1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 2b20a3503SChristoph Lameter /* 314e0f9bcSHugh Dickins * Memory Migration functionality - linux/mm/migrate.c 4b20a3503SChristoph Lameter * 5b20a3503SChristoph Lameter * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter 6b20a3503SChristoph Lameter * 7b20a3503SChristoph Lameter * Page migration was first developed in the context of the memory hotplug 8b20a3503SChristoph Lameter * project. The main authors of the migration code are: 9b20a3503SChristoph Lameter * 10b20a3503SChristoph Lameter * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> 11b20a3503SChristoph Lameter * Hirokazu Takahashi <taka@valinux.co.jp> 12b20a3503SChristoph Lameter * Dave Hansen <haveblue@us.ibm.com> 13cde53535SChristoph Lameter * Christoph Lameter 14b20a3503SChristoph Lameter */ 15b20a3503SChristoph Lameter 16b20a3503SChristoph Lameter #include <linux/migrate.h> 17b95f1b31SPaul Gortmaker #include <linux/export.h> 18b20a3503SChristoph Lameter #include <linux/swap.h> 190697212aSChristoph Lameter #include <linux/swapops.h> 20b20a3503SChristoph Lameter #include <linux/pagemap.h> 21e23ca00bSChristoph Lameter #include <linux/buffer_head.h> 22b20a3503SChristoph Lameter #include <linux/mm_inline.h> 23b488893aSPavel Emelyanov #include <linux/nsproxy.h> 24b20a3503SChristoph Lameter #include <linux/pagevec.h> 25e9995ef9SHugh Dickins #include <linux/ksm.h> 26b20a3503SChristoph Lameter #include <linux/rmap.h> 27b20a3503SChristoph Lameter #include <linux/topology.h> 28b20a3503SChristoph Lameter #include <linux/cpu.h> 29b20a3503SChristoph Lameter #include <linux/cpuset.h> 3004e62a29SChristoph Lameter #include <linux/writeback.h> 31742755a1SChristoph Lameter #include <linux/mempolicy.h> 32742755a1SChristoph Lameter #include <linux/vmalloc.h> 3386c3a764SDavid Quigley #include <linux/security.h> 3442cb14b1SHugh Dickins #include <linux/backing-dev.h> 35bda807d4SMinchan Kim #include <linux/compaction.h> 364f5ca265SAdrian Bunk #include <linux/syscalls.h> 377addf443SDominik Brodowski #include <linux/compat.h> 38290408d4SNaoya Horiguchi #include <linux/hugetlb.h> 398e6ac7faSAneesh Kumar K.V #include <linux/hugetlb_cgroup.h> 405a0e3ad6STejun Heo #include <linux/gfp.h> 41df6ad698SJérôme Glisse #include <linux/pfn_t.h> 42a5430ddaSJérôme Glisse #include <linux/memremap.h> 438315ada7SJérôme Glisse #include <linux/userfaultfd_k.h> 44bf6bddf1SRafael Aquini #include <linux/balloon_compaction.h> 4533c3fc71SVladimir Davydov #include <linux/page_idle.h> 46d435edcaSVlastimil Babka #include <linux/page_owner.h> 476e84f315SIngo Molnar #include <linux/sched/mm.h> 48197e7e52SLinus Torvalds #include <linux/ptrace.h> 4934290e2cSRalph Campbell #include <linux/oom.h> 50884a6e5dSDave Hansen #include <linux/memory.h> 51ac16ec83SBaolin Wang #include <linux/random.h> 52c574bbe9SHuang Ying #include <linux/sched/sysctl.h> 53467b171aSAneesh Kumar K.V #include <linux/memory-tiers.h> 54b20a3503SChristoph Lameter 550d1836c3SMichal Nazarewicz #include <asm/tlbflush.h> 560d1836c3SMichal Nazarewicz 577b2a2d4aSMel Gorman #include <trace/events/migrate.h> 587b2a2d4aSMel Gorman 59b20a3503SChristoph Lameter #include "internal.h" 60b20a3503SChristoph Lameter 619e5bcd61SYisheng Xie int isolate_movable_page(struct page *page, isolate_mode_t mode) 62bda807d4SMinchan Kim { 6368f2736aSMatthew Wilcox (Oracle) const struct movable_operations *mops; 64bda807d4SMinchan Kim 65bda807d4SMinchan Kim /* 66bda807d4SMinchan Kim * Avoid burning cycles with pages that are yet under __free_pages(), 67bda807d4SMinchan Kim * or just got freed under us. 68bda807d4SMinchan Kim * 69bda807d4SMinchan Kim * In case we 'win' a race for a movable page being freed under us and 70bda807d4SMinchan Kim * raise its refcount preventing __free_pages() from doing its job 71bda807d4SMinchan Kim * the put_page() at the end of this block will take care of 72bda807d4SMinchan Kim * release this page, thus avoiding a nasty leakage. 73bda807d4SMinchan Kim */ 74bda807d4SMinchan Kim if (unlikely(!get_page_unless_zero(page))) 75bda807d4SMinchan Kim goto out; 76bda807d4SMinchan Kim 77bda807d4SMinchan Kim /* 78bda807d4SMinchan Kim * Check PageMovable before holding a PG_lock because page's owner 79bda807d4SMinchan Kim * assumes anybody doesn't touch PG_lock of newly allocated page 808bb4e7a2SWei Yang * so unconditionally grabbing the lock ruins page's owner side. 81bda807d4SMinchan Kim */ 82bda807d4SMinchan Kim if (unlikely(!__PageMovable(page))) 83bda807d4SMinchan Kim goto out_putpage; 84bda807d4SMinchan Kim /* 85bda807d4SMinchan Kim * As movable pages are not isolated from LRU lists, concurrent 86bda807d4SMinchan Kim * compaction threads can race against page migration functions 87bda807d4SMinchan Kim * as well as race against the releasing a page. 88bda807d4SMinchan Kim * 89bda807d4SMinchan Kim * In order to avoid having an already isolated movable page 90bda807d4SMinchan Kim * being (wrongly) re-isolated while it is under migration, 91bda807d4SMinchan Kim * or to avoid attempting to isolate pages being released, 92bda807d4SMinchan Kim * lets be sure we have the page lock 93bda807d4SMinchan Kim * before proceeding with the movable page isolation steps. 94bda807d4SMinchan Kim */ 95bda807d4SMinchan Kim if (unlikely(!trylock_page(page))) 96bda807d4SMinchan Kim goto out_putpage; 97bda807d4SMinchan Kim 98bda807d4SMinchan Kim if (!PageMovable(page) || PageIsolated(page)) 99bda807d4SMinchan Kim goto out_no_isolated; 100bda807d4SMinchan Kim 10168f2736aSMatthew Wilcox (Oracle) mops = page_movable_ops(page); 10268f2736aSMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(!mops, page); 103bda807d4SMinchan Kim 10468f2736aSMatthew Wilcox (Oracle) if (!mops->isolate_page(page, mode)) 105bda807d4SMinchan Kim goto out_no_isolated; 106bda807d4SMinchan Kim 107bda807d4SMinchan Kim /* Driver shouldn't use PG_isolated bit of page->flags */ 108bda807d4SMinchan Kim WARN_ON_ONCE(PageIsolated(page)); 109356ea386Sandrew.yang SetPageIsolated(page); 110bda807d4SMinchan Kim unlock_page(page); 111bda807d4SMinchan Kim 1129e5bcd61SYisheng Xie return 0; 113bda807d4SMinchan Kim 114bda807d4SMinchan Kim out_no_isolated: 115bda807d4SMinchan Kim unlock_page(page); 116bda807d4SMinchan Kim out_putpage: 117bda807d4SMinchan Kim put_page(page); 118bda807d4SMinchan Kim out: 1199e5bcd61SYisheng Xie return -EBUSY; 120bda807d4SMinchan Kim } 121bda807d4SMinchan Kim 122606a6f71SMiaohe Lin static void putback_movable_page(struct page *page) 123bda807d4SMinchan Kim { 12468f2736aSMatthew Wilcox (Oracle) const struct movable_operations *mops = page_movable_ops(page); 125bda807d4SMinchan Kim 12668f2736aSMatthew Wilcox (Oracle) mops->putback_page(page); 127356ea386Sandrew.yang ClearPageIsolated(page); 128bda807d4SMinchan Kim } 129bda807d4SMinchan Kim 130b20a3503SChristoph Lameter /* 1315733c7d1SRafael Aquini * Put previously isolated pages back onto the appropriate lists 1325733c7d1SRafael Aquini * from where they were once taken off for compaction/migration. 1335733c7d1SRafael Aquini * 13459c82b70SJoonsoo Kim * This function shall be used whenever the isolated pageset has been 13559c82b70SJoonsoo Kim * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() 1367ce82f4cSMiaohe Lin * and isolate_hugetlb(). 1375733c7d1SRafael Aquini */ 1385733c7d1SRafael Aquini void putback_movable_pages(struct list_head *l) 1395733c7d1SRafael Aquini { 1405733c7d1SRafael Aquini struct page *page; 1415733c7d1SRafael Aquini struct page *page2; 1425733c7d1SRafael Aquini 1435733c7d1SRafael Aquini list_for_each_entry_safe(page, page2, l, lru) { 14431caf665SNaoya Horiguchi if (unlikely(PageHuge(page))) { 14531caf665SNaoya Horiguchi putback_active_hugepage(page); 14631caf665SNaoya Horiguchi continue; 14731caf665SNaoya Horiguchi } 1485733c7d1SRafael Aquini list_del(&page->lru); 149bda807d4SMinchan Kim /* 150bda807d4SMinchan Kim * We isolated non-lru movable page so here we can use 151bda807d4SMinchan Kim * __PageMovable because LRU page's mapping cannot have 152bda807d4SMinchan Kim * PAGE_MAPPING_MOVABLE. 153bda807d4SMinchan Kim */ 154b1123ea6SMinchan Kim if (unlikely(__PageMovable(page))) { 155bda807d4SMinchan Kim VM_BUG_ON_PAGE(!PageIsolated(page), page); 156bda807d4SMinchan Kim lock_page(page); 157bda807d4SMinchan Kim if (PageMovable(page)) 158bda807d4SMinchan Kim putback_movable_page(page); 159bf6bddf1SRafael Aquini else 160356ea386Sandrew.yang ClearPageIsolated(page); 161bda807d4SMinchan Kim unlock_page(page); 162bda807d4SMinchan Kim put_page(page); 163bda807d4SMinchan Kim } else { 164e8db67ebSNaoya Horiguchi mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + 1656c357848SMatthew Wilcox (Oracle) page_is_file_lru(page), -thp_nr_pages(page)); 166fc280fe8SRabin Vincent putback_lru_page(page); 167b20a3503SChristoph Lameter } 168b20a3503SChristoph Lameter } 169bda807d4SMinchan Kim } 170b20a3503SChristoph Lameter 1710697212aSChristoph Lameter /* 1720697212aSChristoph Lameter * Restore a potential migration pte to a working pte entry 1730697212aSChristoph Lameter */ 1742f031c6fSMatthew Wilcox (Oracle) static bool remove_migration_pte(struct folio *folio, 1752f031c6fSMatthew Wilcox (Oracle) struct vm_area_struct *vma, unsigned long addr, void *old) 1760697212aSChristoph Lameter { 1774eecb8b9SMatthew Wilcox (Oracle) DEFINE_FOLIO_VMA_WALK(pvmw, old, vma, addr, PVMW_SYNC | PVMW_MIGRATION); 1780697212aSChristoph Lameter 1793fe87967SKirill A. Shutemov while (page_vma_mapped_walk(&pvmw)) { 1806c287605SDavid Hildenbrand rmap_t rmap_flags = RMAP_NONE; 1810697212aSChristoph Lameter pte_t pte; 1820697212aSChristoph Lameter swp_entry_t entry; 1834eecb8b9SMatthew Wilcox (Oracle) struct page *new; 1844eecb8b9SMatthew Wilcox (Oracle) unsigned long idx = 0; 1850697212aSChristoph Lameter 1864eecb8b9SMatthew Wilcox (Oracle) /* pgoff is invalid for ksm pages, but they are never large */ 1874eecb8b9SMatthew Wilcox (Oracle) if (folio_test_large(folio) && !folio_test_hugetlb(folio)) 1884eecb8b9SMatthew Wilcox (Oracle) idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff; 1894eecb8b9SMatthew Wilcox (Oracle) new = folio_page(folio, idx); 1900697212aSChristoph Lameter 191616b8371SZi Yan #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 192616b8371SZi Yan /* PMD-mapped THP migration entry */ 193616b8371SZi Yan if (!pvmw.pte) { 1944eecb8b9SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_hugetlb(folio) || 1954eecb8b9SMatthew Wilcox (Oracle) !folio_test_pmd_mappable(folio), folio); 196616b8371SZi Yan remove_migration_pmd(&pvmw, new); 197616b8371SZi Yan continue; 198616b8371SZi Yan } 199616b8371SZi Yan #endif 200616b8371SZi Yan 2014eecb8b9SMatthew Wilcox (Oracle) folio_get(folio); 2022e346877SPeter Xu pte = mk_pte(new, READ_ONCE(vma->vm_page_prot)); 2033fe87967SKirill A. Shutemov if (pte_swp_soft_dirty(*pvmw.pte)) 204c3d16e16SCyrill Gorcunov pte = pte_mksoft_dirty(pte); 205d3cb8bf6SMel Gorman 2063fe87967SKirill A. Shutemov /* 2073fe87967SKirill A. Shutemov * Recheck VMA as permissions can change since migration started 2083fe87967SKirill A. Shutemov */ 2093fe87967SKirill A. Shutemov entry = pte_to_swp_entry(*pvmw.pte); 2102e346877SPeter Xu if (!is_migration_entry_young(entry)) 2112e346877SPeter Xu pte = pte_mkold(pte); 2122e346877SPeter Xu if (folio_test_dirty(folio) && is_migration_entry_dirty(entry)) 2132e346877SPeter Xu pte = pte_mkdirty(pte); 2144dd845b5SAlistair Popple if (is_writable_migration_entry(entry)) 215d3cb8bf6SMel Gorman pte = maybe_mkwrite(pte, vma); 216f45ec5ffSPeter Xu else if (pte_swp_uffd_wp(*pvmw.pte)) 217f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 218d3cb8bf6SMel Gorman 2196c287605SDavid Hildenbrand if (folio_test_anon(folio) && !is_readable_migration_entry(entry)) 2206c287605SDavid Hildenbrand rmap_flags |= RMAP_EXCLUSIVE; 2216c287605SDavid Hildenbrand 2226128763fSRalph Campbell if (unlikely(is_device_private_page(new))) { 2234dd845b5SAlistair Popple if (pte_write(pte)) 2244dd845b5SAlistair Popple entry = make_writable_device_private_entry( 2254dd845b5SAlistair Popple page_to_pfn(new)); 2264dd845b5SAlistair Popple else 2274dd845b5SAlistair Popple entry = make_readable_device_private_entry( 2284dd845b5SAlistair Popple page_to_pfn(new)); 229a5430ddaSJérôme Glisse pte = swp_entry_to_pte(entry); 2303d321bf8SRalph Campbell if (pte_swp_soft_dirty(*pvmw.pte)) 2313d321bf8SRalph Campbell pte = pte_swp_mksoft_dirty(pte); 232f45ec5ffSPeter Xu if (pte_swp_uffd_wp(*pvmw.pte)) 233ebdf8321SAlistair Popple pte = pte_swp_mkuffd_wp(pte); 234df6ad698SJérôme Glisse } 235a5430ddaSJérôme Glisse 2363ef8fd7fSAndi Kleen #ifdef CONFIG_HUGETLB_PAGE 2374eecb8b9SMatthew Wilcox (Oracle) if (folio_test_hugetlb(folio)) { 23879c1c594SChristophe Leroy unsigned int shift = huge_page_shift(hstate_vma(vma)); 23979c1c594SChristophe Leroy 240290408d4SNaoya Horiguchi pte = pte_mkhuge(pte); 24179c1c594SChristophe Leroy pte = arch_make_huge_pte(pte, shift, vma->vm_flags); 2424eecb8b9SMatthew Wilcox (Oracle) if (folio_test_anon(folio)) 24328c5209dSDavid Hildenbrand hugepage_add_anon_rmap(new, vma, pvmw.address, 2446c287605SDavid Hildenbrand rmap_flags); 245290408d4SNaoya Horiguchi else 246fb3d824dSDavid Hildenbrand page_dup_file_rmap(new, true); 2471eba86c0SPasha Tatashin set_huge_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); 248383321abSAneesh Kumar K.V } else 249383321abSAneesh Kumar K.V #endif 250383321abSAneesh Kumar K.V { 2514eecb8b9SMatthew Wilcox (Oracle) if (folio_test_anon(folio)) 252f1e2db12SDavid Hildenbrand page_add_anon_rmap(new, vma, pvmw.address, 2536c287605SDavid Hildenbrand rmap_flags); 25404e62a29SChristoph Lameter else 255cea86fe2SHugh Dickins page_add_file_rmap(new, vma, false); 2561eba86c0SPasha Tatashin set_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); 257383321abSAneesh Kumar K.V } 258b7435507SHugh Dickins if (vma->vm_flags & VM_LOCKED) 259adb11e78SSebastian Andrzej Siewior mlock_page_drain_local(); 260e125fe40SKirill A. Shutemov 2614cc79b33SAnshuman Khandual trace_remove_migration_pte(pvmw.address, pte_val(pte), 2624cc79b33SAnshuman Khandual compound_order(new)); 2634cc79b33SAnshuman Khandual 26404e62a29SChristoph Lameter /* No need to invalidate - it was non-present before */ 2653fe87967SKirill A. Shutemov update_mmu_cache(vma, pvmw.address, pvmw.pte); 2663fe87967SKirill A. Shutemov } 2673fe87967SKirill A. Shutemov 268e4b82222SMinchan Kim return true; 2690697212aSChristoph Lameter } 2700697212aSChristoph Lameter 2710697212aSChristoph Lameter /* 27204e62a29SChristoph Lameter * Get rid of all migration entries and replace them by 27304e62a29SChristoph Lameter * references to the indicated page. 27404e62a29SChristoph Lameter */ 2754eecb8b9SMatthew Wilcox (Oracle) void remove_migration_ptes(struct folio *src, struct folio *dst, bool locked) 27604e62a29SChristoph Lameter { 277051ac83aSJoonsoo Kim struct rmap_walk_control rwc = { 278051ac83aSJoonsoo Kim .rmap_one = remove_migration_pte, 2794eecb8b9SMatthew Wilcox (Oracle) .arg = src, 280051ac83aSJoonsoo Kim }; 281051ac83aSJoonsoo Kim 282e388466dSKirill A. Shutemov if (locked) 2832f031c6fSMatthew Wilcox (Oracle) rmap_walk_locked(dst, &rwc); 284e388466dSKirill A. Shutemov else 2852f031c6fSMatthew Wilcox (Oracle) rmap_walk(dst, &rwc); 28604e62a29SChristoph Lameter } 28704e62a29SChristoph Lameter 28804e62a29SChristoph Lameter /* 2890697212aSChristoph Lameter * Something used the pte of a page under migration. We need to 2900697212aSChristoph Lameter * get to the page and wait until migration is finished. 2910697212aSChristoph Lameter * When we return from this function the fault will be retried. 2920697212aSChristoph Lameter */ 293e66f17ffSNaoya Horiguchi void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep, 29430dad309SNaoya Horiguchi spinlock_t *ptl) 2950697212aSChristoph Lameter { 29630dad309SNaoya Horiguchi pte_t pte; 2970697212aSChristoph Lameter swp_entry_t entry; 2980697212aSChristoph Lameter 29930dad309SNaoya Horiguchi spin_lock(ptl); 3000697212aSChristoph Lameter pte = *ptep; 3010697212aSChristoph Lameter if (!is_swap_pte(pte)) 3020697212aSChristoph Lameter goto out; 3030697212aSChristoph Lameter 3040697212aSChristoph Lameter entry = pte_to_swp_entry(pte); 3050697212aSChristoph Lameter if (!is_migration_entry(entry)) 3060697212aSChristoph Lameter goto out; 3070697212aSChristoph Lameter 308ffa65753SAlistair Popple migration_entry_wait_on_locked(entry, ptep, ptl); 3090697212aSChristoph Lameter return; 3100697212aSChristoph Lameter out: 3110697212aSChristoph Lameter pte_unmap_unlock(ptep, ptl); 3120697212aSChristoph Lameter } 3130697212aSChristoph Lameter 31430dad309SNaoya Horiguchi void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, 31530dad309SNaoya Horiguchi unsigned long address) 31630dad309SNaoya Horiguchi { 31730dad309SNaoya Horiguchi spinlock_t *ptl = pte_lockptr(mm, pmd); 31830dad309SNaoya Horiguchi pte_t *ptep = pte_offset_map(pmd, address); 31930dad309SNaoya Horiguchi __migration_entry_wait(mm, ptep, ptl); 32030dad309SNaoya Horiguchi } 32130dad309SNaoya Horiguchi 322ad1ac596SMiaohe Lin #ifdef CONFIG_HUGETLB_PAGE 323ad1ac596SMiaohe Lin void __migration_entry_wait_huge(pte_t *ptep, spinlock_t *ptl) 32430dad309SNaoya Horiguchi { 325ad1ac596SMiaohe Lin pte_t pte; 326ad1ac596SMiaohe Lin 327ad1ac596SMiaohe Lin spin_lock(ptl); 328ad1ac596SMiaohe Lin pte = huge_ptep_get(ptep); 329ad1ac596SMiaohe Lin 330ad1ac596SMiaohe Lin if (unlikely(!is_hugetlb_entry_migration(pte))) 331ad1ac596SMiaohe Lin spin_unlock(ptl); 332ad1ac596SMiaohe Lin else 333ad1ac596SMiaohe Lin migration_entry_wait_on_locked(pte_to_swp_entry(pte), NULL, ptl); 33430dad309SNaoya Horiguchi } 33530dad309SNaoya Horiguchi 336ad1ac596SMiaohe Lin void migration_entry_wait_huge(struct vm_area_struct *vma, pte_t *pte) 337ad1ac596SMiaohe Lin { 338ad1ac596SMiaohe Lin spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), vma->vm_mm, pte); 339ad1ac596SMiaohe Lin 340ad1ac596SMiaohe Lin __migration_entry_wait_huge(pte, ptl); 341ad1ac596SMiaohe Lin } 342ad1ac596SMiaohe Lin #endif 343ad1ac596SMiaohe Lin 344616b8371SZi Yan #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 345616b8371SZi Yan void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd) 346616b8371SZi Yan { 347616b8371SZi Yan spinlock_t *ptl; 348616b8371SZi Yan 349616b8371SZi Yan ptl = pmd_lock(mm, pmd); 350616b8371SZi Yan if (!is_pmd_migration_entry(*pmd)) 351616b8371SZi Yan goto unlock; 352ffa65753SAlistair Popple migration_entry_wait_on_locked(pmd_to_swp_entry(*pmd), NULL, ptl); 353616b8371SZi Yan return; 354616b8371SZi Yan unlock: 355616b8371SZi Yan spin_unlock(ptl); 356616b8371SZi Yan } 357616b8371SZi Yan #endif 358616b8371SZi Yan 359108ca835SMatthew Wilcox (Oracle) static int folio_expected_refs(struct address_space *mapping, 360108ca835SMatthew Wilcox (Oracle) struct folio *folio) 3610b3901b3SJan Kara { 362108ca835SMatthew Wilcox (Oracle) int refs = 1; 363108ca835SMatthew Wilcox (Oracle) if (!mapping) 364108ca835SMatthew Wilcox (Oracle) return refs; 3650b3901b3SJan Kara 366108ca835SMatthew Wilcox (Oracle) refs += folio_nr_pages(folio); 367108ca835SMatthew Wilcox (Oracle) if (folio_test_private(folio)) 368108ca835SMatthew Wilcox (Oracle) refs++; 369108ca835SMatthew Wilcox (Oracle) 370108ca835SMatthew Wilcox (Oracle) return refs; 3710b3901b3SJan Kara } 3720b3901b3SJan Kara 373b20a3503SChristoph Lameter /* 374c3fcf8a5SChristoph Lameter * Replace the page in the mapping. 3755b5c7120SChristoph Lameter * 3765b5c7120SChristoph Lameter * The number of remaining references must be: 3775b5c7120SChristoph Lameter * 1 for anonymous pages without a mapping 3785b5c7120SChristoph Lameter * 2 for pages with a mapping 379266cf658SDavid Howells * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. 380b20a3503SChristoph Lameter */ 3813417013eSMatthew Wilcox (Oracle) int folio_migrate_mapping(struct address_space *mapping, 3823417013eSMatthew Wilcox (Oracle) struct folio *newfolio, struct folio *folio, int extra_count) 383b20a3503SChristoph Lameter { 3843417013eSMatthew Wilcox (Oracle) XA_STATE(xas, &mapping->i_pages, folio_index(folio)); 38542cb14b1SHugh Dickins struct zone *oldzone, *newzone; 38642cb14b1SHugh Dickins int dirty; 387108ca835SMatthew Wilcox (Oracle) int expected_count = folio_expected_refs(mapping, folio) + extra_count; 3883417013eSMatthew Wilcox (Oracle) long nr = folio_nr_pages(folio); 3898763cb45SJérôme Glisse 3906c5240aeSChristoph Lameter if (!mapping) { 3910e8c7d0fSChristoph Lameter /* Anonymous page without mapping */ 3923417013eSMatthew Wilcox (Oracle) if (folio_ref_count(folio) != expected_count) 3936c5240aeSChristoph Lameter return -EAGAIN; 394cf4b769aSHugh Dickins 395cf4b769aSHugh Dickins /* No turning back from here */ 3963417013eSMatthew Wilcox (Oracle) newfolio->index = folio->index; 3973417013eSMatthew Wilcox (Oracle) newfolio->mapping = folio->mapping; 3983417013eSMatthew Wilcox (Oracle) if (folio_test_swapbacked(folio)) 3993417013eSMatthew Wilcox (Oracle) __folio_set_swapbacked(newfolio); 400cf4b769aSHugh Dickins 40178bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 4026c5240aeSChristoph Lameter } 4036c5240aeSChristoph Lameter 4043417013eSMatthew Wilcox (Oracle) oldzone = folio_zone(folio); 4053417013eSMatthew Wilcox (Oracle) newzone = folio_zone(newfolio); 40642cb14b1SHugh Dickins 40789eb946aSMatthew Wilcox xas_lock_irq(&xas); 4083417013eSMatthew Wilcox (Oracle) if (!folio_ref_freeze(folio, expected_count)) { 40989eb946aSMatthew Wilcox xas_unlock_irq(&xas); 410e286781dSNick Piggin return -EAGAIN; 411e286781dSNick Piggin } 412e286781dSNick Piggin 413b20a3503SChristoph Lameter /* 4143417013eSMatthew Wilcox (Oracle) * Now we know that no one else is looking at the folio: 415cf4b769aSHugh Dickins * no turning back from here. 416b20a3503SChristoph Lameter */ 4173417013eSMatthew Wilcox (Oracle) newfolio->index = folio->index; 4183417013eSMatthew Wilcox (Oracle) newfolio->mapping = folio->mapping; 4193417013eSMatthew Wilcox (Oracle) folio_ref_add(newfolio, nr); /* add cache reference */ 4203417013eSMatthew Wilcox (Oracle) if (folio_test_swapbacked(folio)) { 4213417013eSMatthew Wilcox (Oracle) __folio_set_swapbacked(newfolio); 4223417013eSMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) { 4233417013eSMatthew Wilcox (Oracle) folio_set_swapcache(newfolio); 4243417013eSMatthew Wilcox (Oracle) newfolio->private = folio_get_private(folio); 425b20a3503SChristoph Lameter } 4266326fec1SNicholas Piggin } else { 4273417013eSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio); 4286326fec1SNicholas Piggin } 429b20a3503SChristoph Lameter 43042cb14b1SHugh Dickins /* Move dirty while page refs frozen and newpage not yet exposed */ 4313417013eSMatthew Wilcox (Oracle) dirty = folio_test_dirty(folio); 43242cb14b1SHugh Dickins if (dirty) { 4333417013eSMatthew Wilcox (Oracle) folio_clear_dirty(folio); 4343417013eSMatthew Wilcox (Oracle) folio_set_dirty(newfolio); 43542cb14b1SHugh Dickins } 43642cb14b1SHugh Dickins 4373417013eSMatthew Wilcox (Oracle) xas_store(&xas, newfolio); 4387cf9c2c7SNick Piggin 4397cf9c2c7SNick Piggin /* 440937a94c9SJacobo Giralt * Drop cache reference from old page by unfreezing 441937a94c9SJacobo Giralt * to one less reference. 4427cf9c2c7SNick Piggin * We know this isn't the last reference. 4437cf9c2c7SNick Piggin */ 4443417013eSMatthew Wilcox (Oracle) folio_ref_unfreeze(folio, expected_count - nr); 4457cf9c2c7SNick Piggin 44689eb946aSMatthew Wilcox xas_unlock(&xas); 44742cb14b1SHugh Dickins /* Leave irq disabled to prevent preemption while updating stats */ 44842cb14b1SHugh Dickins 4490e8c7d0fSChristoph Lameter /* 4500e8c7d0fSChristoph Lameter * If moved to a different zone then also account 4510e8c7d0fSChristoph Lameter * the page for that zone. Other VM counters will be 4520e8c7d0fSChristoph Lameter * taken care of when we establish references to the 4530e8c7d0fSChristoph Lameter * new page and drop references to the old page. 4540e8c7d0fSChristoph Lameter * 4550e8c7d0fSChristoph Lameter * Note that anonymous pages are accounted for 4564b9d0fabSMel Gorman * via NR_FILE_PAGES and NR_ANON_MAPPED if they 4570e8c7d0fSChristoph Lameter * are mapped to swap space. 4580e8c7d0fSChristoph Lameter */ 45942cb14b1SHugh Dickins if (newzone != oldzone) { 4600d1c2072SJohannes Weiner struct lruvec *old_lruvec, *new_lruvec; 4610d1c2072SJohannes Weiner struct mem_cgroup *memcg; 4620d1c2072SJohannes Weiner 4633417013eSMatthew Wilcox (Oracle) memcg = folio_memcg(folio); 4640d1c2072SJohannes Weiner old_lruvec = mem_cgroup_lruvec(memcg, oldzone->zone_pgdat); 4650d1c2072SJohannes Weiner new_lruvec = mem_cgroup_lruvec(memcg, newzone->zone_pgdat); 4660d1c2072SJohannes Weiner 4675c447d27SShakeel Butt __mod_lruvec_state(old_lruvec, NR_FILE_PAGES, -nr); 4685c447d27SShakeel Butt __mod_lruvec_state(new_lruvec, NR_FILE_PAGES, nr); 4693417013eSMatthew Wilcox (Oracle) if (folio_test_swapbacked(folio) && !folio_test_swapcache(folio)) { 4705c447d27SShakeel Butt __mod_lruvec_state(old_lruvec, NR_SHMEM, -nr); 4715c447d27SShakeel Butt __mod_lruvec_state(new_lruvec, NR_SHMEM, nr); 4724b02108aSKOSAKI Motohiro } 473b6038942SShakeel Butt #ifdef CONFIG_SWAP 4743417013eSMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) { 475b6038942SShakeel Butt __mod_lruvec_state(old_lruvec, NR_SWAPCACHE, -nr); 476b6038942SShakeel Butt __mod_lruvec_state(new_lruvec, NR_SWAPCACHE, nr); 477b6038942SShakeel Butt } 478b6038942SShakeel Butt #endif 479f56753acSChristoph Hellwig if (dirty && mapping_can_writeback(mapping)) { 4805c447d27SShakeel Butt __mod_lruvec_state(old_lruvec, NR_FILE_DIRTY, -nr); 4815c447d27SShakeel Butt __mod_zone_page_state(oldzone, NR_ZONE_WRITE_PENDING, -nr); 4825c447d27SShakeel Butt __mod_lruvec_state(new_lruvec, NR_FILE_DIRTY, nr); 4835c447d27SShakeel Butt __mod_zone_page_state(newzone, NR_ZONE_WRITE_PENDING, nr); 48442cb14b1SHugh Dickins } 48542cb14b1SHugh Dickins } 48642cb14b1SHugh Dickins local_irq_enable(); 487b20a3503SChristoph Lameter 48878bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 489b20a3503SChristoph Lameter } 4903417013eSMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_migrate_mapping); 491b20a3503SChristoph Lameter 492b20a3503SChristoph Lameter /* 493290408d4SNaoya Horiguchi * The expected number of remaining references is the same as that 4943417013eSMatthew Wilcox (Oracle) * of folio_migrate_mapping(). 495290408d4SNaoya Horiguchi */ 496290408d4SNaoya Horiguchi int migrate_huge_page_move_mapping(struct address_space *mapping, 497b890ec2aSMatthew Wilcox (Oracle) struct folio *dst, struct folio *src) 498290408d4SNaoya Horiguchi { 499b890ec2aSMatthew Wilcox (Oracle) XA_STATE(xas, &mapping->i_pages, folio_index(src)); 500290408d4SNaoya Horiguchi int expected_count; 501290408d4SNaoya Horiguchi 50289eb946aSMatthew Wilcox xas_lock_irq(&xas); 503b890ec2aSMatthew Wilcox (Oracle) expected_count = 2 + folio_has_private(src); 504b890ec2aSMatthew Wilcox (Oracle) if (!folio_ref_freeze(src, expected_count)) { 50589eb946aSMatthew Wilcox xas_unlock_irq(&xas); 506290408d4SNaoya Horiguchi return -EAGAIN; 507290408d4SNaoya Horiguchi } 508290408d4SNaoya Horiguchi 509b890ec2aSMatthew Wilcox (Oracle) dst->index = src->index; 510b890ec2aSMatthew Wilcox (Oracle) dst->mapping = src->mapping; 5116a93ca8fSJohannes Weiner 512b890ec2aSMatthew Wilcox (Oracle) folio_get(dst); 513290408d4SNaoya Horiguchi 514b890ec2aSMatthew Wilcox (Oracle) xas_store(&xas, dst); 515290408d4SNaoya Horiguchi 516b890ec2aSMatthew Wilcox (Oracle) folio_ref_unfreeze(src, expected_count - 1); 517290408d4SNaoya Horiguchi 51889eb946aSMatthew Wilcox xas_unlock_irq(&xas); 5196a93ca8fSJohannes Weiner 52078bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 521290408d4SNaoya Horiguchi } 522290408d4SNaoya Horiguchi 523290408d4SNaoya Horiguchi /* 52419138349SMatthew Wilcox (Oracle) * Copy the flags and some other ancillary information 525b20a3503SChristoph Lameter */ 52619138349SMatthew Wilcox (Oracle) void folio_migrate_flags(struct folio *newfolio, struct folio *folio) 527b20a3503SChristoph Lameter { 5287851a45cSRik van Riel int cpupid; 5297851a45cSRik van Riel 53019138349SMatthew Wilcox (Oracle) if (folio_test_error(folio)) 53119138349SMatthew Wilcox (Oracle) folio_set_error(newfolio); 53219138349SMatthew Wilcox (Oracle) if (folio_test_referenced(folio)) 53319138349SMatthew Wilcox (Oracle) folio_set_referenced(newfolio); 53419138349SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 53519138349SMatthew Wilcox (Oracle) folio_mark_uptodate(newfolio); 53619138349SMatthew Wilcox (Oracle) if (folio_test_clear_active(folio)) { 53719138349SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_unevictable(folio), folio); 53819138349SMatthew Wilcox (Oracle) folio_set_active(newfolio); 53919138349SMatthew Wilcox (Oracle) } else if (folio_test_clear_unevictable(folio)) 54019138349SMatthew Wilcox (Oracle) folio_set_unevictable(newfolio); 54119138349SMatthew Wilcox (Oracle) if (folio_test_workingset(folio)) 54219138349SMatthew Wilcox (Oracle) folio_set_workingset(newfolio); 54319138349SMatthew Wilcox (Oracle) if (folio_test_checked(folio)) 54419138349SMatthew Wilcox (Oracle) folio_set_checked(newfolio); 5456c287605SDavid Hildenbrand /* 5466c287605SDavid Hildenbrand * PG_anon_exclusive (-> PG_mappedtodisk) is always migrated via 5476c287605SDavid Hildenbrand * migration entries. We can still have PG_anon_exclusive set on an 5486c287605SDavid Hildenbrand * effectively unmapped and unreferenced first sub-pages of an 5496c287605SDavid Hildenbrand * anonymous THP: we can simply copy it here via PG_mappedtodisk. 5506c287605SDavid Hildenbrand */ 55119138349SMatthew Wilcox (Oracle) if (folio_test_mappedtodisk(folio)) 55219138349SMatthew Wilcox (Oracle) folio_set_mappedtodisk(newfolio); 553b20a3503SChristoph Lameter 5543417013eSMatthew Wilcox (Oracle) /* Move dirty on pages not done by folio_migrate_mapping() */ 55519138349SMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) 55619138349SMatthew Wilcox (Oracle) folio_set_dirty(newfolio); 557b20a3503SChristoph Lameter 55819138349SMatthew Wilcox (Oracle) if (folio_test_young(folio)) 55919138349SMatthew Wilcox (Oracle) folio_set_young(newfolio); 56019138349SMatthew Wilcox (Oracle) if (folio_test_idle(folio)) 56119138349SMatthew Wilcox (Oracle) folio_set_idle(newfolio); 56233c3fc71SVladimir Davydov 5637851a45cSRik van Riel /* 5647851a45cSRik van Riel * Copy NUMA information to the new page, to prevent over-eager 5657851a45cSRik van Riel * future migrations of this same page. 5667851a45cSRik van Riel */ 56719138349SMatthew Wilcox (Oracle) cpupid = page_cpupid_xchg_last(&folio->page, -1); 56833024536SHuang Ying /* 56933024536SHuang Ying * For memory tiering mode, when migrate between slow and fast 57033024536SHuang Ying * memory node, reset cpupid, because that is used to record 57133024536SHuang Ying * page access time in slow memory node. 57233024536SHuang Ying */ 57333024536SHuang Ying if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) { 57433024536SHuang Ying bool f_toptier = node_is_toptier(page_to_nid(&folio->page)); 57533024536SHuang Ying bool t_toptier = node_is_toptier(page_to_nid(&newfolio->page)); 57633024536SHuang Ying 57733024536SHuang Ying if (f_toptier != t_toptier) 57833024536SHuang Ying cpupid = -1; 57933024536SHuang Ying } 58019138349SMatthew Wilcox (Oracle) page_cpupid_xchg_last(&newfolio->page, cpupid); 5817851a45cSRik van Riel 58219138349SMatthew Wilcox (Oracle) folio_migrate_ksm(newfolio, folio); 583c8d6553bSHugh Dickins /* 584c8d6553bSHugh Dickins * Please do not reorder this without considering how mm/ksm.c's 585c8d6553bSHugh Dickins * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache(). 586c8d6553bSHugh Dickins */ 58719138349SMatthew Wilcox (Oracle) if (folio_test_swapcache(folio)) 58819138349SMatthew Wilcox (Oracle) folio_clear_swapcache(folio); 58919138349SMatthew Wilcox (Oracle) folio_clear_private(folio); 590ad2fa371SMuchun Song 591ad2fa371SMuchun Song /* page->private contains hugetlb specific flags */ 59219138349SMatthew Wilcox (Oracle) if (!folio_test_hugetlb(folio)) 59319138349SMatthew Wilcox (Oracle) folio->private = NULL; 594b20a3503SChristoph Lameter 595b20a3503SChristoph Lameter /* 596b20a3503SChristoph Lameter * If any waiters have accumulated on the new page then 597b20a3503SChristoph Lameter * wake them up. 598b20a3503SChristoph Lameter */ 59919138349SMatthew Wilcox (Oracle) if (folio_test_writeback(newfolio)) 60019138349SMatthew Wilcox (Oracle) folio_end_writeback(newfolio); 601d435edcaSVlastimil Babka 6026aeff241SYang Shi /* 6036aeff241SYang Shi * PG_readahead shares the same bit with PG_reclaim. The above 6046aeff241SYang Shi * end_page_writeback() may clear PG_readahead mistakenly, so set the 6056aeff241SYang Shi * bit after that. 6066aeff241SYang Shi */ 60719138349SMatthew Wilcox (Oracle) if (folio_test_readahead(folio)) 60819138349SMatthew Wilcox (Oracle) folio_set_readahead(newfolio); 6096aeff241SYang Shi 61019138349SMatthew Wilcox (Oracle) folio_copy_owner(newfolio, folio); 61174485cf2SJohannes Weiner 61219138349SMatthew Wilcox (Oracle) if (!folio_test_hugetlb(folio)) 613d21bba2bSMatthew Wilcox (Oracle) mem_cgroup_migrate(folio, newfolio); 614b20a3503SChristoph Lameter } 61519138349SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_migrate_flags); 6162916ecc0SJérôme Glisse 617715cbfd6SMatthew Wilcox (Oracle) void folio_migrate_copy(struct folio *newfolio, struct folio *folio) 6182916ecc0SJérôme Glisse { 619715cbfd6SMatthew Wilcox (Oracle) folio_copy(newfolio, folio); 620715cbfd6SMatthew Wilcox (Oracle) folio_migrate_flags(newfolio, folio); 6212916ecc0SJérôme Glisse } 622715cbfd6SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_migrate_copy); 623b20a3503SChristoph Lameter 6241d8b85ccSChristoph Lameter /************************************************************ 6251d8b85ccSChristoph Lameter * Migration functions 6261d8b85ccSChristoph Lameter ***********************************************************/ 6271d8b85ccSChristoph Lameter 62854184650SMatthew Wilcox (Oracle) /** 62954184650SMatthew Wilcox (Oracle) * migrate_folio() - Simple folio migration. 63054184650SMatthew Wilcox (Oracle) * @mapping: The address_space containing the folio. 63154184650SMatthew Wilcox (Oracle) * @dst: The folio to migrate the data to. 63254184650SMatthew Wilcox (Oracle) * @src: The folio containing the current data. 63354184650SMatthew Wilcox (Oracle) * @mode: How to migrate the page. 634b20a3503SChristoph Lameter * 63554184650SMatthew Wilcox (Oracle) * Common logic to directly migrate a single LRU folio suitable for 63654184650SMatthew Wilcox (Oracle) * folios that do not use PagePrivate/PagePrivate2. 63754184650SMatthew Wilcox (Oracle) * 63854184650SMatthew Wilcox (Oracle) * Folios are locked upon entry and exit. 639b20a3503SChristoph Lameter */ 64054184650SMatthew Wilcox (Oracle) int migrate_folio(struct address_space *mapping, struct folio *dst, 64154184650SMatthew Wilcox (Oracle) struct folio *src, enum migrate_mode mode) 642b20a3503SChristoph Lameter { 643b20a3503SChristoph Lameter int rc; 644b20a3503SChristoph Lameter 64554184650SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(src)); /* Writeback must be complete */ 646b20a3503SChristoph Lameter 64754184650SMatthew Wilcox (Oracle) rc = folio_migrate_mapping(mapping, dst, src, 0); 648b20a3503SChristoph Lameter 64978bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 650b20a3503SChristoph Lameter return rc; 651b20a3503SChristoph Lameter 6522916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 65354184650SMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 6542916ecc0SJérôme Glisse else 65554184650SMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 65678bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 657b20a3503SChristoph Lameter } 65854184650SMatthew Wilcox (Oracle) EXPORT_SYMBOL(migrate_folio); 659b20a3503SChristoph Lameter 6609361401eSDavid Howells #ifdef CONFIG_BLOCK 66184ade7c1SJan Kara /* Returns true if all buffers are successfully locked */ 66284ade7c1SJan Kara static bool buffer_migrate_lock_buffers(struct buffer_head *head, 66384ade7c1SJan Kara enum migrate_mode mode) 66484ade7c1SJan Kara { 66584ade7c1SJan Kara struct buffer_head *bh = head; 66684ade7c1SJan Kara 66784ade7c1SJan Kara /* Simple case, sync compaction */ 66884ade7c1SJan Kara if (mode != MIGRATE_ASYNC) { 66984ade7c1SJan Kara do { 67084ade7c1SJan Kara lock_buffer(bh); 67184ade7c1SJan Kara bh = bh->b_this_page; 67284ade7c1SJan Kara 67384ade7c1SJan Kara } while (bh != head); 67484ade7c1SJan Kara 67584ade7c1SJan Kara return true; 67684ade7c1SJan Kara } 67784ade7c1SJan Kara 67884ade7c1SJan Kara /* async case, we cannot block on lock_buffer so use trylock_buffer */ 67984ade7c1SJan Kara do { 68084ade7c1SJan Kara if (!trylock_buffer(bh)) { 68184ade7c1SJan Kara /* 68284ade7c1SJan Kara * We failed to lock the buffer and cannot stall in 68384ade7c1SJan Kara * async migration. Release the taken locks 68484ade7c1SJan Kara */ 68584ade7c1SJan Kara struct buffer_head *failed_bh = bh; 68684ade7c1SJan Kara bh = head; 68784ade7c1SJan Kara while (bh != failed_bh) { 68884ade7c1SJan Kara unlock_buffer(bh); 68984ade7c1SJan Kara bh = bh->b_this_page; 69084ade7c1SJan Kara } 69184ade7c1SJan Kara return false; 69284ade7c1SJan Kara } 69384ade7c1SJan Kara 69484ade7c1SJan Kara bh = bh->b_this_page; 69584ade7c1SJan Kara } while (bh != head); 69684ade7c1SJan Kara return true; 69784ade7c1SJan Kara } 69884ade7c1SJan Kara 69967235182SMatthew Wilcox (Oracle) static int __buffer_migrate_folio(struct address_space *mapping, 70067235182SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode, 70189cb0888SJan Kara bool check_refs) 7021d8b85ccSChristoph Lameter { 7031d8b85ccSChristoph Lameter struct buffer_head *bh, *head; 7041d8b85ccSChristoph Lameter int rc; 705cc4f11e6SJan Kara int expected_count; 7061d8b85ccSChristoph Lameter 70767235182SMatthew Wilcox (Oracle) head = folio_buffers(src); 70867235182SMatthew Wilcox (Oracle) if (!head) 70954184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 7101d8b85ccSChristoph Lameter 711cc4f11e6SJan Kara /* Check whether page does not have extra refs before we do more work */ 712108ca835SMatthew Wilcox (Oracle) expected_count = folio_expected_refs(mapping, src); 71367235182SMatthew Wilcox (Oracle) if (folio_ref_count(src) != expected_count) 714cc4f11e6SJan Kara return -EAGAIN; 715cc4f11e6SJan Kara 716cc4f11e6SJan Kara if (!buffer_migrate_lock_buffers(head, mode)) 717cc4f11e6SJan Kara return -EAGAIN; 7181d8b85ccSChristoph Lameter 71989cb0888SJan Kara if (check_refs) { 72089cb0888SJan Kara bool busy; 72189cb0888SJan Kara bool invalidated = false; 72289cb0888SJan Kara 72389cb0888SJan Kara recheck_buffers: 72489cb0888SJan Kara busy = false; 72589cb0888SJan Kara spin_lock(&mapping->private_lock); 72689cb0888SJan Kara bh = head; 72789cb0888SJan Kara do { 72889cb0888SJan Kara if (atomic_read(&bh->b_count)) { 72989cb0888SJan Kara busy = true; 73089cb0888SJan Kara break; 73189cb0888SJan Kara } 73289cb0888SJan Kara bh = bh->b_this_page; 73389cb0888SJan Kara } while (bh != head); 73489cb0888SJan Kara if (busy) { 73589cb0888SJan Kara if (invalidated) { 73689cb0888SJan Kara rc = -EAGAIN; 73789cb0888SJan Kara goto unlock_buffers; 73889cb0888SJan Kara } 739ebdf4de5SJan Kara spin_unlock(&mapping->private_lock); 74089cb0888SJan Kara invalidate_bh_lrus(); 74189cb0888SJan Kara invalidated = true; 74289cb0888SJan Kara goto recheck_buffers; 74389cb0888SJan Kara } 74489cb0888SJan Kara } 74589cb0888SJan Kara 74667235182SMatthew Wilcox (Oracle) rc = folio_migrate_mapping(mapping, dst, src, 0); 74778bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 748cc4f11e6SJan Kara goto unlock_buffers; 7491d8b85ccSChristoph Lameter 75067235182SMatthew Wilcox (Oracle) folio_attach_private(dst, folio_detach_private(src)); 7511d8b85ccSChristoph Lameter 7521d8b85ccSChristoph Lameter bh = head; 7531d8b85ccSChristoph Lameter do { 75467235182SMatthew Wilcox (Oracle) set_bh_page(bh, &dst->page, bh_offset(bh)); 7551d8b85ccSChristoph Lameter bh = bh->b_this_page; 7561d8b85ccSChristoph Lameter } while (bh != head); 7571d8b85ccSChristoph Lameter 7582916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 75967235182SMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 7602916ecc0SJérôme Glisse else 76167235182SMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 7621d8b85ccSChristoph Lameter 763cc4f11e6SJan Kara rc = MIGRATEPAGE_SUCCESS; 764cc4f11e6SJan Kara unlock_buffers: 765ebdf4de5SJan Kara if (check_refs) 766ebdf4de5SJan Kara spin_unlock(&mapping->private_lock); 7671d8b85ccSChristoph Lameter bh = head; 7681d8b85ccSChristoph Lameter do { 7691d8b85ccSChristoph Lameter unlock_buffer(bh); 7701d8b85ccSChristoph Lameter bh = bh->b_this_page; 7711d8b85ccSChristoph Lameter } while (bh != head); 7721d8b85ccSChristoph Lameter 773cc4f11e6SJan Kara return rc; 7741d8b85ccSChristoph Lameter } 77589cb0888SJan Kara 77667235182SMatthew Wilcox (Oracle) /** 77767235182SMatthew Wilcox (Oracle) * buffer_migrate_folio() - Migration function for folios with buffers. 77867235182SMatthew Wilcox (Oracle) * @mapping: The address space containing @src. 77967235182SMatthew Wilcox (Oracle) * @dst: The folio to migrate to. 78067235182SMatthew Wilcox (Oracle) * @src: The folio to migrate from. 78167235182SMatthew Wilcox (Oracle) * @mode: How to migrate the folio. 78267235182SMatthew Wilcox (Oracle) * 78367235182SMatthew Wilcox (Oracle) * This function can only be used if the underlying filesystem guarantees 78467235182SMatthew Wilcox (Oracle) * that no other references to @src exist. For example attached buffer 78567235182SMatthew Wilcox (Oracle) * heads are accessed only under the folio lock. If your filesystem cannot 78667235182SMatthew Wilcox (Oracle) * provide this guarantee, buffer_migrate_folio_norefs() may be more 78767235182SMatthew Wilcox (Oracle) * appropriate. 78867235182SMatthew Wilcox (Oracle) * 78967235182SMatthew Wilcox (Oracle) * Return: 0 on success or a negative errno on failure. 79089cb0888SJan Kara */ 79167235182SMatthew Wilcox (Oracle) int buffer_migrate_folio(struct address_space *mapping, 79267235182SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 79389cb0888SJan Kara { 79467235182SMatthew Wilcox (Oracle) return __buffer_migrate_folio(mapping, dst, src, mode, false); 79589cb0888SJan Kara } 79667235182SMatthew Wilcox (Oracle) EXPORT_SYMBOL(buffer_migrate_folio); 79789cb0888SJan Kara 79867235182SMatthew Wilcox (Oracle) /** 79967235182SMatthew Wilcox (Oracle) * buffer_migrate_folio_norefs() - Migration function for folios with buffers. 80067235182SMatthew Wilcox (Oracle) * @mapping: The address space containing @src. 80167235182SMatthew Wilcox (Oracle) * @dst: The folio to migrate to. 80267235182SMatthew Wilcox (Oracle) * @src: The folio to migrate from. 80367235182SMatthew Wilcox (Oracle) * @mode: How to migrate the folio. 80467235182SMatthew Wilcox (Oracle) * 80567235182SMatthew Wilcox (Oracle) * Like buffer_migrate_folio() except that this variant is more careful 80667235182SMatthew Wilcox (Oracle) * and checks that there are also no buffer head references. This function 80767235182SMatthew Wilcox (Oracle) * is the right one for mappings where buffer heads are directly looked 80867235182SMatthew Wilcox (Oracle) * up and referenced (such as block device mappings). 80967235182SMatthew Wilcox (Oracle) * 81067235182SMatthew Wilcox (Oracle) * Return: 0 on success or a negative errno on failure. 81189cb0888SJan Kara */ 81267235182SMatthew Wilcox (Oracle) int buffer_migrate_folio_norefs(struct address_space *mapping, 81367235182SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 81489cb0888SJan Kara { 81567235182SMatthew Wilcox (Oracle) return __buffer_migrate_folio(mapping, dst, src, mode, true); 81689cb0888SJan Kara } 8179361401eSDavid Howells #endif 8181d8b85ccSChristoph Lameter 8192ec810d5SMatthew Wilcox (Oracle) int filemap_migrate_folio(struct address_space *mapping, 8202ec810d5SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 8212ec810d5SMatthew Wilcox (Oracle) { 8222ec810d5SMatthew Wilcox (Oracle) int ret; 8232ec810d5SMatthew Wilcox (Oracle) 8242ec810d5SMatthew Wilcox (Oracle) ret = folio_migrate_mapping(mapping, dst, src, 0); 8252ec810d5SMatthew Wilcox (Oracle) if (ret != MIGRATEPAGE_SUCCESS) 8262ec810d5SMatthew Wilcox (Oracle) return ret; 8272ec810d5SMatthew Wilcox (Oracle) 8282ec810d5SMatthew Wilcox (Oracle) if (folio_get_private(src)) 8292ec810d5SMatthew Wilcox (Oracle) folio_attach_private(dst, folio_detach_private(src)); 8302ec810d5SMatthew Wilcox (Oracle) 8312ec810d5SMatthew Wilcox (Oracle) if (mode != MIGRATE_SYNC_NO_COPY) 8322ec810d5SMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 8332ec810d5SMatthew Wilcox (Oracle) else 8342ec810d5SMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 8352ec810d5SMatthew Wilcox (Oracle) return MIGRATEPAGE_SUCCESS; 8362ec810d5SMatthew Wilcox (Oracle) } 8372ec810d5SMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(filemap_migrate_folio); 8382ec810d5SMatthew Wilcox (Oracle) 83904e62a29SChristoph Lameter /* 8402be7fa10SMatthew Wilcox (Oracle) * Writeback a folio to clean the dirty state 84104e62a29SChristoph Lameter */ 8422be7fa10SMatthew Wilcox (Oracle) static int writeout(struct address_space *mapping, struct folio *folio) 84304e62a29SChristoph Lameter { 84404e62a29SChristoph Lameter struct writeback_control wbc = { 84504e62a29SChristoph Lameter .sync_mode = WB_SYNC_NONE, 84604e62a29SChristoph Lameter .nr_to_write = 1, 84704e62a29SChristoph Lameter .range_start = 0, 84804e62a29SChristoph Lameter .range_end = LLONG_MAX, 84904e62a29SChristoph Lameter .for_reclaim = 1 85004e62a29SChristoph Lameter }; 85104e62a29SChristoph Lameter int rc; 85204e62a29SChristoph Lameter 85304e62a29SChristoph Lameter if (!mapping->a_ops->writepage) 85404e62a29SChristoph Lameter /* No write method for the address space */ 85504e62a29SChristoph Lameter return -EINVAL; 85604e62a29SChristoph Lameter 8572be7fa10SMatthew Wilcox (Oracle) if (!folio_clear_dirty_for_io(folio)) 85804e62a29SChristoph Lameter /* Someone else already triggered a write */ 85904e62a29SChristoph Lameter return -EAGAIN; 86004e62a29SChristoph Lameter 86104e62a29SChristoph Lameter /* 8622be7fa10SMatthew Wilcox (Oracle) * A dirty folio may imply that the underlying filesystem has 8632be7fa10SMatthew Wilcox (Oracle) * the folio on some queue. So the folio must be clean for 8642be7fa10SMatthew Wilcox (Oracle) * migration. Writeout may mean we lose the lock and the 8652be7fa10SMatthew Wilcox (Oracle) * folio state is no longer what we checked for earlier. 86604e62a29SChristoph Lameter * At this point we know that the migration attempt cannot 86704e62a29SChristoph Lameter * be successful. 86804e62a29SChristoph Lameter */ 8694eecb8b9SMatthew Wilcox (Oracle) remove_migration_ptes(folio, folio, false); 87004e62a29SChristoph Lameter 8712be7fa10SMatthew Wilcox (Oracle) rc = mapping->a_ops->writepage(&folio->page, &wbc); 87204e62a29SChristoph Lameter 87304e62a29SChristoph Lameter if (rc != AOP_WRITEPAGE_ACTIVATE) 87404e62a29SChristoph Lameter /* unlocked. Relock */ 8752be7fa10SMatthew Wilcox (Oracle) folio_lock(folio); 87604e62a29SChristoph Lameter 877bda8550dSHugh Dickins return (rc < 0) ? -EIO : -EAGAIN; 87804e62a29SChristoph Lameter } 87904e62a29SChristoph Lameter 88004e62a29SChristoph Lameter /* 88104e62a29SChristoph Lameter * Default handling if a filesystem does not provide a migration function. 88204e62a29SChristoph Lameter */ 8838faa8ef5SMatthew Wilcox (Oracle) static int fallback_migrate_folio(struct address_space *mapping, 8848faa8ef5SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 8858351a6e4SChristoph Lameter { 8868faa8ef5SMatthew Wilcox (Oracle) if (folio_test_dirty(src)) { 8878faa8ef5SMatthew Wilcox (Oracle) /* Only writeback folios in full synchronous migration */ 8882916ecc0SJérôme Glisse switch (mode) { 8892916ecc0SJérôme Glisse case MIGRATE_SYNC: 8902916ecc0SJérôme Glisse case MIGRATE_SYNC_NO_COPY: 8912916ecc0SJérôme Glisse break; 8922916ecc0SJérôme Glisse default: 893b969c4abSMel Gorman return -EBUSY; 8942916ecc0SJérôme Glisse } 8952be7fa10SMatthew Wilcox (Oracle) return writeout(mapping, src); 896b969c4abSMel Gorman } 8978351a6e4SChristoph Lameter 8988351a6e4SChristoph Lameter /* 8998351a6e4SChristoph Lameter * Buffers may be managed in a filesystem specific way. 9008351a6e4SChristoph Lameter * We must have no buffers or drop them. 9018351a6e4SChristoph Lameter */ 9028faa8ef5SMatthew Wilcox (Oracle) if (folio_test_private(src) && 9038faa8ef5SMatthew Wilcox (Oracle) !filemap_release_folio(src, GFP_KERNEL)) 904806031bbSMel Gorman return mode == MIGRATE_SYNC ? -EAGAIN : -EBUSY; 9058351a6e4SChristoph Lameter 90654184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 9078351a6e4SChristoph Lameter } 9088351a6e4SChristoph Lameter 9091d8b85ccSChristoph Lameter /* 910e24f0b8fSChristoph Lameter * Move a page to a newly allocated page 911e24f0b8fSChristoph Lameter * The page is locked and all ptes have been successfully removed. 912b20a3503SChristoph Lameter * 913e24f0b8fSChristoph Lameter * The new page will have replaced the old page if this function 914e24f0b8fSChristoph Lameter * is successful. 915894bc310SLee Schermerhorn * 916894bc310SLee Schermerhorn * Return value: 917894bc310SLee Schermerhorn * < 0 - error code 91878bd5209SRafael Aquini * MIGRATEPAGE_SUCCESS - success 919b20a3503SChristoph Lameter */ 920e7e3ffebSMatthew Wilcox (Oracle) static int move_to_new_folio(struct folio *dst, struct folio *src, 9215c3f9a67SHugh Dickins enum migrate_mode mode) 922b20a3503SChristoph Lameter { 923bda807d4SMinchan Kim int rc = -EAGAIN; 924e7e3ffebSMatthew Wilcox (Oracle) bool is_lru = !__PageMovable(&src->page); 925b20a3503SChristoph Lameter 926e7e3ffebSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_locked(src), src); 927e7e3ffebSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_locked(dst), dst); 928b20a3503SChristoph Lameter 929bda807d4SMinchan Kim if (likely(is_lru)) { 93068f2736aSMatthew Wilcox (Oracle) struct address_space *mapping = folio_mapping(src); 93168f2736aSMatthew Wilcox (Oracle) 932b20a3503SChristoph Lameter if (!mapping) 93354184650SMatthew Wilcox (Oracle) rc = migrate_folio(mapping, dst, src, mode); 9345490da4fSMatthew Wilcox (Oracle) else if (mapping->a_ops->migrate_folio) 935b20a3503SChristoph Lameter /* 9365490da4fSMatthew Wilcox (Oracle) * Most folios have a mapping and most filesystems 9375490da4fSMatthew Wilcox (Oracle) * provide a migrate_folio callback. Anonymous folios 938bda807d4SMinchan Kim * are part of swap space which also has its own 9395490da4fSMatthew Wilcox (Oracle) * migrate_folio callback. This is the most common path 940bda807d4SMinchan Kim * for page migration. 941b20a3503SChristoph Lameter */ 9425490da4fSMatthew Wilcox (Oracle) rc = mapping->a_ops->migrate_folio(mapping, dst, src, 9435490da4fSMatthew Wilcox (Oracle) mode); 9448351a6e4SChristoph Lameter else 9458faa8ef5SMatthew Wilcox (Oracle) rc = fallback_migrate_folio(mapping, dst, src, mode); 946bda807d4SMinchan Kim } else { 94768f2736aSMatthew Wilcox (Oracle) const struct movable_operations *mops; 94868f2736aSMatthew Wilcox (Oracle) 949bda807d4SMinchan Kim /* 950bda807d4SMinchan Kim * In case of non-lru page, it could be released after 951bda807d4SMinchan Kim * isolation step. In that case, we shouldn't try migration. 952bda807d4SMinchan Kim */ 953e7e3ffebSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); 954e7e3ffebSMatthew Wilcox (Oracle) if (!folio_test_movable(src)) { 955bda807d4SMinchan Kim rc = MIGRATEPAGE_SUCCESS; 956e7e3ffebSMatthew Wilcox (Oracle) folio_clear_isolated(src); 957bda807d4SMinchan Kim goto out; 958bda807d4SMinchan Kim } 959bda807d4SMinchan Kim 96068f2736aSMatthew Wilcox (Oracle) mops = page_movable_ops(&src->page); 96168f2736aSMatthew Wilcox (Oracle) rc = mops->migrate_page(&dst->page, &src->page, mode); 962bda807d4SMinchan Kim WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && 963e7e3ffebSMatthew Wilcox (Oracle) !folio_test_isolated(src)); 964bda807d4SMinchan Kim } 965b20a3503SChristoph Lameter 9665c3f9a67SHugh Dickins /* 967e7e3ffebSMatthew Wilcox (Oracle) * When successful, old pagecache src->mapping must be cleared before 968e7e3ffebSMatthew Wilcox (Oracle) * src is freed; but stats require that PageAnon be left as PageAnon. 9695c3f9a67SHugh Dickins */ 9705c3f9a67SHugh Dickins if (rc == MIGRATEPAGE_SUCCESS) { 971e7e3ffebSMatthew Wilcox (Oracle) if (__PageMovable(&src->page)) { 972e7e3ffebSMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); 973bda807d4SMinchan Kim 974bda807d4SMinchan Kim /* 975bda807d4SMinchan Kim * We clear PG_movable under page_lock so any compactor 976bda807d4SMinchan Kim * cannot try to migrate this page. 977bda807d4SMinchan Kim */ 978e7e3ffebSMatthew Wilcox (Oracle) folio_clear_isolated(src); 979bda807d4SMinchan Kim } 980bda807d4SMinchan Kim 981bda807d4SMinchan Kim /* 982e7e3ffebSMatthew Wilcox (Oracle) * Anonymous and movable src->mapping will be cleared by 983bda807d4SMinchan Kim * free_pages_prepare so don't reset it here for keeping 984bda807d4SMinchan Kim * the type to work PageAnon, for example. 985bda807d4SMinchan Kim */ 986e7e3ffebSMatthew Wilcox (Oracle) if (!folio_mapping_flags(src)) 987e7e3ffebSMatthew Wilcox (Oracle) src->mapping = NULL; 988d2b2c6ddSLars Persson 989e7e3ffebSMatthew Wilcox (Oracle) if (likely(!folio_is_zone_device(dst))) 990e7e3ffebSMatthew Wilcox (Oracle) flush_dcache_folio(dst); 9913fe2011fSMel Gorman } 992bda807d4SMinchan Kim out: 993e24f0b8fSChristoph Lameter return rc; 994e24f0b8fSChristoph Lameter } 995e24f0b8fSChristoph Lameter 996682a71a1SMatthew Wilcox (Oracle) static int __unmap_and_move(struct folio *src, struct folio *dst, 9979c620e2bSHugh Dickins int force, enum migrate_mode mode) 998e24f0b8fSChristoph Lameter { 9990dabec93SMinchan Kim int rc = -EAGAIN; 1000213ecb31SBaolin Wang bool page_was_mapped = false; 10013f6c8272SMel Gorman struct anon_vma *anon_vma = NULL; 1002682a71a1SMatthew Wilcox (Oracle) bool is_lru = !__PageMovable(&src->page); 100395a402c3SChristoph Lameter 1004682a71a1SMatthew Wilcox (Oracle) if (!folio_trylock(src)) { 1005a6bc32b8SMel Gorman if (!force || mode == MIGRATE_ASYNC) 10060dabec93SMinchan Kim goto out; 10073e7d3449SMel Gorman 10083e7d3449SMel Gorman /* 10093e7d3449SMel Gorman * It's not safe for direct compaction to call lock_page. 10103e7d3449SMel Gorman * For example, during page readahead pages are added locked 10113e7d3449SMel Gorman * to the LRU. Later, when the IO completes the pages are 10123e7d3449SMel Gorman * marked uptodate and unlocked. However, the queueing 10133e7d3449SMel Gorman * could be merging multiple pages for one bio (e.g. 1014d4388340SMatthew Wilcox (Oracle) * mpage_readahead). If an allocation happens for the 10153e7d3449SMel Gorman * second or third page, the process can end up locking 10163e7d3449SMel Gorman * the same page twice and deadlocking. Rather than 10173e7d3449SMel Gorman * trying to be clever about what pages can be locked, 10183e7d3449SMel Gorman * avoid the use of lock_page for direct compaction 10193e7d3449SMel Gorman * altogether. 10203e7d3449SMel Gorman */ 10213e7d3449SMel Gorman if (current->flags & PF_MEMALLOC) 10220dabec93SMinchan Kim goto out; 10233e7d3449SMel Gorman 1024682a71a1SMatthew Wilcox (Oracle) folio_lock(src); 1025e24f0b8fSChristoph Lameter } 1026e24f0b8fSChristoph Lameter 1027682a71a1SMatthew Wilcox (Oracle) if (folio_test_writeback(src)) { 102811bc82d6SAndrea Arcangeli /* 1029fed5b64aSJianguo Wu * Only in the case of a full synchronous migration is it 1030a6bc32b8SMel Gorman * necessary to wait for PageWriteback. In the async case, 1031a6bc32b8SMel Gorman * the retry loop is too short and in the sync-light case, 1032a6bc32b8SMel Gorman * the overhead of stalling is too much 103311bc82d6SAndrea Arcangeli */ 10342916ecc0SJérôme Glisse switch (mode) { 10352916ecc0SJérôme Glisse case MIGRATE_SYNC: 10362916ecc0SJérôme Glisse case MIGRATE_SYNC_NO_COPY: 10372916ecc0SJérôme Glisse break; 10382916ecc0SJérôme Glisse default: 103911bc82d6SAndrea Arcangeli rc = -EBUSY; 10400a31bc97SJohannes Weiner goto out_unlock; 104111bc82d6SAndrea Arcangeli } 104211bc82d6SAndrea Arcangeli if (!force) 10430a31bc97SJohannes Weiner goto out_unlock; 1044682a71a1SMatthew Wilcox (Oracle) folio_wait_writeback(src); 1045e24f0b8fSChristoph Lameter } 104603f15c86SHugh Dickins 1047e24f0b8fSChristoph Lameter /* 1048682a71a1SMatthew Wilcox (Oracle) * By try_to_migrate(), src->mapcount goes down to 0 here. In this case, 1049682a71a1SMatthew Wilcox (Oracle) * we cannot notice that anon_vma is freed while we migrate a page. 10501ce82b69SHugh Dickins * This get_anon_vma() delays freeing anon_vma pointer until the end 1051dc386d4dSKAMEZAWA Hiroyuki * of migration. File cache pages are no problem because of page_lock() 1052989f89c5SKAMEZAWA Hiroyuki * File Caches may use write_page() or lock_page() in migration, then, 1053989f89c5SKAMEZAWA Hiroyuki * just care Anon page here. 10543fe2011fSMel Gorman * 105503f15c86SHugh Dickins * Only page_get_anon_vma() understands the subtleties of 105603f15c86SHugh Dickins * getting a hold on an anon_vma from outside one of its mms. 105703f15c86SHugh Dickins * But if we cannot get anon_vma, then we won't need it anyway, 105803f15c86SHugh Dickins * because that implies that the anon page is no longer mapped 105903f15c86SHugh Dickins * (and cannot be remapped so long as we hold the page lock). 10603fe2011fSMel Gorman */ 1061682a71a1SMatthew Wilcox (Oracle) if (folio_test_anon(src) && !folio_test_ksm(src)) 1062682a71a1SMatthew Wilcox (Oracle) anon_vma = page_get_anon_vma(&src->page); 106362e1c553SShaohua Li 10647db7671fSHugh Dickins /* 10657db7671fSHugh Dickins * Block others from accessing the new page when we get around to 10667db7671fSHugh Dickins * establishing additional references. We are usually the only one 1067682a71a1SMatthew Wilcox (Oracle) * holding a reference to dst at this point. We used to have a BUG 1068682a71a1SMatthew Wilcox (Oracle) * here if folio_trylock(dst) fails, but would like to allow for 1069682a71a1SMatthew Wilcox (Oracle) * cases where there might be a race with the previous use of dst. 10707db7671fSHugh Dickins * This is much like races on refcount of oldpage: just don't BUG(). 10717db7671fSHugh Dickins */ 1072682a71a1SMatthew Wilcox (Oracle) if (unlikely(!folio_trylock(dst))) 10737db7671fSHugh Dickins goto out_unlock; 10747db7671fSHugh Dickins 1075bda807d4SMinchan Kim if (unlikely(!is_lru)) { 1076682a71a1SMatthew Wilcox (Oracle) rc = move_to_new_folio(dst, src, mode); 1077bda807d4SMinchan Kim goto out_unlock_both; 1078bda807d4SMinchan Kim } 1079bda807d4SMinchan Kim 1080dc386d4dSKAMEZAWA Hiroyuki /* 108162e1c553SShaohua Li * Corner case handling: 108262e1c553SShaohua Li * 1. When a new swap-cache page is read into, it is added to the LRU 108362e1c553SShaohua Li * and treated as swapcache but it has no rmap yet. 1084682a71a1SMatthew Wilcox (Oracle) * Calling try_to_unmap() against a src->mapping==NULL page will 108562e1c553SShaohua Li * trigger a BUG. So handle it here. 1086d12b8951SYang Shi * 2. An orphaned page (see truncate_cleanup_page) might have 108762e1c553SShaohua Li * fs-private metadata. The page can be picked up due to memory 108862e1c553SShaohua Li * offlining. Everywhere else except page reclaim, the page is 108962e1c553SShaohua Li * invisible to the vm, so the page can not be migrated. So try to 109062e1c553SShaohua Li * free the metadata, so the page can be freed. 1091dc386d4dSKAMEZAWA Hiroyuki */ 1092682a71a1SMatthew Wilcox (Oracle) if (!src->mapping) { 1093682a71a1SMatthew Wilcox (Oracle) if (folio_test_private(src)) { 1094682a71a1SMatthew Wilcox (Oracle) try_to_free_buffers(src); 10957db7671fSHugh Dickins goto out_unlock_both; 109662e1c553SShaohua Li } 1097682a71a1SMatthew Wilcox (Oracle) } else if (folio_mapped(src)) { 10987db7671fSHugh Dickins /* Establish migration ptes */ 1099682a71a1SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_anon(src) && 1100682a71a1SMatthew Wilcox (Oracle) !folio_test_ksm(src) && !anon_vma, src); 1101682a71a1SMatthew Wilcox (Oracle) try_to_migrate(src, 0); 1102213ecb31SBaolin Wang page_was_mapped = true; 11032ebba6b7SHugh Dickins } 1104dc386d4dSKAMEZAWA Hiroyuki 1105682a71a1SMatthew Wilcox (Oracle) if (!folio_mapped(src)) 1106682a71a1SMatthew Wilcox (Oracle) rc = move_to_new_folio(dst, src, mode); 1107e24f0b8fSChristoph Lameter 1108c3096e67SHugh Dickins /* 1109682a71a1SMatthew Wilcox (Oracle) * When successful, push dst to LRU immediately: so that if it 1110c3096e67SHugh Dickins * turns out to be an mlocked page, remove_migration_ptes() will 1111682a71a1SMatthew Wilcox (Oracle) * automatically build up the correct dst->mlock_count for it. 1112c3096e67SHugh Dickins * 1113c3096e67SHugh Dickins * We would like to do something similar for the old page, when 1114c3096e67SHugh Dickins * unsuccessful, and other cases when a page has been temporarily 1115c3096e67SHugh Dickins * isolated from the unevictable LRU: but this case is the easiest. 1116c3096e67SHugh Dickins */ 1117c3096e67SHugh Dickins if (rc == MIGRATEPAGE_SUCCESS) { 1118682a71a1SMatthew Wilcox (Oracle) folio_add_lru(dst); 11195c3f9a67SHugh Dickins if (page_was_mapped) 1120c3096e67SHugh Dickins lru_add_drain(); 1121c3096e67SHugh Dickins } 1122c3096e67SHugh Dickins 11235c3f9a67SHugh Dickins if (page_was_mapped) 1124682a71a1SMatthew Wilcox (Oracle) remove_migration_ptes(src, 1125682a71a1SMatthew Wilcox (Oracle) rc == MIGRATEPAGE_SUCCESS ? dst : src, false); 11263f6c8272SMel Gorman 11277db7671fSHugh Dickins out_unlock_both: 1128682a71a1SMatthew Wilcox (Oracle) folio_unlock(dst); 11297db7671fSHugh Dickins out_unlock: 11303f6c8272SMel Gorman /* Drop an anon_vma reference if we took one */ 113176545066SRik van Riel if (anon_vma) 11329e60109fSPeter Zijlstra put_anon_vma(anon_vma); 1133682a71a1SMatthew Wilcox (Oracle) folio_unlock(src); 11340dabec93SMinchan Kim out: 1135c6c919ebSMinchan Kim /* 1136682a71a1SMatthew Wilcox (Oracle) * If migration is successful, decrease refcount of dst, 1137c6c919ebSMinchan Kim * which will not free the page because new page owner increased 1138c3096e67SHugh Dickins * refcounter. 1139c6c919ebSMinchan Kim */ 1140c3096e67SHugh Dickins if (rc == MIGRATEPAGE_SUCCESS) 1141682a71a1SMatthew Wilcox (Oracle) folio_put(dst); 1142c6c919ebSMinchan Kim 11430dabec93SMinchan Kim return rc; 11440dabec93SMinchan Kim } 114595a402c3SChristoph Lameter 11460dabec93SMinchan Kim /* 11470dabec93SMinchan Kim * Obtain the lock on page, remove all ptes and migrate the page 11480dabec93SMinchan Kim * to the newly allocated page in newpage. 11490dabec93SMinchan Kim */ 11506ec4476aSLinus Torvalds static int unmap_and_move(new_page_t get_new_page, 1151ef2a5153SGeert Uytterhoeven free_page_t put_new_page, 1152ef2a5153SGeert Uytterhoeven unsigned long private, struct page *page, 1153add05cecSNaoya Horiguchi int force, enum migrate_mode mode, 1154dd4ae78aSYang Shi enum migrate_reason reason, 1155dd4ae78aSYang Shi struct list_head *ret) 11560dabec93SMinchan Kim { 1157682a71a1SMatthew Wilcox (Oracle) struct folio *dst, *src = page_folio(page); 11582def7424SHugh Dickins int rc = MIGRATEPAGE_SUCCESS; 115974d4a579SYang Shi struct page *newpage = NULL; 11600dabec93SMinchan Kim 116194723aafSMichal Hocko if (!thp_migration_supported() && PageTransHuge(page)) 1162d532e2e5SYang Shi return -ENOSYS; 116394723aafSMichal Hocko 11640dabec93SMinchan Kim if (page_count(page) == 1) { 1165160088b3SMiaohe Lin /* Page was freed from under us. So we are done. */ 1166c6c919ebSMinchan Kim ClearPageActive(page); 1167c6c919ebSMinchan Kim ClearPageUnevictable(page); 1168160088b3SMiaohe Lin /* free_pages_prepare() will clear PG_isolated. */ 11690dabec93SMinchan Kim goto out; 11700dabec93SMinchan Kim } 11710dabec93SMinchan Kim 117274d4a579SYang Shi newpage = get_new_page(page, private); 117374d4a579SYang Shi if (!newpage) 117474d4a579SYang Shi return -ENOMEM; 1175682a71a1SMatthew Wilcox (Oracle) dst = page_folio(newpage); 117674d4a579SYang Shi 1177b653db77SMatthew Wilcox (Oracle) newpage->private = 0; 1178682a71a1SMatthew Wilcox (Oracle) rc = __unmap_and_move(src, dst, force, mode); 1179c6c919ebSMinchan Kim if (rc == MIGRATEPAGE_SUCCESS) 11807cd12b4aSVlastimil Babka set_page_owner_migrate_reason(newpage, reason); 1181bf6bddf1SRafael Aquini 11820dabec93SMinchan Kim out: 1183e24f0b8fSChristoph Lameter if (rc != -EAGAIN) { 1184aaa994b3SChristoph Lameter /* 1185aaa994b3SChristoph Lameter * A page that has been migrated has all references 1186aaa994b3SChristoph Lameter * removed and will be freed. A page that has not been 1187c23a0c99SRalph Campbell * migrated will have kept its references and be restored. 1188aaa994b3SChristoph Lameter */ 1189aaa994b3SChristoph Lameter list_del(&page->lru); 1190e24f0b8fSChristoph Lameter } 119168711a74SDavid Rientjes 119295a402c3SChristoph Lameter /* 1193c6c919ebSMinchan Kim * If migration is successful, releases reference grabbed during 1194c6c919ebSMinchan Kim * isolation. Otherwise, restore the page to right list unless 1195c6c919ebSMinchan Kim * we want to retry. 119695a402c3SChristoph Lameter */ 1197c6c919ebSMinchan Kim if (rc == MIGRATEPAGE_SUCCESS) { 1198dd4ae78aSYang Shi /* 1199dd4ae78aSYang Shi * Compaction can migrate also non-LRU pages which are 1200dd4ae78aSYang Shi * not accounted to NR_ISOLATED_*. They can be recognized 1201dd4ae78aSYang Shi * as __PageMovable 1202dd4ae78aSYang Shi */ 1203dd4ae78aSYang Shi if (likely(!__PageMovable(page))) 1204dd4ae78aSYang Shi mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + 1205dd4ae78aSYang Shi page_is_file_lru(page), -thp_nr_pages(page)); 1206dd4ae78aSYang Shi 120779f5f8faSOscar Salvador if (reason != MR_MEMORY_FAILURE) 1208c6c919ebSMinchan Kim /* 120979f5f8faSOscar Salvador * We release the page in page_handle_poison. 1210c6c919ebSMinchan Kim */ 121179f5f8faSOscar Salvador put_page(page); 1212c6c919ebSMinchan Kim } else { 1213dd4ae78aSYang Shi if (rc != -EAGAIN) 1214dd4ae78aSYang Shi list_add_tail(&page->lru, ret); 1215bda807d4SMinchan Kim 1216cf4b769aSHugh Dickins if (put_new_page) 121768711a74SDavid Rientjes put_new_page(newpage, private); 1218c6c919ebSMinchan Kim else 1219d6d86c0aSKonstantin Khlebnikov put_page(newpage); 1220c6c919ebSMinchan Kim } 122168711a74SDavid Rientjes 1222e24f0b8fSChristoph Lameter return rc; 1223e24f0b8fSChristoph Lameter } 1224b20a3503SChristoph Lameter 1225e24f0b8fSChristoph Lameter /* 1226290408d4SNaoya Horiguchi * Counterpart of unmap_and_move_page() for hugepage migration. 1227290408d4SNaoya Horiguchi * 1228290408d4SNaoya Horiguchi * This function doesn't wait the completion of hugepage I/O 1229290408d4SNaoya Horiguchi * because there is no race between I/O and migration for hugepage. 1230290408d4SNaoya Horiguchi * Note that currently hugepage I/O occurs only in direct I/O 1231290408d4SNaoya Horiguchi * where no lock is held and PG_writeback is irrelevant, 1232290408d4SNaoya Horiguchi * and writeback status of all subpages are counted in the reference 1233290408d4SNaoya Horiguchi * count of the head page (i.e. if all subpages of a 2MB hugepage are 1234290408d4SNaoya Horiguchi * under direct I/O, the reference of the head page is 512 and a bit more.) 1235290408d4SNaoya Horiguchi * This means that when we try to migrate hugepage whose subpages are 1236290408d4SNaoya Horiguchi * doing direct I/O, some references remain after try_to_unmap() and 1237290408d4SNaoya Horiguchi * hugepage migration fails without data corruption. 1238290408d4SNaoya Horiguchi * 1239290408d4SNaoya Horiguchi * There is also no race when direct I/O is issued on the page under migration, 1240290408d4SNaoya Horiguchi * because then pte is replaced with migration swap entry and direct I/O code 1241290408d4SNaoya Horiguchi * will wait in the page fault for migration to complete. 1242290408d4SNaoya Horiguchi */ 1243290408d4SNaoya Horiguchi static int unmap_and_move_huge_page(new_page_t get_new_page, 124468711a74SDavid Rientjes free_page_t put_new_page, unsigned long private, 124568711a74SDavid Rientjes struct page *hpage, int force, 1246dd4ae78aSYang Shi enum migrate_mode mode, int reason, 1247dd4ae78aSYang Shi struct list_head *ret) 1248290408d4SNaoya Horiguchi { 12494eecb8b9SMatthew Wilcox (Oracle) struct folio *dst, *src = page_folio(hpage); 12502def7424SHugh Dickins int rc = -EAGAIN; 12512ebba6b7SHugh Dickins int page_was_mapped = 0; 125232665f2bSJoonsoo Kim struct page *new_hpage; 1253290408d4SNaoya Horiguchi struct anon_vma *anon_vma = NULL; 1254c0d0381aSMike Kravetz struct address_space *mapping = NULL; 1255290408d4SNaoya Horiguchi 125683467efbSNaoya Horiguchi /* 12577ed2c31dSAnshuman Khandual * Migratability of hugepages depends on architectures and their size. 125883467efbSNaoya Horiguchi * This check is necessary because some callers of hugepage migration 125983467efbSNaoya Horiguchi * like soft offline and memory hotremove don't walk through page 126083467efbSNaoya Horiguchi * tables or check whether the hugepage is pmd-based or not before 126183467efbSNaoya Horiguchi * kicking migration. 126283467efbSNaoya Horiguchi */ 1263577be05cSHuang Ying if (!hugepage_migration_supported(page_hstate(hpage))) 126483467efbSNaoya Horiguchi return -ENOSYS; 126583467efbSNaoya Horiguchi 1266*c33db292SMatthew Wilcox (Oracle) if (folio_ref_count(src) == 1) { 126771a64f61SMuchun Song /* page was freed from under us. So we are done. */ 126871a64f61SMuchun Song putback_active_hugepage(hpage); 126971a64f61SMuchun Song return MIGRATEPAGE_SUCCESS; 127071a64f61SMuchun Song } 127171a64f61SMuchun Song 1272666feb21SMichal Hocko new_hpage = get_new_page(hpage, private); 1273290408d4SNaoya Horiguchi if (!new_hpage) 1274290408d4SNaoya Horiguchi return -ENOMEM; 12754eecb8b9SMatthew Wilcox (Oracle) dst = page_folio(new_hpage); 1276290408d4SNaoya Horiguchi 1277*c33db292SMatthew Wilcox (Oracle) if (!folio_trylock(src)) { 12782916ecc0SJérôme Glisse if (!force) 1279290408d4SNaoya Horiguchi goto out; 12802916ecc0SJérôme Glisse switch (mode) { 12812916ecc0SJérôme Glisse case MIGRATE_SYNC: 12822916ecc0SJérôme Glisse case MIGRATE_SYNC_NO_COPY: 12832916ecc0SJérôme Glisse break; 12842916ecc0SJérôme Glisse default: 12852916ecc0SJérôme Glisse goto out; 12862916ecc0SJérôme Glisse } 1287*c33db292SMatthew Wilcox (Oracle) folio_lock(src); 1288290408d4SNaoya Horiguchi } 1289290408d4SNaoya Horiguchi 1290cb6acd01SMike Kravetz /* 1291cb6acd01SMike Kravetz * Check for pages which are in the process of being freed. Without 1292*c33db292SMatthew Wilcox (Oracle) * folio_mapping() set, hugetlbfs specific move page routine will not 1293cb6acd01SMike Kravetz * be called and we could leak usage counts for subpools. 1294cb6acd01SMike Kravetz */ 1295*c33db292SMatthew Wilcox (Oracle) if (hugetlb_page_subpool(hpage) && !folio_mapping(src)) { 1296cb6acd01SMike Kravetz rc = -EBUSY; 1297cb6acd01SMike Kravetz goto out_unlock; 1298cb6acd01SMike Kravetz } 1299cb6acd01SMike Kravetz 1300*c33db292SMatthew Wilcox (Oracle) if (folio_test_anon(src)) 1301*c33db292SMatthew Wilcox (Oracle) anon_vma = page_get_anon_vma(&src->page); 1302290408d4SNaoya Horiguchi 1303*c33db292SMatthew Wilcox (Oracle) if (unlikely(!folio_trylock(dst))) 13047db7671fSHugh Dickins goto put_anon; 13057db7671fSHugh Dickins 1306*c33db292SMatthew Wilcox (Oracle) if (folio_mapped(src)) { 1307a98a2f0cSAlistair Popple enum ttu_flags ttu = 0; 1308336bf30eSMike Kravetz 1309*c33db292SMatthew Wilcox (Oracle) if (!folio_test_anon(src)) { 1310c0d0381aSMike Kravetz /* 1311336bf30eSMike Kravetz * In shared mappings, try_to_unmap could potentially 1312336bf30eSMike Kravetz * call huge_pmd_unshare. Because of this, take 1313336bf30eSMike Kravetz * semaphore in write mode here and set TTU_RMAP_LOCKED 1314336bf30eSMike Kravetz * to let lower levels know we have taken the lock. 1315c0d0381aSMike Kravetz */ 1316c0d0381aSMike Kravetz mapping = hugetlb_page_mapping_lock_write(hpage); 1317c0d0381aSMike Kravetz if (unlikely(!mapping)) 1318c0d0381aSMike Kravetz goto unlock_put_anon; 1319c0d0381aSMike Kravetz 13205202978bSMiaohe Lin ttu = TTU_RMAP_LOCKED; 1321336bf30eSMike Kravetz } 1322336bf30eSMike Kravetz 13234b8554c5SMatthew Wilcox (Oracle) try_to_migrate(src, ttu); 13242ebba6b7SHugh Dickins page_was_mapped = 1; 1325336bf30eSMike Kravetz 13265202978bSMiaohe Lin if (ttu & TTU_RMAP_LOCKED) 1327336bf30eSMike Kravetz i_mmap_unlock_write(mapping); 13282ebba6b7SHugh Dickins } 1329290408d4SNaoya Horiguchi 1330*c33db292SMatthew Wilcox (Oracle) if (!folio_mapped(src)) 1331e7e3ffebSMatthew Wilcox (Oracle) rc = move_to_new_folio(dst, src, mode); 1332290408d4SNaoya Horiguchi 1333336bf30eSMike Kravetz if (page_was_mapped) 13344eecb8b9SMatthew Wilcox (Oracle) remove_migration_ptes(src, 13354eecb8b9SMatthew Wilcox (Oracle) rc == MIGRATEPAGE_SUCCESS ? dst : src, false); 1336290408d4SNaoya Horiguchi 1337c0d0381aSMike Kravetz unlock_put_anon: 1338*c33db292SMatthew Wilcox (Oracle) folio_unlock(dst); 13397db7671fSHugh Dickins 13407db7671fSHugh Dickins put_anon: 1341fd4a4663SHugh Dickins if (anon_vma) 13429e60109fSPeter Zijlstra put_anon_vma(anon_vma); 13438e6ac7faSAneesh Kumar K.V 13442def7424SHugh Dickins if (rc == MIGRATEPAGE_SUCCESS) { 1345ab5ac90aSMichal Hocko move_hugetlb_state(hpage, new_hpage, reason); 13462def7424SHugh Dickins put_new_page = NULL; 13472def7424SHugh Dickins } 13488e6ac7faSAneesh Kumar K.V 1349cb6acd01SMike Kravetz out_unlock: 1350*c33db292SMatthew Wilcox (Oracle) folio_unlock(src); 135109761333SHillf Danton out: 1352dd4ae78aSYang Shi if (rc == MIGRATEPAGE_SUCCESS) 1353b8ec1ceeSNaoya Horiguchi putback_active_hugepage(hpage); 1354a04840c6SMiaohe Lin else if (rc != -EAGAIN) 1355*c33db292SMatthew Wilcox (Oracle) list_move_tail(&src->lru, ret); 135668711a74SDavid Rientjes 135768711a74SDavid Rientjes /* 135868711a74SDavid Rientjes * If migration was not successful and there's a freeing callback, use 135968711a74SDavid Rientjes * it. Otherwise, put_page() will drop the reference grabbed during 136068711a74SDavid Rientjes * isolation. 136168711a74SDavid Rientjes */ 13622def7424SHugh Dickins if (put_new_page) 136368711a74SDavid Rientjes put_new_page(new_hpage, private); 136468711a74SDavid Rientjes else 13653aaa76e1SNaoya Horiguchi putback_active_hugepage(new_hpage); 136668711a74SDavid Rientjes 1367290408d4SNaoya Horiguchi return rc; 1368290408d4SNaoya Horiguchi } 1369290408d4SNaoya Horiguchi 13709c62ff00SHuang Ying static inline int try_split_thp(struct page *page, struct list_head *split_pages) 1371d532e2e5SYang Shi { 13729c62ff00SHuang Ying int rc; 1373d532e2e5SYang Shi 1374d532e2e5SYang Shi lock_page(page); 13759c62ff00SHuang Ying rc = split_huge_page_to_list(page, split_pages); 1376d532e2e5SYang Shi unlock_page(page); 1377e6fa8a79SHuang Ying if (!rc) 1378e6fa8a79SHuang Ying list_move_tail(&page->lru, split_pages); 1379d532e2e5SYang Shi 1380d532e2e5SYang Shi return rc; 1381d532e2e5SYang Shi } 1382d532e2e5SYang Shi 1383290408d4SNaoya Horiguchi /* 1384c73e5c9cSSrivatsa S. Bhat * migrate_pages - migrate the pages specified in a list, to the free pages 1385c73e5c9cSSrivatsa S. Bhat * supplied as the target for the page migration 1386e24f0b8fSChristoph Lameter * 1387c73e5c9cSSrivatsa S. Bhat * @from: The list of pages to be migrated. 1388c73e5c9cSSrivatsa S. Bhat * @get_new_page: The function used to allocate free pages to be used 1389c73e5c9cSSrivatsa S. Bhat * as the target of the page migration. 139068711a74SDavid Rientjes * @put_new_page: The function used to free target pages if migration 139168711a74SDavid Rientjes * fails, or NULL if no special handling is necessary. 1392c73e5c9cSSrivatsa S. Bhat * @private: Private data to be passed on to get_new_page() 1393c73e5c9cSSrivatsa S. Bhat * @mode: The migration mode that specifies the constraints for 1394c73e5c9cSSrivatsa S. Bhat * page migration, if any. 1395c73e5c9cSSrivatsa S. Bhat * @reason: The reason for page migration. 1396b5bade97SBaolin Wang * @ret_succeeded: Set to the number of normal pages migrated successfully if 13975ac95884SYang Shi * the caller passes a non-NULL pointer. 1398e24f0b8fSChristoph Lameter * 1399c73e5c9cSSrivatsa S. Bhat * The function returns after 10 attempts or if no pages are movable any more 1400c73e5c9cSSrivatsa S. Bhat * because the list has become empty or no retryable pages exist any more. 1401dd4ae78aSYang Shi * It is caller's responsibility to call putback_movable_pages() to return pages 1402dd4ae78aSYang Shi * to the LRU or free list only if ret != 0. 1403e24f0b8fSChristoph Lameter * 14045d39a7ebSBaolin Wang * Returns the number of {normal page, THP, hugetlb} that were not migrated, or 14055d39a7ebSBaolin Wang * an error code. The number of THP splits will be considered as the number of 14065d39a7ebSBaolin Wang * non-migrated THP, no matter how many subpages of the THP are migrated successfully. 1407e24f0b8fSChristoph Lameter */ 14089c620e2bSHugh Dickins int migrate_pages(struct list_head *from, new_page_t get_new_page, 140968711a74SDavid Rientjes free_page_t put_new_page, unsigned long private, 14105ac95884SYang Shi enum migrate_mode mode, int reason, unsigned int *ret_succeeded) 1411e24f0b8fSChristoph Lameter { 1412e24f0b8fSChristoph Lameter int retry = 1; 14131a5bae25SAnshuman Khandual int thp_retry = 1; 1414e24f0b8fSChristoph Lameter int nr_failed = 0; 1415b5bade97SBaolin Wang int nr_failed_pages = 0; 1416077309bcSHuang Ying int nr_retry_pages = 0; 14175647bc29SMel Gorman int nr_succeeded = 0; 14181a5bae25SAnshuman Khandual int nr_thp_succeeded = 0; 14191a5bae25SAnshuman Khandual int nr_thp_failed = 0; 14201a5bae25SAnshuman Khandual int nr_thp_split = 0; 1421e24f0b8fSChristoph Lameter int pass = 0; 14221a5bae25SAnshuman Khandual bool is_thp = false; 1423e24f0b8fSChristoph Lameter struct page *page; 1424e24f0b8fSChristoph Lameter struct page *page2; 14251a5bae25SAnshuman Khandual int rc, nr_subpages; 1426dd4ae78aSYang Shi LIST_HEAD(ret_pages); 1427b5bade97SBaolin Wang LIST_HEAD(thp_split_pages); 1428b0b515bfSYang Shi bool nosplit = (reason == MR_NUMA_MISPLACED); 1429b5bade97SBaolin Wang bool no_subpage_counting = false; 14302d1db3b1SChristoph Lameter 14317bc1aec5SLiam Mark trace_mm_migrate_pages_start(mode, reason); 14327bc1aec5SLiam Mark 1433b5bade97SBaolin Wang thp_subpage_migration: 14341a5bae25SAnshuman Khandual for (pass = 0; pass < 10 && (retry || thp_retry); pass++) { 1435e24f0b8fSChristoph Lameter retry = 0; 14361a5bae25SAnshuman Khandual thp_retry = 0; 1437077309bcSHuang Ying nr_retry_pages = 0; 1438e24f0b8fSChristoph Lameter 1439e24f0b8fSChristoph Lameter list_for_each_entry_safe(page, page2, from, lru) { 14401a5bae25SAnshuman Khandual /* 14411a5bae25SAnshuman Khandual * THP statistics is based on the source huge page. 14421a5bae25SAnshuman Khandual * Capture required information that might get lost 14431a5bae25SAnshuman Khandual * during migration. 14441a5bae25SAnshuman Khandual */ 14456c5c7b9fSZi Yan is_thp = PageTransHuge(page) && !PageHuge(page); 14465d39a7ebSBaolin Wang nr_subpages = compound_nr(page); 1447e24f0b8fSChristoph Lameter cond_resched(); 1448e24f0b8fSChristoph Lameter 144931caf665SNaoya Horiguchi if (PageHuge(page)) 145031caf665SNaoya Horiguchi rc = unmap_and_move_huge_page(get_new_page, 145168711a74SDavid Rientjes put_new_page, private, page, 1452dd4ae78aSYang Shi pass > 2, mode, reason, 1453dd4ae78aSYang Shi &ret_pages); 145431caf665SNaoya Horiguchi else 145568711a74SDavid Rientjes rc = unmap_and_move(get_new_page, put_new_page, 1456add05cecSNaoya Horiguchi private, page, pass > 2, mode, 1457dd4ae78aSYang Shi reason, &ret_pages); 1458dd4ae78aSYang Shi /* 1459dd4ae78aSYang Shi * The rules are: 1460dd4ae78aSYang Shi * Success: non hugetlb page will be freed, hugetlb 1461dd4ae78aSYang Shi * page will be put back 1462dd4ae78aSYang Shi * -EAGAIN: stay on the from list 1463dd4ae78aSYang Shi * -ENOMEM: stay on the from list 1464577be05cSHuang Ying * -ENOSYS: stay on the from list 1465dd4ae78aSYang Shi * Other errno: put on ret_pages list then splice to 1466dd4ae78aSYang Shi * from list 1467dd4ae78aSYang Shi */ 1468e24f0b8fSChristoph Lameter switch(rc) { 146994723aafSMichal Hocko /* 147094723aafSMichal Hocko * THP migration might be unsupported or the 147194723aafSMichal Hocko * allocation could've failed so we should 147294723aafSMichal Hocko * retry on the same page with the THP split 147394723aafSMichal Hocko * to base pages. 147494723aafSMichal Hocko * 1475e6fa8a79SHuang Ying * Sub-pages are put in thp_split_pages, and 1476e6fa8a79SHuang Ying * we will migrate them after the rest of the 1477e6fa8a79SHuang Ying * list is processed. 147894723aafSMichal Hocko */ 1479d532e2e5SYang Shi case -ENOSYS: 1480d532e2e5SYang Shi /* THP migration is unsupported */ 14816c5c7b9fSZi Yan if (is_thp) { 1482b5bade97SBaolin Wang nr_thp_failed++; 14839c62ff00SHuang Ying if (!try_split_thp(page, &thp_split_pages)) { 1484d532e2e5SYang Shi nr_thp_split++; 1485e6fa8a79SHuang Ying break; 1486d532e2e5SYang Shi } 1487f430893bSMiaohe Lin /* Hugetlb migration is unsupported */ 1488f430893bSMiaohe Lin } else if (!no_subpage_counting) { 1489f430893bSMiaohe Lin nr_failed++; 1490d532e2e5SYang Shi } 1491d532e2e5SYang Shi 14925d39a7ebSBaolin Wang nr_failed_pages += nr_subpages; 1493577be05cSHuang Ying list_move_tail(&page->lru, &ret_pages); 1494d532e2e5SYang Shi break; 1495d532e2e5SYang Shi case -ENOMEM: 1496d532e2e5SYang Shi /* 1497d532e2e5SYang Shi * When memory is low, don't bother to try to migrate 1498d532e2e5SYang Shi * other pages, just exit. 1499d532e2e5SYang Shi */ 1500fbed53b4SHuang Ying if (is_thp) { 1501b5bade97SBaolin Wang nr_thp_failed++; 1502fbed53b4SHuang Ying /* THP NUMA faulting doesn't split THP to retry. */ 1503fbed53b4SHuang Ying if (!nosplit && !try_split_thp(page, &thp_split_pages)) { 15041a5bae25SAnshuman Khandual nr_thp_split++; 1505e6fa8a79SHuang Ying break; 150694723aafSMichal Hocko } 1507f430893bSMiaohe Lin } else if (!no_subpage_counting) { 1508f430893bSMiaohe Lin nr_failed++; 15091a5bae25SAnshuman Khandual } 1510b5bade97SBaolin Wang 1511077309bcSHuang Ying nr_failed_pages += nr_subpages + nr_retry_pages; 151269a041ffSMiaohe Lin /* 151369a041ffSMiaohe Lin * There might be some subpages of fail-to-migrate THPs 151469a041ffSMiaohe Lin * left in thp_split_pages list. Move them back to migration 151569a041ffSMiaohe Lin * list so that they could be put back to the right list by 151669a041ffSMiaohe Lin * the caller otherwise the page refcnt will be leaked. 151769a041ffSMiaohe Lin */ 151869a041ffSMiaohe Lin list_splice_init(&thp_split_pages, from); 1519fbed53b4SHuang Ying /* nr_failed isn't updated for not used */ 152069a041ffSMiaohe Lin nr_thp_failed += thp_retry; 152195a402c3SChristoph Lameter goto out; 1522e24f0b8fSChristoph Lameter case -EAGAIN: 1523f430893bSMiaohe Lin if (is_thp) 15241a5bae25SAnshuman Khandual thp_retry++; 15257047b5a4SBaolin Wang else if (!no_subpage_counting) 1526b20a3503SChristoph Lameter retry++; 1527077309bcSHuang Ying nr_retry_pages += nr_subpages; 1528e24f0b8fSChristoph Lameter break; 152978bd5209SRafael Aquini case MIGRATEPAGE_SUCCESS: 15305d39a7ebSBaolin Wang nr_succeeded += nr_subpages; 1531f430893bSMiaohe Lin if (is_thp) 15321a5bae25SAnshuman Khandual nr_thp_succeeded++; 15331a5bae25SAnshuman Khandual break; 1534e24f0b8fSChristoph Lameter default: 1535354a3363SNaoya Horiguchi /* 1536d532e2e5SYang Shi * Permanent failure (-EBUSY, etc.): 1537354a3363SNaoya Horiguchi * unlike -EAGAIN case, the failed page is 1538354a3363SNaoya Horiguchi * removed from migration page list and not 1539354a3363SNaoya Horiguchi * retried in the next outer loop. 1540354a3363SNaoya Horiguchi */ 1541f430893bSMiaohe Lin if (is_thp) 15421a5bae25SAnshuman Khandual nr_thp_failed++; 1543f430893bSMiaohe Lin else if (!no_subpage_counting) 1544b20a3503SChristoph Lameter nr_failed++; 1545f430893bSMiaohe Lin 15465d39a7ebSBaolin Wang nr_failed_pages += nr_subpages; 1547e24f0b8fSChristoph Lameter break; 1548b20a3503SChristoph Lameter } 1549b20a3503SChristoph Lameter } 1550e24f0b8fSChristoph Lameter } 1551b5bade97SBaolin Wang nr_failed += retry; 15521a5bae25SAnshuman Khandual nr_thp_failed += thp_retry; 1553077309bcSHuang Ying nr_failed_pages += nr_retry_pages; 1554b5bade97SBaolin Wang /* 1555b5bade97SBaolin Wang * Try to migrate subpages of fail-to-migrate THPs, no nr_failed 1556b5bade97SBaolin Wang * counting in this round, since all subpages of a THP is counted 1557b5bade97SBaolin Wang * as 1 failure in the first round. 1558b5bade97SBaolin Wang */ 1559b5bade97SBaolin Wang if (!list_empty(&thp_split_pages)) { 1560b5bade97SBaolin Wang /* 1561b5bade97SBaolin Wang * Move non-migrated pages (after 10 retries) to ret_pages 1562b5bade97SBaolin Wang * to avoid migrating them again. 1563b5bade97SBaolin Wang */ 1564b5bade97SBaolin Wang list_splice_init(from, &ret_pages); 1565b5bade97SBaolin Wang list_splice_init(&thp_split_pages, from); 1566b5bade97SBaolin Wang no_subpage_counting = true; 1567b5bade97SBaolin Wang retry = 1; 1568b5bade97SBaolin Wang goto thp_subpage_migration; 1569b5bade97SBaolin Wang } 1570b5bade97SBaolin Wang 1571b5bade97SBaolin Wang rc = nr_failed + nr_thp_failed; 157295a402c3SChristoph Lameter out: 1573dd4ae78aSYang Shi /* 1574dd4ae78aSYang Shi * Put the permanent failure page back to migration list, they 1575dd4ae78aSYang Shi * will be put back to the right list by the caller. 1576dd4ae78aSYang Shi */ 1577dd4ae78aSYang Shi list_splice(&ret_pages, from); 1578dd4ae78aSYang Shi 15795647bc29SMel Gorman count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded); 1580b5bade97SBaolin Wang count_vm_events(PGMIGRATE_FAIL, nr_failed_pages); 15811a5bae25SAnshuman Khandual count_vm_events(THP_MIGRATION_SUCCESS, nr_thp_succeeded); 15821a5bae25SAnshuman Khandual count_vm_events(THP_MIGRATION_FAIL, nr_thp_failed); 15831a5bae25SAnshuman Khandual count_vm_events(THP_MIGRATION_SPLIT, nr_thp_split); 1584b5bade97SBaolin Wang trace_mm_migrate_pages(nr_succeeded, nr_failed_pages, nr_thp_succeeded, 15851a5bae25SAnshuman Khandual nr_thp_failed, nr_thp_split, mode, reason); 15867b2a2d4aSMel Gorman 15875ac95884SYang Shi if (ret_succeeded) 15885ac95884SYang Shi *ret_succeeded = nr_succeeded; 15895ac95884SYang Shi 159095a402c3SChristoph Lameter return rc; 1591b20a3503SChristoph Lameter } 1592b20a3503SChristoph Lameter 159319fc7bedSJoonsoo Kim struct page *alloc_migration_target(struct page *page, unsigned long private) 1594b4b38223SJoonsoo Kim { 1595ffe06786SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 159619fc7bedSJoonsoo Kim struct migration_target_control *mtc; 159719fc7bedSJoonsoo Kim gfp_t gfp_mask; 1598b4b38223SJoonsoo Kim unsigned int order = 0; 1599ffe06786SMatthew Wilcox (Oracle) struct folio *new_folio = NULL; 160019fc7bedSJoonsoo Kim int nid; 160119fc7bedSJoonsoo Kim int zidx; 160219fc7bedSJoonsoo Kim 160319fc7bedSJoonsoo Kim mtc = (struct migration_target_control *)private; 160419fc7bedSJoonsoo Kim gfp_mask = mtc->gfp_mask; 160519fc7bedSJoonsoo Kim nid = mtc->nid; 160619fc7bedSJoonsoo Kim if (nid == NUMA_NO_NODE) 1607ffe06786SMatthew Wilcox (Oracle) nid = folio_nid(folio); 1608b4b38223SJoonsoo Kim 1609ffe06786SMatthew Wilcox (Oracle) if (folio_test_hugetlb(folio)) { 1610ffe06786SMatthew Wilcox (Oracle) struct hstate *h = page_hstate(&folio->page); 1611d92bbc27SJoonsoo Kim 161219fc7bedSJoonsoo Kim gfp_mask = htlb_modify_alloc_mask(h, gfp_mask); 161319fc7bedSJoonsoo Kim return alloc_huge_page_nodemask(h, nid, mtc->nmask, gfp_mask); 1614d92bbc27SJoonsoo Kim } 1615b4b38223SJoonsoo Kim 1616ffe06786SMatthew Wilcox (Oracle) if (folio_test_large(folio)) { 16179933a0c8SJoonsoo Kim /* 16189933a0c8SJoonsoo Kim * clear __GFP_RECLAIM to make the migration callback 16199933a0c8SJoonsoo Kim * consistent with regular THP allocations. 16209933a0c8SJoonsoo Kim */ 16219933a0c8SJoonsoo Kim gfp_mask &= ~__GFP_RECLAIM; 1622b4b38223SJoonsoo Kim gfp_mask |= GFP_TRANSHUGE; 1623ffe06786SMatthew Wilcox (Oracle) order = folio_order(folio); 1624b4b38223SJoonsoo Kim } 1625ffe06786SMatthew Wilcox (Oracle) zidx = zone_idx(folio_zone(folio)); 162619fc7bedSJoonsoo Kim if (is_highmem_idx(zidx) || zidx == ZONE_MOVABLE) 1627b4b38223SJoonsoo Kim gfp_mask |= __GFP_HIGHMEM; 1628b4b38223SJoonsoo Kim 1629ffe06786SMatthew Wilcox (Oracle) new_folio = __folio_alloc(gfp_mask, order, nid, mtc->nmask); 1630b4b38223SJoonsoo Kim 1631ffe06786SMatthew Wilcox (Oracle) return &new_folio->page; 1632b4b38223SJoonsoo Kim } 1633b4b38223SJoonsoo Kim 1634742755a1SChristoph Lameter #ifdef CONFIG_NUMA 1635742755a1SChristoph Lameter 1636a49bd4d7SMichal Hocko static int store_status(int __user *status, int start, int value, int nr) 1637742755a1SChristoph Lameter { 1638a49bd4d7SMichal Hocko while (nr-- > 0) { 1639a49bd4d7SMichal Hocko if (put_user(value, status + start)) 1640a49bd4d7SMichal Hocko return -EFAULT; 1641a49bd4d7SMichal Hocko start++; 1642a49bd4d7SMichal Hocko } 1643742755a1SChristoph Lameter 1644a49bd4d7SMichal Hocko return 0; 1645a49bd4d7SMichal Hocko } 1646742755a1SChristoph Lameter 1647a49bd4d7SMichal Hocko static int do_move_pages_to_node(struct mm_struct *mm, 1648a49bd4d7SMichal Hocko struct list_head *pagelist, int node) 1649a49bd4d7SMichal Hocko { 1650a49bd4d7SMichal Hocko int err; 1651a0976311SJoonsoo Kim struct migration_target_control mtc = { 1652a0976311SJoonsoo Kim .nid = node, 1653a0976311SJoonsoo Kim .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 1654a0976311SJoonsoo Kim }; 1655742755a1SChristoph Lameter 1656a0976311SJoonsoo Kim err = migrate_pages(pagelist, alloc_migration_target, NULL, 16575ac95884SYang Shi (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); 1658a49bd4d7SMichal Hocko if (err) 1659a49bd4d7SMichal Hocko putback_movable_pages(pagelist); 1660a49bd4d7SMichal Hocko return err; 1661742755a1SChristoph Lameter } 1662742755a1SChristoph Lameter 1663742755a1SChristoph Lameter /* 1664a49bd4d7SMichal Hocko * Resolves the given address to a struct page, isolates it from the LRU and 1665a49bd4d7SMichal Hocko * puts it to the given pagelist. 1666e0153fc2SYang Shi * Returns: 1667e0153fc2SYang Shi * errno - if the page cannot be found/isolated 1668e0153fc2SYang Shi * 0 - when it doesn't have to be migrated because it is already on the 1669e0153fc2SYang Shi * target node 1670e0153fc2SYang Shi * 1 - when it has been queued 1671742755a1SChristoph Lameter */ 1672a49bd4d7SMichal Hocko static int add_page_for_migration(struct mm_struct *mm, unsigned long addr, 1673a49bd4d7SMichal Hocko int node, struct list_head *pagelist, bool migrate_all) 1674742755a1SChristoph Lameter { 1675742755a1SChristoph Lameter struct vm_area_struct *vma; 1676742755a1SChristoph Lameter struct page *page; 1677a49bd4d7SMichal Hocko int err; 1678742755a1SChristoph Lameter 1679d8ed45c5SMichel Lespinasse mmap_read_lock(mm); 1680742755a1SChristoph Lameter err = -EFAULT; 1681cb1c37b1SMiaohe Lin vma = vma_lookup(mm, addr); 1682cb1c37b1SMiaohe Lin if (!vma || !vma_migratable(vma)) 1683a49bd4d7SMichal Hocko goto out; 1684742755a1SChristoph Lameter 1685d899844eSKirill A. Shutemov /* FOLL_DUMP to ignore special (like zero) pages */ 168687d2762eSMiaohe Lin page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); 168789f5b7daSLinus Torvalds 168889f5b7daSLinus Torvalds err = PTR_ERR(page); 168989f5b7daSLinus Torvalds if (IS_ERR(page)) 1690a49bd4d7SMichal Hocko goto out; 169189f5b7daSLinus Torvalds 1692742755a1SChristoph Lameter err = -ENOENT; 1693f7091ed6SHaiyue Wang if (!page) 1694a49bd4d7SMichal Hocko goto out; 1695742755a1SChristoph Lameter 1696f7091ed6SHaiyue Wang if (is_zone_device_page(page)) 1697f7091ed6SHaiyue Wang goto out_putpage; 1698f7091ed6SHaiyue Wang 1699a49bd4d7SMichal Hocko err = 0; 1700a49bd4d7SMichal Hocko if (page_to_nid(page) == node) 1701a49bd4d7SMichal Hocko goto out_putpage; 1702742755a1SChristoph Lameter 1703742755a1SChristoph Lameter err = -EACCES; 1704a49bd4d7SMichal Hocko if (page_mapcount(page) > 1 && !migrate_all) 1705a49bd4d7SMichal Hocko goto out_putpage; 1706742755a1SChristoph Lameter 1707e632a938SNaoya Horiguchi if (PageHuge(page)) { 1708e8db67ebSNaoya Horiguchi if (PageHead(page)) { 17097ce82f4cSMiaohe Lin err = isolate_hugetlb(page, pagelist); 17107ce82f4cSMiaohe Lin if (!err) 1711e0153fc2SYang Shi err = 1; 1712e8db67ebSNaoya Horiguchi } 1713a49bd4d7SMichal Hocko } else { 1714a49bd4d7SMichal Hocko struct page *head; 1715e632a938SNaoya Horiguchi 1716e8db67ebSNaoya Horiguchi head = compound_head(page); 1717e8db67ebSNaoya Horiguchi err = isolate_lru_page(head); 1718a49bd4d7SMichal Hocko if (err) 1719a49bd4d7SMichal Hocko goto out_putpage; 1720a49bd4d7SMichal Hocko 1721e0153fc2SYang Shi err = 1; 1722a49bd4d7SMichal Hocko list_add_tail(&head->lru, pagelist); 1723e8db67ebSNaoya Horiguchi mod_node_page_state(page_pgdat(head), 17249de4f22aSHuang Ying NR_ISOLATED_ANON + page_is_file_lru(head), 17256c357848SMatthew Wilcox (Oracle) thp_nr_pages(head)); 17266d9c285aSKOSAKI Motohiro } 1727a49bd4d7SMichal Hocko out_putpage: 1728742755a1SChristoph Lameter /* 1729742755a1SChristoph Lameter * Either remove the duplicate refcount from 1730742755a1SChristoph Lameter * isolate_lru_page() or drop the page ref if it was 1731742755a1SChristoph Lameter * not isolated. 1732742755a1SChristoph Lameter */ 1733742755a1SChristoph Lameter put_page(page); 1734a49bd4d7SMichal Hocko out: 1735d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 1736742755a1SChristoph Lameter return err; 1737742755a1SChristoph Lameter } 1738742755a1SChristoph Lameter 17397ca8783aSWei Yang static int move_pages_and_store_status(struct mm_struct *mm, int node, 17407ca8783aSWei Yang struct list_head *pagelist, int __user *status, 17417ca8783aSWei Yang int start, int i, unsigned long nr_pages) 17427ca8783aSWei Yang { 17437ca8783aSWei Yang int err; 17447ca8783aSWei Yang 17455d7ae891SWei Yang if (list_empty(pagelist)) 17465d7ae891SWei Yang return 0; 17475d7ae891SWei Yang 17487ca8783aSWei Yang err = do_move_pages_to_node(mm, pagelist, node); 17497ca8783aSWei Yang if (err) { 17507ca8783aSWei Yang /* 17517ca8783aSWei Yang * Positive err means the number of failed 17527ca8783aSWei Yang * pages to migrate. Since we are going to 17537ca8783aSWei Yang * abort and return the number of non-migrated 1754ab9dd4f8SLong Li * pages, so need to include the rest of the 17557ca8783aSWei Yang * nr_pages that have not been attempted as 17567ca8783aSWei Yang * well. 17577ca8783aSWei Yang */ 17587ca8783aSWei Yang if (err > 0) 1759a7504ed1SHuang Ying err += nr_pages - i; 17607ca8783aSWei Yang return err; 17617ca8783aSWei Yang } 17627ca8783aSWei Yang return store_status(status, start, node, i - start); 17637ca8783aSWei Yang } 17647ca8783aSWei Yang 1765742755a1SChristoph Lameter /* 17665e9a0f02SBrice Goglin * Migrate an array of page address onto an array of nodes and fill 17675e9a0f02SBrice Goglin * the corresponding array of status. 17685e9a0f02SBrice Goglin */ 17693268c63eSChristoph Lameter static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, 17705e9a0f02SBrice Goglin unsigned long nr_pages, 17715e9a0f02SBrice Goglin const void __user * __user *pages, 17725e9a0f02SBrice Goglin const int __user *nodes, 17735e9a0f02SBrice Goglin int __user *status, int flags) 17745e9a0f02SBrice Goglin { 1775a49bd4d7SMichal Hocko int current_node = NUMA_NO_NODE; 1776a49bd4d7SMichal Hocko LIST_HEAD(pagelist); 1777a49bd4d7SMichal Hocko int start, i; 1778a49bd4d7SMichal Hocko int err = 0, err1; 177935282a2dSBrice Goglin 1780361a2a22SMinchan Kim lru_cache_disable(); 178135282a2dSBrice Goglin 1782a49bd4d7SMichal Hocko for (i = start = 0; i < nr_pages; i++) { 17835e9a0f02SBrice Goglin const void __user *p; 1784a49bd4d7SMichal Hocko unsigned long addr; 17855e9a0f02SBrice Goglin int node; 17865e9a0f02SBrice Goglin 17873140a227SBrice Goglin err = -EFAULT; 1788a49bd4d7SMichal Hocko if (get_user(p, pages + i)) 1789a49bd4d7SMichal Hocko goto out_flush; 1790a49bd4d7SMichal Hocko if (get_user(node, nodes + i)) 1791a49bd4d7SMichal Hocko goto out_flush; 1792057d3389SAndrey Konovalov addr = (unsigned long)untagged_addr(p); 17935e9a0f02SBrice Goglin 17945e9a0f02SBrice Goglin err = -ENODEV; 17956f5a55f1SLinus Torvalds if (node < 0 || node >= MAX_NUMNODES) 1796a49bd4d7SMichal Hocko goto out_flush; 1797389162c2SLai Jiangshan if (!node_state(node, N_MEMORY)) 1798a49bd4d7SMichal Hocko goto out_flush; 17995e9a0f02SBrice Goglin 18005e9a0f02SBrice Goglin err = -EACCES; 18015e9a0f02SBrice Goglin if (!node_isset(node, task_nodes)) 1802a49bd4d7SMichal Hocko goto out_flush; 18035e9a0f02SBrice Goglin 1804a49bd4d7SMichal Hocko if (current_node == NUMA_NO_NODE) { 1805a49bd4d7SMichal Hocko current_node = node; 1806a49bd4d7SMichal Hocko start = i; 1807a49bd4d7SMichal Hocko } else if (node != current_node) { 18087ca8783aSWei Yang err = move_pages_and_store_status(mm, current_node, 18097ca8783aSWei Yang &pagelist, status, start, i, nr_pages); 1810a49bd4d7SMichal Hocko if (err) 1811a49bd4d7SMichal Hocko goto out; 1812a49bd4d7SMichal Hocko start = i; 1813a49bd4d7SMichal Hocko current_node = node; 18145e9a0f02SBrice Goglin } 18155e9a0f02SBrice Goglin 1816a49bd4d7SMichal Hocko /* 1817a49bd4d7SMichal Hocko * Errors in the page lookup or isolation are not fatal and we simply 1818a49bd4d7SMichal Hocko * report them via status 1819a49bd4d7SMichal Hocko */ 1820a49bd4d7SMichal Hocko err = add_page_for_migration(mm, addr, current_node, 1821a49bd4d7SMichal Hocko &pagelist, flags & MPOL_MF_MOVE_ALL); 1822e0153fc2SYang Shi 1823d08221a0SWei Yang if (err > 0) { 1824e0153fc2SYang Shi /* The page is successfully queued for migration */ 1825e0153fc2SYang Shi continue; 1826e0153fc2SYang Shi } 18273140a227SBrice Goglin 1828d08221a0SWei Yang /* 182965462462SJohn Hubbard * The move_pages() man page does not have an -EEXIST choice, so 183065462462SJohn Hubbard * use -EFAULT instead. 183165462462SJohn Hubbard */ 183265462462SJohn Hubbard if (err == -EEXIST) 183365462462SJohn Hubbard err = -EFAULT; 183465462462SJohn Hubbard 183565462462SJohn Hubbard /* 1836d08221a0SWei Yang * If the page is already on the target node (!err), store the 1837d08221a0SWei Yang * node, otherwise, store the err. 1838d08221a0SWei Yang */ 1839d08221a0SWei Yang err = store_status(status, i, err ? : current_node, 1); 1840a49bd4d7SMichal Hocko if (err) 1841a49bd4d7SMichal Hocko goto out_flush; 18423140a227SBrice Goglin 18437ca8783aSWei Yang err = move_pages_and_store_status(mm, current_node, &pagelist, 18447ca8783aSWei Yang status, start, i, nr_pages); 1845a7504ed1SHuang Ying if (err) { 1846a7504ed1SHuang Ying /* We have accounted for page i */ 1847a7504ed1SHuang Ying if (err > 0) 1848a7504ed1SHuang Ying err--; 1849a49bd4d7SMichal Hocko goto out; 1850a7504ed1SHuang Ying } 1851a49bd4d7SMichal Hocko current_node = NUMA_NO_NODE; 18523140a227SBrice Goglin } 1853a49bd4d7SMichal Hocko out_flush: 1854a49bd4d7SMichal Hocko /* Make sure we do not overwrite the existing error */ 18557ca8783aSWei Yang err1 = move_pages_and_store_status(mm, current_node, &pagelist, 18567ca8783aSWei Yang status, start, i, nr_pages); 1857dfe9aa23SWei Yang if (err >= 0) 1858a49bd4d7SMichal Hocko err = err1; 18595e9a0f02SBrice Goglin out: 1860361a2a22SMinchan Kim lru_cache_enable(); 18615e9a0f02SBrice Goglin return err; 18625e9a0f02SBrice Goglin } 18635e9a0f02SBrice Goglin 18645e9a0f02SBrice Goglin /* 18652f007e74SBrice Goglin * Determine the nodes of an array of pages and store it in an array of status. 1866742755a1SChristoph Lameter */ 186780bba129SBrice Goglin static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, 186880bba129SBrice Goglin const void __user **pages, int *status) 1869742755a1SChristoph Lameter { 18702f007e74SBrice Goglin unsigned long i; 1871742755a1SChristoph Lameter 1872d8ed45c5SMichel Lespinasse mmap_read_lock(mm); 18732f007e74SBrice Goglin 18742f007e74SBrice Goglin for (i = 0; i < nr_pages; i++) { 187580bba129SBrice Goglin unsigned long addr = (unsigned long)(*pages); 187683156821SHaiyue Wang unsigned int foll_flags = FOLL_DUMP; 18772f007e74SBrice Goglin struct vm_area_struct *vma; 18782f007e74SBrice Goglin struct page *page; 1879c095adbcSKOSAKI Motohiro int err = -EFAULT; 18802f007e74SBrice Goglin 1881059b8b48SLiam Howlett vma = vma_lookup(mm, addr); 1882059b8b48SLiam Howlett if (!vma) 1883742755a1SChristoph Lameter goto set_status; 1884742755a1SChristoph Lameter 188583156821SHaiyue Wang /* Not all huge page follow APIs support 'FOLL_GET' */ 188683156821SHaiyue Wang if (!is_vm_hugetlb_page(vma)) 188783156821SHaiyue Wang foll_flags |= FOLL_GET; 188883156821SHaiyue Wang 1889d899844eSKirill A. Shutemov /* FOLL_DUMP to ignore special (like zero) pages */ 189083156821SHaiyue Wang page = follow_page(vma, addr, foll_flags); 189189f5b7daSLinus Torvalds 189289f5b7daSLinus Torvalds err = PTR_ERR(page); 189389f5b7daSLinus Torvalds if (IS_ERR(page)) 189489f5b7daSLinus Torvalds goto set_status; 189589f5b7daSLinus Torvalds 1896f7091ed6SHaiyue Wang err = -ENOENT; 1897f7091ed6SHaiyue Wang if (!page) 1898f7091ed6SHaiyue Wang goto set_status; 1899f7091ed6SHaiyue Wang 1900f7091ed6SHaiyue Wang if (!is_zone_device_page(page)) 19014cd61484SMiaohe Lin err = page_to_nid(page); 1902f7091ed6SHaiyue Wang 190383156821SHaiyue Wang if (foll_flags & FOLL_GET) 19044cd61484SMiaohe Lin put_page(page); 1905742755a1SChristoph Lameter set_status: 190680bba129SBrice Goglin *status = err; 190780bba129SBrice Goglin 190880bba129SBrice Goglin pages++; 190980bba129SBrice Goglin status++; 191080bba129SBrice Goglin } 191180bba129SBrice Goglin 1912d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 191380bba129SBrice Goglin } 191480bba129SBrice Goglin 19155b1b561bSArnd Bergmann static int get_compat_pages_array(const void __user *chunk_pages[], 19165b1b561bSArnd Bergmann const void __user * __user *pages, 19175b1b561bSArnd Bergmann unsigned long chunk_nr) 19185b1b561bSArnd Bergmann { 19195b1b561bSArnd Bergmann compat_uptr_t __user *pages32 = (compat_uptr_t __user *)pages; 19205b1b561bSArnd Bergmann compat_uptr_t p; 19215b1b561bSArnd Bergmann int i; 19225b1b561bSArnd Bergmann 19235b1b561bSArnd Bergmann for (i = 0; i < chunk_nr; i++) { 19245b1b561bSArnd Bergmann if (get_user(p, pages32 + i)) 19255b1b561bSArnd Bergmann return -EFAULT; 19265b1b561bSArnd Bergmann chunk_pages[i] = compat_ptr(p); 19275b1b561bSArnd Bergmann } 19285b1b561bSArnd Bergmann 19295b1b561bSArnd Bergmann return 0; 19305b1b561bSArnd Bergmann } 19315b1b561bSArnd Bergmann 193280bba129SBrice Goglin /* 193380bba129SBrice Goglin * Determine the nodes of a user array of pages and store it in 193480bba129SBrice Goglin * a user array of status. 193580bba129SBrice Goglin */ 193680bba129SBrice Goglin static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, 193780bba129SBrice Goglin const void __user * __user *pages, 193880bba129SBrice Goglin int __user *status) 193980bba129SBrice Goglin { 19403eefb826SMiaohe Lin #define DO_PAGES_STAT_CHUNK_NR 16UL 194180bba129SBrice Goglin const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; 194280bba129SBrice Goglin int chunk_status[DO_PAGES_STAT_CHUNK_NR]; 194380bba129SBrice Goglin 194487b8d1adSH. Peter Anvin while (nr_pages) { 19453eefb826SMiaohe Lin unsigned long chunk_nr = min(nr_pages, DO_PAGES_STAT_CHUNK_NR); 194687b8d1adSH. Peter Anvin 19475b1b561bSArnd Bergmann if (in_compat_syscall()) { 19485b1b561bSArnd Bergmann if (get_compat_pages_array(chunk_pages, pages, 19495b1b561bSArnd Bergmann chunk_nr)) 195087b8d1adSH. Peter Anvin break; 19515b1b561bSArnd Bergmann } else { 19525b1b561bSArnd Bergmann if (copy_from_user(chunk_pages, pages, 19535b1b561bSArnd Bergmann chunk_nr * sizeof(*chunk_pages))) 19545b1b561bSArnd Bergmann break; 19555b1b561bSArnd Bergmann } 195680bba129SBrice Goglin 195780bba129SBrice Goglin do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); 195880bba129SBrice Goglin 195987b8d1adSH. Peter Anvin if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) 196087b8d1adSH. Peter Anvin break; 1961742755a1SChristoph Lameter 196287b8d1adSH. Peter Anvin pages += chunk_nr; 196387b8d1adSH. Peter Anvin status += chunk_nr; 196487b8d1adSH. Peter Anvin nr_pages -= chunk_nr; 196587b8d1adSH. Peter Anvin } 196687b8d1adSH. Peter Anvin return nr_pages ? -EFAULT : 0; 1967742755a1SChristoph Lameter } 1968742755a1SChristoph Lameter 19694dc200ceSMiaohe Lin static struct mm_struct *find_mm_struct(pid_t pid, nodemask_t *mem_nodes) 19704dc200ceSMiaohe Lin { 19714dc200ceSMiaohe Lin struct task_struct *task; 19724dc200ceSMiaohe Lin struct mm_struct *mm; 19734dc200ceSMiaohe Lin 19744dc200ceSMiaohe Lin /* 19754dc200ceSMiaohe Lin * There is no need to check if current process has the right to modify 19764dc200ceSMiaohe Lin * the specified process when they are same. 19774dc200ceSMiaohe Lin */ 19784dc200ceSMiaohe Lin if (!pid) { 19794dc200ceSMiaohe Lin mmget(current->mm); 19804dc200ceSMiaohe Lin *mem_nodes = cpuset_mems_allowed(current); 19814dc200ceSMiaohe Lin return current->mm; 19824dc200ceSMiaohe Lin } 19834dc200ceSMiaohe Lin 19844dc200ceSMiaohe Lin /* Find the mm_struct */ 19854dc200ceSMiaohe Lin rcu_read_lock(); 19864dc200ceSMiaohe Lin task = find_task_by_vpid(pid); 19874dc200ceSMiaohe Lin if (!task) { 19884dc200ceSMiaohe Lin rcu_read_unlock(); 19894dc200ceSMiaohe Lin return ERR_PTR(-ESRCH); 19904dc200ceSMiaohe Lin } 19914dc200ceSMiaohe Lin get_task_struct(task); 19924dc200ceSMiaohe Lin 19934dc200ceSMiaohe Lin /* 19944dc200ceSMiaohe Lin * Check if this process has the right to modify the specified 19954dc200ceSMiaohe Lin * process. Use the regular "ptrace_may_access()" checks. 19964dc200ceSMiaohe Lin */ 19974dc200ceSMiaohe Lin if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { 19984dc200ceSMiaohe Lin rcu_read_unlock(); 19994dc200ceSMiaohe Lin mm = ERR_PTR(-EPERM); 20004dc200ceSMiaohe Lin goto out; 20014dc200ceSMiaohe Lin } 20024dc200ceSMiaohe Lin rcu_read_unlock(); 20034dc200ceSMiaohe Lin 20044dc200ceSMiaohe Lin mm = ERR_PTR(security_task_movememory(task)); 20054dc200ceSMiaohe Lin if (IS_ERR(mm)) 20064dc200ceSMiaohe Lin goto out; 20074dc200ceSMiaohe Lin *mem_nodes = cpuset_mems_allowed(task); 20084dc200ceSMiaohe Lin mm = get_task_mm(task); 20094dc200ceSMiaohe Lin out: 20104dc200ceSMiaohe Lin put_task_struct(task); 20114dc200ceSMiaohe Lin if (!mm) 20124dc200ceSMiaohe Lin mm = ERR_PTR(-EINVAL); 20134dc200ceSMiaohe Lin return mm; 20144dc200ceSMiaohe Lin } 20154dc200ceSMiaohe Lin 2016742755a1SChristoph Lameter /* 2017742755a1SChristoph Lameter * Move a list of pages in the address space of the currently executing 2018742755a1SChristoph Lameter * process. 2019742755a1SChristoph Lameter */ 20207addf443SDominik Brodowski static int kernel_move_pages(pid_t pid, unsigned long nr_pages, 20217addf443SDominik Brodowski const void __user * __user *pages, 20227addf443SDominik Brodowski const int __user *nodes, 20237addf443SDominik Brodowski int __user *status, int flags) 2024742755a1SChristoph Lameter { 2025742755a1SChristoph Lameter struct mm_struct *mm; 20265e9a0f02SBrice Goglin int err; 20273268c63eSChristoph Lameter nodemask_t task_nodes; 2028742755a1SChristoph Lameter 2029742755a1SChristoph Lameter /* Check flags */ 2030742755a1SChristoph Lameter if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) 2031742755a1SChristoph Lameter return -EINVAL; 2032742755a1SChristoph Lameter 2033742755a1SChristoph Lameter if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) 2034742755a1SChristoph Lameter return -EPERM; 2035742755a1SChristoph Lameter 20364dc200ceSMiaohe Lin mm = find_mm_struct(pid, &task_nodes); 20374dc200ceSMiaohe Lin if (IS_ERR(mm)) 20384dc200ceSMiaohe Lin return PTR_ERR(mm); 20396e8b09eaSSasha Levin 20403268c63eSChristoph Lameter if (nodes) 20413268c63eSChristoph Lameter err = do_pages_move(mm, task_nodes, nr_pages, pages, 20423268c63eSChristoph Lameter nodes, status, flags); 20433268c63eSChristoph Lameter else 20445e9a0f02SBrice Goglin err = do_pages_stat(mm, nr_pages, pages, status); 20453268c63eSChristoph Lameter 20463268c63eSChristoph Lameter mmput(mm); 20473268c63eSChristoph Lameter return err; 2048742755a1SChristoph Lameter } 2049742755a1SChristoph Lameter 20507addf443SDominik Brodowski SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, 20517addf443SDominik Brodowski const void __user * __user *, pages, 20527addf443SDominik Brodowski const int __user *, nodes, 20537addf443SDominik Brodowski int __user *, status, int, flags) 20547addf443SDominik Brodowski { 20557addf443SDominik Brodowski return kernel_move_pages(pid, nr_pages, pages, nodes, status, flags); 20567addf443SDominik Brodowski } 20577addf443SDominik Brodowski 20587039e1dbSPeter Zijlstra #ifdef CONFIG_NUMA_BALANCING 20597039e1dbSPeter Zijlstra /* 20607039e1dbSPeter Zijlstra * Returns true if this is a safe migration target node for misplaced NUMA 2061bc53008eSWei Yang * pages. Currently it only checks the watermarks which is crude. 20627039e1dbSPeter Zijlstra */ 20637039e1dbSPeter Zijlstra static bool migrate_balanced_pgdat(struct pglist_data *pgdat, 20643abef4e6SMel Gorman unsigned long nr_migrate_pages) 20657039e1dbSPeter Zijlstra { 20667039e1dbSPeter Zijlstra int z; 2067599d0c95SMel Gorman 20687039e1dbSPeter Zijlstra for (z = pgdat->nr_zones - 1; z >= 0; z--) { 20697039e1dbSPeter Zijlstra struct zone *zone = pgdat->node_zones + z; 20707039e1dbSPeter Zijlstra 2071bc53008eSWei Yang if (!managed_zone(zone)) 20727039e1dbSPeter Zijlstra continue; 20737039e1dbSPeter Zijlstra 20747039e1dbSPeter Zijlstra /* Avoid waking kswapd by allocating pages_to_migrate pages. */ 20757039e1dbSPeter Zijlstra if (!zone_watermark_ok(zone, 0, 20767039e1dbSPeter Zijlstra high_wmark_pages(zone) + 20777039e1dbSPeter Zijlstra nr_migrate_pages, 2078bfe9d006SHuang Ying ZONE_MOVABLE, 0)) 20797039e1dbSPeter Zijlstra continue; 20807039e1dbSPeter Zijlstra return true; 20817039e1dbSPeter Zijlstra } 20827039e1dbSPeter Zijlstra return false; 20837039e1dbSPeter Zijlstra } 20847039e1dbSPeter Zijlstra 20857039e1dbSPeter Zijlstra static struct page *alloc_misplaced_dst_page(struct page *page, 2086666feb21SMichal Hocko unsigned long data) 20877039e1dbSPeter Zijlstra { 20887039e1dbSPeter Zijlstra int nid = (int) data; 2089c185e494SMatthew Wilcox (Oracle) int order = compound_order(page); 2090c185e494SMatthew Wilcox (Oracle) gfp_t gfp = __GFP_THISNODE; 2091c185e494SMatthew Wilcox (Oracle) struct folio *new; 20927039e1dbSPeter Zijlstra 2093c185e494SMatthew Wilcox (Oracle) if (order > 0) 2094c185e494SMatthew Wilcox (Oracle) gfp |= GFP_TRANSHUGE_LIGHT; 2095c185e494SMatthew Wilcox (Oracle) else { 2096c185e494SMatthew Wilcox (Oracle) gfp |= GFP_HIGHUSER_MOVABLE | __GFP_NOMEMALLOC | __GFP_NORETRY | 2097c185e494SMatthew Wilcox (Oracle) __GFP_NOWARN; 2098c185e494SMatthew Wilcox (Oracle) gfp &= ~__GFP_RECLAIM; 20997039e1dbSPeter Zijlstra } 2100c185e494SMatthew Wilcox (Oracle) new = __folio_alloc_node(gfp, order, nid); 21017039e1dbSPeter Zijlstra 2102c185e494SMatthew Wilcox (Oracle) return &new->page; 2103c5b5a3ddSYang Shi } 2104c5b5a3ddSYang Shi 21051c30e017SMel Gorman static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page) 2106b32967ffSMel Gorman { 21072b9b624fSBaolin Wang int nr_pages = thp_nr_pages(page); 2108c574bbe9SHuang Ying int order = compound_order(page); 2109b32967ffSMel Gorman 2110c574bbe9SHuang Ying VM_BUG_ON_PAGE(order && !PageTransHuge(page), page); 21113abef4e6SMel Gorman 2112662aeea7SYang Shi /* Do not migrate THP mapped by multiple processes */ 2113662aeea7SYang Shi if (PageTransHuge(page) && total_mapcount(page) > 1) 2114662aeea7SYang Shi return 0; 2115662aeea7SYang Shi 2116b32967ffSMel Gorman /* Avoid migrating to a node that is nearly full */ 2117c574bbe9SHuang Ying if (!migrate_balanced_pgdat(pgdat, nr_pages)) { 2118c574bbe9SHuang Ying int z; 2119c574bbe9SHuang Ying 2120c574bbe9SHuang Ying if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING)) 2121340ef390SHugh Dickins return 0; 2122c574bbe9SHuang Ying for (z = pgdat->nr_zones - 1; z >= 0; z--) { 2123bc53008eSWei Yang if (managed_zone(pgdat->node_zones + z)) 2124c574bbe9SHuang Ying break; 2125c574bbe9SHuang Ying } 2126c574bbe9SHuang Ying wakeup_kswapd(pgdat->node_zones + z, 0, order, ZONE_MOVABLE); 2127c574bbe9SHuang Ying return 0; 2128c574bbe9SHuang Ying } 2129b32967ffSMel Gorman 2130340ef390SHugh Dickins if (isolate_lru_page(page)) 2131340ef390SHugh Dickins return 0; 2132340ef390SHugh Dickins 2133b75454e1SMiaohe Lin mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_is_file_lru(page), 21342b9b624fSBaolin Wang nr_pages); 2135b32967ffSMel Gorman 2136b32967ffSMel Gorman /* 2137340ef390SHugh Dickins * Isolating the page has taken another reference, so the 2138340ef390SHugh Dickins * caller's reference can be safely dropped without the page 2139340ef390SHugh Dickins * disappearing underneath us during migration. 2140b32967ffSMel Gorman */ 2141b32967ffSMel Gorman put_page(page); 2142340ef390SHugh Dickins return 1; 2143b32967ffSMel Gorman } 2144b32967ffSMel Gorman 2145a8f60772SMel Gorman /* 21467039e1dbSPeter Zijlstra * Attempt to migrate a misplaced page to the specified destination 21477039e1dbSPeter Zijlstra * node. Caller is expected to have an elevated reference count on 21487039e1dbSPeter Zijlstra * the page that will be dropped by this function before returning. 21497039e1dbSPeter Zijlstra */ 21501bc115d8SMel Gorman int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma, 21511bc115d8SMel Gorman int node) 21527039e1dbSPeter Zijlstra { 2153a8f60772SMel Gorman pg_data_t *pgdat = NODE_DATA(node); 2154340ef390SHugh Dickins int isolated; 2155b32967ffSMel Gorman int nr_remaining; 2156e39bb6beSHuang Ying unsigned int nr_succeeded; 21577039e1dbSPeter Zijlstra LIST_HEAD(migratepages); 2158b5916c02SAneesh Kumar K.V int nr_pages = thp_nr_pages(page); 2159c5b5a3ddSYang Shi 2160c5b5a3ddSYang Shi /* 21611bc115d8SMel Gorman * Don't migrate file pages that are mapped in multiple processes 21621bc115d8SMel Gorman * with execute permissions as they are probably shared libraries. 21637039e1dbSPeter Zijlstra */ 21647ee820eeSMiaohe Lin if (page_mapcount(page) != 1 && page_is_file_lru(page) && 21657ee820eeSMiaohe Lin (vma->vm_flags & VM_EXEC)) 21667039e1dbSPeter Zijlstra goto out; 21677039e1dbSPeter Zijlstra 2168a8f60772SMel Gorman /* 216909a913a7SMel Gorman * Also do not migrate dirty pages as not all filesystems can move 217009a913a7SMel Gorman * dirty pages in MIGRATE_ASYNC mode which is a waste of cycles. 217109a913a7SMel Gorman */ 21729de4f22aSHuang Ying if (page_is_file_lru(page) && PageDirty(page)) 217309a913a7SMel Gorman goto out; 217409a913a7SMel Gorman 2175b32967ffSMel Gorman isolated = numamigrate_isolate_page(pgdat, page); 2176b32967ffSMel Gorman if (!isolated) 21777039e1dbSPeter Zijlstra goto out; 21787039e1dbSPeter Zijlstra 21797039e1dbSPeter Zijlstra list_add(&page->lru, &migratepages); 2180c185e494SMatthew Wilcox (Oracle) nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page, 2181c185e494SMatthew Wilcox (Oracle) NULL, node, MIGRATE_ASYNC, 2182c185e494SMatthew Wilcox (Oracle) MR_NUMA_MISPLACED, &nr_succeeded); 21837039e1dbSPeter Zijlstra if (nr_remaining) { 218459c82b70SJoonsoo Kim if (!list_empty(&migratepages)) { 218559c82b70SJoonsoo Kim list_del(&page->lru); 2186c5fc5c3aSYang Shi mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + 2187c5fc5c3aSYang Shi page_is_file_lru(page), -nr_pages); 218859c82b70SJoonsoo Kim putback_lru_page(page); 218959c82b70SJoonsoo Kim } 21907039e1dbSPeter Zijlstra isolated = 0; 2191e39bb6beSHuang Ying } 2192e39bb6beSHuang Ying if (nr_succeeded) { 2193e39bb6beSHuang Ying count_vm_numa_events(NUMA_PAGE_MIGRATE, nr_succeeded); 2194e39bb6beSHuang Ying if (!node_is_toptier(page_to_nid(page)) && node_is_toptier(node)) 2195e39bb6beSHuang Ying mod_node_page_state(pgdat, PGPROMOTE_SUCCESS, 2196e39bb6beSHuang Ying nr_succeeded); 2197e39bb6beSHuang Ying } 21987039e1dbSPeter Zijlstra BUG_ON(!list_empty(&migratepages)); 21997039e1dbSPeter Zijlstra return isolated; 2200340ef390SHugh Dickins 2201340ef390SHugh Dickins out: 2202340ef390SHugh Dickins put_page(page); 2203340ef390SHugh Dickins return 0; 22047039e1dbSPeter Zijlstra } 2205220018d3SMel Gorman #endif /* CONFIG_NUMA_BALANCING */ 22067d6e2d96SOscar Salvador #endif /* CONFIG_NUMA */ 2207