Lines Matching +full:chg +full:- +full:status
1 // SPDX-License-Identifier: GPL-2.0-only
83 return page_private(head + 4) == -1UL; in PageHugeFreed()
88 set_page_private(head + 4, -1UL); in SetPageHugeFreed()
101 bool free = (spool->count == 0) && (spool->used_hpages == 0); in unlock_or_release_subpool()
103 spin_unlock(&spool->lock); in unlock_or_release_subpool()
109 if (spool->min_hpages != -1) in unlock_or_release_subpool()
110 hugetlb_acct_memory(spool->hstate, in unlock_or_release_subpool()
111 -spool->min_hpages); in unlock_or_release_subpool()
125 spin_lock_init(&spool->lock); in hugepage_new_subpool()
126 spool->count = 1; in hugepage_new_subpool()
127 spool->max_hpages = max_hpages; in hugepage_new_subpool()
128 spool->hstate = h; in hugepage_new_subpool()
129 spool->min_hpages = min_hpages; in hugepage_new_subpool()
131 if (min_hpages != -1 && hugetlb_acct_memory(h, min_hpages)) { in hugepage_new_subpool()
135 spool->rsv_hpages = min_hpages; in hugepage_new_subpool()
142 spin_lock(&spool->lock); in hugepage_put_subpool()
143 BUG_ON(!spool->count); in hugepage_put_subpool()
144 spool->count--; in hugepage_put_subpool()
150 * Return -ENOMEM if there are not enough resources to satisfy the
164 spin_lock(&spool->lock); in hugepage_subpool_get_pages()
166 if (spool->max_hpages != -1) { /* maximum size accounting */ in hugepage_subpool_get_pages()
167 if ((spool->used_hpages + delta) <= spool->max_hpages) in hugepage_subpool_get_pages()
168 spool->used_hpages += delta; in hugepage_subpool_get_pages()
170 ret = -ENOMEM; in hugepage_subpool_get_pages()
176 if (spool->min_hpages != -1 && spool->rsv_hpages) { in hugepage_subpool_get_pages()
177 if (delta > spool->rsv_hpages) { in hugepage_subpool_get_pages()
182 ret = delta - spool->rsv_hpages; in hugepage_subpool_get_pages()
183 spool->rsv_hpages = 0; in hugepage_subpool_get_pages()
186 spool->rsv_hpages -= delta; in hugepage_subpool_get_pages()
191 spin_unlock(&spool->lock); in hugepage_subpool_get_pages()
209 spin_lock(&spool->lock); in hugepage_subpool_put_pages()
211 if (spool->max_hpages != -1) /* maximum size accounting */ in hugepage_subpool_put_pages()
212 spool->used_hpages -= delta; in hugepage_subpool_put_pages()
215 if (spool->min_hpages != -1 && spool->used_hpages < spool->min_hpages) { in hugepage_subpool_put_pages()
216 if (spool->rsv_hpages + delta <= spool->min_hpages) in hugepage_subpool_put_pages()
219 ret = spool->rsv_hpages + delta - spool->min_hpages; in hugepage_subpool_put_pages()
221 spool->rsv_hpages += delta; in hugepage_subpool_put_pages()
222 if (spool->rsv_hpages > spool->min_hpages) in hugepage_subpool_put_pages()
223 spool->rsv_hpages = spool->min_hpages; in hugepage_subpool_put_pages()
237 return HUGETLBFS_SB(inode->i_sb)->spool; in subpool_inode()
242 return subpool_inode(file_inode(vma->vm_file)); in subpool_vma()
253 VM_BUG_ON(resv->region_cache_count <= 0); in get_file_region_entry_from_cache()
255 resv->region_cache_count--; in get_file_region_entry_from_cache()
256 nrg = list_first_entry(&resv->region_cache, struct file_region, link); in get_file_region_entry_from_cache()
257 list_del(&nrg->link); in get_file_region_entry_from_cache()
259 nrg->from = from; in get_file_region_entry_from_cache()
260 nrg->to = to; in get_file_region_entry_from_cache()
269 nrg->reservation_counter = rg->reservation_counter; in copy_hugetlb_cgroup_uncharge_info()
270 nrg->css = rg->css; in copy_hugetlb_cgroup_uncharge_info()
271 if (rg->css) in copy_hugetlb_cgroup_uncharge_info()
272 css_get(rg->css); in copy_hugetlb_cgroup_uncharge_info()
284 nrg->reservation_counter = in record_hugetlb_cgroup_uncharge_info()
285 &h_cg->rsvd_hugepage[hstate_index(h)]; in record_hugetlb_cgroup_uncharge_info()
286 nrg->css = &h_cg->css; in record_hugetlb_cgroup_uncharge_info()
288 * The caller will hold exactly one h_cg->css reference for the in record_hugetlb_cgroup_uncharge_info()
293 * exactly one h_cg->css reference, we should do css_get for in record_hugetlb_cgroup_uncharge_info()
297 css_get(&h_cg->css); in record_hugetlb_cgroup_uncharge_info()
298 if (!resv->pages_per_hpage) in record_hugetlb_cgroup_uncharge_info()
299 resv->pages_per_hpage = pages_per_huge_page(h); in record_hugetlb_cgroup_uncharge_info()
303 VM_BUG_ON(resv->pages_per_hpage != pages_per_huge_page(h)); in record_hugetlb_cgroup_uncharge_info()
305 nrg->reservation_counter = NULL; in record_hugetlb_cgroup_uncharge_info()
306 nrg->css = NULL; in record_hugetlb_cgroup_uncharge_info()
314 if (rg->css) in put_uncharge_info()
315 css_put(rg->css); in put_uncharge_info()
324 rg->reservation_counter == org->reservation_counter && in has_same_uncharge_info()
325 rg->css == org->css; in has_same_uncharge_info()
337 if (&prg->link != &resv->regions && prg->to == rg->from && in coalesce_file_region()
339 prg->to = rg->to; in coalesce_file_region()
341 list_del(&rg->link); in coalesce_file_region()
349 if (&nrg->link != &resv->regions && nrg->from == rg->to && in coalesce_file_region()
351 nrg->from = rg->from; in coalesce_file_region()
353 list_del(&rg->link); in coalesce_file_region()
360 * Must be called with resv->lock held.
372 struct list_head *head = &resv->regions; in add_reservation_in_range()
380 * [last_accounted_offset, rg->from), at every iteration, with some in add_reservation_in_range()
385 if (rg->from < f) { in add_reservation_in_range()
389 if (rg->to > last_accounted_offset) in add_reservation_in_range()
390 last_accounted_offset = rg->to; in add_reservation_in_range()
397 if (rg->from > t) in add_reservation_in_range()
400 /* Add an entry for last_accounted_offset -> rg->from, and in add_reservation_in_range()
403 if (rg->from > last_accounted_offset) { in add_reservation_in_range()
404 add += rg->from - last_accounted_offset; in add_reservation_in_range()
407 resv, last_accounted_offset, rg->from); in add_reservation_in_range()
410 list_add(&nrg->link, rg->link.prev); in add_reservation_in_range()
416 last_accounted_offset = rg->to; in add_reservation_in_range()
423 add += t - last_accounted_offset; in add_reservation_in_range()
428 list_add(&nrg->link, rg->link.prev); in add_reservation_in_range()
438 /* Must be called with resv->lock acquired. Will drop lock to allocate entries.
442 __must_hold(&resv->lock) in allocate_file_region_entries()
461 while (resv->region_cache_count < in allocate_file_region_entries()
462 (resv->adds_in_progress + regions_needed)) { in allocate_file_region_entries()
463 to_allocate = resv->adds_in_progress + regions_needed - in allocate_file_region_entries()
464 resv->region_cache_count; in allocate_file_region_entries()
470 VM_BUG_ON(resv->region_cache_count < resv->adds_in_progress); in allocate_file_region_entries()
472 spin_unlock(&resv->lock); in allocate_file_region_entries()
477 list_add(&trg->link, &allocated_regions); in allocate_file_region_entries()
480 spin_lock(&resv->lock); in allocate_file_region_entries()
482 list_splice(&allocated_regions, &resv->region_cache); in allocate_file_region_entries()
483 resv->region_cache_count += to_allocate; in allocate_file_region_entries()
490 list_del(&rg->link); in allocate_file_region_entries()
493 return -ENOMEM; in allocate_file_region_entries()
508 * this operation and we were not able to allocate, it returns -ENOMEM.
519 spin_lock(&resv->lock); in region_add()
536 resv->region_cache_count < in region_add()
537 resv->adds_in_progress + in region_add()
538 (actual_regions_needed - in_regions_needed)) { in region_add()
542 VM_BUG_ON(t - f <= 1); in region_add()
545 resv, actual_regions_needed - in_regions_needed)) { in region_add()
546 return -ENOMEM; in region_add()
554 resv->adds_in_progress -= in_regions_needed; in region_add()
556 spin_unlock(&resv->lock); in region_add()
573 * resv->adds_in_progress. This value needs to be provided to a follow up call
578 * zero. -ENOMEM is returned if a new file_region structure or cache entry
584 long chg = 0; in region_chg() local
586 spin_lock(&resv->lock); in region_chg()
589 chg = add_reservation_in_range(resv, f, t, NULL, NULL, in region_chg()
596 return -ENOMEM; in region_chg()
598 resv->adds_in_progress += *out_regions_needed; in region_chg()
600 spin_unlock(&resv->lock); in region_chg()
601 return chg; in region_chg()
620 spin_lock(&resv->lock); in region_abort()
621 VM_BUG_ON(!resv->region_cache_count); in region_abort()
622 resv->adds_in_progress -= regions_needed; in region_abort()
623 spin_unlock(&resv->lock); in region_abort()
635 * be allocated. If the allocation fails, -ENOMEM will be returned.
637 * a region and possibly return -ENOMEM. Callers specifying
638 * t == LONG_MAX do not need to check for -ENOMEM error.
642 struct list_head *head = &resv->regions; in region_del()
648 spin_lock(&resv->lock); in region_del()
657 if (rg->to <= f && (rg->to != rg->from || rg->to != f)) in region_del()
660 if (rg->from >= t) in region_del()
663 if (f > rg->from && t < rg->to) { /* Must split region */ in region_del()
669 resv->region_cache_count > resv->adds_in_progress) { in region_del()
670 nrg = list_first_entry(&resv->region_cache, in region_del()
673 list_del(&nrg->link); in region_del()
674 resv->region_cache_count--; in region_del()
678 spin_unlock(&resv->lock); in region_del()
681 return -ENOMEM; in region_del()
685 del += t - f; in region_del()
687 resv, rg, t - f, false); in region_del()
690 nrg->from = t; in region_del()
691 nrg->to = rg->to; in region_del()
695 INIT_LIST_HEAD(&nrg->link); in region_del()
698 rg->to = f; in region_del()
700 list_add(&nrg->link, &rg->link); in region_del()
705 if (f <= rg->from && t >= rg->to) { /* Remove entire region */ in region_del()
706 del += rg->to - rg->from; in region_del()
708 rg->to - rg->from, true); in region_del()
709 list_del(&rg->link); in region_del()
714 if (f <= rg->from) { /* Trim beginning of region */ in region_del()
716 t - rg->from, false); in region_del()
718 del += t - rg->from; in region_del()
719 rg->from = t; in region_del()
722 rg->to - f, false); in region_del()
724 del += rg->to - f; in region_del()
725 rg->to = f; in region_del()
729 spin_unlock(&resv->lock); in region_del()
769 struct list_head *head = &resv->regions; in region_count()
771 long chg = 0; in region_count() local
773 spin_lock(&resv->lock); in region_count()
779 if (rg->to <= f) in region_count()
781 if (rg->from >= t) in region_count()
784 seg_from = max(rg->from, f); in region_count()
785 seg_to = min(rg->to, t); in region_count()
787 chg += seg_to - seg_from; in region_count()
789 spin_unlock(&resv->lock); in region_count()
791 return chg; in region_count()
801 return ((address - vma->vm_start) >> huge_page_shift(h)) + in vma_hugecache_offset()
802 (vma->vm_pgoff >> huge_page_order(h)); in vma_hugecache_offset()
818 if (vma->vm_ops && vma->vm_ops->pagesize) in vma_kernel_pagesize()
819 return vma->vm_ops->pagesize(vma); in vma_kernel_pagesize()
827 * architectures where it differs, an architecture-specific 'strong'
865 return (unsigned long)vma->vm_private_data; in get_vma_private_data()
871 vma->vm_private_data = (void *)value; in set_vma_private_data()
881 resv_map->reservation_counter = NULL; in resv_map_set_hugetlb_cgroup_uncharge_info()
882 resv_map->pages_per_hpage = 0; in resv_map_set_hugetlb_cgroup_uncharge_info()
883 resv_map->css = NULL; in resv_map_set_hugetlb_cgroup_uncharge_info()
885 resv_map->reservation_counter = in resv_map_set_hugetlb_cgroup_uncharge_info()
886 &h_cg->rsvd_hugepage[hstate_index(h)]; in resv_map_set_hugetlb_cgroup_uncharge_info()
887 resv_map->pages_per_hpage = pages_per_huge_page(h); in resv_map_set_hugetlb_cgroup_uncharge_info()
888 resv_map->css = &h_cg->css; in resv_map_set_hugetlb_cgroup_uncharge_info()
904 kref_init(&resv_map->refs); in resv_map_alloc()
905 spin_lock_init(&resv_map->lock); in resv_map_alloc()
906 INIT_LIST_HEAD(&resv_map->regions); in resv_map_alloc()
908 resv_map->adds_in_progress = 0; in resv_map_alloc()
912 * re-initialized to the proper values, to indicate that hugetlb cgroup in resv_map_alloc()
913 * reservations are to be un-charged from here. in resv_map_alloc()
917 INIT_LIST_HEAD(&resv_map->region_cache); in resv_map_alloc()
918 list_add(&rg->link, &resv_map->region_cache); in resv_map_alloc()
919 resv_map->region_cache_count = 1; in resv_map_alloc()
927 struct list_head *head = &resv_map->region_cache; in resv_map_release()
935 list_del(&rg->link); in resv_map_release()
939 VM_BUG_ON(resv_map->adds_in_progress); in resv_map_release()
951 * The VERY common case is inode->mapping == &inode->i_data but, in inode_resv_map()
954 return (struct resv_map *)(&inode->i_data)->private_data; in inode_resv_map()
960 if (vma->vm_flags & VM_MAYSHARE) { in vma_resv_map()
961 struct address_space *mapping = vma->vm_file->f_mapping; in vma_resv_map()
962 struct inode *inode = mapping->host; in vma_resv_map()
975 VM_BUG_ON_VMA(vma->vm_flags & VM_MAYSHARE, vma); in set_vma_resv_map()
984 VM_BUG_ON_VMA(vma->vm_flags & VM_MAYSHARE, vma); in set_vma_resv_flags()
1000 if (!(vma->vm_flags & VM_MAYSHARE)) in reset_vma_resv_huge_pages()
1001 vma->vm_private_data = (void *)0; in reset_vma_resv_huge_pages()
1005 static bool vma_has_reserves(struct vm_area_struct *vma, long chg) in vma_has_reserves() argument
1007 if (vma->vm_flags & VM_NORESERVE) { in vma_has_reserves()
1009 * This address is already reserved by other process(chg == 0), in vma_has_reserves()
1015 * properly, so add work-around here. in vma_has_reserves()
1017 if (vma->vm_flags & VM_MAYSHARE && chg == 0) in vma_has_reserves()
1024 if (vma->vm_flags & VM_MAYSHARE) { in vma_has_reserves()
1030 * use. This situation is indicated if chg != 0. in vma_has_reserves()
1032 if (chg) in vma_has_reserves()
1046 * Examine the value of chg to determine if reserves in vma_has_reserves()
1048 * Very Subtle - The value of chg comes from a previous in vma_has_reserves()
1053 * account. Therefore, the meaning of chg is the same in vma_has_reserves()
1058 if (chg) in vma_has_reserves()
1070 list_move(&page->lru, &h->hugepage_freelists[nid]); in enqueue_huge_page()
1071 h->free_huge_pages++; in enqueue_huge_page()
1072 h->free_huge_pages_node[nid]++; in enqueue_huge_page()
1079 bool nocma = !!(current->flags & PF_MEMALLOC_NOCMA); in dequeue_huge_page_node_exact()
1081 list_for_each_entry(page, &h->hugepage_freelists[nid], lru) { in dequeue_huge_page_node_exact()
1088 list_move(&page->lru, &h->hugepage_activelist); in dequeue_huge_page_node_exact()
1091 h->free_huge_pages--; in dequeue_huge_page_node_exact()
1092 h->free_huge_pages_node[nid]--; in dequeue_huge_page_node_exact()
1138 long chg) in dequeue_huge_page_vma() argument
1151 if (!vma_has_reserves(vma, chg) && in dequeue_huge_page_vma()
1152 h->free_huge_pages - h->resv_huge_pages == 0) in dequeue_huge_page_vma()
1156 if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) in dequeue_huge_page_vma()
1162 if (page && !avoid_reserve && vma_has_reserves(vma, chg)) { in dequeue_huge_page_vma()
1164 h->resv_huge_pages--; in dequeue_huge_page_vma()
1177 * nodes_allowed previously, so h->next_node_to_{alloc|free} might
1209 nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); in hstate_next_node_to_alloc()
1210 h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); in hstate_next_node_to_alloc()
1216 * helper for free_pool_huge_page() - return the previously saved
1227 nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); in hstate_next_node_to_free()
1228 h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); in hstate_next_node_to_free()
1237 nr_nodes--)
1243 nr_nodes--)
1346 h->nr_huge_pages--; in update_and_free_page()
1347 h->nr_huge_pages_node[page_to_nid(page)]--; in update_and_free_page()
1350 subpage->flags &= ~(1 << PG_locked | 1 << PG_error | in update_and_free_page()
1385 * Test to determine whether the hugepage is "active/in-use" (i.e. being linked
1386 * to hstate->hugepage_activelist.)
1417 return (unsigned long)page[2].mapping == -1U; in PageHugeTemporary()
1422 page[2].mapping = (void *)-1U; in SetPageHugeTemporary()
1446 page->mapping = NULL; in __free_huge_page()
1476 h->resv_huge_pages++; in __free_huge_page()
1479 list_del(&page->lru); in __free_huge_page()
1482 } else if (h->surplus_huge_pages_node[nid]) { in __free_huge_page()
1484 list_del(&page->lru); in __free_huge_page()
1486 h->surplus_huge_pages--; in __free_huge_page()
1487 h->surplus_huge_pages_node[nid]--; in __free_huge_page()
1496 * As free_huge_page() can be called from a non-task context, we have
1501 * be freed and frees them one-by-one. As the page->mapping pointer is
1517 node = node->next; in free_hpage_workfn()
1526 * Defer freeing if in non-task context to avoid hugetlb_lock deadlock. in free_huge_page()
1534 if (llist_add((struct llist_node *)&page->mapping, in free_huge_page()
1545 INIT_LIST_HEAD(&page->lru); in prep_new_huge_page()
1550 h->nr_huge_pages++; in prep_new_huge_page()
1551 h->nr_huge_pages_node[nid]++; in prep_new_huge_page()
1569 * boot, it's safer to be consistent with the not-gigantic in prep_compound_gigantic_page()
1583 atomic_set(compound_mapcount_ptr(page), -1); in prep_compound_gigantic_page()
1641 compound_idx = page_to_pfn(page) - page_to_pfn(page_head); in hugetlb_basepage_index()
1643 compound_idx = page - page_head; in hugetlb_basepage_index()
1763 if ((!acct_surplus || h->surplus_huge_pages_node[node]) && in free_pool_huge_page()
1764 !list_empty(&h->hugepage_freelists[node])) { in free_pool_huge_page()
1766 list_entry(h->hugepage_freelists[node].next, in free_pool_huge_page()
1768 list_del(&page->lru); in free_pool_huge_page()
1769 h->free_huge_pages--; in free_pool_huge_page()
1770 h->free_huge_pages_node[node]--; in free_pool_huge_page()
1772 h->surplus_huge_pages--; in free_pool_huge_page()
1773 h->surplus_huge_pages_node[node]--; in free_pool_huge_page()
1786 * nothing for in-use hugepages and non-hugepages.
1789 * -EBUSY: failed to dissolved free hugepages or the hugepage is in-use
1796 int rc = -EBUSY; in dissolve_free_huge_page()
1813 if (h->free_huge_pages - h->resv_huge_pages == 0) in dissolve_free_huge_page()
1825 * Theoretically, we should return -EBUSY when we in dissolve_free_huge_page()
1843 list_del(&head->lru); in dissolve_free_huge_page()
1844 h->free_huge_pages--; in dissolve_free_huge_page()
1845 h->free_huge_pages_node[nid]--; in dissolve_free_huge_page()
1846 h->max_huge_pages--; in dissolve_free_huge_page()
1894 if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) in alloc_surplus_huge_page()
1910 if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { in alloc_surplus_huge_page()
1916 h->surplus_huge_pages++; in alloc_surplus_huge_page()
1917 h->surplus_huge_pages_node[page_to_nid(page)]++; in alloc_surplus_huge_page()
1972 if (h->free_huge_pages - h->resv_huge_pages > 0) { in alloc_huge_page_nodemask()
2017 needed = (h->resv_huge_pages + delta) - h->free_huge_pages; in gather_surplus_pages()
2019 h->resv_huge_pages += delta; in gather_surplus_pages()
2026 ret = -ENOMEM; in gather_surplus_pages()
2036 list_add(&page->lru, &surplus_list); in gather_surplus_pages()
2046 needed = (h->resv_huge_pages + delta) - in gather_surplus_pages()
2047 (h->free_huge_pages + allocated); in gather_surplus_pages()
2067 h->resv_huge_pages += delta; in gather_surplus_pages()
2072 if ((--needed) < 0) in gather_surplus_pages()
2076 * no users -- drop the buddy allocator's reference. in gather_surplus_pages()
2118 * by pre-allocated pages. Only free surplus pages. in return_unused_surplus_pages()
2120 nr_pages = min(unused_resv_pages, h->surplus_huge_pages); in return_unused_surplus_pages()
2128 * on-line nodes with memory and will handle the hstate accounting. in return_unused_surplus_pages()
2134 while (nr_pages--) { in return_unused_surplus_pages()
2135 h->resv_huge_pages--; in return_unused_surplus_pages()
2136 unused_resv_pages--; in return_unused_surplus_pages()
2144 h->resv_huge_pages -= unused_resv_pages; in return_unused_surplus_pages()
2211 if (vma->vm_flags & VM_MAYSHARE) { in __vma_reservation_common()
2224 if (vma->vm_flags & VM_MAYSHARE) in __vma_reservation_common()
2232 * Subtle - The reserve map for private mappings has the in __vma_reservation_common()
2337 return ERR_PTR(-ENOMEM); in alloc_huge_page()
2350 return ERR_PTR(-ENOSPC); in alloc_huge_page()
2394 h->resv_huge_pages--; in alloc_huge_page()
2396 list_add(&page->lru, &h->hugepage_activelist); in alloc_huge_page()
2426 hugetlb_acct_memory(h, -rsv_adjust); in alloc_huge_page()
2443 return ERR_PTR(-ENOSPC); in alloc_huge_page()
2474 INIT_LIST_HEAD(&m->list); in __alloc_bootmem_huge_page()
2475 list_add(&m->list, &huge_boot_pages); in __alloc_bootmem_huge_page()
2476 m->hstate = h; in __alloc_bootmem_huge_page()
2490 struct hstate *h = m->hstate; in gather_bootmem_prealloc()
2502 * other side-effects, like CommitLimit going negative. in gather_bootmem_prealloc()
2516 * Bit mask controlling how hard we retry per-node allocations. in hugetlb_hstate_alloc_pages()
2528 /* bit mask controlling how hard we retry per-node allocations */ in hugetlb_hstate_alloc_pages()
2532 for (i = 0; i < h->max_huge_pages; ++i) { in hugetlb_hstate_alloc_pages()
2546 if (i < h->max_huge_pages) { in hugetlb_hstate_alloc_pages()
2551 h->max_huge_pages, buf, i); in hugetlb_hstate_alloc_pages()
2552 h->max_huge_pages = i; in hugetlb_hstate_alloc_pages()
2581 pr_info("HugeTLB registered %s page size, pre-allocated %ld pages\n", in report_hugepages()
2582 buf, h->free_huge_pages); in report_hugepages()
2597 struct list_head *freel = &h->hugepage_freelists[i]; in try_to_free_low()
2599 if (count >= h->nr_huge_pages) in try_to_free_low()
2603 list_del(&page->lru); in try_to_free_low()
2605 h->free_huge_pages--; in try_to_free_low()
2606 h->free_huge_pages_node[page_to_nid(page)]--; in try_to_free_low()
2618 * Increment or decrement surplus_huge_pages. Keep node-specific counters
2619 * balanced by operating on them in a round-robin fashion.
2627 VM_BUG_ON(delta != -1 && delta != 1); in adjust_pool_surplus()
2631 if (h->surplus_huge_pages_node[node]) in adjust_pool_surplus()
2636 if (h->surplus_huge_pages_node[node] < in adjust_pool_surplus()
2637 h->nr_huge_pages_node[node]) in adjust_pool_surplus()
2644 h->surplus_huge_pages += delta; in adjust_pool_surplus()
2645 h->surplus_huge_pages_node[node] += delta; in adjust_pool_surplus()
2649 #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages)
2657 * Bit mask controlling how hard we retry per-node allocations. in set_max_huge_pages()
2664 return -ENOMEM; in set_max_huge_pages()
2677 count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; in set_max_huge_pages()
2699 return -EINVAL; in set_max_huge_pages()
2715 while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { in set_max_huge_pages()
2716 if (!adjust_pool_surplus(h, nodes_allowed, -1)) in set_max_huge_pages()
2737 /* Bail for signals. Probably ctrl-c from user */ in set_max_huge_pages()
2757 min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; in set_max_huge_pages()
2770 h->max_huge_pages = persistent_huge_pages(h); in set_max_huge_pages()
2813 nr_huge_pages = h->nr_huge_pages; in nr_hugepages_show_common()
2815 nr_huge_pages = h->nr_huge_pages_node[nid]; in nr_hugepages_show_common()
2828 return -EINVAL; in __nr_hugepages_store_common()
2886 * hstate attribute for optionally mempolicy-based constraint on persistent
2908 return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); in nr_overcommit_hugepages_show()
2919 return -EINVAL; in nr_overcommit_hugepages_store()
2926 h->nr_overcommit_huge_pages = input; in nr_overcommit_hugepages_store()
2942 free_huge_pages = h->free_huge_pages; in free_hugepages_show()
2944 free_huge_pages = h->free_huge_pages_node[nid]; in free_hugepages_show()
2954 return sprintf(buf, "%lu\n", h->resv_huge_pages); in resv_hugepages_show()
2967 surplus_huge_pages = h->surplus_huge_pages; in surplus_hugepages_show()
2969 surplus_huge_pages = h->surplus_huge_pages_node[nid]; in surplus_hugepages_show()
2998 hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); in hugetlb_sysfs_add_hstate()
3000 return -ENOMEM; in hugetlb_sysfs_add_hstate()
3024 pr_err("HugeTLB: Unable to add hstate %s", h->name); in hugetlb_sysfs_init()
3031 * node_hstate/s - associate per node hstate attributes, via their kobjects,
3058 * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj.
3059 * Returns node id via non-NULL nidp.
3069 if (nhs->hstate_kobjs[i] == kobj) { in kobj_to_node_hstate()
3082 * No-op if no hstate attributes attached.
3087 struct node_hstate *nhs = &node_hstates[node->dev.id]; in hugetlb_unregister_node()
3089 if (!nhs->hugepages_kobj) in hugetlb_unregister_node()
3094 if (nhs->hstate_kobjs[idx]) { in hugetlb_unregister_node()
3095 kobject_put(nhs->hstate_kobjs[idx]); in hugetlb_unregister_node()
3096 nhs->hstate_kobjs[idx] = NULL; in hugetlb_unregister_node()
3100 kobject_put(nhs->hugepages_kobj); in hugetlb_unregister_node()
3101 nhs->hugepages_kobj = NULL; in hugetlb_unregister_node()
3107 * No-op if attributes already registered.
3112 struct node_hstate *nhs = &node_hstates[node->dev.id]; in hugetlb_register_node()
3115 if (nhs->hugepages_kobj) in hugetlb_register_node()
3118 nhs->hugepages_kobj = kobject_create_and_add("hugepages", in hugetlb_register_node()
3119 &node->dev.kobj); in hugetlb_register_node()
3120 if (!nhs->hugepages_kobj) in hugetlb_register_node()
3124 err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, in hugetlb_register_node()
3125 nhs->hstate_kobjs, in hugetlb_register_node()
3129 h->name, node->dev.id); in hugetlb_register_node()
3138 * devices of nodes that have memory. All on-line nodes should have
3147 if (node->dev.id == nid) in hugetlb_register_all_nodes()
3164 *nidp = -1; in kobj_to_node_hstate()
3178 pr_warn("HugeTLB: huge pages not supported, ignoring associated command-line parameters\n"); in hugetlb_init()
3255 h->order = order; in hugetlb_add_hstate()
3256 h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); in hugetlb_add_hstate()
3257 h->nr_huge_pages = 0; in hugetlb_add_hstate()
3258 h->free_huge_pages = 0; in hugetlb_add_hstate()
3260 INIT_LIST_HEAD(&h->hugepage_freelists[i]); in hugetlb_add_hstate()
3261 INIT_LIST_HEAD(&h->hugepage_activelist); in hugetlb_add_hstate()
3262 h->next_nid_to_alloc = first_memory_node; in hugetlb_add_hstate()
3263 h->next_nid_to_free = first_memory_node; in hugetlb_add_hstate()
3264 snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", in hugetlb_add_hstate()
3297 mhp = &parsed_hstate->max_huge_pages; in hugepages_setup()
3312 if (hugetlb_max_hstate && parsed_hstate->order >= MAX_ORDER) in hugepages_setup()
3366 hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT); in hugepagesz_setup()
3393 hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT); in default_hugepagesz_setup()
3421 unsigned int *array = h->free_huge_pages_node; in allowed_mems_nr()
3457 unsigned long tmp = h->max_huge_pages; in hugetlb_sysctl_handler_common()
3461 return -EOPNOTSUPP; in hugetlb_sysctl_handler_common()
3500 return -EOPNOTSUPP; in hugetlb_overcommit_handler()
3502 tmp = h->nr_overcommit_huge_pages; in hugetlb_overcommit_handler()
3505 return -EINVAL; in hugetlb_overcommit_handler()
3514 h->nr_overcommit_huge_pages = tmp; in hugetlb_overcommit_handler()
3532 unsigned long count = h->nr_huge_pages; in hugetlb_report_meminfo()
3544 h->free_huge_pages, in hugetlb_report_meminfo()
3545 h->resv_huge_pages, in hugetlb_report_meminfo()
3546 h->surplus_huge_pages, in hugetlb_report_meminfo()
3564 nid, h->nr_huge_pages_node[nid], in hugetlb_report_node_meminfo()
3565 nid, h->free_huge_pages_node[nid], in hugetlb_report_node_meminfo()
3566 nid, h->surplus_huge_pages_node[nid]); in hugetlb_report_node_meminfo()
3581 h->nr_huge_pages_node[nid], in hugetlb_show_meminfo()
3582 h->free_huge_pages_node[nid], in hugetlb_show_meminfo()
3583 h->surplus_huge_pages_node[nid], in hugetlb_show_meminfo()
3584 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); in hugetlb_show_meminfo()
3590 atomic_long_read(&mm->hugetlb_usage) << (PAGE_SHIFT - 10)); in hugetlb_report_usage()
3600 nr_total_pages += h->nr_huge_pages * pages_per_huge_page(h); in hugetlb_total_pages()
3606 int ret = -ENOMEM; in hugetlb_acct_memory()
3644 return_unused_surplus_pages(h, (unsigned long) -delta); in hugetlb_acct_memory()
3665 kref_get(&resv->refs); in hugetlb_vm_op_open()
3680 start = vma_hugecache_offset(h, vma, vma->vm_start); in hugetlb_vm_op_close()
3681 end = vma_hugecache_offset(h, vma, vma->vm_end); in hugetlb_vm_op_close()
3683 reserve = (end - start) - region_count(resv, start, end); in hugetlb_vm_op_close()
3691 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_vm_op_close()
3694 kref_put(&resv->refs, resv_map_release); in hugetlb_vm_op_close()
3700 return -EINVAL; in hugetlb_vm_op_split()
3713 * handle_mm_fault() to try to instantiate regular-sized pages in the
3745 vma->vm_page_prot))); in make_huge_pte()
3748 vma->vm_page_prot)); in make_huge_pte()
3802 struct address_space *mapping = vma->vm_file->f_mapping; in copy_hugetlb_page_range()
3806 cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; in copy_hugetlb_page_range()
3810 vma->vm_start, in copy_hugetlb_page_range()
3811 vma->vm_end); in copy_hugetlb_page_range()
3823 for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { in copy_hugetlb_page_range()
3830 ret = -ENOMEM; in copy_hugetlb_page_range()
3908 struct mm_struct *mm = vma->vm_mm; in __unmap_hugepage_range()
4038 vma->vm_flags &= ~VM_MAYSHARE; in __unmap_hugepage_range_final()
4058 mm = vma->vm_mm; in unmap_hugepage_range()
4084 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + in unmap_ref_private()
4085 vma->vm_pgoff; in unmap_ref_private()
4086 mapping = vma->vm_file->f_mapping; in unmap_ref_private()
4094 vma_interval_tree_foreach(iter_vma, &mapping->i_mmap, pgoff, pgoff) { in unmap_ref_private()
4104 if (iter_vma->vm_flags & VM_MAYSHARE) in unmap_ref_private()
4110 * areas. This is because a future no-page fault on this VMA in unmap_ref_private()
4143 /* If no-one else is actually using this page, avoid the copy in hugetlb_cow()
4182 struct address_space *mapping = vma->vm_file->f_mapping; in hugetlb_cow()
4212 * race occurs while re-acquiring page table in hugetlb_cow()
4278 mapping = vma->vm_file->f_mapping; in hugetlbfs_pagecache_page()
4295 mapping = vma->vm_file->f_mapping; in hugetlbfs_pagecache_present()
4307 struct inode *inode = mapping->host; in huge_add_to_page_cache()
4317 * by non-hugetlbfs specific code paths. in huge_add_to_page_cache()
4321 spin_lock(&inode->i_lock); in huge_add_to_page_cache()
4322 inode->i_blocks += blocks_per_huge_page(h); in huge_add_to_page_cache()
4323 spin_unlock(&inode->i_lock); in huge_add_to_page_cache()
4382 current->pid); in hugetlb_no_page()
4391 size = i_size_read(mapping->host) >> huge_page_shift(h); in hugetlb_no_page()
4434 if (vma->vm_flags & VM_MAYSHARE) { in hugetlb_no_page()
4438 if (err == -EEXIST) in hugetlb_no_page()
4479 if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { in hugetlb_no_page()
4498 new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) in hugetlb_no_page()
4499 && (vma->vm_flags & VM_SHARED))); in hugetlb_no_page()
4503 if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { in hugetlb_no_page()
4544 return hash & (num_fault_mutexes - 1); in hugetlb_fault_mutex_hash()
4599 mapping = vma->vm_file->f_mapping; in hugetlb_fault()
4652 if (!(vma->vm_flags & VM_MAYSHARE)) in hugetlb_fault()
4732 int vm_shared = dst_vma->vm_flags & VM_SHARED; in hugetlb_mcopy_atomic_pte()
4740 mapping = dst_vma->vm_file->f_mapping; in hugetlb_mcopy_atomic_pte()
4744 ret = -EFAULT; in hugetlb_mcopy_atomic_pte()
4750 * a non-missing case. Return -EEXIST. in hugetlb_mcopy_atomic_pte()
4754 ret = -EEXIST; in hugetlb_mcopy_atomic_pte()
4760 ret = -ENOMEM; in hugetlb_mcopy_atomic_pte()
4770 ret = -ENOENT; in hugetlb_mcopy_atomic_pte()
4789 size = i_size_read(mapping->host) >> huge_page_shift(h); in hugetlb_mcopy_atomic_pte()
4790 ret = -EFAULT; in hugetlb_mcopy_atomic_pte()
4817 size = i_size_read(mapping->host) >> huge_page_shift(h); in hugetlb_mcopy_atomic_pte()
4818 ret = -EFAULT; in hugetlb_mcopy_atomic_pte()
4822 ret = -EEXIST; in hugetlb_mcopy_atomic_pte()
4833 /* For CONTINUE on a non-shared VMA, don't set VM_WRITE for CoW. */ in hugetlb_mcopy_atomic_pte()
4837 writable = dst_vma->vm_flags & VM_WRITE; in hugetlb_mcopy_atomic_pte()
4847 dst_vma->vm_flags & VM_WRITE); in hugetlb_mcopy_atomic_pte()
4850 /* No need to invalidate - it was non-present before */ in hugetlb_mcopy_atomic_pte()
4880 int err = -EFAULT; in follow_hugetlb_page()
4882 while (vaddr < vma->vm_end && remainder) { in follow_hugetlb_page()
4954 * FAULT_FLAG_TRIED can co-exist in follow_hugetlb_page()
4991 (vaddr + huge_page_size(h) < vma->vm_end) && in follow_hugetlb_page()
4994 remainder -= pages_per_huge_page(h); in follow_hugetlb_page()
5016 err = -ENOMEM; in follow_hugetlb_page()
5026 --remainder; in follow_hugetlb_page()
5028 if (vaddr < vma->vm_end && remainder && in follow_hugetlb_page()
5052 struct mm_struct *mm = vma->vm_mm; in hugetlb_change_protection()
5074 i_mmap_lock_write(vma->vm_file->f_mapping); in hugetlb_change_protection()
5135 i_mmap_unlock_write(vma->vm_file->f_mapping); in hugetlb_change_protection()
5138 return pages << h->order; in hugetlb_change_protection()
5146 long ret, chg, add = -1; in hugetlb_reserve_pages() local
5156 return -EINVAL; in hugetlb_reserve_pages()
5170 * to reserve the full area even if read-only as mprotect() may be in hugetlb_reserve_pages()
5171 * called to make the mapping read-write. Assume !vma is a shm mapping in hugetlb_reserve_pages()
5173 if (!vma || vma->vm_flags & VM_MAYSHARE) { in hugetlb_reserve_pages()
5181 chg = region_chg(resv_map, from, to, ®ions_needed); in hugetlb_reserve_pages()
5187 return -ENOMEM; in hugetlb_reserve_pages()
5189 chg = to - from; in hugetlb_reserve_pages()
5195 if (chg < 0) { in hugetlb_reserve_pages()
5196 ret = chg; in hugetlb_reserve_pages()
5201 hstate_index(h), chg * pages_per_huge_page(h), &h_cg); in hugetlb_reserve_pages()
5204 ret = -ENOMEM; in hugetlb_reserve_pages()
5208 if (vma && !(vma->vm_flags & VM_MAYSHARE) && h_cg) { in hugetlb_reserve_pages()
5220 gbl_reserve = hugepage_subpool_get_pages(spool, chg); in hugetlb_reserve_pages()
5222 ret = -ENOSPC; in hugetlb_reserve_pages()
5240 * the reservation was consumed. Private mappings are per-VMA and in hugetlb_reserve_pages()
5246 if (!vma || vma->vm_flags & VM_MAYSHARE) { in hugetlb_reserve_pages()
5250 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_reserve_pages()
5253 } else if (unlikely(chg > add)) { in hugetlb_reserve_pages()
5265 * reference to h_cg->css. See comment below for detail. in hugetlb_reserve_pages()
5269 (chg - add) * pages_per_huge_page(h), h_cg); in hugetlb_reserve_pages()
5272 chg - add); in hugetlb_reserve_pages()
5273 hugetlb_acct_memory(h, -rsv_adjust); in hugetlb_reserve_pages()
5277 * h_cg->css. So we should release the reference held in hugetlb_reserve_pages()
5286 /* put back original number of pages, chg */ in hugetlb_reserve_pages()
5287 (void)hugepage_subpool_put_pages(spool, chg); in hugetlb_reserve_pages()
5290 chg * pages_per_huge_page(h), h_cg); in hugetlb_reserve_pages()
5292 if (!vma || vma->vm_flags & VM_MAYSHARE) in hugetlb_reserve_pages()
5296 if (chg >= 0 && add < 0) in hugetlb_reserve_pages()
5299 kref_put(&resv_map->refs, resv_map_release); in hugetlb_reserve_pages()
5308 long chg = 0; in hugetlb_unreserve_pages() local
5317 chg = region_del(resv_map, start, end); in hugetlb_unreserve_pages()
5323 if (chg < 0) in hugetlb_unreserve_pages()
5324 return chg; in hugetlb_unreserve_pages()
5327 spin_lock(&inode->i_lock); in hugetlb_unreserve_pages()
5328 inode->i_blocks -= (blocks_per_huge_page(h) * freed); in hugetlb_unreserve_pages()
5329 spin_unlock(&inode->i_lock); in hugetlb_unreserve_pages()
5335 gbl_reserve = hugepage_subpool_put_pages(spool, (chg - freed)); in hugetlb_unreserve_pages()
5336 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_unreserve_pages()
5346 unsigned long saddr = ((idx - svma->vm_pgoff) << PAGE_SHIFT) + in page_table_shareable()
5347 svma->vm_start; in page_table_shareable()
5352 unsigned long vm_flags = vma->vm_flags & VM_LOCKED_CLEAR_MASK; in page_table_shareable()
5353 unsigned long svm_flags = svma->vm_flags & VM_LOCKED_CLEAR_MASK; in page_table_shareable()
5361 sbase < svma->vm_start || svma->vm_end < s_end) in page_table_shareable()
5375 if (vma->vm_flags & VM_MAYSHARE && range_in_vma(vma, base, end)) in vma_shareable()
5397 unsigned long v_start = ALIGN(vma->vm_start, PUD_SIZE), in adjust_range_if_pmd_sharing_possible()
5398 v_end = ALIGN_DOWN(vma->vm_end, PUD_SIZE); in adjust_range_if_pmd_sharing_possible()
5404 if (!(vma->vm_flags & VM_MAYSHARE) || !(v_end > v_start) || in adjust_range_if_pmd_sharing_possible()
5436 struct address_space *mapping = vma->vm_file->f_mapping; in huge_pmd_share()
5437 pgoff_t idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + in huge_pmd_share()
5438 vma->vm_pgoff; in huge_pmd_share()
5446 vma_interval_tree_foreach(svma, &mapping->i_mmap, idx, idx) { in huge_pmd_share()
5452 spte = huge_pte_offset(svma->vm_mm, saddr, in huge_pmd_share()
5497 i_mmap_assert_write_locked(vma->vm_file->f_mapping); in huge_pmd_unshare()
5512 *addr |= PUD_SIZE - PMD_SIZE; in huge_pmd_unshare()
5571 * huge_pte_offset() - Walk the page table to resolve the hugepage
5596 /* must be pud huge, non-present or none */ in huge_pte_offset()
5603 /* must be pmd huge, non-present or none */ in huge_pte_offset()
5617 return ERR_PTR(-EINVAL); in follow_huge_addr()
5632 struct mm_struct *mm = vma->vm_mm; in follow_huge_pmd_pte()
5710 list_move_tail(&page->lru, list); in isolate_huge_page()
5721 list_move_tail(&page->lru, &(page_hstate(page))->hugepage_activelist); in putback_active_hugepage()
5737 * the temporary status. in move_hugetlb_state()
5739 * Also note that we have to transfer the per-node surplus state in move_hugetlb_state()
5741 * the per-node's. in move_hugetlb_state()
5751 if (h->surplus_huge_pages_node[old_nid]) { in move_hugetlb_state()
5752 h->surplus_huge_pages_node[old_nid]--; in move_hugetlb_state()
5753 h->surplus_huge_pages_node[new_nid]++; in move_hugetlb_state()
5767 struct mm_struct *mm = vma->vm_mm; in hugetlb_unshare_all_pmds()
5773 if (!(vma->vm_flags & VM_MAYSHARE)) in hugetlb_unshare_all_pmds()
5776 start = ALIGN(vma->vm_start, PUD_SIZE); in hugetlb_unshare_all_pmds()
5777 end = ALIGN_DOWN(vma->vm_end, PUD_SIZE); in hugetlb_unshare_all_pmds()
5790 i_mmap_lock_write(vma->vm_file->f_mapping); in hugetlb_unshare_all_pmds()
5803 i_mmap_unlock_write(vma->vm_file->f_mapping); in hugetlb_unshare_all_pmds()
5851 size = min(per_node, hugetlb_cma_size - reserved); in hugetlb_cma_reserve()