Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * IBM System z Huge TLB Page Support for Kernel.
4 *
5 * Copyright IBM Corp. 2007,2016
6 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
7 */
8
9#define KMSG_COMPONENT "hugetlb"
10#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
11
12#include <linux/mm.h>
13#include <linux/hugetlb.h>
14
15/*
16 * If the bit selected by single-bit bitmask "a" is set within "x", move
17 * it to the position indicated by single-bit bitmask "b".
18 */
19#define move_set_bit(x, a, b) (((x) & (a)) >> ilog2(a) << ilog2(b))
20
21static inline unsigned long __pte_to_rste(pte_t pte)
22{
23 unsigned long rste;
24
25 /*
26 * Convert encoding pte bits pmd / pud bits
27 * lIR.uswrdy.p dy..R...I...wr
28 * empty 010.000000.0 -> 00..0...1...00
29 * prot-none, clean, old 111.000000.1 -> 00..1...1...00
30 * prot-none, clean, young 111.000001.1 -> 01..1...1...00
31 * prot-none, dirty, old 111.000010.1 -> 10..1...1...00
32 * prot-none, dirty, young 111.000011.1 -> 11..1...1...00
33 * read-only, clean, old 111.000100.1 -> 00..1...1...01
34 * read-only, clean, young 101.000101.1 -> 01..1...0...01
35 * read-only, dirty, old 111.000110.1 -> 10..1...1...01
36 * read-only, dirty, young 101.000111.1 -> 11..1...0...01
37 * read-write, clean, old 111.001100.1 -> 00..1...1...11
38 * read-write, clean, young 101.001101.1 -> 01..1...0...11
39 * read-write, dirty, old 110.001110.1 -> 10..0...1...11
40 * read-write, dirty, young 100.001111.1 -> 11..0...0...11
41 * HW-bits: R read-only, I invalid
42 * SW-bits: p present, y young, d dirty, r read, w write, s special,
43 * u unused, l large
44 */
45 if (pte_present(pte)) {
46 rste = pte_val(pte) & PAGE_MASK;
47 rste |= move_set_bit(pte_val(pte), _PAGE_READ,
48 _SEGMENT_ENTRY_READ);
49 rste |= move_set_bit(pte_val(pte), _PAGE_WRITE,
50 _SEGMENT_ENTRY_WRITE);
51 rste |= move_set_bit(pte_val(pte), _PAGE_INVALID,
52 _SEGMENT_ENTRY_INVALID);
53 rste |= move_set_bit(pte_val(pte), _PAGE_PROTECT,
54 _SEGMENT_ENTRY_PROTECT);
55 rste |= move_set_bit(pte_val(pte), _PAGE_DIRTY,
56 _SEGMENT_ENTRY_DIRTY);
57 rste |= move_set_bit(pte_val(pte), _PAGE_YOUNG,
58 _SEGMENT_ENTRY_YOUNG);
59#ifdef CONFIG_MEM_SOFT_DIRTY
60 rste |= move_set_bit(pte_val(pte), _PAGE_SOFT_DIRTY,
61 _SEGMENT_ENTRY_SOFT_DIRTY);
62#endif
63 rste |= move_set_bit(pte_val(pte), _PAGE_NOEXEC,
64 _SEGMENT_ENTRY_NOEXEC);
65 } else
66 rste = _SEGMENT_ENTRY_EMPTY;
67 return rste;
68}
69
70static inline pte_t __rste_to_pte(unsigned long rste)
71{
72 int present;
73 pte_t pte;
74
75 if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
76 present = pud_present(__pud(rste));
77 else
78 present = pmd_present(__pmd(rste));
79
80 /*
81 * Convert encoding pmd / pud bits pte bits
82 * dy..R...I...wr lIR.uswrdy.p
83 * empty 00..0...1...00 -> 010.000000.0
84 * prot-none, clean, old 00..1...1...00 -> 111.000000.1
85 * prot-none, clean, young 01..1...1...00 -> 111.000001.1
86 * prot-none, dirty, old 10..1...1...00 -> 111.000010.1
87 * prot-none, dirty, young 11..1...1...00 -> 111.000011.1
88 * read-only, clean, old 00..1...1...01 -> 111.000100.1
89 * read-only, clean, young 01..1...0...01 -> 101.000101.1
90 * read-only, dirty, old 10..1...1...01 -> 111.000110.1
91 * read-only, dirty, young 11..1...0...01 -> 101.000111.1
92 * read-write, clean, old 00..1...1...11 -> 111.001100.1
93 * read-write, clean, young 01..1...0...11 -> 101.001101.1
94 * read-write, dirty, old 10..0...1...11 -> 110.001110.1
95 * read-write, dirty, young 11..0...0...11 -> 100.001111.1
96 * HW-bits: R read-only, I invalid
97 * SW-bits: p present, y young, d dirty, r read, w write, s special,
98 * u unused, l large
99 */
100 if (present) {
101 pte_val(pte) = rste & _SEGMENT_ENTRY_ORIGIN_LARGE;
102 pte_val(pte) |= _PAGE_LARGE | _PAGE_PRESENT;
103 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_READ,
104 _PAGE_READ);
105 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_WRITE,
106 _PAGE_WRITE);
107 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_INVALID,
108 _PAGE_INVALID);
109 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_PROTECT,
110 _PAGE_PROTECT);
111 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_DIRTY,
112 _PAGE_DIRTY);
113 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_YOUNG,
114 _PAGE_YOUNG);
115#ifdef CONFIG_MEM_SOFT_DIRTY
116 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_SOFT_DIRTY,
117 _PAGE_DIRTY);
118#endif
119 pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_NOEXEC,
120 _PAGE_NOEXEC);
121 } else
122 pte_val(pte) = _PAGE_INVALID;
123 return pte;
124}
125
126static void clear_huge_pte_skeys(struct mm_struct *mm, unsigned long rste)
127{
128 struct page *page;
129 unsigned long size, paddr;
130
131 if (!mm_uses_skeys(mm) ||
132 rste & _SEGMENT_ENTRY_INVALID)
133 return;
134
135 if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) {
136 page = pud_page(__pud(rste));
137 size = PUD_SIZE;
138 paddr = rste & PUD_MASK;
139 } else {
140 page = pmd_page(__pmd(rste));
141 size = PMD_SIZE;
142 paddr = rste & PMD_MASK;
143 }
144
145 if (!test_and_set_bit(PG_arch_1, &page->flags))
146 __storage_key_init_range(paddr, paddr + size - 1);
147}
148
149void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
150 pte_t *ptep, pte_t pte)
151{
152 unsigned long rste;
153
154 rste = __pte_to_rste(pte);
155 if (!MACHINE_HAS_NX)
156 rste &= ~_SEGMENT_ENTRY_NOEXEC;
157
158 /* Set correct table type for 2G hugepages */
159 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
160 rste |= _REGION_ENTRY_TYPE_R3 | _REGION3_ENTRY_LARGE;
161 else
162 rste |= _SEGMENT_ENTRY_LARGE;
163 clear_huge_pte_skeys(mm, rste);
164 pte_val(*ptep) = rste;
165}
166
167pte_t huge_ptep_get(pte_t *ptep)
168{
169 return __rste_to_pte(pte_val(*ptep));
170}
171
172pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
173 unsigned long addr, pte_t *ptep)
174{
175 pte_t pte = huge_ptep_get(ptep);
176 pmd_t *pmdp = (pmd_t *) ptep;
177 pud_t *pudp = (pud_t *) ptep;
178
179 if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
180 pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY));
181 else
182 pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
183 return pte;
184}
185
186pte_t *huge_pte_alloc(struct mm_struct *mm,
187 unsigned long addr, unsigned long sz)
188{
189 pgd_t *pgdp;
190 p4d_t *p4dp;
191 pud_t *pudp;
192 pmd_t *pmdp = NULL;
193
194 pgdp = pgd_offset(mm, addr);
195 p4dp = p4d_alloc(mm, pgdp, addr);
196 if (p4dp) {
197 pudp = pud_alloc(mm, p4dp, addr);
198 if (pudp) {
199 if (sz == PUD_SIZE)
200 return (pte_t *) pudp;
201 else if (sz == PMD_SIZE)
202 pmdp = pmd_alloc(mm, pudp, addr);
203 }
204 }
205 return (pte_t *) pmdp;
206}
207
208pte_t *huge_pte_offset(struct mm_struct *mm,
209 unsigned long addr, unsigned long sz)
210{
211 pgd_t *pgdp;
212 p4d_t *p4dp;
213 pud_t *pudp;
214 pmd_t *pmdp = NULL;
215
216 pgdp = pgd_offset(mm, addr);
217 if (pgd_present(*pgdp)) {
218 p4dp = p4d_offset(pgdp, addr);
219 if (p4d_present(*p4dp)) {
220 pudp = pud_offset(p4dp, addr);
221 if (pud_present(*pudp)) {
222 if (pud_large(*pudp))
223 return (pte_t *) pudp;
224 pmdp = pmd_offset(pudp, addr);
225 }
226 }
227 }
228 return (pte_t *) pmdp;
229}
230
231int pmd_huge(pmd_t pmd)
232{
233 return pmd_large(pmd);
234}
235
236int pud_huge(pud_t pud)
237{
238 return pud_large(pud);
239}
240
241struct page *
242follow_huge_pud(struct mm_struct *mm, unsigned long address,
243 pud_t *pud, int flags)
244{
245 if (flags & FOLL_GET)
246 return NULL;
247
248 return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
249}
250
251static __init int setup_hugepagesz(char *opt)
252{
253 unsigned long size;
254 char *string = opt;
255
256 size = memparse(opt, &opt);
257 if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) {
258 hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
259 } else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) {
260 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
261 } else {
262 hugetlb_bad_size();
263 pr_err("hugepagesz= specifies an unsupported page size %s\n",
264 string);
265 return 0;
266 }
267 return 1;
268}
269__setup("hugepagesz=", setup_hugepagesz);
1/*
2 * IBM System z Huge TLB Page Support for Kernel.
3 *
4 * Copyright IBM Corp. 2007
5 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
6 */
7
8#include <linux/mm.h>
9#include <linux/hugetlb.h>
10
11static inline pmd_t __pte_to_pmd(pte_t pte)
12{
13 int none, young, prot;
14 pmd_t pmd;
15
16 /*
17 * Convert encoding pte bits pmd bits
18 * .IR...wrdytp ..R...I...y.
19 * empty .10...000000 -> ..0...1...0.
20 * prot-none, clean, old .11...000001 -> ..0...1...1.
21 * prot-none, clean, young .11...000101 -> ..1...1...1.
22 * prot-none, dirty, old .10...001001 -> ..0...1...1.
23 * prot-none, dirty, young .10...001101 -> ..1...1...1.
24 * read-only, clean, old .11...010001 -> ..1...1...0.
25 * read-only, clean, young .01...010101 -> ..1...0...1.
26 * read-only, dirty, old .11...011001 -> ..1...1...0.
27 * read-only, dirty, young .01...011101 -> ..1...0...1.
28 * read-write, clean, old .11...110001 -> ..0...1...0.
29 * read-write, clean, young .01...110101 -> ..0...0...1.
30 * read-write, dirty, old .10...111001 -> ..0...1...0.
31 * read-write, dirty, young .00...111101 -> ..0...0...1.
32 * Huge ptes are dirty by definition, a clean pte is made dirty
33 * by the conversion.
34 */
35 if (pte_present(pte)) {
36 pmd_val(pmd) = pte_val(pte) & PAGE_MASK;
37 if (pte_val(pte) & _PAGE_INVALID)
38 pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
39 none = (pte_val(pte) & _PAGE_PRESENT) &&
40 !(pte_val(pte) & _PAGE_READ) &&
41 !(pte_val(pte) & _PAGE_WRITE);
42 prot = (pte_val(pte) & _PAGE_PROTECT) &&
43 !(pte_val(pte) & _PAGE_WRITE);
44 young = pte_val(pte) & _PAGE_YOUNG;
45 if (none || young)
46 pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
47 if (prot || (none && young))
48 pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
49 } else
50 pmd_val(pmd) = _SEGMENT_ENTRY_INVALID;
51 return pmd;
52}
53
54static inline pte_t __pmd_to_pte(pmd_t pmd)
55{
56 pte_t pte;
57
58 /*
59 * Convert encoding pmd bits pte bits
60 * ..R...I...y. .IR...wrdytp
61 * empty ..0...1...0. -> .10...000000
62 * prot-none, old ..0...1...1. -> .10...001001
63 * prot-none, young ..1...1...1. -> .10...001101
64 * read-only, old ..1...1...0. -> .11...011001
65 * read-only, young ..1...0...1. -> .01...011101
66 * read-write, old ..0...1...0. -> .10...111001
67 * read-write, young ..0...0...1. -> .00...111101
68 * Huge ptes are dirty by definition
69 */
70 if (pmd_present(pmd)) {
71 pte_val(pte) = _PAGE_PRESENT | _PAGE_LARGE | _PAGE_DIRTY |
72 (pmd_val(pmd) & PAGE_MASK);
73 if (pmd_val(pmd) & _SEGMENT_ENTRY_INVALID)
74 pte_val(pte) |= _PAGE_INVALID;
75 if (pmd_prot_none(pmd)) {
76 if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
77 pte_val(pte) |= _PAGE_YOUNG;
78 } else {
79 pte_val(pte) |= _PAGE_READ;
80 if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
81 pte_val(pte) |= _PAGE_PROTECT;
82 else
83 pte_val(pte) |= _PAGE_WRITE;
84 if (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG)
85 pte_val(pte) |= _PAGE_YOUNG;
86 }
87 } else
88 pte_val(pte) = _PAGE_INVALID;
89 return pte;
90}
91
92void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
93 pte_t *ptep, pte_t pte)
94{
95 pmd_t pmd;
96
97 pmd = __pte_to_pmd(pte);
98 if (!MACHINE_HAS_HPAGE) {
99 pmd_val(pmd) &= ~_SEGMENT_ENTRY_ORIGIN;
100 pmd_val(pmd) |= pte_page(pte)[1].index;
101 } else
102 pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO;
103 *(pmd_t *) ptep = pmd;
104}
105
106pte_t huge_ptep_get(pte_t *ptep)
107{
108 unsigned long origin;
109 pmd_t pmd;
110
111 pmd = *(pmd_t *) ptep;
112 if (!MACHINE_HAS_HPAGE && pmd_present(pmd)) {
113 origin = pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN;
114 pmd_val(pmd) &= ~_SEGMENT_ENTRY_ORIGIN;
115 pmd_val(pmd) |= *(unsigned long *) origin;
116 }
117 return __pmd_to_pte(pmd);
118}
119
120pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
121 unsigned long addr, pte_t *ptep)
122{
123 pmd_t *pmdp = (pmd_t *) ptep;
124 pte_t pte = huge_ptep_get(ptep);
125
126 pmdp_flush_direct(mm, addr, pmdp);
127 pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
128 return pte;
129}
130
131int arch_prepare_hugepage(struct page *page)
132{
133 unsigned long addr = page_to_phys(page);
134 pte_t pte;
135 pte_t *ptep;
136 int i;
137
138 if (MACHINE_HAS_HPAGE)
139 return 0;
140
141 ptep = (pte_t *) pte_alloc_one(&init_mm, addr);
142 if (!ptep)
143 return -ENOMEM;
144
145 pte_val(pte) = addr;
146 for (i = 0; i < PTRS_PER_PTE; i++) {
147 set_pte_at(&init_mm, addr + i * PAGE_SIZE, ptep + i, pte);
148 pte_val(pte) += PAGE_SIZE;
149 }
150 page[1].index = (unsigned long) ptep;
151 return 0;
152}
153
154void arch_release_hugepage(struct page *page)
155{
156 pte_t *ptep;
157
158 if (MACHINE_HAS_HPAGE)
159 return;
160
161 ptep = (pte_t *) page[1].index;
162 if (!ptep)
163 return;
164 clear_table((unsigned long *) ptep, _PAGE_INVALID,
165 PTRS_PER_PTE * sizeof(pte_t));
166 page_table_free(&init_mm, (unsigned long *) ptep);
167 page[1].index = 0;
168}
169
170pte_t *huge_pte_alloc(struct mm_struct *mm,
171 unsigned long addr, unsigned long sz)
172{
173 pgd_t *pgdp;
174 pud_t *pudp;
175 pmd_t *pmdp = NULL;
176
177 pgdp = pgd_offset(mm, addr);
178 pudp = pud_alloc(mm, pgdp, addr);
179 if (pudp)
180 pmdp = pmd_alloc(mm, pudp, addr);
181 return (pte_t *) pmdp;
182}
183
184pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
185{
186 pgd_t *pgdp;
187 pud_t *pudp;
188 pmd_t *pmdp = NULL;
189
190 pgdp = pgd_offset(mm, addr);
191 if (pgd_present(*pgdp)) {
192 pudp = pud_offset(pgdp, addr);
193 if (pud_present(*pudp))
194 pmdp = pmd_offset(pudp, addr);
195 }
196 return (pte_t *) pmdp;
197}
198
199int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
200{
201 return 0;
202}
203
204struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
205 int write)
206{
207 return ERR_PTR(-EINVAL);
208}
209
210int pmd_huge(pmd_t pmd)
211{
212 if (!MACHINE_HAS_HPAGE)
213 return 0;
214
215 return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
216}
217
218int pud_huge(pud_t pud)
219{
220 return 0;
221}
222
223int pmd_huge_support(void)
224{
225 return 1;
226}
227
228struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
229 pmd_t *pmdp, int write)
230{
231 struct page *page;
232
233 if (!MACHINE_HAS_HPAGE)
234 return NULL;
235
236 page = pmd_page(*pmdp);
237 if (page)
238 page += ((address & ~HPAGE_MASK) >> PAGE_SHIFT);
239 return page;
240}