2 * IBM System z Huge TLB Page Support for Kernel.
4 * Copyright IBM Corp. 2007
5 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
9 #include <linux/hugetlb.h>
11 static inline pmd_t
__pte_to_pmd(pte_t pte
)
16 * Convert encoding pte bits pmd bits
17 * .IR...wrdytp dy..R...I...wr
18 * empty .10...000000 -> 00..0...1...00
19 * prot-none, clean, old .11...000001 -> 00..1...1...00
20 * prot-none, clean, young .11...000101 -> 01..1...1...00
21 * prot-none, dirty, old .10...001001 -> 10..1...1...00
22 * prot-none, dirty, young .10...001101 -> 11..1...1...00
23 * read-only, clean, old .11...010001 -> 00..1...1...01
24 * read-only, clean, young .01...010101 -> 01..1...0...01
25 * read-only, dirty, old .11...011001 -> 10..1...1...01
26 * read-only, dirty, young .01...011101 -> 11..1...0...01
27 * read-write, clean, old .11...110001 -> 00..0...1...11
28 * read-write, clean, young .01...110101 -> 01..0...0...11
29 * read-write, dirty, old .10...111001 -> 10..0...1...11
30 * read-write, dirty, young .00...111101 -> 11..0...0...11
32 if (pte_present(pte
)) {
33 pmd_val(pmd
) = pte_val(pte
) & PAGE_MASK
;
34 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_READ
) >> 4;
35 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_WRITE
) >> 4;
36 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_INVALID
) >> 5;
37 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_PROTECT
);
38 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_DIRTY
) << 10;
39 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_YOUNG
) << 10;
41 pmd_val(pmd
) = _SEGMENT_ENTRY_INVALID
;
45 static inline pte_t
__pmd_to_pte(pmd_t pmd
)
50 * Convert encoding pmd bits pte bits
51 * dy..R...I...wr .IR...wrdytp
52 * empty 00..0...1...00 -> .10...001100
53 * prot-none, clean, old 00..0...1...00 -> .10...000001
54 * prot-none, clean, young 01..0...1...00 -> .10...000101
55 * prot-none, dirty, old 10..0...1...00 -> .10...001001
56 * prot-none, dirty, young 11..0...1...00 -> .10...001101
57 * read-only, clean, old 00..1...1...01 -> .11...010001
58 * read-only, clean, young 01..1...1...01 -> .11...010101
59 * read-only, dirty, old 10..1...1...01 -> .11...011001
60 * read-only, dirty, young 11..1...1...01 -> .11...011101
61 * read-write, clean, old 00..0...1...11 -> .10...110001
62 * read-write, clean, young 01..0...1...11 -> .10...110101
63 * read-write, dirty, old 10..0...1...11 -> .10...111001
64 * read-write, dirty, young 11..0...1...11 -> .10...111101
66 if (pmd_present(pmd
)) {
67 pte_val(pte
) = pmd_val(pmd
) & _SEGMENT_ENTRY_ORIGIN_LARGE
;
68 pte_val(pte
) |= _PAGE_LARGE
| _PAGE_PRESENT
;
69 pte_val(pte
) |= (pmd_val(pmd
) & _SEGMENT_ENTRY_READ
) << 4;
70 pte_val(pte
) |= (pmd_val(pmd
) & _SEGMENT_ENTRY_WRITE
) << 4;
71 pte_val(pte
) |= (pmd_val(pmd
) & _SEGMENT_ENTRY_INVALID
) << 5;
72 pte_val(pte
) |= (pmd_val(pmd
) & _SEGMENT_ENTRY_PROTECT
);
73 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_DIRTY
) << 10;
74 pmd_val(pmd
) |= (pte_val(pte
) & _PAGE_YOUNG
) << 10;
76 pte_val(pte
) = _PAGE_INVALID
;
80 void set_huge_pte_at(struct mm_struct
*mm
, unsigned long addr
,
81 pte_t
*ptep
, pte_t pte
)
85 pmd
= __pte_to_pmd(pte
);
86 if (!MACHINE_HAS_HPAGE
) {
87 /* Emulated huge ptes loose the dirty and young bit */
88 pmd_val(pmd
) &= ~_SEGMENT_ENTRY_ORIGIN
;
89 pmd_val(pmd
) |= pte_page(pte
)[1].index
;
91 pmd_val(pmd
) |= _SEGMENT_ENTRY_LARGE
;
92 *(pmd_t
*) ptep
= pmd
;
95 pte_t
huge_ptep_get(pte_t
*ptep
)
100 pmd
= *(pmd_t
*) ptep
;
101 if (!MACHINE_HAS_HPAGE
&& pmd_present(pmd
)) {
102 origin
= pmd_val(pmd
) & _SEGMENT_ENTRY_ORIGIN
;
103 pmd_val(pmd
) &= ~_SEGMENT_ENTRY_ORIGIN
;
104 pmd_val(pmd
) |= *(unsigned long *) origin
;
105 /* Emulated huge ptes are young and dirty by definition */
106 pmd_val(pmd
) |= _SEGMENT_ENTRY_YOUNG
| _SEGMENT_ENTRY_DIRTY
;
108 return __pmd_to_pte(pmd
);
111 pte_t
huge_ptep_get_and_clear(struct mm_struct
*mm
,
112 unsigned long addr
, pte_t
*ptep
)
114 pmd_t
*pmdp
= (pmd_t
*) ptep
;
115 pte_t pte
= huge_ptep_get(ptep
);
117 pmdp_flush_direct(mm
, addr
, pmdp
);
118 pmd_val(*pmdp
) = _SEGMENT_ENTRY_EMPTY
;
122 int arch_prepare_hugepage(struct page
*page
)
124 unsigned long addr
= page_to_phys(page
);
129 if (MACHINE_HAS_HPAGE
)
132 ptep
= (pte_t
*) pte_alloc_one(&init_mm
, addr
);
137 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
138 set_pte_at(&init_mm
, addr
+ i
* PAGE_SIZE
, ptep
+ i
, pte
);
139 pte_val(pte
) += PAGE_SIZE
;
141 page
[1].index
= (unsigned long) ptep
;
145 void arch_release_hugepage(struct page
*page
)
149 if (MACHINE_HAS_HPAGE
)
152 ptep
= (pte_t
*) page
[1].index
;
155 clear_table((unsigned long *) ptep
, _PAGE_INVALID
,
156 PTRS_PER_PTE
* sizeof(pte_t
));
157 page_table_free(&init_mm
, (unsigned long *) ptep
);
161 pte_t
*huge_pte_alloc(struct mm_struct
*mm
,
162 unsigned long addr
, unsigned long sz
)
168 pgdp
= pgd_offset(mm
, addr
);
169 pudp
= pud_alloc(mm
, pgdp
, addr
);
171 pmdp
= pmd_alloc(mm
, pudp
, addr
);
172 return (pte_t
*) pmdp
;
175 pte_t
*huge_pte_offset(struct mm_struct
*mm
, unsigned long addr
)
181 pgdp
= pgd_offset(mm
, addr
);
182 if (pgd_present(*pgdp
)) {
183 pudp
= pud_offset(pgdp
, addr
);
184 if (pud_present(*pudp
))
185 pmdp
= pmd_offset(pudp
, addr
);
187 return (pte_t
*) pmdp
;
190 int huge_pmd_unshare(struct mm_struct
*mm
, unsigned long *addr
, pte_t
*ptep
)
195 struct page
*follow_huge_addr(struct mm_struct
*mm
, unsigned long address
,
198 return ERR_PTR(-EINVAL
);
201 int pmd_huge(pmd_t pmd
)
203 if (!MACHINE_HAS_HPAGE
)
206 return !!(pmd_val(pmd
) & _SEGMENT_ENTRY_LARGE
);
209 int pud_huge(pud_t pud
)
214 struct page
*follow_huge_pmd(struct mm_struct
*mm
, unsigned long address
,
215 pmd_t
*pmdp
, int write
)
219 if (!MACHINE_HAS_HPAGE
)
222 page
= pmd_page(*pmdp
);
224 page
+= ((address
& ~HPAGE_MASK
) >> PAGE_SHIFT
);