xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision a06c488d)
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_ATTRINDX(MT_DEVICE_nGnRnE))
71 #define PROT_DEVICE_nGnRE	(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_ATTRINDX(MT_DEVICE_nGnRE))
72 #define PROT_NORMAL_NC		(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_ATTRINDX(MT_NORMAL_NC))
73 #define PROT_NORMAL_WT		(PROT_DEFAULT | PTE_PXN | PTE_UXN | PTE_ATTRINDX(MT_NORMAL_WT))
74 #define PROT_NORMAL		(PROT_DEFAULT | PTE_PXN | PTE_UXN | 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 
157 #ifdef CONFIG_ARM64_HW_AFDBM
158 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
159 #else
160 #define pte_hw_dirty(pte)	(0)
161 #endif
162 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
163 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
164 
165 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
166 #define pte_valid_user(pte) \
167 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
168 #define pte_valid_not_user(pte) \
169 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == PTE_VALID)
170 #define pte_valid_young(pte) \
171 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
172 
173 /*
174  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
175  * so that we don't erroneously return false for pages that have been
176  * remapped as PROT_NONE but are yet to be flushed from the TLB.
177  */
178 #define pte_accessible(mm, pte)	\
179 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
180 
181 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
182 {
183 	pte_val(pte) &= ~pgprot_val(prot);
184 	return pte;
185 }
186 
187 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
188 {
189 	pte_val(pte) |= pgprot_val(prot);
190 	return pte;
191 }
192 
193 static inline pte_t pte_wrprotect(pte_t pte)
194 {
195 	return clear_pte_bit(pte, __pgprot(PTE_WRITE));
196 }
197 
198 static inline pte_t pte_mkwrite(pte_t pte)
199 {
200 	return set_pte_bit(pte, __pgprot(PTE_WRITE));
201 }
202 
203 static inline pte_t pte_mkclean(pte_t pte)
204 {
205 	return clear_pte_bit(pte, __pgprot(PTE_DIRTY));
206 }
207 
208 static inline pte_t pte_mkdirty(pte_t pte)
209 {
210 	return set_pte_bit(pte, __pgprot(PTE_DIRTY));
211 }
212 
213 static inline pte_t pte_mkold(pte_t pte)
214 {
215 	return clear_pte_bit(pte, __pgprot(PTE_AF));
216 }
217 
218 static inline pte_t pte_mkyoung(pte_t pte)
219 {
220 	return set_pte_bit(pte, __pgprot(PTE_AF));
221 }
222 
223 static inline pte_t pte_mkspecial(pte_t pte)
224 {
225 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
226 }
227 
228 static inline pte_t pte_mkcont(pte_t pte)
229 {
230 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
231 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
232 }
233 
234 static inline pte_t pte_mknoncont(pte_t pte)
235 {
236 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
237 }
238 
239 static inline pmd_t pmd_mkcont(pmd_t pmd)
240 {
241 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
242 }
243 
244 static inline void set_pte(pte_t *ptep, pte_t pte)
245 {
246 	*ptep = pte;
247 
248 	/*
249 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
250 	 * or update_mmu_cache() have the necessary barriers.
251 	 */
252 	if (pte_valid_not_user(pte)) {
253 		dsb(ishst);
254 		isb();
255 	}
256 }
257 
258 struct mm_struct;
259 struct vm_area_struct;
260 
261 extern void __sync_icache_dcache(pte_t pteval, unsigned long addr);
262 
263 /*
264  * PTE bits configuration in the presence of hardware Dirty Bit Management
265  * (PTE_WRITE == PTE_DBM):
266  *
267  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
268  *   0      0      |   1           0          0
269  *   0      1      |   1           1          0
270  *   1      0      |   1           0          1
271  *   1      1      |   0           1          x
272  *
273  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
274  * the page fault mechanism. Checking the dirty status of a pte becomes:
275  *
276  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
277  */
278 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
279 			      pte_t *ptep, pte_t pte)
280 {
281 	if (pte_valid_user(pte)) {
282 		if (!pte_special(pte) && pte_exec(pte))
283 			__sync_icache_dcache(pte, addr);
284 		if (pte_sw_dirty(pte) && pte_write(pte))
285 			pte_val(pte) &= ~PTE_RDONLY;
286 		else
287 			pte_val(pte) |= PTE_RDONLY;
288 	}
289 
290 	/*
291 	 * If the existing pte is valid, check for potential race with
292 	 * hardware updates of the pte (ptep_set_access_flags safely changes
293 	 * valid ptes without going through an invalid entry).
294 	 */
295 	if (IS_ENABLED(CONFIG_ARM64_HW_AFDBM) &&
296 	    pte_valid(*ptep) && pte_valid(pte)) {
297 		VM_WARN_ONCE(!pte_young(pte),
298 			     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
299 			     __func__, pte_val(*ptep), pte_val(pte));
300 		VM_WARN_ONCE(pte_write(*ptep) && !pte_dirty(pte),
301 			     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
302 			     __func__, pte_val(*ptep), pte_val(pte));
303 	}
304 
305 	set_pte(ptep, pte);
306 }
307 
308 /*
309  * Huge pte definitions.
310  */
311 #define pte_huge(pte)		(!(pte_val(pte) & PTE_TABLE_BIT))
312 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
313 
314 /*
315  * Hugetlb definitions.
316  */
317 #define HUGE_MAX_HSTATE		4
318 #define HPAGE_SHIFT		PMD_SHIFT
319 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
320 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
321 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
322 
323 #define __HAVE_ARCH_PTE_SPECIAL
324 
325 static inline pte_t pud_pte(pud_t pud)
326 {
327 	return __pte(pud_val(pud));
328 }
329 
330 static inline pmd_t pud_pmd(pud_t pud)
331 {
332 	return __pmd(pud_val(pud));
333 }
334 
335 static inline pte_t pmd_pte(pmd_t pmd)
336 {
337 	return __pte(pmd_val(pmd));
338 }
339 
340 static inline pmd_t pte_pmd(pte_t pte)
341 {
342 	return __pmd(pte_val(pte));
343 }
344 
345 static inline pgprot_t mk_sect_prot(pgprot_t prot)
346 {
347 	return __pgprot(pgprot_val(prot) & ~PTE_TABLE_BIT);
348 }
349 
350 /*
351  * THP definitions.
352  */
353 
354 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
355 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT))
356 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
357 
358 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
359 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
360 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
361 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
362 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
363 #define pmd_mkclean(pmd)       pte_pmd(pte_mkclean(pmd_pte(pmd)))
364 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
365 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
366 #define pmd_mknotpresent(pmd)	(__pmd(pmd_val(pmd) & ~PMD_TYPE_MASK))
367 
368 #define __HAVE_ARCH_PMD_WRITE
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_pfn(pmd)		(((pmd_val(pmd) & PMD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
374 #define pfn_pmd(pfn,prot)	(__pmd(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
375 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
376 
377 #define pud_write(pud)		pte_write(pud_pte(pud))
378 #define pud_pfn(pud)		(((pud_val(pud) & PUD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
379 
380 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
381 
382 static inline int has_transparent_hugepage(void)
383 {
384 	return 1;
385 }
386 
387 #define __pgprot_modify(prot,mask,bits) \
388 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
389 
390 /*
391  * Mark the prot value as uncacheable and unbufferable.
392  */
393 #define pgprot_noncached(prot) \
394 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
395 #define pgprot_writecombine(prot) \
396 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
397 #define pgprot_device(prot) \
398 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
399 #define __HAVE_PHYS_MEM_ACCESS_PROT
400 struct file;
401 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
402 				     unsigned long size, pgprot_t vma_prot);
403 
404 #define pmd_none(pmd)		(!pmd_val(pmd))
405 #define pmd_present(pmd)	(pmd_val(pmd))
406 
407 #define pmd_bad(pmd)		(!(pmd_val(pmd) & 2))
408 
409 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
410 				 PMD_TYPE_TABLE)
411 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
412 				 PMD_TYPE_SECT)
413 
414 #ifdef CONFIG_ARM64_64K_PAGES
415 #define pud_sect(pud)		(0)
416 #define pud_table(pud)		(1)
417 #else
418 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
419 				 PUD_TYPE_SECT)
420 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
421 				 PUD_TYPE_TABLE)
422 #endif
423 
424 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
425 {
426 	*pmdp = pmd;
427 	dsb(ishst);
428 	isb();
429 }
430 
431 static inline void pmd_clear(pmd_t *pmdp)
432 {
433 	set_pmd(pmdp, __pmd(0));
434 }
435 
436 static inline pte_t *pmd_page_vaddr(pmd_t pmd)
437 {
438 	return __va(pmd_val(pmd) & PHYS_MASK & (s32)PAGE_MASK);
439 }
440 
441 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(pmd_val(pmd) & PHYS_MASK))
442 
443 /*
444  * Conversion functions: convert a page and protection to a page entry,
445  * and a page entry and page directory to the page they refer to.
446  */
447 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
448 
449 #if CONFIG_PGTABLE_LEVELS > 2
450 
451 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
452 
453 #define pud_none(pud)		(!pud_val(pud))
454 #define pud_bad(pud)		(!(pud_val(pud) & 2))
455 #define pud_present(pud)	(pud_val(pud))
456 
457 static inline void set_pud(pud_t *pudp, pud_t pud)
458 {
459 	*pudp = pud;
460 	dsb(ishst);
461 	isb();
462 }
463 
464 static inline void pud_clear(pud_t *pudp)
465 {
466 	set_pud(pudp, __pud(0));
467 }
468 
469 static inline pmd_t *pud_page_vaddr(pud_t pud)
470 {
471 	return __va(pud_val(pud) & PHYS_MASK & (s32)PAGE_MASK);
472 }
473 
474 /* Find an entry in the second-level page table. */
475 #define pmd_index(addr)		(((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
476 
477 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
478 {
479 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(addr);
480 }
481 
482 #define pud_page(pud)		pfn_to_page(__phys_to_pfn(pud_val(pud) & PHYS_MASK))
483 
484 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
485 
486 #if CONFIG_PGTABLE_LEVELS > 3
487 
488 #define pud_ERROR(pud)		__pud_error(__FILE__, __LINE__, pud_val(pud))
489 
490 #define pgd_none(pgd)		(!pgd_val(pgd))
491 #define pgd_bad(pgd)		(!(pgd_val(pgd) & 2))
492 #define pgd_present(pgd)	(pgd_val(pgd))
493 
494 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
495 {
496 	*pgdp = pgd;
497 	dsb(ishst);
498 }
499 
500 static inline void pgd_clear(pgd_t *pgdp)
501 {
502 	set_pgd(pgdp, __pgd(0));
503 }
504 
505 static inline pud_t *pgd_page_vaddr(pgd_t pgd)
506 {
507 	return __va(pgd_val(pgd) & PHYS_MASK & (s32)PAGE_MASK);
508 }
509 
510 /* Find an entry in the frst-level page table. */
511 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
512 
513 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long addr)
514 {
515 	return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(addr);
516 }
517 
518 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(pgd_val(pgd) & PHYS_MASK))
519 
520 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
521 
522 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
523 
524 /* to find an entry in a page-table-directory */
525 #define pgd_index(addr)		(((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
526 
527 #define pgd_offset(mm, addr)	((mm)->pgd+pgd_index(addr))
528 
529 /* to find an entry in a kernel page-table-directory */
530 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
531 
532 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
533 {
534 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
535 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE;
536 	/* preserve the hardware dirty information */
537 	if (pte_hw_dirty(pte))
538 		pte = pte_mkdirty(pte);
539 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
540 	return pte;
541 }
542 
543 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
544 {
545 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
546 }
547 
548 #ifdef CONFIG_ARM64_HW_AFDBM
549 /*
550  * Atomic pte/pmd modifications.
551  */
552 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
553 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
554 					    unsigned long address,
555 					    pte_t *ptep)
556 {
557 	pteval_t pteval;
558 	unsigned int tmp, res;
559 
560 	asm volatile("//	ptep_test_and_clear_young\n"
561 	"	prfm	pstl1strm, %2\n"
562 	"1:	ldxr	%0, %2\n"
563 	"	ubfx	%w3, %w0, %5, #1	// extract PTE_AF (young)\n"
564 	"	and	%0, %0, %4		// clear PTE_AF\n"
565 	"	stxr	%w1, %0, %2\n"
566 	"	cbnz	%w1, 1b\n"
567 	: "=&r" (pteval), "=&r" (tmp), "+Q" (pte_val(*ptep)), "=&r" (res)
568 	: "L" (~PTE_AF), "I" (ilog2(PTE_AF)));
569 
570 	return res;
571 }
572 
573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
574 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
575 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
576 					    unsigned long address,
577 					    pmd_t *pmdp)
578 {
579 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
580 }
581 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
582 
583 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
584 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
585 				       unsigned long address, pte_t *ptep)
586 {
587 	pteval_t old_pteval;
588 	unsigned int tmp;
589 
590 	asm volatile("//	ptep_get_and_clear\n"
591 	"	prfm	pstl1strm, %2\n"
592 	"1:	ldxr	%0, %2\n"
593 	"	stxr	%w1, xzr, %2\n"
594 	"	cbnz	%w1, 1b\n"
595 	: "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep)));
596 
597 	return __pte(old_pteval);
598 }
599 
600 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
601 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
602 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
603 				       unsigned long address, pmd_t *pmdp)
604 {
605 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
606 }
607 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
608 
609 /*
610  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
611  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
612  */
613 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
614 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
615 {
616 	pteval_t pteval;
617 	unsigned long tmp;
618 
619 	asm volatile("//	ptep_set_wrprotect\n"
620 	"	prfm	pstl1strm, %2\n"
621 	"1:	ldxr	%0, %2\n"
622 	"	tst	%0, %4			// check for hw dirty (!PTE_RDONLY)\n"
623 	"	csel	%1, %3, xzr, eq		// set PTE_DIRTY|PTE_RDONLY if dirty\n"
624 	"	orr	%0, %0, %1		// if !dirty, PTE_RDONLY is already set\n"
625 	"	and	%0, %0, %5		// clear PTE_WRITE/PTE_DBM\n"
626 	"	stxr	%w1, %0, %2\n"
627 	"	cbnz	%w1, 1b\n"
628 	: "=&r" (pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
629 	: "r" (PTE_DIRTY|PTE_RDONLY), "L" (PTE_RDONLY), "L" (~PTE_WRITE)
630 	: "cc");
631 }
632 
633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
634 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
635 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
636 				      unsigned long address, pmd_t *pmdp)
637 {
638 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
639 }
640 #endif
641 #endif	/* CONFIG_ARM64_HW_AFDBM */
642 
643 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
644 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
645 
646 /*
647  * Encode and decode a swap entry:
648  *	bits 0-1:	present (must be zero)
649  *	bits 2-7:	swap type
650  *	bits 8-57:	swap offset
651  */
652 #define __SWP_TYPE_SHIFT	2
653 #define __SWP_TYPE_BITS		6
654 #define __SWP_OFFSET_BITS	50
655 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
656 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
657 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
658 
659 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
660 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
661 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
662 
663 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
664 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
665 
666 /*
667  * Ensure that there are not more swap files than can be encoded in the kernel
668  * PTEs.
669  */
670 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
671 
672 extern int kern_addr_valid(unsigned long addr);
673 
674 #include <asm-generic/pgtable.h>
675 
676 void pgd_cache_init(void);
677 #define pgtable_cache_init	pgd_cache_init
678 
679 /*
680  * On AArch64, the cache coherency is handled via the set_pte_at() function.
681  */
682 static inline void update_mmu_cache(struct vm_area_struct *vma,
683 				    unsigned long addr, pte_t *ptep)
684 {
685 	/*
686 	 * We don't do anything here, so there's a very small chance of
687 	 * us retaking a user fault which we just fixed up. The alternative
688 	 * is doing a dsb(ishst), but that penalises the fastpath.
689 	 */
690 }
691 
692 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
693 
694 #define kc_vaddr_to_offset(v)	((v) & ~VA_START)
695 #define kc_offset_to_vaddr(o)	((o) | VA_START)
696 
697 #endif /* !__ASSEMBLY__ */
698 
699 #endif /* __ASM_PGTABLE_H */
700