xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 2b076054)
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		(- PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
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_young(pte) \
119 	((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF))
120 #define pte_valid_user(pte) \
121 	((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER))
122 
123 /*
124  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
125  * so that we don't erroneously return false for pages that have been
126  * remapped as PROT_NONE but are yet to be flushed from the TLB.
127  */
128 #define pte_accessible(mm, pte)	\
129 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte))
130 
131 /*
132  * p??_access_permitted() is true for valid user mappings (subject to the
133  * write permission check). PROT_NONE mappings do not have the PTE_VALID bit
134  * set.
135  */
136 #define pte_access_permitted(pte, write) \
137 	(pte_valid_user(pte) && (!(write) || pte_write(pte)))
138 #define pmd_access_permitted(pmd, write) \
139 	(pte_access_permitted(pmd_pte(pmd), (write)))
140 #define pud_access_permitted(pud, write) \
141 	(pte_access_permitted(pud_pte(pud), (write)))
142 
143 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
144 {
145 	pte_val(pte) &= ~pgprot_val(prot);
146 	return pte;
147 }
148 
149 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
150 {
151 	pte_val(pte) |= pgprot_val(prot);
152 	return pte;
153 }
154 
155 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
156 {
157 	pmd_val(pmd) &= ~pgprot_val(prot);
158 	return pmd;
159 }
160 
161 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
162 {
163 	pmd_val(pmd) |= pgprot_val(prot);
164 	return pmd;
165 }
166 
167 static inline pte_t pte_wrprotect(pte_t pte)
168 {
169 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
170 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
171 	return pte;
172 }
173 
174 static inline pte_t pte_mkwrite(pte_t pte)
175 {
176 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
177 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
178 	return pte;
179 }
180 
181 static inline pte_t pte_mkclean(pte_t pte)
182 {
183 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
184 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
185 
186 	return pte;
187 }
188 
189 static inline pte_t pte_mkdirty(pte_t pte)
190 {
191 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
192 
193 	if (pte_write(pte))
194 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
195 
196 	return pte;
197 }
198 
199 static inline pte_t pte_mkold(pte_t pte)
200 {
201 	return clear_pte_bit(pte, __pgprot(PTE_AF));
202 }
203 
204 static inline pte_t pte_mkyoung(pte_t pte)
205 {
206 	return set_pte_bit(pte, __pgprot(PTE_AF));
207 }
208 
209 static inline pte_t pte_mkspecial(pte_t pte)
210 {
211 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
212 }
213 
214 static inline pte_t pte_mkcont(pte_t pte)
215 {
216 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
217 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
218 }
219 
220 static inline pte_t pte_mknoncont(pte_t pte)
221 {
222 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
223 }
224 
225 static inline pte_t pte_mkpresent(pte_t pte)
226 {
227 	return set_pte_bit(pte, __pgprot(PTE_VALID));
228 }
229 
230 static inline pmd_t pmd_mkcont(pmd_t pmd)
231 {
232 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
233 }
234 
235 static inline pte_t pte_mkdevmap(pte_t pte)
236 {
237 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
238 }
239 
240 static inline void set_pte(pte_t *ptep, pte_t pte)
241 {
242 	WRITE_ONCE(*ptep, pte);
243 
244 	/*
245 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
246 	 * or update_mmu_cache() have the necessary barriers.
247 	 */
248 	if (pte_valid_not_user(pte)) {
249 		dsb(ishst);
250 		isb();
251 	}
252 }
253 
254 extern void __sync_icache_dcache(pte_t pteval);
255 
256 /*
257  * PTE bits configuration in the presence of hardware Dirty Bit Management
258  * (PTE_WRITE == PTE_DBM):
259  *
260  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
261  *   0      0      |   1           0          0
262  *   0      1      |   1           1          0
263  *   1      0      |   1           0          1
264  *   1      1      |   0           1          x
265  *
266  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
267  * the page fault mechanism. Checking the dirty status of a pte becomes:
268  *
269  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
270  */
271 
272 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
273 					   pte_t pte)
274 {
275 	pte_t old_pte;
276 
277 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
278 		return;
279 
280 	old_pte = READ_ONCE(*ptep);
281 
282 	if (!pte_valid(old_pte) || !pte_valid(pte))
283 		return;
284 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
285 		return;
286 
287 	/*
288 	 * Check for potential race with hardware updates of the pte
289 	 * (ptep_set_access_flags safely changes valid ptes without going
290 	 * through an invalid entry).
291 	 */
292 	VM_WARN_ONCE(!pte_young(pte),
293 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
294 		     __func__, pte_val(old_pte), pte_val(pte));
295 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
296 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
297 		     __func__, pte_val(old_pte), pte_val(pte));
298 }
299 
300 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
301 			      pte_t *ptep, pte_t pte)
302 {
303 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
304 		__sync_icache_dcache(pte);
305 
306 	if (system_supports_mte() &&
307 	    pte_present(pte) && pte_tagged(pte) && !pte_special(pte))
308 		mte_sync_tags(ptep, pte);
309 
310 	__check_racy_pte_update(mm, ptep, pte);
311 
312 	set_pte(ptep, pte);
313 }
314 
315 /*
316  * Huge pte definitions.
317  */
318 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
319 
320 /*
321  * Hugetlb definitions.
322  */
323 #define HUGE_MAX_HSTATE		4
324 #define HPAGE_SHIFT		PMD_SHIFT
325 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
326 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
327 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
328 
329 static inline pte_t pgd_pte(pgd_t pgd)
330 {
331 	return __pte(pgd_val(pgd));
332 }
333 
334 static inline pte_t p4d_pte(p4d_t p4d)
335 {
336 	return __pte(p4d_val(p4d));
337 }
338 
339 static inline pte_t pud_pte(pud_t pud)
340 {
341 	return __pte(pud_val(pud));
342 }
343 
344 static inline pud_t pte_pud(pte_t pte)
345 {
346 	return __pud(pte_val(pte));
347 }
348 
349 static inline pmd_t pud_pmd(pud_t pud)
350 {
351 	return __pmd(pud_val(pud));
352 }
353 
354 static inline pte_t pmd_pte(pmd_t pmd)
355 {
356 	return __pte(pmd_val(pmd));
357 }
358 
359 static inline pmd_t pte_pmd(pte_t pte)
360 {
361 	return __pmd(pte_val(pte));
362 }
363 
364 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
365 {
366 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
367 }
368 
369 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
370 {
371 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
372 }
373 
374 #ifdef CONFIG_NUMA_BALANCING
375 /*
376  * See the comment in include/linux/pgtable.h
377  */
378 static inline int pte_protnone(pte_t pte)
379 {
380 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
381 }
382 
383 static inline int pmd_protnone(pmd_t pmd)
384 {
385 	return pte_protnone(pmd_pte(pmd));
386 }
387 #endif
388 
389 #define pmd_present_invalid(pmd)     (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
390 
391 static inline int pmd_present(pmd_t pmd)
392 {
393 	return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
394 }
395 
396 /*
397  * THP definitions.
398  */
399 
400 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
401 static inline int pmd_trans_huge(pmd_t pmd)
402 {
403 	return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
404 }
405 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
406 
407 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
408 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
409 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
410 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
411 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
412 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
413 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
414 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
415 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
416 
417 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
418 {
419 	pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
420 	pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
421 
422 	return pmd;
423 }
424 
425 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
426 
427 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
428 
429 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
430 
431 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
432 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
433 #endif
434 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
435 {
436 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
437 }
438 
439 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
440 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
441 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
442 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
443 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
444 
445 #define pud_young(pud)		pte_young(pud_pte(pud))
446 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
447 #define pud_write(pud)		pte_write(pud_pte(pud))
448 
449 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
450 
451 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
452 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
453 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
454 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
455 
456 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
457 
458 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
459 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
460 
461 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
462 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
463 
464 #define __pgprot_modify(prot,mask,bits) \
465 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
466 
467 #define pgprot_nx(prot) \
468 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
469 
470 /*
471  * Mark the prot value as uncacheable and unbufferable.
472  */
473 #define pgprot_noncached(prot) \
474 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
475 #define pgprot_writecombine(prot) \
476 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
477 #define pgprot_device(prot) \
478 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
479 /*
480  * DMA allocations for non-coherent devices use what the Arm architecture calls
481  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
482  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
483  * is intended for MMIO and thus forbids speculation, preserves access size,
484  * requires strict alignment and can also force write responses to come from the
485  * endpoint.
486  */
487 #define pgprot_dmacoherent(prot) \
488 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
489 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
490 
491 #define __HAVE_PHYS_MEM_ACCESS_PROT
492 struct file;
493 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
494 				     unsigned long size, pgprot_t vma_prot);
495 
496 #define pmd_none(pmd)		(!pmd_val(pmd))
497 
498 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
499 
500 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
501 				 PMD_TYPE_TABLE)
502 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
503 				 PMD_TYPE_SECT)
504 #define pmd_leaf(pmd)		pmd_sect(pmd)
505 
506 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
507 static inline bool pud_sect(pud_t pud) { return false; }
508 static inline bool pud_table(pud_t pud) { return true; }
509 #else
510 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
511 				 PUD_TYPE_SECT)
512 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
513 				 PUD_TYPE_TABLE)
514 #endif
515 
516 extern pgd_t init_pg_dir[PTRS_PER_PGD];
517 extern pgd_t init_pg_end[];
518 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
519 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
520 extern pgd_t idmap_pg_end[];
521 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
522 
523 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
524 
525 static inline bool in_swapper_pgdir(void *addr)
526 {
527 	return ((unsigned long)addr & PAGE_MASK) ==
528 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
529 }
530 
531 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
532 {
533 #ifdef __PAGETABLE_PMD_FOLDED
534 	if (in_swapper_pgdir(pmdp)) {
535 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
536 		return;
537 	}
538 #endif /* __PAGETABLE_PMD_FOLDED */
539 
540 	WRITE_ONCE(*pmdp, pmd);
541 
542 	if (pmd_valid(pmd)) {
543 		dsb(ishst);
544 		isb();
545 	}
546 }
547 
548 static inline void pmd_clear(pmd_t *pmdp)
549 {
550 	set_pmd(pmdp, __pmd(0));
551 }
552 
553 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
554 {
555 	return __pmd_to_phys(pmd);
556 }
557 
558 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
559 {
560 	return (unsigned long)__va(pmd_page_paddr(pmd));
561 }
562 
563 /* Find an entry in the third-level page table. */
564 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
565 
566 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
567 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
568 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
569 
570 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
571 
572 /* use ONLY for statically allocated translation tables */
573 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
574 
575 /*
576  * Conversion functions: convert a page and protection to a page entry,
577  * and a page entry and page directory to the page they refer to.
578  */
579 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
580 
581 #if CONFIG_PGTABLE_LEVELS > 2
582 
583 #define pmd_ERROR(e)	\
584 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
585 
586 #define pud_none(pud)		(!pud_val(pud))
587 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
588 #define pud_present(pud)	pte_present(pud_pte(pud))
589 #define pud_leaf(pud)		pud_sect(pud)
590 #define pud_valid(pud)		pte_valid(pud_pte(pud))
591 
592 static inline void set_pud(pud_t *pudp, pud_t pud)
593 {
594 #ifdef __PAGETABLE_PUD_FOLDED
595 	if (in_swapper_pgdir(pudp)) {
596 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
597 		return;
598 	}
599 #endif /* __PAGETABLE_PUD_FOLDED */
600 
601 	WRITE_ONCE(*pudp, pud);
602 
603 	if (pud_valid(pud)) {
604 		dsb(ishst);
605 		isb();
606 	}
607 }
608 
609 static inline void pud_clear(pud_t *pudp)
610 {
611 	set_pud(pudp, __pud(0));
612 }
613 
614 static inline phys_addr_t pud_page_paddr(pud_t pud)
615 {
616 	return __pud_to_phys(pud);
617 }
618 
619 static inline unsigned long pud_page_vaddr(pud_t pud)
620 {
621 	return (unsigned long)__va(pud_page_paddr(pud));
622 }
623 
624 /* Find an entry in the second-level page table. */
625 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
626 
627 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
628 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
629 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
630 
631 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
632 
633 /* use ONLY for statically allocated translation tables */
634 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
635 
636 #else
637 
638 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
639 
640 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
641 #define pmd_set_fixmap(addr)		NULL
642 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
643 #define pmd_clear_fixmap()
644 
645 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
646 
647 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
648 
649 #if CONFIG_PGTABLE_LEVELS > 3
650 
651 #define pud_ERROR(e)	\
652 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
653 
654 #define p4d_none(p4d)		(!p4d_val(p4d))
655 #define p4d_bad(p4d)		(!(p4d_val(p4d) & 2))
656 #define p4d_present(p4d)	(p4d_val(p4d))
657 
658 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
659 {
660 	if (in_swapper_pgdir(p4dp)) {
661 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
662 		return;
663 	}
664 
665 	WRITE_ONCE(*p4dp, p4d);
666 	dsb(ishst);
667 	isb();
668 }
669 
670 static inline void p4d_clear(p4d_t *p4dp)
671 {
672 	set_p4d(p4dp, __p4d(0));
673 }
674 
675 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
676 {
677 	return __p4d_to_phys(p4d);
678 }
679 
680 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
681 {
682 	return (unsigned long)__va(p4d_page_paddr(p4d));
683 }
684 
685 /* Find an entry in the frst-level page table. */
686 #define pud_offset_phys(dir, addr)	(p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
687 
688 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
689 #define pud_set_fixmap_offset(p4d, addr)	pud_set_fixmap(pud_offset_phys(p4d, addr))
690 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
691 
692 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
693 
694 /* use ONLY for statically allocated translation tables */
695 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
696 
697 #else
698 
699 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
700 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
701 
702 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
703 #define pud_set_fixmap(addr)		NULL
704 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
705 #define pud_clear_fixmap()
706 
707 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
708 
709 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
710 
711 #define pgd_ERROR(e)	\
712 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
713 
714 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
715 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
716 
717 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
718 {
719 	/*
720 	 * Normal and Normal-Tagged are two different memory types and indices
721 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
722 	 */
723 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
724 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
725 			      PTE_ATTRINDX_MASK;
726 	/* preserve the hardware dirty information */
727 	if (pte_hw_dirty(pte))
728 		pte = pte_mkdirty(pte);
729 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
730 	return pte;
731 }
732 
733 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
734 {
735 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
736 }
737 
738 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
739 extern int ptep_set_access_flags(struct vm_area_struct *vma,
740 				 unsigned long address, pte_t *ptep,
741 				 pte_t entry, int dirty);
742 
743 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
744 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
745 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
746 					unsigned long address, pmd_t *pmdp,
747 					pmd_t entry, int dirty)
748 {
749 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
750 }
751 
752 static inline int pud_devmap(pud_t pud)
753 {
754 	return 0;
755 }
756 
757 static inline int pgd_devmap(pgd_t pgd)
758 {
759 	return 0;
760 }
761 #endif
762 
763 /*
764  * Atomic pte/pmd modifications.
765  */
766 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
767 static inline int __ptep_test_and_clear_young(pte_t *ptep)
768 {
769 	pte_t old_pte, pte;
770 
771 	pte = READ_ONCE(*ptep);
772 	do {
773 		old_pte = pte;
774 		pte = pte_mkold(pte);
775 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
776 					       pte_val(old_pte), pte_val(pte));
777 	} while (pte_val(pte) != pte_val(old_pte));
778 
779 	return pte_young(pte);
780 }
781 
782 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
783 					    unsigned long address,
784 					    pte_t *ptep)
785 {
786 	return __ptep_test_and_clear_young(ptep);
787 }
788 
789 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
790 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
791 					 unsigned long address, pte_t *ptep)
792 {
793 	int young = ptep_test_and_clear_young(vma, address, ptep);
794 
795 	if (young) {
796 		/*
797 		 * We can elide the trailing DSB here since the worst that can
798 		 * happen is that a CPU continues to use the young entry in its
799 		 * TLB and we mistakenly reclaim the associated page. The
800 		 * window for such an event is bounded by the next
801 		 * context-switch, which provides a DSB to complete the TLB
802 		 * invalidation.
803 		 */
804 		flush_tlb_page_nosync(vma, address);
805 	}
806 
807 	return young;
808 }
809 
810 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
811 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
812 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
813 					    unsigned long address,
814 					    pmd_t *pmdp)
815 {
816 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
817 }
818 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
819 
820 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
821 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
822 				       unsigned long address, pte_t *ptep)
823 {
824 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
825 }
826 
827 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
828 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
829 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
830 					    unsigned long address, pmd_t *pmdp)
831 {
832 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
833 }
834 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
835 
836 /*
837  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
838  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
839  */
840 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
841 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
842 {
843 	pte_t old_pte, pte;
844 
845 	pte = READ_ONCE(*ptep);
846 	do {
847 		old_pte = pte;
848 		/*
849 		 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
850 		 * clear), set the PTE_DIRTY bit.
851 		 */
852 		if (pte_hw_dirty(pte))
853 			pte = pte_mkdirty(pte);
854 		pte = pte_wrprotect(pte);
855 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
856 					       pte_val(old_pte), pte_val(pte));
857 	} while (pte_val(pte) != pte_val(old_pte));
858 }
859 
860 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
861 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
862 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
863 				      unsigned long address, pmd_t *pmdp)
864 {
865 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
866 }
867 
868 #define pmdp_establish pmdp_establish
869 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
870 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
871 {
872 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
873 }
874 #endif
875 
876 /*
877  * Encode and decode a swap entry:
878  *	bits 0-1:	present (must be zero)
879  *	bits 2-7:	swap type
880  *	bits 8-57:	swap offset
881  *	bit  58:	PTE_PROT_NONE (must be zero)
882  */
883 #define __SWP_TYPE_SHIFT	2
884 #define __SWP_TYPE_BITS		6
885 #define __SWP_OFFSET_BITS	50
886 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
887 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
888 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
889 
890 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
891 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
892 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
893 
894 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
895 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
896 
897 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
898 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
899 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
900 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
901 
902 /*
903  * Ensure that there are not more swap files than can be encoded in the kernel
904  * PTEs.
905  */
906 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
907 
908 extern int kern_addr_valid(unsigned long addr);
909 
910 #ifdef CONFIG_ARM64_MTE
911 
912 #define __HAVE_ARCH_PREPARE_TO_SWAP
913 static inline int arch_prepare_to_swap(struct page *page)
914 {
915 	if (system_supports_mte())
916 		return mte_save_tags(page);
917 	return 0;
918 }
919 
920 #define __HAVE_ARCH_SWAP_INVALIDATE
921 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
922 {
923 	if (system_supports_mte())
924 		mte_invalidate_tags(type, offset);
925 }
926 
927 static inline void arch_swap_invalidate_area(int type)
928 {
929 	if (system_supports_mte())
930 		mte_invalidate_tags_area(type);
931 }
932 
933 #define __HAVE_ARCH_SWAP_RESTORE
934 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
935 {
936 	if (system_supports_mte() && mte_restore_tags(entry, page))
937 		set_bit(PG_mte_tagged, &page->flags);
938 }
939 
940 #endif /* CONFIG_ARM64_MTE */
941 
942 /*
943  * On AArch64, the cache coherency is handled via the set_pte_at() function.
944  */
945 static inline void update_mmu_cache(struct vm_area_struct *vma,
946 				    unsigned long addr, pte_t *ptep)
947 {
948 	/*
949 	 * We don't do anything here, so there's a very small chance of
950 	 * us retaking a user fault which we just fixed up. The alternative
951 	 * is doing a dsb(ishst), but that penalises the fastpath.
952 	 */
953 }
954 
955 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
956 
957 #ifdef CONFIG_ARM64_PA_BITS_52
958 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
959 #else
960 #define phys_to_ttbr(addr)	(addr)
961 #endif
962 
963 /*
964  * On arm64 without hardware Access Flag, copying from user will fail because
965  * the pte is old and cannot be marked young. So we always end up with zeroed
966  * page after fork() + CoW for pfn mappings. We don't always have a
967  * hardware-managed access flag on arm64.
968  */
969 static inline bool arch_faults_on_old_pte(void)
970 {
971 	WARN_ON(preemptible());
972 
973 	return !cpu_has_hw_af();
974 }
975 #define arch_faults_on_old_pte arch_faults_on_old_pte
976 
977 #endif /* !__ASSEMBLY__ */
978 
979 #endif /* __ASM_PGTABLE_H */
980