xref: /linux/arch/powerpc/include/asm/book3s/64/pgtable.h (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_POWERPC_BOOK3S_64_PGTABLE_H_
3 #define _ASM_POWERPC_BOOK3S_64_PGTABLE_H_
4 
5 #include <asm-generic/pgtable-nop4d.h>
6 
7 #ifndef __ASSEMBLY__
8 #include <linux/mmdebug.h>
9 #include <linux/bug.h>
10 #include <linux/sizes.h>
11 #endif
12 
13 /*
14  * Common bits between hash and Radix page table
15  */
16 
17 #define _PAGE_EXEC		0x00001 /* execute permission */
18 #define _PAGE_WRITE		0x00002 /* write access allowed */
19 #define _PAGE_READ		0x00004	/* read access allowed */
20 #define _PAGE_PRIVILEGED	0x00008 /* kernel access only */
21 #define _PAGE_SAO		0x00010 /* Strong access order */
22 #define _PAGE_NON_IDEMPOTENT	0x00020 /* non idempotent memory */
23 #define _PAGE_TOLERANT		0x00030 /* tolerant memory, cache inhibited */
24 #define _PAGE_DIRTY		0x00080 /* C: page changed */
25 #define _PAGE_ACCESSED		0x00100 /* R: page referenced */
26 /*
27  * Software bits
28  */
29 #define _RPAGE_SW0		0x2000000000000000UL
30 #define _RPAGE_SW1		0x00800
31 #define _RPAGE_SW2		0x00400
32 #define _RPAGE_SW3		0x00200
33 #define _RPAGE_RSV1		0x00040UL
34 
35 #define _RPAGE_PKEY_BIT4	0x1000000000000000UL
36 #define _RPAGE_PKEY_BIT3	0x0800000000000000UL
37 #define _RPAGE_PKEY_BIT2	0x0400000000000000UL
38 #define _RPAGE_PKEY_BIT1	0x0200000000000000UL
39 #define _RPAGE_PKEY_BIT0	0x0100000000000000UL
40 
41 #define _PAGE_PTE		0x4000000000000000UL	/* distinguishes PTEs from pointers */
42 #define _PAGE_PRESENT		0x8000000000000000UL	/* pte contains a translation */
43 /*
44  * We need to mark a pmd pte invalid while splitting. We can do that by clearing
45  * the _PAGE_PRESENT bit. But then that will be taken as a swap pte. In order to
46  * differentiate between two use a SW field when invalidating.
47  *
48  * We do that temporary invalidate for regular pte entry in ptep_set_access_flags
49  *
50  * This is used only when _PAGE_PRESENT is cleared.
51  */
52 #define _PAGE_INVALID		_RPAGE_SW0
53 
54 /*
55  * Top and bottom bits of RPN which can be used by hash
56  * translation mode, because we expect them to be zero
57  * otherwise.
58  */
59 #define _RPAGE_RPN0		0x01000
60 #define _RPAGE_RPN1		0x02000
61 #define _RPAGE_RPN43		0x0080000000000000UL
62 #define _RPAGE_RPN42		0x0040000000000000UL
63 #define _RPAGE_RPN41		0x0020000000000000UL
64 
65 /* Max physical address bit as per radix table */
66 #define _RPAGE_PA_MAX		56
67 
68 /*
69  * Max physical address bit we will use for now.
70  *
71  * This is mostly a hardware limitation and for now Power9 has
72  * a 51 bit limit.
73  *
74  * This is different from the number of physical bit required to address
75  * the last byte of memory. That is defined by MAX_PHYSMEM_BITS.
76  * MAX_PHYSMEM_BITS is a linux limitation imposed by the maximum
77  * number of sections we can support (SECTIONS_SHIFT).
78  *
79  * This is different from Radix page table limitation above and
80  * should always be less than that. The limit is done such that
81  * we can overload the bits between _RPAGE_PA_MAX and _PAGE_PA_MAX
82  * for hash linux page table specific bits.
83  *
84  * In order to be compatible with future hardware generations we keep
85  * some offsets and limit this for now to 53
86  */
87 #define _PAGE_PA_MAX		53
88 
89 #define _PAGE_SOFT_DIRTY	_RPAGE_SW3 /* software: software dirty tracking */
90 #define _PAGE_SPECIAL		_RPAGE_SW2 /* software: special page */
91 #define _PAGE_DEVMAP		_RPAGE_SW1 /* software: ZONE_DEVICE page */
92 
93 /*
94  * Drivers request for cache inhibited pte mapping using _PAGE_NO_CACHE
95  * Instead of fixing all of them, add an alternate define which
96  * maps CI pte mapping.
97  */
98 #define _PAGE_NO_CACHE		_PAGE_TOLERANT
99 /*
100  * We support _RPAGE_PA_MAX bit real address in pte. On the linux side
101  * we are limited by _PAGE_PA_MAX. Clear everything above _PAGE_PA_MAX
102  * and every thing below PAGE_SHIFT;
103  */
104 #define PTE_RPN_MASK	(((1UL << _PAGE_PA_MAX) - 1) & (PAGE_MASK))
105 #define PTE_RPN_SHIFT	PAGE_SHIFT
106 /*
107  * set of bits not changed in pmd_modify. Even though we have hash specific bits
108  * in here, on radix we expect them to be zero.
109  */
110 #define _HPAGE_CHG_MASK (PTE_RPN_MASK | _PAGE_HPTEFLAGS | _PAGE_DIRTY | \
111 			 _PAGE_ACCESSED | H_PAGE_THP_HUGE | _PAGE_PTE | \
112 			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP)
113 /*
114  * user access blocked by key
115  */
116 #define _PAGE_KERNEL_RW		(_PAGE_PRIVILEGED | _PAGE_RW | _PAGE_DIRTY)
117 #define _PAGE_KERNEL_RO		 (_PAGE_PRIVILEGED | _PAGE_READ)
118 #define _PAGE_KERNEL_ROX	 (_PAGE_PRIVILEGED | _PAGE_READ | _PAGE_EXEC)
119 #define _PAGE_KERNEL_RWX	(_PAGE_PRIVILEGED | _PAGE_DIRTY | _PAGE_RW | _PAGE_EXEC)
120 /*
121  * _PAGE_CHG_MASK masks of bits that are to be preserved across
122  * pgprot changes
123  */
124 #define _PAGE_CHG_MASK	(PTE_RPN_MASK | _PAGE_HPTEFLAGS | _PAGE_DIRTY | \
125 			 _PAGE_ACCESSED | _PAGE_SPECIAL | _PAGE_PTE |	\
126 			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP)
127 
128 /*
129  * We define 2 sets of base prot bits, one for basic pages (ie,
130  * cacheable kernel and user pages) and one for non cacheable
131  * pages. We always set _PAGE_COHERENT when SMP is enabled or
132  * the processor might need it for DMA coherency.
133  */
134 #define _PAGE_BASE_NC	(_PAGE_PRESENT | _PAGE_ACCESSED)
135 #define _PAGE_BASE	(_PAGE_BASE_NC)
136 
137 #include <asm/pgtable-masks.h>
138 
139 /* Permission masks used for kernel mappings */
140 #define PAGE_KERNEL	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RW)
141 #define PAGE_KERNEL_NC	__pgprot(_PAGE_BASE_NC | _PAGE_KERNEL_RW | _PAGE_TOLERANT)
142 #define PAGE_KERNEL_NCG	__pgprot(_PAGE_BASE_NC | _PAGE_KERNEL_RW | _PAGE_NON_IDEMPOTENT)
143 #define PAGE_KERNEL_X	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RWX)
144 #define PAGE_KERNEL_RO	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RO)
145 #define PAGE_KERNEL_ROX	__pgprot(_PAGE_BASE | _PAGE_KERNEL_ROX)
146 
147 #ifndef __ASSEMBLY__
148 /*
149  * page table defines
150  */
151 extern unsigned long __pte_index_size;
152 extern unsigned long __pmd_index_size;
153 extern unsigned long __pud_index_size;
154 extern unsigned long __pgd_index_size;
155 extern unsigned long __pud_cache_index;
156 #define PTE_INDEX_SIZE  __pte_index_size
157 #define PMD_INDEX_SIZE  __pmd_index_size
158 #define PUD_INDEX_SIZE  __pud_index_size
159 #define PGD_INDEX_SIZE  __pgd_index_size
160 /* pmd table use page table fragments */
161 #define PMD_CACHE_INDEX  0
162 #define PUD_CACHE_INDEX __pud_cache_index
163 /*
164  * Because of use of pte fragments and THP, size of page table
165  * are not always derived out of index size above.
166  */
167 extern unsigned long __pte_table_size;
168 extern unsigned long __pmd_table_size;
169 extern unsigned long __pud_table_size;
170 extern unsigned long __pgd_table_size;
171 #define PTE_TABLE_SIZE	__pte_table_size
172 #define PMD_TABLE_SIZE	__pmd_table_size
173 #define PUD_TABLE_SIZE	__pud_table_size
174 #define PGD_TABLE_SIZE	__pgd_table_size
175 
176 extern unsigned long __pmd_val_bits;
177 extern unsigned long __pud_val_bits;
178 extern unsigned long __pgd_val_bits;
179 #define PMD_VAL_BITS	__pmd_val_bits
180 #define PUD_VAL_BITS	__pud_val_bits
181 #define PGD_VAL_BITS	__pgd_val_bits
182 
183 extern unsigned long __pte_frag_nr;
184 #define PTE_FRAG_NR __pte_frag_nr
185 extern unsigned long __pte_frag_size_shift;
186 #define PTE_FRAG_SIZE_SHIFT __pte_frag_size_shift
187 #define PTE_FRAG_SIZE (1UL << PTE_FRAG_SIZE_SHIFT)
188 
189 extern unsigned long __pmd_frag_nr;
190 #define PMD_FRAG_NR __pmd_frag_nr
191 extern unsigned long __pmd_frag_size_shift;
192 #define PMD_FRAG_SIZE_SHIFT __pmd_frag_size_shift
193 #define PMD_FRAG_SIZE (1UL << PMD_FRAG_SIZE_SHIFT)
194 
195 #define PTRS_PER_PTE	(1 << PTE_INDEX_SIZE)
196 #define PTRS_PER_PMD	(1 << PMD_INDEX_SIZE)
197 #define PTRS_PER_PUD	(1 << PUD_INDEX_SIZE)
198 #define PTRS_PER_PGD	(1 << PGD_INDEX_SIZE)
199 
200 #define MAX_PTRS_PER_PTE ((H_PTRS_PER_PTE > R_PTRS_PER_PTE) ? H_PTRS_PER_PTE : R_PTRS_PER_PTE)
201 #define MAX_PTRS_PER_PMD ((H_PTRS_PER_PMD > R_PTRS_PER_PMD) ? H_PTRS_PER_PMD : R_PTRS_PER_PMD)
202 #define MAX_PTRS_PER_PUD ((H_PTRS_PER_PUD > R_PTRS_PER_PUD) ? H_PTRS_PER_PUD : R_PTRS_PER_PUD)
203 #define MAX_PTRS_PER_PGD	(1 << (H_PGD_INDEX_SIZE > RADIX_PGD_INDEX_SIZE ? \
204 				       H_PGD_INDEX_SIZE : RADIX_PGD_INDEX_SIZE))
205 
206 /* PMD_SHIFT determines what a second-level page table entry can map */
207 #define PMD_SHIFT	(PAGE_SHIFT + PTE_INDEX_SIZE)
208 #define PMD_SIZE	(1UL << PMD_SHIFT)
209 #define PMD_MASK	(~(PMD_SIZE-1))
210 
211 /* PUD_SHIFT determines what a third-level page table entry can map */
212 #define PUD_SHIFT	(PMD_SHIFT + PMD_INDEX_SIZE)
213 #define PUD_SIZE	(1UL << PUD_SHIFT)
214 #define PUD_MASK	(~(PUD_SIZE-1))
215 
216 /* PGDIR_SHIFT determines what a fourth-level page table entry can map */
217 #define PGDIR_SHIFT	(PUD_SHIFT + PUD_INDEX_SIZE)
218 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
219 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
220 
221 /* Bits to mask out from a PMD to get to the PTE page */
222 #define PMD_MASKED_BITS		0xc0000000000000ffUL
223 /* Bits to mask out from a PUD to get to the PMD page */
224 #define PUD_MASKED_BITS		0xc0000000000000ffUL
225 /* Bits to mask out from a PGD to get to the PUD page */
226 #define P4D_MASKED_BITS		0xc0000000000000ffUL
227 
228 /*
229  * Used as an indicator for rcu callback functions
230  */
231 enum pgtable_index {
232 	PTE_INDEX = 0,
233 	PMD_INDEX,
234 	PUD_INDEX,
235 	PGD_INDEX,
236 	/*
237 	 * Below are used with 4k page size and hugetlb
238 	 */
239 	HTLB_16M_INDEX,
240 	HTLB_16G_INDEX,
241 };
242 
243 extern unsigned long __vmalloc_start;
244 extern unsigned long __vmalloc_end;
245 #define VMALLOC_START	__vmalloc_start
246 #define VMALLOC_END	__vmalloc_end
247 
248 static inline unsigned int ioremap_max_order(void)
249 {
250 	if (radix_enabled())
251 		return PUD_SHIFT;
252 	return 7 + PAGE_SHIFT; /* default from linux/vmalloc.h */
253 }
254 #define IOREMAP_MAX_ORDER ioremap_max_order()
255 
256 extern unsigned long __kernel_virt_start;
257 extern unsigned long __kernel_io_start;
258 extern unsigned long __kernel_io_end;
259 #define KERN_VIRT_START __kernel_virt_start
260 #define KERN_IO_START  __kernel_io_start
261 #define KERN_IO_END __kernel_io_end
262 
263 extern struct page *vmemmap;
264 extern unsigned long pci_io_base;
265 #endif /* __ASSEMBLY__ */
266 
267 #include <asm/book3s/64/hash.h>
268 #include <asm/book3s/64/radix.h>
269 
270 #if H_MAX_PHYSMEM_BITS > R_MAX_PHYSMEM_BITS
271 #define  MAX_PHYSMEM_BITS	H_MAX_PHYSMEM_BITS
272 #else
273 #define  MAX_PHYSMEM_BITS	R_MAX_PHYSMEM_BITS
274 #endif
275 
276 
277 #ifdef CONFIG_PPC_64K_PAGES
278 #include <asm/book3s/64/pgtable-64k.h>
279 #else
280 #include <asm/book3s/64/pgtable-4k.h>
281 #endif
282 
283 #include <asm/barrier.h>
284 /*
285  * IO space itself carved into the PIO region (ISA and PHB IO space) and
286  * the ioremap space
287  *
288  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
289  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
290  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
291  */
292 #define FULL_IO_SIZE	0x80000000ul
293 #define  ISA_IO_BASE	(KERN_IO_START)
294 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
295 #define  PHB_IO_BASE	(ISA_IO_END)
296 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
297 #define IOREMAP_BASE	(PHB_IO_END)
298 #define IOREMAP_START	(ioremap_bot)
299 #define IOREMAP_END	(KERN_IO_END - FIXADDR_SIZE)
300 #define FIXADDR_SIZE	SZ_32M
301 #define FIXADDR_TOP	(IOREMAP_END + FIXADDR_SIZE)
302 
303 #ifndef __ASSEMBLY__
304 
305 /*
306  * This is the default implementation of various PTE accessors, it's
307  * used in all cases except Book3S with 64K pages where we have a
308  * concept of sub-pages
309  */
310 #ifndef __real_pte
311 
312 #define __real_pte(e, p, o)		((real_pte_t){(e)})
313 #define __rpte_to_pte(r)	((r).pte)
314 #define __rpte_to_hidx(r,index)	(pte_val(__rpte_to_pte(r)) >> H_PAGE_F_GIX_SHIFT)
315 
316 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift)       \
317 	do {							         \
318 		index = 0;					         \
319 		shift = mmu_psize_defs[psize].shift;		         \
320 
321 #define pte_iterate_hashed_end() } while(0)
322 
323 /*
324  * We expect this to be called only for user addresses or kernel virtual
325  * addresses other than the linear mapping.
326  */
327 #define pte_pagesize_index(mm, addr, pte)	MMU_PAGE_4K
328 
329 #endif /* __real_pte */
330 
331 static inline unsigned long pte_update(struct mm_struct *mm, unsigned long addr,
332 				       pte_t *ptep, unsigned long clr,
333 				       unsigned long set, int huge)
334 {
335 	if (radix_enabled())
336 		return radix__pte_update(mm, addr, ptep, clr, set, huge);
337 	return hash__pte_update(mm, addr, ptep, clr, set, huge);
338 }
339 /*
340  * For hash even if we have _PAGE_ACCESSED = 0, we do a pte_update.
341  * We currently remove entries from the hashtable regardless of whether
342  * the entry was young or dirty.
343  *
344  * We should be more intelligent about this but for the moment we override
345  * these functions and force a tlb flush unconditionally
346  * For radix: H_PAGE_HASHPTE should be zero. Hence we can use the same
347  * function for both hash and radix.
348  */
349 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
350 					      unsigned long addr, pte_t *ptep)
351 {
352 	unsigned long old;
353 
354 	if ((pte_raw(*ptep) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
355 		return 0;
356 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
357 	return (old & _PAGE_ACCESSED) != 0;
358 }
359 
360 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
361 #define ptep_test_and_clear_young(__vma, __addr, __ptep)	\
362 ({								\
363 	__ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
364 })
365 
366 /*
367  * On Book3S CPUs, clearing the accessed bit without a TLB flush
368  * doesn't cause data corruption. [ It could cause incorrect
369  * page aging and the (mistaken) reclaim of hot pages, but the
370  * chance of that should be relatively low. ]
371  *
372  * So as a performance optimization don't flush the TLB when
373  * clearing the accessed bit, it will eventually be flushed by
374  * a context switch or a VM operation anyway. [ In the rare
375  * event of it not getting flushed for a long time the delay
376  * shouldn't really matter because there's no real memory
377  * pressure for swapout to react to. ]
378  *
379  * Note: this optimisation also exists in pte_needs_flush() and
380  * huge_pmd_needs_flush().
381  */
382 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
383 #define ptep_clear_flush_young ptep_test_and_clear_young
384 
385 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
386 #define pmdp_clear_flush_young pmdp_test_and_clear_young
387 
388 static inline int pte_write(pte_t pte)
389 {
390 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_WRITE));
391 }
392 
393 static inline int pte_read(pte_t pte)
394 {
395 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_READ));
396 }
397 
398 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
399 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
400 				      pte_t *ptep)
401 {
402 	if (pte_write(*ptep))
403 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 0);
404 }
405 
406 #define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
407 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
408 					   unsigned long addr, pte_t *ptep)
409 {
410 	if (pte_write(*ptep))
411 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 1);
412 }
413 
414 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
415 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
416 				       unsigned long addr, pte_t *ptep)
417 {
418 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
419 	return __pte(old);
420 }
421 
422 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
423 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
424 					    unsigned long addr,
425 					    pte_t *ptep, int full)
426 {
427 	if (full && radix_enabled()) {
428 		/*
429 		 * We know that this is a full mm pte clear and
430 		 * hence can be sure there is no parallel set_pte.
431 		 */
432 		return radix__ptep_get_and_clear_full(mm, addr, ptep, full);
433 	}
434 	return ptep_get_and_clear(mm, addr, ptep);
435 }
436 
437 
438 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
439 			     pte_t * ptep)
440 {
441 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
442 }
443 
444 static inline int pte_dirty(pte_t pte)
445 {
446 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DIRTY));
447 }
448 
449 static inline int pte_young(pte_t pte)
450 {
451 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_ACCESSED));
452 }
453 
454 static inline int pte_special(pte_t pte)
455 {
456 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SPECIAL));
457 }
458 
459 static inline bool pte_exec(pte_t pte)
460 {
461 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_EXEC));
462 }
463 
464 
465 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
466 static inline bool pte_soft_dirty(pte_t pte)
467 {
468 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SOFT_DIRTY));
469 }
470 
471 static inline pte_t pte_mksoft_dirty(pte_t pte)
472 {
473 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SOFT_DIRTY));
474 }
475 
476 static inline pte_t pte_clear_soft_dirty(pte_t pte)
477 {
478 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SOFT_DIRTY));
479 }
480 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
481 
482 #ifdef CONFIG_NUMA_BALANCING
483 static inline int pte_protnone(pte_t pte)
484 {
485 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE | _PAGE_RWX)) ==
486 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
487 }
488 #endif /* CONFIG_NUMA_BALANCING */
489 
490 static inline bool pte_hw_valid(pte_t pte)
491 {
492 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE)) ==
493 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
494 }
495 
496 static inline int pte_present(pte_t pte)
497 {
498 	/*
499 	 * A pte is considerent present if _PAGE_PRESENT is set.
500 	 * We also need to consider the pte present which is marked
501 	 * invalid during ptep_set_access_flags. Hence we look for _PAGE_INVALID
502 	 * if we find _PAGE_PRESENT cleared.
503 	 */
504 
505 	if (pte_hw_valid(pte))
506 		return true;
507 	return (pte_raw(pte) & cpu_to_be64(_PAGE_INVALID | _PAGE_PTE)) ==
508 		cpu_to_be64(_PAGE_INVALID | _PAGE_PTE);
509 }
510 
511 #ifdef CONFIG_PPC_MEM_KEYS
512 extern bool arch_pte_access_permitted(u64 pte, bool write, bool execute);
513 #else
514 static inline bool arch_pte_access_permitted(u64 pte, bool write, bool execute)
515 {
516 	return true;
517 }
518 #endif /* CONFIG_PPC_MEM_KEYS */
519 
520 static inline bool pte_user(pte_t pte)
521 {
522 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_PRIVILEGED));
523 }
524 
525 #define pte_access_permitted pte_access_permitted
526 static inline bool pte_access_permitted(pte_t pte, bool write)
527 {
528 	/*
529 	 * _PAGE_READ is needed for any access and will be cleared for
530 	 * PROT_NONE. Execute-only mapping via PROT_EXEC also returns false.
531 	 */
532 	if (!pte_present(pte) || !pte_user(pte) || !pte_read(pte))
533 		return false;
534 
535 	if (write && !pte_write(pte))
536 		return false;
537 
538 	return arch_pte_access_permitted(pte_val(pte), write, 0);
539 }
540 
541 /*
542  * Conversion functions: convert a page and protection to a page entry,
543  * and a page entry and page directory to the page they refer to.
544  *
545  * Even if PTEs can be unsigned long long, a PFN is always an unsigned
546  * long for now.
547  */
548 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot)
549 {
550 	VM_BUG_ON(pfn >> (64 - PAGE_SHIFT));
551 	VM_BUG_ON((pfn << PAGE_SHIFT) & ~PTE_RPN_MASK);
552 
553 	return __pte(((pte_basic_t)pfn << PAGE_SHIFT) | pgprot_val(pgprot) | _PAGE_PTE);
554 }
555 
556 /* Generic modifiers for PTE bits */
557 static inline pte_t pte_wrprotect(pte_t pte)
558 {
559 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
560 }
561 
562 static inline pte_t pte_exprotect(pte_t pte)
563 {
564 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_EXEC));
565 }
566 
567 static inline pte_t pte_mkclean(pte_t pte)
568 {
569 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_DIRTY));
570 }
571 
572 static inline pte_t pte_mkold(pte_t pte)
573 {
574 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_ACCESSED));
575 }
576 
577 static inline pte_t pte_mkexec(pte_t pte)
578 {
579 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_EXEC));
580 }
581 
582 static inline pte_t pte_mkwrite_novma(pte_t pte)
583 {
584 	/*
585 	 * write implies read, hence set both
586 	 */
587 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_RW));
588 }
589 
590 static inline pte_t pte_mkdirty(pte_t pte)
591 {
592 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_DIRTY | _PAGE_SOFT_DIRTY));
593 }
594 
595 static inline pte_t pte_mkyoung(pte_t pte)
596 {
597 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_ACCESSED));
598 }
599 
600 static inline pte_t pte_mkspecial(pte_t pte)
601 {
602 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL));
603 }
604 
605 static inline pte_t pte_mkhuge(pte_t pte)
606 {
607 	return pte;
608 }
609 
610 static inline pte_t pte_mkdevmap(pte_t pte)
611 {
612 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL | _PAGE_DEVMAP));
613 }
614 
615 /*
616  * This is potentially called with a pmd as the argument, in which case it's not
617  * safe to check _PAGE_DEVMAP unless we also confirm that _PAGE_PTE is set.
618  * That's because the bit we use for _PAGE_DEVMAP is not reserved for software
619  * use in page directory entries (ie. non-ptes).
620  */
621 static inline int pte_devmap(pte_t pte)
622 {
623 	__be64 mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
624 
625 	return (pte_raw(pte) & mask) == mask;
626 }
627 
628 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
629 {
630 	/* FIXME!! check whether this need to be a conditional */
631 	return __pte_raw((pte_raw(pte) & cpu_to_be64(_PAGE_CHG_MASK)) |
632 			 cpu_to_be64(pgprot_val(newprot)));
633 }
634 
635 /* Encode and de-code a swap entry */
636 #define MAX_SWAPFILES_CHECK() do { \
637 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
638 	/*							\
639 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
640 	 * We filter HPTEFLAGS on set_pte.			\
641 	 */							\
642 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & SWP_TYPE_MASK); \
643 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_SOFT_DIRTY);	\
644 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_EXCLUSIVE);	\
645 	} while (0)
646 
647 #define SWP_TYPE_BITS 5
648 #define SWP_TYPE_MASK		((1UL << SWP_TYPE_BITS) - 1)
649 #define __swp_type(x)		((x).val & SWP_TYPE_MASK)
650 #define __swp_offset(x)		(((x).val & PTE_RPN_MASK) >> PAGE_SHIFT)
651 #define __swp_entry(type, offset)	((swp_entry_t) { \
652 				(type) | (((offset) << PAGE_SHIFT) & PTE_RPN_MASK)})
653 /*
654  * swp_entry_t must be independent of pte bits. We build a swp_entry_t from
655  * swap type and offset we get from swap and convert that to pte to find a
656  * matching pte in linux page table.
657  * Clear bits not found in swap entries here.
658  */
659 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val((pte)) & ~_PAGE_PTE })
660 #define __swp_entry_to_pte(x)	__pte((x).val | _PAGE_PTE)
661 #define __pmd_to_swp_entry(pmd)	(__pte_to_swp_entry(pmd_pte(pmd)))
662 #define __swp_entry_to_pmd(x)	(pte_pmd(__swp_entry_to_pte(x)))
663 
664 #ifdef CONFIG_MEM_SOFT_DIRTY
665 #define _PAGE_SWP_SOFT_DIRTY	_PAGE_SOFT_DIRTY
666 #else
667 #define _PAGE_SWP_SOFT_DIRTY	0UL
668 #endif /* CONFIG_MEM_SOFT_DIRTY */
669 
670 #define _PAGE_SWP_EXCLUSIVE	_PAGE_NON_IDEMPOTENT
671 
672 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
673 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
674 {
675 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
676 }
677 
678 static inline bool pte_swp_soft_dirty(pte_t pte)
679 {
680 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
681 }
682 
683 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
684 {
685 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_SOFT_DIRTY));
686 }
687 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
688 
689 static inline pte_t pte_swp_mkexclusive(pte_t pte)
690 {
691 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_EXCLUSIVE));
692 }
693 
694 static inline int pte_swp_exclusive(pte_t pte)
695 {
696 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_EXCLUSIVE));
697 }
698 
699 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
700 {
701 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_EXCLUSIVE));
702 }
703 
704 static inline bool check_pte_access(unsigned long access, unsigned long ptev)
705 {
706 	/*
707 	 * This check for _PAGE_RWX and _PAGE_PRESENT bits
708 	 */
709 	if (access & ~ptev)
710 		return false;
711 	/*
712 	 * This check for access to privilege space
713 	 */
714 	if ((access & _PAGE_PRIVILEGED) != (ptev & _PAGE_PRIVILEGED))
715 		return false;
716 
717 	return true;
718 }
719 /*
720  * Generic functions with hash/radix callbacks
721  */
722 
723 static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
724 					   pte_t *ptep, pte_t entry,
725 					   unsigned long address,
726 					   int psize)
727 {
728 	if (radix_enabled())
729 		return radix__ptep_set_access_flags(vma, ptep, entry,
730 						    address, psize);
731 	return hash__ptep_set_access_flags(ptep, entry);
732 }
733 
734 #define __HAVE_ARCH_PTE_SAME
735 static inline int pte_same(pte_t pte_a, pte_t pte_b)
736 {
737 	if (radix_enabled())
738 		return radix__pte_same(pte_a, pte_b);
739 	return hash__pte_same(pte_a, pte_b);
740 }
741 
742 static inline int pte_none(pte_t pte)
743 {
744 	if (radix_enabled())
745 		return radix__pte_none(pte);
746 	return hash__pte_none(pte);
747 }
748 
749 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
750 				pte_t *ptep, pte_t pte, int percpu)
751 {
752 
753 	VM_WARN_ON(!(pte_raw(pte) & cpu_to_be64(_PAGE_PTE)));
754 	/*
755 	 * Keep the _PAGE_PTE added till we are sure we handle _PAGE_PTE
756 	 * in all the callers.
757 	 */
758 	pte = __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PTE));
759 
760 	if (radix_enabled())
761 		return radix__set_pte_at(mm, addr, ptep, pte, percpu);
762 	return hash__set_pte_at(mm, addr, ptep, pte, percpu);
763 }
764 
765 #define _PAGE_CACHE_CTL	(_PAGE_SAO | _PAGE_NON_IDEMPOTENT | _PAGE_TOLERANT)
766 
767 #define pgprot_noncached pgprot_noncached
768 static inline pgprot_t pgprot_noncached(pgprot_t prot)
769 {
770 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
771 			_PAGE_NON_IDEMPOTENT);
772 }
773 
774 #define pgprot_noncached_wc pgprot_noncached_wc
775 static inline pgprot_t pgprot_noncached_wc(pgprot_t prot)
776 {
777 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
778 			_PAGE_TOLERANT);
779 }
780 
781 #define pgprot_cached pgprot_cached
782 static inline pgprot_t pgprot_cached(pgprot_t prot)
783 {
784 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL));
785 }
786 
787 #define pgprot_writecombine pgprot_writecombine
788 static inline pgprot_t pgprot_writecombine(pgprot_t prot)
789 {
790 	return pgprot_noncached_wc(prot);
791 }
792 /*
793  * check a pte mapping have cache inhibited property
794  */
795 static inline bool pte_ci(pte_t pte)
796 {
797 	__be64 pte_v = pte_raw(pte);
798 
799 	if (((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_TOLERANT)) ||
800 	    ((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_NON_IDEMPOTENT)))
801 		return true;
802 	return false;
803 }
804 
805 static inline void pmd_clear(pmd_t *pmdp)
806 {
807 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
808 		/*
809 		 * Don't use this if we can possibly have a hash page table
810 		 * entry mapping this.
811 		 */
812 		WARN_ON((pmd_val(*pmdp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
813 	}
814 	*pmdp = __pmd(0);
815 }
816 
817 static inline int pmd_none(pmd_t pmd)
818 {
819 	return !pmd_raw(pmd);
820 }
821 
822 static inline int pmd_present(pmd_t pmd)
823 {
824 	/*
825 	 * A pmd is considerent present if _PAGE_PRESENT is set.
826 	 * We also need to consider the pmd present which is marked
827 	 * invalid during a split. Hence we look for _PAGE_INVALID
828 	 * if we find _PAGE_PRESENT cleared.
829 	 */
830 	if (pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID))
831 		return true;
832 
833 	return false;
834 }
835 
836 static inline int pmd_is_serializing(pmd_t pmd)
837 {
838 	/*
839 	 * If the pmd is undergoing a split, the _PAGE_PRESENT bit is clear
840 	 * and _PAGE_INVALID is set (see pmd_present, pmdp_invalidate).
841 	 *
842 	 * This condition may also occur when flushing a pmd while flushing
843 	 * it (see ptep_modify_prot_start), so callers must ensure this
844 	 * case is fine as well.
845 	 */
846 	if ((pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID)) ==
847 						cpu_to_be64(_PAGE_INVALID))
848 		return true;
849 
850 	return false;
851 }
852 
853 static inline int pmd_bad(pmd_t pmd)
854 {
855 	if (radix_enabled())
856 		return radix__pmd_bad(pmd);
857 	return hash__pmd_bad(pmd);
858 }
859 
860 static inline void pud_clear(pud_t *pudp)
861 {
862 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
863 		/*
864 		 * Don't use this if we can possibly have a hash page table
865 		 * entry mapping this.
866 		 */
867 		WARN_ON((pud_val(*pudp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
868 	}
869 	*pudp = __pud(0);
870 }
871 
872 static inline int pud_none(pud_t pud)
873 {
874 	return !pud_raw(pud);
875 }
876 
877 static inline int pud_present(pud_t pud)
878 {
879 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PRESENT));
880 }
881 
882 extern struct page *pud_page(pud_t pud);
883 extern struct page *pmd_page(pmd_t pmd);
884 static inline pte_t pud_pte(pud_t pud)
885 {
886 	return __pte_raw(pud_raw(pud));
887 }
888 
889 static inline pud_t pte_pud(pte_t pte)
890 {
891 	return __pud_raw(pte_raw(pte));
892 }
893 
894 static inline pte_t *pudp_ptep(pud_t *pud)
895 {
896 	return (pte_t *)pud;
897 }
898 
899 #define pud_pfn(pud)		pte_pfn(pud_pte(pud))
900 #define pud_dirty(pud)		pte_dirty(pud_pte(pud))
901 #define pud_young(pud)		pte_young(pud_pte(pud))
902 #define pud_mkold(pud)		pte_pud(pte_mkold(pud_pte(pud)))
903 #define pud_wrprotect(pud)	pte_pud(pte_wrprotect(pud_pte(pud)))
904 #define pud_mkdirty(pud)	pte_pud(pte_mkdirty(pud_pte(pud)))
905 #define pud_mkclean(pud)	pte_pud(pte_mkclean(pud_pte(pud)))
906 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
907 #define pud_mkwrite(pud)	pte_pud(pte_mkwrite_novma(pud_pte(pud)))
908 #define pud_write(pud)		pte_write(pud_pte(pud))
909 
910 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
911 #define pud_soft_dirty(pmd)    pte_soft_dirty(pud_pte(pud))
912 #define pud_mksoft_dirty(pmd)  pte_pud(pte_mksoft_dirty(pud_pte(pud)))
913 #define pud_clear_soft_dirty(pmd) pte_pud(pte_clear_soft_dirty(pud_pte(pud)))
914 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
915 
916 static inline int pud_bad(pud_t pud)
917 {
918 	if (radix_enabled())
919 		return radix__pud_bad(pud);
920 	return hash__pud_bad(pud);
921 }
922 
923 #define pud_access_permitted pud_access_permitted
924 static inline bool pud_access_permitted(pud_t pud, bool write)
925 {
926 	return pte_access_permitted(pud_pte(pud), write);
927 }
928 
929 #define __p4d_raw(x)	((p4d_t) { __pgd_raw(x) })
930 static inline __be64 p4d_raw(p4d_t x)
931 {
932 	return pgd_raw(x.pgd);
933 }
934 
935 #define p4d_write(p4d)		pte_write(p4d_pte(p4d))
936 
937 static inline void p4d_clear(p4d_t *p4dp)
938 {
939 	*p4dp = __p4d(0);
940 }
941 
942 static inline int p4d_none(p4d_t p4d)
943 {
944 	return !p4d_raw(p4d);
945 }
946 
947 static inline int p4d_present(p4d_t p4d)
948 {
949 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PRESENT));
950 }
951 
952 static inline pte_t p4d_pte(p4d_t p4d)
953 {
954 	return __pte_raw(p4d_raw(p4d));
955 }
956 
957 static inline p4d_t pte_p4d(pte_t pte)
958 {
959 	return __p4d_raw(pte_raw(pte));
960 }
961 
962 static inline int p4d_bad(p4d_t p4d)
963 {
964 	if (radix_enabled())
965 		return radix__p4d_bad(p4d);
966 	return hash__p4d_bad(p4d);
967 }
968 
969 #define p4d_access_permitted p4d_access_permitted
970 static inline bool p4d_access_permitted(p4d_t p4d, bool write)
971 {
972 	return pte_access_permitted(p4d_pte(p4d), write);
973 }
974 
975 extern struct page *p4d_page(p4d_t p4d);
976 
977 /* Pointers in the page table tree are physical addresses */
978 #define __pgtable_ptr_val(ptr)	__pa(ptr)
979 
980 static inline pud_t *p4d_pgtable(p4d_t p4d)
981 {
982 	return (pud_t *)__va(p4d_val(p4d) & ~P4D_MASKED_BITS);
983 }
984 
985 static inline pmd_t *pud_pgtable(pud_t pud)
986 {
987 	return (pmd_t *)__va(pud_val(pud) & ~PUD_MASKED_BITS);
988 }
989 
990 #define pmd_ERROR(e) \
991 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
992 #define pud_ERROR(e) \
993 	pr_err("%s:%d: bad pud %08lx.\n", __FILE__, __LINE__, pud_val(e))
994 #define pgd_ERROR(e) \
995 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
996 
997 static inline int map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot)
998 {
999 	if (radix_enabled()) {
1000 #if defined(CONFIG_PPC_RADIX_MMU) && defined(DEBUG_VM)
1001 		unsigned long page_size = 1 << mmu_psize_defs[mmu_io_psize].shift;
1002 		WARN((page_size != PAGE_SIZE), "I/O page size != PAGE_SIZE");
1003 #endif
1004 		return radix__map_kernel_page(ea, pa, prot, PAGE_SIZE);
1005 	}
1006 	return hash__map_kernel_page(ea, pa, prot);
1007 }
1008 
1009 void unmap_kernel_page(unsigned long va);
1010 
1011 static inline int __meminit vmemmap_create_mapping(unsigned long start,
1012 						   unsigned long page_size,
1013 						   unsigned long phys)
1014 {
1015 	if (radix_enabled())
1016 		return radix__vmemmap_create_mapping(start, page_size, phys);
1017 	return hash__vmemmap_create_mapping(start, page_size, phys);
1018 }
1019 
1020 #ifdef CONFIG_MEMORY_HOTPLUG
1021 static inline void vmemmap_remove_mapping(unsigned long start,
1022 					  unsigned long page_size)
1023 {
1024 	if (radix_enabled())
1025 		return radix__vmemmap_remove_mapping(start, page_size);
1026 	return hash__vmemmap_remove_mapping(start, page_size);
1027 }
1028 #endif
1029 
1030 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE)
1031 static inline void __kernel_map_pages(struct page *page, int numpages, int enable)
1032 {
1033 	if (radix_enabled())
1034 		radix__kernel_map_pages(page, numpages, enable);
1035 	else
1036 		hash__kernel_map_pages(page, numpages, enable);
1037 }
1038 #endif
1039 
1040 static inline pte_t pmd_pte(pmd_t pmd)
1041 {
1042 	return __pte_raw(pmd_raw(pmd));
1043 }
1044 
1045 static inline pmd_t pte_pmd(pte_t pte)
1046 {
1047 	return __pmd_raw(pte_raw(pte));
1048 }
1049 
1050 static inline pte_t *pmdp_ptep(pmd_t *pmd)
1051 {
1052 	return (pte_t *)pmd;
1053 }
1054 #define pmd_pfn(pmd)		pte_pfn(pmd_pte(pmd))
1055 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
1056 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
1057 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
1058 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
1059 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
1060 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
1061 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
1062 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
1063 
1064 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1065 #define pmd_soft_dirty(pmd)    pte_soft_dirty(pmd_pte(pmd))
1066 #define pmd_mksoft_dirty(pmd)  pte_pmd(pte_mksoft_dirty(pmd_pte(pmd)))
1067 #define pmd_clear_soft_dirty(pmd) pte_pmd(pte_clear_soft_dirty(pmd_pte(pmd)))
1068 
1069 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1070 #define pmd_swp_mksoft_dirty(pmd)	pte_pmd(pte_swp_mksoft_dirty(pmd_pte(pmd)))
1071 #define pmd_swp_soft_dirty(pmd)		pte_swp_soft_dirty(pmd_pte(pmd))
1072 #define pmd_swp_clear_soft_dirty(pmd)	pte_pmd(pte_swp_clear_soft_dirty(pmd_pte(pmd)))
1073 #endif
1074 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
1075 
1076 #ifdef CONFIG_NUMA_BALANCING
1077 static inline int pmd_protnone(pmd_t pmd)
1078 {
1079 	return pte_protnone(pmd_pte(pmd));
1080 }
1081 #endif /* CONFIG_NUMA_BALANCING */
1082 
1083 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
1084 
1085 #define pmd_access_permitted pmd_access_permitted
1086 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1087 {
1088 	/*
1089 	 * pmdp_invalidate sets this combination (which is not caught by
1090 	 * !pte_present() check in pte_access_permitted), to prevent
1091 	 * lock-free lookups, as part of the serialize_against_pte_lookup()
1092 	 * synchronisation.
1093 	 *
1094 	 * This also catches the case where the PTE's hardware PRESENT bit is
1095 	 * cleared while TLB is flushed, which is suboptimal but should not
1096 	 * be frequent.
1097 	 */
1098 	if (pmd_is_serializing(pmd))
1099 		return false;
1100 
1101 	return pte_access_permitted(pmd_pte(pmd), write);
1102 }
1103 
1104 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1105 extern pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot);
1106 extern pud_t pfn_pud(unsigned long pfn, pgprot_t pgprot);
1107 extern pmd_t mk_pmd(struct page *page, pgprot_t pgprot);
1108 extern pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot);
1109 extern void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1110 		       pmd_t *pmdp, pmd_t pmd);
1111 extern void set_pud_at(struct mm_struct *mm, unsigned long addr,
1112 		       pud_t *pudp, pud_t pud);
1113 
1114 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1115 					unsigned long addr, pmd_t *pmd)
1116 {
1117 }
1118 
1119 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1120 					unsigned long addr, pud_t *pud)
1121 {
1122 }
1123 
1124 extern int hash__has_transparent_hugepage(void);
1125 static inline int has_transparent_hugepage(void)
1126 {
1127 	if (radix_enabled())
1128 		return radix__has_transparent_hugepage();
1129 	return hash__has_transparent_hugepage();
1130 }
1131 #define has_transparent_hugepage has_transparent_hugepage
1132 
1133 static inline int has_transparent_pud_hugepage(void)
1134 {
1135 	if (radix_enabled())
1136 		return radix__has_transparent_pud_hugepage();
1137 	return 0;
1138 }
1139 #define has_transparent_pud_hugepage has_transparent_pud_hugepage
1140 
1141 static inline unsigned long
1142 pmd_hugepage_update(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp,
1143 		    unsigned long clr, unsigned long set)
1144 {
1145 	if (radix_enabled())
1146 		return radix__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1147 	return hash__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1148 }
1149 
1150 static inline unsigned long
1151 pud_hugepage_update(struct mm_struct *mm, unsigned long addr, pud_t *pudp,
1152 		    unsigned long clr, unsigned long set)
1153 {
1154 	if (radix_enabled())
1155 		return radix__pud_hugepage_update(mm, addr, pudp, clr, set);
1156 	BUG();
1157 	return pud_val(*pudp);
1158 }
1159 
1160 /*
1161  * returns true for pmd migration entries, THP, devmap, hugetlb
1162  * But compile time dependent on THP config
1163  */
1164 static inline int pmd_large(pmd_t pmd)
1165 {
1166 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1167 }
1168 
1169 static inline int pud_large(pud_t pud)
1170 {
1171 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PTE));
1172 }
1173 
1174 /*
1175  * For radix we should always find H_PAGE_HASHPTE zero. Hence
1176  * the below will work for radix too
1177  */
1178 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
1179 					      unsigned long addr, pmd_t *pmdp)
1180 {
1181 	unsigned long old;
1182 
1183 	if ((pmd_raw(*pmdp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1184 		return 0;
1185 	old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
1186 	return ((old & _PAGE_ACCESSED) != 0);
1187 }
1188 
1189 static inline int __pudp_test_and_clear_young(struct mm_struct *mm,
1190 					      unsigned long addr, pud_t *pudp)
1191 {
1192 	unsigned long old;
1193 
1194 	if ((pud_raw(*pudp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1195 		return 0;
1196 	old = pud_hugepage_update(mm, addr, pudp, _PAGE_ACCESSED, 0);
1197 	return ((old & _PAGE_ACCESSED) != 0);
1198 }
1199 
1200 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1201 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1202 				      pmd_t *pmdp)
1203 {
1204 	if (pmd_write(*pmdp))
1205 		pmd_hugepage_update(mm, addr, pmdp, _PAGE_WRITE, 0);
1206 }
1207 
1208 #define __HAVE_ARCH_PUDP_SET_WRPROTECT
1209 static inline void pudp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1210 				      pud_t *pudp)
1211 {
1212 	if (pud_write(*pudp))
1213 		pud_hugepage_update(mm, addr, pudp, _PAGE_WRITE, 0);
1214 }
1215 
1216 /*
1217  * Only returns true for a THP. False for pmd migration entry.
1218  * We also need to return true when we come across a pte that
1219  * in between a thp split. While splitting THP, we mark the pmd
1220  * invalid (pmdp_invalidate()) before we set it with pte page
1221  * address. A pmd_trans_huge() check against a pmd entry during that time
1222  * should return true.
1223  * We should not call this on a hugetlb entry. We should check for HugeTLB
1224  * entry using vma->vm_flags
1225  * The page table walk rule is explained in Documentation/mm/transhuge.rst
1226  */
1227 static inline int pmd_trans_huge(pmd_t pmd)
1228 {
1229 	if (!pmd_present(pmd))
1230 		return false;
1231 
1232 	if (radix_enabled())
1233 		return radix__pmd_trans_huge(pmd);
1234 	return hash__pmd_trans_huge(pmd);
1235 }
1236 
1237 static inline int pud_trans_huge(pud_t pud)
1238 {
1239 	if (!pud_present(pud))
1240 		return false;
1241 
1242 	if (radix_enabled())
1243 		return radix__pud_trans_huge(pud);
1244 	return 0;
1245 }
1246 
1247 
1248 #define __HAVE_ARCH_PMD_SAME
1249 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1250 {
1251 	if (radix_enabled())
1252 		return radix__pmd_same(pmd_a, pmd_b);
1253 	return hash__pmd_same(pmd_a, pmd_b);
1254 }
1255 
1256 #define pud_same pud_same
1257 static inline int pud_same(pud_t pud_a, pud_t pud_b)
1258 {
1259 	if (radix_enabled())
1260 		return radix__pud_same(pud_a, pud_b);
1261 	return hash__pud_same(pud_a, pud_b);
1262 }
1263 
1264 
1265 static inline pmd_t __pmd_mkhuge(pmd_t pmd)
1266 {
1267 	if (radix_enabled())
1268 		return radix__pmd_mkhuge(pmd);
1269 	return hash__pmd_mkhuge(pmd);
1270 }
1271 
1272 static inline pud_t __pud_mkhuge(pud_t pud)
1273 {
1274 	if (radix_enabled())
1275 		return radix__pud_mkhuge(pud);
1276 	BUG();
1277 	return pud;
1278 }
1279 
1280 /*
1281  * pfn_pmd return a pmd_t that can be used as pmd pte entry.
1282  */
1283 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1284 {
1285 #ifdef CONFIG_DEBUG_VM
1286 	if (radix_enabled())
1287 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE)) == 0);
1288 	else
1289 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE)) !=
1290 			cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE));
1291 #endif
1292 	return pmd;
1293 }
1294 
1295 static inline pud_t pud_mkhuge(pud_t pud)
1296 {
1297 #ifdef CONFIG_DEBUG_VM
1298 	if (radix_enabled())
1299 		WARN_ON((pud_raw(pud) & cpu_to_be64(_PAGE_PTE)) == 0);
1300 	else
1301 		WARN_ON(1);
1302 #endif
1303 	return pud;
1304 }
1305 
1306 
1307 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1308 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1309 				 unsigned long address, pmd_t *pmdp,
1310 				 pmd_t entry, int dirty);
1311 #define __HAVE_ARCH_PUDP_SET_ACCESS_FLAGS
1312 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1313 				 unsigned long address, pud_t *pudp,
1314 				 pud_t entry, int dirty);
1315 
1316 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1317 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1318 				     unsigned long address, pmd_t *pmdp);
1319 #define __HAVE_ARCH_PUDP_TEST_AND_CLEAR_YOUNG
1320 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1321 				     unsigned long address, pud_t *pudp);
1322 
1323 
1324 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1325 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1326 					    unsigned long addr, pmd_t *pmdp)
1327 {
1328 	if (radix_enabled())
1329 		return radix__pmdp_huge_get_and_clear(mm, addr, pmdp);
1330 	return hash__pmdp_huge_get_and_clear(mm, addr, pmdp);
1331 }
1332 
1333 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1334 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1335 					    unsigned long addr, pud_t *pudp)
1336 {
1337 	if (radix_enabled())
1338 		return radix__pudp_huge_get_and_clear(mm, addr, pudp);
1339 	BUG();
1340 	return *pudp;
1341 }
1342 
1343 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1344 					unsigned long address, pmd_t *pmdp)
1345 {
1346 	if (radix_enabled())
1347 		return radix__pmdp_collapse_flush(vma, address, pmdp);
1348 	return hash__pmdp_collapse_flush(vma, address, pmdp);
1349 }
1350 #define pmdp_collapse_flush pmdp_collapse_flush
1351 
1352 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1353 pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1354 				   unsigned long addr,
1355 				   pmd_t *pmdp, int full);
1356 
1357 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
1358 pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
1359 				   unsigned long addr,
1360 				   pud_t *pudp, int full);
1361 
1362 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1363 static inline void pgtable_trans_huge_deposit(struct mm_struct *mm,
1364 					      pmd_t *pmdp, pgtable_t pgtable)
1365 {
1366 	if (radix_enabled())
1367 		return radix__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1368 	return hash__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1369 }
1370 
1371 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1372 static inline pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm,
1373 						    pmd_t *pmdp)
1374 {
1375 	if (radix_enabled())
1376 		return radix__pgtable_trans_huge_withdraw(mm, pmdp);
1377 	return hash__pgtable_trans_huge_withdraw(mm, pmdp);
1378 }
1379 
1380 #define __HAVE_ARCH_PMDP_INVALIDATE
1381 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
1382 			     pmd_t *pmdp);
1383 
1384 #define pmd_move_must_withdraw pmd_move_must_withdraw
1385 struct spinlock;
1386 extern int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
1387 				  struct spinlock *old_pmd_ptl,
1388 				  struct vm_area_struct *vma);
1389 /*
1390  * Hash translation mode use the deposited table to store hash pte
1391  * slot information.
1392  */
1393 #define arch_needs_pgtable_deposit arch_needs_pgtable_deposit
1394 static inline bool arch_needs_pgtable_deposit(void)
1395 {
1396 	if (radix_enabled())
1397 		return false;
1398 	return true;
1399 }
1400 extern void serialize_against_pte_lookup(struct mm_struct *mm);
1401 
1402 
1403 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
1404 {
1405 	if (radix_enabled())
1406 		return radix__pmd_mkdevmap(pmd);
1407 	return hash__pmd_mkdevmap(pmd);
1408 }
1409 
1410 static inline pud_t pud_mkdevmap(pud_t pud)
1411 {
1412 	if (radix_enabled())
1413 		return radix__pud_mkdevmap(pud);
1414 	BUG();
1415 	return pud;
1416 }
1417 
1418 static inline int pmd_devmap(pmd_t pmd)
1419 {
1420 	return pte_devmap(pmd_pte(pmd));
1421 }
1422 
1423 static inline int pud_devmap(pud_t pud)
1424 {
1425 	return pte_devmap(pud_pte(pud));
1426 }
1427 
1428 static inline int pgd_devmap(pgd_t pgd)
1429 {
1430 	return 0;
1431 }
1432 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1433 
1434 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1435 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1436 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1437 			     pte_t *, pte_t, pte_t);
1438 
1439 /*
1440  * Returns true for a R -> RW upgrade of pte
1441  */
1442 static inline bool is_pte_rw_upgrade(unsigned long old_val, unsigned long new_val)
1443 {
1444 	if (!(old_val & _PAGE_READ))
1445 		return false;
1446 
1447 	if ((!(old_val & _PAGE_WRITE)) && (new_val & _PAGE_WRITE))
1448 		return true;
1449 
1450 	return false;
1451 }
1452 
1453 /*
1454  * Like pmd_huge() and pmd_large(), but works regardless of config options
1455  */
1456 #define pmd_is_leaf pmd_is_leaf
1457 #define pmd_leaf pmd_is_leaf
1458 static inline bool pmd_is_leaf(pmd_t pmd)
1459 {
1460 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1461 }
1462 
1463 #define pud_is_leaf pud_is_leaf
1464 #define pud_leaf pud_is_leaf
1465 static inline bool pud_is_leaf(pud_t pud)
1466 {
1467 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PTE));
1468 }
1469 
1470 #endif /* __ASSEMBLY__ */
1471 #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */
1472