11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/memory.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 91da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* 131da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 141da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 171da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 181da177e4SLinus Torvalds * far as I could see. 191da177e4SLinus Torvalds * 201da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 211da177e4SLinus Torvalds */ 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds /* 241da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 251da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 261da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 271da177e4SLinus Torvalds * Found it. Everything seems to work now. 281da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 331da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 341da177e4SLinus Torvalds * 351da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 361da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 371da177e4SLinus Torvalds * 381da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 391da177e4SLinus Torvalds */ 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/kernel_stat.h> 421da177e4SLinus Torvalds #include <linux/mm.h> 431da177e4SLinus Torvalds #include <linux/hugetlb.h> 441da177e4SLinus Torvalds #include <linux/mman.h> 451da177e4SLinus Torvalds #include <linux/swap.h> 461da177e4SLinus Torvalds #include <linux/highmem.h> 471da177e4SLinus Torvalds #include <linux/pagemap.h> 489a840895SHugh Dickins #include <linux/ksm.h> 491da177e4SLinus Torvalds #include <linux/rmap.h> 50b95f1b31SPaul Gortmaker #include <linux/export.h> 510ff92245SShailabh Nagar #include <linux/delayacct.h> 521da177e4SLinus Torvalds #include <linux/init.h> 53edc79b2aSPeter Zijlstra #include <linux/writeback.h> 548a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 55cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 563dc14741SHugh Dickins #include <linux/kallsyms.h> 573dc14741SHugh Dickins #include <linux/swapops.h> 583dc14741SHugh Dickins #include <linux/elf.h> 595a0e3ad6STejun Heo #include <linux/gfp.h> 604daae3b4SMel Gorman #include <linux/migrate.h> 612fbc57c5SAndy Shevchenko #include <linux/string.h> 620abdd7a8SDan Williams #include <linux/dma-debug.h> 631592eef0SKirill A. Shutemov #include <linux/debugfs.h> 646b251fc9SAndrea Arcangeli #include <linux/userfaultfd_k.h> 651da177e4SLinus Torvalds 666952b61dSAlexey Dobriyan #include <asm/io.h> 671da177e4SLinus Torvalds #include <asm/pgalloc.h> 681da177e4SLinus Torvalds #include <asm/uaccess.h> 691da177e4SLinus Torvalds #include <asm/tlb.h> 701da177e4SLinus Torvalds #include <asm/tlbflush.h> 711da177e4SLinus Torvalds #include <asm/pgtable.h> 721da177e4SLinus Torvalds 7342b77728SJan Beulich #include "internal.h" 7442b77728SJan Beulich 7590572890SPeter Zijlstra #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 7690572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 7775980e97SPeter Zijlstra #endif 7875980e97SPeter Zijlstra 79d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 801da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 811da177e4SLinus Torvalds unsigned long max_mapnr; 821da177e4SLinus Torvalds struct page *mem_map; 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 851da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 861da177e4SLinus Torvalds #endif 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* 891da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 901da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 911da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 921da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 931da177e4SLinus Torvalds * and ZONE_HIGHMEM. 941da177e4SLinus Torvalds */ 951da177e4SLinus Torvalds void * high_memory; 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 981da177e4SLinus Torvalds 9932a93233SIngo Molnar /* 10032a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 10132a93233SIngo Molnar * 10232a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 10332a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 10432a93233SIngo Molnar */ 10532a93233SIngo Molnar int randomize_va_space __read_mostly = 10632a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 10732a93233SIngo Molnar 1; 10832a93233SIngo Molnar #else 10932a93233SIngo Molnar 2; 11032a93233SIngo Molnar #endif 111a62eaf15SAndi Kleen 112a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 113a62eaf15SAndi Kleen { 114a62eaf15SAndi Kleen randomize_va_space = 0; 1159b41046cSOGAWA Hirofumi return 1; 116a62eaf15SAndi Kleen } 117a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 118a62eaf15SAndi Kleen 11962eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 12003f6462aSHugh Dickins unsigned long highest_memmap_pfn __read_mostly; 121a13ea5b7SHugh Dickins 1220b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1230b70068eSArd Biesheuvel 124a13ea5b7SHugh Dickins /* 125a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 126a13ea5b7SHugh Dickins */ 127a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 128a13ea5b7SHugh Dickins { 129a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 130a13ea5b7SHugh Dickins return 0; 131a13ea5b7SHugh Dickins } 132a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 133a62eaf15SAndi Kleen 134d559db08SKAMEZAWA Hiroyuki 13534e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 13634e55232SKAMEZAWA Hiroyuki 137ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 13834e55232SKAMEZAWA Hiroyuki { 13934e55232SKAMEZAWA Hiroyuki int i; 14034e55232SKAMEZAWA Hiroyuki 14134e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 14205af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 14305af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 14405af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 14534e55232SKAMEZAWA Hiroyuki } 14634e55232SKAMEZAWA Hiroyuki } 14705af2e10SDavid Rientjes current->rss_stat.events = 0; 14834e55232SKAMEZAWA Hiroyuki } 14934e55232SKAMEZAWA Hiroyuki 15034e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 15134e55232SKAMEZAWA Hiroyuki { 15234e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 15334e55232SKAMEZAWA Hiroyuki 15434e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 15534e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 15634e55232SKAMEZAWA Hiroyuki else 15734e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 15834e55232SKAMEZAWA Hiroyuki } 15934e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 16034e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 16134e55232SKAMEZAWA Hiroyuki 16234e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 16334e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 16434e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 16534e55232SKAMEZAWA Hiroyuki { 16634e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 16734e55232SKAMEZAWA Hiroyuki return; 16834e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 169ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 17034e55232SKAMEZAWA Hiroyuki } 1719547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 17234e55232SKAMEZAWA Hiroyuki 17334e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 17434e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 17534e55232SKAMEZAWA Hiroyuki 17634e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 17734e55232SKAMEZAWA Hiroyuki { 17834e55232SKAMEZAWA Hiroyuki } 17934e55232SKAMEZAWA Hiroyuki 1809547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 1819547d01bSPeter Zijlstra 1829547d01bSPeter Zijlstra #ifdef HAVE_GENERIC_MMU_GATHER 1839547d01bSPeter Zijlstra 184ca1d6c7dSNicholas Krause static bool tlb_next_batch(struct mmu_gather *tlb) 1859547d01bSPeter Zijlstra { 1869547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 1879547d01bSPeter Zijlstra 1889547d01bSPeter Zijlstra batch = tlb->active; 1899547d01bSPeter Zijlstra if (batch->next) { 1909547d01bSPeter Zijlstra tlb->active = batch->next; 191ca1d6c7dSNicholas Krause return true; 1929547d01bSPeter Zijlstra } 1939547d01bSPeter Zijlstra 19453a59fc6SMichal Hocko if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) 195ca1d6c7dSNicholas Krause return false; 19653a59fc6SMichal Hocko 1979547d01bSPeter Zijlstra batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 1989547d01bSPeter Zijlstra if (!batch) 199ca1d6c7dSNicholas Krause return false; 2009547d01bSPeter Zijlstra 20153a59fc6SMichal Hocko tlb->batch_count++; 2029547d01bSPeter Zijlstra batch->next = NULL; 2039547d01bSPeter Zijlstra batch->nr = 0; 2049547d01bSPeter Zijlstra batch->max = MAX_GATHER_BATCH; 2059547d01bSPeter Zijlstra 2069547d01bSPeter Zijlstra tlb->active->next = batch; 2079547d01bSPeter Zijlstra tlb->active = batch; 2089547d01bSPeter Zijlstra 209ca1d6c7dSNicholas Krause return true; 2109547d01bSPeter Zijlstra } 2119547d01bSPeter Zijlstra 2129547d01bSPeter Zijlstra /* tlb_gather_mmu 2139547d01bSPeter Zijlstra * Called to initialize an (on-stack) mmu_gather structure for page-table 2149547d01bSPeter Zijlstra * tear-down from @mm. The @fullmm argument is used when @mm is without 2159547d01bSPeter Zijlstra * users and we're going to destroy the full address space (exit/execve). 2169547d01bSPeter Zijlstra */ 2172b047252SLinus Torvalds void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) 2189547d01bSPeter Zijlstra { 2199547d01bSPeter Zijlstra tlb->mm = mm; 2209547d01bSPeter Zijlstra 2212b047252SLinus Torvalds /* Is it from 0 to ~0? */ 2222b047252SLinus Torvalds tlb->fullmm = !(start | (end+1)); 2231de14c3cSDave Hansen tlb->need_flush_all = 0; 2249547d01bSPeter Zijlstra tlb->local.next = NULL; 2259547d01bSPeter Zijlstra tlb->local.nr = 0; 2269547d01bSPeter Zijlstra tlb->local.max = ARRAY_SIZE(tlb->__pages); 2279547d01bSPeter Zijlstra tlb->active = &tlb->local; 22853a59fc6SMichal Hocko tlb->batch_count = 0; 2299547d01bSPeter Zijlstra 2309547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2319547d01bSPeter Zijlstra tlb->batch = NULL; 23234e55232SKAMEZAWA Hiroyuki #endif 233fb7332a9SWill Deacon 234fb7332a9SWill Deacon __tlb_reset_range(tlb); 2359547d01bSPeter Zijlstra } 2369547d01bSPeter Zijlstra 2371cf35d47SLinus Torvalds static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) 2389547d01bSPeter Zijlstra { 239721c21c1SWill Deacon if (!tlb->end) 240721c21c1SWill Deacon return; 241721c21c1SWill Deacon 2429547d01bSPeter Zijlstra tlb_flush(tlb); 24334ee645eSJoerg Roedel mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); 2449547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2459547d01bSPeter Zijlstra tlb_table_flush(tlb); 2469547d01bSPeter Zijlstra #endif 247fb7332a9SWill Deacon __tlb_reset_range(tlb); 2481cf35d47SLinus Torvalds } 2491cf35d47SLinus Torvalds 2501cf35d47SLinus Torvalds static void tlb_flush_mmu_free(struct mmu_gather *tlb) 2511cf35d47SLinus Torvalds { 2521cf35d47SLinus Torvalds struct mmu_gather_batch *batch; 2539547d01bSPeter Zijlstra 254721c21c1SWill Deacon for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { 2559547d01bSPeter Zijlstra free_pages_and_swap_cache(batch->pages, batch->nr); 2569547d01bSPeter Zijlstra batch->nr = 0; 2579547d01bSPeter Zijlstra } 2589547d01bSPeter Zijlstra tlb->active = &tlb->local; 2599547d01bSPeter Zijlstra } 2609547d01bSPeter Zijlstra 2611cf35d47SLinus Torvalds void tlb_flush_mmu(struct mmu_gather *tlb) 2621cf35d47SLinus Torvalds { 2631cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 2641cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 2651cf35d47SLinus Torvalds } 2661cf35d47SLinus Torvalds 2679547d01bSPeter Zijlstra /* tlb_finish_mmu 2689547d01bSPeter Zijlstra * Called at the end of the shootdown operation to free up any resources 2699547d01bSPeter Zijlstra * that were required. 2709547d01bSPeter Zijlstra */ 2719547d01bSPeter Zijlstra void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 2729547d01bSPeter Zijlstra { 2739547d01bSPeter Zijlstra struct mmu_gather_batch *batch, *next; 2749547d01bSPeter Zijlstra 2759547d01bSPeter Zijlstra tlb_flush_mmu(tlb); 2769547d01bSPeter Zijlstra 2779547d01bSPeter Zijlstra /* keep the page table cache within bounds */ 2789547d01bSPeter Zijlstra check_pgt_cache(); 2799547d01bSPeter Zijlstra 2809547d01bSPeter Zijlstra for (batch = tlb->local.next; batch; batch = next) { 2819547d01bSPeter Zijlstra next = batch->next; 2829547d01bSPeter Zijlstra free_pages((unsigned long)batch, 0); 2839547d01bSPeter Zijlstra } 2849547d01bSPeter Zijlstra tlb->local.next = NULL; 2859547d01bSPeter Zijlstra } 2869547d01bSPeter Zijlstra 2879547d01bSPeter Zijlstra /* __tlb_remove_page 2889547d01bSPeter Zijlstra * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while 2899547d01bSPeter Zijlstra * handling the additional races in SMP caused by other CPUs caching valid 2909547d01bSPeter Zijlstra * mappings in their TLBs. Returns the number of free page slots left. 2919547d01bSPeter Zijlstra * When out of page slots we must call tlb_flush_mmu(). 2929547d01bSPeter Zijlstra */ 2939547d01bSPeter Zijlstra int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 2949547d01bSPeter Zijlstra { 2959547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2969547d01bSPeter Zijlstra 297fb7332a9SWill Deacon VM_BUG_ON(!tlb->end); 2989547d01bSPeter Zijlstra 2999547d01bSPeter Zijlstra batch = tlb->active; 3009547d01bSPeter Zijlstra batch->pages[batch->nr++] = page; 3019547d01bSPeter Zijlstra if (batch->nr == batch->max) { 3029547d01bSPeter Zijlstra if (!tlb_next_batch(tlb)) 3039547d01bSPeter Zijlstra return 0; 3040b43c3aaSShaohua Li batch = tlb->active; 3059547d01bSPeter Zijlstra } 306309381feSSasha Levin VM_BUG_ON_PAGE(batch->nr > batch->max, page); 3079547d01bSPeter Zijlstra 3089547d01bSPeter Zijlstra return batch->max - batch->nr; 3099547d01bSPeter Zijlstra } 3109547d01bSPeter Zijlstra 3119547d01bSPeter Zijlstra #endif /* HAVE_GENERIC_MMU_GATHER */ 31234e55232SKAMEZAWA Hiroyuki 31326723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 31426723911SPeter Zijlstra 31526723911SPeter Zijlstra /* 31626723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 31726723911SPeter Zijlstra */ 31826723911SPeter Zijlstra 31926723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 32026723911SPeter Zijlstra { 32126723911SPeter Zijlstra /* Simply deliver the interrupt */ 32226723911SPeter Zijlstra } 32326723911SPeter Zijlstra 32426723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 32526723911SPeter Zijlstra { 32626723911SPeter Zijlstra /* 32726723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 32826723911SPeter Zijlstra * assumed to be actually RCU-freed. 32926723911SPeter Zijlstra * 33026723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 33126723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 33226723911SPeter Zijlstra */ 33326723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 33426723911SPeter Zijlstra __tlb_remove_table(table); 33526723911SPeter Zijlstra } 33626723911SPeter Zijlstra 33726723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 33826723911SPeter Zijlstra { 33926723911SPeter Zijlstra struct mmu_table_batch *batch; 34026723911SPeter Zijlstra int i; 34126723911SPeter Zijlstra 34226723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 34326723911SPeter Zijlstra 34426723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 34526723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 34626723911SPeter Zijlstra 34726723911SPeter Zijlstra free_page((unsigned long)batch); 34826723911SPeter Zijlstra } 34926723911SPeter Zijlstra 35026723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 35126723911SPeter Zijlstra { 35226723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 35326723911SPeter Zijlstra 35426723911SPeter Zijlstra if (*batch) { 35526723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 35626723911SPeter Zijlstra *batch = NULL; 35726723911SPeter Zijlstra } 35826723911SPeter Zijlstra } 35926723911SPeter Zijlstra 36026723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 36126723911SPeter Zijlstra { 36226723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 36326723911SPeter Zijlstra 36426723911SPeter Zijlstra /* 36526723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 36626723911SPeter Zijlstra * concurrent page-table walk. 36726723911SPeter Zijlstra */ 36826723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 36926723911SPeter Zijlstra __tlb_remove_table(table); 37026723911SPeter Zijlstra return; 37126723911SPeter Zijlstra } 37226723911SPeter Zijlstra 37326723911SPeter Zijlstra if (*batch == NULL) { 37426723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 37526723911SPeter Zijlstra if (*batch == NULL) { 37626723911SPeter Zijlstra tlb_remove_table_one(table); 37726723911SPeter Zijlstra return; 37826723911SPeter Zijlstra } 37926723911SPeter Zijlstra (*batch)->nr = 0; 38026723911SPeter Zijlstra } 38126723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 38226723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 38326723911SPeter Zijlstra tlb_table_flush(tlb); 38426723911SPeter Zijlstra } 38526723911SPeter Zijlstra 3869547d01bSPeter Zijlstra #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ 38726723911SPeter Zijlstra 3881da177e4SLinus Torvalds /* 3891da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 3901da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 3911da177e4SLinus Torvalds */ 3929e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 3939e1b32caSBenjamin Herrenschmidt unsigned long addr) 3941da177e4SLinus Torvalds { 3952f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 3961da177e4SLinus Torvalds pmd_clear(pmd); 3979e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 398e1f56c89SKirill A. Shutemov atomic_long_dec(&tlb->mm->nr_ptes); 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 401e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 402e0da382cSHugh Dickins unsigned long addr, unsigned long end, 403e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4041da177e4SLinus Torvalds { 4051da177e4SLinus Torvalds pmd_t *pmd; 4061da177e4SLinus Torvalds unsigned long next; 407e0da382cSHugh Dickins unsigned long start; 4081da177e4SLinus Torvalds 409e0da382cSHugh Dickins start = addr; 4101da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 4111da177e4SLinus Torvalds do { 4121da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 4131da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 4141da177e4SLinus Torvalds continue; 4159e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 4161da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 4171da177e4SLinus Torvalds 418e0da382cSHugh Dickins start &= PUD_MASK; 419e0da382cSHugh Dickins if (start < floor) 420e0da382cSHugh Dickins return; 421e0da382cSHugh Dickins if (ceiling) { 422e0da382cSHugh Dickins ceiling &= PUD_MASK; 423e0da382cSHugh Dickins if (!ceiling) 424e0da382cSHugh Dickins return; 4251da177e4SLinus Torvalds } 426e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 427e0da382cSHugh Dickins return; 428e0da382cSHugh Dickins 429e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 430e0da382cSHugh Dickins pud_clear(pud); 4319e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 432dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 436e0da382cSHugh Dickins unsigned long addr, unsigned long end, 437e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4381da177e4SLinus Torvalds { 4391da177e4SLinus Torvalds pud_t *pud; 4401da177e4SLinus Torvalds unsigned long next; 441e0da382cSHugh Dickins unsigned long start; 4421da177e4SLinus Torvalds 443e0da382cSHugh Dickins start = addr; 4441da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 4451da177e4SLinus Torvalds do { 4461da177e4SLinus Torvalds next = pud_addr_end(addr, end); 4471da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 4481da177e4SLinus Torvalds continue; 449e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 4501da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 4511da177e4SLinus Torvalds 452e0da382cSHugh Dickins start &= PGDIR_MASK; 453e0da382cSHugh Dickins if (start < floor) 454e0da382cSHugh Dickins return; 455e0da382cSHugh Dickins if (ceiling) { 456e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 457e0da382cSHugh Dickins if (!ceiling) 458e0da382cSHugh Dickins return; 4591da177e4SLinus Torvalds } 460e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 461e0da382cSHugh Dickins return; 462e0da382cSHugh Dickins 463e0da382cSHugh Dickins pud = pud_offset(pgd, start); 464e0da382cSHugh Dickins pgd_clear(pgd); 4659e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds /* 469e0da382cSHugh Dickins * This function frees user-level page tables of a process. 4701da177e4SLinus Torvalds */ 47142b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 472e0da382cSHugh Dickins unsigned long addr, unsigned long end, 473e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4741da177e4SLinus Torvalds { 4751da177e4SLinus Torvalds pgd_t *pgd; 4761da177e4SLinus Torvalds unsigned long next; 4771da177e4SLinus Torvalds 478e0da382cSHugh Dickins /* 479e0da382cSHugh Dickins * The next few lines have given us lots of grief... 480e0da382cSHugh Dickins * 481e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 482e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 483e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 484e0da382cSHugh Dickins * 485e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 486e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 487e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 488e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 489e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 490e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 491e0da382cSHugh Dickins * that end 0 case should be mythical). 492e0da382cSHugh Dickins * 493e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 494e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 495e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 496e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 497e0da382cSHugh Dickins * 498e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 499e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 500e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 501e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 502e0da382cSHugh Dickins */ 503e0da382cSHugh Dickins 504e0da382cSHugh Dickins addr &= PMD_MASK; 505e0da382cSHugh Dickins if (addr < floor) { 506e0da382cSHugh Dickins addr += PMD_SIZE; 507e0da382cSHugh Dickins if (!addr) 508e0da382cSHugh Dickins return; 509e0da382cSHugh Dickins } 510e0da382cSHugh Dickins if (ceiling) { 511e0da382cSHugh Dickins ceiling &= PMD_MASK; 512e0da382cSHugh Dickins if (!ceiling) 513e0da382cSHugh Dickins return; 514e0da382cSHugh Dickins } 515e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 516e0da382cSHugh Dickins end -= PMD_SIZE; 517e0da382cSHugh Dickins if (addr > end - 1) 518e0da382cSHugh Dickins return; 519e0da382cSHugh Dickins 52042b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 5211da177e4SLinus Torvalds do { 5221da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 5231da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 5241da177e4SLinus Torvalds continue; 52542b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 5261da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 527e0da382cSHugh Dickins } 528e0da382cSHugh Dickins 52942b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 530e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 531e0da382cSHugh Dickins { 532e0da382cSHugh Dickins while (vma) { 533e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 534e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 535e0da382cSHugh Dickins 5368f4f8c16SHugh Dickins /* 53725d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 53825d9e2d1Snpiggin@suse.de * pgtables 5398f4f8c16SHugh Dickins */ 5405beb4930SRik van Riel unlink_anon_vmas(vma); 5418f4f8c16SHugh Dickins unlink_file_vma(vma); 5428f4f8c16SHugh Dickins 5439da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 5443bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 5453bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 5463bf5ee95SHugh Dickins } else { 5473bf5ee95SHugh Dickins /* 5483bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 5493bf5ee95SHugh Dickins */ 5503bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 5514866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 552e0da382cSHugh Dickins vma = next; 553e0da382cSHugh Dickins next = vma->vm_next; 5545beb4930SRik van Riel unlink_anon_vmas(vma); 5558f4f8c16SHugh Dickins unlink_file_vma(vma); 556e0da382cSHugh Dickins } 5573bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 558e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 5593bf5ee95SHugh Dickins } 560e0da382cSHugh Dickins vma = next; 561e0da382cSHugh Dickins } 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds 5648ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 5658ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 5661da177e4SLinus Torvalds { 567c4088ebdSKirill A. Shutemov spinlock_t *ptl; 5682f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 5691da177e4SLinus Torvalds if (!new) 5701bb3630eSHugh Dickins return -ENOMEM; 5711bb3630eSHugh Dickins 572362a61adSNick Piggin /* 573362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 574362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 575362a61adSNick Piggin * put into page tables. 576362a61adSNick Piggin * 577362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 578362a61adSNick Piggin * table walking code (when walking the page table without locking; 579362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 580362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 581362a61adSNick Piggin * being the notable exception) will already guarantee loads are 582362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 583362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 584362a61adSNick Piggin */ 585362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 586362a61adSNick Piggin 587c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 5888ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 589e1f56c89SKirill A. Shutemov atomic_long_inc(&mm->nr_ptes); 5901da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 5912f569afdSMartin Schwidefsky new = NULL; 592*4b471e88SKirill A. Shutemov } 593c4088ebdSKirill A. Shutemov spin_unlock(ptl); 5942f569afdSMartin Schwidefsky if (new) 5952f569afdSMartin Schwidefsky pte_free(mm, new); 5961bb3630eSHugh Dickins return 0; 5971da177e4SLinus Torvalds } 5981da177e4SLinus Torvalds 5991bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 6001da177e4SLinus Torvalds { 6011bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 6021da177e4SLinus Torvalds if (!new) 6031bb3630eSHugh Dickins return -ENOMEM; 6041da177e4SLinus Torvalds 605362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 606362a61adSNick Piggin 607872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 6088ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 609872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 6102f569afdSMartin Schwidefsky new = NULL; 611*4b471e88SKirill A. Shutemov } 612872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 6132f569afdSMartin Schwidefsky if (new) 6142f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 6151bb3630eSHugh Dickins return 0; 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 618d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 619ae859762SHugh Dickins { 620d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 621d559db08SKAMEZAWA Hiroyuki } 622d559db08SKAMEZAWA Hiroyuki 623d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 624d559db08SKAMEZAWA Hiroyuki { 625d559db08SKAMEZAWA Hiroyuki int i; 626d559db08SKAMEZAWA Hiroyuki 62734e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 62805af2e10SDavid Rientjes sync_mm_rss(mm); 629d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 630d559db08SKAMEZAWA Hiroyuki if (rss[i]) 631d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 632ae859762SHugh Dickins } 633ae859762SHugh Dickins 6341da177e4SLinus Torvalds /* 6356aab341eSLinus Torvalds * This function is called to print an error when a bad pte 6366aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 6376aab341eSLinus Torvalds * a region that doesn't allow it. 638b5810039SNick Piggin * 639b5810039SNick Piggin * The calling function must still handle the error. 640b5810039SNick Piggin */ 6413dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 6423dc14741SHugh Dickins pte_t pte, struct page *page) 643b5810039SNick Piggin { 6443dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 6453dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 6463dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 6473dc14741SHugh Dickins struct address_space *mapping; 6483dc14741SHugh Dickins pgoff_t index; 649d936cf9bSHugh Dickins static unsigned long resume; 650d936cf9bSHugh Dickins static unsigned long nr_shown; 651d936cf9bSHugh Dickins static unsigned long nr_unshown; 652d936cf9bSHugh Dickins 653d936cf9bSHugh Dickins /* 654d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 655d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 656d936cf9bSHugh Dickins */ 657d936cf9bSHugh Dickins if (nr_shown == 60) { 658d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 659d936cf9bSHugh Dickins nr_unshown++; 660d936cf9bSHugh Dickins return; 661d936cf9bSHugh Dickins } 662d936cf9bSHugh Dickins if (nr_unshown) { 6631e9e6365SHugh Dickins printk(KERN_ALERT 6641e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 665d936cf9bSHugh Dickins nr_unshown); 666d936cf9bSHugh Dickins nr_unshown = 0; 667d936cf9bSHugh Dickins } 668d936cf9bSHugh Dickins nr_shown = 0; 669d936cf9bSHugh Dickins } 670d936cf9bSHugh Dickins if (nr_shown++ == 0) 671d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 6723dc14741SHugh Dickins 6733dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 6743dc14741SHugh Dickins index = linear_page_index(vma, addr); 6753dc14741SHugh Dickins 6761e9e6365SHugh Dickins printk(KERN_ALERT 6771e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 6783dc14741SHugh Dickins current->comm, 6793dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 680718a3821SWu Fengguang if (page) 681f0b791a3SDave Hansen dump_page(page, "bad pte"); 6821e9e6365SHugh Dickins printk(KERN_ALERT 6833dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 6843dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 6853dc14741SHugh Dickins /* 6863dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 6873dc14741SHugh Dickins */ 6882682582aSKonstantin Khlebnikov pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", 6892682582aSKonstantin Khlebnikov vma->vm_file, 6902682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 6912682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 6922682582aSKonstantin Khlebnikov mapping ? mapping->a_ops->readpage : NULL); 693b5810039SNick Piggin dump_stack(); 694373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 695b5810039SNick Piggin } 696b5810039SNick Piggin 697b5810039SNick Piggin /* 6987e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 6996aab341eSLinus Torvalds * 7007e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7017e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7027e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 703b379d790SJared Hulbert * 7047e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7057e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7067e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7077e675137SNick Piggin * described below. 7087e675137SNick Piggin * 7097e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7107e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7117e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7126aab341eSLinus Torvalds * 713b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 714b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7157e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7167e675137SNick Piggin * mapping will always honor the rule 7176aab341eSLinus Torvalds * 7186aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7196aab341eSLinus Torvalds * 7207e675137SNick Piggin * And for normal mappings this is false. 721b379d790SJared Hulbert * 7227e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7237e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7247e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7257e675137SNick Piggin * special (because none can have been COWed). 726b379d790SJared Hulbert * 727b379d790SJared Hulbert * 7287e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7297e675137SNick Piggin * 730b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 731b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 732b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 733b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 734b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 735b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 736b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 737b379d790SJared Hulbert * 738ee498ed7SHugh Dickins */ 7397e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7407e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7417e675137SNick Piggin #else 7427e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7437e675137SNick Piggin #endif 7447e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7457e675137SNick Piggin pte_t pte) 746ee498ed7SHugh Dickins { 74722b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7487e675137SNick Piggin 7497e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 750b38af472SHugh Dickins if (likely(!pte_special(pte))) 75122b31eecSHugh Dickins goto check_pfn; 752667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 753667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 754a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 755a13ea5b7SHugh Dickins return NULL; 75662eede62SHugh Dickins if (!is_zero_pfn(pfn)) 75722b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7587e675137SNick Piggin return NULL; 7597e675137SNick Piggin } 7607e675137SNick Piggin 7617e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7627e675137SNick Piggin 763b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 764b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 765b379d790SJared Hulbert if (!pfn_valid(pfn)) 766b379d790SJared Hulbert return NULL; 767b379d790SJared Hulbert goto out; 768b379d790SJared Hulbert } else { 7697e675137SNick Piggin unsigned long off; 7707e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7716aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7726aab341eSLinus Torvalds return NULL; 77367121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 774fb155c16SLinus Torvalds return NULL; 7756aab341eSLinus Torvalds } 776b379d790SJared Hulbert } 7776aab341eSLinus Torvalds 778b38af472SHugh Dickins if (is_zero_pfn(pfn)) 779b38af472SHugh Dickins return NULL; 78022b31eecSHugh Dickins check_pfn: 78122b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 78222b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 78322b31eecSHugh Dickins return NULL; 78422b31eecSHugh Dickins } 7856aab341eSLinus Torvalds 7866aab341eSLinus Torvalds /* 7877e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7887e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7896aab341eSLinus Torvalds */ 790b379d790SJared Hulbert out: 7916aab341eSLinus Torvalds return pfn_to_page(pfn); 792ee498ed7SHugh Dickins } 793ee498ed7SHugh Dickins 794ee498ed7SHugh Dickins /* 7951da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7961da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7971da177e4SLinus Torvalds * covered by this vma. 7981da177e4SLinus Torvalds */ 7991da177e4SLinus Torvalds 800570a335bSHugh Dickins static inline unsigned long 8011da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 802b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8038c103762SHugh Dickins unsigned long addr, int *rss) 8041da177e4SLinus Torvalds { 805b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8061da177e4SLinus Torvalds pte_t pte = *src_pte; 8071da177e4SLinus Torvalds struct page *page; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8101da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8110697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8120697212aSChristoph Lameter 8132022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 814570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 815570a335bSHugh Dickins return entry.val; 816570a335bSHugh Dickins 8171da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8181da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8191da177e4SLinus Torvalds spin_lock(&mmlist_lock); 820f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 821f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 822f412ac08SHugh Dickins &src_mm->mmlist); 8231da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8241da177e4SLinus Torvalds } 825b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8262022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 8279f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8289f9f1acdSKonstantin Khlebnikov 829eca56ff9SJerome Marchand rss[mm_counter(page)]++; 8309f9f1acdSKonstantin Khlebnikov 8319f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8320697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8330697212aSChristoph Lameter /* 8349f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8359f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8360697212aSChristoph Lameter */ 8370697212aSChristoph Lameter make_migration_entry_read(&entry); 8380697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 839c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 840c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 8410697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8420697212aSChristoph Lameter } 8431da177e4SLinus Torvalds } 844ae859762SHugh Dickins goto out_set_pte; 8451da177e4SLinus Torvalds } 8461da177e4SLinus Torvalds 8471da177e4SLinus Torvalds /* 8481da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8491da177e4SLinus Torvalds * in the parent and the child 8501da177e4SLinus Torvalds */ 85167121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8521da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8533dc90795SZachary Amsden pte = pte_wrprotect(pte); 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds /* 8571da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8581da177e4SLinus Torvalds * the child 8591da177e4SLinus Torvalds */ 8601da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8611da177e4SLinus Torvalds pte = pte_mkclean(pte); 8621da177e4SLinus Torvalds pte = pte_mkold(pte); 8636aab341eSLinus Torvalds 8646aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 8656aab341eSLinus Torvalds if (page) { 8661da177e4SLinus Torvalds get_page(page); 86721333b2bSHugh Dickins page_dup_rmap(page); 868eca56ff9SJerome Marchand rss[mm_counter(page)]++; 8696aab341eSLinus Torvalds } 870ae859762SHugh Dickins 871ae859762SHugh Dickins out_set_pte: 872ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 873570a335bSHugh Dickins return 0; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 87621bda264SJerome Marchand static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8771da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8781da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8791da177e4SLinus Torvalds { 880c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8811da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 882c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 883e040f218SHugh Dickins int progress = 0; 884d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 885570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 8861da177e4SLinus Torvalds 8871da177e4SLinus Torvalds again: 888d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 889d559db08SKAMEZAWA Hiroyuki 890c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 8911da177e4SLinus Torvalds if (!dst_pte) 8921da177e4SLinus Torvalds return -ENOMEM; 893ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 8944c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 895f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 896c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 897c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 8986606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds do { 9011da177e4SLinus Torvalds /* 9021da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9031da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9041da177e4SLinus Torvalds */ 905e040f218SHugh Dickins if (progress >= 32) { 906e040f218SHugh Dickins progress = 0; 907e040f218SHugh Dickins if (need_resched() || 90895c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9091da177e4SLinus Torvalds break; 910e040f218SHugh Dickins } 9111da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9121da177e4SLinus Torvalds progress++; 9131da177e4SLinus Torvalds continue; 9141da177e4SLinus Torvalds } 915570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 916570a335bSHugh Dickins vma, addr, rss); 917570a335bSHugh Dickins if (entry.val) 918570a335bSHugh Dickins break; 9191da177e4SLinus Torvalds progress += 8; 9201da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9211da177e4SLinus Torvalds 9226606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 923c74df32cSHugh Dickins spin_unlock(src_ptl); 924ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 925d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 926c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 927c74df32cSHugh Dickins cond_resched(); 928570a335bSHugh Dickins 929570a335bSHugh Dickins if (entry.val) { 930570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 931570a335bSHugh Dickins return -ENOMEM; 932570a335bSHugh Dickins progress = 0; 933570a335bSHugh Dickins } 9341da177e4SLinus Torvalds if (addr != end) 9351da177e4SLinus Torvalds goto again; 9361da177e4SLinus Torvalds return 0; 9371da177e4SLinus Torvalds } 9381da177e4SLinus Torvalds 9391da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9401da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9411da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9421da177e4SLinus Torvalds { 9431da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9441da177e4SLinus Torvalds unsigned long next; 9451da177e4SLinus Torvalds 9461da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9471da177e4SLinus Torvalds if (!dst_pmd) 9481da177e4SLinus Torvalds return -ENOMEM; 9491da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9501da177e4SLinus Torvalds do { 9511da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 95271e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 95371e3aac0SAndrea Arcangeli int err; 95414d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 95571e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 95671e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 95771e3aac0SAndrea Arcangeli if (err == -ENOMEM) 95871e3aac0SAndrea Arcangeli return -ENOMEM; 95971e3aac0SAndrea Arcangeli if (!err) 96071e3aac0SAndrea Arcangeli continue; 96171e3aac0SAndrea Arcangeli /* fall through */ 96271e3aac0SAndrea Arcangeli } 9631da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 9641da177e4SLinus Torvalds continue; 9651da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 9661da177e4SLinus Torvalds vma, addr, next)) 9671da177e4SLinus Torvalds return -ENOMEM; 9681da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 9691da177e4SLinus Torvalds return 0; 9701da177e4SLinus Torvalds } 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9731da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9741da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9751da177e4SLinus Torvalds { 9761da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9771da177e4SLinus Torvalds unsigned long next; 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9801da177e4SLinus Torvalds if (!dst_pud) 9811da177e4SLinus Torvalds return -ENOMEM; 9821da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9831da177e4SLinus Torvalds do { 9841da177e4SLinus Torvalds next = pud_addr_end(addr, end); 9851da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 9861da177e4SLinus Torvalds continue; 9871da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 9881da177e4SLinus Torvalds vma, addr, next)) 9891da177e4SLinus Torvalds return -ENOMEM; 9901da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 9911da177e4SLinus Torvalds return 0; 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9951da177e4SLinus Torvalds struct vm_area_struct *vma) 9961da177e4SLinus Torvalds { 9971da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 9981da177e4SLinus Torvalds unsigned long next; 9991da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10001da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10012ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10022ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10032ec74c3eSSagi Grimberg bool is_cow; 1004cddb8a5cSAndrea Arcangeli int ret; 10051da177e4SLinus Torvalds 1006d992895bSNick Piggin /* 1007d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1008d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1009d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1010d992895bSNick Piggin * efficient than faulting. 1011d992895bSNick Piggin */ 10120661a336SKirill A. Shutemov if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && 10130661a336SKirill A. Shutemov !vma->anon_vma) 1014d992895bSNick Piggin return 0; 1015d992895bSNick Piggin 10161da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10171da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10181da177e4SLinus Torvalds 1019b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10202ab64037Svenkatesh.pallipadi@intel.com /* 10212ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10222ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10232ab64037Svenkatesh.pallipadi@intel.com */ 10245180da41SSuresh Siddha ret = track_pfn_copy(vma); 10252ab64037Svenkatesh.pallipadi@intel.com if (ret) 10262ab64037Svenkatesh.pallipadi@intel.com return ret; 10272ab64037Svenkatesh.pallipadi@intel.com } 10282ab64037Svenkatesh.pallipadi@intel.com 1029cddb8a5cSAndrea Arcangeli /* 1030cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1031cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1032cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1033cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1034cddb8a5cSAndrea Arcangeli */ 10352ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10362ec74c3eSSagi Grimberg mmun_start = addr; 10372ec74c3eSSagi Grimberg mmun_end = end; 10382ec74c3eSSagi Grimberg if (is_cow) 10392ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10402ec74c3eSSagi Grimberg mmun_end); 1041cddb8a5cSAndrea Arcangeli 1042cddb8a5cSAndrea Arcangeli ret = 0; 10431da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10441da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10451da177e4SLinus Torvalds do { 10461da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10471da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10481da177e4SLinus Torvalds continue; 1049cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1050cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1051cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1052cddb8a5cSAndrea Arcangeli break; 1053cddb8a5cSAndrea Arcangeli } 10541da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1055cddb8a5cSAndrea Arcangeli 10562ec74c3eSSagi Grimberg if (is_cow) 10572ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1058cddb8a5cSAndrea Arcangeli return ret; 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds 106151c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1062b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 10631da177e4SLinus Torvalds unsigned long addr, unsigned long end, 106497a89413SPeter Zijlstra struct zap_details *details) 10651da177e4SLinus Torvalds { 1066b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1067d16dfc55SPeter Zijlstra int force_flush = 0; 1068d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 106997a89413SPeter Zijlstra spinlock_t *ptl; 10705f1a1907SSteven Rostedt pte_t *start_pte; 107197a89413SPeter Zijlstra pte_t *pte; 10728a5f14a2SKirill A. Shutemov swp_entry_t entry; 1073d559db08SKAMEZAWA Hiroyuki 1074d16dfc55SPeter Zijlstra again: 1075e303297eSPeter Zijlstra init_rss_vec(rss); 10765f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10775f1a1907SSteven Rostedt pte = start_pte; 10786606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10791da177e4SLinus Torvalds do { 10801da177e4SLinus Torvalds pte_t ptent = *pte; 108151c6f666SRobin Holt if (pte_none(ptent)) { 10821da177e4SLinus Torvalds continue; 108351c6f666SRobin Holt } 108451c6f666SRobin Holt 10856f5e6b9eSHugh Dickins if (pte_present(ptent)) { 10866f5e6b9eSHugh Dickins struct page *page; 10876f5e6b9eSHugh Dickins 10886aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 10891da177e4SLinus Torvalds if (unlikely(details) && page) { 10901da177e4SLinus Torvalds /* 10911da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 10921da177e4SLinus Torvalds * invalidate cache without truncating: 10931da177e4SLinus Torvalds * unmap shared but keep private pages. 10941da177e4SLinus Torvalds */ 10951da177e4SLinus Torvalds if (details->check_mapping && 10961da177e4SLinus Torvalds details->check_mapping != page->mapping) 10971da177e4SLinus Torvalds continue; 10981da177e4SLinus Torvalds } 1099b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1100a600388dSZachary Amsden tlb->fullmm); 11011da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11021da177e4SLinus Torvalds if (unlikely(!page)) 11031da177e4SLinus Torvalds continue; 1104eca56ff9SJerome Marchand 1105eca56ff9SJerome Marchand if (!PageAnon(page)) { 11061cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 11071cf35d47SLinus Torvalds force_flush = 1; 11086237bcd9SHugh Dickins set_page_dirty(page); 11091cf35d47SLinus Torvalds } 11104917e5d0SJohannes Weiner if (pte_young(ptent) && 111164363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1112bf3f3bc5SNick Piggin mark_page_accessed(page); 11136237bcd9SHugh Dickins } 1114eca56ff9SJerome Marchand rss[mm_counter(page)]--; 1115d281ee61SKirill A. Shutemov page_remove_rmap(page, false); 11163dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11173dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 11181cf35d47SLinus Torvalds if (unlikely(!__tlb_remove_page(tlb, page))) { 11191cf35d47SLinus Torvalds force_flush = 1; 1120ce9ec37bSWill Deacon addr += PAGE_SIZE; 1121d16dfc55SPeter Zijlstra break; 11221cf35d47SLinus Torvalds } 11231da177e4SLinus Torvalds continue; 11241da177e4SLinus Torvalds } 11258a5f14a2SKirill A. Shutemov /* If details->check_mapping, we leave swap entries. */ 11261da177e4SLinus Torvalds if (unlikely(details)) 11271da177e4SLinus Torvalds continue; 1128b084d435SKAMEZAWA Hiroyuki 11298a5f14a2SKirill A. Shutemov entry = pte_to_swp_entry(ptent); 1130b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1131b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11329f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11339f9f1acdSKonstantin Khlebnikov struct page *page; 11349f9f1acdSKonstantin Khlebnikov 11359f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 1136eca56ff9SJerome Marchand rss[mm_counter(page)]--; 11379f9f1acdSKonstantin Khlebnikov } 1138b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 11392509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 11409888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 114197a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1142ae859762SHugh Dickins 1143d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 11446606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 114551c6f666SRobin Holt 11461cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 1147fb7332a9SWill Deacon if (force_flush) 11481cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 11491cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 11501cf35d47SLinus Torvalds 11511cf35d47SLinus Torvalds /* 11521cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 11531cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 11541cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 11551cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 11561cf35d47SLinus Torvalds */ 11571cf35d47SLinus Torvalds if (force_flush) { 11581cf35d47SLinus Torvalds force_flush = 0; 11591cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 11602b047252SLinus Torvalds 11612b047252SLinus Torvalds if (addr != end) 1162d16dfc55SPeter Zijlstra goto again; 1163d16dfc55SPeter Zijlstra } 1164d16dfc55SPeter Zijlstra 116551c6f666SRobin Holt return addr; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 116851c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1169b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 11701da177e4SLinus Torvalds unsigned long addr, unsigned long end, 117197a89413SPeter Zijlstra struct zap_details *details) 11721da177e4SLinus Torvalds { 11731da177e4SLinus Torvalds pmd_t *pmd; 11741da177e4SLinus Torvalds unsigned long next; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 11771da177e4SLinus Torvalds do { 11781da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 117971e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 118014d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1181e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1182e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1183e0897d75SDavid Rientjes pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n", 1184e0897d75SDavid Rientjes __func__, addr, end, 1185e0897d75SDavid Rientjes vma->vm_start, 1186e0897d75SDavid Rientjes vma->vm_end); 1187e0897d75SDavid Rientjes BUG(); 1188e0897d75SDavid Rientjes } 1189e0897d75SDavid Rientjes #endif 119078ddc534SKirill A. Shutemov split_huge_pmd(vma, pmd, addr); 1191f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 11921a5a9906SAndrea Arcangeli goto next; 119371e3aac0SAndrea Arcangeli /* fall through */ 119471e3aac0SAndrea Arcangeli } 11951a5a9906SAndrea Arcangeli /* 11961a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 11971a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 11981a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 11991a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12001a5a9906SAndrea Arcangeli * mode. 12011a5a9906SAndrea Arcangeli */ 12021a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12031a5a9906SAndrea Arcangeli goto next; 120497a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12051a5a9906SAndrea Arcangeli next: 120697a89413SPeter Zijlstra cond_resched(); 120797a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 120851c6f666SRobin Holt 120951c6f666SRobin Holt return addr; 12101da177e4SLinus Torvalds } 12111da177e4SLinus Torvalds 121251c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1213b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12141da177e4SLinus Torvalds unsigned long addr, unsigned long end, 121597a89413SPeter Zijlstra struct zap_details *details) 12161da177e4SLinus Torvalds { 12171da177e4SLinus Torvalds pud_t *pud; 12181da177e4SLinus Torvalds unsigned long next; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12211da177e4SLinus Torvalds do { 12221da177e4SLinus Torvalds next = pud_addr_end(addr, end); 122397a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12241da177e4SLinus Torvalds continue; 122597a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 122697a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 122751c6f666SRobin Holt 122851c6f666SRobin Holt return addr; 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 1231038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 123251c6f666SRobin Holt struct vm_area_struct *vma, 12331da177e4SLinus Torvalds unsigned long addr, unsigned long end, 123497a89413SPeter Zijlstra struct zap_details *details) 12351da177e4SLinus Torvalds { 12361da177e4SLinus Torvalds pgd_t *pgd; 12371da177e4SLinus Torvalds unsigned long next; 12381da177e4SLinus Torvalds 12398a5f14a2SKirill A. Shutemov if (details && !details->check_mapping) 12401da177e4SLinus Torvalds details = NULL; 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds BUG_ON(addr >= end); 12431da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 12441da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 12451da177e4SLinus Torvalds do { 12461da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 124797a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 12481da177e4SLinus Torvalds continue; 124997a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 125097a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 12511da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 12521da177e4SLinus Torvalds } 12531da177e4SLinus Torvalds 1254f5cc4eefSAl Viro 1255f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 12561da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 12574f74d2c8SLinus Torvalds unsigned long end_addr, 12581da177e4SLinus Torvalds struct zap_details *details) 12591da177e4SLinus Torvalds { 1260f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 12611da177e4SLinus Torvalds unsigned long end; 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds if (start >= vma->vm_end) 1264f5cc4eefSAl Viro return; 12651da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 12661da177e4SLinus Torvalds if (end <= vma->vm_start) 1267f5cc4eefSAl Viro return; 12681da177e4SLinus Torvalds 1269cbc91f71SSrikar Dronamraju if (vma->vm_file) 1270cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1271cbc91f71SSrikar Dronamraju 1272b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 12735180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 12742ab64037Svenkatesh.pallipadi@intel.com 12758b2a1238SAl Viro if (start != end) { 127651c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1277a137e1ccSAndi Kleen /* 1278a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1279a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1280a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 12817aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1282a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 12837aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1284a137e1ccSAndi Kleen * before calling this function to clean up. 1285a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1286a137e1ccSAndi Kleen * safe to do nothing in this case. 1287a137e1ccSAndi Kleen */ 128824669e58SAneesh Kumar K.V if (vma->vm_file) { 128983cde9e8SDavidlohr Bueso i_mmap_lock_write(vma->vm_file->f_mapping); 1290d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 129183cde9e8SDavidlohr Bueso i_mmap_unlock_write(vma->vm_file->f_mapping); 129224669e58SAneesh Kumar K.V } 129351c6f666SRobin Holt } else 1294038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 129597a89413SPeter Zijlstra } 129651c6f666SRobin Holt } 12971da177e4SLinus Torvalds 1298f5cc4eefSAl Viro /** 1299f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1300f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1301f5cc4eefSAl Viro * @vma: the starting vma 1302f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1303f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1304f5cc4eefSAl Viro * 1305f5cc4eefSAl Viro * Unmap all pages in the vma list. 1306f5cc4eefSAl Viro * 1307f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1308f5cc4eefSAl Viro * 1309f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1310f5cc4eefSAl Viro * 1311f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1312f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1313f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1314f5cc4eefSAl Viro * drops the lock and schedules. 1315f5cc4eefSAl Viro */ 1316f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1317f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13184f74d2c8SLinus Torvalds unsigned long end_addr) 1319f5cc4eefSAl Viro { 1320f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1321f5cc4eefSAl Viro 1322f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1323f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13244f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1325cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds /** 13291da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13301da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1331eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13321da177e4SLinus Torvalds * @size: number of bytes to zap 13338a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1334f5cc4eefSAl Viro * 1335f5cc4eefSAl Viro * Caller must protect the VMA list 13361da177e4SLinus Torvalds */ 13377e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 13381da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 13391da177e4SLinus Torvalds { 13401da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1341d16dfc55SPeter Zijlstra struct mmu_gather tlb; 13427e027b14SLinus Torvalds unsigned long end = start + size; 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds lru_add_drain(); 13452b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 1346365e9c87SHugh Dickins update_hiwater_rss(mm); 13477e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 13487e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 13494f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 13507e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 13517e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 1354c627f9ccSJack Steiner /** 1355f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1356f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1357f5cc4eefSAl Viro * @address: starting address of pages to zap 1358f5cc4eefSAl Viro * @size: number of bytes to zap 13598a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1360f5cc4eefSAl Viro * 1361f5cc4eefSAl Viro * The range must fit into one VMA. 1362f5cc4eefSAl Viro */ 1363f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1364f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1365f5cc4eefSAl Viro { 1366f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1367f5cc4eefSAl Viro struct mmu_gather tlb; 1368f5cc4eefSAl Viro unsigned long end = address + size; 1369f5cc4eefSAl Viro 1370f5cc4eefSAl Viro lru_add_drain(); 13712b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, address, end); 1372f5cc4eefSAl Viro update_hiwater_rss(mm); 1373f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 13744f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1375f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1376f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds 1379c627f9ccSJack Steiner /** 1380c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1381c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1382c627f9ccSJack Steiner * @address: starting address of pages to zap 1383c627f9ccSJack Steiner * @size: number of bytes to zap 1384c627f9ccSJack Steiner * 1385c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1386c627f9ccSJack Steiner * 1387c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1388c627f9ccSJack Steiner * 1389c627f9ccSJack Steiner * Returns 0 if successful. 1390c627f9ccSJack Steiner */ 1391c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1392c627f9ccSJack Steiner unsigned long size) 1393c627f9ccSJack Steiner { 1394c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1395c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1396c627f9ccSJack Steiner return -1; 1397f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1398c627f9ccSJack Steiner return 0; 1399c627f9ccSJack Steiner } 1400c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1401c627f9ccSJack Steiner 140225ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 1403920c7a5dSHarvey Harrison spinlock_t **ptl) 1404c9cfcddfSLinus Torvalds { 1405c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1406c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1407c9cfcddfSLinus Torvalds if (pud) { 140849c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1409f66055abSAndrea Arcangeli if (pmd) { 1410f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 1411c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1412c9cfcddfSLinus Torvalds } 1413f66055abSAndrea Arcangeli } 1414c9cfcddfSLinus Torvalds return NULL; 1415c9cfcddfSLinus Torvalds } 1416c9cfcddfSLinus Torvalds 14171da177e4SLinus Torvalds /* 1418238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1419238f58d8SLinus Torvalds * 1420238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1421238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1422238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1423238f58d8SLinus Torvalds */ 1424423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1425423bad60SNick Piggin struct page *page, pgprot_t prot) 1426238f58d8SLinus Torvalds { 1427423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1428238f58d8SLinus Torvalds int retval; 1429238f58d8SLinus Torvalds pte_t *pte; 1430238f58d8SLinus Torvalds spinlock_t *ptl; 1431238f58d8SLinus Torvalds 1432238f58d8SLinus Torvalds retval = -EINVAL; 1433a145dd41SLinus Torvalds if (PageAnon(page)) 14345b4e655eSKAMEZAWA Hiroyuki goto out; 1435238f58d8SLinus Torvalds retval = -ENOMEM; 1436238f58d8SLinus Torvalds flush_dcache_page(page); 1437c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1438238f58d8SLinus Torvalds if (!pte) 14395b4e655eSKAMEZAWA Hiroyuki goto out; 1440238f58d8SLinus Torvalds retval = -EBUSY; 1441238f58d8SLinus Torvalds if (!pte_none(*pte)) 1442238f58d8SLinus Torvalds goto out_unlock; 1443238f58d8SLinus Torvalds 1444238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1445238f58d8SLinus Torvalds get_page(page); 1446eca56ff9SJerome Marchand inc_mm_counter_fast(mm, mm_counter_file(page)); 1447238f58d8SLinus Torvalds page_add_file_rmap(page); 1448238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1449238f58d8SLinus Torvalds 1450238f58d8SLinus Torvalds retval = 0; 14518a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 14528a9f3ccdSBalbir Singh return retval; 1453238f58d8SLinus Torvalds out_unlock: 1454238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1455238f58d8SLinus Torvalds out: 1456238f58d8SLinus Torvalds return retval; 1457238f58d8SLinus Torvalds } 1458238f58d8SLinus Torvalds 1459bfa5bf6dSRolf Eike Beer /** 1460bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1461bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1462bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1463bfa5bf6dSRolf Eike Beer * @page: source kernel page 1464bfa5bf6dSRolf Eike Beer * 1465a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1466a145dd41SLinus Torvalds * into a user vma. 1467a145dd41SLinus Torvalds * 1468a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1469a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1470a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 14718dfcc9baSNick Piggin * (see split_page()). 1472a145dd41SLinus Torvalds * 1473a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1474a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1475a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1476a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1477a145dd41SLinus Torvalds * ask for a shared writable mapping! 1478a145dd41SLinus Torvalds * 1479a145dd41SLinus Torvalds * The page does not need to be reserved. 14804b6e1e37SKonstantin Khlebnikov * 14814b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 14824b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 14834b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 14844b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 1485a145dd41SLinus Torvalds */ 1486423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 1487423bad60SNick Piggin struct page *page) 1488a145dd41SLinus Torvalds { 1489a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1490a145dd41SLinus Torvalds return -EFAULT; 1491a145dd41SLinus Torvalds if (!page_count(page)) 1492a145dd41SLinus Torvalds return -EINVAL; 14934b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 14944b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 14954b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 14964b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 14974b6e1e37SKonstantin Khlebnikov } 1498423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1499a145dd41SLinus Torvalds } 1500e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1501a145dd41SLinus Torvalds 1502423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1503423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 1504423bad60SNick Piggin { 1505423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1506423bad60SNick Piggin int retval; 1507423bad60SNick Piggin pte_t *pte, entry; 1508423bad60SNick Piggin spinlock_t *ptl; 1509423bad60SNick Piggin 1510423bad60SNick Piggin retval = -ENOMEM; 1511423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1512423bad60SNick Piggin if (!pte) 1513423bad60SNick Piggin goto out; 1514423bad60SNick Piggin retval = -EBUSY; 1515423bad60SNick Piggin if (!pte_none(*pte)) 1516423bad60SNick Piggin goto out_unlock; 1517423bad60SNick Piggin 1518423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 1519423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 1520423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 15214b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 1522423bad60SNick Piggin 1523423bad60SNick Piggin retval = 0; 1524423bad60SNick Piggin out_unlock: 1525423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 1526423bad60SNick Piggin out: 1527423bad60SNick Piggin return retval; 1528423bad60SNick Piggin } 1529423bad60SNick Piggin 1530e0dc0d8fSNick Piggin /** 1531e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 1532e0dc0d8fSNick Piggin * @vma: user vma to map to 1533e0dc0d8fSNick Piggin * @addr: target user address of this page 1534e0dc0d8fSNick Piggin * @pfn: source kernel pfn 1535e0dc0d8fSNick Piggin * 1536c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 1537e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 1538e0dc0d8fSNick Piggin * 1539e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 1540e0dc0d8fSNick Piggin * in that case the handler should return NULL. 15410d71d10aSNick Piggin * 15420d71d10aSNick Piggin * vma cannot be a COW mapping. 15430d71d10aSNick Piggin * 15440d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 15450d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 1546e0dc0d8fSNick Piggin */ 1547e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1548e0dc0d8fSNick Piggin unsigned long pfn) 1549e0dc0d8fSNick Piggin { 15502ab64037Svenkatesh.pallipadi@intel.com int ret; 1551e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 15527e675137SNick Piggin /* 15537e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 15547e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 15557e675137SNick Piggin * consistency in testing and feature parity among all, so we should 15567e675137SNick Piggin * try to keep these invariants in place for everybody. 15577e675137SNick Piggin */ 1558b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 1559b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 1560b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 1561b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 1562b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 1563e0dc0d8fSNick Piggin 1564423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1565423bad60SNick Piggin return -EFAULT; 15665180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 15672ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 15682ab64037Svenkatesh.pallipadi@intel.com 1569e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 15702ab64037Svenkatesh.pallipadi@intel.com 15712ab64037Svenkatesh.pallipadi@intel.com return ret; 1572e0dc0d8fSNick Piggin } 1573e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 1574e0dc0d8fSNick Piggin 1575423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 1576423bad60SNick Piggin unsigned long pfn) 1577423bad60SNick Piggin { 1578423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 1579423bad60SNick Piggin 1580423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1581423bad60SNick Piggin return -EFAULT; 1582423bad60SNick Piggin 1583423bad60SNick Piggin /* 1584423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 1585423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 1586423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 158762eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 158862eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 1589423bad60SNick Piggin */ 1590423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 1591423bad60SNick Piggin struct page *page; 1592423bad60SNick Piggin 1593423bad60SNick Piggin page = pfn_to_page(pfn); 1594423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1595423bad60SNick Piggin } 1596423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 1597423bad60SNick Piggin } 1598423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 1599423bad60SNick Piggin 1600a145dd41SLinus Torvalds /* 16011da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 16021da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 16031da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 16041da177e4SLinus Torvalds */ 16051da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 16061da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16071da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16081da177e4SLinus Torvalds { 16091da177e4SLinus Torvalds pte_t *pte; 1610c74df32cSHugh Dickins spinlock_t *ptl; 16111da177e4SLinus Torvalds 1612c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 16131da177e4SLinus Torvalds if (!pte) 16141da177e4SLinus Torvalds return -ENOMEM; 16156606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 16161da177e4SLinus Torvalds do { 16171da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 16187e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 16191da177e4SLinus Torvalds pfn++; 16201da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 16216606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1622c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 16231da177e4SLinus Torvalds return 0; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 16271da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16281da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16291da177e4SLinus Torvalds { 16301da177e4SLinus Torvalds pmd_t *pmd; 16311da177e4SLinus Torvalds unsigned long next; 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16341da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 16351da177e4SLinus Torvalds if (!pmd) 16361da177e4SLinus Torvalds return -ENOMEM; 1637f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 16381da177e4SLinus Torvalds do { 16391da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 16401da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 16411da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16421da177e4SLinus Torvalds return -ENOMEM; 16431da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 16441da177e4SLinus Torvalds return 0; 16451da177e4SLinus Torvalds } 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 16481da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16491da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16501da177e4SLinus Torvalds { 16511da177e4SLinus Torvalds pud_t *pud; 16521da177e4SLinus Torvalds unsigned long next; 16531da177e4SLinus Torvalds 16541da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16551da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 16561da177e4SLinus Torvalds if (!pud) 16571da177e4SLinus Torvalds return -ENOMEM; 16581da177e4SLinus Torvalds do { 16591da177e4SLinus Torvalds next = pud_addr_end(addr, end); 16601da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 16611da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16621da177e4SLinus Torvalds return -ENOMEM; 16631da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 16641da177e4SLinus Torvalds return 0; 16651da177e4SLinus Torvalds } 16661da177e4SLinus Torvalds 1667bfa5bf6dSRolf Eike Beer /** 1668bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 1669bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1670bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 1671bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 1672bfa5bf6dSRolf Eike Beer * @size: size of map area 1673bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 1674bfa5bf6dSRolf Eike Beer * 1675bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 1676bfa5bf6dSRolf Eike Beer */ 16771da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 16781da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 16791da177e4SLinus Torvalds { 16801da177e4SLinus Torvalds pgd_t *pgd; 16811da177e4SLinus Torvalds unsigned long next; 16822d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 16831da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 16841da177e4SLinus Torvalds int err; 16851da177e4SLinus Torvalds 16861da177e4SLinus Torvalds /* 16871da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 16881da177e4SLinus Torvalds * rest of the world about it: 16891da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 16901da177e4SLinus Torvalds * (accesses can have side effects). 16916aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 16926aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 16936aab341eSLinus Torvalds * with them. 1694314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 1695314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 1696314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 1697314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 1698fb155c16SLinus Torvalds * 1699fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 1700fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 1701fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 1702b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 17031da177e4SLinus Torvalds */ 1704b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 1705b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 1706b3b9c293SKonstantin Khlebnikov return -EINVAL; 17076aab341eSLinus Torvalds vma->vm_pgoff = pfn; 1708b3b9c293SKonstantin Khlebnikov } 1709b3b9c293SKonstantin Khlebnikov 1710b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 1711b3b9c293SKonstantin Khlebnikov if (err) 17123c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 1713fb155c16SLinus Torvalds 1714314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds BUG_ON(addr >= end); 17171da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 17181da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 17191da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 17201da177e4SLinus Torvalds do { 17211da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 17221da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 17231da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 17241da177e4SLinus Torvalds if (err) 17251da177e4SLinus Torvalds break; 17261da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 17272ab64037Svenkatesh.pallipadi@intel.com 17282ab64037Svenkatesh.pallipadi@intel.com if (err) 17295180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 17302ab64037Svenkatesh.pallipadi@intel.com 17311da177e4SLinus Torvalds return err; 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 17341da177e4SLinus Torvalds 1735b4cbb197SLinus Torvalds /** 1736b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 1737b4cbb197SLinus Torvalds * @vma: user vma to map to 1738b4cbb197SLinus Torvalds * @start: start of area 1739b4cbb197SLinus Torvalds * @len: size of area 1740b4cbb197SLinus Torvalds * 1741b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 1742b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 1743b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 1744b4cbb197SLinus Torvalds * 1745b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 1746b4cbb197SLinus Torvalds * whatever write-combining details or similar. 1747b4cbb197SLinus Torvalds */ 1748b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 1749b4cbb197SLinus Torvalds { 1750b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 1751b4cbb197SLinus Torvalds 1752b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 1753b4cbb197SLinus Torvalds if (start + len < start) 1754b4cbb197SLinus Torvalds return -EINVAL; 1755b4cbb197SLinus Torvalds /* 1756b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 1757b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 1758b4cbb197SLinus Torvalds * just have smaller alignment. 1759b4cbb197SLinus Torvalds */ 1760b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 1761b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 1762b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 1763b4cbb197SLinus Torvalds if (pfn + pages < pfn) 1764b4cbb197SLinus Torvalds return -EINVAL; 1765b4cbb197SLinus Torvalds 1766b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 1767b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 1768b4cbb197SLinus Torvalds return -EINVAL; 1769b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 1770b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 1771b4cbb197SLinus Torvalds 1772b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 1773b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 1774b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 1775b4cbb197SLinus Torvalds return -EINVAL; 1776b4cbb197SLinus Torvalds 1777b4cbb197SLinus Torvalds /* Ok, let it rip */ 1778b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 1779b4cbb197SLinus Torvalds } 1780b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 1781b4cbb197SLinus Torvalds 1782aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 1783aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1784aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1785aee16b3cSJeremy Fitzhardinge { 1786aee16b3cSJeremy Fitzhardinge pte_t *pte; 1787aee16b3cSJeremy Fitzhardinge int err; 17882f569afdSMartin Schwidefsky pgtable_t token; 178994909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 1790aee16b3cSJeremy Fitzhardinge 1791aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 1792aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 1793aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 1794aee16b3cSJeremy Fitzhardinge if (!pte) 1795aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1796aee16b3cSJeremy Fitzhardinge 1797aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 1798aee16b3cSJeremy Fitzhardinge 179938e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 180038e0edb1SJeremy Fitzhardinge 18012f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 1802aee16b3cSJeremy Fitzhardinge 1803aee16b3cSJeremy Fitzhardinge do { 1804c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 1805aee16b3cSJeremy Fitzhardinge if (err) 1806aee16b3cSJeremy Fitzhardinge break; 1807c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 1808aee16b3cSJeremy Fitzhardinge 180938e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 181038e0edb1SJeremy Fitzhardinge 1811aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 1812aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 1813aee16b3cSJeremy Fitzhardinge return err; 1814aee16b3cSJeremy Fitzhardinge } 1815aee16b3cSJeremy Fitzhardinge 1816aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 1817aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1818aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1819aee16b3cSJeremy Fitzhardinge { 1820aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 1821aee16b3cSJeremy Fitzhardinge unsigned long next; 1822aee16b3cSJeremy Fitzhardinge int err; 1823aee16b3cSJeremy Fitzhardinge 1824ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 1825ceb86879SAndi Kleen 1826aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 1827aee16b3cSJeremy Fitzhardinge if (!pmd) 1828aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1829aee16b3cSJeremy Fitzhardinge do { 1830aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 1831aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 1832aee16b3cSJeremy Fitzhardinge if (err) 1833aee16b3cSJeremy Fitzhardinge break; 1834aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 1835aee16b3cSJeremy Fitzhardinge return err; 1836aee16b3cSJeremy Fitzhardinge } 1837aee16b3cSJeremy Fitzhardinge 1838aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 1839aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1840aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1841aee16b3cSJeremy Fitzhardinge { 1842aee16b3cSJeremy Fitzhardinge pud_t *pud; 1843aee16b3cSJeremy Fitzhardinge unsigned long next; 1844aee16b3cSJeremy Fitzhardinge int err; 1845aee16b3cSJeremy Fitzhardinge 1846aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 1847aee16b3cSJeremy Fitzhardinge if (!pud) 1848aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1849aee16b3cSJeremy Fitzhardinge do { 1850aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 1851aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 1852aee16b3cSJeremy Fitzhardinge if (err) 1853aee16b3cSJeremy Fitzhardinge break; 1854aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 1855aee16b3cSJeremy Fitzhardinge return err; 1856aee16b3cSJeremy Fitzhardinge } 1857aee16b3cSJeremy Fitzhardinge 1858aee16b3cSJeremy Fitzhardinge /* 1859aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 1860aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 1861aee16b3cSJeremy Fitzhardinge */ 1862aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 1863aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 1864aee16b3cSJeremy Fitzhardinge { 1865aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 1866aee16b3cSJeremy Fitzhardinge unsigned long next; 186757250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 1868aee16b3cSJeremy Fitzhardinge int err; 1869aee16b3cSJeremy Fitzhardinge 1870aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 1871aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 1872aee16b3cSJeremy Fitzhardinge do { 1873aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 1874aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 1875aee16b3cSJeremy Fitzhardinge if (err) 1876aee16b3cSJeremy Fitzhardinge break; 1877aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 187857250a5bSJeremy Fitzhardinge 1879aee16b3cSJeremy Fitzhardinge return err; 1880aee16b3cSJeremy Fitzhardinge } 1881aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 1882aee16b3cSJeremy Fitzhardinge 18831da177e4SLinus Torvalds /* 18849b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 18859b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 18869b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 18879b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 18889b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 1889a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 18908f4e2101SHugh Dickins */ 18914c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 18928f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 18938f4e2101SHugh Dickins { 18948f4e2101SHugh Dickins int same = 1; 18958f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 18968f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 18974c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 18984c21e2f2SHugh Dickins spin_lock(ptl); 18998f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 19004c21e2f2SHugh Dickins spin_unlock(ptl); 19018f4e2101SHugh Dickins } 19028f4e2101SHugh Dickins #endif 19038f4e2101SHugh Dickins pte_unmap(page_table); 19048f4e2101SHugh Dickins return same; 19058f4e2101SHugh Dickins } 19068f4e2101SHugh Dickins 19079de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 19086aab341eSLinus Torvalds { 19090abdd7a8SDan Williams debug_dma_assert_idle(src); 19100abdd7a8SDan Williams 19116aab341eSLinus Torvalds /* 19126aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 19136aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 19146aab341eSLinus Torvalds * just copying from the original user address. If that 19156aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 19166aab341eSLinus Torvalds */ 19176aab341eSLinus Torvalds if (unlikely(!src)) { 19189b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 19195d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 19205d2a2dbbSLinus Torvalds 19215d2a2dbbSLinus Torvalds /* 19225d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 19235d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 19245d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 19255d2a2dbbSLinus Torvalds * zeroes. 19265d2a2dbbSLinus Torvalds */ 19275d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 19283ecb01dfSJan Beulich clear_page(kaddr); 19299b04c5feSCong Wang kunmap_atomic(kaddr); 1930c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 19310ed361deSNick Piggin } else 19329de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 19336aab341eSLinus Torvalds } 19346aab341eSLinus Torvalds 1935c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 1936c20cd45eSMichal Hocko { 1937c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 1938c20cd45eSMichal Hocko 1939c20cd45eSMichal Hocko if (vm_file) 1940c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 1941c20cd45eSMichal Hocko 1942c20cd45eSMichal Hocko /* 1943c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 1944c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 1945c20cd45eSMichal Hocko */ 1946c20cd45eSMichal Hocko return GFP_KERNEL; 1947c20cd45eSMichal Hocko } 1948c20cd45eSMichal Hocko 19491da177e4SLinus Torvalds /* 1950fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 1951fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 1952fb09a464SKirill A. Shutemov * 1953fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 1954fb09a464SKirill A. Shutemov */ 1955fb09a464SKirill A. Shutemov static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, 1956fb09a464SKirill A. Shutemov unsigned long address) 1957fb09a464SKirill A. Shutemov { 1958fb09a464SKirill A. Shutemov struct vm_fault vmf; 1959fb09a464SKirill A. Shutemov int ret; 1960fb09a464SKirill A. Shutemov 1961fb09a464SKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 1962fb09a464SKirill A. Shutemov vmf.pgoff = page->index; 1963fb09a464SKirill A. Shutemov vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 1964c20cd45eSMichal Hocko vmf.gfp_mask = __get_fault_gfp_mask(vma); 1965fb09a464SKirill A. Shutemov vmf.page = page; 19662e4cdab0SMatthew Wilcox vmf.cow_page = NULL; 1967fb09a464SKirill A. Shutemov 1968fb09a464SKirill A. Shutemov ret = vma->vm_ops->page_mkwrite(vma, &vmf); 1969fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 1970fb09a464SKirill A. Shutemov return ret; 1971fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 1972fb09a464SKirill A. Shutemov lock_page(page); 1973fb09a464SKirill A. Shutemov if (!page->mapping) { 1974fb09a464SKirill A. Shutemov unlock_page(page); 1975fb09a464SKirill A. Shutemov return 0; /* retry */ 1976fb09a464SKirill A. Shutemov } 1977fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 1978fb09a464SKirill A. Shutemov } else 1979fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 1980fb09a464SKirill A. Shutemov return ret; 1981fb09a464SKirill A. Shutemov } 1982fb09a464SKirill A. Shutemov 1983fb09a464SKirill A. Shutemov /* 19844e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 19854e047f89SShachar Raindel * 19864e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 19874e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 19884e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 19894e047f89SShachar Raindel * any related book-keeping. 19904e047f89SShachar Raindel */ 19914e047f89SShachar Raindel static inline int wp_page_reuse(struct mm_struct *mm, 19924e047f89SShachar Raindel struct vm_area_struct *vma, unsigned long address, 19934e047f89SShachar Raindel pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, 19944e047f89SShachar Raindel struct page *page, int page_mkwrite, 19954e047f89SShachar Raindel int dirty_shared) 19964e047f89SShachar Raindel __releases(ptl) 19974e047f89SShachar Raindel { 19984e047f89SShachar Raindel pte_t entry; 19994e047f89SShachar Raindel /* 20004e047f89SShachar Raindel * Clear the pages cpupid information as the existing 20014e047f89SShachar Raindel * information potentially belongs to a now completely 20024e047f89SShachar Raindel * unrelated process. 20034e047f89SShachar Raindel */ 20044e047f89SShachar Raindel if (page) 20054e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 20064e047f89SShachar Raindel 20074e047f89SShachar Raindel flush_cache_page(vma, address, pte_pfn(orig_pte)); 20084e047f89SShachar Raindel entry = pte_mkyoung(orig_pte); 20094e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 20104e047f89SShachar Raindel if (ptep_set_access_flags(vma, address, page_table, entry, 1)) 20114e047f89SShachar Raindel update_mmu_cache(vma, address, page_table); 20124e047f89SShachar Raindel pte_unmap_unlock(page_table, ptl); 20134e047f89SShachar Raindel 20144e047f89SShachar Raindel if (dirty_shared) { 20154e047f89SShachar Raindel struct address_space *mapping; 20164e047f89SShachar Raindel int dirtied; 20174e047f89SShachar Raindel 20184e047f89SShachar Raindel if (!page_mkwrite) 20194e047f89SShachar Raindel lock_page(page); 20204e047f89SShachar Raindel 20214e047f89SShachar Raindel dirtied = set_page_dirty(page); 20224e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 20234e047f89SShachar Raindel mapping = page->mapping; 20244e047f89SShachar Raindel unlock_page(page); 20254e047f89SShachar Raindel page_cache_release(page); 20264e047f89SShachar Raindel 20274e047f89SShachar Raindel if ((dirtied || page_mkwrite) && mapping) { 20284e047f89SShachar Raindel /* 20294e047f89SShachar Raindel * Some device drivers do not set page.mapping 20304e047f89SShachar Raindel * but still dirty their pages 20314e047f89SShachar Raindel */ 20324e047f89SShachar Raindel balance_dirty_pages_ratelimited(mapping); 20334e047f89SShachar Raindel } 20344e047f89SShachar Raindel 20354e047f89SShachar Raindel if (!page_mkwrite) 20364e047f89SShachar Raindel file_update_time(vma->vm_file); 20374e047f89SShachar Raindel } 20384e047f89SShachar Raindel 20394e047f89SShachar Raindel return VM_FAULT_WRITE; 20404e047f89SShachar Raindel } 20414e047f89SShachar Raindel 20424e047f89SShachar Raindel /* 20432f38ab2cSShachar Raindel * Handle the case of a page which we actually need to copy to a new page. 20442f38ab2cSShachar Raindel * 20452f38ab2cSShachar Raindel * Called with mmap_sem locked and the old page referenced, but 20462f38ab2cSShachar Raindel * without the ptl held. 20472f38ab2cSShachar Raindel * 20482f38ab2cSShachar Raindel * High level logic flow: 20492f38ab2cSShachar Raindel * 20502f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 20512f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 20522f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 20532f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 20542f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 20552f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 20562f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 20572f38ab2cSShachar Raindel */ 20582f38ab2cSShachar Raindel static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma, 20592f38ab2cSShachar Raindel unsigned long address, pte_t *page_table, pmd_t *pmd, 20602f38ab2cSShachar Raindel pte_t orig_pte, struct page *old_page) 20612f38ab2cSShachar Raindel { 20622f38ab2cSShachar Raindel struct page *new_page = NULL; 20632f38ab2cSShachar Raindel spinlock_t *ptl = NULL; 20642f38ab2cSShachar Raindel pte_t entry; 20652f38ab2cSShachar Raindel int page_copied = 0; 20662f38ab2cSShachar Raindel const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */ 20672f38ab2cSShachar Raindel const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */ 20682f38ab2cSShachar Raindel struct mem_cgroup *memcg; 20692f38ab2cSShachar Raindel 20702f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 20712f38ab2cSShachar Raindel goto oom; 20722f38ab2cSShachar Raindel 20732f38ab2cSShachar Raindel if (is_zero_pfn(pte_pfn(orig_pte))) { 20742f38ab2cSShachar Raindel new_page = alloc_zeroed_user_highpage_movable(vma, address); 20752f38ab2cSShachar Raindel if (!new_page) 20762f38ab2cSShachar Raindel goto oom; 20772f38ab2cSShachar Raindel } else { 20782f38ab2cSShachar Raindel new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 20792f38ab2cSShachar Raindel if (!new_page) 20802f38ab2cSShachar Raindel goto oom; 20812f38ab2cSShachar Raindel cow_user_page(new_page, old_page, address, vma); 20822f38ab2cSShachar Raindel } 20832f38ab2cSShachar Raindel 2084f627c2f5SKirill A. Shutemov if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) 20852f38ab2cSShachar Raindel goto oom_free_new; 20862f38ab2cSShachar Raindel 2087eb3c24f3SMel Gorman __SetPageUptodate(new_page); 2088eb3c24f3SMel Gorman 20892f38ab2cSShachar Raindel mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 20902f38ab2cSShachar Raindel 20912f38ab2cSShachar Raindel /* 20922f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 20932f38ab2cSShachar Raindel */ 20942f38ab2cSShachar Raindel page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 20952f38ab2cSShachar Raindel if (likely(pte_same(*page_table, orig_pte))) { 20962f38ab2cSShachar Raindel if (old_page) { 20972f38ab2cSShachar Raindel if (!PageAnon(old_page)) { 2098eca56ff9SJerome Marchand dec_mm_counter_fast(mm, 2099eca56ff9SJerome Marchand mm_counter_file(old_page)); 21002f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 21012f38ab2cSShachar Raindel } 21022f38ab2cSShachar Raindel } else { 21032f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 21042f38ab2cSShachar Raindel } 21052f38ab2cSShachar Raindel flush_cache_page(vma, address, pte_pfn(orig_pte)); 21062f38ab2cSShachar Raindel entry = mk_pte(new_page, vma->vm_page_prot); 21072f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 21082f38ab2cSShachar Raindel /* 21092f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 21102f38ab2cSShachar Raindel * pte with the new entry. This will avoid a race condition 21112f38ab2cSShachar Raindel * seen in the presence of one thread doing SMC and another 21122f38ab2cSShachar Raindel * thread doing COW. 21132f38ab2cSShachar Raindel */ 21142f38ab2cSShachar Raindel ptep_clear_flush_notify(vma, address, page_table); 2115d281ee61SKirill A. Shutemov page_add_new_anon_rmap(new_page, vma, address, false); 2116f627c2f5SKirill A. Shutemov mem_cgroup_commit_charge(new_page, memcg, false, false); 21172f38ab2cSShachar Raindel lru_cache_add_active_or_unevictable(new_page, vma); 21182f38ab2cSShachar Raindel /* 21192f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 21202f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 21212f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 21222f38ab2cSShachar Raindel */ 21232f38ab2cSShachar Raindel set_pte_at_notify(mm, address, page_table, entry); 21242f38ab2cSShachar Raindel update_mmu_cache(vma, address, page_table); 21252f38ab2cSShachar Raindel if (old_page) { 21262f38ab2cSShachar Raindel /* 21272f38ab2cSShachar Raindel * Only after switching the pte to the new page may 21282f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 21292f38ab2cSShachar Raindel * process may come and find the rmap count decremented 21302f38ab2cSShachar Raindel * before the pte is switched to the new page, and 21312f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 21322f38ab2cSShachar Raindel * here still points into it and can be read by other 21332f38ab2cSShachar Raindel * threads. 21342f38ab2cSShachar Raindel * 21352f38ab2cSShachar Raindel * The critical issue is to order this 21362f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 21372f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 21382f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 21392f38ab2cSShachar Raindel * in page_remove_rmap. 21402f38ab2cSShachar Raindel * 21412f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 21422f38ab2cSShachar Raindel * no process can access the old page before the 21432f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 21442f38ab2cSShachar Raindel * cannot be reused until after the decremented 21452f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 21462f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 21472f38ab2cSShachar Raindel */ 2148d281ee61SKirill A. Shutemov page_remove_rmap(old_page, false); 21492f38ab2cSShachar Raindel } 21502f38ab2cSShachar Raindel 21512f38ab2cSShachar Raindel /* Free the old page.. */ 21522f38ab2cSShachar Raindel new_page = old_page; 21532f38ab2cSShachar Raindel page_copied = 1; 21542f38ab2cSShachar Raindel } else { 2155f627c2f5SKirill A. Shutemov mem_cgroup_cancel_charge(new_page, memcg, false); 21562f38ab2cSShachar Raindel } 21572f38ab2cSShachar Raindel 21582f38ab2cSShachar Raindel if (new_page) 21592f38ab2cSShachar Raindel page_cache_release(new_page); 21602f38ab2cSShachar Raindel 21612f38ab2cSShachar Raindel pte_unmap_unlock(page_table, ptl); 21622f38ab2cSShachar Raindel mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 21637479df6dSKirill A. Shutemov /* THP pages are never mlocked */ 21647479df6dSKirill A. Shutemov if (old_page && !PageTransCompound(old_page)) { 21652f38ab2cSShachar Raindel /* 21662f38ab2cSShachar Raindel * Don't let another task, with possibly unlocked vma, 21672f38ab2cSShachar Raindel * keep the mlocked page. 21682f38ab2cSShachar Raindel */ 21692f38ab2cSShachar Raindel if (page_copied && (vma->vm_flags & VM_LOCKED)) { 21702f38ab2cSShachar Raindel lock_page(old_page); /* LRU manipulation */ 21712f38ab2cSShachar Raindel munlock_vma_page(old_page); 21722f38ab2cSShachar Raindel unlock_page(old_page); 21732f38ab2cSShachar Raindel } 21742f38ab2cSShachar Raindel page_cache_release(old_page); 21752f38ab2cSShachar Raindel } 21762f38ab2cSShachar Raindel return page_copied ? VM_FAULT_WRITE : 0; 21772f38ab2cSShachar Raindel oom_free_new: 21782f38ab2cSShachar Raindel page_cache_release(new_page); 21792f38ab2cSShachar Raindel oom: 21802f38ab2cSShachar Raindel if (old_page) 21812f38ab2cSShachar Raindel page_cache_release(old_page); 21822f38ab2cSShachar Raindel return VM_FAULT_OOM; 21832f38ab2cSShachar Raindel } 21842f38ab2cSShachar Raindel 2185dd906184SBoaz Harrosh /* 2186dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 2187dd906184SBoaz Harrosh * mapping 2188dd906184SBoaz Harrosh */ 2189dd906184SBoaz Harrosh static int wp_pfn_shared(struct mm_struct *mm, 2190dd906184SBoaz Harrosh struct vm_area_struct *vma, unsigned long address, 2191dd906184SBoaz Harrosh pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, 2192dd906184SBoaz Harrosh pmd_t *pmd) 2193dd906184SBoaz Harrosh { 2194dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 2195dd906184SBoaz Harrosh struct vm_fault vmf = { 2196dd906184SBoaz Harrosh .page = NULL, 2197dd906184SBoaz Harrosh .pgoff = linear_page_index(vma, address), 2198dd906184SBoaz Harrosh .virtual_address = (void __user *)(address & PAGE_MASK), 2199dd906184SBoaz Harrosh .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE, 2200dd906184SBoaz Harrosh }; 2201dd906184SBoaz Harrosh int ret; 2202dd906184SBoaz Harrosh 2203dd906184SBoaz Harrosh pte_unmap_unlock(page_table, ptl); 2204dd906184SBoaz Harrosh ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); 2205dd906184SBoaz Harrosh if (ret & VM_FAULT_ERROR) 2206dd906184SBoaz Harrosh return ret; 2207dd906184SBoaz Harrosh page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2208dd906184SBoaz Harrosh /* 2209dd906184SBoaz Harrosh * We might have raced with another page fault while we 2210dd906184SBoaz Harrosh * released the pte_offset_map_lock. 2211dd906184SBoaz Harrosh */ 2212dd906184SBoaz Harrosh if (!pte_same(*page_table, orig_pte)) { 2213dd906184SBoaz Harrosh pte_unmap_unlock(page_table, ptl); 2214dd906184SBoaz Harrosh return 0; 2215dd906184SBoaz Harrosh } 2216dd906184SBoaz Harrosh } 2217dd906184SBoaz Harrosh return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte, 2218dd906184SBoaz Harrosh NULL, 0, 0); 2219dd906184SBoaz Harrosh } 2220dd906184SBoaz Harrosh 222193e478d4SShachar Raindel static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma, 222293e478d4SShachar Raindel unsigned long address, pte_t *page_table, 222393e478d4SShachar Raindel pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte, 222493e478d4SShachar Raindel struct page *old_page) 222593e478d4SShachar Raindel __releases(ptl) 222693e478d4SShachar Raindel { 222793e478d4SShachar Raindel int page_mkwrite = 0; 222893e478d4SShachar Raindel 222993e478d4SShachar Raindel page_cache_get(old_page); 223093e478d4SShachar Raindel 223193e478d4SShachar Raindel /* 223293e478d4SShachar Raindel * Only catch write-faults on shared writable pages, 223393e478d4SShachar Raindel * read-only shared pages can get COWed by 223493e478d4SShachar Raindel * get_user_pages(.write=1, .force=1). 223593e478d4SShachar Raindel */ 223693e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 223793e478d4SShachar Raindel int tmp; 223893e478d4SShachar Raindel 223993e478d4SShachar Raindel pte_unmap_unlock(page_table, ptl); 224093e478d4SShachar Raindel tmp = do_page_mkwrite(vma, old_page, address); 224193e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 224293e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 224393e478d4SShachar Raindel page_cache_release(old_page); 224493e478d4SShachar Raindel return tmp; 224593e478d4SShachar Raindel } 224693e478d4SShachar Raindel /* 224793e478d4SShachar Raindel * Since we dropped the lock we need to revalidate 224893e478d4SShachar Raindel * the PTE as someone else may have changed it. If 224993e478d4SShachar Raindel * they did, we just return, as we can count on the 225093e478d4SShachar Raindel * MMU to tell us if they didn't also make it writable. 225193e478d4SShachar Raindel */ 225293e478d4SShachar Raindel page_table = pte_offset_map_lock(mm, pmd, address, 225393e478d4SShachar Raindel &ptl); 225493e478d4SShachar Raindel if (!pte_same(*page_table, orig_pte)) { 225593e478d4SShachar Raindel unlock_page(old_page); 225693e478d4SShachar Raindel pte_unmap_unlock(page_table, ptl); 225793e478d4SShachar Raindel page_cache_release(old_page); 225893e478d4SShachar Raindel return 0; 225993e478d4SShachar Raindel } 226093e478d4SShachar Raindel page_mkwrite = 1; 226193e478d4SShachar Raindel } 226293e478d4SShachar Raindel 226393e478d4SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 226493e478d4SShachar Raindel orig_pte, old_page, page_mkwrite, 1); 226593e478d4SShachar Raindel } 226693e478d4SShachar Raindel 22672f38ab2cSShachar Raindel /* 22681da177e4SLinus Torvalds * This routine handles present pages, when users try to write 22691da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 22701da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 22711da177e4SLinus Torvalds * 22721da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 22731da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 22741da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 22751da177e4SLinus Torvalds * COW. 22761da177e4SLinus Torvalds * 22771da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 22781da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 22791da177e4SLinus Torvalds * and potentially makes it more efficient. 22801da177e4SLinus Torvalds * 22818f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 22828f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 22838f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 22841da177e4SLinus Torvalds */ 22851da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 228665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 22878f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2288e6219ec8SNamhyung Kim __releases(ptl) 22891da177e4SLinus Torvalds { 22902f38ab2cSShachar Raindel struct page *old_page; 22911da177e4SLinus Torvalds 22926aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2293251b97f5SPeter Zijlstra if (!old_page) { 2294251b97f5SPeter Zijlstra /* 229564e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 229664e45507SPeter Feiner * VM_PFNMAP VMA. 2297251b97f5SPeter Zijlstra * 2298251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2299dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 2300251b97f5SPeter Zijlstra */ 2301251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2302251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2303dd906184SBoaz Harrosh return wp_pfn_shared(mm, vma, address, page_table, ptl, 2304dd906184SBoaz Harrosh orig_pte, pmd); 23052f38ab2cSShachar Raindel 23062f38ab2cSShachar Raindel pte_unmap_unlock(page_table, ptl); 23072f38ab2cSShachar Raindel return wp_page_copy(mm, vma, address, page_table, pmd, 23082f38ab2cSShachar Raindel orig_pte, old_page); 2309251b97f5SPeter Zijlstra } 23101da177e4SLinus Torvalds 2311d08b3851SPeter Zijlstra /* 2312ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2313ee6a6457SPeter Zijlstra * not dirty accountable. 2314d08b3851SPeter Zijlstra */ 23159a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2316ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2317ab967d86SHugh Dickins page_cache_get(old_page); 2318ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2319ab967d86SHugh Dickins lock_page(old_page); 2320ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2321ab967d86SHugh Dickins &ptl); 2322ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2323ab967d86SHugh Dickins unlock_page(old_page); 232428766805SShachar Raindel pte_unmap_unlock(page_table, ptl); 232528766805SShachar Raindel page_cache_release(old_page); 232628766805SShachar Raindel return 0; 2327ab967d86SHugh Dickins } 2328ab967d86SHugh Dickins page_cache_release(old_page); 2329ab967d86SHugh Dickins } 2330b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2331c44b6743SRik van Riel /* 2332c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2333c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2334c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2335c44b6743SRik van Riel */ 2336c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2337ee6a6457SPeter Zijlstra unlock_page(old_page); 23384e047f89SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 23394e047f89SShachar Raindel orig_pte, old_page, 0, 0); 2340b009c024SMichel Lespinasse } 2341b009c024SMichel Lespinasse unlock_page(old_page); 2342ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2343d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 234493e478d4SShachar Raindel return wp_page_shared(mm, vma, address, page_table, pmd, 234593e478d4SShachar Raindel ptl, orig_pte, old_page); 23461da177e4SLinus Torvalds } 23471da177e4SLinus Torvalds 23481da177e4SLinus Torvalds /* 23491da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 23501da177e4SLinus Torvalds */ 23511da177e4SLinus Torvalds page_cache_get(old_page); 235228766805SShachar Raindel 23538f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 23542f38ab2cSShachar Raindel return wp_page_copy(mm, vma, address, page_table, pmd, 23552f38ab2cSShachar Raindel orig_pte, old_page); 23561da177e4SLinus Torvalds } 23571da177e4SLinus Torvalds 235897a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 23591da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 23601da177e4SLinus Torvalds struct zap_details *details) 23611da177e4SLinus Torvalds { 2362f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 23631da177e4SLinus Torvalds } 23641da177e4SLinus Torvalds 23656b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 23661da177e4SLinus Torvalds struct zap_details *details) 23671da177e4SLinus Torvalds { 23681da177e4SLinus Torvalds struct vm_area_struct *vma; 23691da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 23701da177e4SLinus Torvalds 23716b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 23721da177e4SLinus Torvalds details->first_index, details->last_index) { 23731da177e4SLinus Torvalds 23741da177e4SLinus Torvalds vba = vma->vm_pgoff; 2375d6e93217SLibin vea = vba + vma_pages(vma) - 1; 23761da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 23771da177e4SLinus Torvalds zba = details->first_index; 23781da177e4SLinus Torvalds if (zba < vba) 23791da177e4SLinus Torvalds zba = vba; 23801da177e4SLinus Torvalds zea = details->last_index; 23811da177e4SLinus Torvalds if (zea > vea) 23821da177e4SLinus Torvalds zea = vea; 23831da177e4SLinus Torvalds 238497a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 23851da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 23861da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 238797a89413SPeter Zijlstra details); 23881da177e4SLinus Torvalds } 23891da177e4SLinus Torvalds } 23901da177e4SLinus Torvalds 23911da177e4SLinus Torvalds /** 23928a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 23938a5f14a2SKirill A. Shutemov * address_space corresponding to the specified page range in the underlying 23948a5f14a2SKirill A. Shutemov * file. 23958a5f14a2SKirill A. Shutemov * 23963d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 23971da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 23981da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 239925d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 24001da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 24011da177e4SLinus Torvalds * partial pages. 24021da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 24031da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 24041da177e4SLinus Torvalds * end of the file. 24051da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 24061da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 24071da177e4SLinus Torvalds */ 24081da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 24091da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 24101da177e4SLinus Torvalds { 24111da177e4SLinus Torvalds struct zap_details details; 24121da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 24131da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds /* Check for overflow. */ 24161da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 24171da177e4SLinus Torvalds long long holeend = 24181da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 24191da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 24201da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 24211da177e4SLinus Torvalds } 24221da177e4SLinus Torvalds 24231da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 24241da177e4SLinus Torvalds details.first_index = hba; 24251da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 24261da177e4SLinus Torvalds if (details.last_index < details.first_index) 24271da177e4SLinus Torvalds details.last_index = ULONG_MAX; 24281da177e4SLinus Torvalds 24290f90cc66SRoss Zwisler 24300f90cc66SRoss Zwisler /* DAX uses i_mmap_lock to serialise file truncate vs page fault */ 243148ec833bSKirill A. Shutemov i_mmap_lock_write(mapping); 24326b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 24331da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 243448ec833bSKirill A. Shutemov i_mmap_unlock_write(mapping); 24351da177e4SLinus Torvalds } 24361da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds /* 24398f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 24408f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 24419a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 24429a95f3cfSPaul Cassella * 24439a95f3cfSPaul Cassella * We return with the mmap_sem locked or unlocked in the same cases 24449a95f3cfSPaul Cassella * as does filemap_fault(). 24451da177e4SLinus Torvalds */ 244665500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 244765500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 244830c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 24491da177e4SLinus Torvalds { 24508f4e2101SHugh Dickins spinlock_t *ptl; 245156f31801SHugh Dickins struct page *page, *swapcache; 245200501b53SJohannes Weiner struct mem_cgroup *memcg; 245365500d23SHugh Dickins swp_entry_t entry; 24541da177e4SLinus Torvalds pte_t pte; 2455d065bd81SMichel Lespinasse int locked; 2456ad8c2ee8SRik van Riel int exclusive = 0; 245783c54070SNick Piggin int ret = 0; 24581da177e4SLinus Torvalds 24594c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 24608f4e2101SHugh Dickins goto out; 246165500d23SHugh Dickins 246265500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 2463d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 24640697212aSChristoph Lameter if (is_migration_entry(entry)) { 24650697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 2466d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 2467d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2468d1737fdbSAndi Kleen } else { 2469d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 2470d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 2471d1737fdbSAndi Kleen } 24720697212aSChristoph Lameter goto out; 24730697212aSChristoph Lameter } 24740ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 24751da177e4SLinus Torvalds page = lookup_swap_cache(entry); 24761da177e4SLinus Torvalds if (!page) { 247702098feaSHugh Dickins page = swapin_readahead(entry, 247802098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 24791da177e4SLinus Torvalds if (!page) { 24801da177e4SLinus Torvalds /* 24818f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 24828f4e2101SHugh Dickins * while we released the pte lock. 24831da177e4SLinus Torvalds */ 24848f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 24851da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 24861da177e4SLinus Torvalds ret = VM_FAULT_OOM; 24870ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 248865500d23SHugh Dickins goto unlock; 24891da177e4SLinus Torvalds } 24901da177e4SLinus Torvalds 24911da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 24921da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2493f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 2494456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 2495d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 249671f72525SWu Fengguang /* 249771f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 249871f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 249971f72525SWu Fengguang */ 2500d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2501d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 250256f31801SHugh Dickins swapcache = page; 25034779cb31SAndi Kleen goto out_release; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds 250656f31801SHugh Dickins swapcache = page; 2507d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 2508e709ffd6SRik van Riel 250920a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 2510d065bd81SMichel Lespinasse if (!locked) { 2511d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 2512d065bd81SMichel Lespinasse goto out_release; 2513d065bd81SMichel Lespinasse } 25141da177e4SLinus Torvalds 25154969c119SAndrea Arcangeli /* 251631c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 251731c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 251831c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 251931c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 25204969c119SAndrea Arcangeli */ 252131c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 25224969c119SAndrea Arcangeli goto out_page; 25234969c119SAndrea Arcangeli 2524cbf86cfeSHugh Dickins page = ksm_might_need_to_copy(page, vma, address); 25254969c119SAndrea Arcangeli if (unlikely(!page)) { 25265ad64688SHugh Dickins ret = VM_FAULT_OOM; 25274969c119SAndrea Arcangeli page = swapcache; 25284969c119SAndrea Arcangeli goto out_page; 25294969c119SAndrea Arcangeli } 25305ad64688SHugh Dickins 2531f627c2f5SKirill A. Shutemov if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) { 2532073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 2533bc43f75cSJohannes Weiner goto out_page; 2534073e587eSKAMEZAWA Hiroyuki } 2535073e587eSKAMEZAWA Hiroyuki 25361da177e4SLinus Torvalds /* 25378f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 25381da177e4SLinus Torvalds */ 25398f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 25409e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 2541b8107480SKirill Korotaev goto out_nomap; 2542b8107480SKirill Korotaev 2543b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 2544b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 2545b8107480SKirill Korotaev goto out_nomap; 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds 25488c7c6e34SKAMEZAWA Hiroyuki /* 25498c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 25508c7c6e34SKAMEZAWA Hiroyuki * 25518c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 25528c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 25538c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 25548c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 25558c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 25568c7c6e34SKAMEZAWA Hiroyuki */ 25571da177e4SLinus Torvalds 255834e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2559b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 25601da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 256130c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 25621da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 256330c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 25649a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 2565d281ee61SKirill A. Shutemov exclusive = RMAP_EXCLUSIVE; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds flush_icache_page(vma, page); 2568179ef71cSCyrill Gorcunov if (pte_swp_soft_dirty(orig_pte)) 2569179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 25701da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 257100501b53SJohannes Weiner if (page == swapcache) { 2572ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 2573f627c2f5SKirill A. Shutemov mem_cgroup_commit_charge(page, memcg, true, false); 257400501b53SJohannes Weiner } else { /* ksm created a completely new copy */ 2575d281ee61SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address, false); 2576f627c2f5SKirill A. Shutemov mem_cgroup_commit_charge(page, memcg, false, false); 257700501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 257800501b53SJohannes Weiner } 25791da177e4SLinus Torvalds 2580c475a8abSHugh Dickins swap_free(entry); 2581b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 2582a2c43eedSHugh Dickins try_to_free_swap(page); 2583c475a8abSHugh Dickins unlock_page(page); 258456f31801SHugh Dickins if (page != swapcache) { 25854969c119SAndrea Arcangeli /* 25864969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 25874969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 25884969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 25894969c119SAndrea Arcangeli * further safety release the lock after the swap_free 25904969c119SAndrea Arcangeli * so that the swap count won't change under a 25914969c119SAndrea Arcangeli * parallel locked swapcache. 25924969c119SAndrea Arcangeli */ 25934969c119SAndrea Arcangeli unlock_page(swapcache); 25944969c119SAndrea Arcangeli page_cache_release(swapcache); 25954969c119SAndrea Arcangeli } 2596c475a8abSHugh Dickins 259730c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 259861469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 259961469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 260061469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 26011da177e4SLinus Torvalds goto out; 26021da177e4SLinus Torvalds } 26031da177e4SLinus Torvalds 26041da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 26054b3073e1SRussell King update_mmu_cache(vma, address, page_table); 260665500d23SHugh Dickins unlock: 26078f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 26081da177e4SLinus Torvalds out: 26091da177e4SLinus Torvalds return ret; 2610b8107480SKirill Korotaev out_nomap: 2611f627c2f5SKirill A. Shutemov mem_cgroup_cancel_charge(page, memcg, false); 26128f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2613bc43f75cSJohannes Weiner out_page: 2614b8107480SKirill Korotaev unlock_page(page); 26154779cb31SAndi Kleen out_release: 2616b8107480SKirill Korotaev page_cache_release(page); 261756f31801SHugh Dickins if (page != swapcache) { 26184969c119SAndrea Arcangeli unlock_page(swapcache); 26194969c119SAndrea Arcangeli page_cache_release(swapcache); 26204969c119SAndrea Arcangeli } 262165500d23SHugh Dickins return ret; 26221da177e4SLinus Torvalds } 26231da177e4SLinus Torvalds 26241da177e4SLinus Torvalds /* 26258ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 26268ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 2627320b2b8dSLinus Torvalds * doesn't hit another vma. 2628320b2b8dSLinus Torvalds */ 2629320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 2630320b2b8dSLinus Torvalds { 2631320b2b8dSLinus Torvalds address &= PAGE_MASK; 2632320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 26330e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 2634320b2b8dSLinus Torvalds 26350e8e50e2SLinus Torvalds /* 26360e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 26370e8e50e2SLinus Torvalds * 26380e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 26390e8e50e2SLinus Torvalds * that has gotten split.. 26400e8e50e2SLinus Torvalds */ 26410e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 26420e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 26430e8e50e2SLinus Torvalds 2644fee7e49dSLinus Torvalds return expand_downwards(vma, address - PAGE_SIZE); 2645320b2b8dSLinus Torvalds } 26468ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 26478ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 26488ca3eb08SLuck, Tony 26498ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 26508ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 26518ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 26528ca3eb08SLuck, Tony 2653fee7e49dSLinus Torvalds return expand_upwards(vma, address + PAGE_SIZE); 26548ca3eb08SLuck, Tony } 2655320b2b8dSLinus Torvalds return 0; 2656320b2b8dSLinus Torvalds } 2657320b2b8dSLinus Torvalds 2658320b2b8dSLinus Torvalds /* 26598f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 26608f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 26618f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 26621da177e4SLinus Torvalds */ 266365500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 266465500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 266530c9f3a9SLinus Torvalds unsigned int flags) 26661da177e4SLinus Torvalds { 266700501b53SJohannes Weiner struct mem_cgroup *memcg; 26688f4e2101SHugh Dickins struct page *page; 26698f4e2101SHugh Dickins spinlock_t *ptl; 26701da177e4SLinus Torvalds pte_t entry; 26711da177e4SLinus Torvalds 26725528f913SLinus Torvalds pte_unmap(page_table); 2673320b2b8dSLinus Torvalds 26746b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 26756b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 26766b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 26776b7339f4SKirill A. Shutemov 267811ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 267911ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 26809c145c56SLinus Torvalds return VM_FAULT_SIGSEGV; 268111ac5524SLinus Torvalds 268211ac5524SLinus Torvalds /* Use the zero-page for reads */ 2683593befa6SDominik Dingel if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) { 268462eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 268562eede62SHugh Dickins vma->vm_page_prot)); 268611ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2687a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 2688a13ea5b7SHugh Dickins goto unlock; 26896b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 26906b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 26916b251fc9SAndrea Arcangeli pte_unmap_unlock(page_table, ptl); 26926b251fc9SAndrea Arcangeli return handle_userfault(vma, address, flags, 26936b251fc9SAndrea Arcangeli VM_UFFD_MISSING); 26946b251fc9SAndrea Arcangeli } 2695a13ea5b7SHugh Dickins goto setpte; 2696a13ea5b7SHugh Dickins } 2697a13ea5b7SHugh Dickins 26981da177e4SLinus Torvalds /* Allocate our own private page. */ 26991da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 270065500d23SHugh Dickins goto oom; 2701769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 27021da177e4SLinus Torvalds if (!page) 270365500d23SHugh Dickins goto oom; 2704eb3c24f3SMel Gorman 2705f627c2f5SKirill A. Shutemov if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) 2706eb3c24f3SMel Gorman goto oom_free_page; 2707eb3c24f3SMel Gorman 270852f37629SMinchan Kim /* 270952f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 271052f37629SMinchan Kim * preceeding stores to the page contents become visible before 271152f37629SMinchan Kim * the set_pte_at() write. 271252f37629SMinchan Kim */ 27130ed361deSNick Piggin __SetPageUptodate(page); 27141da177e4SLinus Torvalds 271565500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 27161ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 27171ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 27188f4e2101SHugh Dickins 27198f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 27201c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 27218f4e2101SHugh Dickins goto release; 27229ba69294SHugh Dickins 27236b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 27246b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 27256b251fc9SAndrea Arcangeli pte_unmap_unlock(page_table, ptl); 2726f627c2f5SKirill A. Shutemov mem_cgroup_cancel_charge(page, memcg, false); 27276b251fc9SAndrea Arcangeli page_cache_release(page); 27286b251fc9SAndrea Arcangeli return handle_userfault(vma, address, flags, 27296b251fc9SAndrea Arcangeli VM_UFFD_MISSING); 27306b251fc9SAndrea Arcangeli } 27316b251fc9SAndrea Arcangeli 273234e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2733d281ee61SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address, false); 2734f627c2f5SKirill A. Shutemov mem_cgroup_commit_charge(page, memcg, false, false); 273500501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 2736a13ea5b7SHugh Dickins setpte: 273765500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 27381da177e4SLinus Torvalds 27391da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 27404b3073e1SRussell King update_mmu_cache(vma, address, page_table); 274165500d23SHugh Dickins unlock: 27428f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 274383c54070SNick Piggin return 0; 27448f4e2101SHugh Dickins release: 2745f627c2f5SKirill A. Shutemov mem_cgroup_cancel_charge(page, memcg, false); 27468f4e2101SHugh Dickins page_cache_release(page); 27478f4e2101SHugh Dickins goto unlock; 27488a9f3ccdSBalbir Singh oom_free_page: 27496dbf6d3bSHugh Dickins page_cache_release(page); 275065500d23SHugh Dickins oom: 27511da177e4SLinus Torvalds return VM_FAULT_OOM; 27521da177e4SLinus Torvalds } 27531da177e4SLinus Torvalds 27549a95f3cfSPaul Cassella /* 27559a95f3cfSPaul Cassella * The mmap_sem must have been held on entry, and may have been 27569a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 27579a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 27589a95f3cfSPaul Cassella */ 27597eae74afSKirill A. Shutemov static int __do_fault(struct vm_area_struct *vma, unsigned long address, 27602e4cdab0SMatthew Wilcox pgoff_t pgoff, unsigned int flags, 27612e4cdab0SMatthew Wilcox struct page *cow_page, struct page **page) 27627eae74afSKirill A. Shutemov { 27637eae74afSKirill A. Shutemov struct vm_fault vmf; 27647eae74afSKirill A. Shutemov int ret; 27657eae74afSKirill A. Shutemov 27667eae74afSKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 27677eae74afSKirill A. Shutemov vmf.pgoff = pgoff; 27687eae74afSKirill A. Shutemov vmf.flags = flags; 27697eae74afSKirill A. Shutemov vmf.page = NULL; 2770c20cd45eSMichal Hocko vmf.gfp_mask = __get_fault_gfp_mask(vma); 27712e4cdab0SMatthew Wilcox vmf.cow_page = cow_page; 27727eae74afSKirill A. Shutemov 27737eae74afSKirill A. Shutemov ret = vma->vm_ops->fault(vma, &vmf); 27747eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 27757eae74afSKirill A. Shutemov return ret; 27762e4cdab0SMatthew Wilcox if (!vmf.page) 27772e4cdab0SMatthew Wilcox goto out; 27787eae74afSKirill A. Shutemov 27797eae74afSKirill A. Shutemov if (unlikely(PageHWPoison(vmf.page))) { 27807eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 27817eae74afSKirill A. Shutemov unlock_page(vmf.page); 27827eae74afSKirill A. Shutemov page_cache_release(vmf.page); 27837eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 27847eae74afSKirill A. Shutemov } 27857eae74afSKirill A. Shutemov 27867eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 27877eae74afSKirill A. Shutemov lock_page(vmf.page); 27887eae74afSKirill A. Shutemov else 27897eae74afSKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); 27907eae74afSKirill A. Shutemov 27912e4cdab0SMatthew Wilcox out: 27927eae74afSKirill A. Shutemov *page = vmf.page; 27937eae74afSKirill A. Shutemov return ret; 27947eae74afSKirill A. Shutemov } 27957eae74afSKirill A. Shutemov 27968c6e50b0SKirill A. Shutemov /** 27978c6e50b0SKirill A. Shutemov * do_set_pte - setup new PTE entry for given page and add reverse page mapping. 27988c6e50b0SKirill A. Shutemov * 27998c6e50b0SKirill A. Shutemov * @vma: virtual memory area 28008c6e50b0SKirill A. Shutemov * @address: user virtual address 28018c6e50b0SKirill A. Shutemov * @page: page to map 28028c6e50b0SKirill A. Shutemov * @pte: pointer to target page table entry 28038c6e50b0SKirill A. Shutemov * @write: true, if new entry is writable 28048c6e50b0SKirill A. Shutemov * @anon: true, if it's anonymous page 28058c6e50b0SKirill A. Shutemov * 28068c6e50b0SKirill A. Shutemov * Caller must hold page table lock relevant for @pte. 28078c6e50b0SKirill A. Shutemov * 28088c6e50b0SKirill A. Shutemov * Target users are page handler itself and implementations of 28098c6e50b0SKirill A. Shutemov * vm_ops->map_pages. 28108c6e50b0SKirill A. Shutemov */ 28118c6e50b0SKirill A. Shutemov void do_set_pte(struct vm_area_struct *vma, unsigned long address, 28123bb97794SKirill A. Shutemov struct page *page, pte_t *pte, bool write, bool anon) 28133bb97794SKirill A. Shutemov { 28143bb97794SKirill A. Shutemov pte_t entry; 28153bb97794SKirill A. Shutemov 28163bb97794SKirill A. Shutemov flush_icache_page(vma, page); 28173bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 28183bb97794SKirill A. Shutemov if (write) 28193bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 28203bb97794SKirill A. Shutemov if (anon) { 28213bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 2822d281ee61SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address, false); 28233bb97794SKirill A. Shutemov } else { 2824eca56ff9SJerome Marchand inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); 28253bb97794SKirill A. Shutemov page_add_file_rmap(page); 28263bb97794SKirill A. Shutemov } 28273bb97794SKirill A. Shutemov set_pte_at(vma->vm_mm, address, pte, entry); 28283bb97794SKirill A. Shutemov 28293bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 28303bb97794SKirill A. Shutemov update_mmu_cache(vma, address, pte); 28313bb97794SKirill A. Shutemov } 28323bb97794SKirill A. Shutemov 28333a91053aSKirill A. Shutemov static unsigned long fault_around_bytes __read_mostly = 28343a91053aSKirill A. Shutemov rounddown_pow_of_two(65536); 2835a9b0f861SKirill A. Shutemov 28361592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 2837a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 28381592eef0SKirill A. Shutemov { 2839a9b0f861SKirill A. Shutemov *val = fault_around_bytes; 28401592eef0SKirill A. Shutemov return 0; 28411592eef0SKirill A. Shutemov } 28421592eef0SKirill A. Shutemov 2843b4903d6eSAndrey Ryabinin /* 2844b4903d6eSAndrey Ryabinin * fault_around_pages() and fault_around_mask() expects fault_around_bytes 2845b4903d6eSAndrey Ryabinin * rounded down to nearest page order. It's what do_fault_around() expects to 2846b4903d6eSAndrey Ryabinin * see. 2847b4903d6eSAndrey Ryabinin */ 2848a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 28491592eef0SKirill A. Shutemov { 2850a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 28511592eef0SKirill A. Shutemov return -EINVAL; 2852b4903d6eSAndrey Ryabinin if (val > PAGE_SIZE) 2853b4903d6eSAndrey Ryabinin fault_around_bytes = rounddown_pow_of_two(val); 2854b4903d6eSAndrey Ryabinin else 2855b4903d6eSAndrey Ryabinin fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ 28561592eef0SKirill A. Shutemov return 0; 28571592eef0SKirill A. Shutemov } 2858a9b0f861SKirill A. Shutemov DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, 2859a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 28601592eef0SKirill A. Shutemov 28611592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 28621592eef0SKirill A. Shutemov { 28631592eef0SKirill A. Shutemov void *ret; 28641592eef0SKirill A. Shutemov 2865a9b0f861SKirill A. Shutemov ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, 2866a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 28671592eef0SKirill A. Shutemov if (!ret) 2868a9b0f861SKirill A. Shutemov pr_warn("Failed to create fault_around_bytes in debugfs"); 28691592eef0SKirill A. Shutemov return 0; 28701592eef0SKirill A. Shutemov } 28711592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 28721592eef0SKirill A. Shutemov #endif 28738c6e50b0SKirill A. Shutemov 28741fdb412bSKirill A. Shutemov /* 28751fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 28761fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 28771fdb412bSKirill A. Shutemov * faults to handle. 28781fdb412bSKirill A. Shutemov * 28791fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 28801fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 28811fdb412bSKirill A. Shutemov * 28821fdb412bSKirill A. Shutemov * This function is called with the page table lock taken. In the split ptlock 28831fdb412bSKirill A. Shutemov * case the page table lock only protects only those entries which belong to 28841fdb412bSKirill A. Shutemov * the page table corresponding to the fault address. 28851fdb412bSKirill A. Shutemov * 28861fdb412bSKirill A. Shutemov * This function doesn't cross the VMA boundaries, in order to call map_pages() 28871fdb412bSKirill A. Shutemov * only once. 28881fdb412bSKirill A. Shutemov * 28891fdb412bSKirill A. Shutemov * fault_around_pages() defines how many pages we'll try to map. 28901fdb412bSKirill A. Shutemov * do_fault_around() expects it to return a power of two less than or equal to 28911fdb412bSKirill A. Shutemov * PTRS_PER_PTE. 28921fdb412bSKirill A. Shutemov * 28931fdb412bSKirill A. Shutemov * The virtual address of the area that we map is naturally aligned to the 28941fdb412bSKirill A. Shutemov * fault_around_pages() value (and therefore to page order). This way it's 28951fdb412bSKirill A. Shutemov * easier to guarantee that we don't cross page table boundaries. 28961fdb412bSKirill A. Shutemov */ 28978c6e50b0SKirill A. Shutemov static void do_fault_around(struct vm_area_struct *vma, unsigned long address, 28988c6e50b0SKirill A. Shutemov pte_t *pte, pgoff_t pgoff, unsigned int flags) 28998c6e50b0SKirill A. Shutemov { 2900aecd6f44SKirill A. Shutemov unsigned long start_addr, nr_pages, mask; 29018c6e50b0SKirill A. Shutemov pgoff_t max_pgoff; 29028c6e50b0SKirill A. Shutemov struct vm_fault vmf; 29038c6e50b0SKirill A. Shutemov int off; 29048c6e50b0SKirill A. Shutemov 29054db0c3c2SJason Low nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; 2906aecd6f44SKirill A. Shutemov mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; 2907aecd6f44SKirill A. Shutemov 2908aecd6f44SKirill A. Shutemov start_addr = max(address & mask, vma->vm_start); 29098c6e50b0SKirill A. Shutemov off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 29108c6e50b0SKirill A. Shutemov pte -= off; 29118c6e50b0SKirill A. Shutemov pgoff -= off; 29128c6e50b0SKirill A. Shutemov 29138c6e50b0SKirill A. Shutemov /* 29148c6e50b0SKirill A. Shutemov * max_pgoff is either end of page table or end of vma 2915850e9c69SKirill A. Shutemov * or fault_around_pages() from pgoff, depending what is nearest. 29168c6e50b0SKirill A. Shutemov */ 29178c6e50b0SKirill A. Shutemov max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + 29188c6e50b0SKirill A. Shutemov PTRS_PER_PTE - 1; 29198c6e50b0SKirill A. Shutemov max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, 2920aecd6f44SKirill A. Shutemov pgoff + nr_pages - 1); 29218c6e50b0SKirill A. Shutemov 29228c6e50b0SKirill A. Shutemov /* Check if it makes any sense to call ->map_pages */ 29238c6e50b0SKirill A. Shutemov while (!pte_none(*pte)) { 29248c6e50b0SKirill A. Shutemov if (++pgoff > max_pgoff) 29258c6e50b0SKirill A. Shutemov return; 29268c6e50b0SKirill A. Shutemov start_addr += PAGE_SIZE; 29278c6e50b0SKirill A. Shutemov if (start_addr >= vma->vm_end) 29288c6e50b0SKirill A. Shutemov return; 29298c6e50b0SKirill A. Shutemov pte++; 29308c6e50b0SKirill A. Shutemov } 29318c6e50b0SKirill A. Shutemov 29328c6e50b0SKirill A. Shutemov vmf.virtual_address = (void __user *) start_addr; 29338c6e50b0SKirill A. Shutemov vmf.pte = pte; 29348c6e50b0SKirill A. Shutemov vmf.pgoff = pgoff; 29358c6e50b0SKirill A. Shutemov vmf.max_pgoff = max_pgoff; 29368c6e50b0SKirill A. Shutemov vmf.flags = flags; 2937c20cd45eSMichal Hocko vmf.gfp_mask = __get_fault_gfp_mask(vma); 29388c6e50b0SKirill A. Shutemov vma->vm_ops->map_pages(vma, &vmf); 29398c6e50b0SKirill A. Shutemov } 29408c6e50b0SKirill A. Shutemov 2941e655fb29SKirill A. Shutemov static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2942e655fb29SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2943e655fb29SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2944e655fb29SKirill A. Shutemov { 2945e655fb29SKirill A. Shutemov struct page *fault_page; 2946e655fb29SKirill A. Shutemov spinlock_t *ptl; 29473bb97794SKirill A. Shutemov pte_t *pte; 29488c6e50b0SKirill A. Shutemov int ret = 0; 29498c6e50b0SKirill A. Shutemov 29508c6e50b0SKirill A. Shutemov /* 29518c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 29528c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 29538c6e50b0SKirill A. Shutemov * something). 29548c6e50b0SKirill A. Shutemov */ 29559b4bdd2fSKirill A. Shutemov if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { 29568c6e50b0SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 29578c6e50b0SKirill A. Shutemov do_fault_around(vma, address, pte, pgoff, flags); 29588c6e50b0SKirill A. Shutemov if (!pte_same(*pte, orig_pte)) 29598c6e50b0SKirill A. Shutemov goto unlock_out; 29608c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 29618c6e50b0SKirill A. Shutemov } 2962e655fb29SKirill A. Shutemov 29632e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 2964e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2965e655fb29SKirill A. Shutemov return ret; 2966e655fb29SKirill A. Shutemov 2967e655fb29SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2968e655fb29SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2969e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2970e655fb29SKirill A. Shutemov unlock_page(fault_page); 2971e655fb29SKirill A. Shutemov page_cache_release(fault_page); 2972e655fb29SKirill A. Shutemov return ret; 2973e655fb29SKirill A. Shutemov } 29743bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, false, false); 2975e655fb29SKirill A. Shutemov unlock_page(fault_page); 29768c6e50b0SKirill A. Shutemov unlock_out: 29778c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2978e655fb29SKirill A. Shutemov return ret; 2979e655fb29SKirill A. Shutemov } 2980e655fb29SKirill A. Shutemov 2981ec47c3b9SKirill A. Shutemov static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2982ec47c3b9SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2983ec47c3b9SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2984ec47c3b9SKirill A. Shutemov { 2985ec47c3b9SKirill A. Shutemov struct page *fault_page, *new_page; 298600501b53SJohannes Weiner struct mem_cgroup *memcg; 2987ec47c3b9SKirill A. Shutemov spinlock_t *ptl; 29883bb97794SKirill A. Shutemov pte_t *pte; 2989ec47c3b9SKirill A. Shutemov int ret; 2990ec47c3b9SKirill A. Shutemov 2991ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 2992ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2993ec47c3b9SKirill A. Shutemov 2994ec47c3b9SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 2995ec47c3b9SKirill A. Shutemov if (!new_page) 2996ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2997ec47c3b9SKirill A. Shutemov 2998f627c2f5SKirill A. Shutemov if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) { 2999ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3000ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3001ec47c3b9SKirill A. Shutemov } 3002ec47c3b9SKirill A. Shutemov 30032e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page); 3004ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3005ec47c3b9SKirill A. Shutemov goto uncharge_out; 3006ec47c3b9SKirill A. Shutemov 30072e4cdab0SMatthew Wilcox if (fault_page) 3008ec47c3b9SKirill A. Shutemov copy_user_highpage(new_page, fault_page, address, vma); 3009ec47c3b9SKirill A. Shutemov __SetPageUptodate(new_page); 3010ec47c3b9SKirill A. Shutemov 3011ec47c3b9SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3012ec47c3b9SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3013ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 30142e4cdab0SMatthew Wilcox if (fault_page) { 3015ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3016ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 30172e4cdab0SMatthew Wilcox } else { 30182e4cdab0SMatthew Wilcox /* 30192e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 30200df9d41aSYigal Korman * i_mmap_lock for read to protect against truncate. 30212e4cdab0SMatthew Wilcox */ 30220df9d41aSYigal Korman i_mmap_unlock_read(vma->vm_file->f_mapping); 30232e4cdab0SMatthew Wilcox } 3024ec47c3b9SKirill A. Shutemov goto uncharge_out; 3025ec47c3b9SKirill A. Shutemov } 30263bb97794SKirill A. Shutemov do_set_pte(vma, address, new_page, pte, true, true); 3027f627c2f5SKirill A. Shutemov mem_cgroup_commit_charge(new_page, memcg, false, false); 302800501b53SJohannes Weiner lru_cache_add_active_or_unevictable(new_page, vma); 3029ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 30302e4cdab0SMatthew Wilcox if (fault_page) { 3031ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3032ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 30332e4cdab0SMatthew Wilcox } else { 30342e4cdab0SMatthew Wilcox /* 30352e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 30360df9d41aSYigal Korman * i_mmap_lock for read to protect against truncate. 30372e4cdab0SMatthew Wilcox */ 30380df9d41aSYigal Korman i_mmap_unlock_read(vma->vm_file->f_mapping); 30392e4cdab0SMatthew Wilcox } 3040ec47c3b9SKirill A. Shutemov return ret; 3041ec47c3b9SKirill A. Shutemov uncharge_out: 3042f627c2f5SKirill A. Shutemov mem_cgroup_cancel_charge(new_page, memcg, false); 3043ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3044ec47c3b9SKirill A. Shutemov return ret; 3045ec47c3b9SKirill A. Shutemov } 3046ec47c3b9SKirill A. Shutemov 3047f0c6d4d2SKirill A. Shutemov static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, 304816abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 304954cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 30501da177e4SLinus Torvalds { 3051f0c6d4d2SKirill A. Shutemov struct page *fault_page; 3052f0c6d4d2SKirill A. Shutemov struct address_space *mapping; 30538f4e2101SHugh Dickins spinlock_t *ptl; 30543bb97794SKirill A. Shutemov pte_t *pte; 3055f0c6d4d2SKirill A. Shutemov int dirtied = 0; 3056f0c6d4d2SKirill A. Shutemov int ret, tmp; 30571d65f86dSKAMEZAWA Hiroyuki 30582e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 30597eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3060f0c6d4d2SKirill A. Shutemov return ret; 30611da177e4SLinus Torvalds 30621da177e4SLinus Torvalds /* 3063f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 3064f0c6d4d2SKirill A. Shutemov * about to become writable 30651da177e4SLinus Torvalds */ 3066fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 3067f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3068fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, fault_page, address); 3069fb09a464SKirill A. Shutemov if (unlikely(!tmp || 3070fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3071f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3072f0c6d4d2SKirill A. Shutemov return tmp; 307369676147SMark Fasheh } 3074d0217ac0SNick Piggin } 3075fb09a464SKirill A. Shutemov 3076f0c6d4d2SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3077f0c6d4d2SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3078f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3079f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3080f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3081f0c6d4d2SKirill A. Shutemov return ret; 30829637a5efSDavid Howells } 30833bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, true, false); 3084f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3085d00806b1SNick Piggin 3086f0c6d4d2SKirill A. Shutemov if (set_page_dirty(fault_page)) 308741c4d25fSJan Kara dirtied = 1; 3088d82fa87dSAndrew Morton /* 3089d82fa87dSAndrew Morton * Take a local copy of the address_space - page.mapping may be zeroed 3090d82fa87dSAndrew Morton * by truncate after unlock_page(). The address_space itself remains 3091d82fa87dSAndrew Morton * pinned by vma->vm_file's reference. We rely on unlock_page()'s 3092d82fa87dSAndrew Morton * release semantics to prevent the compiler from undoing this copying. 3093d82fa87dSAndrew Morton */ 30941c290f64SKirill A. Shutemov mapping = page_rmapping(fault_page); 3095f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3096f0c6d4d2SKirill A. Shutemov if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { 3097b827e496SNick Piggin /* 3098b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3099b827e496SNick Piggin * dirty their pages 3100b827e496SNick Piggin */ 3101b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3102d08b3851SPeter Zijlstra } 3103d00806b1SNick Piggin 310474ec6751SJohannes Weiner if (!vma->vm_ops->page_mkwrite) 3105b827e496SNick Piggin file_update_time(vma->vm_file); 3106b827e496SNick Piggin 3107b827e496SNick Piggin return ret; 310854cb8821SNick Piggin } 3109d00806b1SNick Piggin 31109a95f3cfSPaul Cassella /* 31119a95f3cfSPaul Cassella * We enter with non-exclusive mmap_sem (to exclude vma changes, 31129a95f3cfSPaul Cassella * but allow concurrent faults). 31139a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 31149a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 31159a95f3cfSPaul Cassella */ 31169b4bdd2fSKirill A. Shutemov static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 311754cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 311830c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 311954cb8821SNick Piggin { 312054cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 31210da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 312254cb8821SNick Piggin 312316abfa08SHugh Dickins pte_unmap(page_table); 31246b7339f4SKirill A. Shutemov /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ 31256b7339f4SKirill A. Shutemov if (!vma->vm_ops->fault) 31266b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 3127e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3128e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3129e655fb29SKirill A. Shutemov orig_pte); 3130ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3131ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3132ec47c3b9SKirill A. Shutemov orig_pte); 3133f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 313454cb8821SNick Piggin } 313554cb8821SNick Piggin 3136b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 313704bb2f94SRik van Riel unsigned long addr, int page_nid, 313804bb2f94SRik van Riel int *flags) 31399532fec1SMel Gorman { 31409532fec1SMel Gorman get_page(page); 31419532fec1SMel Gorman 31429532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 314304bb2f94SRik van Riel if (page_nid == numa_node_id()) { 31449532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 314504bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 314604bb2f94SRik van Riel } 31479532fec1SMel Gorman 31489532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 31499532fec1SMel Gorman } 31509532fec1SMel Gorman 3151b19a9939SRashika Kheria static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3152d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3153d10e63f2SMel Gorman { 31544daae3b4SMel Gorman struct page *page = NULL; 3155d10e63f2SMel Gorman spinlock_t *ptl; 31568191acbdSMel Gorman int page_nid = -1; 315790572890SPeter Zijlstra int last_cpupid; 3158cbee9f88SPeter Zijlstra int target_nid; 3159b8593bfdSMel Gorman bool migrated = false; 3160b191f9b1SMel Gorman bool was_writable = pte_write(pte); 31616688cc05SPeter Zijlstra int flags = 0; 3162d10e63f2SMel Gorman 3163c0e7cad9SMel Gorman /* A PROT_NONE fault should not end up here */ 3164c0e7cad9SMel Gorman BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))); 3165c0e7cad9SMel Gorman 3166d10e63f2SMel Gorman /* 3167d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3168d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3169d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3170d10e63f2SMel Gorman * 31714d942466SMel Gorman * We can safely just do a "set_pte_at()", because the old 31724d942466SMel Gorman * page table entry is not accessible, so there would be no 31734d942466SMel Gorman * concurrent hardware modifications to the PTE. 3174d10e63f2SMel Gorman */ 3175d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3176d10e63f2SMel Gorman spin_lock(ptl); 31774daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 31784daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 31794daae3b4SMel Gorman goto out; 31804daae3b4SMel Gorman } 31814daae3b4SMel Gorman 31824d942466SMel Gorman /* Make it present again */ 31834d942466SMel Gorman pte = pte_modify(pte, vma->vm_page_prot); 31844d942466SMel Gorman pte = pte_mkyoung(pte); 3185b191f9b1SMel Gorman if (was_writable) 3186b191f9b1SMel Gorman pte = pte_mkwrite(pte); 3187d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3188d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3189d10e63f2SMel Gorman 3190d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3191d10e63f2SMel Gorman if (!page) { 3192d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3193d10e63f2SMel Gorman return 0; 3194d10e63f2SMel Gorman } 3195d10e63f2SMel Gorman 31966688cc05SPeter Zijlstra /* 3197bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 3198bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 3199bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 3200bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 3201bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 3202bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 32036688cc05SPeter Zijlstra */ 3204bea66fbdSMel Gorman if (!(vma->vm_flags & VM_WRITE)) 32056688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 32066688cc05SPeter Zijlstra 3207dabe1d99SRik van Riel /* 3208dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 3209dabe1d99SRik van Riel * is later used when determining whether to group tasks together 3210dabe1d99SRik van Riel */ 3211dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 3212dabe1d99SRik van Riel flags |= TNF_SHARED; 3213dabe1d99SRik van Riel 321490572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 32158191acbdSMel Gorman page_nid = page_to_nid(page); 321604bb2f94SRik van Riel target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); 3217d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 32184daae3b4SMel Gorman if (target_nid == -1) { 32194daae3b4SMel Gorman put_page(page); 32204daae3b4SMel Gorman goto out; 32214daae3b4SMel Gorman } 32224daae3b4SMel Gorman 32234daae3b4SMel Gorman /* Migrate to the requested node */ 32241bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 32256688cc05SPeter Zijlstra if (migrated) { 32268191acbdSMel Gorman page_nid = target_nid; 32276688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 3228074c2381SMel Gorman } else 3229074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 32304daae3b4SMel Gorman 32314daae3b4SMel Gorman out: 32328191acbdSMel Gorman if (page_nid != -1) 32336688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 3234d10e63f2SMel Gorman return 0; 3235d10e63f2SMel Gorman } 3236d10e63f2SMel Gorman 3237b96375f7SMatthew Wilcox static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, 3238b96375f7SMatthew Wilcox unsigned long address, pmd_t *pmd, unsigned int flags) 3239b96375f7SMatthew Wilcox { 3240fb6dd5faSKirill A. Shutemov if (vma_is_anonymous(vma)) 3241b96375f7SMatthew Wilcox return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags); 3242b96375f7SMatthew Wilcox if (vma->vm_ops->pmd_fault) 3243b96375f7SMatthew Wilcox return vma->vm_ops->pmd_fault(vma, address, pmd, flags); 3244b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 3245b96375f7SMatthew Wilcox } 3246b96375f7SMatthew Wilcox 3247b96375f7SMatthew Wilcox static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, 3248b96375f7SMatthew Wilcox unsigned long address, pmd_t *pmd, pmd_t orig_pmd, 3249b96375f7SMatthew Wilcox unsigned int flags) 3250b96375f7SMatthew Wilcox { 3251fb6dd5faSKirill A. Shutemov if (vma_is_anonymous(vma)) 3252b96375f7SMatthew Wilcox return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd); 3253b96375f7SMatthew Wilcox if (vma->vm_ops->pmd_fault) 3254b96375f7SMatthew Wilcox return vma->vm_ops->pmd_fault(vma, address, pmd, flags); 3255b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 3256b96375f7SMatthew Wilcox } 3257b96375f7SMatthew Wilcox 32581da177e4SLinus Torvalds /* 32591da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 32601da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 32611da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 32621da177e4SLinus Torvalds * 32631da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 32641da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 32651da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 32661da177e4SLinus Torvalds * 3267c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3268c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 32699a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 32709a95f3cfSPaul Cassella * 32719a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 32729a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 32731da177e4SLinus Torvalds */ 3274c0292554SKirill A. Shutemov static int handle_pte_fault(struct mm_struct *mm, 32751da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 327630c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 32771da177e4SLinus Torvalds { 32781da177e4SLinus Torvalds pte_t entry; 32798f4e2101SHugh Dickins spinlock_t *ptl; 32801da177e4SLinus Torvalds 3281e37c6982SChristian Borntraeger /* 3282e37c6982SChristian Borntraeger * some architectures can have larger ptes than wordsize, 3283e37c6982SChristian Borntraeger * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y, 3284e37c6982SChristian Borntraeger * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses. 3285e37c6982SChristian Borntraeger * The code below just needs a consistent view for the ifs and 3286e37c6982SChristian Borntraeger * we later double check anyway with the ptl lock held. So here 3287e37c6982SChristian Borntraeger * a barrier will do. 3288e37c6982SChristian Borntraeger */ 3289e37c6982SChristian Borntraeger entry = *pte; 3290e37c6982SChristian Borntraeger barrier(); 32911da177e4SLinus Torvalds if (!pte_present(entry)) { 329265500d23SHugh Dickins if (pte_none(entry)) { 3293b5330628SOleg Nesterov if (vma_is_anonymous(vma)) 3294b5330628SOleg Nesterov return do_anonymous_page(mm, vma, address, 3295b5330628SOleg Nesterov pte, pmd, flags); 3296b5330628SOleg Nesterov else 32976b7339f4SKirill A. Shutemov return do_fault(mm, vma, address, pte, pmd, 32986b7339f4SKirill A. Shutemov flags, entry); 329965500d23SHugh Dickins } 330065500d23SHugh Dickins return do_swap_page(mm, vma, address, 330130c9f3a9SLinus Torvalds pte, pmd, flags, entry); 33021da177e4SLinus Torvalds } 33031da177e4SLinus Torvalds 33048a0516edSMel Gorman if (pte_protnone(entry)) 3305d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3306d10e63f2SMel Gorman 33074c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 33088f4e2101SHugh Dickins spin_lock(ptl); 33098f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 33108f4e2101SHugh Dickins goto unlock; 331130c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 33121da177e4SLinus Torvalds if (!pte_write(entry)) 33138f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 33148f4e2101SHugh Dickins pte, pmd, ptl, entry); 33151da177e4SLinus Torvalds entry = pte_mkdirty(entry); 33161da177e4SLinus Torvalds } 33171da177e4SLinus Torvalds entry = pte_mkyoung(entry); 331830c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 33194b3073e1SRussell King update_mmu_cache(vma, address, pte); 33201a44e149SAndrea Arcangeli } else { 33211a44e149SAndrea Arcangeli /* 33221a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 33231a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 33241a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 33251a44e149SAndrea Arcangeli * with threads. 33261a44e149SAndrea Arcangeli */ 332730c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 332861c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 33291a44e149SAndrea Arcangeli } 33308f4e2101SHugh Dickins unlock: 33318f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 333283c54070SNick Piggin return 0; 33331da177e4SLinus Torvalds } 33341da177e4SLinus Torvalds 33351da177e4SLinus Torvalds /* 33361da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 33379a95f3cfSPaul Cassella * 33389a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 33399a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 33401da177e4SLinus Torvalds */ 3341519e5247SJohannes Weiner static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3342d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 33431da177e4SLinus Torvalds { 33441da177e4SLinus Torvalds pgd_t *pgd; 33451da177e4SLinus Torvalds pud_t *pud; 33461da177e4SLinus Torvalds pmd_t *pmd; 33471da177e4SLinus Torvalds pte_t *pte; 33481da177e4SLinus Torvalds 3349ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 335030c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 33511da177e4SLinus Torvalds 33521da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 33531da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 33541da177e4SLinus Torvalds if (!pud) 3355c74df32cSHugh Dickins return VM_FAULT_OOM; 33561da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 33571da177e4SLinus Torvalds if (!pmd) 3358c74df32cSHugh Dickins return VM_FAULT_OOM; 335971e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 3360b96375f7SMatthew Wilcox int ret = create_huge_pmd(mm, vma, address, pmd, flags); 3361c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 3362c0292554SKirill A. Shutemov return ret; 336371e3aac0SAndrea Arcangeli } else { 336471e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 33651f1d06c3SDavid Rientjes int ret; 33661f1d06c3SDavid Rientjes 336771e3aac0SAndrea Arcangeli barrier(); 336871e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3369a1dd450bSWill Deacon unsigned int dirty = flags & FAULT_FLAG_WRITE; 3370a1dd450bSWill Deacon 33718a0516edSMel Gorman if (pmd_protnone(orig_pmd)) 33724daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3373d10e63f2SMel Gorman orig_pmd, pmd); 3374d10e63f2SMel Gorman 33753d59eebcSLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 3376b96375f7SMatthew Wilcox ret = wp_huge_pmd(mm, vma, address, pmd, 3377b96375f7SMatthew Wilcox orig_pmd, flags); 33789845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 33791f1d06c3SDavid Rientjes return ret; 3380a1dd450bSWill Deacon } else { 3381a1dd450bSWill Deacon huge_pmd_set_accessed(mm, vma, address, pmd, 3382a1dd450bSWill Deacon orig_pmd, dirty); 338371e3aac0SAndrea Arcangeli return 0; 338471e3aac0SAndrea Arcangeli } 338571e3aac0SAndrea Arcangeli } 33869845cbbdSKirill A. Shutemov } 338771e3aac0SAndrea Arcangeli 338871e3aac0SAndrea Arcangeli /* 338971e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 339071e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 339171e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 339271e3aac0SAndrea Arcangeli */ 33934fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 33944fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3395c74df32cSHugh Dickins return VM_FAULT_OOM; 339671e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 339771e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 339871e3aac0SAndrea Arcangeli return 0; 339971e3aac0SAndrea Arcangeli /* 340071e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 340171e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 340271e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 340371e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 340471e3aac0SAndrea Arcangeli */ 340571e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 34061da177e4SLinus Torvalds 340730c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 34081da177e4SLinus Torvalds } 34091da177e4SLinus Torvalds 34109a95f3cfSPaul Cassella /* 34119a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 34129a95f3cfSPaul Cassella * 34139a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 34149a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 34159a95f3cfSPaul Cassella */ 3416519e5247SJohannes Weiner int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3417519e5247SJohannes Weiner unsigned long address, unsigned int flags) 3418519e5247SJohannes Weiner { 3419519e5247SJohannes Weiner int ret; 3420519e5247SJohannes Weiner 3421519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 3422519e5247SJohannes Weiner 3423519e5247SJohannes Weiner count_vm_event(PGFAULT); 3424519e5247SJohannes Weiner mem_cgroup_count_vm_event(mm, PGFAULT); 3425519e5247SJohannes Weiner 3426519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 3427519e5247SJohannes Weiner check_sync_rss_stat(current); 3428519e5247SJohannes Weiner 3429519e5247SJohannes Weiner /* 3430519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 3431519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 3432519e5247SJohannes Weiner */ 3433519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 343449426420SJohannes Weiner mem_cgroup_oom_enable(); 3435519e5247SJohannes Weiner 3436519e5247SJohannes Weiner ret = __handle_mm_fault(mm, vma, address, flags); 3437519e5247SJohannes Weiner 343849426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 343949426420SJohannes Weiner mem_cgroup_oom_disable(); 344049426420SJohannes Weiner /* 344149426420SJohannes Weiner * The task may have entered a memcg OOM situation but 344249426420SJohannes Weiner * if the allocation error was handled gracefully (no 344349426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 344449426420SJohannes Weiner * Just clean up the OOM state peacefully. 344549426420SJohannes Weiner */ 344649426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 344749426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 344849426420SJohannes Weiner } 34493812c8c8SJohannes Weiner 3450519e5247SJohannes Weiner return ret; 3451519e5247SJohannes Weiner } 3452e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 3453519e5247SJohannes Weiner 34541da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 34551da177e4SLinus Torvalds /* 34561da177e4SLinus Torvalds * Allocate page upper directory. 3457872fec16SHugh Dickins * We've already handled the fast-path in-line. 34581da177e4SLinus Torvalds */ 34591bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 34601da177e4SLinus Torvalds { 3461c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3462c74df32cSHugh Dickins if (!new) 34631bb3630eSHugh Dickins return -ENOMEM; 34641da177e4SLinus Torvalds 3465362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3466362a61adSNick Piggin 3467872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 34681bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 34695e541973SBenjamin Herrenschmidt pud_free(mm, new); 34701bb3630eSHugh Dickins else 34711da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3472872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 34731bb3630eSHugh Dickins return 0; 34741da177e4SLinus Torvalds } 34751da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 34761da177e4SLinus Torvalds 34771da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 34781da177e4SLinus Torvalds /* 34791da177e4SLinus Torvalds * Allocate page middle directory. 3480872fec16SHugh Dickins * We've already handled the fast-path in-line. 34811da177e4SLinus Torvalds */ 34821bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 34831da177e4SLinus Torvalds { 3484c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3485c74df32cSHugh Dickins if (!new) 34861bb3630eSHugh Dickins return -ENOMEM; 34871da177e4SLinus Torvalds 3488362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3489362a61adSNick Piggin 3490872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 34911da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 3492dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 3493dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 34941da177e4SLinus Torvalds pud_populate(mm, pud, new); 3495dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 34965e541973SBenjamin Herrenschmidt pmd_free(mm, new); 3497dc6c9a35SKirill A. Shutemov #else 3498dc6c9a35SKirill A. Shutemov if (!pgd_present(*pud)) { 3499dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 35001da177e4SLinus Torvalds pgd_populate(mm, pud, new); 3501dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 3502dc6c9a35SKirill A. Shutemov pmd_free(mm, new); 35031da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3504872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 35051bb3630eSHugh Dickins return 0; 35061da177e4SLinus Torvalds } 35071da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 35081da177e4SLinus Torvalds 35091b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3510f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3511f8ad0f49SJohannes Weiner { 3512f8ad0f49SJohannes Weiner pgd_t *pgd; 3513f8ad0f49SJohannes Weiner pud_t *pud; 3514f8ad0f49SJohannes Weiner pmd_t *pmd; 3515f8ad0f49SJohannes Weiner pte_t *ptep; 3516f8ad0f49SJohannes Weiner 3517f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3518f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3519f8ad0f49SJohannes Weiner goto out; 3520f8ad0f49SJohannes Weiner 3521f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3522f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3523f8ad0f49SJohannes Weiner goto out; 3524f8ad0f49SJohannes Weiner 3525f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3526f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3527f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3528f8ad0f49SJohannes Weiner goto out; 3529f8ad0f49SJohannes Weiner 3530f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3531f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3532f8ad0f49SJohannes Weiner goto out; 3533f8ad0f49SJohannes Weiner 3534f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3535f8ad0f49SJohannes Weiner if (!ptep) 3536f8ad0f49SJohannes Weiner goto out; 3537f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3538f8ad0f49SJohannes Weiner goto unlock; 3539f8ad0f49SJohannes Weiner *ptepp = ptep; 3540f8ad0f49SJohannes Weiner return 0; 3541f8ad0f49SJohannes Weiner unlock: 3542f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3543f8ad0f49SJohannes Weiner out: 3544f8ad0f49SJohannes Weiner return -EINVAL; 3545f8ad0f49SJohannes Weiner } 3546f8ad0f49SJohannes Weiner 35471b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 35481b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 35491b36ba81SNamhyung Kim { 35501b36ba81SNamhyung Kim int res; 35511b36ba81SNamhyung Kim 35521b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 35531b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 35541b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 35551b36ba81SNamhyung Kim return res; 35561b36ba81SNamhyung Kim } 35571b36ba81SNamhyung Kim 35583b6748e2SJohannes Weiner /** 35593b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 35603b6748e2SJohannes Weiner * @vma: memory mapping 35613b6748e2SJohannes Weiner * @address: user virtual address 35623b6748e2SJohannes Weiner * @pfn: location to store found PFN 35633b6748e2SJohannes Weiner * 35643b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 35653b6748e2SJohannes Weiner * 35663b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 35673b6748e2SJohannes Weiner */ 35683b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 35693b6748e2SJohannes Weiner unsigned long *pfn) 35703b6748e2SJohannes Weiner { 35713b6748e2SJohannes Weiner int ret = -EINVAL; 35723b6748e2SJohannes Weiner spinlock_t *ptl; 35733b6748e2SJohannes Weiner pte_t *ptep; 35743b6748e2SJohannes Weiner 35753b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 35763b6748e2SJohannes Weiner return ret; 35773b6748e2SJohannes Weiner 35783b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 35793b6748e2SJohannes Weiner if (ret) 35803b6748e2SJohannes Weiner return ret; 35813b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 35823b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 35833b6748e2SJohannes Weiner return 0; 35843b6748e2SJohannes Weiner } 35853b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 35863b6748e2SJohannes Weiner 358728b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 3588d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 358928b2ee20SRik van Riel unsigned long address, unsigned int flags, 3590d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 359128b2ee20SRik van Riel { 359203668a4dSJohannes Weiner int ret = -EINVAL; 359328b2ee20SRik van Riel pte_t *ptep, pte; 359428b2ee20SRik van Riel spinlock_t *ptl; 359528b2ee20SRik van Riel 3596d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 3597d87fe660Svenkatesh.pallipadi@intel.com goto out; 359828b2ee20SRik van Riel 359903668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 3600d87fe660Svenkatesh.pallipadi@intel.com goto out; 360128b2ee20SRik van Riel pte = *ptep; 360203668a4dSJohannes Weiner 360328b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 360428b2ee20SRik van Riel goto unlock; 360528b2ee20SRik van Riel 360628b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 360703668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 360828b2ee20SRik van Riel 360903668a4dSJohannes Weiner ret = 0; 361028b2ee20SRik van Riel unlock: 361128b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 361228b2ee20SRik van Riel out: 3613d87fe660Svenkatesh.pallipadi@intel.com return ret; 361428b2ee20SRik van Riel } 361528b2ee20SRik van Riel 361628b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 361728b2ee20SRik van Riel void *buf, int len, int write) 361828b2ee20SRik van Riel { 361928b2ee20SRik van Riel resource_size_t phys_addr; 362028b2ee20SRik van Riel unsigned long prot = 0; 36212bc7273bSKOSAKI Motohiro void __iomem *maddr; 362228b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 362328b2ee20SRik van Riel 3624d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 362528b2ee20SRik van Riel return -EINVAL; 362628b2ee20SRik van Riel 36279cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 362828b2ee20SRik van Riel if (write) 362928b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 363028b2ee20SRik van Riel else 363128b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 363228b2ee20SRik van Riel iounmap(maddr); 363328b2ee20SRik van Riel 363428b2ee20SRik van Riel return len; 363528b2ee20SRik van Riel } 36365a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 363728b2ee20SRik van Riel #endif 363828b2ee20SRik van Riel 36390ec76a11SDavid Howells /* 3640206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 3641206cb636SStephen Wilson * given task for page fault accounting. 36420ec76a11SDavid Howells */ 3643206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 3644206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 36450ec76a11SDavid Howells { 36460ec76a11SDavid Howells struct vm_area_struct *vma; 36470ec76a11SDavid Howells void *old_buf = buf; 36480ec76a11SDavid Howells 36490ec76a11SDavid Howells down_read(&mm->mmap_sem); 3650183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 36510ec76a11SDavid Howells while (len) { 36520ec76a11SDavid Howells int bytes, ret, offset; 36530ec76a11SDavid Howells void *maddr; 365428b2ee20SRik van Riel struct page *page = NULL; 36550ec76a11SDavid Howells 36560ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 36570ec76a11SDavid Howells write, 1, &page, &vma); 365828b2ee20SRik van Riel if (ret <= 0) { 3659dbffcd03SRik van Riel #ifndef CONFIG_HAVE_IOREMAP_PROT 3660dbffcd03SRik van Riel break; 3661dbffcd03SRik van Riel #else 366228b2ee20SRik van Riel /* 366328b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 366428b2ee20SRik van Riel * we can access using slightly different code. 366528b2ee20SRik van Riel */ 366628b2ee20SRik van Riel vma = find_vma(mm, addr); 3667fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 36680ec76a11SDavid Howells break; 366928b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 367028b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 367128b2ee20SRik van Riel len, write); 367228b2ee20SRik van Riel if (ret <= 0) 367328b2ee20SRik van Riel break; 367428b2ee20SRik van Riel bytes = ret; 3675dbffcd03SRik van Riel #endif 367628b2ee20SRik van Riel } else { 36770ec76a11SDavid Howells bytes = len; 36780ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 36790ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 36800ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 36810ec76a11SDavid Howells 36820ec76a11SDavid Howells maddr = kmap(page); 36830ec76a11SDavid Howells if (write) { 36840ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 36850ec76a11SDavid Howells maddr + offset, buf, bytes); 36860ec76a11SDavid Howells set_page_dirty_lock(page); 36870ec76a11SDavid Howells } else { 36880ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 36890ec76a11SDavid Howells buf, maddr + offset, bytes); 36900ec76a11SDavid Howells } 36910ec76a11SDavid Howells kunmap(page); 36920ec76a11SDavid Howells page_cache_release(page); 369328b2ee20SRik van Riel } 36940ec76a11SDavid Howells len -= bytes; 36950ec76a11SDavid Howells buf += bytes; 36960ec76a11SDavid Howells addr += bytes; 36970ec76a11SDavid Howells } 36980ec76a11SDavid Howells up_read(&mm->mmap_sem); 36990ec76a11SDavid Howells 37000ec76a11SDavid Howells return buf - old_buf; 37010ec76a11SDavid Howells } 370203252919SAndi Kleen 37035ddd36b9SStephen Wilson /** 3704ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 37055ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 37065ddd36b9SStephen Wilson * @addr: start address to access 37075ddd36b9SStephen Wilson * @buf: source or destination buffer 37085ddd36b9SStephen Wilson * @len: number of bytes to transfer 37095ddd36b9SStephen Wilson * @write: whether the access is a write 37105ddd36b9SStephen Wilson * 37115ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 37125ddd36b9SStephen Wilson */ 37135ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 37145ddd36b9SStephen Wilson void *buf, int len, int write) 37155ddd36b9SStephen Wilson { 37165ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 37175ddd36b9SStephen Wilson } 37185ddd36b9SStephen Wilson 371903252919SAndi Kleen /* 3720206cb636SStephen Wilson * Access another process' address space. 3721206cb636SStephen Wilson * Source/target buffer must be kernel space, 3722206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 3723206cb636SStephen Wilson */ 3724206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 3725206cb636SStephen Wilson void *buf, int len, int write) 3726206cb636SStephen Wilson { 3727206cb636SStephen Wilson struct mm_struct *mm; 3728206cb636SStephen Wilson int ret; 3729206cb636SStephen Wilson 3730206cb636SStephen Wilson mm = get_task_mm(tsk); 3731206cb636SStephen Wilson if (!mm) 3732206cb636SStephen Wilson return 0; 3733206cb636SStephen Wilson 3734206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 3735206cb636SStephen Wilson mmput(mm); 3736206cb636SStephen Wilson 3737206cb636SStephen Wilson return ret; 3738206cb636SStephen Wilson } 3739206cb636SStephen Wilson 374003252919SAndi Kleen /* 374103252919SAndi Kleen * Print the name of a VMA. 374203252919SAndi Kleen */ 374303252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 374403252919SAndi Kleen { 374503252919SAndi Kleen struct mm_struct *mm = current->mm; 374603252919SAndi Kleen struct vm_area_struct *vma; 374703252919SAndi Kleen 3748e8bff74aSIngo Molnar /* 3749e8bff74aSIngo Molnar * Do not print if we are in atomic 3750e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 3751e8bff74aSIngo Molnar */ 3752e8bff74aSIngo Molnar if (preempt_count()) 3753e8bff74aSIngo Molnar return; 3754e8bff74aSIngo Molnar 375503252919SAndi Kleen down_read(&mm->mmap_sem); 375603252919SAndi Kleen vma = find_vma(mm, ip); 375703252919SAndi Kleen if (vma && vma->vm_file) { 375803252919SAndi Kleen struct file *f = vma->vm_file; 375903252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 376003252919SAndi Kleen if (buf) { 37612fbc57c5SAndy Shevchenko char *p; 376203252919SAndi Kleen 37639bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 376403252919SAndi Kleen if (IS_ERR(p)) 376503252919SAndi Kleen p = "?"; 37662fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 376703252919SAndi Kleen vma->vm_start, 376803252919SAndi Kleen vma->vm_end - vma->vm_start); 376903252919SAndi Kleen free_page((unsigned long)buf); 377003252919SAndi Kleen } 377103252919SAndi Kleen } 377251a07e50SJeff Liu up_read(&mm->mmap_sem); 377303252919SAndi Kleen } 37743ee1afa3SNick Piggin 3775662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 37769ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 37773ee1afa3SNick Piggin { 377895156f00SPeter Zijlstra /* 377995156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 378095156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 378195156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 378295156f00SPeter Zijlstra * below annotations will generate false positives. 378395156f00SPeter Zijlstra */ 378495156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 378595156f00SPeter Zijlstra return; 37869ec23531SDavid Hildenbrand if (pagefault_disabled()) 3787662bbcb2SMichael S. Tsirkin return; 37889ec23531SDavid Hildenbrand __might_sleep(file, line, 0); 37899ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 3790662bbcb2SMichael S. Tsirkin if (current->mm) 37913ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 37929ec23531SDavid Hildenbrand #endif 37933ee1afa3SNick Piggin } 37949ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 37953ee1afa3SNick Piggin #endif 379647ad8475SAndrea Arcangeli 379747ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 379847ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 379947ad8475SAndrea Arcangeli unsigned long addr, 380047ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 380147ad8475SAndrea Arcangeli { 380247ad8475SAndrea Arcangeli int i; 380347ad8475SAndrea Arcangeli struct page *p = page; 380447ad8475SAndrea Arcangeli 380547ad8475SAndrea Arcangeli might_sleep(); 380647ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 380747ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 380847ad8475SAndrea Arcangeli cond_resched(); 380947ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 381047ad8475SAndrea Arcangeli } 381147ad8475SAndrea Arcangeli } 381247ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 381347ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 381447ad8475SAndrea Arcangeli { 381547ad8475SAndrea Arcangeli int i; 381647ad8475SAndrea Arcangeli 381747ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 381847ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 381947ad8475SAndrea Arcangeli return; 382047ad8475SAndrea Arcangeli } 382147ad8475SAndrea Arcangeli 382247ad8475SAndrea Arcangeli might_sleep(); 382347ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 382447ad8475SAndrea Arcangeli cond_resched(); 382547ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 382647ad8475SAndrea Arcangeli } 382747ad8475SAndrea Arcangeli } 382847ad8475SAndrea Arcangeli 382947ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 383047ad8475SAndrea Arcangeli unsigned long addr, 383147ad8475SAndrea Arcangeli struct vm_area_struct *vma, 383247ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 383347ad8475SAndrea Arcangeli { 383447ad8475SAndrea Arcangeli int i; 383547ad8475SAndrea Arcangeli struct page *dst_base = dst; 383647ad8475SAndrea Arcangeli struct page *src_base = src; 383747ad8475SAndrea Arcangeli 383847ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 383947ad8475SAndrea Arcangeli cond_resched(); 384047ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 384147ad8475SAndrea Arcangeli 384247ad8475SAndrea Arcangeli i++; 384347ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 384447ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 384547ad8475SAndrea Arcangeli } 384647ad8475SAndrea Arcangeli } 384747ad8475SAndrea Arcangeli 384847ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 384947ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 385047ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 385147ad8475SAndrea Arcangeli { 385247ad8475SAndrea Arcangeli int i; 385347ad8475SAndrea Arcangeli 385447ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 385547ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 385647ad8475SAndrea Arcangeli pages_per_huge_page); 385747ad8475SAndrea Arcangeli return; 385847ad8475SAndrea Arcangeli } 385947ad8475SAndrea Arcangeli 386047ad8475SAndrea Arcangeli might_sleep(); 386147ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 386247ad8475SAndrea Arcangeli cond_resched(); 386347ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 386447ad8475SAndrea Arcangeli } 386547ad8475SAndrea Arcangeli } 386647ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 386749076ec2SKirill A. Shutemov 386840b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 3869b35f1819SKirill A. Shutemov 3870b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 3871b35f1819SKirill A. Shutemov 3872b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 3873b35f1819SKirill A. Shutemov { 3874b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 3875b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 3876b35f1819SKirill A. Shutemov } 3877b35f1819SKirill A. Shutemov 3878539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 387949076ec2SKirill A. Shutemov { 388049076ec2SKirill A. Shutemov spinlock_t *ptl; 388149076ec2SKirill A. Shutemov 3882b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 388349076ec2SKirill A. Shutemov if (!ptl) 388449076ec2SKirill A. Shutemov return false; 3885539edb58SPeter Zijlstra page->ptl = ptl; 388649076ec2SKirill A. Shutemov return true; 388749076ec2SKirill A. Shutemov } 388849076ec2SKirill A. Shutemov 3889539edb58SPeter Zijlstra void ptlock_free(struct page *page) 389049076ec2SKirill A. Shutemov { 3891b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 389249076ec2SKirill A. Shutemov } 389349076ec2SKirill A. Shutemov #endif 3894