xref: /openbmc/linux/arch/s390/mm/hugetlbpage.c (revision 37744fee)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  IBM System z Huge TLB Page Support for Kernel.
4  *
5  *    Copyright IBM Corp. 2007,2020
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 #include <linux/mman.h>
15 #include <linux/sched/mm.h>
16 #include <linux/security.h>
17 
18 /*
19  * If the bit selected by single-bit bitmask "a" is set within "x", move
20  * it to the position indicated by single-bit bitmask "b".
21  */
22 #define move_set_bit(x, a, b)	(((x) & (a)) >> ilog2(a) << ilog2(b))
23 
24 static inline unsigned long __pte_to_rste(pte_t pte)
25 {
26 	unsigned long rste;
27 
28 	/*
29 	 * Convert encoding		  pte bits	pmd / pud bits
30 	 *				lIR.uswrdy.p	dy..R...I...wr
31 	 * empty			010.000000.0 -> 00..0...1...00
32 	 * prot-none, clean, old	111.000000.1 -> 00..1...1...00
33 	 * prot-none, clean, young	111.000001.1 -> 01..1...1...00
34 	 * prot-none, dirty, old	111.000010.1 -> 10..1...1...00
35 	 * prot-none, dirty, young	111.000011.1 -> 11..1...1...00
36 	 * read-only, clean, old	111.000100.1 -> 00..1...1...01
37 	 * read-only, clean, young	101.000101.1 -> 01..1...0...01
38 	 * read-only, dirty, old	111.000110.1 -> 10..1...1...01
39 	 * read-only, dirty, young	101.000111.1 -> 11..1...0...01
40 	 * read-write, clean, old	111.001100.1 -> 00..1...1...11
41 	 * read-write, clean, young	101.001101.1 -> 01..1...0...11
42 	 * read-write, dirty, old	110.001110.1 -> 10..0...1...11
43 	 * read-write, dirty, young	100.001111.1 -> 11..0...0...11
44 	 * HW-bits: R read-only, I invalid
45 	 * SW-bits: p present, y young, d dirty, r read, w write, s special,
46 	 *	    u unused, l large
47 	 */
48 	if (pte_present(pte)) {
49 		rste = pte_val(pte) & PAGE_MASK;
50 		rste |= move_set_bit(pte_val(pte), _PAGE_READ,
51 				     _SEGMENT_ENTRY_READ);
52 		rste |= move_set_bit(pte_val(pte), _PAGE_WRITE,
53 				     _SEGMENT_ENTRY_WRITE);
54 		rste |= move_set_bit(pte_val(pte), _PAGE_INVALID,
55 				     _SEGMENT_ENTRY_INVALID);
56 		rste |= move_set_bit(pte_val(pte), _PAGE_PROTECT,
57 				     _SEGMENT_ENTRY_PROTECT);
58 		rste |= move_set_bit(pte_val(pte), _PAGE_DIRTY,
59 				     _SEGMENT_ENTRY_DIRTY);
60 		rste |= move_set_bit(pte_val(pte), _PAGE_YOUNG,
61 				     _SEGMENT_ENTRY_YOUNG);
62 #ifdef CONFIG_MEM_SOFT_DIRTY
63 		rste |= move_set_bit(pte_val(pte), _PAGE_SOFT_DIRTY,
64 				     _SEGMENT_ENTRY_SOFT_DIRTY);
65 #endif
66 		rste |= move_set_bit(pte_val(pte), _PAGE_NOEXEC,
67 				     _SEGMENT_ENTRY_NOEXEC);
68 	} else
69 		rste = _SEGMENT_ENTRY_EMPTY;
70 	return rste;
71 }
72 
73 static inline pte_t __rste_to_pte(unsigned long rste)
74 {
75 	int present;
76 	pte_t pte;
77 
78 	if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
79 		present = pud_present(__pud(rste));
80 	else
81 		present = pmd_present(__pmd(rste));
82 
83 	/*
84 	 * Convert encoding		pmd / pud bits	    pte bits
85 	 *				dy..R...I...wr	  lIR.uswrdy.p
86 	 * empty			00..0...1...00 -> 010.000000.0
87 	 * prot-none, clean, old	00..1...1...00 -> 111.000000.1
88 	 * prot-none, clean, young	01..1...1...00 -> 111.000001.1
89 	 * prot-none, dirty, old	10..1...1...00 -> 111.000010.1
90 	 * prot-none, dirty, young	11..1...1...00 -> 111.000011.1
91 	 * read-only, clean, old	00..1...1...01 -> 111.000100.1
92 	 * read-only, clean, young	01..1...0...01 -> 101.000101.1
93 	 * read-only, dirty, old	10..1...1...01 -> 111.000110.1
94 	 * read-only, dirty, young	11..1...0...01 -> 101.000111.1
95 	 * read-write, clean, old	00..1...1...11 -> 111.001100.1
96 	 * read-write, clean, young	01..1...0...11 -> 101.001101.1
97 	 * read-write, dirty, old	10..0...1...11 -> 110.001110.1
98 	 * read-write, dirty, young	11..0...0...11 -> 100.001111.1
99 	 * HW-bits: R read-only, I invalid
100 	 * SW-bits: p present, y young, d dirty, r read, w write, s special,
101 	 *	    u unused, l large
102 	 */
103 	if (present) {
104 		pte_val(pte) = rste & _SEGMENT_ENTRY_ORIGIN_LARGE;
105 		pte_val(pte) |= _PAGE_LARGE | _PAGE_PRESENT;
106 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_READ,
107 					     _PAGE_READ);
108 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_WRITE,
109 					     _PAGE_WRITE);
110 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_INVALID,
111 					     _PAGE_INVALID);
112 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_PROTECT,
113 					     _PAGE_PROTECT);
114 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_DIRTY,
115 					     _PAGE_DIRTY);
116 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_YOUNG,
117 					     _PAGE_YOUNG);
118 #ifdef CONFIG_MEM_SOFT_DIRTY
119 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_SOFT_DIRTY,
120 					     _PAGE_DIRTY);
121 #endif
122 		pte_val(pte) |= move_set_bit(rste, _SEGMENT_ENTRY_NOEXEC,
123 					     _PAGE_NOEXEC);
124 	} else
125 		pte_val(pte) = _PAGE_INVALID;
126 	return pte;
127 }
128 
129 static void clear_huge_pte_skeys(struct mm_struct *mm, unsigned long rste)
130 {
131 	struct page *page;
132 	unsigned long size, paddr;
133 
134 	if (!mm_uses_skeys(mm) ||
135 	    rste & _SEGMENT_ENTRY_INVALID)
136 		return;
137 
138 	if ((rste & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) {
139 		page = pud_page(__pud(rste));
140 		size = PUD_SIZE;
141 		paddr = rste & PUD_MASK;
142 	} else {
143 		page = pmd_page(__pmd(rste));
144 		size = PMD_SIZE;
145 		paddr = rste & PMD_MASK;
146 	}
147 
148 	if (!test_and_set_bit(PG_arch_1, &page->flags))
149 		__storage_key_init_range(paddr, paddr + size - 1);
150 }
151 
152 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
153 		     pte_t *ptep, pte_t pte)
154 {
155 	unsigned long rste;
156 
157 	rste = __pte_to_rste(pte);
158 	if (!MACHINE_HAS_NX)
159 		rste &= ~_SEGMENT_ENTRY_NOEXEC;
160 
161 	/* Set correct table type for 2G hugepages */
162 	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3) {
163 		if (likely(pte_present(pte)))
164 			rste |= _REGION3_ENTRY_LARGE;
165 		rste |= _REGION_ENTRY_TYPE_R3;
166 	} else if (likely(pte_present(pte)))
167 		rste |= _SEGMENT_ENTRY_LARGE;
168 
169 	clear_huge_pte_skeys(mm, rste);
170 	pte_val(*ptep) = rste;
171 }
172 
173 pte_t huge_ptep_get(pte_t *ptep)
174 {
175 	return __rste_to_pte(pte_val(*ptep));
176 }
177 
178 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
179 			      unsigned long addr, pte_t *ptep)
180 {
181 	pte_t pte = huge_ptep_get(ptep);
182 	pmd_t *pmdp = (pmd_t *) ptep;
183 	pud_t *pudp = (pud_t *) ptep;
184 
185 	if ((pte_val(*ptep) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
186 		pudp_xchg_direct(mm, addr, pudp, __pud(_REGION3_ENTRY_EMPTY));
187 	else
188 		pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
189 	return pte;
190 }
191 
192 pte_t *huge_pte_alloc(struct mm_struct *mm,
193 			unsigned long addr, unsigned long sz)
194 {
195 	pgd_t *pgdp;
196 	p4d_t *p4dp;
197 	pud_t *pudp;
198 	pmd_t *pmdp = NULL;
199 
200 	pgdp = pgd_offset(mm, addr);
201 	p4dp = p4d_alloc(mm, pgdp, addr);
202 	if (p4dp) {
203 		pudp = pud_alloc(mm, p4dp, addr);
204 		if (pudp) {
205 			if (sz == PUD_SIZE)
206 				return (pte_t *) pudp;
207 			else if (sz == PMD_SIZE)
208 				pmdp = pmd_alloc(mm, pudp, addr);
209 		}
210 	}
211 	return (pte_t *) pmdp;
212 }
213 
214 pte_t *huge_pte_offset(struct mm_struct *mm,
215 		       unsigned long addr, unsigned long sz)
216 {
217 	pgd_t *pgdp;
218 	p4d_t *p4dp;
219 	pud_t *pudp;
220 	pmd_t *pmdp = NULL;
221 
222 	pgdp = pgd_offset(mm, addr);
223 	if (pgd_present(*pgdp)) {
224 		p4dp = p4d_offset(pgdp, addr);
225 		if (p4d_present(*p4dp)) {
226 			pudp = pud_offset(p4dp, addr);
227 			if (pud_present(*pudp)) {
228 				if (pud_large(*pudp))
229 					return (pte_t *) pudp;
230 				pmdp = pmd_offset(pudp, addr);
231 			}
232 		}
233 	}
234 	return (pte_t *) pmdp;
235 }
236 
237 int pmd_huge(pmd_t pmd)
238 {
239 	return pmd_large(pmd);
240 }
241 
242 int pud_huge(pud_t pud)
243 {
244 	return pud_large(pud);
245 }
246 
247 struct page *
248 follow_huge_pud(struct mm_struct *mm, unsigned long address,
249 		pud_t *pud, int flags)
250 {
251 	if (flags & FOLL_GET)
252 		return NULL;
253 
254 	return pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
255 }
256 
257 static __init int setup_hugepagesz(char *opt)
258 {
259 	unsigned long size;
260 	char *string = opt;
261 
262 	size = memparse(opt, &opt);
263 	if (MACHINE_HAS_EDAT1 && size == PMD_SIZE) {
264 		hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
265 	} else if (MACHINE_HAS_EDAT2 && size == PUD_SIZE) {
266 		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
267 	} else {
268 		hugetlb_bad_size();
269 		pr_err("hugepagesz= specifies an unsupported page size %s\n",
270 			string);
271 		return 0;
272 	}
273 	return 1;
274 }
275 __setup("hugepagesz=", setup_hugepagesz);
276 
277 static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file,
278 		unsigned long addr, unsigned long len,
279 		unsigned long pgoff, unsigned long flags)
280 {
281 	struct hstate *h = hstate_file(file);
282 	struct vm_unmapped_area_info info;
283 
284 	info.flags = 0;
285 	info.length = len;
286 	info.low_limit = current->mm->mmap_base;
287 	info.high_limit = TASK_SIZE;
288 	info.align_mask = PAGE_MASK & ~huge_page_mask(h);
289 	info.align_offset = 0;
290 	return vm_unmapped_area(&info);
291 }
292 
293 static unsigned long hugetlb_get_unmapped_area_topdown(struct file *file,
294 		unsigned long addr0, unsigned long len,
295 		unsigned long pgoff, unsigned long flags)
296 {
297 	struct hstate *h = hstate_file(file);
298 	struct vm_unmapped_area_info info;
299 	unsigned long addr;
300 
301 	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
302 	info.length = len;
303 	info.low_limit = max(PAGE_SIZE, mmap_min_addr);
304 	info.high_limit = current->mm->mmap_base;
305 	info.align_mask = PAGE_MASK & ~huge_page_mask(h);
306 	info.align_offset = 0;
307 	addr = vm_unmapped_area(&info);
308 
309 	/*
310 	 * A failed mmap() very likely causes application failure,
311 	 * so fall back to the bottom-up function here. This scenario
312 	 * can happen with large stack limits and large mmap()
313 	 * allocations.
314 	 */
315 	if (addr & ~PAGE_MASK) {
316 		VM_BUG_ON(addr != -ENOMEM);
317 		info.flags = 0;
318 		info.low_limit = TASK_UNMAPPED_BASE;
319 		info.high_limit = TASK_SIZE;
320 		addr = vm_unmapped_area(&info);
321 	}
322 
323 	return addr;
324 }
325 
326 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
327 		unsigned long len, unsigned long pgoff, unsigned long flags)
328 {
329 	struct hstate *h = hstate_file(file);
330 	struct mm_struct *mm = current->mm;
331 	struct vm_area_struct *vma;
332 
333 	if (len & ~huge_page_mask(h))
334 		return -EINVAL;
335 	if (len > TASK_SIZE - mmap_min_addr)
336 		return -ENOMEM;
337 
338 	if (flags & MAP_FIXED) {
339 		if (prepare_hugepage_range(file, addr, len))
340 			return -EINVAL;
341 		goto check_asce_limit;
342 	}
343 
344 	if (addr) {
345 		addr = ALIGN(addr, huge_page_size(h));
346 		vma = find_vma(mm, addr);
347 		if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
348 		    (!vma || addr + len <= vm_start_gap(vma)))
349 			goto check_asce_limit;
350 	}
351 
352 	if (mm->get_unmapped_area == arch_get_unmapped_area)
353 		addr = hugetlb_get_unmapped_area_bottomup(file, addr, len,
354 				pgoff, flags);
355 	else
356 		addr = hugetlb_get_unmapped_area_topdown(file, addr, len,
357 				pgoff, flags);
358 	if (offset_in_page(addr))
359 		return addr;
360 
361 check_asce_limit:
362 	return check_asce_limit(mm, addr, len);
363 }
364