xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision a5907065)
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/mte.h>
13 #include <asm/pgtable-hwdef.h>
14 #include <asm/pgtable-prot.h>
15 #include <asm/tlbflush.h>
16 
17 /*
18  * VMALLOC range.
19  *
20  * VMALLOC_START: beginning of the kernel vmalloc space
21  * VMALLOC_END: extends to the available space below vmemmap, PCI I/O space
22  *	and fixed mappings
23  */
24 #define VMALLOC_START		(MODULES_END)
25 #define VMALLOC_END		(VMEMMAP_START - SZ_256M)
26 
27 #define vmemmap			((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
28 
29 #define FIRST_USER_ADDRESS	0UL
30 
31 #ifndef __ASSEMBLY__
32 
33 #include <asm/cmpxchg.h>
34 #include <asm/fixmap.h>
35 #include <linux/mmdebug.h>
36 #include <linux/mm_types.h>
37 #include <linux/sched.h>
38 
39 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
40 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
41 
42 /* Set stride and tlb_level in flush_*_tlb_range */
43 #define flush_pmd_tlb_range(vma, addr, end)	\
44 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
45 #define flush_pud_tlb_range(vma, addr, end)	\
46 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
47 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
48 
49 /*
50  * Outside of a few very special situations (e.g. hibernation), we always
51  * use broadcast TLB invalidation instructions, therefore a spurious page
52  * fault on one CPU which has been handled concurrently by another CPU
53  * does not need to perform additional invalidation.
54  */
55 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
56 
57 /*
58  * ZERO_PAGE is a global shared page that is always zero: used
59  * for zero-mapped memory areas etc..
60  */
61 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
62 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
63 
64 #define pte_ERROR(e)	\
65 	pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
66 
67 /*
68  * Macros to convert between a physical address and its placement in a
69  * page table entry, taking care of 52-bit addresses.
70  */
71 #ifdef CONFIG_ARM64_PA_BITS_52
72 #define __pte_to_phys(pte)	\
73 	((pte_val(pte) & PTE_ADDR_LOW) | ((pte_val(pte) & PTE_ADDR_HIGH) << 36))
74 #define __phys_to_pte_val(phys)	(((phys) | ((phys) >> 36)) & PTE_ADDR_MASK)
75 #else
76 #define __pte_to_phys(pte)	(pte_val(pte) & PTE_ADDR_MASK)
77 #define __phys_to_pte_val(phys)	(phys)
78 #endif
79 
80 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
81 #define pfn_pte(pfn,prot)	\
82 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
83 
84 #define pte_none(pte)		(!pte_val(pte))
85 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
86 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
87 
88 /*
89  * The following only work if pte_present(). Undefined behaviour otherwise.
90  */
91 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
92 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
93 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
94 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
95 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
96 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
97 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
98 #define pte_tagged(pte)		((pte_val(pte) & PTE_ATTRINDX_MASK) == \
99 				 PTE_ATTRINDX(MT_NORMAL_TAGGED))
100 
101 #define pte_cont_addr_end(addr, end)						\
102 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
103 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
104 })
105 
106 #define pmd_cont_addr_end(addr, end)						\
107 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
108 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
109 })
110 
111 #define pte_hw_dirty(pte)	(pte_write(pte) && !(pte_val(pte) & PTE_RDONLY))
112 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
113 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
114 
115 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
116 #define pte_valid_not_user(pte) \
117 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == PTE_VALID)
118 #define pte_valid_user(pte) \
119 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
120 
121 /*
122  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
123  * so that we don't erroneously return false for pages that have been
124  * remapped as PROT_NONE but are yet to be flushed from the TLB.
125  * Note that we can't make any assumptions based on the state of the access
126  * flag, since ptep_clear_flush_young() elides a DSB when invalidating the
127  * TLB.
128  */
129 #define pte_accessible(mm, pte)	\
130 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
131 
132 /*
133  * p??_access_permitted() is true for valid user mappings (subject to the
134  * write permission check). PROT_NONE mappings do not have the PTE_VALID bit
135  * set.
136  */
137 #define pte_access_permitted(pte, write) \
138 	(pte_valid_user(pte) && (!(write) || pte_write(pte)))
139 #define pmd_access_permitted(pmd, write) \
140 	(pte_access_permitted(pmd_pte(pmd), (write)))
141 #define pud_access_permitted(pud, write) \
142 	(pte_access_permitted(pud_pte(pud), (write)))
143 
144 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
145 {
146 	pte_val(pte) &= ~pgprot_val(prot);
147 	return pte;
148 }
149 
150 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
151 {
152 	pte_val(pte) |= pgprot_val(prot);
153 	return pte;
154 }
155 
156 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
157 {
158 	pmd_val(pmd) &= ~pgprot_val(prot);
159 	return pmd;
160 }
161 
162 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
163 {
164 	pmd_val(pmd) |= pgprot_val(prot);
165 	return pmd;
166 }
167 
168 static inline pte_t pte_mkwrite(pte_t pte)
169 {
170 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
171 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
172 	return pte;
173 }
174 
175 static inline pte_t pte_mkclean(pte_t pte)
176 {
177 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
178 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
179 
180 	return pte;
181 }
182 
183 static inline pte_t pte_mkdirty(pte_t pte)
184 {
185 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
186 
187 	if (pte_write(pte))
188 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
189 
190 	return pte;
191 }
192 
193 static inline pte_t pte_wrprotect(pte_t pte)
194 {
195 	/*
196 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
197 	 * clear), set the PTE_DIRTY bit.
198 	 */
199 	if (pte_hw_dirty(pte))
200 		pte = pte_mkdirty(pte);
201 
202 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
203 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
204 	return pte;
205 }
206 
207 static inline pte_t pte_mkold(pte_t pte)
208 {
209 	return clear_pte_bit(pte, __pgprot(PTE_AF));
210 }
211 
212 static inline pte_t pte_mkyoung(pte_t pte)
213 {
214 	return set_pte_bit(pte, __pgprot(PTE_AF));
215 }
216 
217 static inline pte_t pte_mkspecial(pte_t pte)
218 {
219 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
220 }
221 
222 static inline pte_t pte_mkcont(pte_t pte)
223 {
224 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
225 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
226 }
227 
228 static inline pte_t pte_mknoncont(pte_t pte)
229 {
230 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
231 }
232 
233 static inline pte_t pte_mkpresent(pte_t pte)
234 {
235 	return set_pte_bit(pte, __pgprot(PTE_VALID));
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 pte_t pte_mkdevmap(pte_t pte)
244 {
245 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
246 }
247 
248 static inline void set_pte(pte_t *ptep, pte_t pte)
249 {
250 	WRITE_ONCE(*ptep, pte);
251 
252 	/*
253 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
254 	 * or update_mmu_cache() have the necessary barriers.
255 	 */
256 	if (pte_valid_not_user(pte)) {
257 		dsb(ishst);
258 		isb();
259 	}
260 }
261 
262 extern void __sync_icache_dcache(pte_t pteval);
263 
264 /*
265  * PTE bits configuration in the presence of hardware Dirty Bit Management
266  * (PTE_WRITE == PTE_DBM):
267  *
268  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
269  *   0      0      |   1           0          0
270  *   0      1      |   1           1          0
271  *   1      0      |   1           0          1
272  *   1      1      |   0           1          x
273  *
274  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
275  * the page fault mechanism. Checking the dirty status of a pte becomes:
276  *
277  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
278  */
279 
280 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
281 					   pte_t pte)
282 {
283 	pte_t old_pte;
284 
285 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
286 		return;
287 
288 	old_pte = READ_ONCE(*ptep);
289 
290 	if (!pte_valid(old_pte) || !pte_valid(pte))
291 		return;
292 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
293 		return;
294 
295 	/*
296 	 * Check for potential race with hardware updates of the pte
297 	 * (ptep_set_access_flags safely changes valid ptes without going
298 	 * through an invalid entry).
299 	 */
300 	VM_WARN_ONCE(!pte_young(pte),
301 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
302 		     __func__, pte_val(old_pte), pte_val(pte));
303 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
304 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
305 		     __func__, pte_val(old_pte), pte_val(pte));
306 }
307 
308 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
309 			      pte_t *ptep, pte_t pte)
310 {
311 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
312 		__sync_icache_dcache(pte);
313 
314 	if (system_supports_mte() &&
315 	    pte_present(pte) && pte_tagged(pte) && !pte_special(pte))
316 		mte_sync_tags(ptep, pte);
317 
318 	__check_racy_pte_update(mm, ptep, pte);
319 
320 	set_pte(ptep, pte);
321 }
322 
323 /*
324  * Huge pte definitions.
325  */
326 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
327 
328 /*
329  * Hugetlb definitions.
330  */
331 #define HUGE_MAX_HSTATE		4
332 #define HPAGE_SHIFT		PMD_SHIFT
333 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
334 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
335 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
336 
337 static inline pte_t pgd_pte(pgd_t pgd)
338 {
339 	return __pte(pgd_val(pgd));
340 }
341 
342 static inline pte_t p4d_pte(p4d_t p4d)
343 {
344 	return __pte(p4d_val(p4d));
345 }
346 
347 static inline pte_t pud_pte(pud_t pud)
348 {
349 	return __pte(pud_val(pud));
350 }
351 
352 static inline pud_t pte_pud(pte_t pte)
353 {
354 	return __pud(pte_val(pte));
355 }
356 
357 static inline pmd_t pud_pmd(pud_t pud)
358 {
359 	return __pmd(pud_val(pud));
360 }
361 
362 static inline pte_t pmd_pte(pmd_t pmd)
363 {
364 	return __pte(pmd_val(pmd));
365 }
366 
367 static inline pmd_t pte_pmd(pte_t pte)
368 {
369 	return __pmd(pte_val(pte));
370 }
371 
372 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
373 {
374 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
375 }
376 
377 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
378 {
379 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
380 }
381 
382 #ifdef CONFIG_NUMA_BALANCING
383 /*
384  * See the comment in include/linux/pgtable.h
385  */
386 static inline int pte_protnone(pte_t pte)
387 {
388 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
389 }
390 
391 static inline int pmd_protnone(pmd_t pmd)
392 {
393 	return pte_protnone(pmd_pte(pmd));
394 }
395 #endif
396 
397 #define pmd_present_invalid(pmd)     (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
398 
399 static inline int pmd_present(pmd_t pmd)
400 {
401 	return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
402 }
403 
404 /*
405  * THP definitions.
406  */
407 
408 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
409 static inline int pmd_trans_huge(pmd_t pmd)
410 {
411 	return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
412 }
413 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
414 
415 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
416 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
417 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
418 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
419 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
420 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
421 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
422 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
423 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
424 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
425 
426 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
427 {
428 	pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
429 	pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
430 
431 	return pmd;
432 }
433 
434 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
435 
436 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
437 
438 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
439 
440 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
441 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
442 #endif
443 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
444 {
445 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
446 }
447 
448 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
449 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
450 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
451 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
452 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
453 
454 #define pud_young(pud)		pte_young(pud_pte(pud))
455 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
456 #define pud_write(pud)		pte_write(pud_pte(pud))
457 
458 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
459 
460 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
461 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
462 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
463 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
464 
465 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
466 #define set_pud_at(mm, addr, pudp, pud)	set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud))
467 
468 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
469 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
470 
471 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
472 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
473 
474 #define __pgprot_modify(prot,mask,bits) \
475 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
476 
477 #define pgprot_nx(prot) \
478 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
479 
480 /*
481  * Mark the prot value as uncacheable and unbufferable.
482  */
483 #define pgprot_noncached(prot) \
484 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
485 #define pgprot_writecombine(prot) \
486 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
487 #define pgprot_device(prot) \
488 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
489 /*
490  * DMA allocations for non-coherent devices use what the Arm architecture calls
491  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
492  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
493  * is intended for MMIO and thus forbids speculation, preserves access size,
494  * requires strict alignment and can also force write responses to come from the
495  * endpoint.
496  */
497 #define pgprot_dmacoherent(prot) \
498 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
499 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
500 
501 #define __HAVE_PHYS_MEM_ACCESS_PROT
502 struct file;
503 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
504 				     unsigned long size, pgprot_t vma_prot);
505 
506 #define pmd_none(pmd)		(!pmd_val(pmd))
507 
508 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
509 
510 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
511 				 PMD_TYPE_TABLE)
512 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
513 				 PMD_TYPE_SECT)
514 #define pmd_leaf(pmd)		pmd_sect(pmd)
515 
516 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
517 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
518 
519 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
520 static inline bool pud_sect(pud_t pud) { return false; }
521 static inline bool pud_table(pud_t pud) { return true; }
522 #else
523 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
524 				 PUD_TYPE_SECT)
525 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
526 				 PUD_TYPE_TABLE)
527 #endif
528 
529 extern pgd_t init_pg_dir[PTRS_PER_PGD];
530 extern pgd_t init_pg_end[];
531 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
532 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
533 extern pgd_t idmap_pg_end[];
534 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
535 extern pgd_t reserved_pg_dir[PTRS_PER_PGD];
536 
537 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
538 
539 static inline bool in_swapper_pgdir(void *addr)
540 {
541 	return ((unsigned long)addr & PAGE_MASK) ==
542 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
543 }
544 
545 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
546 {
547 #ifdef __PAGETABLE_PMD_FOLDED
548 	if (in_swapper_pgdir(pmdp)) {
549 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
550 		return;
551 	}
552 #endif /* __PAGETABLE_PMD_FOLDED */
553 
554 	WRITE_ONCE(*pmdp, pmd);
555 
556 	if (pmd_valid(pmd)) {
557 		dsb(ishst);
558 		isb();
559 	}
560 }
561 
562 static inline void pmd_clear(pmd_t *pmdp)
563 {
564 	set_pmd(pmdp, __pmd(0));
565 }
566 
567 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
568 {
569 	return __pmd_to_phys(pmd);
570 }
571 
572 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
573 {
574 	return (unsigned long)__va(pmd_page_paddr(pmd));
575 }
576 
577 /* Find an entry in the third-level page table. */
578 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
579 
580 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
581 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
582 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
583 
584 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
585 
586 /* use ONLY for statically allocated translation tables */
587 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
588 
589 /*
590  * Conversion functions: convert a page and protection to a page entry,
591  * and a page entry and page directory to the page they refer to.
592  */
593 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
594 
595 #if CONFIG_PGTABLE_LEVELS > 2
596 
597 #define pmd_ERROR(e)	\
598 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
599 
600 #define pud_none(pud)		(!pud_val(pud))
601 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
602 #define pud_present(pud)	pte_present(pud_pte(pud))
603 #define pud_leaf(pud)		pud_sect(pud)
604 #define pud_valid(pud)		pte_valid(pud_pte(pud))
605 
606 static inline void set_pud(pud_t *pudp, pud_t pud)
607 {
608 #ifdef __PAGETABLE_PUD_FOLDED
609 	if (in_swapper_pgdir(pudp)) {
610 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
611 		return;
612 	}
613 #endif /* __PAGETABLE_PUD_FOLDED */
614 
615 	WRITE_ONCE(*pudp, pud);
616 
617 	if (pud_valid(pud)) {
618 		dsb(ishst);
619 		isb();
620 	}
621 }
622 
623 static inline void pud_clear(pud_t *pudp)
624 {
625 	set_pud(pudp, __pud(0));
626 }
627 
628 static inline phys_addr_t pud_page_paddr(pud_t pud)
629 {
630 	return __pud_to_phys(pud);
631 }
632 
633 static inline unsigned long pud_page_vaddr(pud_t pud)
634 {
635 	return (unsigned long)__va(pud_page_paddr(pud));
636 }
637 
638 /* Find an entry in the second-level page table. */
639 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
640 
641 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
642 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
643 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
644 
645 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
646 
647 /* use ONLY for statically allocated translation tables */
648 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
649 
650 #else
651 
652 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
653 
654 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
655 #define pmd_set_fixmap(addr)		NULL
656 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
657 #define pmd_clear_fixmap()
658 
659 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
660 
661 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
662 
663 #if CONFIG_PGTABLE_LEVELS > 3
664 
665 #define pud_ERROR(e)	\
666 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
667 
668 #define p4d_none(p4d)		(!p4d_val(p4d))
669 #define p4d_bad(p4d)		(!(p4d_val(p4d) & 2))
670 #define p4d_present(p4d)	(p4d_val(p4d))
671 
672 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
673 {
674 	if (in_swapper_pgdir(p4dp)) {
675 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
676 		return;
677 	}
678 
679 	WRITE_ONCE(*p4dp, p4d);
680 	dsb(ishst);
681 	isb();
682 }
683 
684 static inline void p4d_clear(p4d_t *p4dp)
685 {
686 	set_p4d(p4dp, __p4d(0));
687 }
688 
689 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
690 {
691 	return __p4d_to_phys(p4d);
692 }
693 
694 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
695 {
696 	return (unsigned long)__va(p4d_page_paddr(p4d));
697 }
698 
699 /* Find an entry in the frst-level page table. */
700 #define pud_offset_phys(dir, addr)	(p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
701 
702 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
703 #define pud_set_fixmap_offset(p4d, addr)	pud_set_fixmap(pud_offset_phys(p4d, addr))
704 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
705 
706 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
707 
708 /* use ONLY for statically allocated translation tables */
709 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
710 
711 #else
712 
713 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
714 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
715 
716 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
717 #define pud_set_fixmap(addr)		NULL
718 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
719 #define pud_clear_fixmap()
720 
721 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
722 
723 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
724 
725 #define pgd_ERROR(e)	\
726 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
727 
728 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
729 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
730 
731 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
732 {
733 	/*
734 	 * Normal and Normal-Tagged are two different memory types and indices
735 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
736 	 */
737 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
738 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
739 			      PTE_ATTRINDX_MASK;
740 	/* preserve the hardware dirty information */
741 	if (pte_hw_dirty(pte))
742 		pte = pte_mkdirty(pte);
743 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
744 	return pte;
745 }
746 
747 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
748 {
749 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
750 }
751 
752 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
753 extern int ptep_set_access_flags(struct vm_area_struct *vma,
754 				 unsigned long address, pte_t *ptep,
755 				 pte_t entry, int dirty);
756 
757 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
758 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
759 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
760 					unsigned long address, pmd_t *pmdp,
761 					pmd_t entry, int dirty)
762 {
763 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
764 }
765 
766 static inline int pud_devmap(pud_t pud)
767 {
768 	return 0;
769 }
770 
771 static inline int pgd_devmap(pgd_t pgd)
772 {
773 	return 0;
774 }
775 #endif
776 
777 /*
778  * Atomic pte/pmd modifications.
779  */
780 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
781 static inline int __ptep_test_and_clear_young(pte_t *ptep)
782 {
783 	pte_t old_pte, pte;
784 
785 	pte = READ_ONCE(*ptep);
786 	do {
787 		old_pte = pte;
788 		pte = pte_mkold(pte);
789 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
790 					       pte_val(old_pte), pte_val(pte));
791 	} while (pte_val(pte) != pte_val(old_pte));
792 
793 	return pte_young(pte);
794 }
795 
796 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
797 					    unsigned long address,
798 					    pte_t *ptep)
799 {
800 	return __ptep_test_and_clear_young(ptep);
801 }
802 
803 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
804 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
805 					 unsigned long address, pte_t *ptep)
806 {
807 	int young = ptep_test_and_clear_young(vma, address, ptep);
808 
809 	if (young) {
810 		/*
811 		 * We can elide the trailing DSB here since the worst that can
812 		 * happen is that a CPU continues to use the young entry in its
813 		 * TLB and we mistakenly reclaim the associated page. The
814 		 * window for such an event is bounded by the next
815 		 * context-switch, which provides a DSB to complete the TLB
816 		 * invalidation.
817 		 */
818 		flush_tlb_page_nosync(vma, address);
819 	}
820 
821 	return young;
822 }
823 
824 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
825 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
826 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
827 					    unsigned long address,
828 					    pmd_t *pmdp)
829 {
830 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
831 }
832 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
833 
834 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
835 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
836 				       unsigned long address, pte_t *ptep)
837 {
838 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
839 }
840 
841 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
842 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
843 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
844 					    unsigned long address, pmd_t *pmdp)
845 {
846 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
847 }
848 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
849 
850 /*
851  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
852  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
853  */
854 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
855 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
856 {
857 	pte_t old_pte, pte;
858 
859 	pte = READ_ONCE(*ptep);
860 	do {
861 		old_pte = pte;
862 		pte = pte_wrprotect(pte);
863 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
864 					       pte_val(old_pte), pte_val(pte));
865 	} while (pte_val(pte) != pte_val(old_pte));
866 }
867 
868 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
869 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
870 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
871 				      unsigned long address, pmd_t *pmdp)
872 {
873 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
874 }
875 
876 #define pmdp_establish pmdp_establish
877 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
878 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
879 {
880 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
881 }
882 #endif
883 
884 /*
885  * Encode and decode a swap entry:
886  *	bits 0-1:	present (must be zero)
887  *	bits 2-7:	swap type
888  *	bits 8-57:	swap offset
889  *	bit  58:	PTE_PROT_NONE (must be zero)
890  */
891 #define __SWP_TYPE_SHIFT	2
892 #define __SWP_TYPE_BITS		6
893 #define __SWP_OFFSET_BITS	50
894 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
895 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
896 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
897 
898 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
899 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
900 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
901 
902 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
903 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
904 
905 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
906 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
907 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
908 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
909 
910 /*
911  * Ensure that there are not more swap files than can be encoded in the kernel
912  * PTEs.
913  */
914 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
915 
916 extern int kern_addr_valid(unsigned long addr);
917 
918 #ifdef CONFIG_ARM64_MTE
919 
920 #define __HAVE_ARCH_PREPARE_TO_SWAP
921 static inline int arch_prepare_to_swap(struct page *page)
922 {
923 	if (system_supports_mte())
924 		return mte_save_tags(page);
925 	return 0;
926 }
927 
928 #define __HAVE_ARCH_SWAP_INVALIDATE
929 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
930 {
931 	if (system_supports_mte())
932 		mte_invalidate_tags(type, offset);
933 }
934 
935 static inline void arch_swap_invalidate_area(int type)
936 {
937 	if (system_supports_mte())
938 		mte_invalidate_tags_area(type);
939 }
940 
941 #define __HAVE_ARCH_SWAP_RESTORE
942 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
943 {
944 	if (system_supports_mte() && mte_restore_tags(entry, page))
945 		set_bit(PG_mte_tagged, &page->flags);
946 }
947 
948 #endif /* CONFIG_ARM64_MTE */
949 
950 /*
951  * On AArch64, the cache coherency is handled via the set_pte_at() function.
952  */
953 static inline void update_mmu_cache(struct vm_area_struct *vma,
954 				    unsigned long addr, pte_t *ptep)
955 {
956 	/*
957 	 * We don't do anything here, so there's a very small chance of
958 	 * us retaking a user fault which we just fixed up. The alternative
959 	 * is doing a dsb(ishst), but that penalises the fastpath.
960 	 */
961 }
962 
963 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
964 
965 #ifdef CONFIG_ARM64_PA_BITS_52
966 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
967 #else
968 #define phys_to_ttbr(addr)	(addr)
969 #endif
970 
971 /*
972  * On arm64 without hardware Access Flag, copying from user will fail because
973  * the pte is old and cannot be marked young. So we always end up with zeroed
974  * page after fork() + CoW for pfn mappings. We don't always have a
975  * hardware-managed access flag on arm64.
976  */
977 static inline bool arch_faults_on_old_pte(void)
978 {
979 	WARN_ON(preemptible());
980 
981 	return !cpu_has_hw_af();
982 }
983 #define arch_faults_on_old_pte		arch_faults_on_old_pte
984 
985 /*
986  * Experimentally, it's cheap to set the access flag in hardware and we
987  * benefit from prefaulting mappings as 'old' to start with.
988  */
989 static inline bool arch_wants_old_prefaulted_pte(void)
990 {
991 	return !arch_faults_on_old_pte();
992 }
993 #define arch_wants_old_prefaulted_pte	arch_wants_old_prefaulted_pte
994 
995 #endif /* !__ASSEMBLY__ */
996 
997 #endif /* __ASM_PGTABLE_H */
998