Lines Matching +full:trim +full:- +full:hs +full:- +full:current
1 // SPDX-License-Identifier: GPL-2.0-only
51 #include <linux/page-isolation.h>
107 if (spool->count) in subpool_is_free()
109 if (spool->max_hpages != -1) in subpool_is_free()
110 return spool->used_hpages == 0; in subpool_is_free()
111 if (spool->min_hpages != -1) in subpool_is_free()
112 return spool->rsv_hpages == spool->min_hpages; in subpool_is_free()
120 spin_unlock_irqrestore(&spool->lock, irq_flags); in unlock_or_release_subpool()
126 if (spool->min_hpages != -1) in unlock_or_release_subpool()
127 hugetlb_acct_memory(spool->hstate, in unlock_or_release_subpool()
128 -spool->min_hpages); in unlock_or_release_subpool()
142 spin_lock_init(&spool->lock); in hugepage_new_subpool()
143 spool->count = 1; in hugepage_new_subpool()
144 spool->max_hpages = max_hpages; in hugepage_new_subpool()
145 spool->hstate = h; in hugepage_new_subpool()
146 spool->min_hpages = min_hpages; in hugepage_new_subpool()
148 if (min_hpages != -1 && hugetlb_acct_memory(h, min_hpages)) { in hugepage_new_subpool()
152 spool->rsv_hpages = min_hpages; in hugepage_new_subpool()
161 spin_lock_irqsave(&spool->lock, flags); in hugepage_put_subpool()
162 BUG_ON(!spool->count); in hugepage_put_subpool()
163 spool->count--; in hugepage_put_subpool()
169 * Return -ENOMEM if there are not enough resources to satisfy the
183 spin_lock_irq(&spool->lock); in hugepage_subpool_get_pages()
185 if (spool->max_hpages != -1) { /* maximum size accounting */ in hugepage_subpool_get_pages()
186 if ((spool->used_hpages + delta) <= spool->max_hpages) in hugepage_subpool_get_pages()
187 spool->used_hpages += delta; in hugepage_subpool_get_pages()
189 ret = -ENOMEM; in hugepage_subpool_get_pages()
195 if (spool->min_hpages != -1 && spool->rsv_hpages) { in hugepage_subpool_get_pages()
196 if (delta > spool->rsv_hpages) { in hugepage_subpool_get_pages()
201 ret = delta - spool->rsv_hpages; in hugepage_subpool_get_pages()
202 spool->rsv_hpages = 0; in hugepage_subpool_get_pages()
205 spool->rsv_hpages -= delta; in hugepage_subpool_get_pages()
210 spin_unlock_irq(&spool->lock); in hugepage_subpool_get_pages()
229 spin_lock_irqsave(&spool->lock, flags); in hugepage_subpool_put_pages()
231 if (spool->max_hpages != -1) /* maximum size accounting */ in hugepage_subpool_put_pages()
232 spool->used_hpages -= delta; in hugepage_subpool_put_pages()
235 if (spool->min_hpages != -1 && spool->used_hpages < spool->min_hpages) { in hugepage_subpool_put_pages()
236 if (spool->rsv_hpages + delta <= spool->min_hpages) in hugepage_subpool_put_pages()
239 ret = spool->rsv_hpages + delta - spool->min_hpages; in hugepage_subpool_put_pages()
241 spool->rsv_hpages += delta; in hugepage_subpool_put_pages()
242 if (spool->rsv_hpages > spool->min_hpages) in hugepage_subpool_put_pages()
243 spool->rsv_hpages = spool->min_hpages; in hugepage_subpool_put_pages()
257 return HUGETLBFS_SB(inode->i_sb)->spool; in subpool_inode()
262 return subpool_inode(file_inode(vma->vm_file)); in subpool_vma()
271 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_lock_read()
273 down_read(&vma_lock->rw_sema); in hugetlb_vma_lock_read()
277 down_read(&resv_map->rw_sema); in hugetlb_vma_lock_read()
284 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_unlock_read()
286 up_read(&vma_lock->rw_sema); in hugetlb_vma_unlock_read()
290 up_read(&resv_map->rw_sema); in hugetlb_vma_unlock_read()
297 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_lock_write()
299 down_write(&vma_lock->rw_sema); in hugetlb_vma_lock_write()
303 down_write(&resv_map->rw_sema); in hugetlb_vma_lock_write()
310 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_unlock_write()
312 up_write(&vma_lock->rw_sema); in hugetlb_vma_unlock_write()
316 up_write(&resv_map->rw_sema); in hugetlb_vma_unlock_write()
324 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_trylock_write()
326 return down_write_trylock(&vma_lock->rw_sema); in hugetlb_vma_trylock_write()
330 return down_write_trylock(&resv_map->rw_sema); in hugetlb_vma_trylock_write()
339 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_assert_locked()
341 lockdep_assert_held(&vma_lock->rw_sema); in hugetlb_vma_assert_locked()
345 lockdep_assert_held(&resv_map->rw_sema); in hugetlb_vma_assert_locked()
359 struct vm_area_struct *vma = vma_lock->vma; in __hugetlb_vma_unlock_write_put()
364 * Semaphore synchronizes access to vma_lock->vma field. in __hugetlb_vma_unlock_write_put()
366 vma_lock->vma = NULL; in __hugetlb_vma_unlock_write_put()
367 vma->vm_private_data = NULL; in __hugetlb_vma_unlock_write_put()
368 up_write(&vma_lock->rw_sema); in __hugetlb_vma_unlock_write_put()
369 kref_put(&vma_lock->refs, hugetlb_vma_lock_release); in __hugetlb_vma_unlock_write_put()
375 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in __hugetlb_vma_unlock_write_free()
382 up_write(&resv_map->rw_sema); in __hugetlb_vma_unlock_write_free()
394 if (vma->vm_private_data) { in hugetlb_vma_lock_free()
395 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vma_lock_free()
397 down_write(&vma_lock->rw_sema); in hugetlb_vma_lock_free()
407 if (!vma || !(vma->vm_flags & VM_MAYSHARE)) in hugetlb_vma_lock_alloc()
410 /* Should never get here with non-NULL vm_private_data */ in hugetlb_vma_lock_alloc()
411 if (vma->vm_private_data) in hugetlb_vma_lock_alloc()
430 kref_init(&vma_lock->refs); in hugetlb_vma_lock_alloc()
431 init_rwsem(&vma_lock->rw_sema); in hugetlb_vma_lock_alloc()
432 vma_lock->vma = vma; in hugetlb_vma_lock_alloc()
433 vma->vm_private_data = vma_lock; in hugetlb_vma_lock_alloc()
444 VM_BUG_ON(resv->region_cache_count <= 0); in get_file_region_entry_from_cache()
446 resv->region_cache_count--; in get_file_region_entry_from_cache()
447 nrg = list_first_entry(&resv->region_cache, struct file_region, link); in get_file_region_entry_from_cache()
448 list_del(&nrg->link); in get_file_region_entry_from_cache()
450 nrg->from = from; in get_file_region_entry_from_cache()
451 nrg->to = to; in get_file_region_entry_from_cache()
460 nrg->reservation_counter = rg->reservation_counter; in copy_hugetlb_cgroup_uncharge_info()
461 nrg->css = rg->css; in copy_hugetlb_cgroup_uncharge_info()
462 if (rg->css) in copy_hugetlb_cgroup_uncharge_info()
463 css_get(rg->css); in copy_hugetlb_cgroup_uncharge_info()
475 nrg->reservation_counter = in record_hugetlb_cgroup_uncharge_info()
476 &h_cg->rsvd_hugepage[hstate_index(h)]; in record_hugetlb_cgroup_uncharge_info()
477 nrg->css = &h_cg->css; in record_hugetlb_cgroup_uncharge_info()
479 * The caller will hold exactly one h_cg->css reference for the in record_hugetlb_cgroup_uncharge_info()
484 * exactly one h_cg->css reference, we should do css_get for in record_hugetlb_cgroup_uncharge_info()
488 css_get(&h_cg->css); in record_hugetlb_cgroup_uncharge_info()
489 if (!resv->pages_per_hpage) in record_hugetlb_cgroup_uncharge_info()
490 resv->pages_per_hpage = pages_per_huge_page(h); in record_hugetlb_cgroup_uncharge_info()
494 VM_BUG_ON(resv->pages_per_hpage != pages_per_huge_page(h)); in record_hugetlb_cgroup_uncharge_info()
496 nrg->reservation_counter = NULL; in record_hugetlb_cgroup_uncharge_info()
497 nrg->css = NULL; in record_hugetlb_cgroup_uncharge_info()
505 if (rg->css) in put_uncharge_info()
506 css_put(rg->css); in put_uncharge_info()
514 return rg->reservation_counter == org->reservation_counter && in has_same_uncharge_info()
515 rg->css == org->css; in has_same_uncharge_info()
527 if (&prg->link != &resv->regions && prg->to == rg->from && in coalesce_file_region()
529 prg->to = rg->to; in coalesce_file_region()
531 list_del(&rg->link); in coalesce_file_region()
539 if (&nrg->link != &resv->regions && nrg->from == rg->to && in coalesce_file_region()
541 nrg->from = rg->from; in coalesce_file_region()
543 list_del(&rg->link); in coalesce_file_region()
559 list_add(&nrg->link, rg); in hugetlb_resv_map_add()
564 return to - from; in hugetlb_resv_map_add()
568 * Must be called with resv->lock held.
580 struct list_head *head = &resv->regions; in add_reservation_in_range()
589 * [last_accounted_offset, iter->from), at every iteration, with some in add_reservation_in_range()
594 if (iter->from < f) { in add_reservation_in_range()
598 if (iter->to > last_accounted_offset) in add_reservation_in_range()
599 last_accounted_offset = iter->to; in add_reservation_in_range()
606 if (iter->from >= t) { in add_reservation_in_range()
607 rg = iter->link.prev; in add_reservation_in_range()
611 /* Add an entry for last_accounted_offset -> iter->from, and in add_reservation_in_range()
614 if (iter->from > last_accounted_offset) in add_reservation_in_range()
615 add += hugetlb_resv_map_add(resv, iter->link.prev, in add_reservation_in_range()
617 iter->from, h, h_cg, in add_reservation_in_range()
620 last_accounted_offset = iter->to; in add_reservation_in_range()
627 rg = head->prev; in add_reservation_in_range()
635 /* Must be called with resv->lock acquired. Will drop lock to allocate entries.
639 __must_hold(&resv->lock) in allocate_file_region_entries()
656 while (resv->region_cache_count < in allocate_file_region_entries()
657 (resv->adds_in_progress + regions_needed)) { in allocate_file_region_entries()
658 to_allocate = resv->adds_in_progress + regions_needed - in allocate_file_region_entries()
659 resv->region_cache_count; in allocate_file_region_entries()
665 VM_BUG_ON(resv->region_cache_count < resv->adds_in_progress); in allocate_file_region_entries()
667 spin_unlock(&resv->lock); in allocate_file_region_entries()
672 list_add(&trg->link, &allocated_regions); in allocate_file_region_entries()
675 spin_lock(&resv->lock); in allocate_file_region_entries()
677 list_splice(&allocated_regions, &resv->region_cache); in allocate_file_region_entries()
678 resv->region_cache_count += to_allocate; in allocate_file_region_entries()
685 list_del(&rg->link); in allocate_file_region_entries()
688 return -ENOMEM; in allocate_file_region_entries()
703 * this operation and we were not able to allocate, it returns -ENOMEM.
714 spin_lock(&resv->lock); in region_add()
731 resv->region_cache_count < in region_add()
732 resv->adds_in_progress + in region_add()
733 (actual_regions_needed - in_regions_needed)) { in region_add()
737 VM_BUG_ON(t - f <= 1); in region_add()
740 resv, actual_regions_needed - in_regions_needed)) { in region_add()
741 return -ENOMEM; in region_add()
749 resv->adds_in_progress -= in_regions_needed; in region_add()
751 spin_unlock(&resv->lock); in region_add()
767 * resv->adds_in_progress. This value needs to be provided to a follow up call
772 * zero. -ENOMEM is returned if a new file_region structure or cache entry
780 spin_lock(&resv->lock); in region_chg()
790 return -ENOMEM; in region_chg()
792 resv->adds_in_progress += *out_regions_needed; in region_chg()
794 spin_unlock(&resv->lock); in region_chg()
814 spin_lock(&resv->lock); in region_abort()
815 VM_BUG_ON(!resv->region_cache_count); in region_abort()
816 resv->adds_in_progress -= regions_needed; in region_abort()
817 spin_unlock(&resv->lock); in region_abort()
824 * and either trim, delete or split the existing regions.
829 * be allocated. If the allocation fails, -ENOMEM will be returned.
831 * a region and possibly return -ENOMEM. Callers specifying
832 * t == LONG_MAX do not need to check for -ENOMEM error.
836 struct list_head *head = &resv->regions; in region_del()
842 spin_lock(&resv->lock); in region_del()
851 if (rg->to <= f && (rg->to != rg->from || rg->to != f)) in region_del()
854 if (rg->from >= t) in region_del()
857 if (f > rg->from && t < rg->to) { /* Must split region */ in region_del()
863 resv->region_cache_count > resv->adds_in_progress) { in region_del()
864 nrg = list_first_entry(&resv->region_cache, in region_del()
867 list_del(&nrg->link); in region_del()
868 resv->region_cache_count--; in region_del()
872 spin_unlock(&resv->lock); in region_del()
875 return -ENOMEM; in region_del()
879 del += t - f; in region_del()
881 resv, rg, t - f, false); in region_del()
884 nrg->from = t; in region_del()
885 nrg->to = rg->to; in region_del()
889 INIT_LIST_HEAD(&nrg->link); in region_del()
892 rg->to = f; in region_del()
894 list_add(&nrg->link, &rg->link); in region_del()
899 if (f <= rg->from && t >= rg->to) { /* Remove entire region */ in region_del()
900 del += rg->to - rg->from; in region_del()
902 rg->to - rg->from, true); in region_del()
903 list_del(&rg->link); in region_del()
908 if (f <= rg->from) { /* Trim beginning of region */ in region_del()
910 t - rg->from, false); in region_del()
912 del += t - rg->from; in region_del()
913 rg->from = t; in region_del()
914 } else { /* Trim end of region */ in region_del()
916 rg->to - f, false); in region_del()
918 del += rg->to - f; in region_del()
919 rg->to = f; in region_del()
923 spin_unlock(&resv->lock); in region_del()
963 struct list_head *head = &resv->regions; in region_count()
967 spin_lock(&resv->lock); in region_count()
973 if (rg->to <= f) in region_count()
975 if (rg->from >= t) in region_count()
978 seg_from = max(rg->from, f); in region_count()
979 seg_to = min(rg->to, t); in region_count()
981 chg += seg_to - seg_from; in region_count()
983 spin_unlock(&resv->lock); in region_count()
995 return ((address - vma->vm_start) >> huge_page_shift(h)) + in vma_hugecache_offset()
996 (vma->vm_pgoff >> huge_page_order(h)); in vma_hugecache_offset()
1000 * vma_kernel_pagesize - Page size granularity for this VMA.
1010 if (vma->vm_ops && vma->vm_ops->pagesize) in vma_kernel_pagesize()
1011 return vma->vm_ops->pagesize(vma); in vma_kernel_pagesize()
1019 * architectures where it differs, an architecture-specific 'strong'
1057 return (unsigned long)vma->vm_private_data; in get_vma_private_data()
1063 vma->vm_private_data = (void *)value; in set_vma_private_data()
1073 resv_map->reservation_counter = NULL; in resv_map_set_hugetlb_cgroup_uncharge_info()
1074 resv_map->pages_per_hpage = 0; in resv_map_set_hugetlb_cgroup_uncharge_info()
1075 resv_map->css = NULL; in resv_map_set_hugetlb_cgroup_uncharge_info()
1077 resv_map->reservation_counter = in resv_map_set_hugetlb_cgroup_uncharge_info()
1078 &h_cg->rsvd_hugepage[hstate_index(h)]; in resv_map_set_hugetlb_cgroup_uncharge_info()
1079 resv_map->pages_per_hpage = pages_per_huge_page(h); in resv_map_set_hugetlb_cgroup_uncharge_info()
1080 resv_map->css = &h_cg->css; in resv_map_set_hugetlb_cgroup_uncharge_info()
1096 kref_init(&resv_map->refs); in resv_map_alloc()
1097 spin_lock_init(&resv_map->lock); in resv_map_alloc()
1098 INIT_LIST_HEAD(&resv_map->regions); in resv_map_alloc()
1099 init_rwsem(&resv_map->rw_sema); in resv_map_alloc()
1101 resv_map->adds_in_progress = 0; in resv_map_alloc()
1105 * re-initialized to the proper values, to indicate that hugetlb cgroup in resv_map_alloc()
1106 * reservations are to be un-charged from here. in resv_map_alloc()
1110 INIT_LIST_HEAD(&resv_map->region_cache); in resv_map_alloc()
1111 list_add(&rg->link, &resv_map->region_cache); in resv_map_alloc()
1112 resv_map->region_cache_count = 1; in resv_map_alloc()
1120 struct list_head *head = &resv_map->region_cache; in resv_map_release()
1128 list_del(&rg->link); in resv_map_release()
1132 VM_BUG_ON(resv_map->adds_in_progress); in resv_map_release()
1144 * The VERY common case is inode->mapping == &inode->i_data but, in inode_resv_map()
1147 return (struct resv_map *)(&inode->i_data)->i_private_data; in inode_resv_map()
1153 if (vma->vm_flags & VM_MAYSHARE) { in vma_resv_map()
1154 struct address_space *mapping = vma->vm_file->f_mapping; in vma_resv_map()
1155 struct inode *inode = mapping->host; in vma_resv_map()
1168 VM_BUG_ON_VMA(vma->vm_flags & VM_MAYSHARE, vma); in set_vma_resv_map()
1176 VM_BUG_ON_VMA(vma->vm_flags & VM_MAYSHARE, vma); in set_vma_resv_flags()
1190 return !(vma->vm_flags & VM_MAYSHARE) && in __vma_private_lock()
1200 * - For shared mappings this is a per-vma semaphore that may be in hugetlb_dup_vma_private()
1206 * - For MAP_PRIVATE mappings, this is the reserve map which does in hugetlb_dup_vma_private()
1208 * not guaranteed to succeed, even if read-only. in hugetlb_dup_vma_private()
1210 if (vma->vm_flags & VM_MAYSHARE) { in hugetlb_dup_vma_private()
1211 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_dup_vma_private()
1213 if (vma_lock && vma_lock->vma != vma) in hugetlb_dup_vma_private()
1214 vma->vm_private_data = NULL; in hugetlb_dup_vma_private()
1216 vma->vm_private_data = NULL; in hugetlb_dup_vma_private()
1221 * Called with mm->mmap_lock writer semaphore held.
1244 kref_put(&reservations->refs, resv_map_release); in clear_vma_resv_huge_pages()
1257 list_move(&folio->lru, &h->hugepage_freelists[nid]); in enqueue_hugetlb_folio()
1258 h->free_huge_pages++; in enqueue_hugetlb_folio()
1259 h->free_huge_pages_node[nid]++; in enqueue_hugetlb_folio()
1267 bool pin = !!(current->flags & PF_MEMALLOC_PIN); in dequeue_hugetlb_folio_node_exact()
1270 list_for_each_entry(folio, &h->hugepage_freelists[nid], lru) { in dequeue_hugetlb_folio_node_exact()
1277 if (is_migrate_isolate_page(&folio->page)) in dequeue_hugetlb_folio_node_exact()
1280 list_move(&folio->lru, &h->hugepage_activelist); in dequeue_hugetlb_folio_node_exact()
1283 h->free_huge_pages--; in dequeue_hugetlb_folio_node_exact()
1284 h->free_huge_pages_node[nid]--; in dequeue_hugetlb_folio_node_exact()
1333 return h->free_huge_pages - h->resv_huge_pages; in available_huge_pages()
1378 * nodes_allowed previously, so h->next_node_to_{alloc|free} might
1417 * helper for remove_pool_hugetlb_folio() - return the previously saved
1428 nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); in hstate_next_node_to_free()
1429 h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); in hstate_next_node_to_free()
1438 nr_nodes--)
1440 #define for_each_node_mask_to_free(hs, nr_nodes, node, mask) \ argument
1443 ((node = hstate_next_node_to_free(hs, mask)) || 1); \
1444 nr_nodes--)
1532 list_del(&folio->lru); in remove_hugetlb_folio()
1536 h->free_huge_pages--; in remove_hugetlb_folio()
1537 h->free_huge_pages_node[nid]--; in remove_hugetlb_folio()
1540 h->surplus_huge_pages--; in remove_hugetlb_folio()
1541 h->surplus_huge_pages_node[nid]--; in remove_hugetlb_folio()
1552 h->nr_huge_pages--; in remove_hugetlb_folio()
1553 h->nr_huge_pages_node[nid]--; in remove_hugetlb_folio()
1565 INIT_LIST_HEAD(&folio->lru); in add_hugetlb_folio()
1566 h->nr_huge_pages++; in add_hugetlb_folio()
1567 h->nr_huge_pages_node[nid]++; in add_hugetlb_folio()
1570 h->surplus_huge_pages++; in add_hugetlb_folio()
1571 h->surplus_huge_pages_node[nid]++; in add_hugetlb_folio()
1637 INIT_LIST_HEAD(&folio->_deferred_list); in __update_and_free_hugetlb_folio()
1648 * freed and frees them one-by-one. As the page->mapping pointer is going
1666 node = node->next; in free_hpage_workfn()
1667 folio->mapping = NULL; in free_hpage_workfn()
1704 if (llist_add((struct llist_node *)&folio->mapping, &hpage_freelist)) in update_and_free_hugetlb_folio()
1723 list_del(&folio->lru); in bulk_vmemmap_restore_error()
1743 list_del(&folio->lru); in bulk_vmemmap_restore_error()
1748 list_del(&folio->lru); in bulk_vmemmap_restore_error()
1828 __ClearPageAnonExclusive(&folio->page); in free_huge_folio()
1829 folio->mapping = NULL; in free_huge_folio()
1858 lruvec_stat_mod_folio(folio, NR_HUGETLB, -pages_per_huge_page(h)); in free_huge_folio()
1861 h->resv_huge_pages++; in free_huge_folio()
1867 } else if (h->surplus_huge_pages_node[nid]) { in free_huge_folio()
1885 h->nr_huge_pages++; in __prep_account_new_huge_page()
1886 h->nr_huge_pages_node[nid]++; in __prep_account_new_huge_page()
1892 INIT_LIST_HEAD(&folio->lru); in init_new_hugetlb_folio()
2095 if ((!acct_surplus || h->surplus_huge_pages_node[node]) && in remove_pool_hugetlb_folio()
2096 !list_empty(&h->hugepage_freelists[node])) { in remove_pool_hugetlb_folio()
2097 folio = list_entry(h->hugepage_freelists[node].next, in remove_pool_hugetlb_folio()
2109 * does nothing for in-use hugetlb folios and non-hugetlb folios.
2112 * -ENOMEM: failed to allocate vmemmap pages to free the freed hugepages
2116 * -EBUSY: failed to dissolved free hugepages or the hugepage is in-use
2123 int rc = -EBUSY; in dissolve_free_hugetlb_folio()
2152 * Theoretically, we should return -EBUSY when we in dissolve_free_hugetlb_folio()
2162 if (h->surplus_huge_pages_node[folio_nid(folio)]) in dissolve_free_hugetlb_folio()
2165 h->max_huge_pages--; in dissolve_free_hugetlb_folio()
2178 * non-vmemmap optimized hugetlb folios. in dissolve_free_hugetlb_folio()
2185 h->max_huge_pages++; in dissolve_free_hugetlb_folio()
2244 if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) in alloc_surplus_hugetlb_folio()
2260 if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { in alloc_surplus_hugetlb_folio()
2267 h->surplus_huge_pages++; in alloc_surplus_hugetlb_folio()
2268 h->surplus_huge_pages_node[folio_nid(folio)]++; in alloc_surplus_hugetlb_folio()
2337 VM_BUG_ON(!h->resv_huge_pages); in alloc_hugetlb_folio_reserve()
2338 h->resv_huge_pages--; in alloc_hugetlb_folio_reserve()
2362 /* We cannot fallback to other nodes, as we could break the per-node pool. */ in alloc_hugetlb_folio_nodemask()
2372 struct mempolicy *mpol = get_task_policy(current); in policy_mbind_nodemask()
2378 if (mpol->mode == MPOL_BIND && in policy_mbind_nodemask()
2380 cpuset_nodemask_valid_mems_allowed(&mpol->nodes))) in policy_mbind_nodemask()
2381 return &mpol->nodes; in policy_mbind_nodemask()
2409 needed = (h->resv_huge_pages + delta) - h->free_huge_pages; in gather_surplus_pages()
2411 h->resv_huge_pages += delta; in gather_surplus_pages()
2417 ret = -ENOMEM; in gather_surplus_pages()
2423 /* Prioritize current node */ in gather_surplus_pages()
2442 list_add(&folio->lru, &surplus_list); in gather_surplus_pages()
2452 needed = (h->resv_huge_pages + delta) - in gather_surplus_pages()
2453 (h->free_huge_pages + allocated); in gather_surplus_pages()
2473 h->resv_huge_pages += delta; in gather_surplus_pages()
2478 if ((--needed) < 0) in gather_surplus_pages()
2513 h->resv_huge_pages -= unused_resv_pages; in return_unused_surplus_pages()
2520 * by pre-allocated pages. Only free surplus pages. in return_unused_surplus_pages()
2522 nr_pages = min(unused_resv_pages, h->surplus_huge_pages); in return_unused_surplus_pages()
2530 * on-line nodes with memory and will handle the hstate accounting. in return_unused_surplus_pages()
2532 while (nr_pages--) { in return_unused_surplus_pages()
2539 list_add(&folio->lru, &page_list); in return_unused_surplus_pages()
2618 if (vma->vm_flags & VM_MAYSHARE) { in __vma_reservation_common()
2628 if (vma->vm_flags & VM_MAYSHARE) { in __vma_reservation_common()
2641 if (vma->vm_flags & VM_MAYSHARE || mode == VMA_DEL_RESV) in __vma_reservation_common()
2650 * Subtle - The reserve map for private mappings has the in __vma_reservation_common()
2770 if (!(vma->vm_flags & VM_MAYSHARE)) in restore_reserve_on_error()
2789 * alloc_and_dissolve_hugetlb_folio - Allocate a new folio and dissolve
2816 * Fail with -EBUSY if not possible. in alloc_and_dissolve_hugetlb_folio()
2820 ret = isolated ? 0 : -EBUSY; in alloc_and_dissolve_hugetlb_folio()
2838 return -ENOMEM; in alloc_and_dissolve_hugetlb_folio()
2880 int ret = -EBUSY; in isolate_or_dissolve_huge_page()
2898 * alloc_contig_range and them. Return -ENOMEM as this has the effect in isolate_or_dissolve_huge_page()
2902 return -ENOMEM; in isolate_or_dissolve_huge_page()
2913 * replace_free_hugepage_folios - Replace free hugepage folios in a given pfn
2960 * For either 0/1: we checked the per-vma resv map, and one resv
2966 * Cannot use per-vma resv count can be used, hence a new resv
2979 * faults of hugetlb private mappings on top of a non-page-cache folio (in
2998 /* Whether we need a separate per-vma reservation? */ in alloc_hugetlb_folio()
3003 * consume the per-vma resv map even if it's reserved. in alloc_hugetlb_folio()
3014 return ERR_PTR(-ENOMEM); in alloc_hugetlb_folio()
3039 * If this allocation is not consuming a per-vma reservation, in alloc_hugetlb_folio()
3066 list_add(&folio->lru, &h->hugepage_activelist); in alloc_hugetlb_folio()
3072 * Either dequeued or buddy-allocated folio needs to add special in alloc_hugetlb_folio()
3077 h->resv_huge_pages--; in alloc_hugetlb_folio()
3110 hugetlb_acct_memory(h, -rsv_adjust); in alloc_hugetlb_folio()
3129 if (ret == -ENOMEM) { in alloc_hugetlb_folio()
3131 return ERR_PTR(-ENOMEM); in alloc_hugetlb_folio()
3148 return ERR_PTR(-ENOSPC); in alloc_hugetlb_folio()
3167 for_each_node_mask_to_alloc(&h->next_nid_to_alloc, nr_nodes, node, &node_states[N_MEMORY]) { in __alloc_bootmem_huge_page()
3191 huge_page_size(h) - PAGE_SIZE); in __alloc_bootmem_huge_page()
3193 INIT_LIST_HEAD(&m->list); in __alloc_bootmem_huge_page()
3194 list_add(&m->list, &huge_boot_pages[node]); in __alloc_bootmem_huge_page()
3195 m->hstate = h; in __alloc_bootmem_huge_page()
3213 __ClearPageReserved(folio_page(folio, pfn - head_pfn)); in hugetlb_folio_init_tail_vmemmap()
3215 prep_compound_tail((struct page *)folio, pfn - head_pfn); in hugetlb_folio_init_tail_vmemmap()
3280 h = m->hstate; in gather_bootmem_prealloc_node()
3295 list_add(&folio->lru, &folio_list); in gather_bootmem_prealloc_node()
3300 * other side-effects, like CommitLimit going negative. in gather_bootmem_prealloc_node()
3340 for (i = 0; i < h->max_huge_pages_node[nid]; ++i) { in hugetlb_hstate_alloc_pages_onenode()
3352 list_add(&folio->lru, &folio_list); in hugetlb_hstate_alloc_pages_onenode()
3360 if (i == h->max_huge_pages_node[nid]) in hugetlb_hstate_alloc_pages_onenode()
3365 h->max_huge_pages_node[nid], buf, nid, i); in hugetlb_hstate_alloc_pages_onenode()
3366 h->max_huge_pages -= (h->max_huge_pages_node[nid] - i); in hugetlb_hstate_alloc_pages_onenode()
3367 h->max_huge_pages_node[nid] = i; in hugetlb_hstate_alloc_pages_onenode()
3376 if (h->max_huge_pages_node[i] > 0) { in hugetlb_hstate_alloc_pages_specific_nodes()
3387 if (allocated < h->max_huge_pages) { in hugetlb_hstate_alloc_pages_errcheck()
3392 h->max_huge_pages, buf, allocated); in hugetlb_hstate_alloc_pages_errcheck()
3393 h->max_huge_pages = allocated; in hugetlb_hstate_alloc_pages_errcheck()
3400 int i, num = end - start; in hugetlb_pages_alloc_boot_node()
3405 /* Bit mask controlling how hard we retry per-node allocations.*/ in hugetlb_pages_alloc_boot_node()
3414 list_move(&folio->lru, &folio_list); in hugetlb_pages_alloc_boot_node()
3425 for (i = 0; i < h->max_huge_pages; ++i) { in hugetlb_gigantic_pages_alloc_boot()
3444 job.size = h->max_huge_pages; in hugetlb_pages_alloc_boot()
3457 * +------------+-------+-------+-------+-------+-------+ in hugetlb_pages_alloc_boot()
3459 * +------------+-------+-------+-------+-------+-------+ in hugetlb_pages_alloc_boot()
3463 * +------------+-------+-------+-------+-------+-------+ in hugetlb_pages_alloc_boot()
3466 job.min_chunk = h->max_huge_pages / num_node_state(N_MEMORY) / 2; in hugetlb_pages_alloc_boot()
3469 return h->nr_huge_pages; in hugetlb_pages_alloc_boot()
3474 * non-gigantic pages.
3475 * - For gigantic pages, this is called early in the boot process and
3479 * - For non-gigantic pages, this is called later in the boot process after
3527 * h->demote_order is initially 0. in hugetlb_init_hstates()
3528 * - We can not demote gigantic pages if runtime freeing in hugetlb_init_hstates()
3530 * - If CMA allocation is possible, we can not demote in hugetlb_init_hstates()
3535 if (hugetlb_cma_size && h->order <= HUGETLB_PAGE_ORDER) in hugetlb_init_hstates()
3540 if (h2->order < h->order && in hugetlb_init_hstates()
3541 h2->order > h->demote_order) in hugetlb_init_hstates()
3542 h->demote_order = h2->order; in hugetlb_init_hstates()
3555 pr_info("HugeTLB: registered %s page size, pre-allocated %ld pages\n", in report_hugepages()
3556 buf, h->free_huge_pages); in report_hugepages()
3578 struct list_head *freel = &h->hugepage_freelists[i]; in try_to_free_low()
3580 if (count >= h->nr_huge_pages) in try_to_free_low()
3585 list_add(&folio->lru, &page_list); in try_to_free_low()
3602 * Increment or decrement surplus_huge_pages. Keep node-specific counters
3603 * balanced by operating on them in a round-robin fashion.
3612 VM_BUG_ON(delta != -1 && delta != 1); in adjust_pool_surplus()
3615 for_each_node_mask_to_alloc(&h->next_nid_to_alloc, nr_nodes, node, nodes_allowed) { in adjust_pool_surplus()
3616 if (h->surplus_huge_pages_node[node]) in adjust_pool_surplus()
3621 if (h->surplus_huge_pages_node[node] < in adjust_pool_surplus()
3622 h->nr_huge_pages_node[node]) in adjust_pool_surplus()
3629 h->surplus_huge_pages += delta; in adjust_pool_surplus()
3630 h->surplus_huge_pages_node[node] += delta; in adjust_pool_surplus()
3634 #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages)
3645 * Bit mask controlling how hard we retry per-node allocations. in set_max_huge_pages()
3652 return -ENOMEM; in set_max_huge_pages()
3658 mutex_lock(&h->resize_lock); in set_max_huge_pages()
3671 count += persistent_huge_pages(h) - in set_max_huge_pages()
3672 (h->nr_huge_pages_node[nid] - in set_max_huge_pages()
3673 h->surplus_huge_pages_node[nid]); in set_max_huge_pages()
3694 mutex_unlock(&h->resize_lock); in set_max_huge_pages()
3696 return -EINVAL; in set_max_huge_pages()
3712 while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { in set_max_huge_pages()
3713 if (!adjust_pool_surplus(h, nodes_allowed, -1)) in set_max_huge_pages()
3731 &h->next_nid_to_alloc); in set_max_huge_pages()
3738 list_add(&folio->lru, &page_list); in set_max_huge_pages()
3741 /* Bail for signals. Probably ctrl-c from user */ in set_max_huge_pages()
3742 if (signal_pending(current)) { in set_max_huge_pages()
3773 min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; in set_max_huge_pages()
3785 list_add(&folio->lru, &page_list); in set_max_huge_pages()
3798 h->max_huge_pages = persistent_huge_pages(h); in set_max_huge_pages()
3800 mutex_unlock(&h->resize_lock); in set_max_huge_pages()
3823 * Note that we already hold src->resize_lock. To prevent deadlock, in demote_free_hugetlb_folios()
3826 mutex_lock(&dst->resize_lock); in demote_free_hugetlb_folios()
3834 list_del(&folio->lru); in demote_free_hugetlb_folios()
3836 split_page_owner(&folio->page, huge_page_order(src), huge_page_order(dst)); in demote_free_hugetlb_folios()
3845 prep_compound_page(page, dst->order); in demote_free_hugetlb_folios()
3847 new_folio->mapping = NULL; in demote_free_hugetlb_folios()
3849 list_add(&new_folio->lru, &dst_list); in demote_free_hugetlb_folios()
3855 mutex_unlock(&dst->resize_lock); in demote_free_hugetlb_folios()
3872 if (!src->demote_order) { in demote_pool_huge_page()
3874 return -EINVAL; /* internal error */ in demote_pool_huge_page()
3876 dst = size_to_hstate(PAGE_SIZE << src->demote_order); in demote_pool_huge_page()
3882 list_for_each_entry_safe(folio, next, &src->hugepage_freelists[node], lru) { in demote_pool_huge_page()
3887 list_add(&folio->lru, &list); in demote_pool_huge_page()
3900 list_del(&folio->lru); in demote_pool_huge_page()
3903 nr_demoted--; in demote_pool_huge_page()
3914 src->max_huge_pages -= nr_demoted; in demote_pool_huge_page()
3915 dst->max_huge_pages += nr_demoted << (huge_page_order(src) - huge_page_order(dst)); in demote_pool_huge_page()
3924 * Return -EBUSY so that caller will not retry. in demote_pool_huge_page()
3926 return -EBUSY; in demote_pool_huge_page()
3966 nr_huge_pages = h->nr_huge_pages; in nr_hugepages_show_common()
3968 nr_huge_pages = h->nr_huge_pages_node[nid]; in nr_hugepages_show_common()
3981 return -EINVAL; in __nr_hugepages_store_common()
4039 * hstate attribute for optionally mempolicy-based constraint on persistent
4062 return sysfs_emit(buf, "%lu\n", h->nr_overcommit_huge_pages); in nr_overcommit_hugepages_show()
4073 return -EINVAL; in nr_overcommit_hugepages_store()
4080 h->nr_overcommit_huge_pages = input; in nr_overcommit_hugepages_store()
4096 free_huge_pages = h->free_huge_pages; in free_hugepages_show()
4098 free_huge_pages = h->free_huge_pages_node[nid]; in free_hugepages_show()
4108 return sysfs_emit(buf, "%lu\n", h->resv_huge_pages); in resv_hugepages_show()
4121 surplus_huge_pages = h->surplus_huge_pages; in surplus_hugepages_show()
4123 surplus_huge_pages = h->surplus_huge_pages_node[nid]; in surplus_hugepages_show()
4152 mutex_lock(&h->resize_lock); in demote_store()
4163 nr_available = h->free_huge_pages_node[nid]; in demote_store()
4165 nr_available = h->free_huge_pages; in demote_store()
4166 nr_available -= h->resv_huge_pages; in demote_store()
4176 nr_demote -= rc; in demote_store()
4180 mutex_unlock(&h->resize_lock); in demote_store()
4192 unsigned long demote_size = (PAGE_SIZE << h->demote_order) / SZ_1K; in demote_size_show()
4209 return -EINVAL; in demote_size_store()
4210 demote_order = demote_hstate->order; in demote_size_store()
4212 return -EINVAL; in demote_size_store()
4216 if (demote_order >= h->order) in demote_size_store()
4217 return -EINVAL; in demote_size_store()
4220 mutex_lock(&h->resize_lock); in demote_size_store()
4221 h->demote_order = demote_order; in demote_size_store()
4222 mutex_unlock(&h->resize_lock); in demote_size_store()
4261 hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); in hugetlb_sysfs_add_hstate()
4263 return -ENOMEM; in hugetlb_sysfs_add_hstate()
4272 if (h->demote_order) { in hugetlb_sysfs_add_hstate()
4276 pr_warn("HugeTLB unable to create demote interfaces for %s\n", h->name); in hugetlb_sysfs_add_hstate()
4291 * node_hstate/s - associate per node hstate attributes, via their kobjects,
4318 * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj.
4319 * Returns node id via non-NULL nidp.
4329 if (nhs->hstate_kobjs[i] == kobj) { in kobj_to_node_hstate()
4342 * No-op if no hstate attributes attached.
4347 struct node_hstate *nhs = &node_hstates[node->dev.id]; in hugetlb_unregister_node()
4349 if (!nhs->hugepages_kobj) in hugetlb_unregister_node()
4354 struct kobject *hstate_kobj = nhs->hstate_kobjs[idx]; in hugetlb_unregister_node()
4358 if (h->demote_order) in hugetlb_unregister_node()
4362 nhs->hstate_kobjs[idx] = NULL; in hugetlb_unregister_node()
4365 kobject_put(nhs->hugepages_kobj); in hugetlb_unregister_node()
4366 nhs->hugepages_kobj = NULL; in hugetlb_unregister_node()
4372 * No-op if attributes already registered.
4377 struct node_hstate *nhs = &node_hstates[node->dev.id]; in hugetlb_register_node()
4383 if (nhs->hugepages_kobj) in hugetlb_register_node()
4386 nhs->hugepages_kobj = kobject_create_and_add("hugepages", in hugetlb_register_node()
4387 &node->dev.kobj); in hugetlb_register_node()
4388 if (!nhs->hugepages_kobj) in hugetlb_register_node()
4392 err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, in hugetlb_register_node()
4393 nhs->hstate_kobjs, in hugetlb_register_node()
4397 h->name, node->dev.id); in hugetlb_register_node()
4406 * devices of nodes that have memory. All on-line nodes should have
4422 *nidp = -1; in kobj_to_node_hstate()
4451 pr_err("HugeTLB: Unable to add hstate %s", h->name); in hugetlb_sysfs_init()
4475 pr_warn("HugeTLB: huge pages not supported, ignoring associated command-line parameters\n"); in hugetlb_init()
4556 __mutex_init(&h->resize_lock, "resize mutex", &h->resize_key); in hugetlb_add_hstate()
4557 h->order = order; in hugetlb_add_hstate()
4558 h->mask = ~(huge_page_size(h) - 1); in hugetlb_add_hstate()
4560 INIT_LIST_HEAD(&h->hugepage_freelists[i]); in hugetlb_add_hstate()
4561 INIT_LIST_HEAD(&h->hugepage_activelist); in hugetlb_add_hstate()
4562 h->next_nid_to_alloc = first_memory_node; in hugetlb_add_hstate()
4563 h->next_nid_to_free = first_memory_node; in hugetlb_add_hstate()
4564 snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", in hugetlb_add_hstate()
4582 parsed_hstate->max_huge_pages = 0; in hugepages_clear_pages_in_node()
4583 memset(parsed_hstate->max_huge_pages_node, 0, in hugepages_clear_pages_in_node()
4584 sizeof(parsed_hstate->max_huge_pages_node)); in hugepages_clear_pages_in_node()
4619 mhp = &parsed_hstate->max_huge_pages; in hugepages_setup()
4646 parsed_hstate->max_huge_pages_node[node] = tmp; in hugepages_setup()
4725 hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT); in hugepagesz_setup()
4753 hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT); in default_hugepagesz_setup()
4784 unsigned int *array = h->free_huge_pages_node; in allowed_mems_nr()
4818 unsigned long tmp = h->max_huge_pages; in hugetlb_sysctl_handler_common()
4822 return -EOPNOTSUPP; in hugetlb_sysctl_handler_common()
4861 return -EOPNOTSUPP; in hugetlb_overcommit_handler()
4863 tmp = h->nr_overcommit_huge_pages; in hugetlb_overcommit_handler()
4866 return -EINVAL; in hugetlb_overcommit_handler()
4875 h->nr_overcommit_huge_pages = tmp; in hugetlb_overcommit_handler()
4930 unsigned long count = h->nr_huge_pages; in hugetlb_report_meminfo()
4942 h->free_huge_pages, in hugetlb_report_meminfo()
4943 h->resv_huge_pages, in hugetlb_report_meminfo()
4944 h->surplus_huge_pages, in hugetlb_report_meminfo()
4962 nid, h->nr_huge_pages_node[nid], in hugetlb_report_node_meminfo()
4963 nid, h->free_huge_pages_node[nid], in hugetlb_report_node_meminfo()
4964 nid, h->surplus_huge_pages_node[nid]); in hugetlb_report_node_meminfo()
4977 h->nr_huge_pages_node[nid], in hugetlb_show_meminfo_node()
4978 h->free_huge_pages_node[nid], in hugetlb_show_meminfo_node()
4979 h->surplus_huge_pages_node[nid], in hugetlb_show_meminfo_node()
4986 K(atomic_long_read(&mm->hugetlb_usage))); in hugetlb_report_usage()
4996 nr_total_pages += h->nr_huge_pages * pages_per_huge_page(h); in hugetlb_total_pages()
5002 int ret = -ENOMEM; in hugetlb_acct_memory()
5013 * current cpuset. Application can still potentially OOM'ed by kernel in hugetlb_acct_memory()
5021 * we fall back to check against current free page availability as in hugetlb_acct_memory()
5028 * the memory policy of the current task. Similar to the description in hugetlb_acct_memory()
5043 return_unused_surplus_pages(h, (unsigned long) -delta); in hugetlb_acct_memory()
5065 kref_get(&resv->refs); in hugetlb_vm_op_open()
5074 if (vma->vm_flags & VM_MAYSHARE) { in hugetlb_vm_op_open()
5075 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; in hugetlb_vm_op_open()
5078 if (vma_lock->vma != vma) { in hugetlb_vm_op_open()
5079 vma->vm_private_data = NULL; in hugetlb_vm_op_open()
5102 start = vma_hugecache_offset(h, vma, vma->vm_start); in hugetlb_vm_op_close()
5103 end = vma_hugecache_offset(h, vma, vma->vm_end); in hugetlb_vm_op_close()
5105 reserve = (end - start) - region_count(resv, start, end); in hugetlb_vm_op_close()
5113 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_vm_op_close()
5116 kref_put(&resv->refs, resv_map_release); in hugetlb_vm_op_close()
5122 return -EINVAL; in hugetlb_vm_op_split()
5125 * PMD sharing is only possible for PUD_SIZE-aligned address ranges in hugetlb_vm_op_split()
5138 if (floor >= vma->vm_start && ceil <= vma->vm_end) in hugetlb_vm_op_split()
5152 * handle_mm_fault() to try to instantiate regular-sized pages in the
5183 if (try_mkwrite && (vma->vm_flags & VM_WRITE)) { in make_huge_pte()
5185 vma->vm_page_prot))); in make_huge_pte()
5188 vma->vm_page_prot)); in make_huge_pte()
5191 entry = arch_make_huge_pte(entry, shift, vma->vm_flags); in make_huge_pte()
5201 entry = huge_pte_mkwrite(huge_pte_mkdirty(huge_ptep_get(vma->vm_mm, address, ptep))); in set_huge_ptep_writable()
5209 if (vma->vm_flags & VM_WRITE) in set_huge_ptep_maybe_writable()
5243 pte_t newpte = make_huge_pte(vma, &new_folio->page, true); in hugetlb_install_folio()
5249 set_huge_pte_at(vma->vm_mm, addr, ptep, newpte, sz); in hugetlb_install_folio()
5250 hugetlb_count_add(pages_per_huge_page(hstate_vma(vma)), vma->vm_mm); in hugetlb_install_folio()
5261 bool cow = is_cow_mapping(src_vma->vm_flags); in copy_hugetlb_page_range()
5271 src_vma->vm_start, in copy_hugetlb_page_range()
5272 src_vma->vm_end); in copy_hugetlb_page_range()
5275 raw_write_seqcount_begin(&src->write_protect_seq); in copy_hugetlb_page_range()
5287 for (addr = src_vma->vm_start; addr < src_vma->vm_end; addr += sz) { in copy_hugetlb_page_range()
5296 ret = -ENOMEM; in copy_hugetlb_page_range()
5316 entry = huge_ptep_get(src_vma->vm_mm, addr, src_pte); in copy_hugetlb_page_range()
5354 entry = huge_ptep_get(src_vma->vm_mm, addr, src_pte); in copy_hugetlb_page_range()
5364 * When pre-allocating the page or copying data, we in copy_hugetlb_page_range()
5395 entry = huge_ptep_get(src_vma->vm_mm, addr, src_pte); in copy_hugetlb_page_range()
5433 raw_write_seqcount_end(&src->write_protect_seq); in copy_hugetlb_page_range()
5448 struct mm_struct *mm = vma->vm_mm; in move_huge_pte()
5487 struct address_space *mapping = vma->vm_file->f_mapping; in move_hugetlb_page_tables()
5489 struct mm_struct *mm = vma->vm_mm; in move_hugetlb_page_tables()
5537 flush_hugetlb_tlb_range(vma, old_end - len, old_end); in move_hugetlb_page_tables()
5542 return len + old_addr - old_end; in move_hugetlb_page_tables()
5549 struct mm_struct *mm = vma->vm_mm; in __unmap_hugepage_range()
5602 * If the pte was wr-protected by uffd-wp in any of the in __unmap_hugepage_range()
5604 * drop the uffd-wp bit in this zap, then replace the in __unmap_hugepage_range()
5641 /* Leave a uffd-wp pte marker if needed */ in __unmap_hugepage_range()
5656 if (!h->surplus_huge_pages && __vma_private_lock(vma) && in __unmap_hugepage_range()
5668 * resv->adds_in_progress if it succeeds. If this is not done, in __unmap_hugepage_range()
5716 if (!vma->vm_file) /* hugetlbfs_file_mmap error */ in __hugetlb_zap_begin()
5721 if (vma->vm_file) in __hugetlb_zap_begin()
5722 i_mmap_lock_write(vma->vm_file->f_mapping); in __hugetlb_zap_begin()
5728 zap_flags_t zap_flags = details ? details->zap_flags : 0; in __hugetlb_zap_end()
5730 if (!vma->vm_file) /* hugetlbfs_file_mmap error */ in __hugetlb_zap_end()
5748 if (vma->vm_file) in __hugetlb_zap_end()
5749 i_mmap_unlock_write(vma->vm_file->f_mapping); in __hugetlb_zap_end()
5759 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, in unmap_hugepage_range()
5763 tlb_gather_mmu(&tlb, vma->vm_mm); in unmap_hugepage_range()
5790 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + in unmap_ref_private()
5791 vma->vm_pgoff; in unmap_ref_private()
5792 mapping = vma->vm_file->f_mapping; in unmap_ref_private()
5800 vma_interval_tree_foreach(iter_vma, &mapping->i_mmap, pgoff, pgoff) { in unmap_ref_private()
5801 /* Do not unmap the current VMA */ in unmap_ref_private()
5810 if (iter_vma->vm_flags & VM_MAYSHARE) in unmap_ref_private()
5816 * areas. This is because a future no-page fault on this VMA in unmap_ref_private()
5836 struct vm_area_struct *vma = vmf->vma; in hugetlb_wp()
5837 struct mm_struct *mm = vma->vm_mm; in hugetlb_wp()
5838 const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE; in hugetlb_wp()
5839 pte_t pte = huge_ptep_get(mm, vmf->address, vmf->pte); in hugetlb_wp()
5848 * Never handle CoW for uffd-wp protected pages. It should be only in hugetlb_wp()
5849 * handled when the uffd-wp protection is removed. in hugetlb_wp()
5853 * uffd-wp bit first. in hugetlb_wp()
5859 if (vma->vm_flags & VM_MAYSHARE) { in hugetlb_wp()
5860 set_huge_ptep_writable(vma, vmf->address, vmf->pte); in hugetlb_wp()
5870 * If no-one else is actually using this page, we're the exclusive in hugetlb_wp()
5881 if (!PageAnonExclusive(&old_folio->page)) { in hugetlb_wp()
5883 SetPageAnonExclusive(&old_folio->page); in hugetlb_wp()
5886 set_huge_ptep_maybe_writable(vma, vmf->address, in hugetlb_wp()
5887 vmf->pte); in hugetlb_wp()
5893 PageAnonExclusive(&old_folio->page), &old_folio->page); in hugetlb_wp()
5914 spin_unlock(vmf->ptl); in hugetlb_wp()
5915 new_folio = alloc_hugetlb_folio(vma, vmf->address, cow_from_owner); in hugetlb_wp()
5926 struct address_space *mapping = vma->vm_file->f_mapping; in hugetlb_wp()
5940 idx = vma_hugecache_offset(h, vma, vmf->address); in hugetlb_wp()
5945 unmap_ref_private(mm, vma, &old_folio->page, in hugetlb_wp()
5946 vmf->address); in hugetlb_wp()
5950 spin_lock(vmf->ptl); in hugetlb_wp()
5951 vmf->pte = hugetlb_walk(vma, vmf->address, in hugetlb_wp()
5953 if (likely(vmf->pte && in hugetlb_wp()
5954 pte_same(huge_ptep_get(mm, vmf->address, vmf->pte), pte))) in hugetlb_wp()
5957 * race occurs while re-acquiring page table in hugetlb_wp()
5976 if (copy_user_large_folio(new_folio, old_folio, vmf->real_address, vma)) { in hugetlb_wp()
5982 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, mm, vmf->address, in hugetlb_wp()
5983 vmf->address + huge_page_size(h)); in hugetlb_wp()
5990 spin_lock(vmf->ptl); in hugetlb_wp()
5991 vmf->pte = hugetlb_walk(vma, vmf->address, huge_page_size(h)); in hugetlb_wp()
5992 if (likely(vmf->pte && pte_same(huge_ptep_get(mm, vmf->address, vmf->pte), pte))) { in hugetlb_wp()
5993 pte_t newpte = make_huge_pte(vma, &new_folio->page, !unshare); in hugetlb_wp()
5996 huge_ptep_clear_flush(vma, vmf->address, vmf->pte); in hugetlb_wp()
5998 hugetlb_add_new_anon_rmap(new_folio, vma, vmf->address); in hugetlb_wp()
6001 set_huge_pte_at(mm, vmf->address, vmf->pte, newpte, in hugetlb_wp()
6007 spin_unlock(vmf->ptl); in hugetlb_wp()
6015 restore_reserve_on_error(h, vma, vmf->address, new_folio); in hugetlb_wp()
6020 spin_lock(vmf->ptl); /* Caller expects lock to be held */ in hugetlb_wp()
6032 struct address_space *mapping = vma->vm_file->f_mapping; in hugetlbfs_pagecache_present()
6046 struct inode *inode = mapping->host; in hugetlb_add_to_page_cache()
6062 * by non-hugetlbfs specific code paths. in hugetlb_add_to_page_cache()
6066 spin_lock(&inode->i_lock); in hugetlb_add_to_page_cache()
6067 inode->i_blocks += blocks_per_huge_page(h); in hugetlb_add_to_page_cache()
6068 spin_unlock(&inode->i_lock); in hugetlb_add_to_page_cache()
6083 hugetlb_vma_unlock_read(vmf->vma); in hugetlb_handle_userfault()
6084 hash = hugetlb_fault_mutex_hash(mapping, vmf->pgoff); in hugetlb_handle_userfault()
6109 struct vm_area_struct *vma = vmf->vma; in hugetlb_no_page()
6110 struct mm_struct *mm = vma->vm_mm; in hugetlb_no_page()
6118 u32 hash = hugetlb_fault_mutex_hash(mapping, vmf->pgoff); in hugetlb_no_page()
6128 current->pid); in hugetlb_no_page()
6137 folio = filemap_lock_hugetlb_folio(h, mapping, vmf->pgoff); in hugetlb_no_page()
6139 size = i_size_read(mapping->host) >> huge_page_shift(h); in hugetlb_no_page()
6140 if (vmf->pgoff >= size) in hugetlb_no_page()
6146 * without pgtable lock, we need to re-test under in hugetlb_no_page()
6149 * either changed or during-changing ptes and retry in hugetlb_no_page()
6161 if (!hugetlb_pte_stable(h, mm, vmf->address, vmf->pte, vmf->orig_pte)) { in hugetlb_no_page()
6170 if (!(vma->vm_flags & VM_MAYSHARE)) { in hugetlb_no_page()
6176 folio = alloc_hugetlb_folio(vma, vmf->address, false); in hugetlb_no_page()
6190 if (hugetlb_pte_stable(h, mm, vmf->address, vmf->pte, vmf->orig_pte)) in hugetlb_no_page()
6196 folio_zero_user(folio, vmf->real_address); in hugetlb_no_page()
6200 if (vma->vm_flags & VM_MAYSHARE) { in hugetlb_no_page()
6202 vmf->pgoff); in hugetlb_no_page()
6205 * err can't be -EEXIST which implies someone in hugetlb_no_page()
6211 restore_reserve_on_error(h, vma, vmf->address, in hugetlb_no_page()
6239 if (!hugetlb_pte_stable(h, mm, vmf->address, vmf->pte, vmf->orig_pte)) { in hugetlb_no_page()
6254 if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { in hugetlb_no_page()
6255 if (vma_needs_reservation(h, vma, vmf->address) < 0) { in hugetlb_no_page()
6260 vma_end_reservation(h, vma, vmf->address); in hugetlb_no_page()
6263 vmf->ptl = huge_pte_lock(h, mm, vmf->pte); in hugetlb_no_page()
6266 if (!pte_same(huge_ptep_get(mm, vmf->address, vmf->pte), vmf->orig_pte)) in hugetlb_no_page()
6270 hugetlb_add_new_anon_rmap(folio, vma, vmf->address); in hugetlb_no_page()
6273 new_pte = make_huge_pte(vma, &folio->page, vma->vm_flags & VM_SHARED); in hugetlb_no_page()
6275 * If this pte was previously wr-protected, keep it wr-protected even in hugetlb_no_page()
6278 if (unlikely(pte_marker_uffd_wp(vmf->orig_pte))) in hugetlb_no_page()
6280 set_huge_pte_at(mm, vmf->address, vmf->pte, new_pte, huge_page_size(h)); in hugetlb_no_page()
6283 if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { in hugetlb_no_page()
6288 spin_unlock(vmf->ptl); in hugetlb_no_page()
6303 * We must check to release the per-VMA lock. __vmf_anon_prepare() is in hugetlb_no_page()
6313 spin_unlock(vmf->ptl); in hugetlb_no_page()
6316 restore_reserve_on_error(h, vma, vmf->address, folio); in hugetlb_no_page()
6334 return hash & (num_fault_mutexes - 1); in hugetlb_fault_mutex_hash()
6377 mapping = vma->vm_file->f_mapping; in hugetlb_fault()
6455 !(vma->vm_flags & VM_MAYSHARE) && !huge_pte_write(vmf.orig_pte)) { in hugetlb_fault()
6475 /* Handle userfault-wp first, before trying to lock more pages */ in hugetlb_fault()
6536 * We must check to release the per-VMA lock. __vmf_anon_prepare() in in hugetlb_fault()
6546 * the page is not used after unlocked before returning from the current in hugetlb_fault()
6574 * that breaking the per-node hugetlb pool is not allowed in this case. in alloc_hugetlb_folio_vma()
6593 struct mm_struct *dst_mm = dst_vma->vm_mm; in hugetlb_mfill_atomic_pte()
6597 struct address_space *mapping = dst_vma->vm_file->f_mapping; in hugetlb_mfill_atomic_pte()
6600 int vm_shared = dst_vma->vm_flags & VM_SHARED; in hugetlb_mfill_atomic_pte()
6603 int ret = -ENOMEM; in hugetlb_mfill_atomic_pte()
6613 return -EEXIST; in hugetlb_mfill_atomic_pte()
6619 /* No need to invalidate - it was non-present before */ in hugetlb_mfill_atomic_pte()
6627 ret = -EFAULT; in hugetlb_mfill_atomic_pte()
6634 * a non-missing case. Return -EEXIST. in hugetlb_mfill_atomic_pte()
6638 ret = -EEXIST; in hugetlb_mfill_atomic_pte()
6644 ret = -ENOMEM; in hugetlb_mfill_atomic_pte()
6653 ret = -ENOENT; in hugetlb_mfill_atomic_pte()
6665 ret = -ENOMEM; in hugetlb_mfill_atomic_pte()
6679 ret = -EEXIST; in hugetlb_mfill_atomic_pte()
6687 ret = -ENOMEM; in hugetlb_mfill_atomic_pte()
6718 ret = -EFAULT; in hugetlb_mfill_atomic_pte()
6719 if (idx >= (i_size_read(mapping->host) >> huge_page_shift(h))) in hugetlb_mfill_atomic_pte()
6736 ret = -EIO; in hugetlb_mfill_atomic_pte()
6742 * registered, we firstly wr-protect a none pte which has no page cache in hugetlb_mfill_atomic_pte()
6745 ret = -EEXIST; in hugetlb_mfill_atomic_pte()
6755 * For either: (1) CONTINUE on a non-shared VMA, or (2) UFFDIO_COPY in hugetlb_mfill_atomic_pte()
6758 _dst_pte = make_huge_pte(dst_vma, &folio->page, in hugetlb_mfill_atomic_pte()
6776 /* No need to invalidate - it was non-present before */ in hugetlb_mfill_atomic_pte()
6803 struct mm_struct *mm = vma->vm_mm; in hugetlb_change_protection()
6829 i_mmap_lock_write(vma->vm_file->f_mapping); in hugetlb_change_protection()
6840 * Userfaultfd wr-protect requires pgtable in hugetlb_change_protection()
6841 * pre-allocations to install pte markers. in hugetlb_change_protection()
6845 pages = -ENOMEM; in hugetlb_change_protection()
6852 * When uffd-wp is enabled on the vma, unshare in hugetlb_change_protection()
6896 /* Safe to modify directly (non-present->none). */ in hugetlb_change_protection()
6904 pte = arch_make_huge_pte(pte, shift, vma->vm_flags); in hugetlb_change_protection()
6914 /* Safe to modify directly (none->non-present). */ in hugetlb_change_protection()
6939 i_mmap_unlock_write(vma->vm_file->f_mapping); in hugetlb_change_protection()
6943 return pages > 0 ? (pages << h->order) : pages; in hugetlb_change_protection()
6952 long chg = -1, add = -1; in hugetlb_reserve_pages()
6982 * to reserve the full area even if read-only as mprotect() may be in hugetlb_reserve_pages()
6983 * called to make the mapping read-write. Assume !vma is a shm mapping in hugetlb_reserve_pages()
6985 if (!vma || vma->vm_flags & VM_MAYSHARE) { in hugetlb_reserve_pages()
7000 chg = to - from; in hugetlb_reserve_pages()
7013 if (vma && !(vma->vm_flags & VM_MAYSHARE) && h_cg) { in hugetlb_reserve_pages()
7041 * the reservation was consumed. Private mappings are per-VMA and in hugetlb_reserve_pages()
7047 if (!vma || vma->vm_flags & VM_MAYSHARE) { in hugetlb_reserve_pages()
7051 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_reserve_pages()
7065 * reference to h_cg->css. See comment below for detail. in hugetlb_reserve_pages()
7069 (chg - add) * pages_per_huge_page(h), h_cg); in hugetlb_reserve_pages()
7072 chg - add); in hugetlb_reserve_pages()
7073 hugetlb_acct_memory(h, -rsv_adjust); in hugetlb_reserve_pages()
7077 * h_cg->css. So we should release the reference held in hugetlb_reserve_pages()
7094 if (!vma || vma->vm_flags & VM_MAYSHARE) in hugetlb_reserve_pages()
7101 kref_put(&resv_map->refs, resv_map_release); in hugetlb_reserve_pages()
7131 spin_lock(&inode->i_lock); in hugetlb_unreserve_pages()
7132 inode->i_blocks -= (blocks_per_huge_page(h) * freed); in hugetlb_unreserve_pages()
7133 spin_unlock(&inode->i_lock); in hugetlb_unreserve_pages()
7139 * Note that !resv_map implies freed == 0. So (chg - freed) in hugetlb_unreserve_pages()
7142 gbl_reserve = hugepage_subpool_put_pages(spool, (chg - freed)); in hugetlb_unreserve_pages()
7143 hugetlb_acct_memory(h, -gbl_reserve); in hugetlb_unreserve_pages()
7153 unsigned long saddr = ((idx - svma->vm_pgoff) << PAGE_SHIFT) + in page_table_shareable()
7154 svma->vm_start; in page_table_shareable()
7159 unsigned long vm_flags = vma->vm_flags & ~VM_LOCKED_MASK; in page_table_shareable()
7160 unsigned long svm_flags = svma->vm_flags & ~VM_LOCKED_MASK; in page_table_shareable()
7171 !svma->vm_private_data) in page_table_shareable()
7189 if (!(vma->vm_flags & VM_MAYSHARE)) in want_pmd_share()
7191 if (!vma->vm_private_data) /* vma lock required for sharing */ in want_pmd_share()
7206 unsigned long v_start = ALIGN(vma->vm_start, PUD_SIZE), in adjust_range_if_pmd_sharing_possible()
7207 v_end = ALIGN_DOWN(vma->vm_end, PUD_SIZE); in adjust_range_if_pmd_sharing_possible()
7213 if (!(vma->vm_flags & VM_MAYSHARE) || !(v_end > v_start) || in adjust_range_if_pmd_sharing_possible()
7230 * pud has to be populated inside the same i_mmap_rwsem section - otherwise
7237 struct address_space *mapping = vma->vm_file->f_mapping; in huge_pmd_share()
7238 pgoff_t idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + in huge_pmd_share()
7239 vma->vm_pgoff; in huge_pmd_share()
7246 vma_interval_tree_foreach(svma, &mapping->i_mmap, idx, idx) { in huge_pmd_share()
7264 spin_lock(&mm->page_table_lock); in huge_pmd_share()
7272 spin_unlock(&mm->page_table_lock); in huge_pmd_share()
7295 i_mmap_assert_write_locked(vma->vm_file->f_mapping); in huge_pmd_unshare()
7369 * huge_pte_offset() - Walk the page table to resolve the hugepage
7394 /* must be pud huge, non-present or none */ in huge_pte_offset()
7401 /* must be pmd huge, non-present or none */ in huge_pte_offset()
7407 * page in a page table page mapping size. Used to skip non-present
7417 return P4D_SIZE - PUD_SIZE; in hugetlb_mask_last_page()
7419 return PUD_SIZE - PMD_SIZE; in hugetlb_mask_last_page()
7431 return PUD_SIZE - PMD_SIZE; in hugetlb_mask_last_page()
7439 * folio_isolate_hugetlb - try to isolate an allocated hugetlb folio
7444 * isolated/non-migratable, and moving it from the active list to the
7448 * it is already isolated/non-migratable.
7467 list_move_tail(&folio->lru, list); in folio_isolate_hugetlb()
7486 ret = -EBUSY; in get_hwpoison_hugetlb_folio()
7504 * folio_putback_hugetlb - unisolate a hugetlb folio
7507 * Putback/un-isolate the hugetlb folio that was previous isolated using
7508 * folio_isolate_hugetlb(): marking it non-isolated/migratable and putting it
7518 list_move_tail(&folio->lru, &(folio_hstate(folio))->hugepage_activelist); in folio_putback_hugetlb()
7528 set_page_owner_migrate_reason(&new_folio->page, reason); in move_hugetlb_state()
7536 * Also note that we have to transfer the per-node surplus state in move_hugetlb_state()
7538 * the per-node's. in move_hugetlb_state()
7549 * There is no need to transfer the per-node surplus state in move_hugetlb_state()
7555 if (h->surplus_huge_pages_node[old_nid]) { in move_hugetlb_state()
7556 h->surplus_huge_pages_node[old_nid]--; in move_hugetlb_state()
7557 h->surplus_huge_pages_node[new_nid]++; in move_hugetlb_state()
7569 list_move_tail(&new_folio->lru, &(folio_hstate(new_folio))->hugepage_activelist); in move_hugetlb_state()
7579 struct mm_struct *mm = vma->vm_mm; in hugetlb_unshare_pmds()
7585 if (!(vma->vm_flags & VM_MAYSHARE)) in hugetlb_unshare_pmds()
7600 i_mmap_lock_write(vma->vm_file->f_mapping); in hugetlb_unshare_pmds()
7610 i_mmap_unlock_write(vma->vm_file->f_mapping); in hugetlb_unshare_pmds()
7625 hugetlb_unshare_pmds(vma, ALIGN(vma->vm_start, PUD_SIZE), in hugetlb_unshare_all_pmds()
7626 ALIGN_DOWN(vma->vm_end, PUD_SIZE)); in hugetlb_unshare_all_pmds()
7695 hugetlb_cma_size -= hugetlb_cma_size_in_node[nid]; in hugetlb_cma_reserve()
7703 hugetlb_cma_size -= hugetlb_cma_size_in_node[nid]; in hugetlb_cma_reserve()
7742 size = min(per_node, hugetlb_cma_size - reserved); in hugetlb_cma_reserve()
7784 pr_warn("hugetlb_cma: the option isn't supported by current arch\n"); in hugetlb_cma_check()