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