xref: /linux/arch/powerpc/include/asm/book3s/64/pgtable.h (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 
266 #define pmd_leaf pmd_leaf
267 static inline bool pmd_leaf(pmd_t pmd)
268 {
269 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
270 }
271 
272 #define pud_leaf pud_leaf
273 static inline bool pud_leaf(pud_t pud)
274 {
275 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PTE));
276 }
277 #endif /* __ASSEMBLY__ */
278 
279 #include <asm/book3s/64/hash.h>
280 #include <asm/book3s/64/radix.h>
281 
282 #if H_MAX_PHYSMEM_BITS > R_MAX_PHYSMEM_BITS
283 #define  MAX_PHYSMEM_BITS	H_MAX_PHYSMEM_BITS
284 #else
285 #define  MAX_PHYSMEM_BITS	R_MAX_PHYSMEM_BITS
286 #endif
287 
288 
289 #ifdef CONFIG_PPC_64K_PAGES
290 #include <asm/book3s/64/pgtable-64k.h>
291 #else
292 #include <asm/book3s/64/pgtable-4k.h>
293 #endif
294 
295 #include <asm/barrier.h>
296 /*
297  * IO space itself carved into the PIO region (ISA and PHB IO space) and
298  * the ioremap space
299  *
300  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
301  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
302  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
303  */
304 #define FULL_IO_SIZE	0x80000000ul
305 #define  ISA_IO_BASE	(KERN_IO_START)
306 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
307 #define  PHB_IO_BASE	(ISA_IO_END)
308 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
309 #define IOREMAP_BASE	(PHB_IO_END)
310 #define IOREMAP_START	(ioremap_bot)
311 #define IOREMAP_END	(KERN_IO_END - FIXADDR_SIZE)
312 #define FIXADDR_SIZE	SZ_32M
313 #define FIXADDR_TOP	(IOREMAP_END + FIXADDR_SIZE)
314 
315 #ifndef __ASSEMBLY__
316 
317 /*
318  * This is the default implementation of various PTE accessors, it's
319  * used in all cases except Book3S with 64K pages where we have a
320  * concept of sub-pages
321  */
322 #ifndef __real_pte
323 
324 #define __real_pte(e, p, o)		((real_pte_t){(e)})
325 #define __rpte_to_pte(r)	((r).pte)
326 #define __rpte_to_hidx(r,index)	(pte_val(__rpte_to_pte(r)) >> H_PAGE_F_GIX_SHIFT)
327 
328 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift)       \
329 	do {							         \
330 		index = 0;					         \
331 		shift = mmu_psize_defs[psize].shift;		         \
332 
333 #define pte_iterate_hashed_end() } while(0)
334 
335 /*
336  * We expect this to be called only for user addresses or kernel virtual
337  * addresses other than the linear mapping.
338  */
339 #define pte_pagesize_index(mm, addr, pte)	MMU_PAGE_4K
340 
341 #endif /* __real_pte */
342 
343 static inline unsigned long pte_update(struct mm_struct *mm, unsigned long addr,
344 				       pte_t *ptep, unsigned long clr,
345 				       unsigned long set, int huge)
346 {
347 	if (radix_enabled())
348 		return radix__pte_update(mm, addr, ptep, clr, set, huge);
349 	return hash__pte_update(mm, addr, ptep, clr, set, huge);
350 }
351 /*
352  * For hash even if we have _PAGE_ACCESSED = 0, we do a pte_update.
353  * We currently remove entries from the hashtable regardless of whether
354  * the entry was young or dirty.
355  *
356  * We should be more intelligent about this but for the moment we override
357  * these functions and force a tlb flush unconditionally
358  * For radix: H_PAGE_HASHPTE should be zero. Hence we can use the same
359  * function for both hash and radix.
360  */
361 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
362 					      unsigned long addr, pte_t *ptep)
363 {
364 	unsigned long old;
365 
366 	if ((pte_raw(*ptep) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
367 		return 0;
368 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
369 	return (old & _PAGE_ACCESSED) != 0;
370 }
371 
372 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
373 #define ptep_test_and_clear_young(__vma, __addr, __ptep)	\
374 ({								\
375 	__ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
376 })
377 
378 /*
379  * On Book3S CPUs, clearing the accessed bit without a TLB flush
380  * doesn't cause data corruption. [ It could cause incorrect
381  * page aging and the (mistaken) reclaim of hot pages, but the
382  * chance of that should be relatively low. ]
383  *
384  * So as a performance optimization don't flush the TLB when
385  * clearing the accessed bit, it will eventually be flushed by
386  * a context switch or a VM operation anyway. [ In the rare
387  * event of it not getting flushed for a long time the delay
388  * shouldn't really matter because there's no real memory
389  * pressure for swapout to react to. ]
390  *
391  * Note: this optimisation also exists in pte_needs_flush() and
392  * huge_pmd_needs_flush().
393  */
394 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
395 #define ptep_clear_flush_young ptep_test_and_clear_young
396 
397 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
398 #define pmdp_clear_flush_young pmdp_test_and_clear_young
399 
400 static inline int pte_write(pte_t pte)
401 {
402 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_WRITE));
403 }
404 
405 static inline int pte_read(pte_t pte)
406 {
407 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_READ));
408 }
409 
410 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
411 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
412 				      pte_t *ptep)
413 {
414 	if (pte_write(*ptep))
415 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 0);
416 }
417 
418 #define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
419 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
420 					   unsigned long addr, pte_t *ptep)
421 {
422 	if (pte_write(*ptep))
423 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 1);
424 }
425 
426 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
427 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
428 				       unsigned long addr, pte_t *ptep)
429 {
430 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
431 	return __pte(old);
432 }
433 
434 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
435 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
436 					    unsigned long addr,
437 					    pte_t *ptep, int full)
438 {
439 	if (full && radix_enabled()) {
440 		/*
441 		 * We know that this is a full mm pte clear and
442 		 * hence can be sure there is no parallel set_pte.
443 		 */
444 		return radix__ptep_get_and_clear_full(mm, addr, ptep, full);
445 	}
446 	return ptep_get_and_clear(mm, addr, ptep);
447 }
448 
449 
450 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
451 			     pte_t * ptep)
452 {
453 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
454 }
455 
456 static inline int pte_dirty(pte_t pte)
457 {
458 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DIRTY));
459 }
460 
461 static inline int pte_young(pte_t pte)
462 {
463 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_ACCESSED));
464 }
465 
466 static inline int pte_special(pte_t pte)
467 {
468 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SPECIAL));
469 }
470 
471 static inline bool pte_exec(pte_t pte)
472 {
473 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_EXEC));
474 }
475 
476 
477 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
478 static inline bool pte_soft_dirty(pte_t pte)
479 {
480 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SOFT_DIRTY));
481 }
482 
483 static inline pte_t pte_mksoft_dirty(pte_t pte)
484 {
485 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SOFT_DIRTY));
486 }
487 
488 static inline pte_t pte_clear_soft_dirty(pte_t pte)
489 {
490 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SOFT_DIRTY));
491 }
492 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
493 
494 #ifdef CONFIG_NUMA_BALANCING
495 static inline int pte_protnone(pte_t pte)
496 {
497 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE | _PAGE_RWX)) ==
498 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
499 }
500 #endif /* CONFIG_NUMA_BALANCING */
501 
502 static inline bool pte_hw_valid(pte_t pte)
503 {
504 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE)) ==
505 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
506 }
507 
508 static inline int pte_present(pte_t pte)
509 {
510 	/*
511 	 * A pte is considerent present if _PAGE_PRESENT is set.
512 	 * We also need to consider the pte present which is marked
513 	 * invalid during ptep_set_access_flags. Hence we look for _PAGE_INVALID
514 	 * if we find _PAGE_PRESENT cleared.
515 	 */
516 
517 	if (pte_hw_valid(pte))
518 		return true;
519 	return (pte_raw(pte) & cpu_to_be64(_PAGE_INVALID | _PAGE_PTE)) ==
520 		cpu_to_be64(_PAGE_INVALID | _PAGE_PTE);
521 }
522 
523 #ifdef CONFIG_PPC_MEM_KEYS
524 extern bool arch_pte_access_permitted(u64 pte, bool write, bool execute);
525 #else
526 static inline bool arch_pte_access_permitted(u64 pte, bool write, bool execute)
527 {
528 	return true;
529 }
530 #endif /* CONFIG_PPC_MEM_KEYS */
531 
532 static inline bool pte_user(pte_t pte)
533 {
534 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_PRIVILEGED));
535 }
536 
537 #define pte_access_permitted pte_access_permitted
538 static inline bool pte_access_permitted(pte_t pte, bool write)
539 {
540 	/*
541 	 * _PAGE_READ is needed for any access and will be cleared for
542 	 * PROT_NONE. Execute-only mapping via PROT_EXEC also returns false.
543 	 */
544 	if (!pte_present(pte) || !pte_user(pte) || !pte_read(pte))
545 		return false;
546 
547 	if (write && !pte_write(pte))
548 		return false;
549 
550 	return arch_pte_access_permitted(pte_val(pte), write, 0);
551 }
552 
553 /*
554  * Conversion functions: convert a page and protection to a page entry,
555  * and a page entry and page directory to the page they refer to.
556  *
557  * Even if PTEs can be unsigned long long, a PFN is always an unsigned
558  * long for now.
559  */
560 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot)
561 {
562 	VM_BUG_ON(pfn >> (64 - PAGE_SHIFT));
563 	VM_BUG_ON((pfn << PAGE_SHIFT) & ~PTE_RPN_MASK);
564 
565 	return __pte(((pte_basic_t)pfn << PAGE_SHIFT) | pgprot_val(pgprot) | _PAGE_PTE);
566 }
567 
568 /* Generic modifiers for PTE bits */
569 static inline pte_t pte_wrprotect(pte_t pte)
570 {
571 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
572 }
573 
574 static inline pte_t pte_exprotect(pte_t pte)
575 {
576 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_EXEC));
577 }
578 
579 static inline pte_t pte_mkclean(pte_t pte)
580 {
581 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_DIRTY));
582 }
583 
584 static inline pte_t pte_mkold(pte_t pte)
585 {
586 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_ACCESSED));
587 }
588 
589 static inline pte_t pte_mkexec(pte_t pte)
590 {
591 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_EXEC));
592 }
593 
594 static inline pte_t pte_mkwrite_novma(pte_t pte)
595 {
596 	/*
597 	 * write implies read, hence set both
598 	 */
599 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_RW));
600 }
601 
602 static inline pte_t pte_mkdirty(pte_t pte)
603 {
604 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_DIRTY | _PAGE_SOFT_DIRTY));
605 }
606 
607 static inline pte_t pte_mkyoung(pte_t pte)
608 {
609 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_ACCESSED));
610 }
611 
612 static inline pte_t pte_mkspecial(pte_t pte)
613 {
614 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL));
615 }
616 
617 static inline pte_t pte_mkhuge(pte_t pte)
618 {
619 	return pte;
620 }
621 
622 static inline pte_t pte_mkdevmap(pte_t pte)
623 {
624 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL | _PAGE_DEVMAP));
625 }
626 
627 /*
628  * This is potentially called with a pmd as the argument, in which case it's not
629  * safe to check _PAGE_DEVMAP unless we also confirm that _PAGE_PTE is set.
630  * That's because the bit we use for _PAGE_DEVMAP is not reserved for software
631  * use in page directory entries (ie. non-ptes).
632  */
633 static inline int pte_devmap(pte_t pte)
634 {
635 	__be64 mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
636 
637 	return (pte_raw(pte) & mask) == mask;
638 }
639 
640 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
641 {
642 	/* FIXME!! check whether this need to be a conditional */
643 	return __pte_raw((pte_raw(pte) & cpu_to_be64(_PAGE_CHG_MASK)) |
644 			 cpu_to_be64(pgprot_val(newprot)));
645 }
646 
647 /* Encode and de-code a swap entry */
648 #define MAX_SWAPFILES_CHECK() do { \
649 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
650 	/*							\
651 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
652 	 * We filter HPTEFLAGS on set_pte.			\
653 	 */							\
654 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & SWP_TYPE_MASK); \
655 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_SOFT_DIRTY);	\
656 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_EXCLUSIVE);	\
657 	} while (0)
658 
659 #define SWP_TYPE_BITS 5
660 #define SWP_TYPE_MASK		((1UL << SWP_TYPE_BITS) - 1)
661 #define __swp_type(x)		((x).val & SWP_TYPE_MASK)
662 #define __swp_offset(x)		(((x).val & PTE_RPN_MASK) >> PAGE_SHIFT)
663 #define __swp_entry(type, offset)	((swp_entry_t) { \
664 				(type) | (((offset) << PAGE_SHIFT) & PTE_RPN_MASK)})
665 /*
666  * swp_entry_t must be independent of pte bits. We build a swp_entry_t from
667  * swap type and offset we get from swap and convert that to pte to find a
668  * matching pte in linux page table.
669  * Clear bits not found in swap entries here.
670  */
671 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val((pte)) & ~_PAGE_PTE })
672 #define __swp_entry_to_pte(x)	__pte((x).val | _PAGE_PTE)
673 #define __pmd_to_swp_entry(pmd)	(__pte_to_swp_entry(pmd_pte(pmd)))
674 #define __swp_entry_to_pmd(x)	(pte_pmd(__swp_entry_to_pte(x)))
675 
676 #ifdef CONFIG_MEM_SOFT_DIRTY
677 #define _PAGE_SWP_SOFT_DIRTY	_PAGE_SOFT_DIRTY
678 #else
679 #define _PAGE_SWP_SOFT_DIRTY	0UL
680 #endif /* CONFIG_MEM_SOFT_DIRTY */
681 
682 #define _PAGE_SWP_EXCLUSIVE	_PAGE_NON_IDEMPOTENT
683 
684 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
685 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
686 {
687 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
688 }
689 
690 static inline bool pte_swp_soft_dirty(pte_t pte)
691 {
692 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
693 }
694 
695 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
696 {
697 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_SOFT_DIRTY));
698 }
699 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
700 
701 static inline pte_t pte_swp_mkexclusive(pte_t pte)
702 {
703 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_EXCLUSIVE));
704 }
705 
706 static inline int pte_swp_exclusive(pte_t pte)
707 {
708 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_EXCLUSIVE));
709 }
710 
711 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
712 {
713 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_EXCLUSIVE));
714 }
715 
716 static inline bool check_pte_access(unsigned long access, unsigned long ptev)
717 {
718 	/*
719 	 * This check for _PAGE_RWX and _PAGE_PRESENT bits
720 	 */
721 	if (access & ~ptev)
722 		return false;
723 	/*
724 	 * This check for access to privilege space
725 	 */
726 	if ((access & _PAGE_PRIVILEGED) != (ptev & _PAGE_PRIVILEGED))
727 		return false;
728 
729 	return true;
730 }
731 /*
732  * Generic functions with hash/radix callbacks
733  */
734 
735 static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
736 					   pte_t *ptep, pte_t entry,
737 					   unsigned long address,
738 					   int psize)
739 {
740 	if (radix_enabled())
741 		return radix__ptep_set_access_flags(vma, ptep, entry,
742 						    address, psize);
743 	return hash__ptep_set_access_flags(ptep, entry);
744 }
745 
746 #define __HAVE_ARCH_PTE_SAME
747 static inline int pte_same(pte_t pte_a, pte_t pte_b)
748 {
749 	if (radix_enabled())
750 		return radix__pte_same(pte_a, pte_b);
751 	return hash__pte_same(pte_a, pte_b);
752 }
753 
754 static inline int pte_none(pte_t pte)
755 {
756 	if (radix_enabled())
757 		return radix__pte_none(pte);
758 	return hash__pte_none(pte);
759 }
760 
761 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
762 				pte_t *ptep, pte_t pte, int percpu)
763 {
764 
765 	VM_WARN_ON(!(pte_raw(pte) & cpu_to_be64(_PAGE_PTE)));
766 	/*
767 	 * Keep the _PAGE_PTE added till we are sure we handle _PAGE_PTE
768 	 * in all the callers.
769 	 */
770 	pte = __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PTE));
771 
772 	if (radix_enabled())
773 		return radix__set_pte_at(mm, addr, ptep, pte, percpu);
774 	return hash__set_pte_at(mm, addr, ptep, pte, percpu);
775 }
776 
777 #define _PAGE_CACHE_CTL	(_PAGE_SAO | _PAGE_NON_IDEMPOTENT | _PAGE_TOLERANT)
778 
779 #define pgprot_noncached pgprot_noncached
780 static inline pgprot_t pgprot_noncached(pgprot_t prot)
781 {
782 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
783 			_PAGE_NON_IDEMPOTENT);
784 }
785 
786 #define pgprot_noncached_wc pgprot_noncached_wc
787 static inline pgprot_t pgprot_noncached_wc(pgprot_t prot)
788 {
789 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
790 			_PAGE_TOLERANT);
791 }
792 
793 #define pgprot_cached pgprot_cached
794 static inline pgprot_t pgprot_cached(pgprot_t prot)
795 {
796 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL));
797 }
798 
799 #define pgprot_writecombine pgprot_writecombine
800 static inline pgprot_t pgprot_writecombine(pgprot_t prot)
801 {
802 	return pgprot_noncached_wc(prot);
803 }
804 /*
805  * check a pte mapping have cache inhibited property
806  */
807 static inline bool pte_ci(pte_t pte)
808 {
809 	__be64 pte_v = pte_raw(pte);
810 
811 	if (((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_TOLERANT)) ||
812 	    ((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_NON_IDEMPOTENT)))
813 		return true;
814 	return false;
815 }
816 
817 static inline void pmd_clear(pmd_t *pmdp)
818 {
819 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
820 		/*
821 		 * Don't use this if we can possibly have a hash page table
822 		 * entry mapping this.
823 		 */
824 		WARN_ON((pmd_val(*pmdp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
825 	}
826 	*pmdp = __pmd(0);
827 }
828 
829 static inline int pmd_none(pmd_t pmd)
830 {
831 	return !pmd_raw(pmd);
832 }
833 
834 static inline int pmd_present(pmd_t pmd)
835 {
836 	/*
837 	 * A pmd is considerent present if _PAGE_PRESENT is set.
838 	 * We also need to consider the pmd present which is marked
839 	 * invalid during a split. Hence we look for _PAGE_INVALID
840 	 * if we find _PAGE_PRESENT cleared.
841 	 */
842 	if (pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID))
843 		return true;
844 
845 	return false;
846 }
847 
848 static inline int pmd_is_serializing(pmd_t pmd)
849 {
850 	/*
851 	 * If the pmd is undergoing a split, the _PAGE_PRESENT bit is clear
852 	 * and _PAGE_INVALID is set (see pmd_present, pmdp_invalidate).
853 	 *
854 	 * This condition may also occur when flushing a pmd while flushing
855 	 * it (see ptep_modify_prot_start), so callers must ensure this
856 	 * case is fine as well.
857 	 */
858 	if ((pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID)) ==
859 						cpu_to_be64(_PAGE_INVALID))
860 		return true;
861 
862 	return false;
863 }
864 
865 static inline int pmd_bad(pmd_t pmd)
866 {
867 	if (radix_enabled())
868 		return radix__pmd_bad(pmd);
869 	return hash__pmd_bad(pmd);
870 }
871 
872 static inline void pud_clear(pud_t *pudp)
873 {
874 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
875 		/*
876 		 * Don't use this if we can possibly have a hash page table
877 		 * entry mapping this.
878 		 */
879 		WARN_ON((pud_val(*pudp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
880 	}
881 	*pudp = __pud(0);
882 }
883 
884 static inline int pud_none(pud_t pud)
885 {
886 	return !pud_raw(pud);
887 }
888 
889 static inline int pud_present(pud_t pud)
890 {
891 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PRESENT));
892 }
893 
894 extern struct page *pud_page(pud_t pud);
895 extern struct page *pmd_page(pmd_t pmd);
896 static inline pte_t pud_pte(pud_t pud)
897 {
898 	return __pte_raw(pud_raw(pud));
899 }
900 
901 static inline pud_t pte_pud(pte_t pte)
902 {
903 	return __pud_raw(pte_raw(pte));
904 }
905 
906 static inline pte_t *pudp_ptep(pud_t *pud)
907 {
908 	return (pte_t *)pud;
909 }
910 
911 #define pud_pfn(pud)		pte_pfn(pud_pte(pud))
912 #define pud_dirty(pud)		pte_dirty(pud_pte(pud))
913 #define pud_young(pud)		pte_young(pud_pte(pud))
914 #define pud_mkold(pud)		pte_pud(pte_mkold(pud_pte(pud)))
915 #define pud_wrprotect(pud)	pte_pud(pte_wrprotect(pud_pte(pud)))
916 #define pud_mkdirty(pud)	pte_pud(pte_mkdirty(pud_pte(pud)))
917 #define pud_mkclean(pud)	pte_pud(pte_mkclean(pud_pte(pud)))
918 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
919 #define pud_mkwrite(pud)	pte_pud(pte_mkwrite_novma(pud_pte(pud)))
920 #define pud_write(pud)		pte_write(pud_pte(pud))
921 
922 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
923 #define pud_soft_dirty(pmd)    pte_soft_dirty(pud_pte(pud))
924 #define pud_mksoft_dirty(pmd)  pte_pud(pte_mksoft_dirty(pud_pte(pud)))
925 #define pud_clear_soft_dirty(pmd) pte_pud(pte_clear_soft_dirty(pud_pte(pud)))
926 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
927 
928 static inline int pud_bad(pud_t pud)
929 {
930 	if (radix_enabled())
931 		return radix__pud_bad(pud);
932 	return hash__pud_bad(pud);
933 }
934 
935 #define pud_access_permitted pud_access_permitted
936 static inline bool pud_access_permitted(pud_t pud, bool write)
937 {
938 	return pte_access_permitted(pud_pte(pud), write);
939 }
940 
941 #define __p4d_raw(x)	((p4d_t) { __pgd_raw(x) })
942 static inline __be64 p4d_raw(p4d_t x)
943 {
944 	return pgd_raw(x.pgd);
945 }
946 
947 #define p4d_write(p4d)		pte_write(p4d_pte(p4d))
948 
949 static inline void p4d_clear(p4d_t *p4dp)
950 {
951 	*p4dp = __p4d(0);
952 }
953 
954 static inline int p4d_none(p4d_t p4d)
955 {
956 	return !p4d_raw(p4d);
957 }
958 
959 static inline int p4d_present(p4d_t p4d)
960 {
961 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PRESENT));
962 }
963 
964 static inline pte_t p4d_pte(p4d_t p4d)
965 {
966 	return __pte_raw(p4d_raw(p4d));
967 }
968 
969 static inline p4d_t pte_p4d(pte_t pte)
970 {
971 	return __p4d_raw(pte_raw(pte));
972 }
973 
974 static inline int p4d_bad(p4d_t p4d)
975 {
976 	if (radix_enabled())
977 		return radix__p4d_bad(p4d);
978 	return hash__p4d_bad(p4d);
979 }
980 
981 #define p4d_access_permitted p4d_access_permitted
982 static inline bool p4d_access_permitted(p4d_t p4d, bool write)
983 {
984 	return pte_access_permitted(p4d_pte(p4d), write);
985 }
986 
987 extern struct page *p4d_page(p4d_t p4d);
988 
989 /* Pointers in the page table tree are physical addresses */
990 #define __pgtable_ptr_val(ptr)	__pa(ptr)
991 
992 static inline pud_t *p4d_pgtable(p4d_t p4d)
993 {
994 	return (pud_t *)__va(p4d_val(p4d) & ~P4D_MASKED_BITS);
995 }
996 
997 static inline pmd_t *pud_pgtable(pud_t pud)
998 {
999 	return (pmd_t *)__va(pud_val(pud) & ~PUD_MASKED_BITS);
1000 }
1001 
1002 #define pmd_ERROR(e) \
1003 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
1004 #define pud_ERROR(e) \
1005 	pr_err("%s:%d: bad pud %08lx.\n", __FILE__, __LINE__, pud_val(e))
1006 #define pgd_ERROR(e) \
1007 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
1008 
1009 static inline int map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot)
1010 {
1011 	if (radix_enabled()) {
1012 #if defined(CONFIG_PPC_RADIX_MMU) && defined(DEBUG_VM)
1013 		unsigned long page_size = 1 << mmu_psize_defs[mmu_io_psize].shift;
1014 		WARN((page_size != PAGE_SIZE), "I/O page size != PAGE_SIZE");
1015 #endif
1016 		return radix__map_kernel_page(ea, pa, prot, PAGE_SIZE);
1017 	}
1018 	return hash__map_kernel_page(ea, pa, prot);
1019 }
1020 
1021 void unmap_kernel_page(unsigned long va);
1022 
1023 static inline int __meminit vmemmap_create_mapping(unsigned long start,
1024 						   unsigned long page_size,
1025 						   unsigned long phys)
1026 {
1027 	if (radix_enabled())
1028 		return radix__vmemmap_create_mapping(start, page_size, phys);
1029 	return hash__vmemmap_create_mapping(start, page_size, phys);
1030 }
1031 
1032 #ifdef CONFIG_MEMORY_HOTPLUG
1033 static inline void vmemmap_remove_mapping(unsigned long start,
1034 					  unsigned long page_size)
1035 {
1036 	if (radix_enabled())
1037 		return radix__vmemmap_remove_mapping(start, page_size);
1038 	return hash__vmemmap_remove_mapping(start, page_size);
1039 }
1040 #endif
1041 
1042 static inline pte_t pmd_pte(pmd_t pmd)
1043 {
1044 	return __pte_raw(pmd_raw(pmd));
1045 }
1046 
1047 static inline pmd_t pte_pmd(pte_t pte)
1048 {
1049 	return __pmd_raw(pte_raw(pte));
1050 }
1051 
1052 static inline pte_t *pmdp_ptep(pmd_t *pmd)
1053 {
1054 	return (pte_t *)pmd;
1055 }
1056 #define pmd_pfn(pmd)		pte_pfn(pmd_pte(pmd))
1057 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
1058 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
1059 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
1060 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
1061 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
1062 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
1063 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
1064 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
1065 
1066 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1067 #define pmd_soft_dirty(pmd)    pte_soft_dirty(pmd_pte(pmd))
1068 #define pmd_mksoft_dirty(pmd)  pte_pmd(pte_mksoft_dirty(pmd_pte(pmd)))
1069 #define pmd_clear_soft_dirty(pmd) pte_pmd(pte_clear_soft_dirty(pmd_pte(pmd)))
1070 
1071 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1072 #define pmd_swp_mksoft_dirty(pmd)	pte_pmd(pte_swp_mksoft_dirty(pmd_pte(pmd)))
1073 #define pmd_swp_soft_dirty(pmd)		pte_swp_soft_dirty(pmd_pte(pmd))
1074 #define pmd_swp_clear_soft_dirty(pmd)	pte_pmd(pte_swp_clear_soft_dirty(pmd_pte(pmd)))
1075 #endif
1076 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
1077 
1078 #ifdef CONFIG_NUMA_BALANCING
1079 static inline int pmd_protnone(pmd_t pmd)
1080 {
1081 	return pte_protnone(pmd_pte(pmd));
1082 }
1083 #endif /* CONFIG_NUMA_BALANCING */
1084 
1085 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
1086 
1087 #define pmd_access_permitted pmd_access_permitted
1088 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1089 {
1090 	/*
1091 	 * pmdp_invalidate sets this combination (which is not caught by
1092 	 * !pte_present() check in pte_access_permitted), to prevent
1093 	 * lock-free lookups, as part of the serialize_against_pte_lookup()
1094 	 * synchronisation.
1095 	 *
1096 	 * This also catches the case where the PTE's hardware PRESENT bit is
1097 	 * cleared while TLB is flushed, which is suboptimal but should not
1098 	 * be frequent.
1099 	 */
1100 	if (pmd_is_serializing(pmd))
1101 		return false;
1102 
1103 	return pte_access_permitted(pmd_pte(pmd), write);
1104 }
1105 
1106 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1107 extern pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot);
1108 extern pud_t pfn_pud(unsigned long pfn, pgprot_t pgprot);
1109 extern pmd_t mk_pmd(struct page *page, pgprot_t pgprot);
1110 extern pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot);
1111 extern void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1112 		       pmd_t *pmdp, pmd_t pmd);
1113 extern void set_pud_at(struct mm_struct *mm, unsigned long addr,
1114 		       pud_t *pudp, pud_t pud);
1115 
1116 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1117 					unsigned long addr, pmd_t *pmd)
1118 {
1119 }
1120 
1121 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1122 					unsigned long addr, pud_t *pud)
1123 {
1124 }
1125 
1126 extern int hash__has_transparent_hugepage(void);
1127 static inline int has_transparent_hugepage(void)
1128 {
1129 	if (radix_enabled())
1130 		return radix__has_transparent_hugepage();
1131 	return hash__has_transparent_hugepage();
1132 }
1133 #define has_transparent_hugepage has_transparent_hugepage
1134 
1135 static inline int has_transparent_pud_hugepage(void)
1136 {
1137 	if (radix_enabled())
1138 		return radix__has_transparent_pud_hugepage();
1139 	return 0;
1140 }
1141 #define has_transparent_pud_hugepage has_transparent_pud_hugepage
1142 
1143 static inline unsigned long
1144 pmd_hugepage_update(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp,
1145 		    unsigned long clr, unsigned long set)
1146 {
1147 	if (radix_enabled())
1148 		return radix__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1149 	return hash__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1150 }
1151 
1152 static inline unsigned long
1153 pud_hugepage_update(struct mm_struct *mm, unsigned long addr, pud_t *pudp,
1154 		    unsigned long clr, unsigned long set)
1155 {
1156 	if (radix_enabled())
1157 		return radix__pud_hugepage_update(mm, addr, pudp, clr, set);
1158 	BUG();
1159 	return pud_val(*pudp);
1160 }
1161 
1162 /*
1163  * For radix we should always find H_PAGE_HASHPTE zero. Hence
1164  * the below will work for radix too
1165  */
1166 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
1167 					      unsigned long addr, pmd_t *pmdp)
1168 {
1169 	unsigned long old;
1170 
1171 	if ((pmd_raw(*pmdp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1172 		return 0;
1173 	old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
1174 	return ((old & _PAGE_ACCESSED) != 0);
1175 }
1176 
1177 static inline int __pudp_test_and_clear_young(struct mm_struct *mm,
1178 					      unsigned long addr, pud_t *pudp)
1179 {
1180 	unsigned long old;
1181 
1182 	if ((pud_raw(*pudp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1183 		return 0;
1184 	old = pud_hugepage_update(mm, addr, pudp, _PAGE_ACCESSED, 0);
1185 	return ((old & _PAGE_ACCESSED) != 0);
1186 }
1187 
1188 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1189 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1190 				      pmd_t *pmdp)
1191 {
1192 	if (pmd_write(*pmdp))
1193 		pmd_hugepage_update(mm, addr, pmdp, _PAGE_WRITE, 0);
1194 }
1195 
1196 #define __HAVE_ARCH_PUDP_SET_WRPROTECT
1197 static inline void pudp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1198 				      pud_t *pudp)
1199 {
1200 	if (pud_write(*pudp))
1201 		pud_hugepage_update(mm, addr, pudp, _PAGE_WRITE, 0);
1202 }
1203 
1204 /*
1205  * Only returns true for a THP. False for pmd migration entry.
1206  * We also need to return true when we come across a pte that
1207  * in between a thp split. While splitting THP, we mark the pmd
1208  * invalid (pmdp_invalidate()) before we set it with pte page
1209  * address. A pmd_trans_huge() check against a pmd entry during that time
1210  * should return true.
1211  * We should not call this on a hugetlb entry. We should check for HugeTLB
1212  * entry using vma->vm_flags
1213  * The page table walk rule is explained in Documentation/mm/transhuge.rst
1214  */
1215 static inline int pmd_trans_huge(pmd_t pmd)
1216 {
1217 	if (!pmd_present(pmd))
1218 		return false;
1219 
1220 	if (radix_enabled())
1221 		return radix__pmd_trans_huge(pmd);
1222 	return hash__pmd_trans_huge(pmd);
1223 }
1224 
1225 static inline int pud_trans_huge(pud_t pud)
1226 {
1227 	if (!pud_present(pud))
1228 		return false;
1229 
1230 	if (radix_enabled())
1231 		return radix__pud_trans_huge(pud);
1232 	return 0;
1233 }
1234 
1235 
1236 #define __HAVE_ARCH_PMD_SAME
1237 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1238 {
1239 	if (radix_enabled())
1240 		return radix__pmd_same(pmd_a, pmd_b);
1241 	return hash__pmd_same(pmd_a, pmd_b);
1242 }
1243 
1244 #define pud_same pud_same
1245 static inline int pud_same(pud_t pud_a, pud_t pud_b)
1246 {
1247 	if (radix_enabled())
1248 		return radix__pud_same(pud_a, pud_b);
1249 	return hash__pud_same(pud_a, pud_b);
1250 }
1251 
1252 
1253 static inline pmd_t __pmd_mkhuge(pmd_t pmd)
1254 {
1255 	if (radix_enabled())
1256 		return radix__pmd_mkhuge(pmd);
1257 	return hash__pmd_mkhuge(pmd);
1258 }
1259 
1260 static inline pud_t __pud_mkhuge(pud_t pud)
1261 {
1262 	if (radix_enabled())
1263 		return radix__pud_mkhuge(pud);
1264 	BUG();
1265 	return pud;
1266 }
1267 
1268 /*
1269  * pfn_pmd return a pmd_t that can be used as pmd pte entry.
1270  */
1271 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1272 {
1273 #ifdef CONFIG_DEBUG_VM
1274 	if (radix_enabled())
1275 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE)) == 0);
1276 	else
1277 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE)) !=
1278 			cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE));
1279 #endif
1280 	return pmd;
1281 }
1282 
1283 static inline pud_t pud_mkhuge(pud_t pud)
1284 {
1285 #ifdef CONFIG_DEBUG_VM
1286 	if (radix_enabled())
1287 		WARN_ON((pud_raw(pud) & cpu_to_be64(_PAGE_PTE)) == 0);
1288 	else
1289 		WARN_ON(1);
1290 #endif
1291 	return pud;
1292 }
1293 
1294 
1295 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1296 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1297 				 unsigned long address, pmd_t *pmdp,
1298 				 pmd_t entry, int dirty);
1299 #define __HAVE_ARCH_PUDP_SET_ACCESS_FLAGS
1300 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1301 				 unsigned long address, pud_t *pudp,
1302 				 pud_t entry, int dirty);
1303 
1304 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1305 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1306 				     unsigned long address, pmd_t *pmdp);
1307 #define __HAVE_ARCH_PUDP_TEST_AND_CLEAR_YOUNG
1308 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1309 				     unsigned long address, pud_t *pudp);
1310 
1311 
1312 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1313 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1314 					    unsigned long addr, pmd_t *pmdp)
1315 {
1316 	if (radix_enabled())
1317 		return radix__pmdp_huge_get_and_clear(mm, addr, pmdp);
1318 	return hash__pmdp_huge_get_and_clear(mm, addr, pmdp);
1319 }
1320 
1321 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1322 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1323 					    unsigned long addr, pud_t *pudp)
1324 {
1325 	if (radix_enabled())
1326 		return radix__pudp_huge_get_and_clear(mm, addr, pudp);
1327 	BUG();
1328 	return *pudp;
1329 }
1330 
1331 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1332 					unsigned long address, pmd_t *pmdp)
1333 {
1334 	if (radix_enabled())
1335 		return radix__pmdp_collapse_flush(vma, address, pmdp);
1336 	return hash__pmdp_collapse_flush(vma, address, pmdp);
1337 }
1338 #define pmdp_collapse_flush pmdp_collapse_flush
1339 
1340 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1341 pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1342 				   unsigned long addr,
1343 				   pmd_t *pmdp, int full);
1344 
1345 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
1346 pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
1347 				   unsigned long addr,
1348 				   pud_t *pudp, int full);
1349 
1350 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1351 static inline void pgtable_trans_huge_deposit(struct mm_struct *mm,
1352 					      pmd_t *pmdp, pgtable_t pgtable)
1353 {
1354 	if (radix_enabled())
1355 		return radix__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1356 	return hash__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1357 }
1358 
1359 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1360 static inline pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm,
1361 						    pmd_t *pmdp)
1362 {
1363 	if (radix_enabled())
1364 		return radix__pgtable_trans_huge_withdraw(mm, pmdp);
1365 	return hash__pgtable_trans_huge_withdraw(mm, pmdp);
1366 }
1367 
1368 #define __HAVE_ARCH_PMDP_INVALIDATE
1369 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
1370 			     pmd_t *pmdp);
1371 
1372 #define pmd_move_must_withdraw pmd_move_must_withdraw
1373 struct spinlock;
1374 extern int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
1375 				  struct spinlock *old_pmd_ptl,
1376 				  struct vm_area_struct *vma);
1377 /*
1378  * Hash translation mode use the deposited table to store hash pte
1379  * slot information.
1380  */
1381 #define arch_needs_pgtable_deposit arch_needs_pgtable_deposit
1382 static inline bool arch_needs_pgtable_deposit(void)
1383 {
1384 	if (radix_enabled())
1385 		return false;
1386 	return true;
1387 }
1388 extern void serialize_against_pte_lookup(struct mm_struct *mm);
1389 
1390 
1391 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
1392 {
1393 	if (radix_enabled())
1394 		return radix__pmd_mkdevmap(pmd);
1395 	return hash__pmd_mkdevmap(pmd);
1396 }
1397 
1398 static inline pud_t pud_mkdevmap(pud_t pud)
1399 {
1400 	if (radix_enabled())
1401 		return radix__pud_mkdevmap(pud);
1402 	BUG();
1403 	return pud;
1404 }
1405 
1406 static inline int pmd_devmap(pmd_t pmd)
1407 {
1408 	return pte_devmap(pmd_pte(pmd));
1409 }
1410 
1411 static inline int pud_devmap(pud_t pud)
1412 {
1413 	return pte_devmap(pud_pte(pud));
1414 }
1415 
1416 static inline int pgd_devmap(pgd_t pgd)
1417 {
1418 	return 0;
1419 }
1420 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1421 
1422 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1423 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1424 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1425 			     pte_t *, pte_t, pte_t);
1426 
1427 /*
1428  * Returns true for a R -> RW upgrade of pte
1429  */
1430 static inline bool is_pte_rw_upgrade(unsigned long old_val, unsigned long new_val)
1431 {
1432 	if (!(old_val & _PAGE_READ))
1433 		return false;
1434 
1435 	if ((!(old_val & _PAGE_WRITE)) && (new_val & _PAGE_WRITE))
1436 		return true;
1437 
1438 	return false;
1439 }
1440 
1441 #endif /* __ASSEMBLY__ */
1442 #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */
1443