1 #ifndef _ASM_POWERPC_PGTABLE_PPC64_H_
2 #define _ASM_POWERPC_PGTABLE_PPC64_H_
4 * This file contains the functions and defines necessary to modify and use
5 * the ppc64 hashed page table.
8 #ifdef CONFIG_PPC_64K_PAGES
9 #include <asm/pgtable-ppc64-64k.h>
11 #include <asm/pgtable-ppc64-4k.h>
13 #include <asm/barrier.h>
15 #define FIRST_USER_ADDRESS 0
18 * Size of EA range mapped by our pagetables.
20 #define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
21 PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
22 #define PGTABLE_RANGE (ASM_CONST(1) << PGTABLE_EADDR_SIZE)
24 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
25 #define PMD_CACHE_INDEX (PMD_INDEX_SIZE + 1)
27 #define PMD_CACHE_INDEX PMD_INDEX_SIZE
30 * Define the address range of the kernel non-linear virtual area
33 #ifdef CONFIG_PPC_BOOK3E
34 #define KERN_VIRT_START ASM_CONST(0x8000000000000000)
36 #define KERN_VIRT_START ASM_CONST(0xD000000000000000)
38 #define KERN_VIRT_SIZE ASM_CONST(0x0000100000000000)
41 * The vmalloc space starts at the beginning of that region, and
42 * occupies half of it on hash CPUs and a quarter of it on Book3E
43 * (we keep a quarter for the virtual memmap)
45 #define VMALLOC_START KERN_VIRT_START
46 #ifdef CONFIG_PPC_BOOK3E
47 #define VMALLOC_SIZE (KERN_VIRT_SIZE >> 2)
49 #define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1)
51 #define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE)
54 * The second half of the kernel virtual space is used for IO mappings,
55 * it's itself carved into the PIO region (ISA and PHB IO space) and
58 * ISA_IO_BASE = KERN_IO_START, 64K reserved area
59 * PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
60 * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
62 #define KERN_IO_START (KERN_VIRT_START + (KERN_VIRT_SIZE >> 1))
63 #define FULL_IO_SIZE 0x80000000ul
64 #define ISA_IO_BASE (KERN_IO_START)
65 #define ISA_IO_END (KERN_IO_START + 0x10000ul)
66 #define PHB_IO_BASE (ISA_IO_END)
67 #define PHB_IO_END (KERN_IO_START + FULL_IO_SIZE)
68 #define IOREMAP_BASE (PHB_IO_END)
69 #define IOREMAP_END (KERN_VIRT_START + KERN_VIRT_SIZE)
75 #define REGION_SHIFT 60UL
76 #define REGION_MASK (0xfUL << REGION_SHIFT)
77 #define REGION_ID(ea) (((unsigned long)(ea)) >> REGION_SHIFT)
79 #define VMALLOC_REGION_ID (REGION_ID(VMALLOC_START))
80 #define KERNEL_REGION_ID (REGION_ID(PAGE_OFFSET))
81 #define VMEMMAP_REGION_ID (0xfUL) /* Server only */
82 #define USER_REGION_ID (0UL)
85 * Defines the address of the vmemap area, in its own region on
86 * hash table CPUs and after the vmalloc space on Book3E
88 #ifdef CONFIG_PPC_BOOK3E
89 #define VMEMMAP_BASE VMALLOC_END
90 #define VMEMMAP_END KERN_IO_START
92 #define VMEMMAP_BASE (VMEMMAP_REGION_ID << REGION_SHIFT)
94 #define vmemmap ((struct page *)VMEMMAP_BASE)
98 * Include the PTE bits definitions
100 #ifdef CONFIG_PPC_BOOK3S
101 #include <asm/pte-hash64.h>
103 #include <asm/pte-book3e.h>
105 #include <asm/pte-common.h>
107 #ifdef CONFIG_PPC_MM_SLICES
108 #define HAVE_ARCH_UNMAPPED_AREA
109 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
110 #endif /* CONFIG_PPC_MM_SLICES */
115 * This is the default implementation of various PTE accessors, it's
116 * used in all cases except Book3S with 64K pages where we have a
117 * concept of sub-pages
121 #ifdef STRICT_MM_TYPECHECKS
122 #define __real_pte(e,p) ((real_pte_t){(e)})
123 #define __rpte_to_pte(r) ((r).pte)
125 #define __real_pte(e,p) (e)
126 #define __rpte_to_pte(r) (__pte(r))
128 #define __rpte_to_hidx(r,index) (pte_val(__rpte_to_pte(r)) >> 12)
130 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift) \
133 shift = mmu_psize_defs[psize].shift; \
135 #define pte_iterate_hashed_end() } while(0)
137 #ifdef CONFIG_PPC_HAS_HASH_64K
138 #define pte_pagesize_index(mm, addr, pte) get_slice_psize(mm, addr)
140 #define pte_pagesize_index(mm, addr, pte) MMU_PAGE_4K
143 #endif /* __real_pte */
146 /* pte_clear moved to later in this file */
148 #define PMD_BAD_BITS (PTE_TABLE_SIZE-1)
149 #define PUD_BAD_BITS (PMD_TABLE_SIZE-1)
151 #define pmd_set(pmdp, pmdval) (pmd_val(*(pmdp)) = (pmdval))
152 #define pmd_none(pmd) (!pmd_val(pmd))
153 #define pmd_bad(pmd) (!is_kernel_addr(pmd_val(pmd)) \
154 || (pmd_val(pmd) & PMD_BAD_BITS))
155 #define pmd_present(pmd) (pmd_val(pmd) != 0)
156 #define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0)
157 #define pmd_page_vaddr(pmd) (pmd_val(pmd) & ~PMD_MASKED_BITS)
158 extern struct page
*pmd_page(pmd_t pmd
);
160 #define pud_set(pudp, pudval) (pud_val(*(pudp)) = (pudval))
161 #define pud_none(pud) (!pud_val(pud))
162 #define pud_bad(pud) (!is_kernel_addr(pud_val(pud)) \
163 || (pud_val(pud) & PUD_BAD_BITS))
164 #define pud_present(pud) (pud_val(pud) != 0)
165 #define pud_clear(pudp) (pud_val(*(pudp)) = 0)
166 #define pud_page_vaddr(pud) (pud_val(pud) & ~PUD_MASKED_BITS)
167 #define pud_page(pud) virt_to_page(pud_page_vaddr(pud))
169 #define pgd_set(pgdp, pudp) ({pgd_val(*(pgdp)) = (unsigned long)(pudp);})
172 * Find an entry in a page-table-directory. We combine the address region
173 * (the high order N bits) and the pgd portion of the address.
175 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
177 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
179 #define pmd_offset(pudp,addr) \
180 (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
182 #define pte_offset_kernel(dir,addr) \
183 (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
185 #define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr))
186 #define pte_unmap(pte) do { } while(0)
188 /* to find an entry in a kernel page-table-directory */
189 /* This now only contains the vmalloc pages */
190 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
191 extern void hpte_need_flush(struct mm_struct
*mm
, unsigned long addr
,
192 pte_t
*ptep
, unsigned long pte
, int huge
);
194 /* Atomic PTE updates */
195 static inline unsigned long pte_update(struct mm_struct
*mm
,
197 pte_t
*ptep
, unsigned long clr
,
200 #ifdef PTE_ATOMIC_UPDATES
201 unsigned long old
, tmp
;
203 __asm__
__volatile__(
204 "1: ldarx %0,0,%3 # pte_update\n\
210 : "=&r" (old
), "=&r" (tmp
), "=m" (*ptep
)
211 : "r" (ptep
), "r" (clr
), "m" (*ptep
), "i" (_PAGE_BUSY
)
214 unsigned long old
= pte_val(*ptep
);
215 *ptep
= __pte(old
& ~clr
);
217 /* huge pages use the old page table lock */
219 assert_pte_locked(mm
, addr
);
221 #ifdef CONFIG_PPC_STD_MMU_64
222 if (old
& _PAGE_HASHPTE
)
223 hpte_need_flush(mm
, addr
, ptep
, old
, huge
);
229 static inline int __ptep_test_and_clear_young(struct mm_struct
*mm
,
230 unsigned long addr
, pte_t
*ptep
)
234 if ((pte_val(*ptep
) & (_PAGE_ACCESSED
| _PAGE_HASHPTE
)) == 0)
236 old
= pte_update(mm
, addr
, ptep
, _PAGE_ACCESSED
, 0);
237 return (old
& _PAGE_ACCESSED
) != 0;
239 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
240 #define ptep_test_and_clear_young(__vma, __addr, __ptep) \
243 __r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
247 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
248 static inline void ptep_set_wrprotect(struct mm_struct
*mm
, unsigned long addr
,
252 if ((pte_val(*ptep
) & _PAGE_RW
) == 0)
255 pte_update(mm
, addr
, ptep
, _PAGE_RW
, 0);
258 static inline void huge_ptep_set_wrprotect(struct mm_struct
*mm
,
259 unsigned long addr
, pte_t
*ptep
)
261 if ((pte_val(*ptep
) & _PAGE_RW
) == 0)
264 pte_update(mm
, addr
, ptep
, _PAGE_RW
, 1);
268 * We currently remove entries from the hashtable regardless of whether
269 * the entry was young or dirty. The generic routines only flush if the
270 * entry was young or dirty which is not good enough.
272 * We should be more intelligent about this but for the moment we override
273 * these functions and force a tlb flush unconditionally
275 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
276 #define ptep_clear_flush_young(__vma, __address, __ptep) \
278 int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
283 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
284 static inline pte_t
ptep_get_and_clear(struct mm_struct
*mm
,
285 unsigned long addr
, pte_t
*ptep
)
287 unsigned long old
= pte_update(mm
, addr
, ptep
, ~0UL, 0);
291 static inline void pte_clear(struct mm_struct
*mm
, unsigned long addr
,
294 pte_update(mm
, addr
, ptep
, ~0UL, 0);
298 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
299 * function doesn't need to flush the hash entry
301 static inline void __ptep_set_access_flags(pte_t
*ptep
, pte_t entry
)
303 unsigned long bits
= pte_val(entry
) &
304 (_PAGE_DIRTY
| _PAGE_ACCESSED
| _PAGE_RW
| _PAGE_EXEC
);
306 #ifdef PTE_ATOMIC_UPDATES
307 unsigned long old
, tmp
;
309 __asm__
__volatile__(
316 :"=&r" (old
), "=&r" (tmp
), "=m" (*ptep
)
317 :"r" (bits
), "r" (ptep
), "m" (*ptep
), "i" (_PAGE_BUSY
)
320 unsigned long old
= pte_val(*ptep
);
321 *ptep
= __pte(old
| bits
);
325 #define __HAVE_ARCH_PTE_SAME
326 #define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
328 #define pte_ERROR(e) \
329 printk("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
330 #define pmd_ERROR(e) \
331 printk("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
332 #define pgd_ERROR(e) \
333 printk("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
335 /* Encode and de-code a swap entry */
336 #define __swp_type(entry) (((entry).val >> 1) & 0x3f)
337 #define __swp_offset(entry) ((entry).val >> 8)
338 #define __swp_entry(type, offset) ((swp_entry_t){((type)<< 1)|((offset)<<8)})
339 #define __pte_to_swp_entry(pte) ((swp_entry_t){pte_val(pte) >> PTE_RPN_SHIFT})
340 #define __swp_entry_to_pte(x) ((pte_t) { (x).val << PTE_RPN_SHIFT })
341 #define pte_to_pgoff(pte) (pte_val(pte) >> PTE_RPN_SHIFT)
342 #define pgoff_to_pte(off) ((pte_t) {((off) << PTE_RPN_SHIFT)|_PAGE_FILE})
343 #define PTE_FILE_MAX_BITS (BITS_PER_LONG - PTE_RPN_SHIFT)
345 void pgtable_cache_add(unsigned shift
, void (*ctor
)(void *));
346 void pgtable_cache_init(void);
347 #endif /* __ASSEMBLY__ */
350 * THP pages can't be special. So use the _PAGE_SPECIAL
352 #define _PAGE_SPLITTING _PAGE_SPECIAL
355 * We need to differentiate between explicit huge page and THP huge
356 * page, since THP huge page also need to track real subpage details
358 #define _PAGE_THP_HUGE _PAGE_4K_PFN
361 * set of bits not changed in pmd_modify.
363 #define _HPAGE_CHG_MASK (PTE_RPN_MASK | _PAGE_HPTEFLAGS | \
364 _PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_SPLITTING | \
369 * The linux hugepage PMD now include the pmd entries followed by the address
370 * to the stashed pgtable_t. The stashed pgtable_t contains the hpte bits.
371 * [ 1 bit secondary | 3 bit hidx | 1 bit valid | 000]. We use one byte per
372 * each HPTE entry. With 16MB hugepage and 64K HPTE we need 256 entries and
373 * with 4K HPTE we need 4096 entries. Both will fit in a 4K pgtable_t.
375 * The last three bits are intentionally left to zero. This memory location
376 * are also used as normal page PTE pointers. So if we have any pointers
377 * left around while we collapse a hugepage, we need to make sure
378 * _PAGE_PRESENT and _PAGE_FILE bits of that are zero when we look at them
380 static inline unsigned int hpte_valid(unsigned char *hpte_slot_array
, int index
)
382 return (hpte_slot_array
[index
] >> 3) & 0x1;
385 static inline unsigned int hpte_hash_index(unsigned char *hpte_slot_array
,
388 return hpte_slot_array
[index
] >> 4;
391 static inline void mark_hpte_slot_valid(unsigned char *hpte_slot_array
,
392 unsigned int index
, unsigned int hidx
)
394 hpte_slot_array
[index
] = hidx
<< 4 | 0x1 << 3;
397 struct page
*realmode_pfn_to_page(unsigned long pfn
);
399 static inline char *get_hpte_slot_array(pmd_t
*pmdp
)
402 * The hpte hindex is stored in the pgtable whose address is in the
403 * second half of the PMD
405 * Order this load with the test for pmd_trans_huge in the caller
408 return *(char **)(pmdp
+ PTRS_PER_PMD
);
413 extern void hpte_do_hugepage_flush(struct mm_struct
*mm
, unsigned long addr
,
415 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
416 extern pmd_t
pfn_pmd(unsigned long pfn
, pgprot_t pgprot
);
417 extern pmd_t
mk_pmd(struct page
*page
, pgprot_t pgprot
);
418 extern pmd_t
pmd_modify(pmd_t pmd
, pgprot_t newprot
);
419 extern void set_pmd_at(struct mm_struct
*mm
, unsigned long addr
,
420 pmd_t
*pmdp
, pmd_t pmd
);
421 extern void update_mmu_cache_pmd(struct vm_area_struct
*vma
, unsigned long addr
,
424 static inline int pmd_trans_huge(pmd_t pmd
)
427 * leaf pte for huge page, bottom two bits != 00
429 return (pmd_val(pmd
) & 0x3) && (pmd_val(pmd
) & _PAGE_THP_HUGE
);
432 static inline int pmd_large(pmd_t pmd
)
435 * leaf pte for huge page, bottom two bits != 00
437 if (pmd_trans_huge(pmd
))
438 return pmd_val(pmd
) & _PAGE_PRESENT
;
442 static inline int pmd_trans_splitting(pmd_t pmd
)
444 if (pmd_trans_huge(pmd
))
445 return pmd_val(pmd
) & _PAGE_SPLITTING
;
449 extern int has_transparent_hugepage(void);
450 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
452 static inline pte_t
pmd_pte(pmd_t pmd
)
454 return __pte(pmd_val(pmd
));
457 static inline pmd_t
pte_pmd(pte_t pte
)
459 return __pmd(pte_val(pte
));
462 static inline pte_t
*pmdp_ptep(pmd_t
*pmd
)
467 #define pmd_pfn(pmd) pte_pfn(pmd_pte(pmd))
468 #define pmd_young(pmd) pte_young(pmd_pte(pmd))
469 #define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
470 #define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
471 #define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
472 #define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
473 #define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
475 #define __HAVE_ARCH_PMD_WRITE
476 #define pmd_write(pmd) pte_write(pmd_pte(pmd))
478 static inline pmd_t
pmd_mkhuge(pmd_t pmd
)
480 /* Do nothing, mk_pmd() does this part. */
484 static inline pmd_t
pmd_mknotpresent(pmd_t pmd
)
486 pmd_val(pmd
) &= ~_PAGE_PRESENT
;
490 static inline pmd_t
pmd_mksplitting(pmd_t pmd
)
492 pmd_val(pmd
) |= _PAGE_SPLITTING
;
496 #define __HAVE_ARCH_PMD_SAME
497 static inline int pmd_same(pmd_t pmd_a
, pmd_t pmd_b
)
499 return (((pmd_val(pmd_a
) ^ pmd_val(pmd_b
)) & ~_PAGE_HPTEFLAGS
) == 0);
502 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
503 extern int pmdp_set_access_flags(struct vm_area_struct
*vma
,
504 unsigned long address
, pmd_t
*pmdp
,
505 pmd_t entry
, int dirty
);
507 extern unsigned long pmd_hugepage_update(struct mm_struct
*mm
,
509 pmd_t
*pmdp
, unsigned long clr
);
511 static inline int __pmdp_test_and_clear_young(struct mm_struct
*mm
,
512 unsigned long addr
, pmd_t
*pmdp
)
516 if ((pmd_val(*pmdp
) & (_PAGE_ACCESSED
| _PAGE_HASHPTE
)) == 0)
518 old
= pmd_hugepage_update(mm
, addr
, pmdp
, _PAGE_ACCESSED
);
519 return ((old
& _PAGE_ACCESSED
) != 0);
522 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
523 extern int pmdp_test_and_clear_young(struct vm_area_struct
*vma
,
524 unsigned long address
, pmd_t
*pmdp
);
525 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
526 extern int pmdp_clear_flush_young(struct vm_area_struct
*vma
,
527 unsigned long address
, pmd_t
*pmdp
);
529 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
530 extern pmd_t
pmdp_get_and_clear(struct mm_struct
*mm
,
531 unsigned long addr
, pmd_t
*pmdp
);
533 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
534 extern pmd_t
pmdp_clear_flush(struct vm_area_struct
*vma
, unsigned long address
,
537 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
538 static inline void pmdp_set_wrprotect(struct mm_struct
*mm
, unsigned long addr
,
542 if ((pmd_val(*pmdp
) & _PAGE_RW
) == 0)
545 pmd_hugepage_update(mm
, addr
, pmdp
, _PAGE_RW
);
548 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
549 extern void pmdp_splitting_flush(struct vm_area_struct
*vma
,
550 unsigned long address
, pmd_t
*pmdp
);
552 #define __HAVE_ARCH_PGTABLE_DEPOSIT
553 extern void pgtable_trans_huge_deposit(struct mm_struct
*mm
, pmd_t
*pmdp
,
555 #define __HAVE_ARCH_PGTABLE_WITHDRAW
556 extern pgtable_t
pgtable_trans_huge_withdraw(struct mm_struct
*mm
, pmd_t
*pmdp
);
558 #define __HAVE_ARCH_PMDP_INVALIDATE
559 extern void pmdp_invalidate(struct vm_area_struct
*vma
, unsigned long address
,
562 #define pmd_move_must_withdraw pmd_move_must_withdraw
564 static inline int pmd_move_must_withdraw(struct spinlock
*new_pmd_ptl
,
565 struct spinlock
*old_pmd_ptl
)
568 * Archs like ppc64 use pgtable to store per pmd
569 * specific information. So when we switch the pmd,
570 * we should also withdraw and deposit the pgtable
575 #endif /* __ASSEMBLY__ */
576 #endif /* _ASM_POWERPC_PGTABLE_PPC64_H_ */