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 #ifndef __ASSEMBLY__
30 
31 #include <asm/cmpxchg.h>
32 #include <asm/fixmap.h>
33 #include <linux/mmdebug.h>
34 #include <linux/mm_types.h>
35 #include <linux/sched.h>
36 #include <linux/page_table_check.h>
37 
38 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
39 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
40 
41 /* Set stride and tlb_level in flush_*_tlb_range */
42 #define flush_pmd_tlb_range(vma, addr, end)	\
43 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
44 #define flush_pud_tlb_range(vma, addr, end)	\
45 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
46 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
47 
arch_thp_swp_supported(void)48 static inline bool arch_thp_swp_supported(void)
49 {
50 	return !system_supports_mte();
51 }
52 #define arch_thp_swp_supported arch_thp_swp_supported
53 
54 /*
55  * Outside of a few very special situations (e.g. hibernation), we always
56  * use broadcast TLB invalidation instructions, therefore a spurious page
57  * fault on one CPU which has been handled concurrently by another CPU
58  * does not need to perform additional invalidation.
59  */
60 #define flush_tlb_fix_spurious_fault(vma, address, ptep) do { } while (0)
61 
62 /*
63  * ZERO_PAGE is a global shared page that is always zero: used
64  * for zero-mapped memory areas etc..
65  */
66 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
67 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
68 
69 #define pte_ERROR(e)	\
70 	pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
71 
72 /*
73  * Macros to convert between a physical address and its placement in a
74  * page table entry, taking care of 52-bit addresses.
75  */
76 #ifdef CONFIG_ARM64_PA_BITS_52
__pte_to_phys(pte_t pte)77 static inline phys_addr_t __pte_to_phys(pte_t pte)
78 {
79 	return (pte_val(pte) & PTE_ADDR_LOW) |
80 		((pte_val(pte) & PTE_ADDR_HIGH) << PTE_ADDR_HIGH_SHIFT);
81 }
__phys_to_pte_val(phys_addr_t phys)82 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
83 {
84 	return (phys | (phys >> PTE_ADDR_HIGH_SHIFT)) & PTE_ADDR_MASK;
85 }
86 #else
87 #define __pte_to_phys(pte)	(pte_val(pte) & PTE_ADDR_MASK)
88 #define __phys_to_pte_val(phys)	(phys)
89 #endif
90 
91 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
92 #define pfn_pte(pfn,prot)	\
93 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
94 
95 #define pte_none(pte)		(!pte_val(pte))
96 #define pte_clear(mm,addr,ptep)	set_pte(ptep, __pte(0))
97 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
98 
99 /*
100  * The following only work if pte_present(). Undefined behaviour otherwise.
101  */
102 #define pte_present(pte)	(!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)))
103 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
104 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
105 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
106 #define pte_rdonly(pte)		(!!(pte_val(pte) & PTE_RDONLY))
107 #define pte_user(pte)		(!!(pte_val(pte) & PTE_USER))
108 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
109 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
110 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
111 #define pte_tagged(pte)		((pte_val(pte) & PTE_ATTRINDX_MASK) == \
112 				 PTE_ATTRINDX(MT_NORMAL_TAGGED))
113 
114 #define pte_cont_addr_end(addr, end)						\
115 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
116 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
117 })
118 
119 #define pmd_cont_addr_end(addr, end)						\
120 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
121 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
122 })
123 
124 #define pte_hw_dirty(pte)	(pte_write(pte) && !pte_rdonly(pte))
125 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
126 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
127 
128 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
129 /*
130  * Execute-only user mappings do not have the PTE_USER bit set. All valid
131  * kernel mappings have the PTE_UXN bit set.
132  */
133 #define pte_valid_not_user(pte) \
134 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
135 /*
136  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
137  * so that we don't erroneously return false for pages that have been
138  * remapped as PROT_NONE but are yet to be flushed from the TLB.
139  * Note that we can't make any assumptions based on the state of the access
140  * flag, since ptep_clear_flush_young() elides a DSB when invalidating the
141  * TLB.
142  */
143 #define pte_accessible(mm, pte)	\
144 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
145 
146 /*
147  * p??_access_permitted() is true for valid user mappings (PTE_USER
148  * bit set, subject to the write permission check). For execute-only
149  * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
150  * not set) must return false. PROT_NONE mappings do not have the
151  * PTE_VALID bit set.
152  */
153 #define pte_access_permitted(pte, write) \
154 	(((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
155 #define pmd_access_permitted(pmd, write) \
156 	(pte_access_permitted(pmd_pte(pmd), (write)))
157 #define pud_access_permitted(pud, write) \
158 	(pte_access_permitted(pud_pte(pud), (write)))
159 
clear_pte_bit(pte_t pte,pgprot_t prot)160 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
161 {
162 	pte_val(pte) &= ~pgprot_val(prot);
163 	return pte;
164 }
165 
set_pte_bit(pte_t pte,pgprot_t prot)166 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
167 {
168 	pte_val(pte) |= pgprot_val(prot);
169 	return pte;
170 }
171 
clear_pmd_bit(pmd_t pmd,pgprot_t prot)172 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
173 {
174 	pmd_val(pmd) &= ~pgprot_val(prot);
175 	return pmd;
176 }
177 
set_pmd_bit(pmd_t pmd,pgprot_t prot)178 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
179 {
180 	pmd_val(pmd) |= pgprot_val(prot);
181 	return pmd;
182 }
183 
pte_mkwrite_novma(pte_t pte)184 static inline pte_t pte_mkwrite_novma(pte_t pte)
185 {
186 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
187 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
188 	return pte;
189 }
190 
pte_mkclean(pte_t pte)191 static inline pte_t pte_mkclean(pte_t pte)
192 {
193 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
194 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
195 
196 	return pte;
197 }
198 
pte_mkdirty(pte_t pte)199 static inline pte_t pte_mkdirty(pte_t pte)
200 {
201 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
202 
203 	if (pte_write(pte))
204 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
205 
206 	return pte;
207 }
208 
pte_wrprotect(pte_t pte)209 static inline pte_t pte_wrprotect(pte_t pte)
210 {
211 	/*
212 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
213 	 * clear), set the PTE_DIRTY bit.
214 	 */
215 	if (pte_hw_dirty(pte))
216 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
217 
218 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
219 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
220 	return pte;
221 }
222 
pte_mkold(pte_t pte)223 static inline pte_t pte_mkold(pte_t pte)
224 {
225 	return clear_pte_bit(pte, __pgprot(PTE_AF));
226 }
227 
pte_mkyoung(pte_t pte)228 static inline pte_t pte_mkyoung(pte_t pte)
229 {
230 	return set_pte_bit(pte, __pgprot(PTE_AF));
231 }
232 
pte_mkspecial(pte_t pte)233 static inline pte_t pte_mkspecial(pte_t pte)
234 {
235 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
236 }
237 
pte_mkcont(pte_t pte)238 static inline pte_t pte_mkcont(pte_t pte)
239 {
240 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
241 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
242 }
243 
pte_mknoncont(pte_t pte)244 static inline pte_t pte_mknoncont(pte_t pte)
245 {
246 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
247 }
248 
pte_mkpresent(pte_t pte)249 static inline pte_t pte_mkpresent(pte_t pte)
250 {
251 	return set_pte_bit(pte, __pgprot(PTE_VALID));
252 }
253 
pmd_mkcont(pmd_t pmd)254 static inline pmd_t pmd_mkcont(pmd_t pmd)
255 {
256 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
257 }
258 
pte_mkdevmap(pte_t pte)259 static inline pte_t pte_mkdevmap(pte_t pte)
260 {
261 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
262 }
263 
set_pte(pte_t * ptep,pte_t pte)264 static inline void set_pte(pte_t *ptep, pte_t pte)
265 {
266 	WRITE_ONCE(*ptep, pte);
267 
268 	/*
269 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
270 	 * or update_mmu_cache() have the necessary barriers.
271 	 */
272 	if (pte_valid_not_user(pte)) {
273 		dsb(ishst);
274 		isb();
275 	}
276 }
277 
278 extern void __sync_icache_dcache(pte_t pteval);
279 bool pgattr_change_is_safe(u64 old, u64 new);
280 
281 /*
282  * PTE bits configuration in the presence of hardware Dirty Bit Management
283  * (PTE_WRITE == PTE_DBM):
284  *
285  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
286  *   0      0      |   1           0          0
287  *   0      1      |   1           1          0
288  *   1      0      |   1           0          1
289  *   1      1      |   0           1          x
290  *
291  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
292  * the page fault mechanism. Checking the dirty status of a pte becomes:
293  *
294  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
295  */
296 
__check_safe_pte_update(struct mm_struct * mm,pte_t * ptep,pte_t pte)297 static inline void __check_safe_pte_update(struct mm_struct *mm, pte_t *ptep,
298 					   pte_t pte)
299 {
300 	pte_t old_pte;
301 
302 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
303 		return;
304 
305 	old_pte = READ_ONCE(*ptep);
306 
307 	if (!pte_valid(old_pte) || !pte_valid(pte))
308 		return;
309 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
310 		return;
311 
312 	/*
313 	 * Check for potential race with hardware updates of the pte
314 	 * (ptep_set_access_flags safely changes valid ptes without going
315 	 * through an invalid entry).
316 	 */
317 	VM_WARN_ONCE(!pte_young(pte),
318 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
319 		     __func__, pte_val(old_pte), pte_val(pte));
320 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
321 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
322 		     __func__, pte_val(old_pte), pte_val(pte));
323 	VM_WARN_ONCE(!pgattr_change_is_safe(pte_val(old_pte), pte_val(pte)),
324 		     "%s: unsafe attribute change: 0x%016llx -> 0x%016llx",
325 		     __func__, pte_val(old_pte), pte_val(pte));
326 }
327 
__set_pte_at(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)328 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
329 				pte_t *ptep, pte_t pte)
330 {
331 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
332 		__sync_icache_dcache(pte);
333 
334 	/*
335 	 * If the PTE would provide user space access to the tags associated
336 	 * with it then ensure that the MTE tags are synchronised.  Although
337 	 * pte_access_permitted() returns false for exec only mappings, they
338 	 * don't expose tags (instruction fetches don't check tags).
339 	 */
340 	if (system_supports_mte() && pte_access_permitted(pte, false) &&
341 	    !pte_special(pte) && pte_tagged(pte))
342 		mte_sync_tags(pte);
343 
344 	__check_safe_pte_update(mm, ptep, pte);
345 
346 	set_pte(ptep, pte);
347 }
348 
set_ptes(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte,unsigned int nr)349 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
350 			      pte_t *ptep, pte_t pte, unsigned int nr)
351 {
352 	page_table_check_ptes_set(mm, ptep, pte, nr);
353 
354 	for (;;) {
355 		__set_pte_at(mm, addr, ptep, pte);
356 		if (--nr == 0)
357 			break;
358 		ptep++;
359 		addr += PAGE_SIZE;
360 		pte_val(pte) += PAGE_SIZE;
361 	}
362 }
363 #define set_ptes set_ptes
364 
365 /*
366  * Huge pte definitions.
367  */
368 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
369 
370 /*
371  * Hugetlb definitions.
372  */
373 #define HUGE_MAX_HSTATE		4
374 #define HPAGE_SHIFT		PMD_SHIFT
375 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
376 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
377 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
378 
pgd_pte(pgd_t pgd)379 static inline pte_t pgd_pte(pgd_t pgd)
380 {
381 	return __pte(pgd_val(pgd));
382 }
383 
p4d_pte(p4d_t p4d)384 static inline pte_t p4d_pte(p4d_t p4d)
385 {
386 	return __pte(p4d_val(p4d));
387 }
388 
pud_pte(pud_t pud)389 static inline pte_t pud_pte(pud_t pud)
390 {
391 	return __pte(pud_val(pud));
392 }
393 
pte_pud(pte_t pte)394 static inline pud_t pte_pud(pte_t pte)
395 {
396 	return __pud(pte_val(pte));
397 }
398 
pud_pmd(pud_t pud)399 static inline pmd_t pud_pmd(pud_t pud)
400 {
401 	return __pmd(pud_val(pud));
402 }
403 
pmd_pte(pmd_t pmd)404 static inline pte_t pmd_pte(pmd_t pmd)
405 {
406 	return __pte(pmd_val(pmd));
407 }
408 
pte_pmd(pte_t pte)409 static inline pmd_t pte_pmd(pte_t pte)
410 {
411 	return __pmd(pte_val(pte));
412 }
413 
mk_pud_sect_prot(pgprot_t prot)414 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
415 {
416 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
417 }
418 
mk_pmd_sect_prot(pgprot_t prot)419 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
420 {
421 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
422 }
423 
pte_swp_mkexclusive(pte_t pte)424 static inline pte_t pte_swp_mkexclusive(pte_t pte)
425 {
426 	return set_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
427 }
428 
pte_swp_exclusive(pte_t pte)429 static inline int pte_swp_exclusive(pte_t pte)
430 {
431 	return pte_val(pte) & PTE_SWP_EXCLUSIVE;
432 }
433 
pte_swp_clear_exclusive(pte_t pte)434 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
435 {
436 	return clear_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
437 }
438 
439 /*
440  * Select all bits except the pfn
441  */
pte_pgprot(pte_t pte)442 static inline pgprot_t pte_pgprot(pte_t pte)
443 {
444 	unsigned long pfn = pte_pfn(pte);
445 
446 	return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
447 }
448 
449 #ifdef CONFIG_NUMA_BALANCING
450 /*
451  * See the comment in include/linux/pgtable.h
452  */
pte_protnone(pte_t pte)453 static inline int pte_protnone(pte_t pte)
454 {
455 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
456 }
457 
pmd_protnone(pmd_t pmd)458 static inline int pmd_protnone(pmd_t pmd)
459 {
460 	return pte_protnone(pmd_pte(pmd));
461 }
462 #endif
463 
464 #define pmd_present_invalid(pmd)     (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
465 
pmd_present(pmd_t pmd)466 static inline int pmd_present(pmd_t pmd)
467 {
468 	return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
469 }
470 
471 /*
472  * THP definitions.
473  */
474 
475 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmd_trans_huge(pmd_t pmd)476 static inline int pmd_trans_huge(pmd_t pmd)
477 {
478 	return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
479 }
480 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
481 
482 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
483 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
484 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
485 #define pmd_user(pmd)		pte_user(pmd_pte(pmd))
486 #define pmd_user_exec(pmd)	pte_user_exec(pmd_pte(pmd))
487 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
488 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
489 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
490 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
491 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
492 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
493 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
494 
pmd_mkinvalid(pmd_t pmd)495 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
496 {
497 	pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
498 	pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
499 
500 	return pmd;
501 }
502 
503 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
504 
505 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
506 
507 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
508 
509 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
510 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
511 #endif
pmd_mkdevmap(pmd_t pmd)512 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
513 {
514 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
515 }
516 
517 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
518 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
519 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
520 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
521 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
522 
523 #define pud_young(pud)		pte_young(pud_pte(pud))
524 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
525 #define pud_write(pud)		pte_write(pud_pte(pud))
526 
527 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
528 
529 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
530 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
531 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
532 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
533 
set_pmd_at(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t pmd)534 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
535 			      pmd_t *pmdp, pmd_t pmd)
536 {
537 	page_table_check_pmd_set(mm, pmdp, pmd);
538 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd));
539 }
540 
set_pud_at(struct mm_struct * mm,unsigned long addr,pud_t * pudp,pud_t pud)541 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
542 			      pud_t *pudp, pud_t pud)
543 {
544 	page_table_check_pud_set(mm, pudp, pud);
545 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud));
546 }
547 
548 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
549 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
550 
551 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
552 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
553 
554 #define __pgprot_modify(prot,mask,bits) \
555 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
556 
557 #define pgprot_nx(prot) \
558 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
559 
560 /*
561  * Mark the prot value as uncacheable and unbufferable.
562  */
563 #define pgprot_noncached(prot) \
564 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
565 #define pgprot_writecombine(prot) \
566 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
567 #define pgprot_device(prot) \
568 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
569 #define pgprot_tagged(prot) \
570 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
571 #define pgprot_mhp	pgprot_tagged
572 /*
573  * DMA allocations for non-coherent devices use what the Arm architecture calls
574  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
575  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
576  * is intended for MMIO and thus forbids speculation, preserves access size,
577  * requires strict alignment and can also force write responses to come from the
578  * endpoint.
579  */
580 #define pgprot_dmacoherent(prot) \
581 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
582 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
583 
584 #define __HAVE_PHYS_MEM_ACCESS_PROT
585 struct file;
586 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
587 				     unsigned long size, pgprot_t vma_prot);
588 
589 #define pmd_none(pmd)		(!pmd_val(pmd))
590 
591 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
592 				 PMD_TYPE_TABLE)
593 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
594 				 PMD_TYPE_SECT)
595 #define pmd_leaf(pmd)		(pmd_present(pmd) && !pmd_table(pmd))
596 #define pmd_bad(pmd)		(!pmd_table(pmd))
597 
598 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
599 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
600 
601 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
pud_sect(pud_t pud)602 static inline bool pud_sect(pud_t pud) { return false; }
pud_table(pud_t pud)603 static inline bool pud_table(pud_t pud) { return true; }
604 #else
605 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
606 				 PUD_TYPE_SECT)
607 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
608 				 PUD_TYPE_TABLE)
609 #endif
610 
611 extern pgd_t init_pg_dir[PTRS_PER_PGD];
612 extern pgd_t init_pg_end[];
613 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
614 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
615 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
616 extern pgd_t reserved_pg_dir[PTRS_PER_PGD];
617 
618 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
619 
in_swapper_pgdir(void * addr)620 static inline bool in_swapper_pgdir(void *addr)
621 {
622 	return ((unsigned long)addr & PAGE_MASK) ==
623 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
624 }
625 
set_pmd(pmd_t * pmdp,pmd_t pmd)626 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
627 {
628 #ifdef __PAGETABLE_PMD_FOLDED
629 	if (in_swapper_pgdir(pmdp)) {
630 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
631 		return;
632 	}
633 #endif /* __PAGETABLE_PMD_FOLDED */
634 
635 	WRITE_ONCE(*pmdp, pmd);
636 
637 	if (pmd_valid(pmd)) {
638 		dsb(ishst);
639 		isb();
640 	}
641 }
642 
pmd_clear(pmd_t * pmdp)643 static inline void pmd_clear(pmd_t *pmdp)
644 {
645 	set_pmd(pmdp, __pmd(0));
646 }
647 
pmd_page_paddr(pmd_t pmd)648 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
649 {
650 	return __pmd_to_phys(pmd);
651 }
652 
pmd_page_vaddr(pmd_t pmd)653 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
654 {
655 	return (unsigned long)__va(pmd_page_paddr(pmd));
656 }
657 
658 /* Find an entry in the third-level page table. */
659 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
660 
661 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
662 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
663 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
664 
665 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
666 
667 /* use ONLY for statically allocated translation tables */
668 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
669 
670 /*
671  * Conversion functions: convert a page and protection to a page entry,
672  * and a page entry and page directory to the page they refer to.
673  */
674 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
675 
676 #if CONFIG_PGTABLE_LEVELS > 2
677 
678 #define pmd_ERROR(e)	\
679 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
680 
681 #define pud_none(pud)		(!pud_val(pud))
682 #define pud_bad(pud)		(!pud_table(pud))
683 #define pud_present(pud)	pte_present(pud_pte(pud))
684 #define pud_leaf(pud)		(pud_present(pud) && !pud_table(pud))
685 #define pud_valid(pud)		pte_valid(pud_pte(pud))
686 #define pud_user(pud)		pte_user(pud_pte(pud))
687 #define pud_user_exec(pud)	pte_user_exec(pud_pte(pud))
688 
set_pud(pud_t * pudp,pud_t pud)689 static inline void set_pud(pud_t *pudp, pud_t pud)
690 {
691 #ifdef __PAGETABLE_PUD_FOLDED
692 	if (in_swapper_pgdir(pudp)) {
693 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
694 		return;
695 	}
696 #endif /* __PAGETABLE_PUD_FOLDED */
697 
698 	WRITE_ONCE(*pudp, pud);
699 
700 	if (pud_valid(pud)) {
701 		dsb(ishst);
702 		isb();
703 	}
704 }
705 
pud_clear(pud_t * pudp)706 static inline void pud_clear(pud_t *pudp)
707 {
708 	set_pud(pudp, __pud(0));
709 }
710 
pud_page_paddr(pud_t pud)711 static inline phys_addr_t pud_page_paddr(pud_t pud)
712 {
713 	return __pud_to_phys(pud);
714 }
715 
pud_pgtable(pud_t pud)716 static inline pmd_t *pud_pgtable(pud_t pud)
717 {
718 	return (pmd_t *)__va(pud_page_paddr(pud));
719 }
720 
721 /* Find an entry in the second-level page table. */
722 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
723 
724 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
725 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
726 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
727 
728 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
729 
730 /* use ONLY for statically allocated translation tables */
731 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
732 
733 #else
734 
735 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
736 #define pud_user_exec(pud)	pud_user(pud) /* Always 0 with folding */
737 
738 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
739 #define pmd_set_fixmap(addr)		NULL
740 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
741 #define pmd_clear_fixmap()
742 
743 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
744 
745 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
746 
747 #if CONFIG_PGTABLE_LEVELS > 3
748 
749 #define pud_ERROR(e)	\
750 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
751 
752 #define p4d_none(p4d)		(!p4d_val(p4d))
753 #define p4d_bad(p4d)		(!(p4d_val(p4d) & 2))
754 #define p4d_present(p4d)	(p4d_val(p4d))
755 
set_p4d(p4d_t * p4dp,p4d_t p4d)756 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
757 {
758 	if (in_swapper_pgdir(p4dp)) {
759 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
760 		return;
761 	}
762 
763 	WRITE_ONCE(*p4dp, p4d);
764 	dsb(ishst);
765 	isb();
766 }
767 
p4d_clear(p4d_t * p4dp)768 static inline void p4d_clear(p4d_t *p4dp)
769 {
770 	set_p4d(p4dp, __p4d(0));
771 }
772 
p4d_page_paddr(p4d_t p4d)773 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
774 {
775 	return __p4d_to_phys(p4d);
776 }
777 
p4d_pgtable(p4d_t p4d)778 static inline pud_t *p4d_pgtable(p4d_t p4d)
779 {
780 	return (pud_t *)__va(p4d_page_paddr(p4d));
781 }
782 
783 /* Find an entry in the first-level page table. */
784 #define pud_offset_phys(dir, addr)	(p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
785 
786 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
787 #define pud_set_fixmap_offset(p4d, addr)	pud_set_fixmap(pud_offset_phys(p4d, addr))
788 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
789 
790 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
791 
792 /* use ONLY for statically allocated translation tables */
793 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
794 
795 #else
796 
797 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
798 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
799 
800 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
801 #define pud_set_fixmap(addr)		NULL
802 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
803 #define pud_clear_fixmap()
804 
805 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
806 
807 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
808 
809 #define pgd_ERROR(e)	\
810 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
811 
812 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
813 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
814 
pte_modify(pte_t pte,pgprot_t newprot)815 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
816 {
817 	/*
818 	 * Normal and Normal-Tagged are two different memory types and indices
819 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
820 	 */
821 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
822 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
823 			      PTE_ATTRINDX_MASK;
824 	/* preserve the hardware dirty information */
825 	if (pte_hw_dirty(pte))
826 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
827 
828 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
829 	/*
830 	 * If we end up clearing hw dirtiness for a sw-dirty PTE, set hardware
831 	 * dirtiness again.
832 	 */
833 	if (pte_sw_dirty(pte))
834 		pte = pte_mkdirty(pte);
835 	return pte;
836 }
837 
pmd_modify(pmd_t pmd,pgprot_t newprot)838 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
839 {
840 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
841 }
842 
843 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
844 extern int ptep_set_access_flags(struct vm_area_struct *vma,
845 				 unsigned long address, pte_t *ptep,
846 				 pte_t entry, int dirty);
847 
848 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
849 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
pmdp_set_access_flags(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,pmd_t entry,int dirty)850 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
851 					unsigned long address, pmd_t *pmdp,
852 					pmd_t entry, int dirty)
853 {
854 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
855 }
856 
pud_devmap(pud_t pud)857 static inline int pud_devmap(pud_t pud)
858 {
859 	return 0;
860 }
861 
pgd_devmap(pgd_t pgd)862 static inline int pgd_devmap(pgd_t pgd)
863 {
864 	return 0;
865 }
866 #endif
867 
868 #ifdef CONFIG_PAGE_TABLE_CHECK
pte_user_accessible_page(pte_t pte)869 static inline bool pte_user_accessible_page(pte_t pte)
870 {
871 	return pte_present(pte) && (pte_user(pte) || pte_user_exec(pte));
872 }
873 
pmd_user_accessible_page(pmd_t pmd)874 static inline bool pmd_user_accessible_page(pmd_t pmd)
875 {
876 	return pmd_leaf(pmd) && !pmd_present_invalid(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
877 }
878 
pud_user_accessible_page(pud_t pud)879 static inline bool pud_user_accessible_page(pud_t pud)
880 {
881 	return pud_leaf(pud) && (pud_user(pud) || pud_user_exec(pud));
882 }
883 #endif
884 
885 /*
886  * Atomic pte/pmd modifications.
887  */
888 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
__ptep_test_and_clear_young(pte_t * ptep)889 static inline int __ptep_test_and_clear_young(pte_t *ptep)
890 {
891 	pte_t old_pte, pte;
892 
893 	pte = READ_ONCE(*ptep);
894 	do {
895 		old_pte = pte;
896 		pte = pte_mkold(pte);
897 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
898 					       pte_val(old_pte), pte_val(pte));
899 	} while (pte_val(pte) != pte_val(old_pte));
900 
901 	return pte_young(pte);
902 }
903 
ptep_test_and_clear_young(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)904 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
905 					    unsigned long address,
906 					    pte_t *ptep)
907 {
908 	return __ptep_test_and_clear_young(ptep);
909 }
910 
911 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
ptep_clear_flush_young(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)912 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
913 					 unsigned long address, pte_t *ptep)
914 {
915 	int young = ptep_test_and_clear_young(vma, address, ptep);
916 
917 	if (young) {
918 		/*
919 		 * We can elide the trailing DSB here since the worst that can
920 		 * happen is that a CPU continues to use the young entry in its
921 		 * TLB and we mistakenly reclaim the associated page. The
922 		 * window for such an event is bounded by the next
923 		 * context-switch, which provides a DSB to complete the TLB
924 		 * invalidation.
925 		 */
926 		flush_tlb_page_nosync(vma, address);
927 	}
928 
929 	return young;
930 }
931 
932 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
933 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
pmdp_test_and_clear_young(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)934 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
935 					    unsigned long address,
936 					    pmd_t *pmdp)
937 {
938 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
939 }
940 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
941 
942 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
ptep_get_and_clear(struct mm_struct * mm,unsigned long address,pte_t * ptep)943 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
944 				       unsigned long address, pte_t *ptep)
945 {
946 	pte_t pte = __pte(xchg_relaxed(&pte_val(*ptep), 0));
947 
948 	page_table_check_pte_clear(mm, pte);
949 
950 	return pte;
951 }
952 
953 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
954 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
pmdp_huge_get_and_clear(struct mm_struct * mm,unsigned long address,pmd_t * pmdp)955 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
956 					    unsigned long address, pmd_t *pmdp)
957 {
958 	pmd_t pmd = __pmd(xchg_relaxed(&pmd_val(*pmdp), 0));
959 
960 	page_table_check_pmd_clear(mm, pmd);
961 
962 	return pmd;
963 }
964 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
965 
966 /*
967  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
968  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
969  */
970 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
ptep_set_wrprotect(struct mm_struct * mm,unsigned long address,pte_t * ptep)971 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
972 {
973 	pte_t old_pte, pte;
974 
975 	pte = READ_ONCE(*ptep);
976 	do {
977 		old_pte = pte;
978 		pte = pte_wrprotect(pte);
979 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
980 					       pte_val(old_pte), pte_val(pte));
981 	} while (pte_val(pte) != pte_val(old_pte));
982 }
983 
984 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
985 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
pmdp_set_wrprotect(struct mm_struct * mm,unsigned long address,pmd_t * pmdp)986 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
987 				      unsigned long address, pmd_t *pmdp)
988 {
989 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
990 }
991 
992 #define pmdp_establish pmdp_establish
pmdp_establish(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,pmd_t pmd)993 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
994 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
995 {
996 	page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
997 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
998 }
999 #endif
1000 
1001 /*
1002  * Encode and decode a swap entry:
1003  *	bits 0-1:	present (must be zero)
1004  *	bits 2:		remember PG_anon_exclusive
1005  *	bits 3-7:	swap type
1006  *	bits 8-57:	swap offset
1007  *	bit  58:	PTE_PROT_NONE (must be zero)
1008  */
1009 #define __SWP_TYPE_SHIFT	3
1010 #define __SWP_TYPE_BITS		5
1011 #define __SWP_OFFSET_BITS	50
1012 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
1013 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
1014 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
1015 
1016 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
1017 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
1018 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
1019 
1020 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1021 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
1022 
1023 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1024 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
1025 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
1026 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1027 
1028 /*
1029  * Ensure that there are not more swap files than can be encoded in the kernel
1030  * PTEs.
1031  */
1032 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1033 
1034 #ifdef CONFIG_ARM64_MTE
1035 
1036 #define __HAVE_ARCH_PREPARE_TO_SWAP
arch_prepare_to_swap(struct page * page)1037 static inline int arch_prepare_to_swap(struct page *page)
1038 {
1039 	if (system_supports_mte())
1040 		return mte_save_tags(page);
1041 	return 0;
1042 }
1043 
1044 #define __HAVE_ARCH_SWAP_INVALIDATE
arch_swap_invalidate_page(int type,pgoff_t offset)1045 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1046 {
1047 	if (system_supports_mte())
1048 		mte_invalidate_tags(type, offset);
1049 }
1050 
arch_swap_invalidate_area(int type)1051 static inline void arch_swap_invalidate_area(int type)
1052 {
1053 	if (system_supports_mte())
1054 		mte_invalidate_tags_area(type);
1055 }
1056 
1057 #define __HAVE_ARCH_SWAP_RESTORE
arch_swap_restore(swp_entry_t entry,struct folio * folio)1058 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1059 {
1060 	if (system_supports_mte())
1061 		mte_restore_tags(entry, &folio->page);
1062 }
1063 
1064 #endif /* CONFIG_ARM64_MTE */
1065 
1066 /*
1067  * On AArch64, the cache coherency is handled via the set_pte_at() function.
1068  */
update_mmu_cache_range(struct vm_fault * vmf,struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,unsigned int nr)1069 static inline void update_mmu_cache_range(struct vm_fault *vmf,
1070 		struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1071 		unsigned int nr)
1072 {
1073 	/*
1074 	 * We don't do anything here, so there's a very small chance of
1075 	 * us retaking a user fault which we just fixed up. The alternative
1076 	 * is doing a dsb(ishst), but that penalises the fastpath.
1077 	 */
1078 }
1079 
1080 #define update_mmu_cache(vma, addr, ptep) \
1081 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
1082 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
1083 
1084 #ifdef CONFIG_ARM64_PA_BITS_52
1085 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
1086 #else
1087 #define phys_to_ttbr(addr)	(addr)
1088 #endif
1089 
1090 /*
1091  * On arm64 without hardware Access Flag, copying from user will fail because
1092  * the pte is old and cannot be marked young. So we always end up with zeroed
1093  * page after fork() + CoW for pfn mappings. We don't always have a
1094  * hardware-managed access flag on arm64.
1095  */
1096 #define arch_has_hw_pte_young		cpu_has_hw_af
1097 
1098 /*
1099  * Experimentally, it's cheap to set the access flag in hardware and we
1100  * benefit from prefaulting mappings as 'old' to start with.
1101  */
1102 #define arch_wants_old_prefaulted_pte	cpu_has_hw_af
1103 
pud_sect_supported(void)1104 static inline bool pud_sect_supported(void)
1105 {
1106 	return PAGE_SIZE == SZ_4K;
1107 }
1108 
1109 
1110 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1111 #define ptep_modify_prot_start ptep_modify_prot_start
1112 extern pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1113 				    unsigned long addr, pte_t *ptep);
1114 
1115 #define ptep_modify_prot_commit ptep_modify_prot_commit
1116 extern void ptep_modify_prot_commit(struct vm_area_struct *vma,
1117 				    unsigned long addr, pte_t *ptep,
1118 				    pte_t old_pte, pte_t new_pte);
1119 #endif /* !__ASSEMBLY__ */
1120 
1121 #endif /* __ASM_PGTABLE_H */
1122