xref: /openbmc/linux/arch/s390/mm/hugetlbpage.c (revision 791d3ef2)
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 
21 static 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 
70 static 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 
126 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
127 		     pte_t *ptep, pte_t pte)
128 {
129 	unsigned long rste;
130 
131 	rste = __pte_to_rste(pte);
132 	if (!MACHINE_HAS_NX)
133 		rste &= ~_SEGMENT_ENTRY_NOEXEC;
134 
135 	/* Set correct table type for 2G hugepages */
136 	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
137 		rste |= _REGION_ENTRY_TYPE_R3 | _REGION3_ENTRY_LARGE;
138 	else
139 		rste |= _SEGMENT_ENTRY_LARGE;
140 	pte_val(*ptep) = rste;
141 }
142 
143 pte_t huge_ptep_get(pte_t *ptep)
144 {
145 	return __rste_to_pte(pte_val(*ptep));
146 }
147 
148 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
149 			      unsigned long addr, pte_t *ptep)
150 {
151 	pte_t pte = huge_ptep_get(ptep);
152 	pmd_t *pmdp = (pmd_t *) ptep;
153 	pud_t *pudp = (pud_t *) ptep;
154 
155 	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
156 		pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY));
157 	else
158 		pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
159 	return pte;
160 }
161 
162 pte_t *huge_pte_alloc(struct mm_struct *mm,
163 			unsigned long addr, unsigned long sz)
164 {
165 	pgd_t *pgdp;
166 	p4d_t *p4dp;
167 	pud_t *pudp;
168 	pmd_t *pmdp = NULL;
169 
170 	pgdp = pgd_offset(mm, addr);
171 	p4dp = p4d_alloc(mm, pgdp, addr);
172 	if (p4dp) {
173 		pudp = pud_alloc(mm, p4dp, addr);
174 		if (pudp) {
175 			if (sz == PUD_SIZE)
176 				return (pte_t *) pudp;
177 			else if (sz == PMD_SIZE)
178 				pmdp = pmd_alloc(mm, pudp, addr);
179 		}
180 	}
181 	return (pte_t *) pmdp;
182 }
183 
184 pte_t *huge_pte_offset(struct mm_struct *mm,
185 		       unsigned long addr, unsigned long sz)
186 {
187 	pgd_t *pgdp;
188 	p4d_t *p4dp;
189 	pud_t *pudp;
190 	pmd_t *pmdp = NULL;
191 
192 	pgdp = pgd_offset(mm, addr);
193 	if (pgd_present(*pgdp)) {
194 		p4dp = p4d_offset(pgdp, addr);
195 		if (p4d_present(*p4dp)) {
196 			pudp = pud_offset(p4dp, addr);
197 			if (pud_present(*pudp)) {
198 				if (pud_large(*pudp))
199 					return (pte_t *) pudp;
200 				pmdp = pmd_offset(pudp, addr);
201 			}
202 		}
203 	}
204 	return (pte_t *) pmdp;
205 }
206 
207 int pmd_huge(pmd_t pmd)
208 {
209 	return pmd_large(pmd);
210 }
211 
212 int pud_huge(pud_t pud)
213 {
214 	return pud_large(pud);
215 }
216 
217 struct page *
218 follow_huge_pud(struct mm_struct *mm, unsigned long address,
219 		pud_t *pud, int flags)
220 {
221 	if (flags & FOLL_GET)
222 		return NULL;
223 
224 	return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
225 }
226 
227 static __init int setup_hugepagesz(char *opt)
228 {
229 	unsigned long size;
230 	char *string = opt;
231 
232 	size = memparse(opt, &opt);
233 	if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) {
234 		hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
235 	} else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) {
236 		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
237 	} else {
238 		hugetlb_bad_size();
239 		pr_err("hugepagesz= specifies an unsupported page size %s\n",
240 			string);
241 		return 0;
242 	}
243 	return 1;
244 }
245 __setup("hugepagesz=", setup_hugepagesz);
246