEFAULT | PTE_UXN | PTE_DIRTY | PTE_WRITE | PTE_CONT)
87 
88 #define PAGE_HYP		__pgprot(_PAGE_DEFAULT | PTE_HYP)
89 #define PAGE_HYP_DEVICE		__pgprot(PROT_DEVICE_nGnRE | PTE_HYP)
90 
91 #define PAGE_S2			__pgprot(PROT_DEFAULT | PTE_S2_MEMATTR(MT_S2_NORMAL) | PTE_S2_RDONLY)
92 #define PAGE_S2_DEVICE		__pgprot(PROT_DEFAULT | PTE_S2_MEMATTR(MT_S2_DEVICE_nGnRE) | PTE_S2_RDONLY | PTE_UXN)
93 
94 #define PAGE_NONE		__pgprot(((_PAGE_DEFAULT) & ~PTE_VALID) | PTE_PROT_NONE | PTE_PXN | PTE_UXN)
95 #define PAGE_SHARED		__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN | PTE_UXN | PTE_WRITE)
96 #define PAGE_SHARED_EXEC	__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN | PTE_WRITE)
97 #define PAGE_COPY		__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN | PTE_UXN)
98 #define PAGE_COPY_EXEC		__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN)
99 #define PAGE_READONLY		__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN | PTE_UXN)
100 #define PAGE_READONLY_EXEC	__pgprot(_PAGE_DEFAULT | PTE_USER | PTE_NG | PTE_PXN)
101 
102 #define __P000  PAGE_NONE
103 #define __P001  PAGE_READONLY
104 #define __P010  PAGE_COPY
105 #define __P011  PAGE_COPY
106 #define __P100  PAGE_READONLY_EXEC
107 #define __P101  PAGE_READONLY_EXEC
108 #define __P110  PAGE_COPY_EXEC
109 #define __P111  PAGE_COPY_EXEC
110 
111 #define __S000  PAGE_NONE
112 #define __S001  PAGE_READONLY
113 #define __S010  PAGE_SHARED
114 #define __S011  PAGE_SHARED
115 #define __S100  PAGE_READONLY_EXEC
116 #define __S101  PAGE_READONLY_EXEC
117 #define __S110  PAGE_SHARED_EXEC
118 #define __S111  PAGE_SHARED_EXEC
119 
120 /*
121  * ZERO_PAGE is a global shared page that is always zero: used
122  * for zero-mapped memory areas etc..
123  */
124 extern struct page *empty_zero_page;
125 #define ZERO_PAGE(vaddr)	(empty_zero_page)
126 
127 #define pte_ERROR(pte)		__pte_error(__FILE__, __LINE__, pte_val(pte))
128 
129 #define pte_pfn(pte)		((pte_val(pte) & PHYS_MASK) >> PAGE_SHIFT)
130 
131 #define pfn_pte(pfn,prot)	(__pte(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
132 
133 #define pte_none(pte)		(!pte_val(pte))
134 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
135 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
136 
137 /* Find an entry in the third-level page table. */
138 #define pte_index(addr)		(((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
139 
140 #define pte_offset_kernel(dir,addr)	(pmd_page_vaddr(*(dir)) + pte_index(addr))
141 
142 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
143 #define pte_offset_map_nested(dir,addr)	pte_offset_kernel((dir), (addr))
144 #define pte_unmap(pte)			do { } while (0)
145 #define pte_unmap_nested(pte)		do { } while (0)
146 
147 /*
148  * The following only work if pte_present(). Undefined behaviour otherwise.
149  */
150 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
151 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
152 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
153 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
154 #define pte_exec(pte)		(!(pte_val(pte) & PTE_UXN))
155 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
156 #define pte_user(pte)		(!!(pte_val(pte) & PTE_USER))
157 
158 #ifdef CONFIG_ARM64_HW_AFDBM
159 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
160 #else
161 #define pte_hw_dirty(pte)	(0)
162 #endif
163 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
164 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
165 
166 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
167 #define pte_valid_not_user(pte) \
168 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == PTE_VALID)
169 #define pte_valid_young(pte) \
170 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
171 
172 /*
173  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
174  * so that we don't erroneously return false for pages that have been
175  * remapped as PROT_NONE but are yet to be flushed from the TLB.
176  */
177 #define pte_accessible(mm, pte)	\
178 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
179 
180 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
181 {
182 	pte_val(pte) &= ~pgprot_val(prot);
183 	return pte;
184 }
185 
186 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
187 {
188 	pte_val(pte) |= pgprot_val(prot);
189 	return pte;
190 }
191 
192 static inline pte_t pte_wrprotect(pte_t pte)
193 {
194 	return clear_pte_bit(pte, __pgprot(PTE_WRITE));
195 }
196 
197 static inline pte_t pte_mkwrite(pte_t pte)
198 {
199 	return set_pte_bit(pte, __pgprot(PTE_WRITE));
200 }
201 
202 static inline pte_t pte_mkclean(pte_t pte)
203 {
204 	return clear_pte_bit(pte, __pgprot(PTE_DIRTY));
205 }
206 
207 static inline pte_t pte_mkdirty(pte_t pte)
208 {
209 	return set_pte_bit(pte, __pgprot(PTE_DIRTY));
210 }
211 
212 static inline pte_t pte_mkold(pte_t pte)
213 {
214 	return clear_pte_bit(pte, __pgprot(PTE_AF));
215 }
216 
217 static inline pte_t pte_mkyoung(pte_t pte)
218 {
219 	return set_pte_bit(pte, __pgprot(PTE_AF));
220 }
221 
222 static inline pte_t pte_mkspecial(pte_t pte)
223 {
224 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
225 }
226 
227 static inline pte_t pte_mkcont(pte_t pte)
228 {
229 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
230 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
231 }
232 
233 static inline pte_t pte_mknoncont(pte_t pte)
234 {
235 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
236 }
237 
238 static inline pmd_t pmd_mkcont(pmd_t pmd)
239 {
240 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
241 }
242 
243 static inline void set_pte(pte_t *ptep, pte_t pte)
244 {
245 	*ptep = pte;
246 
247 	/*
248 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
249 	 * or update_mmu_cache() have the necessary barriers.
250 	 */
251 	if (pte_valid_not_user(pte)) {
252 		dsb(ishst);
253 		isb();
254 	}
255 }
256 
257 struct mm_struct;
258 struct vm_area_struct;
259 
260 extern void __sync_icache_dcache(pte_t pteval, unsigned long addr);
261 
262 /*
263  * PTE bits configuration in the presence of hardware Dirty Bit Management
264  * (PTE_WRITE == PTE_DBM):
265  *
266  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
267  *   0      0      |   1           0          0
268  *   0      1      |   1           1          0
269  *   1      0      |   1           0          1
270  *   1      1      |   0           1          x
271  *
272  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
273  * the page fault mechanism. Checking the dirty status of a pte becomes:
274  *
275  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
276  */
277 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
278 			      pte_t *ptep, pte_t pte)
279 {
280 	if (pte_valid(pte)) {
281 		if (pte_sw_dirty(pte) && pte_write(pte))
282 			pte_val(pte) &= ~PTE_RDONLY;
283 		else
284 			pte_val(pte) |= PTE_RDONLY;
285 		if (pte_user(pte) && pte_exec(pte) && !pte_special(pte))
286 			__sync_icache_dcache(pte, addr);
287 	}
288 
289 	/*
290 	 * If the existing pte is valid, check for potential race with
291 	 * hardware updates of the pte (ptep_set_access_flags safely changes
292 	 * valid ptes without going through an invalid entry).
293 	 */
294 	if (IS_ENABLED(CONFIG_ARM64_HW_AFDBM) &&
295 	    pte_valid(*ptep) && pte_valid(pte)) {
296 		VM_WARN_ONCE(!pte_young(pte),
297 			     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
298 			     __func__, pte_val(*ptep), pte_val(pte));
299 		VM_WARN_ONCE(pte_write(*ptep) && !pte_dirty(pte),
300 			     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
301 			     __func__, pte_val(*ptep), pte_val(pte));
302 	}
303 
304 	set_pte(ptep, pte);
305 }
306 
307 /*
308  * Huge pte definitions.
309  */
310 #define pte_huge(pte)		(!(pte_val(pte) & PTE_TABLE_BIT))
311 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
312 
313 /*
314  * Hugetlb definitions.
315  */
316 #define HUGE_MAX_HSTATE		4
317 #define HPAGE_SHIFT		PMD_SHIFT
318 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
319 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
320 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
321 
322 #define __HAVE_ARCH_PTE_SPECIAL
323 
324 static inline pte_t pud_pte(pud_t pud)
325 {
326 	return __pte(pud_val(pud));
327 }
328 
329 static inline pmd_t pud_pmd(pud_t pud)
330 {
331 	return __pmd(pud_val(pud));
332 }
333 
334 static inline pte_t pmd_pte(pmd_t pmd)
335 {
336 	return __pte(pmd_val(pmd));
337 }
338 
339 static inline pmd_t pte_pmd(pte_t pte)
340 {
341 	return __pmd(pte_val(pte));
342 }
343 
344 static inline pgprot_t mk_sect_prot(pgprot_t prot)
345 {
346 	return __pgprot(pgprot_val(prot) & ~PTE_TABLE_BIT);
347 }
348 
349 /*
350  * THP definitions.
351  */
352 
353 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
354 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
355 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
356 
357 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
358 #define pmd_young(pmd)		pte_young(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_TYPE_MASK))
366 
367 #define __HAVE_ARCH_PMD_WRITE
368 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
369 
370 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
371 
372 #define pmd_pfn(pmd)		(((pmd_val(pmd) & PMD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
373 #define pfn_pmd(pfn,prot)	(__pmd(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
374 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
375 
376 #define pud_write(pud)		pte_write(pud_pte(pud))
377 #define pud_pfn(pud)		(((pud_val(pud) & PUD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
378 
379 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
380 
381 static inline int has_transparent_hugepage(void)
382 {
383 	return 1;
384 }
385 
386 #define __pgprot_modify(prot,mask,bits) \
387 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
388 
389 /*
390  * Mark the prot value as uncacheable and unbufferable.
391  */
392 #define pgprot_noncached(prot) \
393 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
394 #define pgprot_writecombine(prot) \
395 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
396 #define pgprot_device(prot) \
397 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
398 #define __HAVE_PHYS_MEM_ACCESS_PROT
399 struct file;
400 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
401 				     unsigned long size, pgprot_t vma_prot);
402 
403 #define pmd_none(pmd)		(!pmd_val(pmd))
404 #define pmd_present(pmd)	(pmd_val(pmd))
405 
406 #define pmd_bad(pmd)		(!(pmd_val(pmd) & 2))
407 
408 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
409 				 PMD_TYPE_TABLE)
410 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
411 				 PMD_TYPE_SECT)
412 
413 #ifdef CONFIG_ARM64_64K_PAGES
414 #define pud_sect(pud)		(0)
415 #define pud_table(pud)		(1)
416 #else
417 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
418 				 PUD_TYPE_SECT)
419 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
420 				 PUD_TYPE_TABLE)
421 #endif
422 
423 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
424 {
425 	*pmdp = pmd;
426 	dsb(ishst);
427 	isb();
428 }
429 
430 static inline void pmd_clear(pmd_t *pmdp)
431 {
432 	set_pmd(pmdp, __pmd(0));
433 }
434 
435 static inline pte_t *pmd_page_vaddr(pmd_t pmd)
436 {
437 	return __va(pmd_val(pmd) & PHYS_MASK & (s32)PAGE_MASK);
438 }
439 
440 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(pmd_val(pmd) & PHYS_MASK))
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) & 2))
454 #define pud_present(pud)	(pud_val(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 pmd_t *pud_page_vaddr(pud_t pud)
469 {
470 	return __va(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 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
477 {
478 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(addr);
479 }
480 
481 #define pud_page(pud)		pfn_to_page(__phys_to_pfn(pud_val(pud) & PHYS_MASK))
482 
483 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
484 
485 #if CONFIG_PGTABLE_LEVELS > 3
486 
487 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
488 
489 #define pgd_none(pgd)		(!pgd_val(pgd))
490 #define pgd_bad(pgd)		(!(pgd_val(pgd) & 2))
491 #define pgd_present(pgd)	(pgd_val(pgd))
492 
493 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
494 {
495 	*pgdp = pgd;
496 	dsb(ishst);
497 }
498 
499 static inline void pgd_clear(pgd_t *pgdp)
500 {
501 	set_pgd(pgdp, __pgd(0));
502 }
503 
504 static inline pud_t *pgd_page_vaddr(pgd_t pgd)
505 {
506 	return __va(pgd_val(pgd) & PHYS_MASK & (s32)PAGE_MASK);
507 }
508 
509 /* Find an entry in the frst-level page table. */
510 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
511 
512 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long addr)
513 {
514 	return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(addr);
515 }
516 
517 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(pgd_val(pgd) & PHYS_MASK))
518 
519 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
520 
521 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
522 
523 /* to find an entry in a page-table-directory */
524 #define pgd_index(addr)		(((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
525 
526 #define pgd_offset(mm, addr)	((mm)->pgd+pgd_index(addr))
527 
528 /* to find an entry in a kernel page-table-directory */
529 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
530 
531 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
532 {
533 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
534 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE;
535 	/* preserve the hardware dirty information */
536 	if (pte_hw_dirty(pte))
537 		pte = pte_mkdirty(pte);
538 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
539 	return pte;
540 }
541 
542 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
543 {
544 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
545 }
546 
547 #ifdef CONFIG_ARM64_HW_AFDBM
548 /*
549  * Atomic pte/pmd modifications.
550  */
551 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
552 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
553 					    unsigned long address,
554 					    pte_t *ptep)
555 {
556 	pteval_t pteval;
557 	unsigned int tmp, res;
558 
559 	asm volatile("//	ptep_test_and_clear_young\n"
560 	"	prfm	pstl1strm, %2\n"
561 	"1:	ldxr	%0, %2\n"
562 	"	ubfx	%w3, %w0, %5, #1	// extract PTE_AF (young)\n"
563 	"	and	%0, %0, %4		// clear PTE_AF\n"
564 	"	stxr	%w1, %0, %2\n"
565 	"	cbnz	%w1, 1b\n"
566 	: "=&r" (pteval), "=&r" (tmp), "+Q" (pte_val(*ptep)), "=&r" (res)
567 	: "L" (~PTE_AF), "I" (ilog2(PTE_AF)));
568 
569 	return res;
570 }
571 
572 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
573 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
574 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
575 					    unsigned long address,
576 					    pmd_t *pmdp)
577 {
578 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
579 }
580 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
581 
582 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
583 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
584 				       unsigned long address, pte_t *ptep)
585 {
586 	pteval_t old_pteval;
587 	unsigned int tmp;
588 
589 	asm volatile("//	ptep_get_and_clear\n"
590 	"	prfm	pstl1strm, %2\n"
591 	"1:	ldxr	%0, %2\n"
592 	"	stxr	%w1, xzr, %2\n"
593 	"	cbnz	%w1, 1b\n"
594 	: "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep)));
595 
596 	return __pte(old_pteval);
597 }
598 
599 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
600 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
601 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
602 				       unsigned long address, pmd_t *pmdp)
603 {
604 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
605 }
606 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
607 
608 /*
609  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
610  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
611  */
612 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
613 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
614 {
615 	pteval_t pteval;
616 	unsigned long tmp;
617 
618 	asm volatile("//	ptep_set_wrprotect\n"
619 	"	prfm	pstl1strm, %2\n"
620 	"1:	ldxr	%0, %2\n"
621 	"	tst	%0, %4			// check for hw dirty (!PTE_RDONLY)\n"
622 	"	csel	%1, %3, xzr, eq		// set PTE_DIRTY|PTE_RDONLY if dirty\n"
623 	"	orr	%0, %0, %1		// if !dirty, PTE_RDONLY is already set\n"
624 	"	and	%0, %0, %5		// clear PTE_WRITE/PTE_DBM\n"
625 	"	stxr	%w1, %0, %2\n"
626 	"	cbnz	%w1, 1b\n"
627 	: "=&r" (pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
628 	: "r" (PTE_DIRTY|PTE_RDONLY), "L" (PTE_RDONLY), "L" (~PTE_WRITE)
629 	: "cc");
630 }
631 
632 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
633 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
634 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
635 				      unsigned long address, pmd_t *pmdp)
636 {
637 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
638 }
639 #endif
640 #endif	/* CONFIG_ARM64_HW_AFDBM */
641 
642 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
643 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
644 
645 /*
646  * Encode and decode a swap entry:
647  *	bits 0-1:	present (must be zero)
648  *	bits 2-7:	swap type
649  *	bits 8-57:	swap offset
650  */
651 #define __SWP_TYPE_SHIFT	2
652 #define __SWP_TYPE_BITS		6
653 #define __SWP_OFFSET_BITS	50
654 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
655 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
656 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
657 
658 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
659 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
660 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
661 
662 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
663 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
664 
665 /*
666  * Ensure that there are not more swap files than can be encoded in the kernel
667  * PTEs.
668  */
669 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
670 
671 extern int kern_addr_valid(unsigned long addr);
672 
673 #include <asm-generic/pgtable.h>
674 
675 void pgd_cache_init(void);
676 #define pgtable_cache_init	pgd_cache_init
677 
678 /*
679  * On AArch64, the cache coherency is handled via the set_pte_at() function.
680  */
681 static inline void update_mmu_cache(struct vm_area_struct *vma,
682 				    unsigned long addr, pte_t *ptep)
683 {
684 	/*
685 	 * We don't do anything here, so there's a very small chance of
686 	 * us retaking a user fault which we just fixed up. The alternative
687 	 * is doing a dsb(ishst), but that penalises the fastpath.
688 	 */
689 }
690 
691 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
692 
693 #define kc_vaddr_to_offset(v)	((v) & ~VA_START)
694 #define kc_offset_to_vaddr(o)	((o) | VA_START)
695 
696 #endif /* !__ASSEMBLY__ */
697 
698 #endif /* __ASM_PGTABLE_H */
699