xref: /openbmc/linux/arch/arm64/include/asm/pgtable.h (revision 5f66f73b)
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 /*
117  * Execute-only user mappings do not have the PTE_USER bit set. All valid
118  * kernel mappings have the PTE_UXN bit set.
119  */
120 #define pte_valid_not_user(pte) \
121 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
122 /*
123  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
124  * so that we don't erroneously return false for pages that have been
125  * remapped as PROT_NONE but are yet to be flushed from the TLB.
126  * Note that we can't make any assumptions based on the state of the access
127  * flag, since ptep_clear_flush_young() elides a DSB when invalidating the
128  * TLB.
129  */
130 #define pte_accessible(mm, pte)	\
131 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
132 
133 /*
134  * p??_access_permitted() is true for valid user mappings (PTE_USER
135  * bit set, subject to the write permission check). For execute-only
136  * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
137  * not set) must return false. PROT_NONE mappings do not have the
138  * PTE_VALID bit set.
139  */
140 #define pte_access_permitted(pte, write) \
141 	(((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
142 #define pmd_access_permitted(pmd, write) \
143 	(pte_access_permitted(pmd_pte(pmd), (write)))
144 #define pud_access_permitted(pud, write) \
145 	(pte_access_permitted(pud_pte(pud), (write)))
146 
147 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
148 {
149 	pte_val(pte) &= ~pgprot_val(prot);
150 	return pte;
151 }
152 
153 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
154 {
155 	pte_val(pte) |= pgprot_val(prot);
156 	return pte;
157 }
158 
159 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
160 {
161 	pmd_val(pmd) &= ~pgprot_val(prot);
162 	return pmd;
163 }
164 
165 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
166 {
167 	pmd_val(pmd) |= pgprot_val(prot);
168 	return pmd;
169 }
170 
171 static inline pte_t pte_mkwrite(pte_t pte)
172 {
173 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
174 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
175 	return pte;
176 }
177 
178 static inline pte_t pte_mkclean(pte_t pte)
179 {
180 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
181 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
182 
183 	return pte;
184 }
185 
186 static inline pte_t pte_mkdirty(pte_t pte)
187 {
188 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
189 
190 	if (pte_write(pte))
191 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
192 
193 	return pte;
194 }
195 
196 static inline pte_t pte_wrprotect(pte_t pte)
197 {
198 	/*
199 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
200 	 * clear), set the PTE_DIRTY bit.
201 	 */
202 	if (pte_hw_dirty(pte))
203 		pte = pte_mkdirty(pte);
204 
205 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
206 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
207 	return pte;
208 }
209 
210 static inline pte_t pte_mkold(pte_t pte)
211 {
212 	return clear_pte_bit(pte, __pgprot(PTE_AF));
213 }
214 
215 static inline pte_t pte_mkyoung(pte_t pte)
216 {
217 	return set_pte_bit(pte, __pgprot(PTE_AF));
218 }
219 
220 static inline pte_t pte_mkspecial(pte_t pte)
221 {
222 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
223 }
224 
225 static inline pte_t pte_mkcont(pte_t pte)
226 {
227 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
228 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
229 }
230 
231 static inline pte_t pte_mknoncont(pte_t pte)
232 {
233 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
234 }
235 
236 static inline pte_t pte_mkpresent(pte_t pte)
237 {
238 	return set_pte_bit(pte, __pgprot(PTE_VALID));
239 }
240 
241 static inline pmd_t pmd_mkcont(pmd_t pmd)
242 {
243 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
244 }
245 
246 static inline pte_t pte_mkdevmap(pte_t pte)
247 {
248 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
249 }
250 
251 static inline void set_pte(pte_t *ptep, pte_t pte)
252 {
253 	WRITE_ONCE(*ptep, pte);
254 
255 	/*
256 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
257 	 * or update_mmu_cache() have the necessary barriers.
258 	 */
259 	if (pte_valid_not_user(pte)) {
260 		dsb(ishst);
261 		isb();
262 	}
263 }
264 
265 extern void __sync_icache_dcache(pte_t pteval);
266 
267 /*
268  * PTE bits configuration in the presence of hardware Dirty Bit Management
269  * (PTE_WRITE == PTE_DBM):
270  *
271  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
272  *   0      0      |   1           0          0
273  *   0      1      |   1           1          0
274  *   1      0      |   1           0          1
275  *   1      1      |   0           1          x
276  *
277  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
278  * the page fault mechanism. Checking the dirty status of a pte becomes:
279  *
280  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
281  */
282 
283 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep,
284 					   pte_t pte)
285 {
286 	pte_t old_pte;
287 
288 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
289 		return;
290 
291 	old_pte = READ_ONCE(*ptep);
292 
293 	if (!pte_valid(old_pte) || !pte_valid(pte))
294 		return;
295 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
296 		return;
297 
298 	/*
299 	 * Check for potential race with hardware updates of the pte
300 	 * (ptep_set_access_flags safely changes valid ptes without going
301 	 * through an invalid entry).
302 	 */
303 	VM_WARN_ONCE(!pte_young(pte),
304 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
305 		     __func__, pte_val(old_pte), pte_val(pte));
306 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
307 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
308 		     __func__, pte_val(old_pte), pte_val(pte));
309 }
310 
311 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
312 			      pte_t *ptep, pte_t pte)
313 {
314 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
315 		__sync_icache_dcache(pte);
316 
317 	if (system_supports_mte() &&
318 	    pte_present(pte) && pte_tagged(pte) && !pte_special(pte))
319 		mte_sync_tags(ptep, pte);
320 
321 	__check_racy_pte_update(mm, ptep, pte);
322 
323 	set_pte(ptep, pte);
324 }
325 
326 /*
327  * Huge pte definitions.
328  */
329 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
330 
331 /*
332  * Hugetlb definitions.
333  */
334 #define HUGE_MAX_HSTATE		4
335 #define HPAGE_SHIFT		PMD_SHIFT
336 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
337 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
338 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
339 
340 static inline pte_t pgd_pte(pgd_t pgd)
341 {
342 	return __pte(pgd_val(pgd));
343 }
344 
345 static inline pte_t p4d_pte(p4d_t p4d)
346 {
347 	return __pte(p4d_val(p4d));
348 }
349 
350 static inline pte_t pud_pte(pud_t pud)
351 {
352 	return __pte(pud_val(pud));
353 }
354 
355 static inline pud_t pte_pud(pte_t pte)
356 {
357 	return __pud(pte_val(pte));
358 }
359 
360 static inline pmd_t pud_pmd(pud_t pud)
361 {
362 	return __pmd(pud_val(pud));
363 }
364 
365 static inline pte_t pmd_pte(pmd_t pmd)
366 {
367 	return __pte(pmd_val(pmd));
368 }
369 
370 static inline pmd_t pte_pmd(pte_t pte)
371 {
372 	return __pmd(pte_val(pte));
373 }
374 
375 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
376 {
377 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
378 }
379 
380 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
381 {
382 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
383 }
384 
385 #ifdef CONFIG_NUMA_BALANCING
386 /*
387  * See the comment in include/linux/pgtable.h
388  */
389 static inline int pte_protnone(pte_t pte)
390 {
391 	return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE;
392 }
393 
394 static inline int pmd_protnone(pmd_t pmd)
395 {
396 	return pte_protnone(pmd_pte(pmd));
397 }
398 #endif
399 
400 #define pmd_present_invalid(pmd)     (!!(pmd_val(pmd) & PMD_PRESENT_INVALID))
401 
402 static inline int pmd_present(pmd_t pmd)
403 {
404 	return pte_present(pmd_pte(pmd)) || pmd_present_invalid(pmd);
405 }
406 
407 /*
408  * THP definitions.
409  */
410 
411 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
412 static inline int pmd_trans_huge(pmd_t pmd)
413 {
414 	return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
415 }
416 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
417 
418 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
419 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
420 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
421 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
422 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
423 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
424 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
425 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
426 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
427 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
428 
429 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
430 {
431 	pmd = set_pmd_bit(pmd, __pgprot(PMD_PRESENT_INVALID));
432 	pmd = clear_pmd_bit(pmd, __pgprot(PMD_SECT_VALID));
433 
434 	return pmd;
435 }
436 
437 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
438 
439 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
440 
441 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
442 
443 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
444 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
445 #endif
446 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
447 {
448 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
449 }
450 
451 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
452 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
453 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
454 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
455 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
456 
457 #define pud_young(pud)		pte_young(pud_pte(pud))
458 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
459 #define pud_write(pud)		pte_write(pud_pte(pud))
460 
461 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
462 
463 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
464 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
465 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
466 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
467 
468 #define set_pmd_at(mm, addr, pmdp, pmd)	set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd))
469 #define set_pud_at(mm, addr, pudp, pud)	set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud))
470 
471 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
472 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
473 
474 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
475 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
476 
477 #define __pgprot_modify(prot,mask,bits) \
478 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
479 
480 #define pgprot_nx(prot) \
481 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
482 
483 /*
484  * Mark the prot value as uncacheable and unbufferable.
485  */
486 #define pgprot_noncached(prot) \
487 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
488 #define pgprot_writecombine(prot) \
489 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
490 #define pgprot_device(prot) \
491 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
492 #define pgprot_tagged(prot) \
493 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
494 #define pgprot_mhp	pgprot_tagged
495 /*
496  * DMA allocations for non-coherent devices use what the Arm architecture calls
497  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
498  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
499  * is intended for MMIO and thus forbids speculation, preserves access size,
500  * requires strict alignment and can also force write responses to come from the
501  * endpoint.
502  */
503 #define pgprot_dmacoherent(prot) \
504 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
505 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
506 
507 #define __HAVE_PHYS_MEM_ACCESS_PROT
508 struct file;
509 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
510 				     unsigned long size, pgprot_t vma_prot);
511 
512 #define pmd_none(pmd)		(!pmd_val(pmd))
513 
514 #define pmd_bad(pmd)		(!(pmd_val(pmd) & PMD_TABLE_BIT))
515 
516 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
517 				 PMD_TYPE_TABLE)
518 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
519 				 PMD_TYPE_SECT)
520 #define pmd_leaf(pmd)		pmd_sect(pmd)
521 
522 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
523 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
524 
525 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
526 static inline bool pud_sect(pud_t pud) { return false; }
527 static inline bool pud_table(pud_t pud) { return true; }
528 #else
529 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
530 				 PUD_TYPE_SECT)
531 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
532 				 PUD_TYPE_TABLE)
533 #endif
534 
535 extern pgd_t init_pg_dir[PTRS_PER_PGD];
536 extern pgd_t init_pg_end[];
537 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
538 extern pgd_t idmap_pg_dir[PTRS_PER_PGD];
539 extern pgd_t idmap_pg_end[];
540 extern pgd_t tramp_pg_dir[PTRS_PER_PGD];
541 extern pgd_t reserved_pg_dir[PTRS_PER_PGD];
542 
543 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
544 
545 static inline bool in_swapper_pgdir(void *addr)
546 {
547 	return ((unsigned long)addr & PAGE_MASK) ==
548 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
549 }
550 
551 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
552 {
553 #ifdef __PAGETABLE_PMD_FOLDED
554 	if (in_swapper_pgdir(pmdp)) {
555 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
556 		return;
557 	}
558 #endif /* __PAGETABLE_PMD_FOLDED */
559 
560 	WRITE_ONCE(*pmdp, pmd);
561 
562 	if (pmd_valid(pmd)) {
563 		dsb(ishst);
564 		isb();
565 	}
566 }
567 
568 static inline void pmd_clear(pmd_t *pmdp)
569 {
570 	set_pmd(pmdp, __pmd(0));
571 }
572 
573 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
574 {
575 	return __pmd_to_phys(pmd);
576 }
577 
578 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
579 {
580 	return (unsigned long)__va(pmd_page_paddr(pmd));
581 }
582 
583 /* Find an entry in the third-level page table. */
584 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
585 
586 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
587 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
588 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
589 
590 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
591 
592 /* use ONLY for statically allocated translation tables */
593 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
594 
595 /*
596  * Conversion functions: convert a page and protection to a page entry,
597  * and a page entry and page directory to the page they refer to.
598  */
599 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
600 
601 #if CONFIG_PGTABLE_LEVELS > 2
602 
603 #define pmd_ERROR(e)	\
604 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
605 
606 #define pud_none(pud)		(!pud_val(pud))
607 #define pud_bad(pud)		(!(pud_val(pud) & PUD_TABLE_BIT))
608 #define pud_present(pud)	pte_present(pud_pte(pud))
609 #define pud_leaf(pud)		pud_sect(pud)
610 #define pud_valid(pud)		pte_valid(pud_pte(pud))
611 
612 static inline void set_pud(pud_t *pudp, pud_t pud)
613 {
614 #ifdef __PAGETABLE_PUD_FOLDED
615 	if (in_swapper_pgdir(pudp)) {
616 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
617 		return;
618 	}
619 #endif /* __PAGETABLE_PUD_FOLDED */
620 
621 	WRITE_ONCE(*pudp, pud);
622 
623 	if (pud_valid(pud)) {
624 		dsb(ishst);
625 		isb();
626 	}
627 }
628 
629 static inline void pud_clear(pud_t *pudp)
630 {
631 	set_pud(pudp, __pud(0));
632 }
633 
634 static inline phys_addr_t pud_page_paddr(pud_t pud)
635 {
636 	return __pud_to_phys(pud);
637 }
638 
639 static inline unsigned long pud_page_vaddr(pud_t pud)
640 {
641 	return (unsigned long)__va(pud_page_paddr(pud));
642 }
643 
644 /* Find an entry in the second-level page table. */
645 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
646 
647 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
648 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
649 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
650 
651 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
652 
653 /* use ONLY for statically allocated translation tables */
654 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
655 
656 #else
657 
658 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
659 
660 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
661 #define pmd_set_fixmap(addr)		NULL
662 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
663 #define pmd_clear_fixmap()
664 
665 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
666 
667 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
668 
669 #if CONFIG_PGTABLE_LEVELS > 3
670 
671 #define pud_ERROR(e)	\
672 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
673 
674 #define p4d_none(p4d)		(!p4d_val(p4d))
675 #define p4d_bad(p4d)		(!(p4d_val(p4d) & 2))
676 #define p4d_present(p4d)	(p4d_val(p4d))
677 
678 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
679 {
680 	if (in_swapper_pgdir(p4dp)) {
681 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
682 		return;
683 	}
684 
685 	WRITE_ONCE(*p4dp, p4d);
686 	dsb(ishst);
687 	isb();
688 }
689 
690 static inline void p4d_clear(p4d_t *p4dp)
691 {
692 	set_p4d(p4dp, __p4d(0));
693 }
694 
695 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
696 {
697 	return __p4d_to_phys(p4d);
698 }
699 
700 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
701 {
702 	return (unsigned long)__va(p4d_page_paddr(p4d));
703 }
704 
705 /* Find an entry in the frst-level page table. */
706 #define pud_offset_phys(dir, addr)	(p4d_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t))
707 
708 #define pud_set_fixmap(addr)		((pud_t *)set_fixmap_offset(FIX_PUD, addr))
709 #define pud_set_fixmap_offset(p4d, addr)	pud_set_fixmap(pud_offset_phys(p4d, addr))
710 #define pud_clear_fixmap()		clear_fixmap(FIX_PUD)
711 
712 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
713 
714 /* use ONLY for statically allocated translation tables */
715 #define pud_offset_kimg(dir,addr)	((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr))))
716 
717 #else
718 
719 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
720 #define pgd_page_paddr(pgd)	({ BUILD_BUG(); 0;})
721 
722 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
723 #define pud_set_fixmap(addr)		NULL
724 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
725 #define pud_clear_fixmap()
726 
727 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
728 
729 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
730 
731 #define pgd_ERROR(e)	\
732 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
733 
734 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
735 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
736 
737 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
738 {
739 	/*
740 	 * Normal and Normal-Tagged are two different memory types and indices
741 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
742 	 */
743 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
744 			      PTE_PROT_NONE | PTE_VALID | PTE_WRITE | PTE_GP |
745 			      PTE_ATTRINDX_MASK;
746 	/* preserve the hardware dirty information */
747 	if (pte_hw_dirty(pte))
748 		pte = pte_mkdirty(pte);
749 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
750 	return pte;
751 }
752 
753 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
754 {
755 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
756 }
757 
758 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
759 extern int ptep_set_access_flags(struct vm_area_struct *vma,
760 				 unsigned long address, pte_t *ptep,
761 				 pte_t entry, int dirty);
762 
763 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
764 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
765 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
766 					unsigned long address, pmd_t *pmdp,
767 					pmd_t entry, int dirty)
768 {
769 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
770 }
771 
772 static inline int pud_devmap(pud_t pud)
773 {
774 	return 0;
775 }
776 
777 static inline int pgd_devmap(pgd_t pgd)
778 {
779 	return 0;
780 }
781 #endif
782 
783 /*
784  * Atomic pte/pmd modifications.
785  */
786 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
787 static inline int __ptep_test_and_clear_young(pte_t *ptep)
788 {
789 	pte_t old_pte, pte;
790 
791 	pte = READ_ONCE(*ptep);
792 	do {
793 		old_pte = pte;
794 		pte = pte_mkold(pte);
795 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
796 					       pte_val(old_pte), pte_val(pte));
797 	} while (pte_val(pte) != pte_val(old_pte));
798 
799 	return pte_young(pte);
800 }
801 
802 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
803 					    unsigned long address,
804 					    pte_t *ptep)
805 {
806 	return __ptep_test_and_clear_young(ptep);
807 }
808 
809 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
810 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
811 					 unsigned long address, pte_t *ptep)
812 {
813 	int young = ptep_test_and_clear_young(vma, address, ptep);
814 
815 	if (young) {
816 		/*
817 		 * We can elide the trailing DSB here since the worst that can
818 		 * happen is that a CPU continues to use the young entry in its
819 		 * TLB and we mistakenly reclaim the associated page. The
820 		 * window for such an event is bounded by the next
821 		 * context-switch, which provides a DSB to complete the TLB
822 		 * invalidation.
823 		 */
824 		flush_tlb_page_nosync(vma, address);
825 	}
826 
827 	return young;
828 }
829 
830 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
831 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
832 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
833 					    unsigned long address,
834 					    pmd_t *pmdp)
835 {
836 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
837 }
838 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
839 
840 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
841 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
842 				       unsigned long address, pte_t *ptep)
843 {
844 	return __pte(xchg_relaxed(&pte_val(*ptep), 0));
845 }
846 
847 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
848 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
849 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
850 					    unsigned long address, pmd_t *pmdp)
851 {
852 	return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp));
853 }
854 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
855 
856 /*
857  * ptep_set_wrprotect - mark read-only while trasferring potential hardware
858  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
859  */
860 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
861 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
862 {
863 	pte_t old_pte, pte;
864 
865 	pte = READ_ONCE(*ptep);
866 	do {
867 		old_pte = pte;
868 		pte = pte_wrprotect(pte);
869 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
870 					       pte_val(old_pte), pte_val(pte));
871 	} while (pte_val(pte) != pte_val(old_pte));
872 }
873 
874 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
875 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
876 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
877 				      unsigned long address, pmd_t *pmdp)
878 {
879 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
880 }
881 
882 #define pmdp_establish pmdp_establish
883 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
884 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
885 {
886 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
887 }
888 #endif
889 
890 /*
891  * Encode and decode a swap entry:
892  *	bits 0-1:	present (must be zero)
893  *	bits 2-7:	swap type
894  *	bits 8-57:	swap offset
895  *	bit  58:	PTE_PROT_NONE (must be zero)
896  */
897 #define __SWP_TYPE_SHIFT	2
898 #define __SWP_TYPE_BITS		6
899 #define __SWP_OFFSET_BITS	50
900 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
901 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
902 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
903 
904 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
905 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
906 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
907 
908 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
909 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
910 
911 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
912 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
913 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
914 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
915 
916 /*
917  * Ensure that there are not more swap files than can be encoded in the kernel
918  * PTEs.
919  */
920 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
921 
922 extern int kern_addr_valid(unsigned long addr);
923 
924 #ifdef CONFIG_ARM64_MTE
925 
926 #define __HAVE_ARCH_PREPARE_TO_SWAP
927 static inline int arch_prepare_to_swap(struct page *page)
928 {
929 	if (system_supports_mte())
930 		return mte_save_tags(page);
931 	return 0;
932 }
933 
934 #define __HAVE_ARCH_SWAP_INVALIDATE
935 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
936 {
937 	if (system_supports_mte())
938 		mte_invalidate_tags(type, offset);
939 }
940 
941 static inline void arch_swap_invalidate_area(int type)
942 {
943 	if (system_supports_mte())
944 		mte_invalidate_tags_area(type);
945 }
946 
947 #define __HAVE_ARCH_SWAP_RESTORE
948 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
949 {
950 	if (system_supports_mte() && mte_restore_tags(entry, page))
951 		set_bit(PG_mte_tagged, &page->flags);
952 }
953 
954 #endif /* CONFIG_ARM64_MTE */
955 
956 /*
957  * On AArch64, the cache coherency is handled via the set_pte_at() function.
958  */
959 static inline void update_mmu_cache(struct vm_area_struct *vma,
960 				    unsigned long addr, pte_t *ptep)
961 {
962 	/*
963 	 * We don't do anything here, so there's a very small chance of
964 	 * us retaking a user fault which we just fixed up. The alternative
965 	 * is doing a dsb(ishst), but that penalises the fastpath.
966 	 */
967 }
968 
969 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
970 
971 #ifdef CONFIG_ARM64_PA_BITS_52
972 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
973 #else
974 #define phys_to_ttbr(addr)	(addr)
975 #endif
976 
977 /*
978  * On arm64 without hardware Access Flag, copying from user will fail because
979  * the pte is old and cannot be marked young. So we always end up with zeroed
980  * page after fork() + CoW for pfn mappings. We don't always have a
981  * hardware-managed access flag on arm64.
982  */
983 static inline bool arch_faults_on_old_pte(void)
984 {
985 	WARN_ON(preemptible());
986 
987 	return !cpu_has_hw_af();
988 }
989 #define arch_faults_on_old_pte		arch_faults_on_old_pte
990 
991 /*
992  * Experimentally, it's cheap to set the access flag in hardware and we
993  * benefit from prefaulting mappings as 'old' to start with.
994  */
995 static inline bool arch_wants_old_prefaulted_pte(void)
996 {
997 	return !arch_faults_on_old_pte();
998 }
999 #define arch_wants_old_prefaulted_pte	arch_wants_old_prefaulted_pte
1000 
1001 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
1002 {
1003 	if (cpus_have_const_cap(ARM64_HAS_EPAN))
1004 		return prot;
1005 
1006 	if (pgprot_val(prot) != pgprot_val(PAGE_EXECONLY))
1007 		return prot;
1008 
1009 	return PAGE_READONLY_EXEC;
1010 }
1011 
1012 
1013 #endif /* !__ASSEMBLY__ */
1014 
1015 #endif /* __ASM_PGTABLE_H */
1016