Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SPARC64 Huge TLB page support.
4 *
5 * Copyright (C) 2002, 2003, 2006 David S. Miller (davem@davemloft.net)
6 */
7
8#include <linux/fs.h>
9#include <linux/mm.h>
10#include <linux/sched/mm.h>
11#include <linux/hugetlb.h>
12#include <linux/pagemap.h>
13#include <linux/sysctl.h>
14
15#include <asm/mman.h>
16#include <asm/pgalloc.h>
17#include <asm/tlb.h>
18#include <asm/tlbflush.h>
19#include <asm/cacheflush.h>
20#include <asm/mmu_context.h>
21
22/* Slightly simplified from the non-hugepage variant because by
23 * definition we don't have to worry about any page coloring stuff
24 */
25
26static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *filp,
27 unsigned long addr,
28 unsigned long len,
29 unsigned long pgoff,
30 unsigned long flags)
31{
32 struct hstate *h = hstate_file(filp);
33 unsigned long task_size = TASK_SIZE;
34 struct vm_unmapped_area_info info;
35
36 if (test_thread_flag(TIF_32BIT))
37 task_size = STACK_TOP32;
38
39 info.flags = 0;
40 info.length = len;
41 info.low_limit = TASK_UNMAPPED_BASE;
42 info.high_limit = min(task_size, VA_EXCLUDE_START);
43 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
44 info.align_offset = 0;
45 addr = vm_unmapped_area(&info);
46
47 if ((addr & ~PAGE_MASK) && task_size > VA_EXCLUDE_END) {
48 VM_BUG_ON(addr != -ENOMEM);
49 info.low_limit = VA_EXCLUDE_END;
50 info.high_limit = task_size;
51 addr = vm_unmapped_area(&info);
52 }
53
54 return addr;
55}
56
57static unsigned long
58hugetlb_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
59 const unsigned long len,
60 const unsigned long pgoff,
61 const unsigned long flags)
62{
63 struct hstate *h = hstate_file(filp);
64 struct mm_struct *mm = current->mm;
65 unsigned long addr = addr0;
66 struct vm_unmapped_area_info info;
67
68 /* This should only ever run for 32-bit processes. */
69 BUG_ON(!test_thread_flag(TIF_32BIT));
70
71 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
72 info.length = len;
73 info.low_limit = PAGE_SIZE;
74 info.high_limit = mm->mmap_base;
75 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
76 info.align_offset = 0;
77 addr = vm_unmapped_area(&info);
78
79 /*
80 * A failed mmap() very likely causes application failure,
81 * so fall back to the bottom-up function here. This scenario
82 * can happen with large stack limits and large mmap()
83 * allocations.
84 */
85 if (addr & ~PAGE_MASK) {
86 VM_BUG_ON(addr != -ENOMEM);
87 info.flags = 0;
88 info.low_limit = TASK_UNMAPPED_BASE;
89 info.high_limit = STACK_TOP32;
90 addr = vm_unmapped_area(&info);
91 }
92
93 return addr;
94}
95
96unsigned long
97hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
98 unsigned long len, unsigned long pgoff, unsigned long flags)
99{
100 struct hstate *h = hstate_file(file);
101 struct mm_struct *mm = current->mm;
102 struct vm_area_struct *vma;
103 unsigned long task_size = TASK_SIZE;
104
105 if (test_thread_flag(TIF_32BIT))
106 task_size = STACK_TOP32;
107
108 if (len & ~huge_page_mask(h))
109 return -EINVAL;
110 if (len > task_size)
111 return -ENOMEM;
112
113 if (flags & MAP_FIXED) {
114 if (prepare_hugepage_range(file, addr, len))
115 return -EINVAL;
116 return addr;
117 }
118
119 if (addr) {
120 addr = ALIGN(addr, huge_page_size(h));
121 vma = find_vma(mm, addr);
122 if (task_size - len >= addr &&
123 (!vma || addr + len <= vm_start_gap(vma)))
124 return addr;
125 }
126 if (mm->get_unmapped_area == arch_get_unmapped_area)
127 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
128 pgoff, flags);
129 else
130 return hugetlb_get_unmapped_area_topdown(file, addr, len,
131 pgoff, flags);
132}
133
134static pte_t sun4u_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
135{
136 return entry;
137}
138
139static pte_t sun4v_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
140{
141 unsigned long hugepage_size = _PAGE_SZ4MB_4V;
142
143 pte_val(entry) = pte_val(entry) & ~_PAGE_SZALL_4V;
144
145 switch (shift) {
146 case HPAGE_16GB_SHIFT:
147 hugepage_size = _PAGE_SZ16GB_4V;
148 pte_val(entry) |= _PAGE_PUD_HUGE;
149 break;
150 case HPAGE_2GB_SHIFT:
151 hugepage_size = _PAGE_SZ2GB_4V;
152 pte_val(entry) |= _PAGE_PMD_HUGE;
153 break;
154 case HPAGE_256MB_SHIFT:
155 hugepage_size = _PAGE_SZ256MB_4V;
156 pte_val(entry) |= _PAGE_PMD_HUGE;
157 break;
158 case HPAGE_SHIFT:
159 pte_val(entry) |= _PAGE_PMD_HUGE;
160 break;
161 case HPAGE_64K_SHIFT:
162 hugepage_size = _PAGE_SZ64K_4V;
163 break;
164 default:
165 WARN_ONCE(1, "unsupported hugepage shift=%u\n", shift);
166 }
167
168 pte_val(entry) = pte_val(entry) | hugepage_size;
169 return entry;
170}
171
172static pte_t hugepage_shift_to_tte(pte_t entry, unsigned int shift)
173{
174 if (tlb_type == hypervisor)
175 return sun4v_hugepage_shift_to_tte(entry, shift);
176 else
177 return sun4u_hugepage_shift_to_tte(entry, shift);
178}
179
180pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
181{
182 pte_t pte;
183
184 entry = pte_mkhuge(entry);
185 pte = hugepage_shift_to_tte(entry, shift);
186
187#ifdef CONFIG_SPARC64
188 /* If this vma has ADI enabled on it, turn on TTE.mcd
189 */
190 if (flags & VM_SPARC_ADI)
191 return pte_mkmcd(pte);
192 else
193 return pte_mknotmcd(pte);
194#else
195 return pte;
196#endif
197}
198
199static unsigned int sun4v_huge_tte_to_shift(pte_t entry)
200{
201 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4V;
202 unsigned int shift;
203
204 switch (tte_szbits) {
205 case _PAGE_SZ16GB_4V:
206 shift = HPAGE_16GB_SHIFT;
207 break;
208 case _PAGE_SZ2GB_4V:
209 shift = HPAGE_2GB_SHIFT;
210 break;
211 case _PAGE_SZ256MB_4V:
212 shift = HPAGE_256MB_SHIFT;
213 break;
214 case _PAGE_SZ4MB_4V:
215 shift = REAL_HPAGE_SHIFT;
216 break;
217 case _PAGE_SZ64K_4V:
218 shift = HPAGE_64K_SHIFT;
219 break;
220 default:
221 shift = PAGE_SHIFT;
222 break;
223 }
224 return shift;
225}
226
227static unsigned int sun4u_huge_tte_to_shift(pte_t entry)
228{
229 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4U;
230 unsigned int shift;
231
232 switch (tte_szbits) {
233 case _PAGE_SZ256MB_4U:
234 shift = HPAGE_256MB_SHIFT;
235 break;
236 case _PAGE_SZ4MB_4U:
237 shift = REAL_HPAGE_SHIFT;
238 break;
239 case _PAGE_SZ64K_4U:
240 shift = HPAGE_64K_SHIFT;
241 break;
242 default:
243 shift = PAGE_SHIFT;
244 break;
245 }
246 return shift;
247}
248
249static unsigned long tte_to_shift(pte_t entry)
250{
251 if (tlb_type == hypervisor)
252 return sun4v_huge_tte_to_shift(entry);
253
254 return sun4u_huge_tte_to_shift(entry);
255}
256
257static unsigned int huge_tte_to_shift(pte_t entry)
258{
259 unsigned long shift = tte_to_shift(entry);
260
261 if (shift == PAGE_SHIFT)
262 WARN_ONCE(1, "tto_to_shift: invalid hugepage tte=0x%lx\n",
263 pte_val(entry));
264
265 return shift;
266}
267
268static unsigned long huge_tte_to_size(pte_t pte)
269{
270 unsigned long size = 1UL << huge_tte_to_shift(pte);
271
272 if (size == REAL_HPAGE_SIZE)
273 size = HPAGE_SIZE;
274 return size;
275}
276
277unsigned long pud_leaf_size(pud_t pud) { return 1UL << tte_to_shift(*(pte_t *)&pud); }
278unsigned long pmd_leaf_size(pmd_t pmd) { return 1UL << tte_to_shift(*(pte_t *)&pmd); }
279unsigned long pte_leaf_size(pte_t pte) { return 1UL << tte_to_shift(pte); }
280
281pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
282 unsigned long addr, unsigned long sz)
283{
284 pgd_t *pgd;
285 p4d_t *p4d;
286 pud_t *pud;
287 pmd_t *pmd;
288
289 pgd = pgd_offset(mm, addr);
290 p4d = p4d_offset(pgd, addr);
291 pud = pud_alloc(mm, p4d, addr);
292 if (!pud)
293 return NULL;
294 if (sz >= PUD_SIZE)
295 return (pte_t *)pud;
296 pmd = pmd_alloc(mm, pud, addr);
297 if (!pmd)
298 return NULL;
299 if (sz >= PMD_SIZE)
300 return (pte_t *)pmd;
301 return pte_alloc_map(mm, pmd, addr);
302}
303
304pte_t *huge_pte_offset(struct mm_struct *mm,
305 unsigned long addr, unsigned long sz)
306{
307 pgd_t *pgd;
308 p4d_t *p4d;
309 pud_t *pud;
310 pmd_t *pmd;
311
312 pgd = pgd_offset(mm, addr);
313 if (pgd_none(*pgd))
314 return NULL;
315 p4d = p4d_offset(pgd, addr);
316 if (p4d_none(*p4d))
317 return NULL;
318 pud = pud_offset(p4d, addr);
319 if (pud_none(*pud))
320 return NULL;
321 if (is_hugetlb_pud(*pud))
322 return (pte_t *)pud;
323 pmd = pmd_offset(pud, addr);
324 if (pmd_none(*pmd))
325 return NULL;
326 if (is_hugetlb_pmd(*pmd))
327 return (pte_t *)pmd;
328 return pte_offset_map(pmd, addr);
329}
330
331void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
332 pte_t *ptep, pte_t entry)
333{
334 unsigned int nptes, orig_shift, shift;
335 unsigned long i, size;
336 pte_t orig;
337
338 size = huge_tte_to_size(entry);
339
340 shift = PAGE_SHIFT;
341 if (size >= PUD_SIZE)
342 shift = PUD_SHIFT;
343 else if (size >= PMD_SIZE)
344 shift = PMD_SHIFT;
345 else
346 shift = PAGE_SHIFT;
347
348 nptes = size >> shift;
349
350 if (!pte_present(*ptep) && pte_present(entry))
351 mm->context.hugetlb_pte_count += nptes;
352
353 addr &= ~(size - 1);
354 orig = *ptep;
355 orig_shift = pte_none(orig) ? PAGE_SHIFT : huge_tte_to_shift(orig);
356
357 for (i = 0; i < nptes; i++)
358 ptep[i] = __pte(pte_val(entry) + (i << shift));
359
360 maybe_tlb_batch_add(mm, addr, ptep, orig, 0, orig_shift);
361 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
362 if (size == HPAGE_SIZE)
363 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0,
364 orig_shift);
365}
366
367pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
368 pte_t *ptep)
369{
370 unsigned int i, nptes, orig_shift, shift;
371 unsigned long size;
372 pte_t entry;
373
374 entry = *ptep;
375 size = huge_tte_to_size(entry);
376
377 shift = PAGE_SHIFT;
378 if (size >= PUD_SIZE)
379 shift = PUD_SHIFT;
380 else if (size >= PMD_SIZE)
381 shift = PMD_SHIFT;
382 else
383 shift = PAGE_SHIFT;
384
385 nptes = size >> shift;
386 orig_shift = pte_none(entry) ? PAGE_SHIFT : huge_tte_to_shift(entry);
387
388 if (pte_present(entry))
389 mm->context.hugetlb_pte_count -= nptes;
390
391 addr &= ~(size - 1);
392 for (i = 0; i < nptes; i++)
393 ptep[i] = __pte(0UL);
394
395 maybe_tlb_batch_add(mm, addr, ptep, entry, 0, orig_shift);
396 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
397 if (size == HPAGE_SIZE)
398 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0,
399 orig_shift);
400
401 return entry;
402}
403
404int pmd_huge(pmd_t pmd)
405{
406 return !pmd_none(pmd) &&
407 (pmd_val(pmd) & (_PAGE_VALID|_PAGE_PMD_HUGE)) != _PAGE_VALID;
408}
409
410int pud_huge(pud_t pud)
411{
412 return !pud_none(pud) &&
413 (pud_val(pud) & (_PAGE_VALID|_PAGE_PUD_HUGE)) != _PAGE_VALID;
414}
415
416static void hugetlb_free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
417 unsigned long addr)
418{
419 pgtable_t token = pmd_pgtable(*pmd);
420
421 pmd_clear(pmd);
422 pte_free_tlb(tlb, token, addr);
423 mm_dec_nr_ptes(tlb->mm);
424}
425
426static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
427 unsigned long addr, unsigned long end,
428 unsigned long floor, unsigned long ceiling)
429{
430 pmd_t *pmd;
431 unsigned long next;
432 unsigned long start;
433
434 start = addr;
435 pmd = pmd_offset(pud, addr);
436 do {
437 next = pmd_addr_end(addr, end);
438 if (pmd_none(*pmd))
439 continue;
440 if (is_hugetlb_pmd(*pmd))
441 pmd_clear(pmd);
442 else
443 hugetlb_free_pte_range(tlb, pmd, addr);
444 } while (pmd++, addr = next, addr != end);
445
446 start &= PUD_MASK;
447 if (start < floor)
448 return;
449 if (ceiling) {
450 ceiling &= PUD_MASK;
451 if (!ceiling)
452 return;
453 }
454 if (end - 1 > ceiling - 1)
455 return;
456
457 pmd = pmd_offset(pud, start);
458 pud_clear(pud);
459 pmd_free_tlb(tlb, pmd, start);
460 mm_dec_nr_pmds(tlb->mm);
461}
462
463static void hugetlb_free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
464 unsigned long addr, unsigned long end,
465 unsigned long floor, unsigned long ceiling)
466{
467 pud_t *pud;
468 unsigned long next;
469 unsigned long start;
470
471 start = addr;
472 pud = pud_offset(p4d, addr);
473 do {
474 next = pud_addr_end(addr, end);
475 if (pud_none_or_clear_bad(pud))
476 continue;
477 if (is_hugetlb_pud(*pud))
478 pud_clear(pud);
479 else
480 hugetlb_free_pmd_range(tlb, pud, addr, next, floor,
481 ceiling);
482 } while (pud++, addr = next, addr != end);
483
484 start &= PGDIR_MASK;
485 if (start < floor)
486 return;
487 if (ceiling) {
488 ceiling &= PGDIR_MASK;
489 if (!ceiling)
490 return;
491 }
492 if (end - 1 > ceiling - 1)
493 return;
494
495 pud = pud_offset(p4d, start);
496 p4d_clear(p4d);
497 pud_free_tlb(tlb, pud, start);
498 mm_dec_nr_puds(tlb->mm);
499}
500
501void hugetlb_free_pgd_range(struct mmu_gather *tlb,
502 unsigned long addr, unsigned long end,
503 unsigned long floor, unsigned long ceiling)
504{
505 pgd_t *pgd;
506 p4d_t *p4d;
507 unsigned long next;
508
509 addr &= PMD_MASK;
510 if (addr < floor) {
511 addr += PMD_SIZE;
512 if (!addr)
513 return;
514 }
515 if (ceiling) {
516 ceiling &= PMD_MASK;
517 if (!ceiling)
518 return;
519 }
520 if (end - 1 > ceiling - 1)
521 end -= PMD_SIZE;
522 if (addr > end - 1)
523 return;
524
525 pgd = pgd_offset(tlb->mm, addr);
526 p4d = p4d_offset(pgd, addr);
527 do {
528 next = p4d_addr_end(addr, end);
529 if (p4d_none_or_clear_bad(p4d))
530 continue;
531 hugetlb_free_pud_range(tlb, p4d, addr, next, floor, ceiling);
532 } while (p4d++, addr = next, addr != end);
533}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * SPARC64 Huge TLB page support.
4 *
5 * Copyright (C) 2002, 2003, 2006 David S. Miller (davem@davemloft.net)
6 */
7
8#include <linux/fs.h>
9#include <linux/mm.h>
10#include <linux/sched/mm.h>
11#include <linux/hugetlb.h>
12#include <linux/pagemap.h>
13#include <linux/sysctl.h>
14
15#include <asm/mman.h>
16#include <asm/pgalloc.h>
17#include <asm/tlb.h>
18#include <asm/tlbflush.h>
19#include <asm/cacheflush.h>
20#include <asm/mmu_context.h>
21
22
23static pte_t sun4u_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
24{
25 return entry;
26}
27
28static pte_t sun4v_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
29{
30 unsigned long hugepage_size = _PAGE_SZ4MB_4V;
31
32 pte_val(entry) = pte_val(entry) & ~_PAGE_SZALL_4V;
33
34 switch (shift) {
35 case HPAGE_16GB_SHIFT:
36 hugepage_size = _PAGE_SZ16GB_4V;
37 pte_val(entry) |= _PAGE_PUD_HUGE;
38 break;
39 case HPAGE_2GB_SHIFT:
40 hugepage_size = _PAGE_SZ2GB_4V;
41 pte_val(entry) |= _PAGE_PMD_HUGE;
42 break;
43 case HPAGE_256MB_SHIFT:
44 hugepage_size = _PAGE_SZ256MB_4V;
45 pte_val(entry) |= _PAGE_PMD_HUGE;
46 break;
47 case HPAGE_SHIFT:
48 pte_val(entry) |= _PAGE_PMD_HUGE;
49 break;
50 case HPAGE_64K_SHIFT:
51 hugepage_size = _PAGE_SZ64K_4V;
52 break;
53 default:
54 WARN_ONCE(1, "unsupported hugepage shift=%u\n", shift);
55 }
56
57 pte_val(entry) = pte_val(entry) | hugepage_size;
58 return entry;
59}
60
61static pte_t hugepage_shift_to_tte(pte_t entry, unsigned int shift)
62{
63 if (tlb_type == hypervisor)
64 return sun4v_hugepage_shift_to_tte(entry, shift);
65 else
66 return sun4u_hugepage_shift_to_tte(entry, shift);
67}
68
69pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
70{
71 pte_t pte;
72
73 entry = pte_mkhuge(entry);
74 pte = hugepage_shift_to_tte(entry, shift);
75
76#ifdef CONFIG_SPARC64
77 /* If this vma has ADI enabled on it, turn on TTE.mcd
78 */
79 if (flags & VM_SPARC_ADI)
80 return pte_mkmcd(pte);
81 else
82 return pte_mknotmcd(pte);
83#else
84 return pte;
85#endif
86}
87
88static unsigned int sun4v_huge_tte_to_shift(pte_t entry)
89{
90 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4V;
91 unsigned int shift;
92
93 switch (tte_szbits) {
94 case _PAGE_SZ16GB_4V:
95 shift = HPAGE_16GB_SHIFT;
96 break;
97 case _PAGE_SZ2GB_4V:
98 shift = HPAGE_2GB_SHIFT;
99 break;
100 case _PAGE_SZ256MB_4V:
101 shift = HPAGE_256MB_SHIFT;
102 break;
103 case _PAGE_SZ4MB_4V:
104 shift = REAL_HPAGE_SHIFT;
105 break;
106 case _PAGE_SZ64K_4V:
107 shift = HPAGE_64K_SHIFT;
108 break;
109 default:
110 shift = PAGE_SHIFT;
111 break;
112 }
113 return shift;
114}
115
116static unsigned int sun4u_huge_tte_to_shift(pte_t entry)
117{
118 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4U;
119 unsigned int shift;
120
121 switch (tte_szbits) {
122 case _PAGE_SZ256MB_4U:
123 shift = HPAGE_256MB_SHIFT;
124 break;
125 case _PAGE_SZ4MB_4U:
126 shift = REAL_HPAGE_SHIFT;
127 break;
128 case _PAGE_SZ64K_4U:
129 shift = HPAGE_64K_SHIFT;
130 break;
131 default:
132 shift = PAGE_SHIFT;
133 break;
134 }
135 return shift;
136}
137
138static unsigned long tte_to_shift(pte_t entry)
139{
140 if (tlb_type == hypervisor)
141 return sun4v_huge_tte_to_shift(entry);
142
143 return sun4u_huge_tte_to_shift(entry);
144}
145
146static unsigned int huge_tte_to_shift(pte_t entry)
147{
148 unsigned long shift = tte_to_shift(entry);
149
150 if (shift == PAGE_SHIFT)
151 WARN_ONCE(1, "tto_to_shift: invalid hugepage tte=0x%lx\n",
152 pte_val(entry));
153
154 return shift;
155}
156
157static unsigned long huge_tte_to_size(pte_t pte)
158{
159 unsigned long size = 1UL << huge_tte_to_shift(pte);
160
161 if (size == REAL_HPAGE_SIZE)
162 size = HPAGE_SIZE;
163 return size;
164}
165
166unsigned long pud_leaf_size(pud_t pud) { return 1UL << tte_to_shift(*(pte_t *)&pud); }
167unsigned long pmd_leaf_size(pmd_t pmd) { return 1UL << tte_to_shift(*(pte_t *)&pmd); }
168unsigned long pte_leaf_size(pte_t pte) { return 1UL << tte_to_shift(pte); }
169
170pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
171 unsigned long addr, unsigned long sz)
172{
173 pgd_t *pgd;
174 p4d_t *p4d;
175 pud_t *pud;
176 pmd_t *pmd;
177
178 pgd = pgd_offset(mm, addr);
179 p4d = p4d_offset(pgd, addr);
180 pud = pud_alloc(mm, p4d, addr);
181 if (!pud)
182 return NULL;
183 if (sz >= PUD_SIZE)
184 return (pte_t *)pud;
185 pmd = pmd_alloc(mm, pud, addr);
186 if (!pmd)
187 return NULL;
188 if (sz >= PMD_SIZE)
189 return (pte_t *)pmd;
190 return pte_alloc_huge(mm, pmd, addr);
191}
192
193pte_t *huge_pte_offset(struct mm_struct *mm,
194 unsigned long addr, unsigned long sz)
195{
196 pgd_t *pgd;
197 p4d_t *p4d;
198 pud_t *pud;
199 pmd_t *pmd;
200
201 pgd = pgd_offset(mm, addr);
202 if (pgd_none(*pgd))
203 return NULL;
204 p4d = p4d_offset(pgd, addr);
205 if (p4d_none(*p4d))
206 return NULL;
207 pud = pud_offset(p4d, addr);
208 if (pud_none(*pud))
209 return NULL;
210 if (is_hugetlb_pud(*pud))
211 return (pte_t *)pud;
212 pmd = pmd_offset(pud, addr);
213 if (pmd_none(*pmd))
214 return NULL;
215 if (is_hugetlb_pmd(*pmd))
216 return (pte_t *)pmd;
217 return pte_offset_huge(pmd, addr);
218}
219
220void __set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
221 pte_t *ptep, pte_t entry)
222{
223 unsigned int nptes, orig_shift, shift;
224 unsigned long i, size;
225 pte_t orig;
226
227 size = huge_tte_to_size(entry);
228
229 shift = PAGE_SHIFT;
230 if (size >= PUD_SIZE)
231 shift = PUD_SHIFT;
232 else if (size >= PMD_SIZE)
233 shift = PMD_SHIFT;
234 else
235 shift = PAGE_SHIFT;
236
237 nptes = size >> shift;
238
239 if (!pte_present(*ptep) && pte_present(entry))
240 mm->context.hugetlb_pte_count += nptes;
241
242 addr &= ~(size - 1);
243 orig = *ptep;
244 orig_shift = pte_none(orig) ? PAGE_SHIFT : huge_tte_to_shift(orig);
245
246 for (i = 0; i < nptes; i++)
247 ptep[i] = __pte(pte_val(entry) + (i << shift));
248
249 maybe_tlb_batch_add(mm, addr, ptep, orig, 0, orig_shift);
250 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
251 if (size == HPAGE_SIZE)
252 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0,
253 orig_shift);
254}
255
256void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
257 pte_t *ptep, pte_t entry, unsigned long sz)
258{
259 __set_huge_pte_at(mm, addr, ptep, entry);
260}
261
262pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
263 pte_t *ptep, unsigned long sz)
264{
265 unsigned int i, nptes, orig_shift, shift;
266 unsigned long size;
267 pte_t entry;
268
269 entry = *ptep;
270 size = huge_tte_to_size(entry);
271
272 shift = PAGE_SHIFT;
273 if (size >= PUD_SIZE)
274 shift = PUD_SHIFT;
275 else if (size >= PMD_SIZE)
276 shift = PMD_SHIFT;
277 else
278 shift = PAGE_SHIFT;
279
280 nptes = size >> shift;
281 orig_shift = pte_none(entry) ? PAGE_SHIFT : huge_tte_to_shift(entry);
282
283 if (pte_present(entry))
284 mm->context.hugetlb_pte_count -= nptes;
285
286 addr &= ~(size - 1);
287 for (i = 0; i < nptes; i++)
288 ptep[i] = __pte(0UL);
289
290 maybe_tlb_batch_add(mm, addr, ptep, entry, 0, orig_shift);
291 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
292 if (size == HPAGE_SIZE)
293 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0,
294 orig_shift);
295
296 return entry;
297}
298
299static void hugetlb_free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
300 unsigned long addr)
301{
302 pgtable_t token = pmd_pgtable(*pmd);
303
304 pmd_clear(pmd);
305 pte_free_tlb(tlb, token, addr);
306 mm_dec_nr_ptes(tlb->mm);
307}
308
309static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
310 unsigned long addr, unsigned long end,
311 unsigned long floor, unsigned long ceiling)
312{
313 pmd_t *pmd;
314 unsigned long next;
315 unsigned long start;
316
317 start = addr;
318 pmd = pmd_offset(pud, addr);
319 do {
320 next = pmd_addr_end(addr, end);
321 if (pmd_none(*pmd))
322 continue;
323 if (is_hugetlb_pmd(*pmd))
324 pmd_clear(pmd);
325 else
326 hugetlb_free_pte_range(tlb, pmd, addr);
327 } while (pmd++, addr = next, addr != end);
328
329 start &= PUD_MASK;
330 if (start < floor)
331 return;
332 if (ceiling) {
333 ceiling &= PUD_MASK;
334 if (!ceiling)
335 return;
336 }
337 if (end - 1 > ceiling - 1)
338 return;
339
340 pmd = pmd_offset(pud, start);
341 pud_clear(pud);
342 pmd_free_tlb(tlb, pmd, start);
343 mm_dec_nr_pmds(tlb->mm);
344}
345
346static void hugetlb_free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
347 unsigned long addr, unsigned long end,
348 unsigned long floor, unsigned long ceiling)
349{
350 pud_t *pud;
351 unsigned long next;
352 unsigned long start;
353
354 start = addr;
355 pud = pud_offset(p4d, addr);
356 do {
357 next = pud_addr_end(addr, end);
358 if (pud_none_or_clear_bad(pud))
359 continue;
360 if (is_hugetlb_pud(*pud))
361 pud_clear(pud);
362 else
363 hugetlb_free_pmd_range(tlb, pud, addr, next, floor,
364 ceiling);
365 } while (pud++, addr = next, addr != end);
366
367 start &= PGDIR_MASK;
368 if (start < floor)
369 return;
370 if (ceiling) {
371 ceiling &= PGDIR_MASK;
372 if (!ceiling)
373 return;
374 }
375 if (end - 1 > ceiling - 1)
376 return;
377
378 pud = pud_offset(p4d, start);
379 p4d_clear(p4d);
380 pud_free_tlb(tlb, pud, start);
381 mm_dec_nr_puds(tlb->mm);
382}
383
384void hugetlb_free_pgd_range(struct mmu_gather *tlb,
385 unsigned long addr, unsigned long end,
386 unsigned long floor, unsigned long ceiling)
387{
388 pgd_t *pgd;
389 p4d_t *p4d;
390 unsigned long next;
391
392 addr &= PMD_MASK;
393 if (addr < floor) {
394 addr += PMD_SIZE;
395 if (!addr)
396 return;
397 }
398 if (ceiling) {
399 ceiling &= PMD_MASK;
400 if (!ceiling)
401 return;
402 }
403 if (end - 1 > ceiling - 1)
404 end -= PMD_SIZE;
405 if (addr > end - 1)
406 return;
407
408 pgd = pgd_offset(tlb->mm, addr);
409 p4d = p4d_offset(pgd, addr);
410 do {
411 next = p4d_addr_end(addr, end);
412 if (p4d_none_or_clear_bad(p4d))
413 continue;
414 hugetlb_free_pud_range(tlb, p4d, addr, next, floor, ceiling);
415 } while (p4d++, addr = next, addr != end);
416}