2 * IBM System z Huge TLB Page Support for Kernel.
4 * Copyright IBM Corp. 2007,2016
5 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
8 #define KMSG_COMPONENT "hugetlb"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12 #include <linux/hugetlb.h>
15 * If the bit selected by single-bit bitmask "a" is set within "x", move
16 * it to the position indicated by single-bit bitmask "b".
18 #define move_set_bit(x, a, b) (((x) & (a)) >> ilog2(a) << ilog2(b))
20 static inline unsigned long __pte_to_rste(pte_t pte)
25 * Convert encoding pte bits pmd / pud bits
26 * lIR.uswrdy.p dy..R...I...wr
27 * empty 010.000000.0 -> 00..0...1...00
28 * prot-none, clean, old 111.000000.1 -> 00..1...1...00
29 * prot-none, clean, young 111.000001.1 -> 01..1...1...00
30 * prot-none, dirty, old 111.000010.1 -> 10..1...1...00
31 * prot-none, dirty, young 111.000011.1 -> 11..1...1...00
32 * read-only, clean, old 111.000100.1 -> 00..1...1...01
33 * read-only, clean, young 101.000101.1 -> 01..1...0...01
34 * read-only, dirty, old 111.000110.1 -> 10..1...1...01
35 * read-only, dirty, young 101.000111.1 -> 11..1...0...01
36 * read-write, clean, old 111.001100.1 -> 00..1...1...11
37 * read-write, clean, young 101.001101.1 -> 01..1...0...11
38 * read-write, dirty, old 110.001110.1 -> 10..0...1...11
39 * read-write, dirty, young 100.001111.1 -> 11..0...0...11
40 * HW-bits: R read-only, I invalid
41 * SW-bits: p present, y young, d dirty, r read, w write, s special,
44 if (pte_present(pte)) {
45 rste = pte_val(pte) & PAGE_MASK;
46 rste |= move_set_bit(pte_val(pte), _PAGE_READ,
48 rste |= move_set_bit(pte_val(pte), _PAGE_WRITE,
49 _SEGMENT_ENTRY_WRITE);
50 rste |= move_set_bit(pte_val(pte), _PAGE_INVALID,
51 _SEGMENT_ENTRY_INVALID);
52 rste |= move_set_bit(pte_val(pte), _PAGE_PROTECT,
53 _SEGMENT_ENTRY_PROTECT);
54 rste |= move_set_bit(pte_val(pte), _PAGE_DIRTY,
55 _SEGMENT_ENTRY_DIRTY);
56 rste |= move_set_bit(pte_val(pte), _PAGE_YOUNG,
57 _SEGMENT_ENTRY_YOUNG);
58 #ifdef CONFIG_MEM_SOFT_DIRTY
59 rste |= move_set_bit(pte_val(pte), _PAGE_SOFT_DIRTY,
60 _SEGMENT_ENTRY_SOFT_DIRTY);
63 rste = _SEGMENT_ENTRY_INVALID;
67 static inline pte_t __rste_to_pte(unsigned long rste)
72 if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
73 present = pud_present(__pud(rste));
75 present = pmd_present(__pmd(rste));
78 * Convert encoding pmd / pud bits pte bits
79 * dy..R...I...wr lIR.uswrdy.p
80 * empty 00..0...1...00 -> 010.000000.0
81 * prot-none, clean, old 00..1...1...00 -> 111.000000.1
82 * prot-none, clean, young 01..1...1...00 -> 111.000001.1
83 * prot-none, dirty, old 10..1...1...00 -> 111.000010.1
84 * prot-none, dirty, young 11..1...1...00 -> 111.000011.1
85 * read-only, clean, old 00..1...1...01 -> 111.000100.1
86 * read-only, clean, young 01..1...0...01 -> 101.000101.1
87 * read-only, dirty, old 10..1...1...01 -> 111.000110.1
88 * read-only, dirty, young 11..1...0...01 -> 101.000111.1
89 * read-write, clean, old 00..1...1...11 -> 111.001100.1
90 * read-write, clean, young 01..1...0...11 -> 101.001101.1
91 * read-write, dirty, old 10..0...1...11 -> 110.001110.1
92 * read-write, dirty, young 11..0...0...11 -> 100.001111.1
93 * HW-bits: R read-only, I invalid
94 * SW-bits: p present, y young, d dirty, r read, w write, s special,
98 pte_val(pte) = rste & _SEGMENT_ENTRY_ORIGIN_LARGE;
99 pte_val(pte) |= _PAGE_LARGE | _PAGE_PRESENT;
100 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_READ,
102 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_WRITE,
104 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_INVALID,
106 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_PROTECT,
108 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_DIRTY,
110 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_YOUNG,
112 #ifdef CONFIG_MEM_SOFT_DIRTY
113 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_SOFT_DIRTY,
117 pte_val(pte) = _PAGE_INVALID;
121 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
122 pte_t *ptep, pte_t pte)
124 unsigned long rste = __pte_to_rste(pte);
126 /* Set correct table type for 2G hugepages */
127 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
128 rste |= _REGION_ENTRY_TYPE_R3 | _REGION3_ENTRY_LARGE;
130 rste |= _SEGMENT_ENTRY_LARGE;
131 pte_val(*ptep) = rste;
134 pte_t huge_ptep_get(pte_t *ptep)
136 return __rste_to_pte(pte_val(*ptep));
139 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
140 unsigned long addr, pte_t *ptep)
142 pte_t pte = huge_ptep_get(ptep);
143 pmd_t *pmdp = (pmd_t *) ptep;
144 pud_t *pudp = (pud_t *) ptep;
146 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
147 pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY));
149 pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
153 pte_t *huge_pte_alloc(struct mm_struct *mm,
154 unsigned long addr, unsigned long sz)
160 pgdp = pgd_offset(mm, addr);
161 pudp = pud_alloc(mm, pgdp, addr);
164 return (pte_t *) pudp;
165 else if (sz == PMD_SIZE)
166 pmdp = pmd_alloc(mm, pudp, addr);
168 return (pte_t *) pmdp;
171 pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
177 pgdp = pgd_offset(mm, addr);
178 if (pgd_present(*pgdp)) {
179 pudp = pud_offset(pgdp, addr);
180 if (pud_present(*pudp)) {
181 if (pud_large(*pudp))
182 return (pte_t *) pudp;
183 pmdp = pmd_offset(pudp, addr);
186 return (pte_t *) pmdp;
189 int pmd_huge(pmd_t pmd)
191 return pmd_large(pmd);
194 int pud_huge(pud_t pud)
196 return pud_large(pud);
200 follow_huge_pud(struct mm_struct *mm, unsigned long address,
201 pud_t *pud, int flags)
203 if (flags & FOLL_GET)
206 return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
209 static __init int setup_hugepagesz(char *opt)
214 size = memparse(opt, &opt);
215 if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) {
216 hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
217 } else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) {
218 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
220 pr_err("hugepagesz= specifies an unsupported page size %s\n",
226 __setup("hugepagesz=", setup_hugepagesz);