]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - mm/hugetlb.c
SLUB: consolidate trace code
[mirror_ubuntu-zesty-kernel.git] / mm / hugetlb.c
1 /*
2 * Generic hugetlb support.
3 * (C) William Irwin, April 2004
4 */
5 #include <linux/gfp.h>
6 #include <linux/list.h>
7 #include <linux/init.h>
8 #include <linux/module.h>
9 #include <linux/mm.h>
10 #include <linux/sysctl.h>
11 #include <linux/highmem.h>
12 #include <linux/nodemask.h>
13 #include <linux/pagemap.h>
14 #include <linux/mempolicy.h>
15 #include <linux/cpuset.h>
16 #include <linux/mutex.h>
17
18 #include <asm/page.h>
19 #include <asm/pgtable.h>
20
21 #include <linux/hugetlb.h>
22 #include "internal.h"
23
24 const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL;
25 static unsigned long nr_huge_pages, free_huge_pages, resv_huge_pages;
26 unsigned long max_huge_pages;
27 static struct list_head hugepage_freelists[MAX_NUMNODES];
28 static unsigned int nr_huge_pages_node[MAX_NUMNODES];
29 static unsigned int free_huge_pages_node[MAX_NUMNODES];
30 /*
31 * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages
32 */
33 static DEFINE_SPINLOCK(hugetlb_lock);
34
35 static void clear_huge_page(struct page *page, unsigned long addr)
36 {
37 int i;
38
39 might_sleep();
40 for (i = 0; i < (HPAGE_SIZE/PAGE_SIZE); i++) {
41 cond_resched();
42 clear_user_highpage(page + i, addr);
43 }
44 }
45
46 static void copy_huge_page(struct page *dst, struct page *src,
47 unsigned long addr, struct vm_area_struct *vma)
48 {
49 int i;
50
51 might_sleep();
52 for (i = 0; i < HPAGE_SIZE/PAGE_SIZE; i++) {
53 cond_resched();
54 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
55 }
56 }
57
58 static void enqueue_huge_page(struct page *page)
59 {
60 int nid = page_to_nid(page);
61 list_add(&page->lru, &hugepage_freelists[nid]);
62 free_huge_pages++;
63 free_huge_pages_node[nid]++;
64 }
65
66 static struct page *dequeue_huge_page(struct vm_area_struct *vma,
67 unsigned long address)
68 {
69 int nid = numa_node_id();
70 struct page *page = NULL;
71 struct zonelist *zonelist = huge_zonelist(vma, address);
72 struct zone **z;
73
74 for (z = zonelist->zones; *z; z++) {
75 nid = zone_to_nid(*z);
76 if (cpuset_zone_allowed_softwall(*z, GFP_HIGHUSER) &&
77 !list_empty(&hugepage_freelists[nid]))
78 break;
79 }
80
81 if (*z) {
82 page = list_entry(hugepage_freelists[nid].next,
83 struct page, lru);
84 list_del(&page->lru);
85 free_huge_pages--;
86 free_huge_pages_node[nid]--;
87 }
88 return page;
89 }
90
91 static void free_huge_page(struct page *page)
92 {
93 BUG_ON(page_count(page));
94
95 INIT_LIST_HEAD(&page->lru);
96
97 spin_lock(&hugetlb_lock);
98 enqueue_huge_page(page);
99 spin_unlock(&hugetlb_lock);
100 }
101
102 static int alloc_fresh_huge_page(void)
103 {
104 static int nid = 0;
105 struct page *page;
106 page = alloc_pages_node(nid, GFP_HIGHUSER|__GFP_COMP|__GFP_NOWARN,
107 HUGETLB_PAGE_ORDER);
108 nid = next_node(nid, node_online_map);
109 if (nid == MAX_NUMNODES)
110 nid = first_node(node_online_map);
111 if (page) {
112 set_compound_page_dtor(page, free_huge_page);
113 spin_lock(&hugetlb_lock);
114 nr_huge_pages++;
115 nr_huge_pages_node[page_to_nid(page)]++;
116 spin_unlock(&hugetlb_lock);
117 put_page(page); /* free it into the hugepage allocator */
118 return 1;
119 }
120 return 0;
121 }
122
123 static struct page *alloc_huge_page(struct vm_area_struct *vma,
124 unsigned long addr)
125 {
126 struct page *page;
127
128 spin_lock(&hugetlb_lock);
129 if (vma->vm_flags & VM_MAYSHARE)
130 resv_huge_pages--;
131 else if (free_huge_pages <= resv_huge_pages)
132 goto fail;
133
134 page = dequeue_huge_page(vma, addr);
135 if (!page)
136 goto fail;
137
138 spin_unlock(&hugetlb_lock);
139 set_page_refcounted(page);
140 return page;
141
142 fail:
143 spin_unlock(&hugetlb_lock);
144 return NULL;
145 }
146
147 static int __init hugetlb_init(void)
148 {
149 unsigned long i;
150
151 if (HPAGE_SHIFT == 0)
152 return 0;
153
154 for (i = 0; i < MAX_NUMNODES; ++i)
155 INIT_LIST_HEAD(&hugepage_freelists[i]);
156
157 for (i = 0; i < max_huge_pages; ++i) {
158 if (!alloc_fresh_huge_page())
159 break;
160 }
161 max_huge_pages = free_huge_pages = nr_huge_pages = i;
162 printk("Total HugeTLB memory allocated, %ld\n", free_huge_pages);
163 return 0;
164 }
165 module_init(hugetlb_init);
166
167 static int __init hugetlb_setup(char *s)
168 {
169 if (sscanf(s, "%lu", &max_huge_pages) <= 0)
170 max_huge_pages = 0;
171 return 1;
172 }
173 __setup("hugepages=", hugetlb_setup);
174
175 #ifdef CONFIG_SYSCTL
176 static void update_and_free_page(struct page *page)
177 {
178 int i;
179 nr_huge_pages--;
180 nr_huge_pages_node[page_to_nid(page)]--;
181 for (i = 0; i < (HPAGE_SIZE / PAGE_SIZE); i++) {
182 page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced |
183 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved |
184 1 << PG_private | 1<< PG_writeback);
185 }
186 page[1].lru.next = NULL;
187 set_page_refcounted(page);
188 __free_pages(page, HUGETLB_PAGE_ORDER);
189 }
190
191 #ifdef CONFIG_HIGHMEM
192 static void try_to_free_low(unsigned long count)
193 {
194 int i;
195
196 for (i = 0; i < MAX_NUMNODES; ++i) {
197 struct page *page, *next;
198 list_for_each_entry_safe(page, next, &hugepage_freelists[i], lru) {
199 if (PageHighMem(page))
200 continue;
201 list_del(&page->lru);
202 update_and_free_page(page);
203 free_huge_pages--;
204 free_huge_pages_node[page_to_nid(page)]--;
205 if (count >= nr_huge_pages)
206 return;
207 }
208 }
209 }
210 #else
211 static inline void try_to_free_low(unsigned long count)
212 {
213 }
214 #endif
215
216 static unsigned long set_max_huge_pages(unsigned long count)
217 {
218 while (count > nr_huge_pages) {
219 if (!alloc_fresh_huge_page())
220 return nr_huge_pages;
221 }
222 if (count >= nr_huge_pages)
223 return nr_huge_pages;
224
225 spin_lock(&hugetlb_lock);
226 count = max(count, resv_huge_pages);
227 try_to_free_low(count);
228 while (count < nr_huge_pages) {
229 struct page *page = dequeue_huge_page(NULL, 0);
230 if (!page)
231 break;
232 update_and_free_page(page);
233 }
234 spin_unlock(&hugetlb_lock);
235 return nr_huge_pages;
236 }
237
238 int hugetlb_sysctl_handler(struct ctl_table *table, int write,
239 struct file *file, void __user *buffer,
240 size_t *length, loff_t *ppos)
241 {
242 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
243 max_huge_pages = set_max_huge_pages(max_huge_pages);
244 return 0;
245 }
246 #endif /* CONFIG_SYSCTL */
247
248 int hugetlb_report_meminfo(char *buf)
249 {
250 return sprintf(buf,
251 "HugePages_Total: %5lu\n"
252 "HugePages_Free: %5lu\n"
253 "HugePages_Rsvd: %5lu\n"
254 "Hugepagesize: %5lu kB\n",
255 nr_huge_pages,
256 free_huge_pages,
257 resv_huge_pages,
258 HPAGE_SIZE/1024);
259 }
260
261 int hugetlb_report_node_meminfo(int nid, char *buf)
262 {
263 return sprintf(buf,
264 "Node %d HugePages_Total: %5u\n"
265 "Node %d HugePages_Free: %5u\n",
266 nid, nr_huge_pages_node[nid],
267 nid, free_huge_pages_node[nid]);
268 }
269
270 /* Return the number pages of memory we physically have, in PAGE_SIZE units. */
271 unsigned long hugetlb_total_pages(void)
272 {
273 return nr_huge_pages * (HPAGE_SIZE / PAGE_SIZE);
274 }
275
276 /*
277 * We cannot handle pagefaults against hugetlb pages at all. They cause
278 * handle_mm_fault() to try to instantiate regular-sized pages in the
279 * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get
280 * this far.
281 */
282 static struct page *hugetlb_nopage(struct vm_area_struct *vma,
283 unsigned long address, int *unused)
284 {
285 BUG();
286 return NULL;
287 }
288
289 struct vm_operations_struct hugetlb_vm_ops = {
290 .nopage = hugetlb_nopage,
291 };
292
293 static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page,
294 int writable)
295 {
296 pte_t entry;
297
298 if (writable) {
299 entry =
300 pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
301 } else {
302 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
303 }
304 entry = pte_mkyoung(entry);
305 entry = pte_mkhuge(entry);
306
307 return entry;
308 }
309
310 static void set_huge_ptep_writable(struct vm_area_struct *vma,
311 unsigned long address, pte_t *ptep)
312 {
313 pte_t entry;
314
315 entry = pte_mkwrite(pte_mkdirty(*ptep));
316 ptep_set_access_flags(vma, address, ptep, entry, 1);
317 update_mmu_cache(vma, address, entry);
318 lazy_mmu_prot_update(entry);
319 }
320
321
322 int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
323 struct vm_area_struct *vma)
324 {
325 pte_t *src_pte, *dst_pte, entry;
326 struct page *ptepage;
327 unsigned long addr;
328 int cow;
329
330 cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
331
332 for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
333 src_pte = huge_pte_offset(src, addr);
334 if (!src_pte)
335 continue;
336 dst_pte = huge_pte_alloc(dst, addr);
337 if (!dst_pte)
338 goto nomem;
339 spin_lock(&dst->page_table_lock);
340 spin_lock(&src->page_table_lock);
341 if (!pte_none(*src_pte)) {
342 if (cow)
343 ptep_set_wrprotect(src, addr, src_pte);
344 entry = *src_pte;
345 ptepage = pte_page(entry);
346 get_page(ptepage);
347 set_huge_pte_at(dst, addr, dst_pte, entry);
348 }
349 spin_unlock(&src->page_table_lock);
350 spin_unlock(&dst->page_table_lock);
351 }
352 return 0;
353
354 nomem:
355 return -ENOMEM;
356 }
357
358 void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
359 unsigned long end)
360 {
361 struct mm_struct *mm = vma->vm_mm;
362 unsigned long address;
363 pte_t *ptep;
364 pte_t pte;
365 struct page *page;
366 struct page *tmp;
367 /*
368 * A page gathering list, protected by per file i_mmap_lock. The
369 * lock is used to avoid list corruption from multiple unmapping
370 * of the same page since we are using page->lru.
371 */
372 LIST_HEAD(page_list);
373
374 WARN_ON(!is_vm_hugetlb_page(vma));
375 BUG_ON(start & ~HPAGE_MASK);
376 BUG_ON(end & ~HPAGE_MASK);
377
378 spin_lock(&mm->page_table_lock);
379 for (address = start; address < end; address += HPAGE_SIZE) {
380 ptep = huge_pte_offset(mm, address);
381 if (!ptep)
382 continue;
383
384 if (huge_pmd_unshare(mm, &address, ptep))
385 continue;
386
387 pte = huge_ptep_get_and_clear(mm, address, ptep);
388 if (pte_none(pte))
389 continue;
390
391 page = pte_page(pte);
392 if (pte_dirty(pte))
393 set_page_dirty(page);
394 list_add(&page->lru, &page_list);
395 }
396 spin_unlock(&mm->page_table_lock);
397 flush_tlb_range(vma, start, end);
398 list_for_each_entry_safe(page, tmp, &page_list, lru) {
399 list_del(&page->lru);
400 put_page(page);
401 }
402 }
403
404 void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
405 unsigned long end)
406 {
407 /*
408 * It is undesirable to test vma->vm_file as it should be non-null
409 * for valid hugetlb area. However, vm_file will be NULL in the error
410 * cleanup path of do_mmap_pgoff. When hugetlbfs ->mmap method fails,
411 * do_mmap_pgoff() nullifies vma->vm_file before calling this function
412 * to clean up. Since no pte has actually been setup, it is safe to
413 * do nothing in this case.
414 */
415 if (vma->vm_file) {
416 spin_lock(&vma->vm_file->f_mapping->i_mmap_lock);
417 __unmap_hugepage_range(vma, start, end);
418 spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock);
419 }
420 }
421
422 static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma,
423 unsigned long address, pte_t *ptep, pte_t pte)
424 {
425 struct page *old_page, *new_page;
426 int avoidcopy;
427
428 old_page = pte_page(pte);
429
430 /* If no-one else is actually using this page, avoid the copy
431 * and just make the page writable */
432 avoidcopy = (page_count(old_page) == 1);
433 if (avoidcopy) {
434 set_huge_ptep_writable(vma, address, ptep);
435 return VM_FAULT_MINOR;
436 }
437
438 page_cache_get(old_page);
439 new_page = alloc_huge_page(vma, address);
440
441 if (!new_page) {
442 page_cache_release(old_page);
443 return VM_FAULT_OOM;
444 }
445
446 spin_unlock(&mm->page_table_lock);
447 copy_huge_page(new_page, old_page, address, vma);
448 spin_lock(&mm->page_table_lock);
449
450 ptep = huge_pte_offset(mm, address & HPAGE_MASK);
451 if (likely(pte_same(*ptep, pte))) {
452 /* Break COW */
453 set_huge_pte_at(mm, address, ptep,
454 make_huge_pte(vma, new_page, 1));
455 /* Make the old page be freed below */
456 new_page = old_page;
457 }
458 page_cache_release(new_page);
459 page_cache_release(old_page);
460 return VM_FAULT_MINOR;
461 }
462
463 int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
464 unsigned long address, pte_t *ptep, int write_access)
465 {
466 int ret = VM_FAULT_SIGBUS;
467 unsigned long idx;
468 unsigned long size;
469 struct page *page;
470 struct address_space *mapping;
471 pte_t new_pte;
472
473 mapping = vma->vm_file->f_mapping;
474 idx = ((address - vma->vm_start) >> HPAGE_SHIFT)
475 + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
476
477 /*
478 * Use page lock to guard against racing truncation
479 * before we get page_table_lock.
480 */
481 retry:
482 page = find_lock_page(mapping, idx);
483 if (!page) {
484 size = i_size_read(mapping->host) >> HPAGE_SHIFT;
485 if (idx >= size)
486 goto out;
487 if (hugetlb_get_quota(mapping))
488 goto out;
489 page = alloc_huge_page(vma, address);
490 if (!page) {
491 hugetlb_put_quota(mapping);
492 ret = VM_FAULT_OOM;
493 goto out;
494 }
495 clear_huge_page(page, address);
496
497 if (vma->vm_flags & VM_SHARED) {
498 int err;
499
500 err = add_to_page_cache(page, mapping, idx, GFP_KERNEL);
501 if (err) {
502 put_page(page);
503 hugetlb_put_quota(mapping);
504 if (err == -EEXIST)
505 goto retry;
506 goto out;
507 }
508 } else
509 lock_page(page);
510 }
511
512 spin_lock(&mm->page_table_lock);
513 size = i_size_read(mapping->host) >> HPAGE_SHIFT;
514 if (idx >= size)
515 goto backout;
516
517 ret = VM_FAULT_MINOR;
518 if (!pte_none(*ptep))
519 goto backout;
520
521 new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE)
522 && (vma->vm_flags & VM_SHARED)));
523 set_huge_pte_at(mm, address, ptep, new_pte);
524
525 if (write_access && !(vma->vm_flags & VM_SHARED)) {
526 /* Optimization, do the COW without a second fault */
527 ret = hugetlb_cow(mm, vma, address, ptep, new_pte);
528 }
529
530 spin_unlock(&mm->page_table_lock);
531 unlock_page(page);
532 out:
533 return ret;
534
535 backout:
536 spin_unlock(&mm->page_table_lock);
537 hugetlb_put_quota(mapping);
538 unlock_page(page);
539 put_page(page);
540 goto out;
541 }
542
543 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
544 unsigned long address, int write_access)
545 {
546 pte_t *ptep;
547 pte_t entry;
548 int ret;
549 static DEFINE_MUTEX(hugetlb_instantiation_mutex);
550
551 ptep = huge_pte_alloc(mm, address);
552 if (!ptep)
553 return VM_FAULT_OOM;
554
555 /*
556 * Serialize hugepage allocation and instantiation, so that we don't
557 * get spurious allocation failures if two CPUs race to instantiate
558 * the same page in the page cache.
559 */
560 mutex_lock(&hugetlb_instantiation_mutex);
561 entry = *ptep;
562 if (pte_none(entry)) {
563 ret = hugetlb_no_page(mm, vma, address, ptep, write_access);
564 mutex_unlock(&hugetlb_instantiation_mutex);
565 return ret;
566 }
567
568 ret = VM_FAULT_MINOR;
569
570 spin_lock(&mm->page_table_lock);
571 /* Check for a racing update before calling hugetlb_cow */
572 if (likely(pte_same(entry, *ptep)))
573 if (write_access && !pte_write(entry))
574 ret = hugetlb_cow(mm, vma, address, ptep, entry);
575 spin_unlock(&mm->page_table_lock);
576 mutex_unlock(&hugetlb_instantiation_mutex);
577
578 return ret;
579 }
580
581 int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
582 struct page **pages, struct vm_area_struct **vmas,
583 unsigned long *position, int *length, int i)
584 {
585 unsigned long pfn_offset;
586 unsigned long vaddr = *position;
587 int remainder = *length;
588
589 spin_lock(&mm->page_table_lock);
590 while (vaddr < vma->vm_end && remainder) {
591 pte_t *pte;
592 struct page *page;
593
594 /*
595 * Some archs (sparc64, sh*) have multiple pte_ts to
596 * each hugepage. We have to make * sure we get the
597 * first, for the page indexing below to work.
598 */
599 pte = huge_pte_offset(mm, vaddr & HPAGE_MASK);
600
601 if (!pte || pte_none(*pte)) {
602 int ret;
603
604 spin_unlock(&mm->page_table_lock);
605 ret = hugetlb_fault(mm, vma, vaddr, 0);
606 spin_lock(&mm->page_table_lock);
607 if (ret == VM_FAULT_MINOR)
608 continue;
609
610 remainder = 0;
611 if (!i)
612 i = -EFAULT;
613 break;
614 }
615
616 pfn_offset = (vaddr & ~HPAGE_MASK) >> PAGE_SHIFT;
617 page = pte_page(*pte);
618 same_page:
619 if (pages) {
620 get_page(page);
621 pages[i] = page + pfn_offset;
622 }
623
624 if (vmas)
625 vmas[i] = vma;
626
627 vaddr += PAGE_SIZE;
628 ++pfn_offset;
629 --remainder;
630 ++i;
631 if (vaddr < vma->vm_end && remainder &&
632 pfn_offset < HPAGE_SIZE/PAGE_SIZE) {
633 /*
634 * We use pfn_offset to avoid touching the pageframes
635 * of this compound page.
636 */
637 goto same_page;
638 }
639 }
640 spin_unlock(&mm->page_table_lock);
641 *length = remainder;
642 *position = vaddr;
643
644 return i;
645 }
646
647 void hugetlb_change_protection(struct vm_area_struct *vma,
648 unsigned long address, unsigned long end, pgprot_t newprot)
649 {
650 struct mm_struct *mm = vma->vm_mm;
651 unsigned long start = address;
652 pte_t *ptep;
653 pte_t pte;
654
655 BUG_ON(address >= end);
656 flush_cache_range(vma, address, end);
657
658 spin_lock(&vma->vm_file->f_mapping->i_mmap_lock);
659 spin_lock(&mm->page_table_lock);
660 for (; address < end; address += HPAGE_SIZE) {
661 ptep = huge_pte_offset(mm, address);
662 if (!ptep)
663 continue;
664 if (huge_pmd_unshare(mm, &address, ptep))
665 continue;
666 if (!pte_none(*ptep)) {
667 pte = huge_ptep_get_and_clear(mm, address, ptep);
668 pte = pte_mkhuge(pte_modify(pte, newprot));
669 set_huge_pte_at(mm, address, ptep, pte);
670 lazy_mmu_prot_update(pte);
671 }
672 }
673 spin_unlock(&mm->page_table_lock);
674 spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock);
675
676 flush_tlb_range(vma, start, end);
677 }
678
679 struct file_region {
680 struct list_head link;
681 long from;
682 long to;
683 };
684
685 static long region_add(struct list_head *head, long f, long t)
686 {
687 struct file_region *rg, *nrg, *trg;
688
689 /* Locate the region we are either in or before. */
690 list_for_each_entry(rg, head, link)
691 if (f <= rg->to)
692 break;
693
694 /* Round our left edge to the current segment if it encloses us. */
695 if (f > rg->from)
696 f = rg->from;
697
698 /* Check for and consume any regions we now overlap with. */
699 nrg = rg;
700 list_for_each_entry_safe(rg, trg, rg->link.prev, link) {
701 if (&rg->link == head)
702 break;
703 if (rg->from > t)
704 break;
705
706 /* If this area reaches higher then extend our area to
707 * include it completely. If this is not the first area
708 * which we intend to reuse, free it. */
709 if (rg->to > t)
710 t = rg->to;
711 if (rg != nrg) {
712 list_del(&rg->link);
713 kfree(rg);
714 }
715 }
716 nrg->from = f;
717 nrg->to = t;
718 return 0;
719 }
720
721 static long region_chg(struct list_head *head, long f, long t)
722 {
723 struct file_region *rg, *nrg;
724 long chg = 0;
725
726 /* Locate the region we are before or in. */
727 list_for_each_entry(rg, head, link)
728 if (f <= rg->to)
729 break;
730
731 /* If we are below the current region then a new region is required.
732 * Subtle, allocate a new region at the position but make it zero
733 * size such that we can guarentee to record the reservation. */
734 if (&rg->link == head || t < rg->from) {
735 nrg = kmalloc(sizeof(*nrg), GFP_KERNEL);
736 if (nrg == 0)
737 return -ENOMEM;
738 nrg->from = f;
739 nrg->to = f;
740 INIT_LIST_HEAD(&nrg->link);
741 list_add(&nrg->link, rg->link.prev);
742
743 return t - f;
744 }
745
746 /* Round our left edge to the current segment if it encloses us. */
747 if (f > rg->from)
748 f = rg->from;
749 chg = t - f;
750
751 /* Check for and consume any regions we now overlap with. */
752 list_for_each_entry(rg, rg->link.prev, link) {
753 if (&rg->link == head)
754 break;
755 if (rg->from > t)
756 return chg;
757
758 /* We overlap with this area, if it extends futher than
759 * us then we must extend ourselves. Account for its
760 * existing reservation. */
761 if (rg->to > t) {
762 chg += rg->to - t;
763 t = rg->to;
764 }
765 chg -= rg->to - rg->from;
766 }
767 return chg;
768 }
769
770 static long region_truncate(struct list_head *head, long end)
771 {
772 struct file_region *rg, *trg;
773 long chg = 0;
774
775 /* Locate the region we are either in or before. */
776 list_for_each_entry(rg, head, link)
777 if (end <= rg->to)
778 break;
779 if (&rg->link == head)
780 return 0;
781
782 /* If we are in the middle of a region then adjust it. */
783 if (end > rg->from) {
784 chg = rg->to - end;
785 rg->to = end;
786 rg = list_entry(rg->link.next, typeof(*rg), link);
787 }
788
789 /* Drop any remaining regions. */
790 list_for_each_entry_safe(rg, trg, rg->link.prev, link) {
791 if (&rg->link == head)
792 break;
793 chg += rg->to - rg->from;
794 list_del(&rg->link);
795 kfree(rg);
796 }
797 return chg;
798 }
799
800 static int hugetlb_acct_memory(long delta)
801 {
802 int ret = -ENOMEM;
803
804 spin_lock(&hugetlb_lock);
805 if ((delta + resv_huge_pages) <= free_huge_pages) {
806 resv_huge_pages += delta;
807 ret = 0;
808 }
809 spin_unlock(&hugetlb_lock);
810 return ret;
811 }
812
813 int hugetlb_reserve_pages(struct inode *inode, long from, long to)
814 {
815 long ret, chg;
816
817 chg = region_chg(&inode->i_mapping->private_list, from, to);
818 if (chg < 0)
819 return chg;
820 ret = hugetlb_acct_memory(chg);
821 if (ret < 0)
822 return ret;
823 region_add(&inode->i_mapping->private_list, from, to);
824 return 0;
825 }
826
827 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed)
828 {
829 long chg = region_truncate(&inode->i_mapping->private_list, offset);
830 hugetlb_acct_memory(freed - chg);
831 }