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 #define _PAGE_BIT_SWAP_TYPE	0
17 
18 #define _PAGE_EXEC		0x00001 /* execute permission */
19 #define _PAGE_WRITE		0x00002 /* write access allowed */
20 #define _PAGE_READ		0x00004	/* read access allowed */
21 #define _PAGE_RW		(_PAGE_READ | _PAGE_WRITE)
22 #define _PAGE_RWX		(_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC)
23 #define _PAGE_PRIVILEGED	0x00008 /* kernel access only */
24 #define _PAGE_SAO		0x00010 /* Strong access order */
25 #define _PAGE_NON_IDEMPOTENT	0x00020 /* non idempotent memory */
26 #define _PAGE_TOLERANT		0x00030 /* tolerant memory, cache inhibited */
27 #define _PAGE_DIRTY		0x00080 /* C: page changed */
28 #define _PAGE_ACCESSED		0x00100 /* R: page referenced */
29 /*
30  * Software bits
31  */
32 #define _RPAGE_SW0		0x2000000000000000UL
33 #define _RPAGE_SW1		0x00800
34 #define _RPAGE_SW2		0x00400
35 #define _RPAGE_SW3		0x00200
36 #define _RPAGE_RSV1		0x00040UL
37 
38 #define _RPAGE_PKEY_BIT4	0x1000000000000000UL
39 #define _RPAGE_PKEY_BIT3	0x0800000000000000UL
40 #define _RPAGE_PKEY_BIT2	0x0400000000000000UL
41 #define _RPAGE_PKEY_BIT1	0x0200000000000000UL
42 #define _RPAGE_PKEY_BIT0	0x0100000000000000UL
43 
44 #define _PAGE_PTE		0x4000000000000000UL	/* distinguishes PTEs from pointers */
45 #define _PAGE_PRESENT		0x8000000000000000UL	/* pte contains a translation */
46 /*
47  * We need to mark a pmd pte invalid while splitting. We can do that by clearing
48  * the _PAGE_PRESENT bit. But then that will be taken as a swap pte. In order to
49  * differentiate between two use a SW field when invalidating.
50  *
51  * We do that temporary invalidate for regular pte entry in ptep_set_access_flags
52  *
53  * This is used only when _PAGE_PRESENT is cleared.
54  */
55 #define _PAGE_INVALID		_RPAGE_SW0
56 
57 /*
58  * Top and bottom bits of RPN which can be used by hash
59  * translation mode, because we expect them to be zero
60  * otherwise.
61  */
62 #define _RPAGE_RPN0		0x01000
63 #define _RPAGE_RPN1		0x02000
64 #define _RPAGE_RPN43		0x0080000000000000UL
65 #define _RPAGE_RPN42		0x0040000000000000UL
66 #define _RPAGE_RPN41		0x0020000000000000UL
67 
68 /* Max physical address bit as per radix table */
69 #define _RPAGE_PA_MAX		56
70 
71 /*
72  * Max physical address bit we will use for now.
73  *
74  * This is mostly a hardware limitation and for now Power9 has
75  * a 51 bit limit.
76  *
77  * This is different from the number of physical bit required to address
78  * the last byte of memory. That is defined by MAX_PHYSMEM_BITS.
79  * MAX_PHYSMEM_BITS is a linux limitation imposed by the maximum
80  * number of sections we can support (SECTIONS_SHIFT).
81  *
82  * This is different from Radix page table limitation above and
83  * should always be less than that. The limit is done such that
84  * we can overload the bits between _RPAGE_PA_MAX and _PAGE_PA_MAX
85  * for hash linux page table specific bits.
86  *
87  * In order to be compatible with future hardware generations we keep
88  * some offsets and limit this for now to 53
89  */
90 #define _PAGE_PA_MAX		53
91 
92 #define _PAGE_SOFT_DIRTY	_RPAGE_SW3 /* software: software dirty tracking */
93 #define _PAGE_SPECIAL		_RPAGE_SW2 /* software: special page */
94 #define _PAGE_DEVMAP		_RPAGE_SW1 /* software: ZONE_DEVICE page */
95 
96 /*
97  * Drivers request for cache inhibited pte mapping using _PAGE_NO_CACHE
98  * Instead of fixing all of them, add an alternate define which
99  * maps CI pte mapping.
100  */
101 #define _PAGE_NO_CACHE		_PAGE_TOLERANT
102 /*
103  * We support _RPAGE_PA_MAX bit real address in pte. On the linux side
104  * we are limited by _PAGE_PA_MAX. Clear everything above _PAGE_PA_MAX
105  * and every thing below PAGE_SHIFT;
106  */
107 #define PTE_RPN_MASK	(((1UL << _PAGE_PA_MAX) - 1) & (PAGE_MASK))
108 /*
109  * set of bits not changed in pmd_modify. Even though we have hash specific bits
110  * in here, on radix we expect them to be zero.
111  */
112 #define _HPAGE_CHG_MASK (PTE_RPN_MASK | _PAGE_HPTEFLAGS | _PAGE_DIRTY | \
113 			 _PAGE_ACCESSED | H_PAGE_THP_HUGE | _PAGE_PTE | \
114 			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP)
115 /*
116  * user access blocked by key
117  */
118 #define _PAGE_KERNEL_RW		(_PAGE_PRIVILEGED | _PAGE_RW | _PAGE_DIRTY)
119 #define _PAGE_KERNEL_RO		 (_PAGE_PRIVILEGED | _PAGE_READ)
120 #define _PAGE_KERNEL_ROX	 (_PAGE_PRIVILEGED | _PAGE_READ | _PAGE_EXEC)
121 #define _PAGE_KERNEL_RWX	(_PAGE_PRIVILEGED | _PAGE_DIRTY |	\
122 				 _PAGE_RW | _PAGE_EXEC)
123 /*
124  * _PAGE_CHG_MASK masks of bits that are to be preserved across
125  * pgprot changes
126  */
127 #define _PAGE_CHG_MASK	(PTE_RPN_MASK | _PAGE_HPTEFLAGS | _PAGE_DIRTY | \
128 			 _PAGE_ACCESSED | _PAGE_SPECIAL | _PAGE_PTE |	\
129 			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP)
130 
131 /*
132  * We define 2 sets of base prot bits, one for basic pages (ie,
133  * cacheable kernel and user pages) and one for non cacheable
134  * pages. We always set _PAGE_COHERENT when SMP is enabled or
135  * the processor might need it for DMA coherency.
136  */
137 #define _PAGE_BASE_NC	(_PAGE_PRESENT | _PAGE_ACCESSED)
138 #define _PAGE_BASE	(_PAGE_BASE_NC)
139 
140 /* Permission masks used to generate the __P and __S table,
141  *
142  * Note:__pgprot is defined in arch/powerpc/include/asm/page.h
143  *
144  * Write permissions imply read permissions for now (we could make write-only
145  * pages on BookE but we don't bother for now). Execute permission control is
146  * possible on platforms that define _PAGE_EXEC
147  */
148 #define PAGE_NONE	__pgprot(_PAGE_BASE | _PAGE_PRIVILEGED)
149 #define PAGE_SHARED	__pgprot(_PAGE_BASE | _PAGE_RW)
150 #define PAGE_SHARED_X	__pgprot(_PAGE_BASE | _PAGE_RW | _PAGE_EXEC)
151 #define PAGE_COPY	__pgprot(_PAGE_BASE | _PAGE_READ)
152 #define PAGE_COPY_X	__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
153 #define PAGE_READONLY	__pgprot(_PAGE_BASE | _PAGE_READ)
154 #define PAGE_READONLY_X	__pgprot(_PAGE_BASE | _PAGE_READ | _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 | \
159 				 _PAGE_TOLERANT)
160 #define PAGE_KERNEL_NCG	__pgprot(_PAGE_BASE_NC | _PAGE_KERNEL_RW | \
161 				 _PAGE_NON_IDEMPOTENT)
162 #define PAGE_KERNEL_X	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RWX)
163 #define PAGE_KERNEL_RO	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RO)
164 #define PAGE_KERNEL_ROX	__pgprot(_PAGE_BASE | _PAGE_KERNEL_ROX)
165 
166 /*
167  * Protection used for kernel text. We want the debuggers to be able to
168  * set breakpoints anywhere, so don't write protect the kernel text
169  * on platforms where such control is possible.
170  */
171 #if defined(CONFIG_KGDB) || defined(CONFIG_XMON) || defined(CONFIG_BDI_SWITCH) || \
172 	defined(CONFIG_KPROBES) || defined(CONFIG_DYNAMIC_FTRACE)
173 #define PAGE_KERNEL_TEXT	PAGE_KERNEL_X
174 #else
175 #define PAGE_KERNEL_TEXT	PAGE_KERNEL_ROX
176 #endif
177 
178 /* Make modules code happy. We don't set RO yet */
179 #define PAGE_KERNEL_EXEC	PAGE_KERNEL_X
180 #define PAGE_AGP		(PAGE_KERNEL_NC)
181 
182 #ifndef __ASSEMBLY__
183 /*
184  * page table defines
185  */
186 extern unsigned long __pte_index_size;
187 extern unsigned long __pmd_index_size;
188 extern unsigned long __pud_index_size;
189 extern unsigned long __pgd_index_size;
190 extern unsigned long __pud_cache_index;
191 #define PTE_INDEX_SIZE  __pte_index_size
192 #define PMD_INDEX_SIZE  __pmd_index_size
193 #define PUD_INDEX_SIZE  __pud_index_size
194 #define PGD_INDEX_SIZE  __pgd_index_size
195 /* pmd table use page table fragments */
196 #define PMD_CACHE_INDEX  0
197 #define PUD_CACHE_INDEX __pud_cache_index
198 /*
199  * Because of use of pte fragments and THP, size of page table
200  * are not always derived out of index size above.
201  */
202 extern unsigned long __pte_table_size;
203 extern unsigned long __pmd_table_size;
204 extern unsigned long __pud_table_size;
205 extern unsigned long __pgd_table_size;
206 #define PTE_TABLE_SIZE	__pte_table_size
207 #define PMD_TABLE_SIZE	__pmd_table_size
208 #define PUD_TABLE_SIZE	__pud_table_size
209 #define PGD_TABLE_SIZE	__pgd_table_size
210 
211 extern unsigned long __pmd_val_bits;
212 extern unsigned long __pud_val_bits;
213 extern unsigned long __pgd_val_bits;
214 #define PMD_VAL_BITS	__pmd_val_bits
215 #define PUD_VAL_BITS	__pud_val_bits
216 #define PGD_VAL_BITS	__pgd_val_bits
217 
218 extern unsigned long __pte_frag_nr;
219 #define PTE_FRAG_NR __pte_frag_nr
220 extern unsigned long __pte_frag_size_shift;
221 #define PTE_FRAG_SIZE_SHIFT __pte_frag_size_shift
222 #define PTE_FRAG_SIZE (1UL << PTE_FRAG_SIZE_SHIFT)
223 
224 extern unsigned long __pmd_frag_nr;
225 #define PMD_FRAG_NR __pmd_frag_nr
226 extern unsigned long __pmd_frag_size_shift;
227 #define PMD_FRAG_SIZE_SHIFT __pmd_frag_size_shift
228 #define PMD_FRAG_SIZE (1UL << PMD_FRAG_SIZE_SHIFT)
229 
230 #define PTRS_PER_PTE	(1 << PTE_INDEX_SIZE)
231 #define PTRS_PER_PMD	(1 << PMD_INDEX_SIZE)
232 #define PTRS_PER_PUD	(1 << PUD_INDEX_SIZE)
233 #define PTRS_PER_PGD	(1 << PGD_INDEX_SIZE)
234 
235 #define MAX_PTRS_PER_PGD	(1 << (H_PGD_INDEX_SIZE > RADIX_PGD_INDEX_SIZE ? \
236 				       H_PGD_INDEX_SIZE : RADIX_PGD_INDEX_SIZE))
237 
238 /* PMD_SHIFT determines what a second-level page table entry can map */
239 #define PMD_SHIFT	(PAGE_SHIFT + PTE_INDEX_SIZE)
240 #define PMD_SIZE	(1UL << PMD_SHIFT)
241 #define PMD_MASK	(~(PMD_SIZE-1))
242 
243 /* PUD_SHIFT determines what a third-level page table entry can map */
244 #define PUD_SHIFT	(PMD_SHIFT + PMD_INDEX_SIZE)
245 #define PUD_SIZE	(1UL << PUD_SHIFT)
246 #define PUD_MASK	(~(PUD_SIZE-1))
247 
248 /* PGDIR_SHIFT determines what a fourth-level page table entry can map */
249 #define PGDIR_SHIFT	(PUD_SHIFT + PUD_INDEX_SIZE)
250 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
251 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
252 
253 /* Bits to mask out from a PMD to get to the PTE page */
254 #define PMD_MASKED_BITS		0xc0000000000000ffUL
255 /* Bits to mask out from a PUD to get to the PMD page */
256 #define PUD_MASKED_BITS		0xc0000000000000ffUL
257 /* Bits to mask out from a PGD to get to the PUD page */
258 #define P4D_MASKED_BITS		0xc0000000000000ffUL
259 
260 /*
261  * Used as an indicator for rcu callback functions
262  */
263 enum pgtable_index {
264 	PTE_INDEX = 0,
265 	PMD_INDEX,
266 	PUD_INDEX,
267 	PGD_INDEX,
268 	/*
269 	 * Below are used with 4k page size and hugetlb
270 	 */
271 	HTLB_16M_INDEX,
272 	HTLB_16G_INDEX,
273 };
274 
275 extern unsigned long __vmalloc_start;
276 extern unsigned long __vmalloc_end;
277 #define VMALLOC_START	__vmalloc_start
278 #define VMALLOC_END	__vmalloc_end
279 
280 static inline unsigned int ioremap_max_order(void)
281 {
282 	if (radix_enabled())
283 		return PUD_SHIFT;
284 	return 7 + PAGE_SHIFT; /* default from linux/vmalloc.h */
285 }
286 #define IOREMAP_MAX_ORDER ioremap_max_order()
287 
288 extern unsigned long __kernel_virt_start;
289 extern unsigned long __kernel_io_start;
290 extern unsigned long __kernel_io_end;
291 #define KERN_VIRT_START __kernel_virt_start
292 #define KERN_IO_START  __kernel_io_start
293 #define KERN_IO_END __kernel_io_end
294 
295 extern struct page *vmemmap;
296 extern unsigned long pci_io_base;
297 #endif /* __ASSEMBLY__ */
298 
299 #include <asm/book3s/64/hash.h>
300 #include <asm/book3s/64/radix.h>
301 
302 #if H_MAX_PHYSMEM_BITS > R_MAX_PHYSMEM_BITS
303 #define  MAX_PHYSMEM_BITS	H_MAX_PHYSMEM_BITS
304 #else
305 #define  MAX_PHYSMEM_BITS	R_MAX_PHYSMEM_BITS
306 #endif
307 
308 
309 #ifdef CONFIG_PPC_64K_PAGES
310 #include <asm/book3s/64/pgtable-64k.h>
311 #else
312 #include <asm/book3s/64/pgtable-4k.h>
313 #endif
314 
315 #include <asm/barrier.h>
316 /*
317  * IO space itself carved into the PIO region (ISA and PHB IO space) and
318  * the ioremap space
319  *
320  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
321  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
322  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
323  */
324 #define FULL_IO_SIZE	0x80000000ul
325 #define  ISA_IO_BASE	(KERN_IO_START)
326 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
327 #define  PHB_IO_BASE	(ISA_IO_END)
328 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
329 #define IOREMAP_BASE	(PHB_IO_END)
330 #define IOREMAP_START	(ioremap_bot)
331 #define IOREMAP_END	(KERN_IO_END - FIXADDR_SIZE)
332 #define FIXADDR_SIZE	SZ_32M
333 
334 /* Advertise special mapping type for AGP */
335 #define HAVE_PAGE_AGP
336 
337 #ifndef __ASSEMBLY__
338 
339 /*
340  * This is the default implementation of various PTE accessors, it's
341  * used in all cases except Book3S with 64K pages where we have a
342  * concept of sub-pages
343  */
344 #ifndef __real_pte
345 
346 #define __real_pte(e, p, o)		((real_pte_t){(e)})
347 #define __rpte_to_pte(r)	((r).pte)
348 #define __rpte_to_hidx(r,index)	(pte_val(__rpte_to_pte(r)) >> H_PAGE_F_GIX_SHIFT)
349 
350 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift)       \
351 	do {							         \
352 		index = 0;					         \
353 		shift = mmu_psize_defs[psize].shift;		         \
354 
355 #define pte_iterate_hashed_end() } while(0)
356 
357 /*
358  * We expect this to be called only for user addresses or kernel virtual
359  * addresses other than the linear mapping.
360  */
361 #define pte_pagesize_index(mm, addr, pte)	MMU_PAGE_4K
362 
363 #endif /* __real_pte */
364 
365 static inline unsigned long pte_update(struct mm_struct *mm, unsigned long addr,
366 				       pte_t *ptep, unsigned long clr,
367 				       unsigned long set, int huge)
368 {
369 	if (radix_enabled())
370 		return radix__pte_update(mm, addr, ptep, clr, set, huge);
371 	return hash__pte_update(mm, addr, ptep, clr, set, huge);
372 }
373 /*
374  * For hash even if we have _PAGE_ACCESSED = 0, we do a pte_update.
375  * We currently remove entries from the hashtable regardless of whether
376  * the entry was young or dirty.
377  *
378  * We should be more intelligent about this but for the moment we override
379  * these functions and force a tlb flush unconditionally
380  * For radix: H_PAGE_HASHPTE should be zero. Hence we can use the same
381  * function for both hash and radix.
382  */
383 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
384 					      unsigned long addr, pte_t *ptep)
385 {
386 	unsigned long old;
387 
388 	if ((pte_raw(*ptep) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
389 		return 0;
390 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
391 	return (old & _PAGE_ACCESSED) != 0;
392 }
393 
394 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
395 #define ptep_test_and_clear_young(__vma, __addr, __ptep)	\
396 ({								\
397 	__ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
398 })
399 
400 /*
401  * On Book3S CPUs, clearing the accessed bit without a TLB flush
402  * doesn't cause data corruption. [ It could cause incorrect
403  * page aging and the (mistaken) reclaim of hot pages, but the
404  * chance of that should be relatively low. ]
405  *
406  * So as a performance optimization don't flush the TLB when
407  * clearing the accessed bit, it will eventually be flushed by
408  * a context switch or a VM operation anyway. [ In the rare
409  * event of it not getting flushed for a long time the delay
410  * shouldn't really matter because there's no real memory
411  * pressure for swapout to react to. ]
412  */
413 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
414 #define ptep_clear_flush_young ptep_test_and_clear_young
415 
416 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
417 #define pmdp_clear_flush_young pmdp_test_and_clear_young
418 
419 static inline int __pte_write(pte_t pte)
420 {
421 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_WRITE));
422 }
423 
424 #ifdef CONFIG_NUMA_BALANCING
425 #define pte_savedwrite pte_savedwrite
426 static inline bool pte_savedwrite(pte_t pte)
427 {
428 	/*
429 	 * Saved write ptes are prot none ptes that doesn't have
430 	 * privileged bit sit. We mark prot none as one which has
431 	 * present and pviliged bit set and RWX cleared. To mark
432 	 * protnone which used to have _PAGE_WRITE set we clear
433 	 * the privileged bit.
434 	 */
435 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_RWX | _PAGE_PRIVILEGED));
436 }
437 #else
438 #define pte_savedwrite pte_savedwrite
439 static inline bool pte_savedwrite(pte_t pte)
440 {
441 	return false;
442 }
443 #endif
444 
445 static inline int pte_write(pte_t pte)
446 {
447 	return __pte_write(pte) || pte_savedwrite(pte);
448 }
449 
450 static inline int pte_read(pte_t pte)
451 {
452 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_READ));
453 }
454 
455 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
456 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
457 				      pte_t *ptep)
458 {
459 	if (__pte_write(*ptep))
460 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 0);
461 	else if (unlikely(pte_savedwrite(*ptep)))
462 		pte_update(mm, addr, ptep, 0, _PAGE_PRIVILEGED, 0);
463 }
464 
465 #define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
466 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
467 					   unsigned long addr, pte_t *ptep)
468 {
469 	/*
470 	 * We should not find protnone for hugetlb, but this complete the
471 	 * interface.
472 	 */
473 	if (__pte_write(*ptep))
474 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 1);
475 	else if (unlikely(pte_savedwrite(*ptep)))
476 		pte_update(mm, addr, ptep, 0, _PAGE_PRIVILEGED, 1);
477 }
478 
479 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
480 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
481 				       unsigned long addr, pte_t *ptep)
482 {
483 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
484 	return __pte(old);
485 }
486 
487 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
488 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
489 					    unsigned long addr,
490 					    pte_t *ptep, int full)
491 {
492 	if (full && radix_enabled()) {
493 		/*
494 		 * We know that this is a full mm pte clear and
495 		 * hence can be sure there is no parallel set_pte.
496 		 */
497 		return radix__ptep_get_and_clear_full(mm, addr, ptep, full);
498 	}
499 	return ptep_get_and_clear(mm, addr, ptep);
500 }
501 
502 
503 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
504 			     pte_t * ptep)
505 {
506 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
507 }
508 
509 static inline int pte_dirty(pte_t pte)
510 {
511 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DIRTY));
512 }
513 
514 static inline int pte_young(pte_t pte)
515 {
516 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_ACCESSED));
517 }
518 
519 static inline int pte_special(pte_t pte)
520 {
521 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SPECIAL));
522 }
523 
524 static inline bool pte_exec(pte_t pte)
525 {
526 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_EXEC));
527 }
528 
529 
530 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
531 static inline bool pte_soft_dirty(pte_t pte)
532 {
533 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SOFT_DIRTY));
534 }
535 
536 static inline pte_t pte_mksoft_dirty(pte_t pte)
537 {
538 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SOFT_DIRTY));
539 }
540 
541 static inline pte_t pte_clear_soft_dirty(pte_t pte)
542 {
543 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SOFT_DIRTY));
544 }
545 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
546 
547 #ifdef CONFIG_NUMA_BALANCING
548 static inline int pte_protnone(pte_t pte)
549 {
550 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE | _PAGE_RWX)) ==
551 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
552 }
553 
554 #define pte_mk_savedwrite pte_mk_savedwrite
555 static inline pte_t pte_mk_savedwrite(pte_t pte)
556 {
557 	/*
558 	 * Used by Autonuma subsystem to preserve the write bit
559 	 * while marking the pte PROT_NONE. Only allow this
560 	 * on PROT_NONE pte
561 	 */
562 	VM_BUG_ON((pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_RWX | _PAGE_PRIVILEGED)) !=
563 		  cpu_to_be64(_PAGE_PRESENT | _PAGE_PRIVILEGED));
564 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_PRIVILEGED));
565 }
566 
567 #define pte_clear_savedwrite pte_clear_savedwrite
568 static inline pte_t pte_clear_savedwrite(pte_t pte)
569 {
570 	/*
571 	 * Used by KSM subsystem to make a protnone pte readonly.
572 	 */
573 	VM_BUG_ON(!pte_protnone(pte));
574 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PRIVILEGED));
575 }
576 #else
577 #define pte_clear_savedwrite pte_clear_savedwrite
578 static inline pte_t pte_clear_savedwrite(pte_t pte)
579 {
580 	VM_WARN_ON(1);
581 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
582 }
583 #endif /* CONFIG_NUMA_BALANCING */
584 
585 static inline bool pte_hw_valid(pte_t pte)
586 {
587 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE)) ==
588 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
589 }
590 
591 static inline int pte_present(pte_t pte)
592 {
593 	/*
594 	 * A pte is considerent present if _PAGE_PRESENT is set.
595 	 * We also need to consider the pte present which is marked
596 	 * invalid during ptep_set_access_flags. Hence we look for _PAGE_INVALID
597 	 * if we find _PAGE_PRESENT cleared.
598 	 */
599 
600 	if (pte_hw_valid(pte))
601 		return true;
602 	return (pte_raw(pte) & cpu_to_be64(_PAGE_INVALID | _PAGE_PTE)) ==
603 		cpu_to_be64(_PAGE_INVALID | _PAGE_PTE);
604 }
605 
606 #ifdef CONFIG_PPC_MEM_KEYS
607 extern bool arch_pte_access_permitted(u64 pte, bool write, bool execute);
608 #else
609 static inline bool arch_pte_access_permitted(u64 pte, bool write, bool execute)
610 {
611 	return true;
612 }
613 #endif /* CONFIG_PPC_MEM_KEYS */
614 
615 static inline bool pte_user(pte_t pte)
616 {
617 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_PRIVILEGED));
618 }
619 
620 #define pte_access_permitted pte_access_permitted
621 static inline bool pte_access_permitted(pte_t pte, bool write)
622 {
623 	/*
624 	 * _PAGE_READ is needed for any access and will be
625 	 * cleared for PROT_NONE
626 	 */
627 	if (!pte_present(pte) || !pte_user(pte) || !pte_read(pte))
628 		return false;
629 
630 	if (write && !pte_write(pte))
631 		return false;
632 
633 	return arch_pte_access_permitted(pte_val(pte), write, 0);
634 }
635 
636 /*
637  * Conversion functions: convert a page and protection to a page entry,
638  * and a page entry and page directory to the page they refer to.
639  *
640  * Even if PTEs can be unsigned long long, a PFN is always an unsigned
641  * long for now.
642  */
643 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot)
644 {
645 	VM_BUG_ON(pfn >> (64 - PAGE_SHIFT));
646 	VM_BUG_ON((pfn << PAGE_SHIFT) & ~PTE_RPN_MASK);
647 
648 	return __pte(((pte_basic_t)pfn << PAGE_SHIFT) | pgprot_val(pgprot) | _PAGE_PTE);
649 }
650 
651 static inline unsigned long pte_pfn(pte_t pte)
652 {
653 	return (pte_val(pte) & PTE_RPN_MASK) >> PAGE_SHIFT;
654 }
655 
656 /* Generic modifiers for PTE bits */
657 static inline pte_t pte_wrprotect(pte_t pte)
658 {
659 	if (unlikely(pte_savedwrite(pte)))
660 		return pte_clear_savedwrite(pte);
661 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
662 }
663 
664 static inline pte_t pte_exprotect(pte_t pte)
665 {
666 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_EXEC));
667 }
668 
669 static inline pte_t pte_mkclean(pte_t pte)
670 {
671 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_DIRTY));
672 }
673 
674 static inline pte_t pte_mkold(pte_t pte)
675 {
676 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_ACCESSED));
677 }
678 
679 static inline pte_t pte_mkexec(pte_t pte)
680 {
681 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_EXEC));
682 }
683 
684 static inline pte_t pte_mkwrite(pte_t pte)
685 {
686 	/*
687 	 * write implies read, hence set both
688 	 */
689 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_RW));
690 }
691 
692 static inline pte_t pte_mkdirty(pte_t pte)
693 {
694 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_DIRTY | _PAGE_SOFT_DIRTY));
695 }
696 
697 static inline pte_t pte_mkyoung(pte_t pte)
698 {
699 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_ACCESSED));
700 }
701 
702 static inline pte_t pte_mkspecial(pte_t pte)
703 {
704 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL));
705 }
706 
707 static inline pte_t pte_mkhuge(pte_t pte)
708 {
709 	return pte;
710 }
711 
712 static inline pte_t pte_mkdevmap(pte_t pte)
713 {
714 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL | _PAGE_DEVMAP));
715 }
716 
717 static inline pte_t pte_mkprivileged(pte_t pte)
718 {
719 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PRIVILEGED));
720 }
721 
722 static inline pte_t pte_mkuser(pte_t pte)
723 {
724 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_PRIVILEGED));
725 }
726 
727 /*
728  * This is potentially called with a pmd as the argument, in which case it's not
729  * safe to check _PAGE_DEVMAP unless we also confirm that _PAGE_PTE is set.
730  * That's because the bit we use for _PAGE_DEVMAP is not reserved for software
731  * use in page directory entries (ie. non-ptes).
732  */
733 static inline int pte_devmap(pte_t pte)
734 {
735 	u64 mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
736 
737 	return (pte_raw(pte) & mask) == mask;
738 }
739 
740 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
741 {
742 	/* FIXME!! check whether this need to be a conditional */
743 	return __pte_raw((pte_raw(pte) & cpu_to_be64(_PAGE_CHG_MASK)) |
744 			 cpu_to_be64(pgprot_val(newprot)));
745 }
746 
747 /* Encode and de-code a swap entry */
748 #define MAX_SWAPFILES_CHECK() do { \
749 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
750 	/*							\
751 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
752 	 * We filter HPTEFLAGS on set_pte.			\
753 	 */							\
754 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
755 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_SOFT_DIRTY);	\
756 	} while (0)
757 
758 #define SWP_TYPE_BITS 5
759 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
760 				& ((1UL << SWP_TYPE_BITS) - 1))
761 #define __swp_offset(x)		(((x).val & PTE_RPN_MASK) >> PAGE_SHIFT)
762 #define __swp_entry(type, offset)	((swp_entry_t) { \
763 				((type) << _PAGE_BIT_SWAP_TYPE) \
764 				| (((offset) << PAGE_SHIFT) & PTE_RPN_MASK)})
765 /*
766  * swp_entry_t must be independent of pte bits. We build a swp_entry_t from
767  * swap type and offset we get from swap and convert that to pte to find a
768  * matching pte in linux page table.
769  * Clear bits not found in swap entries here.
770  */
771 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val((pte)) & ~_PAGE_PTE })
772 #define __swp_entry_to_pte(x)	__pte((x).val | _PAGE_PTE)
773 #define __pmd_to_swp_entry(pmd)	(__pte_to_swp_entry(pmd_pte(pmd)))
774 #define __swp_entry_to_pmd(x)	(pte_pmd(__swp_entry_to_pte(x)))
775 
776 #ifdef CONFIG_MEM_SOFT_DIRTY
777 #define _PAGE_SWP_SOFT_DIRTY   (1UL << (SWP_TYPE_BITS + _PAGE_BIT_SWAP_TYPE))
778 #else
779 #define _PAGE_SWP_SOFT_DIRTY	0UL
780 #endif /* CONFIG_MEM_SOFT_DIRTY */
781 
782 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
783 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
784 {
785 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
786 }
787 
788 static inline bool pte_swp_soft_dirty(pte_t pte)
789 {
790 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
791 }
792 
793 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
794 {
795 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_SOFT_DIRTY));
796 }
797 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
798 
799 static inline bool check_pte_access(unsigned long access, unsigned long ptev)
800 {
801 	/*
802 	 * This check for _PAGE_RWX and _PAGE_PRESENT bits
803 	 */
804 	if (access & ~ptev)
805 		return false;
806 	/*
807 	 * This check for access to privilege space
808 	 */
809 	if ((access & _PAGE_PRIVILEGED) != (ptev & _PAGE_PRIVILEGED))
810 		return false;
811 
812 	return true;
813 }
814 /*
815  * Generic functions with hash/radix callbacks
816  */
817 
818 static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
819 					   pte_t *ptep, pte_t entry,
820 					   unsigned long address,
821 					   int psize)
822 {
823 	if (radix_enabled())
824 		return radix__ptep_set_access_flags(vma, ptep, entry,
825 						    address, psize);
826 	return hash__ptep_set_access_flags(ptep, entry);
827 }
828 
829 #define __HAVE_ARCH_PTE_SAME
830 static inline int pte_same(pte_t pte_a, pte_t pte_b)
831 {
832 	if (radix_enabled())
833 		return radix__pte_same(pte_a, pte_b);
834 	return hash__pte_same(pte_a, pte_b);
835 }
836 
837 static inline int pte_none(pte_t pte)
838 {
839 	if (radix_enabled())
840 		return radix__pte_none(pte);
841 	return hash__pte_none(pte);
842 }
843 
844 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
845 				pte_t *ptep, pte_t pte, int percpu)
846 {
847 
848 	VM_WARN_ON(!(pte_raw(pte) & cpu_to_be64(_PAGE_PTE)));
849 	/*
850 	 * Keep the _PAGE_PTE added till we are sure we handle _PAGE_PTE
851 	 * in all the callers.
852 	 */
853 	pte = __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PTE));
854 
855 	if (radix_enabled())
856 		return radix__set_pte_at(mm, addr, ptep, pte, percpu);
857 	return hash__set_pte_at(mm, addr, ptep, pte, percpu);
858 }
859 
860 #define _PAGE_CACHE_CTL	(_PAGE_SAO | _PAGE_NON_IDEMPOTENT | _PAGE_TOLERANT)
861 
862 #define pgprot_noncached pgprot_noncached
863 static inline pgprot_t pgprot_noncached(pgprot_t prot)
864 {
865 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
866 			_PAGE_NON_IDEMPOTENT);
867 }
868 
869 #define pgprot_noncached_wc pgprot_noncached_wc
870 static inline pgprot_t pgprot_noncached_wc(pgprot_t prot)
871 {
872 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
873 			_PAGE_TOLERANT);
874 }
875 
876 #define pgprot_cached pgprot_cached
877 static inline pgprot_t pgprot_cached(pgprot_t prot)
878 {
879 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL));
880 }
881 
882 #define pgprot_writecombine pgprot_writecombine
883 static inline pgprot_t pgprot_writecombine(pgprot_t prot)
884 {
885 	return pgprot_noncached_wc(prot);
886 }
887 /*
888  * check a pte mapping have cache inhibited property
889  */
890 static inline bool pte_ci(pte_t pte)
891 {
892 	__be64 pte_v = pte_raw(pte);
893 
894 	if (((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_TOLERANT)) ||
895 	    ((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_NON_IDEMPOTENT)))
896 		return true;
897 	return false;
898 }
899 
900 static inline void pmd_clear(pmd_t *pmdp)
901 {
902 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
903 		/*
904 		 * Don't use this if we can possibly have a hash page table
905 		 * entry mapping this.
906 		 */
907 		WARN_ON((pmd_val(*pmdp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
908 	}
909 	*pmdp = __pmd(0);
910 }
911 
912 static inline int pmd_none(pmd_t pmd)
913 {
914 	return !pmd_raw(pmd);
915 }
916 
917 static inline int pmd_present(pmd_t pmd)
918 {
919 	/*
920 	 * A pmd is considerent present if _PAGE_PRESENT is set.
921 	 * We also need to consider the pmd present which is marked
922 	 * invalid during a split. Hence we look for _PAGE_INVALID
923 	 * if we find _PAGE_PRESENT cleared.
924 	 */
925 	if (pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID))
926 		return true;
927 
928 	return false;
929 }
930 
931 static inline int pmd_is_serializing(pmd_t pmd)
932 {
933 	/*
934 	 * If the pmd is undergoing a split, the _PAGE_PRESENT bit is clear
935 	 * and _PAGE_INVALID is set (see pmd_present, pmdp_invalidate).
936 	 *
937 	 * This condition may also occur when flushing a pmd while flushing
938 	 * it (see ptep_modify_prot_start), so callers must ensure this
939 	 * case is fine as well.
940 	 */
941 	if ((pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID)) ==
942 						cpu_to_be64(_PAGE_INVALID))
943 		return true;
944 
945 	return false;
946 }
947 
948 static inline int pmd_bad(pmd_t pmd)
949 {
950 	if (radix_enabled())
951 		return radix__pmd_bad(pmd);
952 	return hash__pmd_bad(pmd);
953 }
954 
955 static inline void pud_clear(pud_t *pudp)
956 {
957 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
958 		/*
959 		 * Don't use this if we can possibly have a hash page table
960 		 * entry mapping this.
961 		 */
962 		WARN_ON((pud_val(*pudp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
963 	}
964 	*pudp = __pud(0);
965 }
966 
967 static inline int pud_none(pud_t pud)
968 {
969 	return !pud_raw(pud);
970 }
971 
972 static inline int pud_present(pud_t pud)
973 {
974 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PRESENT));
975 }
976 
977 extern struct page *pud_page(pud_t pud);
978 extern struct page *pmd_page(pmd_t pmd);
979 static inline pte_t pud_pte(pud_t pud)
980 {
981 	return __pte_raw(pud_raw(pud));
982 }
983 
984 static inline pud_t pte_pud(pte_t pte)
985 {
986 	return __pud_raw(pte_raw(pte));
987 }
988 #define pud_write(pud)		pte_write(pud_pte(pud))
989 
990 static inline int pud_bad(pud_t pud)
991 {
992 	if (radix_enabled())
993 		return radix__pud_bad(pud);
994 	return hash__pud_bad(pud);
995 }
996 
997 #define pud_access_permitted pud_access_permitted
998 static inline bool pud_access_permitted(pud_t pud, bool write)
999 {
1000 	return pte_access_permitted(pud_pte(pud), write);
1001 }
1002 
1003 #define __p4d_raw(x)	((p4d_t) { __pgd_raw(x) })
1004 static inline __be64 p4d_raw(p4d_t x)
1005 {
1006 	return pgd_raw(x.pgd);
1007 }
1008 
1009 #define p4d_write(p4d)		pte_write(p4d_pte(p4d))
1010 
1011 static inline void p4d_clear(p4d_t *p4dp)
1012 {
1013 	*p4dp = __p4d(0);
1014 }
1015 
1016 static inline int p4d_none(p4d_t p4d)
1017 {
1018 	return !p4d_raw(p4d);
1019 }
1020 
1021 static inline int p4d_present(p4d_t p4d)
1022 {
1023 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PRESENT));
1024 }
1025 
1026 static inline pte_t p4d_pte(p4d_t p4d)
1027 {
1028 	return __pte_raw(p4d_raw(p4d));
1029 }
1030 
1031 static inline p4d_t pte_p4d(pte_t pte)
1032 {
1033 	return __p4d_raw(pte_raw(pte));
1034 }
1035 
1036 static inline int p4d_bad(p4d_t p4d)
1037 {
1038 	if (radix_enabled())
1039 		return radix__p4d_bad(p4d);
1040 	return hash__p4d_bad(p4d);
1041 }
1042 
1043 #define p4d_access_permitted p4d_access_permitted
1044 static inline bool p4d_access_permitted(p4d_t p4d, bool write)
1045 {
1046 	return pte_access_permitted(p4d_pte(p4d), write);
1047 }
1048 
1049 extern struct page *p4d_page(p4d_t p4d);
1050 
1051 /* Pointers in the page table tree are physical addresses */
1052 #define __pgtable_ptr_val(ptr)	__pa(ptr)
1053 
1054 static inline pud_t *p4d_pgtable(p4d_t p4d)
1055 {
1056 	return (pud_t *)__va(p4d_val(p4d) & ~P4D_MASKED_BITS);
1057 }
1058 
1059 static inline pmd_t *pud_pgtable(pud_t pud)
1060 {
1061 	return (pmd_t *)__va(pud_val(pud) & ~PUD_MASKED_BITS);
1062 }
1063 
1064 #define pte_ERROR(e) \
1065 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
1066 #define pmd_ERROR(e) \
1067 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
1068 #define pud_ERROR(e) \
1069 	pr_err("%s:%d: bad pud %08lx.\n", __FILE__, __LINE__, pud_val(e))
1070 #define pgd_ERROR(e) \
1071 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
1072 
1073 static inline int map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot)
1074 {
1075 	if (radix_enabled()) {
1076 #if defined(CONFIG_PPC_RADIX_MMU) && defined(DEBUG_VM)
1077 		unsigned long page_size = 1 << mmu_psize_defs[mmu_io_psize].shift;
1078 		WARN((page_size != PAGE_SIZE), "I/O page size != PAGE_SIZE");
1079 #endif
1080 		return radix__map_kernel_page(ea, pa, prot, PAGE_SIZE);
1081 	}
1082 	return hash__map_kernel_page(ea, pa, prot);
1083 }
1084 
1085 void unmap_kernel_page(unsigned long va);
1086 
1087 static inline int __meminit vmemmap_create_mapping(unsigned long start,
1088 						   unsigned long page_size,
1089 						   unsigned long phys)
1090 {
1091 	if (radix_enabled())
1092 		return radix__vmemmap_create_mapping(start, page_size, phys);
1093 	return hash__vmemmap_create_mapping(start, page_size, phys);
1094 }
1095 
1096 #ifdef CONFIG_MEMORY_HOTPLUG
1097 static inline void vmemmap_remove_mapping(unsigned long start,
1098 					  unsigned long page_size)
1099 {
1100 	if (radix_enabled())
1101 		return radix__vmemmap_remove_mapping(start, page_size);
1102 	return hash__vmemmap_remove_mapping(start, page_size);
1103 }
1104 #endif
1105 
1106 #ifdef CONFIG_DEBUG_PAGEALLOC
1107 static inline void __kernel_map_pages(struct page *page, int numpages, int enable)
1108 {
1109 	if (radix_enabled())
1110 		radix__kernel_map_pages(page, numpages, enable);
1111 	else
1112 		hash__kernel_map_pages(page, numpages, enable);
1113 }
1114 #endif
1115 
1116 static inline pte_t pmd_pte(pmd_t pmd)
1117 {
1118 	return __pte_raw(pmd_raw(pmd));
1119 }
1120 
1121 static inline pmd_t pte_pmd(pte_t pte)
1122 {
1123 	return __pmd_raw(pte_raw(pte));
1124 }
1125 
1126 static inline pte_t *pmdp_ptep(pmd_t *pmd)
1127 {
1128 	return (pte_t *)pmd;
1129 }
1130 #define pmd_pfn(pmd)		pte_pfn(pmd_pte(pmd))
1131 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
1132 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
1133 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
1134 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
1135 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
1136 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
1137 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
1138 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
1139 #define pmd_mk_savedwrite(pmd)	pte_pmd(pte_mk_savedwrite(pmd_pte(pmd)))
1140 #define pmd_clear_savedwrite(pmd)	pte_pmd(pte_clear_savedwrite(pmd_pte(pmd)))
1141 
1142 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1143 #define pmd_soft_dirty(pmd)    pte_soft_dirty(pmd_pte(pmd))
1144 #define pmd_mksoft_dirty(pmd)  pte_pmd(pte_mksoft_dirty(pmd_pte(pmd)))
1145 #define pmd_clear_soft_dirty(pmd) pte_pmd(pte_clear_soft_dirty(pmd_pte(pmd)))
1146 
1147 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1148 #define pmd_swp_mksoft_dirty(pmd)	pte_pmd(pte_swp_mksoft_dirty(pmd_pte(pmd)))
1149 #define pmd_swp_soft_dirty(pmd)		pte_swp_soft_dirty(pmd_pte(pmd))
1150 #define pmd_swp_clear_soft_dirty(pmd)	pte_pmd(pte_swp_clear_soft_dirty(pmd_pte(pmd)))
1151 #endif
1152 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
1153 
1154 #ifdef CONFIG_NUMA_BALANCING
1155 static inline int pmd_protnone(pmd_t pmd)
1156 {
1157 	return pte_protnone(pmd_pte(pmd));
1158 }
1159 #endif /* CONFIG_NUMA_BALANCING */
1160 
1161 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
1162 #define __pmd_write(pmd)	__pte_write(pmd_pte(pmd))
1163 #define pmd_savedwrite(pmd)	pte_savedwrite(pmd_pte(pmd))
1164 
1165 #define pmd_access_permitted pmd_access_permitted
1166 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1167 {
1168 	/*
1169 	 * pmdp_invalidate sets this combination (which is not caught by
1170 	 * !pte_present() check in pte_access_permitted), to prevent
1171 	 * lock-free lookups, as part of the serialize_against_pte_lookup()
1172 	 * synchronisation.
1173 	 *
1174 	 * This also catches the case where the PTE's hardware PRESENT bit is
1175 	 * cleared while TLB is flushed, which is suboptimal but should not
1176 	 * be frequent.
1177 	 */
1178 	if (pmd_is_serializing(pmd))
1179 		return false;
1180 
1181 	return pte_access_permitted(pmd_pte(pmd), write);
1182 }
1183 
1184 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1185 extern pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot);
1186 extern pmd_t mk_pmd(struct page *page, pgprot_t pgprot);
1187 extern pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot);
1188 extern void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1189 		       pmd_t *pmdp, pmd_t pmd);
1190 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1191 					unsigned long addr, pmd_t *pmd)
1192 {
1193 }
1194 
1195 extern int hash__has_transparent_hugepage(void);
1196 static inline int has_transparent_hugepage(void)
1197 {
1198 	if (radix_enabled())
1199 		return radix__has_transparent_hugepage();
1200 	return hash__has_transparent_hugepage();
1201 }
1202 #define has_transparent_hugepage has_transparent_hugepage
1203 
1204 static inline unsigned long
1205 pmd_hugepage_update(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp,
1206 		    unsigned long clr, unsigned long set)
1207 {
1208 	if (radix_enabled())
1209 		return radix__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1210 	return hash__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1211 }
1212 
1213 /*
1214  * returns true for pmd migration entries, THP, devmap, hugetlb
1215  * But compile time dependent on THP config
1216  */
1217 static inline int pmd_large(pmd_t pmd)
1218 {
1219 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1220 }
1221 
1222 /*
1223  * For radix we should always find H_PAGE_HASHPTE zero. Hence
1224  * the below will work for radix too
1225  */
1226 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
1227 					      unsigned long addr, pmd_t *pmdp)
1228 {
1229 	unsigned long old;
1230 
1231 	if ((pmd_raw(*pmdp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1232 		return 0;
1233 	old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
1234 	return ((old & _PAGE_ACCESSED) != 0);
1235 }
1236 
1237 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1238 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1239 				      pmd_t *pmdp)
1240 {
1241 	if (__pmd_write((*pmdp)))
1242 		pmd_hugepage_update(mm, addr, pmdp, _PAGE_WRITE, 0);
1243 	else if (unlikely(pmd_savedwrite(*pmdp)))
1244 		pmd_hugepage_update(mm, addr, pmdp, 0, _PAGE_PRIVILEGED);
1245 }
1246 
1247 /*
1248  * Only returns true for a THP. False for pmd migration entry.
1249  * We also need to return true when we come across a pte that
1250  * in between a thp split. While splitting THP, we mark the pmd
1251  * invalid (pmdp_invalidate()) before we set it with pte page
1252  * address. A pmd_trans_huge() check against a pmd entry during that time
1253  * should return true.
1254  * We should not call this on a hugetlb entry. We should check for HugeTLB
1255  * entry using vma->vm_flags
1256  * The page table walk rule is explained in Documentation/vm/transhuge.rst
1257  */
1258 static inline int pmd_trans_huge(pmd_t pmd)
1259 {
1260 	if (!pmd_present(pmd))
1261 		return false;
1262 
1263 	if (radix_enabled())
1264 		return radix__pmd_trans_huge(pmd);
1265 	return hash__pmd_trans_huge(pmd);
1266 }
1267 
1268 #define __HAVE_ARCH_PMD_SAME
1269 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1270 {
1271 	if (radix_enabled())
1272 		return radix__pmd_same(pmd_a, pmd_b);
1273 	return hash__pmd_same(pmd_a, pmd_b);
1274 }
1275 
1276 static inline pmd_t __pmd_mkhuge(pmd_t pmd)
1277 {
1278 	if (radix_enabled())
1279 		return radix__pmd_mkhuge(pmd);
1280 	return hash__pmd_mkhuge(pmd);
1281 }
1282 
1283 /*
1284  * pfn_pmd return a pmd_t that can be used as pmd pte entry.
1285  */
1286 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1287 {
1288 #ifdef CONFIG_DEBUG_VM
1289 	if (radix_enabled())
1290 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE)) == 0);
1291 	else
1292 		WARN_ON((pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE)) !=
1293 			cpu_to_be64(_PAGE_PTE | H_PAGE_THP_HUGE));
1294 #endif
1295 	return pmd;
1296 }
1297 
1298 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1299 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1300 				 unsigned long address, pmd_t *pmdp,
1301 				 pmd_t entry, int dirty);
1302 
1303 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1304 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1305 				     unsigned long address, pmd_t *pmdp);
1306 
1307 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1308 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1309 					    unsigned long addr, pmd_t *pmdp)
1310 {
1311 	if (radix_enabled())
1312 		return radix__pmdp_huge_get_and_clear(mm, addr, pmdp);
1313 	return hash__pmdp_huge_get_and_clear(mm, addr, pmdp);
1314 }
1315 
1316 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1317 					unsigned long address, pmd_t *pmdp)
1318 {
1319 	if (radix_enabled())
1320 		return radix__pmdp_collapse_flush(vma, address, pmdp);
1321 	return hash__pmdp_collapse_flush(vma, address, pmdp);
1322 }
1323 #define pmdp_collapse_flush pmdp_collapse_flush
1324 
1325 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1326 pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1327 				   unsigned long addr,
1328 				   pmd_t *pmdp, int full);
1329 
1330 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1331 static inline void pgtable_trans_huge_deposit(struct mm_struct *mm,
1332 					      pmd_t *pmdp, pgtable_t pgtable)
1333 {
1334 	if (radix_enabled())
1335 		return radix__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1336 	return hash__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1337 }
1338 
1339 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1340 static inline pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm,
1341 						    pmd_t *pmdp)
1342 {
1343 	if (radix_enabled())
1344 		return radix__pgtable_trans_huge_withdraw(mm, pmdp);
1345 	return hash__pgtable_trans_huge_withdraw(mm, pmdp);
1346 }
1347 
1348 #define __HAVE_ARCH_PMDP_INVALIDATE
1349 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
1350 			     pmd_t *pmdp);
1351 
1352 #define pmd_move_must_withdraw pmd_move_must_withdraw
1353 struct spinlock;
1354 extern int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
1355 				  struct spinlock *old_pmd_ptl,
1356 				  struct vm_area_struct *vma);
1357 /*
1358  * Hash translation mode use the deposited table to store hash pte
1359  * slot information.
1360  */
1361 #define arch_needs_pgtable_deposit arch_needs_pgtable_deposit
1362 static inline bool arch_needs_pgtable_deposit(void)
1363 {
1364 	if (radix_enabled())
1365 		return false;
1366 	return true;
1367 }
1368 extern void serialize_against_pte_lookup(struct mm_struct *mm);
1369 
1370 
1371 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
1372 {
1373 	if (radix_enabled())
1374 		return radix__pmd_mkdevmap(pmd);
1375 	return hash__pmd_mkdevmap(pmd);
1376 }
1377 
1378 static inline int pmd_devmap(pmd_t pmd)
1379 {
1380 	return pte_devmap(pmd_pte(pmd));
1381 }
1382 
1383 static inline int pud_devmap(pud_t pud)
1384 {
1385 	return 0;
1386 }
1387 
1388 static inline int pgd_devmap(pgd_t pgd)
1389 {
1390 	return 0;
1391 }
1392 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1393 
1394 static inline int pud_pfn(pud_t pud)
1395 {
1396 	/*
1397 	 * Currently all calls to pud_pfn() are gated around a pud_devmap()
1398 	 * check so this should never be used. If it grows another user we
1399 	 * want to know about it.
1400 	 */
1401 	BUILD_BUG();
1402 	return 0;
1403 }
1404 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1405 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1406 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1407 			     pte_t *, pte_t, pte_t);
1408 
1409 /*
1410  * Returns true for a R -> RW upgrade of pte
1411  */
1412 static inline bool is_pte_rw_upgrade(unsigned long old_val, unsigned long new_val)
1413 {
1414 	if (!(old_val & _PAGE_READ))
1415 		return false;
1416 
1417 	if ((!(old_val & _PAGE_WRITE)) && (new_val & _PAGE_WRITE))
1418 		return true;
1419 
1420 	return false;
1421 }
1422 
1423 /*
1424  * Like pmd_huge() and pmd_large(), but works regardless of config options
1425  */
1426 #define pmd_is_leaf pmd_is_leaf
1427 #define pmd_leaf pmd_is_leaf
1428 static inline bool pmd_is_leaf(pmd_t pmd)
1429 {
1430 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1431 }
1432 
1433 #define pud_is_leaf pud_is_leaf
1434 #define pud_leaf pud_is_leaf
1435 static inline bool pud_is_leaf(pud_t pud)
1436 {
1437 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PTE));
1438 }
1439 
1440 #define p4d_is_leaf p4d_is_leaf
1441 #define p4d_leaf p4d_is_leaf
1442 static inline bool p4d_is_leaf(p4d_t p4d)
1443 {
1444 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PTE));
1445 }
1446 
1447 #endif /* __ASSEMBLY__ */
1448 #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */
1449