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 #endif
11 
12 /*
13  * Common bits between hash and Radix page table
14  */
15 #define _PAGE_BIT_SWAP_TYPE	0
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_RWX	(_PAGE_PRIVILEGED | _PAGE_DIRTY |	\
120 				 _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 
154 /* Permission masks used for kernel mappings */
155 #define PAGE_KERNEL	__pgprot(_PAGE_BASE | _PAGE_KERNEL_RW)
156 #define PAGE_KERNEL_NC	__pgprot(_PAGE_BASE_NC | _PAGE_KERNEL_RW | \
157 				 _PAGE_TOLERANT)
158 #define PAGE_KERNEL_NCG	__pgprot(_PAGE_BASE_NC | _PAGE_KERNEL_RW | \
159 				 _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 /*
165  * Protection used for kernel text. We want the debuggers to be able to
166  * set breakpoints anywhere, so don't write protect the kernel text
167  * on platforms where such control is possible.
168  */
169 #if defined(CONFIG_KGDB) || defined(CONFIG_XMON) || defined(CONFIG_BDI_SWITCH) || \
170 	defined(CONFIG_KPROBES) || defined(CONFIG_DYNAMIC_FTRACE)
171 #define PAGE_KERNEL_TEXT	PAGE_KERNEL_X
172 #else
173 #define PAGE_KERNEL_TEXT	PAGE_KERNEL_ROX
174 #endif
175 
176 /* Make modules code happy. We don't set RO yet */
177 #define PAGE_KERNEL_EXEC	PAGE_KERNEL_X
178 #define PAGE_AGP		(PAGE_KERNEL_NC)
179 
180 #ifndef __ASSEMBLY__
181 /*
182  * page table defines
183  */
184 extern unsigned long __pte_index_size;
185 extern unsigned long __pmd_index_size;
186 extern unsigned long __pud_index_size;
187 extern unsigned long __pgd_index_size;
188 extern unsigned long __pud_cache_index;
189 #define PTE_INDEX_SIZE  __pte_index_size
190 #define PMD_INDEX_SIZE  __pmd_index_size
191 #define PUD_INDEX_SIZE  __pud_index_size
192 #define PGD_INDEX_SIZE  __pgd_index_size
193 /* pmd table use page table fragments */
194 #define PMD_CACHE_INDEX  0
195 #define PUD_CACHE_INDEX __pud_cache_index
196 /*
197  * Because of use of pte fragments and THP, size of page table
198  * are not always derived out of index size above.
199  */
200 extern unsigned long __pte_table_size;
201 extern unsigned long __pmd_table_size;
202 extern unsigned long __pud_table_size;
203 extern unsigned long __pgd_table_size;
204 #define PTE_TABLE_SIZE	__pte_table_size
205 #define PMD_TABLE_SIZE	__pmd_table_size
206 #define PUD_TABLE_SIZE	__pud_table_size
207 #define PGD_TABLE_SIZE	__pgd_table_size
208 
209 extern unsigned long __pmd_val_bits;
210 extern unsigned long __pud_val_bits;
211 extern unsigned long __pgd_val_bits;
212 #define PMD_VAL_BITS	__pmd_val_bits
213 #define PUD_VAL_BITS	__pud_val_bits
214 #define PGD_VAL_BITS	__pgd_val_bits
215 
216 extern unsigned long __pte_frag_nr;
217 #define PTE_FRAG_NR __pte_frag_nr
218 extern unsigned long __pte_frag_size_shift;
219 #define PTE_FRAG_SIZE_SHIFT __pte_frag_size_shift
220 #define PTE_FRAG_SIZE (1UL << PTE_FRAG_SIZE_SHIFT)
221 
222 extern unsigned long __pmd_frag_nr;
223 #define PMD_FRAG_NR __pmd_frag_nr
224 extern unsigned long __pmd_frag_size_shift;
225 #define PMD_FRAG_SIZE_SHIFT __pmd_frag_size_shift
226 #define PMD_FRAG_SIZE (1UL << PMD_FRAG_SIZE_SHIFT)
227 
228 #define PTRS_PER_PTE	(1 << PTE_INDEX_SIZE)
229 #define PTRS_PER_PMD	(1 << PMD_INDEX_SIZE)
230 #define PTRS_PER_PUD	(1 << PUD_INDEX_SIZE)
231 #define PTRS_PER_PGD	(1 << PGD_INDEX_SIZE)
232 
233 /* PMD_SHIFT determines what a second-level page table entry can map */
234 #define PMD_SHIFT	(PAGE_SHIFT + PTE_INDEX_SIZE)
235 #define PMD_SIZE	(1UL << PMD_SHIFT)
236 #define PMD_MASK	(~(PMD_SIZE-1))
237 
238 /* PUD_SHIFT determines what a third-level page table entry can map */
239 #define PUD_SHIFT	(PMD_SHIFT + PMD_INDEX_SIZE)
240 #define PUD_SIZE	(1UL << PUD_SHIFT)
241 #define PUD_MASK	(~(PUD_SIZE-1))
242 
243 /* PGDIR_SHIFT determines what a fourth-level page table entry can map */
244 #define PGDIR_SHIFT	(PUD_SHIFT + PUD_INDEX_SIZE)
245 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
246 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
247 
248 /* Bits to mask out from a PMD to get to the PTE page */
249 #define PMD_MASKED_BITS		0xc0000000000000ffUL
250 /* Bits to mask out from a PUD to get to the PMD page */
251 #define PUD_MASKED_BITS		0xc0000000000000ffUL
252 /* Bits to mask out from a PGD to get to the PUD page */
253 #define P4D_MASKED_BITS		0xc0000000000000ffUL
254 
255 /*
256  * Used as an indicator for rcu callback functions
257  */
258 enum pgtable_index {
259 	PTE_INDEX = 0,
260 	PMD_INDEX,
261 	PUD_INDEX,
262 	PGD_INDEX,
263 	/*
264 	 * Below are used with 4k page size and hugetlb
265 	 */
266 	HTLB_16M_INDEX,
267 	HTLB_16G_INDEX,
268 };
269 
270 extern unsigned long __vmalloc_start;
271 extern unsigned long __vmalloc_end;
272 #define VMALLOC_START	__vmalloc_start
273 #define VMALLOC_END	__vmalloc_end
274 
275 static inline unsigned int ioremap_max_order(void)
276 {
277 	if (radix_enabled())
278 		return PUD_SHIFT;
279 	return 7 + PAGE_SHIFT; /* default from linux/vmalloc.h */
280 }
281 #define IOREMAP_MAX_ORDER ioremap_max_order()
282 
283 extern unsigned long __kernel_virt_start;
284 extern unsigned long __kernel_io_start;
285 extern unsigned long __kernel_io_end;
286 #define KERN_VIRT_START __kernel_virt_start
287 #define KERN_IO_START  __kernel_io_start
288 #define KERN_IO_END __kernel_io_end
289 
290 extern struct page *vmemmap;
291 extern unsigned long pci_io_base;
292 #endif /* __ASSEMBLY__ */
293 
294 #include <asm/book3s/64/hash.h>
295 #include <asm/book3s/64/radix.h>
296 
297 #if H_MAX_PHYSMEM_BITS > R_MAX_PHYSMEM_BITS
298 #define  MAX_PHYSMEM_BITS	H_MAX_PHYSMEM_BITS
299 #else
300 #define  MAX_PHYSMEM_BITS	R_MAX_PHYSMEM_BITS
301 #endif
302 
303 
304 #ifdef CONFIG_PPC_64K_PAGES
305 #include <asm/book3s/64/pgtable-64k.h>
306 #else
307 #include <asm/book3s/64/pgtable-4k.h>
308 #endif
309 
310 #include <asm/barrier.h>
311 /*
312  * IO space itself carved into the PIO region (ISA and PHB IO space) and
313  * the ioremap space
314  *
315  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
316  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
317  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
318  */
319 #define FULL_IO_SIZE	0x80000000ul
320 #define  ISA_IO_BASE	(KERN_IO_START)
321 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
322 #define  PHB_IO_BASE	(ISA_IO_END)
323 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
324 #define IOREMAP_BASE	(PHB_IO_END)
325 #define IOREMAP_START	(ioremap_bot)
326 #define IOREMAP_END	(KERN_IO_END)
327 
328 /* Advertise special mapping type for AGP */
329 #define HAVE_PAGE_AGP
330 
331 #ifndef __ASSEMBLY__
332 
333 /*
334  * This is the default implementation of various PTE accessors, it's
335  * used in all cases except Book3S with 64K pages where we have a
336  * concept of sub-pages
337  */
338 #ifndef __real_pte
339 
340 #define __real_pte(e, p, o)		((real_pte_t){(e)})
341 #define __rpte_to_pte(r)	((r).pte)
342 #define __rpte_to_hidx(r,index)	(pte_val(__rpte_to_pte(r)) >> H_PAGE_F_GIX_SHIFT)
343 
344 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift)       \
345 	do {							         \
346 		index = 0;					         \
347 		shift = mmu_psize_defs[psize].shift;		         \
348 
349 #define pte_iterate_hashed_end() } while(0)
350 
351 /*
352  * We expect this to be called only for user addresses or kernel virtual
353  * addresses other than the linear mapping.
354  */
355 #define pte_pagesize_index(mm, addr, pte)	MMU_PAGE_4K
356 
357 #endif /* __real_pte */
358 
359 static inline unsigned long pte_update(struct mm_struct *mm, unsigned long addr,
360 				       pte_t *ptep, unsigned long clr,
361 				       unsigned long set, int huge)
362 {
363 	if (radix_enabled())
364 		return radix__pte_update(mm, addr, ptep, clr, set, huge);
365 	return hash__pte_update(mm, addr, ptep, clr, set, huge);
366 }
367 /*
368  * For hash even if we have _PAGE_ACCESSED = 0, we do a pte_update.
369  * We currently remove entries from the hashtable regardless of whether
370  * the entry was young or dirty.
371  *
372  * We should be more intelligent about this but for the moment we override
373  * these functions and force a tlb flush unconditionally
374  * For radix: H_PAGE_HASHPTE should be zero. Hence we can use the same
375  * function for both hash and radix.
376  */
377 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
378 					      unsigned long addr, pte_t *ptep)
379 {
380 	unsigned long old;
381 
382 	if ((pte_raw(*ptep) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
383 		return 0;
384 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
385 	return (old & _PAGE_ACCESSED) != 0;
386 }
387 
388 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
389 #define ptep_test_and_clear_young(__vma, __addr, __ptep)	\
390 ({								\
391 	int __r;						\
392 	__r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
393 	__r;							\
394 })
395 
396 static inline int __pte_write(pte_t pte)
397 {
398 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_WRITE));
399 }
400 
401 #ifdef CONFIG_NUMA_BALANCING
402 #define pte_savedwrite pte_savedwrite
403 static inline bool pte_savedwrite(pte_t pte)
404 {
405 	/*
406 	 * Saved write ptes are prot none ptes that doesn't have
407 	 * privileged bit sit. We mark prot none as one which has
408 	 * present and pviliged bit set and RWX cleared. To mark
409 	 * protnone which used to have _PAGE_WRITE set we clear
410 	 * the privileged bit.
411 	 */
412 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_RWX | _PAGE_PRIVILEGED));
413 }
414 #else
415 #define pte_savedwrite pte_savedwrite
416 static inline bool pte_savedwrite(pte_t pte)
417 {
418 	return false;
419 }
420 #endif
421 
422 static inline int pte_write(pte_t pte)
423 {
424 	return __pte_write(pte) || pte_savedwrite(pte);
425 }
426 
427 static inline int pte_read(pte_t pte)
428 {
429 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_READ));
430 }
431 
432 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
433 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
434 				      pte_t *ptep)
435 {
436 	if (__pte_write(*ptep))
437 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 0);
438 	else if (unlikely(pte_savedwrite(*ptep)))
439 		pte_update(mm, addr, ptep, 0, _PAGE_PRIVILEGED, 0);
440 }
441 
442 #define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
443 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
444 					   unsigned long addr, pte_t *ptep)
445 {
446 	/*
447 	 * We should not find protnone for hugetlb, but this complete the
448 	 * interface.
449 	 */
450 	if (__pte_write(*ptep))
451 		pte_update(mm, addr, ptep, _PAGE_WRITE, 0, 1);
452 	else if (unlikely(pte_savedwrite(*ptep)))
453 		pte_update(mm, addr, ptep, 0, _PAGE_PRIVILEGED, 1);
454 }
455 
456 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
457 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
458 				       unsigned long addr, pte_t *ptep)
459 {
460 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
461 	return __pte(old);
462 }
463 
464 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
465 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
466 					    unsigned long addr,
467 					    pte_t *ptep, int full)
468 {
469 	if (full && radix_enabled()) {
470 		/*
471 		 * We know that this is a full mm pte clear and
472 		 * hence can be sure there is no parallel set_pte.
473 		 */
474 		return radix__ptep_get_and_clear_full(mm, addr, ptep, full);
475 	}
476 	return ptep_get_and_clear(mm, addr, ptep);
477 }
478 
479 
480 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
481 			     pte_t * ptep)
482 {
483 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
484 }
485 
486 static inline int pte_dirty(pte_t pte)
487 {
488 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DIRTY));
489 }
490 
491 static inline int pte_young(pte_t pte)
492 {
493 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_ACCESSED));
494 }
495 
496 static inline int pte_special(pte_t pte)
497 {
498 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SPECIAL));
499 }
500 
501 static inline bool pte_exec(pte_t pte)
502 {
503 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_EXEC));
504 }
505 
506 
507 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
508 static inline bool pte_soft_dirty(pte_t pte)
509 {
510 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SOFT_DIRTY));
511 }
512 
513 static inline pte_t pte_mksoft_dirty(pte_t pte)
514 {
515 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SOFT_DIRTY));
516 }
517 
518 static inline pte_t pte_clear_soft_dirty(pte_t pte)
519 {
520 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SOFT_DIRTY));
521 }
522 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
523 
524 #ifdef CONFIG_NUMA_BALANCING
525 static inline int pte_protnone(pte_t pte)
526 {
527 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE | _PAGE_RWX)) ==
528 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
529 }
530 
531 #define pte_mk_savedwrite pte_mk_savedwrite
532 static inline pte_t pte_mk_savedwrite(pte_t pte)
533 {
534 	/*
535 	 * Used by Autonuma subsystem to preserve the write bit
536 	 * while marking the pte PROT_NONE. Only allow this
537 	 * on PROT_NONE pte
538 	 */
539 	VM_BUG_ON((pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_RWX | _PAGE_PRIVILEGED)) !=
540 		  cpu_to_be64(_PAGE_PRESENT | _PAGE_PRIVILEGED));
541 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_PRIVILEGED));
542 }
543 
544 #define pte_clear_savedwrite pte_clear_savedwrite
545 static inline pte_t pte_clear_savedwrite(pte_t pte)
546 {
547 	/*
548 	 * Used by KSM subsystem to make a protnone pte readonly.
549 	 */
550 	VM_BUG_ON(!pte_protnone(pte));
551 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PRIVILEGED));
552 }
553 #else
554 #define pte_clear_savedwrite pte_clear_savedwrite
555 static inline pte_t pte_clear_savedwrite(pte_t pte)
556 {
557 	VM_WARN_ON(1);
558 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
559 }
560 #endif /* CONFIG_NUMA_BALANCING */
561 
562 static inline bool pte_hw_valid(pte_t pte)
563 {
564 	return (pte_raw(pte) & cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE)) ==
565 		cpu_to_be64(_PAGE_PRESENT | _PAGE_PTE);
566 }
567 
568 static inline int pte_present(pte_t pte)
569 {
570 	/*
571 	 * A pte is considerent present if _PAGE_PRESENT is set.
572 	 * We also need to consider the pte present which is marked
573 	 * invalid during ptep_set_access_flags. Hence we look for _PAGE_INVALID
574 	 * if we find _PAGE_PRESENT cleared.
575 	 */
576 
577 	if (pte_hw_valid(pte))
578 		return true;
579 	return (pte_raw(pte) & cpu_to_be64(_PAGE_INVALID | _PAGE_PTE)) ==
580 		cpu_to_be64(_PAGE_INVALID | _PAGE_PTE);
581 }
582 
583 #ifdef CONFIG_PPC_MEM_KEYS
584 extern bool arch_pte_access_permitted(u64 pte, bool write, bool execute);
585 #else
586 static inline bool arch_pte_access_permitted(u64 pte, bool write, bool execute)
587 {
588 	return true;
589 }
590 #endif /* CONFIG_PPC_MEM_KEYS */
591 
592 static inline bool pte_user(pte_t pte)
593 {
594 	return !(pte_raw(pte) & cpu_to_be64(_PAGE_PRIVILEGED));
595 }
596 
597 #define pte_access_permitted pte_access_permitted
598 static inline bool pte_access_permitted(pte_t pte, bool write)
599 {
600 	/*
601 	 * _PAGE_READ is needed for any access and will be
602 	 * cleared for PROT_NONE
603 	 */
604 	if (!pte_present(pte) || !pte_user(pte) || !pte_read(pte))
605 		return false;
606 
607 	if (write && !pte_write(pte))
608 		return false;
609 
610 	return arch_pte_access_permitted(pte_val(pte), write, 0);
611 }
612 
613 /*
614  * Conversion functions: convert a page and protection to a page entry,
615  * and a page entry and page directory to the page they refer to.
616  *
617  * Even if PTEs can be unsigned long long, a PFN is always an unsigned
618  * long for now.
619  */
620 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot)
621 {
622 	VM_BUG_ON(pfn >> (64 - PAGE_SHIFT));
623 	VM_BUG_ON((pfn << PAGE_SHIFT) & ~PTE_RPN_MASK);
624 
625 	return __pte(((pte_basic_t)pfn << PAGE_SHIFT) | pgprot_val(pgprot) | _PAGE_PTE);
626 }
627 
628 static inline unsigned long pte_pfn(pte_t pte)
629 {
630 	return (pte_val(pte) & PTE_RPN_MASK) >> PAGE_SHIFT;
631 }
632 
633 /* Generic modifiers for PTE bits */
634 static inline pte_t pte_wrprotect(pte_t pte)
635 {
636 	if (unlikely(pte_savedwrite(pte)))
637 		return pte_clear_savedwrite(pte);
638 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_WRITE));
639 }
640 
641 static inline pte_t pte_exprotect(pte_t pte)
642 {
643 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_EXEC));
644 }
645 
646 static inline pte_t pte_mkclean(pte_t pte)
647 {
648 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_DIRTY));
649 }
650 
651 static inline pte_t pte_mkold(pte_t pte)
652 {
653 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_ACCESSED));
654 }
655 
656 static inline pte_t pte_mkexec(pte_t pte)
657 {
658 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_EXEC));
659 }
660 
661 static inline pte_t pte_mkwrite(pte_t pte)
662 {
663 	/*
664 	 * write implies read, hence set both
665 	 */
666 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_RW));
667 }
668 
669 static inline pte_t pte_mkdirty(pte_t pte)
670 {
671 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_DIRTY | _PAGE_SOFT_DIRTY));
672 }
673 
674 static inline pte_t pte_mkyoung(pte_t pte)
675 {
676 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_ACCESSED));
677 }
678 
679 static inline pte_t pte_mkspecial(pte_t pte)
680 {
681 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL));
682 }
683 
684 static inline pte_t pte_mkhuge(pte_t pte)
685 {
686 	return pte;
687 }
688 
689 static inline pte_t pte_mkdevmap(pte_t pte)
690 {
691 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SPECIAL | _PAGE_DEVMAP));
692 }
693 
694 static inline pte_t pte_mkprivileged(pte_t pte)
695 {
696 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PRIVILEGED));
697 }
698 
699 static inline pte_t pte_mkuser(pte_t pte)
700 {
701 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_PRIVILEGED));
702 }
703 
704 /*
705  * This is potentially called with a pmd as the argument, in which case it's not
706  * safe to check _PAGE_DEVMAP unless we also confirm that _PAGE_PTE is set.
707  * That's because the bit we use for _PAGE_DEVMAP is not reserved for software
708  * use in page directory entries (ie. non-ptes).
709  */
710 static inline int pte_devmap(pte_t pte)
711 {
712 	u64 mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
713 
714 	return (pte_raw(pte) & mask) == mask;
715 }
716 
717 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
718 {
719 	/* FIXME!! check whether this need to be a conditional */
720 	return __pte_raw((pte_raw(pte) & cpu_to_be64(_PAGE_CHG_MASK)) |
721 			 cpu_to_be64(pgprot_val(newprot)));
722 }
723 
724 /* Encode and de-code a swap entry */
725 #define MAX_SWAPFILES_CHECK() do { \
726 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
727 	/*							\
728 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
729 	 * We filter HPTEFLAGS on set_pte.			\
730 	 */							\
731 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
732 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & _PAGE_SWP_SOFT_DIRTY);	\
733 	} while (0)
734 
735 #define SWP_TYPE_BITS 5
736 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
737 				& ((1UL << SWP_TYPE_BITS) - 1))
738 #define __swp_offset(x)		(((x).val & PTE_RPN_MASK) >> PAGE_SHIFT)
739 #define __swp_entry(type, offset)	((swp_entry_t) { \
740 				((type) << _PAGE_BIT_SWAP_TYPE) \
741 				| (((offset) << PAGE_SHIFT) & PTE_RPN_MASK)})
742 /*
743  * swp_entry_t must be independent of pte bits. We build a swp_entry_t from
744  * swap type and offset we get from swap and convert that to pte to find a
745  * matching pte in linux page table.
746  * Clear bits not found in swap entries here.
747  */
748 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val((pte)) & ~_PAGE_PTE })
749 #define __swp_entry_to_pte(x)	__pte((x).val | _PAGE_PTE)
750 #define __pmd_to_swp_entry(pmd)	(__pte_to_swp_entry(pmd_pte(pmd)))
751 #define __swp_entry_to_pmd(x)	(pte_pmd(__swp_entry_to_pte(x)))
752 
753 #ifdef CONFIG_MEM_SOFT_DIRTY
754 #define _PAGE_SWP_SOFT_DIRTY   (1UL << (SWP_TYPE_BITS + _PAGE_BIT_SWAP_TYPE))
755 #else
756 #define _PAGE_SWP_SOFT_DIRTY	0UL
757 #endif /* CONFIG_MEM_SOFT_DIRTY */
758 
759 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
760 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
761 {
762 	return __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
763 }
764 
765 static inline bool pte_swp_soft_dirty(pte_t pte)
766 {
767 	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_SWP_SOFT_DIRTY));
768 }
769 
770 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
771 {
772 	return __pte_raw(pte_raw(pte) & cpu_to_be64(~_PAGE_SWP_SOFT_DIRTY));
773 }
774 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
775 
776 static inline bool check_pte_access(unsigned long access, unsigned long ptev)
777 {
778 	/*
779 	 * This check for _PAGE_RWX and _PAGE_PRESENT bits
780 	 */
781 	if (access & ~ptev)
782 		return false;
783 	/*
784 	 * This check for access to privilege space
785 	 */
786 	if ((access & _PAGE_PRIVILEGED) != (ptev & _PAGE_PRIVILEGED))
787 		return false;
788 
789 	return true;
790 }
791 /*
792  * Generic functions with hash/radix callbacks
793  */
794 
795 static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
796 					   pte_t *ptep, pte_t entry,
797 					   unsigned long address,
798 					   int psize)
799 {
800 	if (radix_enabled())
801 		return radix__ptep_set_access_flags(vma, ptep, entry,
802 						    address, psize);
803 	return hash__ptep_set_access_flags(ptep, entry);
804 }
805 
806 #define __HAVE_ARCH_PTE_SAME
807 static inline int pte_same(pte_t pte_a, pte_t pte_b)
808 {
809 	if (radix_enabled())
810 		return radix__pte_same(pte_a, pte_b);
811 	return hash__pte_same(pte_a, pte_b);
812 }
813 
814 static inline int pte_none(pte_t pte)
815 {
816 	if (radix_enabled())
817 		return radix__pte_none(pte);
818 	return hash__pte_none(pte);
819 }
820 
821 static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
822 				pte_t *ptep, pte_t pte, int percpu)
823 {
824 
825 	VM_WARN_ON(!(pte_raw(pte) & cpu_to_be64(_PAGE_PTE)));
826 	/*
827 	 * Keep the _PAGE_PTE added till we are sure we handle _PAGE_PTE
828 	 * in all the callers.
829 	 */
830 	pte = __pte_raw(pte_raw(pte) | cpu_to_be64(_PAGE_PTE));
831 
832 	if (radix_enabled())
833 		return radix__set_pte_at(mm, addr, ptep, pte, percpu);
834 	return hash__set_pte_at(mm, addr, ptep, pte, percpu);
835 }
836 
837 #define _PAGE_CACHE_CTL	(_PAGE_SAO | _PAGE_NON_IDEMPOTENT | _PAGE_TOLERANT)
838 
839 #define pgprot_noncached pgprot_noncached
840 static inline pgprot_t pgprot_noncached(pgprot_t prot)
841 {
842 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
843 			_PAGE_NON_IDEMPOTENT);
844 }
845 
846 #define pgprot_noncached_wc pgprot_noncached_wc
847 static inline pgprot_t pgprot_noncached_wc(pgprot_t prot)
848 {
849 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL) |
850 			_PAGE_TOLERANT);
851 }
852 
853 #define pgprot_cached pgprot_cached
854 static inline pgprot_t pgprot_cached(pgprot_t prot)
855 {
856 	return __pgprot((pgprot_val(prot) & ~_PAGE_CACHE_CTL));
857 }
858 
859 #define pgprot_writecombine pgprot_writecombine
860 static inline pgprot_t pgprot_writecombine(pgprot_t prot)
861 {
862 	return pgprot_noncached_wc(prot);
863 }
864 /*
865  * check a pte mapping have cache inhibited property
866  */
867 static inline bool pte_ci(pte_t pte)
868 {
869 	__be64 pte_v = pte_raw(pte);
870 
871 	if (((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_TOLERANT)) ||
872 	    ((pte_v & cpu_to_be64(_PAGE_CACHE_CTL)) == cpu_to_be64(_PAGE_NON_IDEMPOTENT)))
873 		return true;
874 	return false;
875 }
876 
877 static inline void pmd_clear(pmd_t *pmdp)
878 {
879 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
880 		/*
881 		 * Don't use this if we can possibly have a hash page table
882 		 * entry mapping this.
883 		 */
884 		WARN_ON((pmd_val(*pmdp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
885 	}
886 	*pmdp = __pmd(0);
887 }
888 
889 static inline int pmd_none(pmd_t pmd)
890 {
891 	return !pmd_raw(pmd);
892 }
893 
894 static inline int pmd_present(pmd_t pmd)
895 {
896 	/*
897 	 * A pmd is considerent present if _PAGE_PRESENT is set.
898 	 * We also need to consider the pmd present which is marked
899 	 * invalid during a split. Hence we look for _PAGE_INVALID
900 	 * if we find _PAGE_PRESENT cleared.
901 	 */
902 	if (pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID))
903 		return true;
904 
905 	return false;
906 }
907 
908 static inline int pmd_is_serializing(pmd_t pmd)
909 {
910 	/*
911 	 * If the pmd is undergoing a split, the _PAGE_PRESENT bit is clear
912 	 * and _PAGE_INVALID is set (see pmd_present, pmdp_invalidate).
913 	 *
914 	 * This condition may also occur when flushing a pmd while flushing
915 	 * it (see ptep_modify_prot_start), so callers must ensure this
916 	 * case is fine as well.
917 	 */
918 	if ((pmd_raw(pmd) & cpu_to_be64(_PAGE_PRESENT | _PAGE_INVALID)) ==
919 						cpu_to_be64(_PAGE_INVALID))
920 		return true;
921 
922 	return false;
923 }
924 
925 static inline int pmd_bad(pmd_t pmd)
926 {
927 	if (radix_enabled())
928 		return radix__pmd_bad(pmd);
929 	return hash__pmd_bad(pmd);
930 }
931 
932 static inline void pud_clear(pud_t *pudp)
933 {
934 	if (IS_ENABLED(CONFIG_DEBUG_VM) && !radix_enabled()) {
935 		/*
936 		 * Don't use this if we can possibly have a hash page table
937 		 * entry mapping this.
938 		 */
939 		WARN_ON((pud_val(*pudp) & (H_PAGE_HASHPTE | _PAGE_PTE)) == (H_PAGE_HASHPTE | _PAGE_PTE));
940 	}
941 	*pudp = __pud(0);
942 }
943 
944 static inline int pud_none(pud_t pud)
945 {
946 	return !pud_raw(pud);
947 }
948 
949 static inline int pud_present(pud_t pud)
950 {
951 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PRESENT));
952 }
953 
954 extern struct page *pud_page(pud_t pud);
955 extern struct page *pmd_page(pmd_t pmd);
956 static inline pte_t pud_pte(pud_t pud)
957 {
958 	return __pte_raw(pud_raw(pud));
959 }
960 
961 static inline pud_t pte_pud(pte_t pte)
962 {
963 	return __pud_raw(pte_raw(pte));
964 }
965 #define pud_write(pud)		pte_write(pud_pte(pud))
966 
967 static inline int pud_bad(pud_t pud)
968 {
969 	if (radix_enabled())
970 		return radix__pud_bad(pud);
971 	return hash__pud_bad(pud);
972 }
973 
974 #define pud_access_permitted pud_access_permitted
975 static inline bool pud_access_permitted(pud_t pud, bool write)
976 {
977 	return pte_access_permitted(pud_pte(pud), write);
978 }
979 
980 #define __p4d_raw(x)	((p4d_t) { __pgd_raw(x) })
981 static inline __be64 p4d_raw(p4d_t x)
982 {
983 	return pgd_raw(x.pgd);
984 }
985 
986 #define p4d_write(p4d)		pte_write(p4d_pte(p4d))
987 
988 static inline void p4d_clear(p4d_t *p4dp)
989 {
990 	*p4dp = __p4d(0);
991 }
992 
993 static inline int p4d_none(p4d_t p4d)
994 {
995 	return !p4d_raw(p4d);
996 }
997 
998 static inline int p4d_present(p4d_t p4d)
999 {
1000 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PRESENT));
1001 }
1002 
1003 static inline pte_t p4d_pte(p4d_t p4d)
1004 {
1005 	return __pte_raw(p4d_raw(p4d));
1006 }
1007 
1008 static inline p4d_t pte_p4d(pte_t pte)
1009 {
1010 	return __p4d_raw(pte_raw(pte));
1011 }
1012 
1013 static inline int p4d_bad(p4d_t p4d)
1014 {
1015 	if (radix_enabled())
1016 		return radix__p4d_bad(p4d);
1017 	return hash__p4d_bad(p4d);
1018 }
1019 
1020 #define p4d_access_permitted p4d_access_permitted
1021 static inline bool p4d_access_permitted(p4d_t p4d, bool write)
1022 {
1023 	return pte_access_permitted(p4d_pte(p4d), write);
1024 }
1025 
1026 extern struct page *p4d_page(p4d_t p4d);
1027 
1028 /* Pointers in the page table tree are physical addresses */
1029 #define __pgtable_ptr_val(ptr)	__pa(ptr)
1030 
1031 #define pud_page_vaddr(pud)	__va(pud_val(pud) & ~PUD_MASKED_BITS)
1032 #define p4d_page_vaddr(p4d)	__va(p4d_val(p4d) & ~P4D_MASKED_BITS)
1033 
1034 #define pte_ERROR(e) \
1035 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
1036 #define pmd_ERROR(e) \
1037 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
1038 #define pud_ERROR(e) \
1039 	pr_err("%s:%d: bad pud %08lx.\n", __FILE__, __LINE__, pud_val(e))
1040 #define pgd_ERROR(e) \
1041 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
1042 
1043 static inline int map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot)
1044 {
1045 	if (radix_enabled()) {
1046 #if defined(CONFIG_PPC_RADIX_MMU) && defined(DEBUG_VM)
1047 		unsigned long page_size = 1 << mmu_psize_defs[mmu_io_psize].shift;
1048 		WARN((page_size != PAGE_SIZE), "I/O page size != PAGE_SIZE");
1049 #endif
1050 		return radix__map_kernel_page(ea, pa, prot, PAGE_SIZE);
1051 	}
1052 	return hash__map_kernel_page(ea, pa, prot);
1053 }
1054 
1055 static inline int __meminit vmemmap_create_mapping(unsigned long start,
1056 						   unsigned long page_size,
1057 						   unsigned long phys)
1058 {
1059 	if (radix_enabled())
1060 		return radix__vmemmap_create_mapping(start, page_size, phys);
1061 	return hash__vmemmap_create_mapping(start, page_size, phys);
1062 }
1063 
1064 #ifdef CONFIG_MEMORY_HOTPLUG
1065 static inline void vmemmap_remove_mapping(unsigned long start,
1066 					  unsigned long page_size)
1067 {
1068 	if (radix_enabled())
1069 		return radix__vmemmap_remove_mapping(start, page_size);
1070 	return hash__vmemmap_remove_mapping(start, page_size);
1071 }
1072 #endif
1073 
1074 static inline pte_t pmd_pte(pmd_t pmd)
1075 {
1076 	return __pte_raw(pmd_raw(pmd));
1077 }
1078 
1079 static inline pmd_t pte_pmd(pte_t pte)
1080 {
1081 	return __pmd_raw(pte_raw(pte));
1082 }
1083 
1084 static inline pte_t *pmdp_ptep(pmd_t *pmd)
1085 {
1086 	return (pte_t *)pmd;
1087 }
1088 #define pmd_pfn(pmd)		pte_pfn(pmd_pte(pmd))
1089 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
1090 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
1091 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
1092 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
1093 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
1094 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
1095 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
1096 #define pmd_mkwrite(pmd)	pte_pmd(pte_mkwrite(pmd_pte(pmd)))
1097 #define pmd_mk_savedwrite(pmd)	pte_pmd(pte_mk_savedwrite(pmd_pte(pmd)))
1098 #define pmd_clear_savedwrite(pmd)	pte_pmd(pte_clear_savedwrite(pmd_pte(pmd)))
1099 
1100 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1101 #define pmd_soft_dirty(pmd)    pte_soft_dirty(pmd_pte(pmd))
1102 #define pmd_mksoft_dirty(pmd)  pte_pmd(pte_mksoft_dirty(pmd_pte(pmd)))
1103 #define pmd_clear_soft_dirty(pmd) pte_pmd(pte_clear_soft_dirty(pmd_pte(pmd)))
1104 
1105 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1106 #define pmd_swp_mksoft_dirty(pmd)	pte_pmd(pte_swp_mksoft_dirty(pmd_pte(pmd)))
1107 #define pmd_swp_soft_dirty(pmd)		pte_swp_soft_dirty(pmd_pte(pmd))
1108 #define pmd_swp_clear_soft_dirty(pmd)	pte_pmd(pte_swp_clear_soft_dirty(pmd_pte(pmd)))
1109 #endif
1110 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
1111 
1112 #ifdef CONFIG_NUMA_BALANCING
1113 static inline int pmd_protnone(pmd_t pmd)
1114 {
1115 	return pte_protnone(pmd_pte(pmd));
1116 }
1117 #endif /* CONFIG_NUMA_BALANCING */
1118 
1119 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
1120 #define __pmd_write(pmd)	__pte_write(pmd_pte(pmd))
1121 #define pmd_savedwrite(pmd)	pte_savedwrite(pmd_pte(pmd))
1122 
1123 #define pmd_access_permitted pmd_access_permitted
1124 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1125 {
1126 	/*
1127 	 * pmdp_invalidate sets this combination (which is not caught by
1128 	 * !pte_present() check in pte_access_permitted), to prevent
1129 	 * lock-free lookups, as part of the serialize_against_pte_lookup()
1130 	 * synchronisation.
1131 	 *
1132 	 * This also catches the case where the PTE's hardware PRESENT bit is
1133 	 * cleared while TLB is flushed, which is suboptimal but should not
1134 	 * be frequent.
1135 	 */
1136 	if (pmd_is_serializing(pmd))
1137 		return false;
1138 
1139 	return pte_access_permitted(pmd_pte(pmd), write);
1140 }
1141 
1142 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1143 extern pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot);
1144 extern pmd_t mk_pmd(struct page *page, pgprot_t pgprot);
1145 extern pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot);
1146 extern void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1147 		       pmd_t *pmdp, pmd_t pmd);
1148 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1149 					unsigned long addr, pmd_t *pmd)
1150 {
1151 }
1152 
1153 extern int hash__has_transparent_hugepage(void);
1154 static inline int has_transparent_hugepage(void)
1155 {
1156 	if (radix_enabled())
1157 		return radix__has_transparent_hugepage();
1158 	return hash__has_transparent_hugepage();
1159 }
1160 #define has_transparent_hugepage has_transparent_hugepage
1161 
1162 static inline unsigned long
1163 pmd_hugepage_update(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp,
1164 		    unsigned long clr, unsigned long set)
1165 {
1166 	if (radix_enabled())
1167 		return radix__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1168 	return hash__pmd_hugepage_update(mm, addr, pmdp, clr, set);
1169 }
1170 
1171 /*
1172  * returns true for pmd migration entries, THP, devmap, hugetlb
1173  * But compile time dependent on THP config
1174  */
1175 static inline int pmd_large(pmd_t pmd)
1176 {
1177 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1178 }
1179 
1180 /*
1181  * For radix we should always find H_PAGE_HASHPTE zero. Hence
1182  * the below will work for radix too
1183  */
1184 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
1185 					      unsigned long addr, pmd_t *pmdp)
1186 {
1187 	unsigned long old;
1188 
1189 	if ((pmd_raw(*pmdp) & cpu_to_be64(_PAGE_ACCESSED | H_PAGE_HASHPTE)) == 0)
1190 		return 0;
1191 	old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
1192 	return ((old & _PAGE_ACCESSED) != 0);
1193 }
1194 
1195 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1196 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
1197 				      pmd_t *pmdp)
1198 {
1199 	if (__pmd_write((*pmdp)))
1200 		pmd_hugepage_update(mm, addr, pmdp, _PAGE_WRITE, 0);
1201 	else if (unlikely(pmd_savedwrite(*pmdp)))
1202 		pmd_hugepage_update(mm, addr, pmdp, 0, _PAGE_PRIVILEGED);
1203 }
1204 
1205 /*
1206  * Only returns true for a THP. False for pmd migration entry.
1207  * We also need to return true when we come across a pte that
1208  * in between a thp split. While splitting THP, we mark the pmd
1209  * invalid (pmdp_invalidate()) before we set it with pte page
1210  * address. A pmd_trans_huge() check against a pmd entry during that time
1211  * should return true.
1212  * We should not call this on a hugetlb entry. We should check for HugeTLB
1213  * entry using vma->vm_flags
1214  * The page table walk rule is explained in Documentation/vm/transhuge.rst
1215  */
1216 static inline int pmd_trans_huge(pmd_t pmd)
1217 {
1218 	if (!pmd_present(pmd))
1219 		return false;
1220 
1221 	if (radix_enabled())
1222 		return radix__pmd_trans_huge(pmd);
1223 	return hash__pmd_trans_huge(pmd);
1224 }
1225 
1226 #define __HAVE_ARCH_PMD_SAME
1227 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1228 {
1229 	if (radix_enabled())
1230 		return radix__pmd_same(pmd_a, pmd_b);
1231 	return hash__pmd_same(pmd_a, pmd_b);
1232 }
1233 
1234 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1235 {
1236 	if (radix_enabled())
1237 		return radix__pmd_mkhuge(pmd);
1238 	return hash__pmd_mkhuge(pmd);
1239 }
1240 
1241 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1242 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1243 				 unsigned long address, pmd_t *pmdp,
1244 				 pmd_t entry, int dirty);
1245 
1246 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1247 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1248 				     unsigned long address, pmd_t *pmdp);
1249 
1250 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1251 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1252 					    unsigned long addr, pmd_t *pmdp)
1253 {
1254 	if (radix_enabled())
1255 		return radix__pmdp_huge_get_and_clear(mm, addr, pmdp);
1256 	return hash__pmdp_huge_get_and_clear(mm, addr, pmdp);
1257 }
1258 
1259 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1260 					unsigned long address, pmd_t *pmdp)
1261 {
1262 	if (radix_enabled())
1263 		return radix__pmdp_collapse_flush(vma, address, pmdp);
1264 	return hash__pmdp_collapse_flush(vma, address, pmdp);
1265 }
1266 #define pmdp_collapse_flush pmdp_collapse_flush
1267 
1268 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1269 pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1270 				   unsigned long addr,
1271 				   pmd_t *pmdp, int full);
1272 
1273 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1274 static inline void pgtable_trans_huge_deposit(struct mm_struct *mm,
1275 					      pmd_t *pmdp, pgtable_t pgtable)
1276 {
1277 	if (radix_enabled())
1278 		return radix__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1279 	return hash__pgtable_trans_huge_deposit(mm, pmdp, pgtable);
1280 }
1281 
1282 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1283 static inline pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm,
1284 						    pmd_t *pmdp)
1285 {
1286 	if (radix_enabled())
1287 		return radix__pgtable_trans_huge_withdraw(mm, pmdp);
1288 	return hash__pgtable_trans_huge_withdraw(mm, pmdp);
1289 }
1290 
1291 #define __HAVE_ARCH_PMDP_INVALIDATE
1292 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
1293 			     pmd_t *pmdp);
1294 
1295 #define pmd_move_must_withdraw pmd_move_must_withdraw
1296 struct spinlock;
1297 extern int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
1298 				  struct spinlock *old_pmd_ptl,
1299 				  struct vm_area_struct *vma);
1300 /*
1301  * Hash translation mode use the deposited table to store hash pte
1302  * slot information.
1303  */
1304 #define arch_needs_pgtable_deposit arch_needs_pgtable_deposit
1305 static inline bool arch_needs_pgtable_deposit(void)
1306 {
1307 	if (radix_enabled())
1308 		return false;
1309 	return true;
1310 }
1311 extern void serialize_against_pte_lookup(struct mm_struct *mm);
1312 
1313 
1314 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
1315 {
1316 	if (radix_enabled())
1317 		return radix__pmd_mkdevmap(pmd);
1318 	return hash__pmd_mkdevmap(pmd);
1319 }
1320 
1321 static inline int pmd_devmap(pmd_t pmd)
1322 {
1323 	return pte_devmap(pmd_pte(pmd));
1324 }
1325 
1326 static inline int pud_devmap(pud_t pud)
1327 {
1328 	return 0;
1329 }
1330 
1331 static inline int pgd_devmap(pgd_t pgd)
1332 {
1333 	return 0;
1334 }
1335 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1336 
1337 static inline int pud_pfn(pud_t pud)
1338 {
1339 	/*
1340 	 * Currently all calls to pud_pfn() are gated around a pud_devmap()
1341 	 * check so this should never be used. If it grows another user we
1342 	 * want to know about it.
1343 	 */
1344 	BUILD_BUG();
1345 	return 0;
1346 }
1347 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1348 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1349 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1350 			     pte_t *, pte_t, pte_t);
1351 
1352 /*
1353  * Returns true for a R -> RW upgrade of pte
1354  */
1355 static inline bool is_pte_rw_upgrade(unsigned long old_val, unsigned long new_val)
1356 {
1357 	if (!(old_val & _PAGE_READ))
1358 		return false;
1359 
1360 	if ((!(old_val & _PAGE_WRITE)) && (new_val & _PAGE_WRITE))
1361 		return true;
1362 
1363 	return false;
1364 }
1365 
1366 /*
1367  * Like pmd_huge() and pmd_large(), but works regardless of config options
1368  */
1369 #define pmd_is_leaf pmd_is_leaf
1370 #define pmd_leaf pmd_is_leaf
1371 static inline bool pmd_is_leaf(pmd_t pmd)
1372 {
1373 	return !!(pmd_raw(pmd) & cpu_to_be64(_PAGE_PTE));
1374 }
1375 
1376 #define pud_is_leaf pud_is_leaf
1377 #define pud_leaf pud_is_leaf
1378 static inline bool pud_is_leaf(pud_t pud)
1379 {
1380 	return !!(pud_raw(pud) & cpu_to_be64(_PAGE_PTE));
1381 }
1382 
1383 #define p4d_is_leaf p4d_is_leaf
1384 #define p4d_leaf p4d_is_leaf
1385 static inline bool p4d_is_leaf(p4d_t p4d)
1386 {
1387 	return !!(p4d_raw(p4d) & cpu_to_be64(_PAGE_PTE));
1388 }
1389 
1390 #endif /* __ASSEMBLY__ */
1391 #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */
1392