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> 631da177e4SLinus Torvalds 646952b61dSAlexey Dobriyan #include <asm/io.h> 651da177e4SLinus Torvalds #include <asm/pgalloc.h> 661da177e4SLinus Torvalds #include <asm/uaccess.h> 671da177e4SLinus Torvalds #include <asm/tlb.h> 681da177e4SLinus Torvalds #include <asm/tlbflush.h> 691da177e4SLinus Torvalds #include <asm/pgtable.h> 701da177e4SLinus Torvalds 7142b77728SJan Beulich #include "internal.h" 7242b77728SJan Beulich 7390572890SPeter Zijlstra #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 7490572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 7575980e97SPeter Zijlstra #endif 7675980e97SPeter Zijlstra 77d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 781da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 791da177e4SLinus Torvalds unsigned long max_mapnr; 801da177e4SLinus Torvalds struct page *mem_map; 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 831da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 841da177e4SLinus Torvalds #endif 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds /* 871da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 881da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 891da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 901da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 911da177e4SLinus Torvalds * and ZONE_HIGHMEM. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds void * high_memory; 941da177e4SLinus Torvalds 951da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 961da177e4SLinus Torvalds 9732a93233SIngo Molnar /* 9832a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 9932a93233SIngo Molnar * 10032a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 10132a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 10232a93233SIngo Molnar */ 10332a93233SIngo Molnar int randomize_va_space __read_mostly = 10432a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 10532a93233SIngo Molnar 1; 10632a93233SIngo Molnar #else 10732a93233SIngo Molnar 2; 10832a93233SIngo Molnar #endif 109a62eaf15SAndi Kleen 110a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 111a62eaf15SAndi Kleen { 112a62eaf15SAndi Kleen randomize_va_space = 0; 1139b41046cSOGAWA Hirofumi return 1; 114a62eaf15SAndi Kleen } 115a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 116a62eaf15SAndi Kleen 11762eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 11803f6462aSHugh Dickins unsigned long highest_memmap_pfn __read_mostly; 119a13ea5b7SHugh Dickins 120a13ea5b7SHugh Dickins /* 121a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 122a13ea5b7SHugh Dickins */ 123a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 124a13ea5b7SHugh Dickins { 125a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 126a13ea5b7SHugh Dickins return 0; 127a13ea5b7SHugh Dickins } 128a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 129a62eaf15SAndi Kleen 130d559db08SKAMEZAWA Hiroyuki 13134e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 13234e55232SKAMEZAWA Hiroyuki 133ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 13434e55232SKAMEZAWA Hiroyuki { 13534e55232SKAMEZAWA Hiroyuki int i; 13634e55232SKAMEZAWA Hiroyuki 13734e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 13805af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 13905af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 14005af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 14134e55232SKAMEZAWA Hiroyuki } 14234e55232SKAMEZAWA Hiroyuki } 14305af2e10SDavid Rientjes current->rss_stat.events = 0; 14434e55232SKAMEZAWA Hiroyuki } 14534e55232SKAMEZAWA Hiroyuki 14634e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 14734e55232SKAMEZAWA Hiroyuki { 14834e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 14934e55232SKAMEZAWA Hiroyuki 15034e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 15134e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 15234e55232SKAMEZAWA Hiroyuki else 15334e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 15434e55232SKAMEZAWA Hiroyuki } 15534e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 15634e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 15734e55232SKAMEZAWA Hiroyuki 15834e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 15934e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 16034e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 16134e55232SKAMEZAWA Hiroyuki { 16234e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 16334e55232SKAMEZAWA Hiroyuki return; 16434e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 165ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 16634e55232SKAMEZAWA Hiroyuki } 1679547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 16834e55232SKAMEZAWA Hiroyuki 16934e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 17034e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 17134e55232SKAMEZAWA Hiroyuki 17234e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 17334e55232SKAMEZAWA Hiroyuki { 17434e55232SKAMEZAWA Hiroyuki } 17534e55232SKAMEZAWA Hiroyuki 1769547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 1779547d01bSPeter Zijlstra 1789547d01bSPeter Zijlstra #ifdef HAVE_GENERIC_MMU_GATHER 1799547d01bSPeter Zijlstra 1809547d01bSPeter Zijlstra static int tlb_next_batch(struct mmu_gather *tlb) 1819547d01bSPeter Zijlstra { 1829547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 1839547d01bSPeter Zijlstra 1849547d01bSPeter Zijlstra batch = tlb->active; 1859547d01bSPeter Zijlstra if (batch->next) { 1869547d01bSPeter Zijlstra tlb->active = batch->next; 1879547d01bSPeter Zijlstra return 1; 1889547d01bSPeter Zijlstra } 1899547d01bSPeter Zijlstra 19053a59fc6SMichal Hocko if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) 19153a59fc6SMichal Hocko return 0; 19253a59fc6SMichal Hocko 1939547d01bSPeter Zijlstra batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 1949547d01bSPeter Zijlstra if (!batch) 1959547d01bSPeter Zijlstra return 0; 1969547d01bSPeter Zijlstra 19753a59fc6SMichal Hocko tlb->batch_count++; 1989547d01bSPeter Zijlstra batch->next = NULL; 1999547d01bSPeter Zijlstra batch->nr = 0; 2009547d01bSPeter Zijlstra batch->max = MAX_GATHER_BATCH; 2019547d01bSPeter Zijlstra 2029547d01bSPeter Zijlstra tlb->active->next = batch; 2039547d01bSPeter Zijlstra tlb->active = batch; 2049547d01bSPeter Zijlstra 2059547d01bSPeter Zijlstra return 1; 2069547d01bSPeter Zijlstra } 2079547d01bSPeter Zijlstra 2089547d01bSPeter Zijlstra /* tlb_gather_mmu 2099547d01bSPeter Zijlstra * Called to initialize an (on-stack) mmu_gather structure for page-table 2109547d01bSPeter Zijlstra * tear-down from @mm. The @fullmm argument is used when @mm is without 2119547d01bSPeter Zijlstra * users and we're going to destroy the full address space (exit/execve). 2129547d01bSPeter Zijlstra */ 2132b047252SLinus Torvalds void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) 2149547d01bSPeter Zijlstra { 2159547d01bSPeter Zijlstra tlb->mm = mm; 2169547d01bSPeter Zijlstra 2172b047252SLinus Torvalds /* Is it from 0 to ~0? */ 2182b047252SLinus Torvalds tlb->fullmm = !(start | (end+1)); 2191de14c3cSDave Hansen tlb->need_flush_all = 0; 2202b047252SLinus Torvalds tlb->start = start; 2212b047252SLinus Torvalds tlb->end = end; 2229547d01bSPeter Zijlstra tlb->need_flush = 0; 2239547d01bSPeter Zijlstra tlb->local.next = NULL; 2249547d01bSPeter Zijlstra tlb->local.nr = 0; 2259547d01bSPeter Zijlstra tlb->local.max = ARRAY_SIZE(tlb->__pages); 2269547d01bSPeter Zijlstra tlb->active = &tlb->local; 22753a59fc6SMichal Hocko tlb->batch_count = 0; 2289547d01bSPeter Zijlstra 2299547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2309547d01bSPeter Zijlstra tlb->batch = NULL; 23134e55232SKAMEZAWA Hiroyuki #endif 2329547d01bSPeter Zijlstra } 2339547d01bSPeter Zijlstra 2349547d01bSPeter Zijlstra void tlb_flush_mmu(struct mmu_gather *tlb) 2359547d01bSPeter Zijlstra { 2369547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2379547d01bSPeter Zijlstra 2389547d01bSPeter Zijlstra if (!tlb->need_flush) 2399547d01bSPeter Zijlstra return; 2409547d01bSPeter Zijlstra tlb->need_flush = 0; 2419547d01bSPeter Zijlstra tlb_flush(tlb); 2429547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2439547d01bSPeter Zijlstra tlb_table_flush(tlb); 2449547d01bSPeter Zijlstra #endif 2459547d01bSPeter Zijlstra 2469547d01bSPeter Zijlstra for (batch = &tlb->local; batch; batch = batch->next) { 2479547d01bSPeter Zijlstra free_pages_and_swap_cache(batch->pages, batch->nr); 2489547d01bSPeter Zijlstra batch->nr = 0; 2499547d01bSPeter Zijlstra } 2509547d01bSPeter Zijlstra tlb->active = &tlb->local; 2519547d01bSPeter Zijlstra } 2529547d01bSPeter Zijlstra 2539547d01bSPeter Zijlstra /* tlb_finish_mmu 2549547d01bSPeter Zijlstra * Called at the end of the shootdown operation to free up any resources 2559547d01bSPeter Zijlstra * that were required. 2569547d01bSPeter Zijlstra */ 2579547d01bSPeter Zijlstra void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 2589547d01bSPeter Zijlstra { 2599547d01bSPeter Zijlstra struct mmu_gather_batch *batch, *next; 2609547d01bSPeter Zijlstra 2619547d01bSPeter Zijlstra tlb_flush_mmu(tlb); 2629547d01bSPeter Zijlstra 2639547d01bSPeter Zijlstra /* keep the page table cache within bounds */ 2649547d01bSPeter Zijlstra check_pgt_cache(); 2659547d01bSPeter Zijlstra 2669547d01bSPeter Zijlstra for (batch = tlb->local.next; batch; batch = next) { 2679547d01bSPeter Zijlstra next = batch->next; 2689547d01bSPeter Zijlstra free_pages((unsigned long)batch, 0); 2699547d01bSPeter Zijlstra } 2709547d01bSPeter Zijlstra tlb->local.next = NULL; 2719547d01bSPeter Zijlstra } 2729547d01bSPeter Zijlstra 2739547d01bSPeter Zijlstra /* __tlb_remove_page 2749547d01bSPeter Zijlstra * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while 2759547d01bSPeter Zijlstra * handling the additional races in SMP caused by other CPUs caching valid 2769547d01bSPeter Zijlstra * mappings in their TLBs. Returns the number of free page slots left. 2779547d01bSPeter Zijlstra * When out of page slots we must call tlb_flush_mmu(). 2789547d01bSPeter Zijlstra */ 2799547d01bSPeter Zijlstra int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 2809547d01bSPeter Zijlstra { 2819547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2829547d01bSPeter Zijlstra 283f21760b1SShaohua Li VM_BUG_ON(!tlb->need_flush); 2849547d01bSPeter Zijlstra 2859547d01bSPeter Zijlstra batch = tlb->active; 2869547d01bSPeter Zijlstra batch->pages[batch->nr++] = page; 2879547d01bSPeter Zijlstra if (batch->nr == batch->max) { 2889547d01bSPeter Zijlstra if (!tlb_next_batch(tlb)) 2899547d01bSPeter Zijlstra return 0; 2900b43c3aaSShaohua Li batch = tlb->active; 2919547d01bSPeter Zijlstra } 292309381feSSasha Levin VM_BUG_ON_PAGE(batch->nr > batch->max, page); 2939547d01bSPeter Zijlstra 2949547d01bSPeter Zijlstra return batch->max - batch->nr; 2959547d01bSPeter Zijlstra } 2969547d01bSPeter Zijlstra 2979547d01bSPeter Zijlstra #endif /* HAVE_GENERIC_MMU_GATHER */ 29834e55232SKAMEZAWA Hiroyuki 29926723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 30026723911SPeter Zijlstra 30126723911SPeter Zijlstra /* 30226723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 30326723911SPeter Zijlstra */ 30426723911SPeter Zijlstra 30526723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 30626723911SPeter Zijlstra { 30726723911SPeter Zijlstra /* Simply deliver the interrupt */ 30826723911SPeter Zijlstra } 30926723911SPeter Zijlstra 31026723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 31126723911SPeter Zijlstra { 31226723911SPeter Zijlstra /* 31326723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 31426723911SPeter Zijlstra * assumed to be actually RCU-freed. 31526723911SPeter Zijlstra * 31626723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 31726723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 31826723911SPeter Zijlstra */ 31926723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 32026723911SPeter Zijlstra __tlb_remove_table(table); 32126723911SPeter Zijlstra } 32226723911SPeter Zijlstra 32326723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 32426723911SPeter Zijlstra { 32526723911SPeter Zijlstra struct mmu_table_batch *batch; 32626723911SPeter Zijlstra int i; 32726723911SPeter Zijlstra 32826723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 32926723911SPeter Zijlstra 33026723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 33126723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 33226723911SPeter Zijlstra 33326723911SPeter Zijlstra free_page((unsigned long)batch); 33426723911SPeter Zijlstra } 33526723911SPeter Zijlstra 33626723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 33726723911SPeter Zijlstra { 33826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 33926723911SPeter Zijlstra 34026723911SPeter Zijlstra if (*batch) { 34126723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 34226723911SPeter Zijlstra *batch = NULL; 34326723911SPeter Zijlstra } 34426723911SPeter Zijlstra } 34526723911SPeter Zijlstra 34626723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 34726723911SPeter Zijlstra { 34826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 34926723911SPeter Zijlstra 35026723911SPeter Zijlstra tlb->need_flush = 1; 35126723911SPeter Zijlstra 35226723911SPeter Zijlstra /* 35326723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 35426723911SPeter Zijlstra * concurrent page-table walk. 35526723911SPeter Zijlstra */ 35626723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 35726723911SPeter Zijlstra __tlb_remove_table(table); 35826723911SPeter Zijlstra return; 35926723911SPeter Zijlstra } 36026723911SPeter Zijlstra 36126723911SPeter Zijlstra if (*batch == NULL) { 36226723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 36326723911SPeter Zijlstra if (*batch == NULL) { 36426723911SPeter Zijlstra tlb_remove_table_one(table); 36526723911SPeter Zijlstra return; 36626723911SPeter Zijlstra } 36726723911SPeter Zijlstra (*batch)->nr = 0; 36826723911SPeter Zijlstra } 36926723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 37026723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 37126723911SPeter Zijlstra tlb_table_flush(tlb); 37226723911SPeter Zijlstra } 37326723911SPeter Zijlstra 3749547d01bSPeter Zijlstra #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ 37526723911SPeter Zijlstra 3761da177e4SLinus Torvalds /* 3771da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 3781da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 3791da177e4SLinus Torvalds */ 3809e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 3819e1b32caSBenjamin Herrenschmidt unsigned long addr) 3821da177e4SLinus Torvalds { 3832f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 3841da177e4SLinus Torvalds pmd_clear(pmd); 3859e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 386e1f56c89SKirill A. Shutemov atomic_long_dec(&tlb->mm->nr_ptes); 3871da177e4SLinus Torvalds } 3881da177e4SLinus Torvalds 389e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 390e0da382cSHugh Dickins unsigned long addr, unsigned long end, 391e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3921da177e4SLinus Torvalds { 3931da177e4SLinus Torvalds pmd_t *pmd; 3941da177e4SLinus Torvalds unsigned long next; 395e0da382cSHugh Dickins unsigned long start; 3961da177e4SLinus Torvalds 397e0da382cSHugh Dickins start = addr; 3981da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 3991da177e4SLinus Torvalds do { 4001da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 4011da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 4021da177e4SLinus Torvalds continue; 4039e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 4041da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 4051da177e4SLinus Torvalds 406e0da382cSHugh Dickins start &= PUD_MASK; 407e0da382cSHugh Dickins if (start < floor) 408e0da382cSHugh Dickins return; 409e0da382cSHugh Dickins if (ceiling) { 410e0da382cSHugh Dickins ceiling &= PUD_MASK; 411e0da382cSHugh Dickins if (!ceiling) 412e0da382cSHugh Dickins return; 4131da177e4SLinus Torvalds } 414e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 415e0da382cSHugh Dickins return; 416e0da382cSHugh Dickins 417e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 418e0da382cSHugh Dickins pud_clear(pud); 4199e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 4201da177e4SLinus Torvalds } 4211da177e4SLinus Torvalds 422e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 423e0da382cSHugh Dickins unsigned long addr, unsigned long end, 424e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4251da177e4SLinus Torvalds { 4261da177e4SLinus Torvalds pud_t *pud; 4271da177e4SLinus Torvalds unsigned long next; 428e0da382cSHugh Dickins unsigned long start; 4291da177e4SLinus Torvalds 430e0da382cSHugh Dickins start = addr; 4311da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 4321da177e4SLinus Torvalds do { 4331da177e4SLinus Torvalds next = pud_addr_end(addr, end); 4341da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 4351da177e4SLinus Torvalds continue; 436e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 4371da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 4381da177e4SLinus Torvalds 439e0da382cSHugh Dickins start &= PGDIR_MASK; 440e0da382cSHugh Dickins if (start < floor) 441e0da382cSHugh Dickins return; 442e0da382cSHugh Dickins if (ceiling) { 443e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 444e0da382cSHugh Dickins if (!ceiling) 445e0da382cSHugh Dickins return; 4461da177e4SLinus Torvalds } 447e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 448e0da382cSHugh Dickins return; 449e0da382cSHugh Dickins 450e0da382cSHugh Dickins pud = pud_offset(pgd, start); 451e0da382cSHugh Dickins pgd_clear(pgd); 4529e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds /* 456e0da382cSHugh Dickins * This function frees user-level page tables of a process. 4571da177e4SLinus Torvalds */ 45842b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 459e0da382cSHugh Dickins unsigned long addr, unsigned long end, 460e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4611da177e4SLinus Torvalds { 4621da177e4SLinus Torvalds pgd_t *pgd; 4631da177e4SLinus Torvalds unsigned long next; 4641da177e4SLinus Torvalds 465e0da382cSHugh Dickins /* 466e0da382cSHugh Dickins * The next few lines have given us lots of grief... 467e0da382cSHugh Dickins * 468e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 469e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 470e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 471e0da382cSHugh Dickins * 472e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 473e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 474e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 475e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 476e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 477e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 478e0da382cSHugh Dickins * that end 0 case should be mythical). 479e0da382cSHugh Dickins * 480e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 481e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 482e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 483e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 484e0da382cSHugh Dickins * 485e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 486e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 487e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 488e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 489e0da382cSHugh Dickins */ 490e0da382cSHugh Dickins 491e0da382cSHugh Dickins addr &= PMD_MASK; 492e0da382cSHugh Dickins if (addr < floor) { 493e0da382cSHugh Dickins addr += PMD_SIZE; 494e0da382cSHugh Dickins if (!addr) 495e0da382cSHugh Dickins return; 496e0da382cSHugh Dickins } 497e0da382cSHugh Dickins if (ceiling) { 498e0da382cSHugh Dickins ceiling &= PMD_MASK; 499e0da382cSHugh Dickins if (!ceiling) 500e0da382cSHugh Dickins return; 501e0da382cSHugh Dickins } 502e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 503e0da382cSHugh Dickins end -= PMD_SIZE; 504e0da382cSHugh Dickins if (addr > end - 1) 505e0da382cSHugh Dickins return; 506e0da382cSHugh Dickins 50742b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 5081da177e4SLinus Torvalds do { 5091da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 5101da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 5111da177e4SLinus Torvalds continue; 51242b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 5131da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 514e0da382cSHugh Dickins } 515e0da382cSHugh Dickins 51642b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 517e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 518e0da382cSHugh Dickins { 519e0da382cSHugh Dickins while (vma) { 520e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 521e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 522e0da382cSHugh Dickins 5238f4f8c16SHugh Dickins /* 52425d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 52525d9e2d1Snpiggin@suse.de * pgtables 5268f4f8c16SHugh Dickins */ 5275beb4930SRik van Riel unlink_anon_vmas(vma); 5288f4f8c16SHugh Dickins unlink_file_vma(vma); 5298f4f8c16SHugh Dickins 5309da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 5313bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 5323bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 5333bf5ee95SHugh Dickins } else { 5343bf5ee95SHugh Dickins /* 5353bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 5363bf5ee95SHugh Dickins */ 5373bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 5384866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 539e0da382cSHugh Dickins vma = next; 540e0da382cSHugh Dickins next = vma->vm_next; 5415beb4930SRik van Riel unlink_anon_vmas(vma); 5428f4f8c16SHugh Dickins unlink_file_vma(vma); 543e0da382cSHugh Dickins } 5443bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 545e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 5463bf5ee95SHugh Dickins } 547e0da382cSHugh Dickins vma = next; 548e0da382cSHugh Dickins } 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5518ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 5528ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 5531da177e4SLinus Torvalds { 554c4088ebdSKirill A. Shutemov spinlock_t *ptl; 5552f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 5568ac1f832SAndrea Arcangeli int wait_split_huge_page; 5571da177e4SLinus Torvalds if (!new) 5581bb3630eSHugh Dickins return -ENOMEM; 5591bb3630eSHugh Dickins 560362a61adSNick Piggin /* 561362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 562362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 563362a61adSNick Piggin * put into page tables. 564362a61adSNick Piggin * 565362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 566362a61adSNick Piggin * table walking code (when walking the page table without locking; 567362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 568362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 569362a61adSNick Piggin * being the notable exception) will already guarantee loads are 570362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 571362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 572362a61adSNick Piggin */ 573362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 574362a61adSNick Piggin 575c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 5768ac1f832SAndrea Arcangeli wait_split_huge_page = 0; 5778ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 578e1f56c89SKirill A. Shutemov atomic_long_inc(&mm->nr_ptes); 5791da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 5802f569afdSMartin Schwidefsky new = NULL; 5818ac1f832SAndrea Arcangeli } else if (unlikely(pmd_trans_splitting(*pmd))) 5828ac1f832SAndrea Arcangeli wait_split_huge_page = 1; 583c4088ebdSKirill A. Shutemov spin_unlock(ptl); 5842f569afdSMartin Schwidefsky if (new) 5852f569afdSMartin Schwidefsky pte_free(mm, new); 5868ac1f832SAndrea Arcangeli if (wait_split_huge_page) 5878ac1f832SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 5881bb3630eSHugh Dickins return 0; 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds 5911bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 5921da177e4SLinus Torvalds { 5931bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 5941da177e4SLinus Torvalds if (!new) 5951bb3630eSHugh Dickins return -ENOMEM; 5961da177e4SLinus Torvalds 597362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 598362a61adSNick Piggin 599872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 6008ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 601872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 6022f569afdSMartin Schwidefsky new = NULL; 6038ac1f832SAndrea Arcangeli } else 6048ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_splitting(*pmd)); 605872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 6062f569afdSMartin Schwidefsky if (new) 6072f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 6081bb3630eSHugh Dickins return 0; 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds 611d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 612ae859762SHugh Dickins { 613d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 614d559db08SKAMEZAWA Hiroyuki } 615d559db08SKAMEZAWA Hiroyuki 616d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 617d559db08SKAMEZAWA Hiroyuki { 618d559db08SKAMEZAWA Hiroyuki int i; 619d559db08SKAMEZAWA Hiroyuki 62034e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 62105af2e10SDavid Rientjes sync_mm_rss(mm); 622d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 623d559db08SKAMEZAWA Hiroyuki if (rss[i]) 624d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 625ae859762SHugh Dickins } 626ae859762SHugh Dickins 6271da177e4SLinus Torvalds /* 6286aab341eSLinus Torvalds * This function is called to print an error when a bad pte 6296aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 6306aab341eSLinus Torvalds * a region that doesn't allow it. 631b5810039SNick Piggin * 632b5810039SNick Piggin * The calling function must still handle the error. 633b5810039SNick Piggin */ 6343dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 6353dc14741SHugh Dickins pte_t pte, struct page *page) 636b5810039SNick Piggin { 6373dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 6383dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 6393dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 6403dc14741SHugh Dickins struct address_space *mapping; 6413dc14741SHugh Dickins pgoff_t index; 642d936cf9bSHugh Dickins static unsigned long resume; 643d936cf9bSHugh Dickins static unsigned long nr_shown; 644d936cf9bSHugh Dickins static unsigned long nr_unshown; 645d936cf9bSHugh Dickins 646d936cf9bSHugh Dickins /* 647d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 648d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 649d936cf9bSHugh Dickins */ 650d936cf9bSHugh Dickins if (nr_shown == 60) { 651d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 652d936cf9bSHugh Dickins nr_unshown++; 653d936cf9bSHugh Dickins return; 654d936cf9bSHugh Dickins } 655d936cf9bSHugh Dickins if (nr_unshown) { 6561e9e6365SHugh Dickins printk(KERN_ALERT 6571e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 658d936cf9bSHugh Dickins nr_unshown); 659d936cf9bSHugh Dickins nr_unshown = 0; 660d936cf9bSHugh Dickins } 661d936cf9bSHugh Dickins nr_shown = 0; 662d936cf9bSHugh Dickins } 663d936cf9bSHugh Dickins if (nr_shown++ == 0) 664d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 6653dc14741SHugh Dickins 6663dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 6673dc14741SHugh Dickins index = linear_page_index(vma, addr); 6683dc14741SHugh Dickins 6691e9e6365SHugh Dickins printk(KERN_ALERT 6701e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 6713dc14741SHugh Dickins current->comm, 6723dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 673718a3821SWu Fengguang if (page) 674f0b791a3SDave Hansen dump_page(page, "bad pte"); 6751e9e6365SHugh Dickins printk(KERN_ALERT 6763dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 6773dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 6783dc14741SHugh Dickins /* 6793dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 6803dc14741SHugh Dickins */ 6813dc14741SHugh Dickins if (vma->vm_ops) 682071361d3SJoe Perches printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n", 683071361d3SJoe Perches vma->vm_ops->fault); 68472c2d531SAl Viro if (vma->vm_file) 685071361d3SJoe Perches printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n", 686071361d3SJoe Perches vma->vm_file->f_op->mmap); 687b5810039SNick Piggin dump_stack(); 688373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 689b5810039SNick Piggin } 690b5810039SNick Piggin 6912ec74c3eSSagi Grimberg static inline bool is_cow_mapping(vm_flags_t flags) 69267121172SLinus Torvalds { 69367121172SLinus Torvalds return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 69467121172SLinus Torvalds } 69567121172SLinus Torvalds 696b5810039SNick Piggin /* 6977e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 6986aab341eSLinus Torvalds * 6997e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7007e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7017e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 702b379d790SJared Hulbert * 7037e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7047e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7057e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7067e675137SNick Piggin * described below. 7077e675137SNick Piggin * 7087e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7097e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7107e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7116aab341eSLinus Torvalds * 712b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 713b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7147e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7157e675137SNick Piggin * mapping will always honor the rule 7166aab341eSLinus Torvalds * 7176aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7186aab341eSLinus Torvalds * 7197e675137SNick Piggin * And for normal mappings this is false. 720b379d790SJared Hulbert * 7217e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7227e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7237e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7247e675137SNick Piggin * special (because none can have been COWed). 725b379d790SJared Hulbert * 726b379d790SJared Hulbert * 7277e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7287e675137SNick Piggin * 729b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 730b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 731b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 732b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 733b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 734b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 735b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 736b379d790SJared Hulbert * 737ee498ed7SHugh Dickins */ 7387e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7397e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7407e675137SNick Piggin #else 7417e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7427e675137SNick Piggin #endif 7437e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7447e675137SNick Piggin pte_t pte) 745ee498ed7SHugh Dickins { 74622b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7477e675137SNick Piggin 7487e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 74922b31eecSHugh Dickins if (likely(!pte_special(pte))) 75022b31eecSHugh Dickins goto check_pfn; 751a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 752a13ea5b7SHugh Dickins return NULL; 75362eede62SHugh Dickins if (!is_zero_pfn(pfn)) 75422b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7557e675137SNick Piggin return NULL; 7567e675137SNick Piggin } 7577e675137SNick Piggin 7587e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7597e675137SNick Piggin 760b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 761b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 762b379d790SJared Hulbert if (!pfn_valid(pfn)) 763b379d790SJared Hulbert return NULL; 764b379d790SJared Hulbert goto out; 765b379d790SJared Hulbert } else { 7667e675137SNick Piggin unsigned long off; 7677e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7686aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7696aab341eSLinus Torvalds return NULL; 77067121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 771fb155c16SLinus Torvalds return NULL; 7726aab341eSLinus Torvalds } 773b379d790SJared Hulbert } 7746aab341eSLinus Torvalds 77562eede62SHugh Dickins if (is_zero_pfn(pfn)) 77662eede62SHugh Dickins return NULL; 77722b31eecSHugh Dickins check_pfn: 77822b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 77922b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 78022b31eecSHugh Dickins return NULL; 78122b31eecSHugh Dickins } 7826aab341eSLinus Torvalds 7836aab341eSLinus Torvalds /* 7847e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7857e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7866aab341eSLinus Torvalds */ 787b379d790SJared Hulbert out: 7886aab341eSLinus Torvalds return pfn_to_page(pfn); 789ee498ed7SHugh Dickins } 790ee498ed7SHugh Dickins 791ee498ed7SHugh Dickins /* 7921da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 7931da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 7941da177e4SLinus Torvalds * covered by this vma. 7951da177e4SLinus Torvalds */ 7961da177e4SLinus Torvalds 797570a335bSHugh Dickins static inline unsigned long 7981da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 799b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8008c103762SHugh Dickins unsigned long addr, int *rss) 8011da177e4SLinus Torvalds { 802b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8031da177e4SLinus Torvalds pte_t pte = *src_pte; 8041da177e4SLinus Torvalds struct page *page; 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8071da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8081da177e4SLinus Torvalds if (!pte_file(pte)) { 8090697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8100697212aSChristoph Lameter 811570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 812570a335bSHugh Dickins return entry.val; 813570a335bSHugh Dickins 8141da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8151da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8161da177e4SLinus Torvalds spin_lock(&mmlist_lock); 817f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 818f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 819f412ac08SHugh Dickins &src_mm->mmlist); 8201da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8211da177e4SLinus Torvalds } 822b084d435SKAMEZAWA Hiroyuki if (likely(!non_swap_entry(entry))) 823b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8249f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 8259f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8269f9f1acdSKonstantin Khlebnikov 8279f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 8289f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]++; 8299f9f1acdSKonstantin Khlebnikov else 8309f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]++; 8319f9f1acdSKonstantin Khlebnikov 8329f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8330697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8340697212aSChristoph Lameter /* 8359f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8369f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8370697212aSChristoph Lameter */ 8380697212aSChristoph Lameter make_migration_entry_read(&entry); 8390697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 840c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 841c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 8420697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8430697212aSChristoph Lameter } 8441da177e4SLinus Torvalds } 8459f9f1acdSKonstantin Khlebnikov } 846ae859762SHugh Dickins goto out_set_pte; 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds 8491da177e4SLinus Torvalds /* 8501da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8511da177e4SLinus Torvalds * in the parent and the child 8521da177e4SLinus Torvalds */ 85367121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8541da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8553dc90795SZachary Amsden pte = pte_wrprotect(pte); 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds /* 8591da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8601da177e4SLinus Torvalds * the child 8611da177e4SLinus Torvalds */ 8621da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8631da177e4SLinus Torvalds pte = pte_mkclean(pte); 8641da177e4SLinus Torvalds pte = pte_mkold(pte); 8656aab341eSLinus Torvalds 8666aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 8676aab341eSLinus Torvalds if (page) { 8681da177e4SLinus Torvalds get_page(page); 86921333b2bSHugh Dickins page_dup_rmap(page); 870d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 871d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 872d559db08SKAMEZAWA Hiroyuki else 873d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 8746aab341eSLinus Torvalds } 875ae859762SHugh Dickins 876ae859762SHugh Dickins out_set_pte: 877ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 878570a335bSHugh Dickins return 0; 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds 88171e3aac0SAndrea Arcangeli int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8821da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8831da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8841da177e4SLinus Torvalds { 885c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8861da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 887c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 888e040f218SHugh Dickins int progress = 0; 889d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 890570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds again: 893d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 894d559db08SKAMEZAWA Hiroyuki 895c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 8961da177e4SLinus Torvalds if (!dst_pte) 8971da177e4SLinus Torvalds return -ENOMEM; 898ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 8994c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 900f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 901c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 902c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9036606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds do { 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9081da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9091da177e4SLinus Torvalds */ 910e040f218SHugh Dickins if (progress >= 32) { 911e040f218SHugh Dickins progress = 0; 912e040f218SHugh Dickins if (need_resched() || 91395c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9141da177e4SLinus Torvalds break; 915e040f218SHugh Dickins } 9161da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9171da177e4SLinus Torvalds progress++; 9181da177e4SLinus Torvalds continue; 9191da177e4SLinus Torvalds } 920570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 921570a335bSHugh Dickins vma, addr, rss); 922570a335bSHugh Dickins if (entry.val) 923570a335bSHugh Dickins break; 9241da177e4SLinus Torvalds progress += 8; 9251da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9261da177e4SLinus Torvalds 9276606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 928c74df32cSHugh Dickins spin_unlock(src_ptl); 929ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 930d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 931c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 932c74df32cSHugh Dickins cond_resched(); 933570a335bSHugh Dickins 934570a335bSHugh Dickins if (entry.val) { 935570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 936570a335bSHugh Dickins return -ENOMEM; 937570a335bSHugh Dickins progress = 0; 938570a335bSHugh Dickins } 9391da177e4SLinus Torvalds if (addr != end) 9401da177e4SLinus Torvalds goto again; 9411da177e4SLinus Torvalds return 0; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds 9441da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9451da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9461da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9471da177e4SLinus Torvalds { 9481da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9491da177e4SLinus Torvalds unsigned long next; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9521da177e4SLinus Torvalds if (!dst_pmd) 9531da177e4SLinus Torvalds return -ENOMEM; 9541da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9551da177e4SLinus Torvalds do { 9561da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 95771e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 95871e3aac0SAndrea Arcangeli int err; 95914d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 96071e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 96171e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 96271e3aac0SAndrea Arcangeli if (err == -ENOMEM) 96371e3aac0SAndrea Arcangeli return -ENOMEM; 96471e3aac0SAndrea Arcangeli if (!err) 96571e3aac0SAndrea Arcangeli continue; 96671e3aac0SAndrea Arcangeli /* fall through */ 96771e3aac0SAndrea Arcangeli } 9681da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 9691da177e4SLinus Torvalds continue; 9701da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 9711da177e4SLinus Torvalds vma, addr, next)) 9721da177e4SLinus Torvalds return -ENOMEM; 9731da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 9741da177e4SLinus Torvalds return 0; 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9781da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9791da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9801da177e4SLinus Torvalds { 9811da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9821da177e4SLinus Torvalds unsigned long next; 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9851da177e4SLinus Torvalds if (!dst_pud) 9861da177e4SLinus Torvalds return -ENOMEM; 9871da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9881da177e4SLinus Torvalds do { 9891da177e4SLinus Torvalds next = pud_addr_end(addr, end); 9901da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 9911da177e4SLinus Torvalds continue; 9921da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 9931da177e4SLinus Torvalds vma, addr, next)) 9941da177e4SLinus Torvalds return -ENOMEM; 9951da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 9961da177e4SLinus Torvalds return 0; 9971da177e4SLinus Torvalds } 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10001da177e4SLinus Torvalds struct vm_area_struct *vma) 10011da177e4SLinus Torvalds { 10021da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 10031da177e4SLinus Torvalds unsigned long next; 10041da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10051da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10062ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10072ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10082ec74c3eSSagi Grimberg bool is_cow; 1009cddb8a5cSAndrea Arcangeli int ret; 10101da177e4SLinus Torvalds 1011d992895bSNick Piggin /* 1012d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1013d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1014d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1015d992895bSNick Piggin * efficient than faulting. 1016d992895bSNick Piggin */ 10174b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR | 10184b6e1e37SKonstantin Khlebnikov VM_PFNMAP | VM_MIXEDMAP))) { 1019d992895bSNick Piggin if (!vma->anon_vma) 1020d992895bSNick Piggin return 0; 1021d992895bSNick Piggin } 1022d992895bSNick Piggin 10231da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10241da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10251da177e4SLinus Torvalds 1026b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10272ab64037Svenkatesh.pallipadi@intel.com /* 10282ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10292ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10302ab64037Svenkatesh.pallipadi@intel.com */ 10315180da41SSuresh Siddha ret = track_pfn_copy(vma); 10322ab64037Svenkatesh.pallipadi@intel.com if (ret) 10332ab64037Svenkatesh.pallipadi@intel.com return ret; 10342ab64037Svenkatesh.pallipadi@intel.com } 10352ab64037Svenkatesh.pallipadi@intel.com 1036cddb8a5cSAndrea Arcangeli /* 1037cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1038cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1039cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1040cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1041cddb8a5cSAndrea Arcangeli */ 10422ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10432ec74c3eSSagi Grimberg mmun_start = addr; 10442ec74c3eSSagi Grimberg mmun_end = end; 10452ec74c3eSSagi Grimberg if (is_cow) 10462ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10472ec74c3eSSagi Grimberg mmun_end); 1048cddb8a5cSAndrea Arcangeli 1049cddb8a5cSAndrea Arcangeli ret = 0; 10501da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10511da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10521da177e4SLinus Torvalds do { 10531da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10541da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10551da177e4SLinus Torvalds continue; 1056cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1057cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1058cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1059cddb8a5cSAndrea Arcangeli break; 1060cddb8a5cSAndrea Arcangeli } 10611da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1062cddb8a5cSAndrea Arcangeli 10632ec74c3eSSagi Grimberg if (is_cow) 10642ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1065cddb8a5cSAndrea Arcangeli return ret; 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds 106851c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1069b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 10701da177e4SLinus Torvalds unsigned long addr, unsigned long end, 107197a89413SPeter Zijlstra struct zap_details *details) 10721da177e4SLinus Torvalds { 1073b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1074d16dfc55SPeter Zijlstra int force_flush = 0; 1075d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 107697a89413SPeter Zijlstra spinlock_t *ptl; 10775f1a1907SSteven Rostedt pte_t *start_pte; 107897a89413SPeter Zijlstra pte_t *pte; 1079d559db08SKAMEZAWA Hiroyuki 1080d16dfc55SPeter Zijlstra again: 1081e303297eSPeter Zijlstra init_rss_vec(rss); 10825f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10835f1a1907SSteven Rostedt pte = start_pte; 10846606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10851da177e4SLinus Torvalds do { 10861da177e4SLinus Torvalds pte_t ptent = *pte; 108751c6f666SRobin Holt if (pte_none(ptent)) { 10881da177e4SLinus Torvalds continue; 108951c6f666SRobin Holt } 109051c6f666SRobin Holt 10916f5e6b9eSHugh Dickins if (pte_present(ptent)) { 10926f5e6b9eSHugh Dickins struct page *page; 10936f5e6b9eSHugh Dickins 10946aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 10951da177e4SLinus Torvalds if (unlikely(details) && page) { 10961da177e4SLinus Torvalds /* 10971da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 10981da177e4SLinus Torvalds * invalidate cache without truncating: 10991da177e4SLinus Torvalds * unmap shared but keep private pages. 11001da177e4SLinus Torvalds */ 11011da177e4SLinus Torvalds if (details->check_mapping && 11021da177e4SLinus Torvalds details->check_mapping != page->mapping) 11031da177e4SLinus Torvalds continue; 11041da177e4SLinus Torvalds /* 11051da177e4SLinus Torvalds * Each page->index must be checked when 11061da177e4SLinus Torvalds * invalidating or truncating nonlinear. 11071da177e4SLinus Torvalds */ 11081da177e4SLinus Torvalds if (details->nonlinear_vma && 11091da177e4SLinus Torvalds (page->index < details->first_index || 11101da177e4SLinus Torvalds page->index > details->last_index)) 11111da177e4SLinus Torvalds continue; 11121da177e4SLinus Torvalds } 1113b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1114a600388dSZachary Amsden tlb->fullmm); 11151da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11161da177e4SLinus Torvalds if (unlikely(!page)) 11171da177e4SLinus Torvalds continue; 11181da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 11191da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 112041bb3476SCyrill Gorcunov addr) != page->index) { 112141bb3476SCyrill Gorcunov pte_t ptfile = pgoff_to_pte(page->index); 112241bb3476SCyrill Gorcunov if (pte_soft_dirty(ptent)) 112341bb3476SCyrill Gorcunov pte_file_mksoft_dirty(ptfile); 112441bb3476SCyrill Gorcunov set_pte_at(mm, addr, pte, ptfile); 112541bb3476SCyrill Gorcunov } 11261da177e4SLinus Torvalds if (PageAnon(page)) 1127d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 11286237bcd9SHugh Dickins else { 11296237bcd9SHugh Dickins if (pte_dirty(ptent)) 11306237bcd9SHugh Dickins set_page_dirty(page); 11314917e5d0SJohannes Weiner if (pte_young(ptent) && 113264363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1133bf3f3bc5SNick Piggin mark_page_accessed(page); 1134d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 11356237bcd9SHugh Dickins } 1136edc315fdSHugh Dickins page_remove_rmap(page); 11373dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11383dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 1139d16dfc55SPeter Zijlstra force_flush = !__tlb_remove_page(tlb, page); 1140d16dfc55SPeter Zijlstra if (force_flush) 1141d16dfc55SPeter Zijlstra break; 11421da177e4SLinus Torvalds continue; 11431da177e4SLinus Torvalds } 11441da177e4SLinus Torvalds /* 11451da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 11461da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 11471da177e4SLinus Torvalds */ 11481da177e4SLinus Torvalds if (unlikely(details)) 11491da177e4SLinus Torvalds continue; 11502509ef26SHugh Dickins if (pte_file(ptent)) { 11512509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) 11522509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1153b084d435SKAMEZAWA Hiroyuki } else { 1154b084d435SKAMEZAWA Hiroyuki swp_entry_t entry = pte_to_swp_entry(ptent); 1155b084d435SKAMEZAWA Hiroyuki 1156b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1157b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11589f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11599f9f1acdSKonstantin Khlebnikov struct page *page; 11609f9f1acdSKonstantin Khlebnikov 11619f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 11629f9f1acdSKonstantin Khlebnikov 11639f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 11649f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]--; 11659f9f1acdSKonstantin Khlebnikov else 11669f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]--; 11679f9f1acdSKonstantin Khlebnikov } 1168b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 11692509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1170b084d435SKAMEZAWA Hiroyuki } 11719888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 117297a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1173ae859762SHugh Dickins 1174d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 11756606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 11765f1a1907SSteven Rostedt pte_unmap_unlock(start_pte, ptl); 117751c6f666SRobin Holt 1178d16dfc55SPeter Zijlstra /* 1179d16dfc55SPeter Zijlstra * mmu_gather ran out of room to batch pages, we break out of 1180d16dfc55SPeter Zijlstra * the PTE lock to avoid doing the potential expensive TLB invalidate 1181d16dfc55SPeter Zijlstra * and page-free while holding it. 1182d16dfc55SPeter Zijlstra */ 1183d16dfc55SPeter Zijlstra if (force_flush) { 11842b047252SLinus Torvalds unsigned long old_end; 11852b047252SLinus Torvalds 1186d16dfc55SPeter Zijlstra force_flush = 0; 1187597e1c35SAlex Shi 11882b047252SLinus Torvalds /* 11892b047252SLinus Torvalds * Flush the TLB just for the previous segment, 11902b047252SLinus Torvalds * then update the range to be the remaining 11912b047252SLinus Torvalds * TLB range. 11922b047252SLinus Torvalds */ 11932b047252SLinus Torvalds old_end = tlb->end; 1194e6c495a9SVineet Gupta tlb->end = addr; 11952b047252SLinus Torvalds 1196d16dfc55SPeter Zijlstra tlb_flush_mmu(tlb); 11972b047252SLinus Torvalds 11982b047252SLinus Torvalds tlb->start = addr; 11992b047252SLinus Torvalds tlb->end = old_end; 12002b047252SLinus Torvalds 12012b047252SLinus Torvalds if (addr != end) 1202d16dfc55SPeter Zijlstra goto again; 1203d16dfc55SPeter Zijlstra } 1204d16dfc55SPeter Zijlstra 120551c6f666SRobin Holt return addr; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 120851c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1209b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 12101da177e4SLinus Torvalds unsigned long addr, unsigned long end, 121197a89413SPeter Zijlstra struct zap_details *details) 12121da177e4SLinus Torvalds { 12131da177e4SLinus Torvalds pmd_t *pmd; 12141da177e4SLinus Torvalds unsigned long next; 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 12171da177e4SLinus Torvalds do { 12181da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 121971e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 122014d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1221e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1222e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1223e0897d75SDavid 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", 1224e0897d75SDavid Rientjes __func__, addr, end, 1225e0897d75SDavid Rientjes vma->vm_start, 1226e0897d75SDavid Rientjes vma->vm_end); 1227e0897d75SDavid Rientjes BUG(); 1228e0897d75SDavid Rientjes } 1229e0897d75SDavid Rientjes #endif 1230e180377fSKirill A. Shutemov split_huge_page_pmd(vma, addr, pmd); 1231f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 12321a5a9906SAndrea Arcangeli goto next; 123371e3aac0SAndrea Arcangeli /* fall through */ 123471e3aac0SAndrea Arcangeli } 12351a5a9906SAndrea Arcangeli /* 12361a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 12371a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 12381a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 12391a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12401a5a9906SAndrea Arcangeli * mode. 12411a5a9906SAndrea Arcangeli */ 12421a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12431a5a9906SAndrea Arcangeli goto next; 124497a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12451a5a9906SAndrea Arcangeli next: 124697a89413SPeter Zijlstra cond_resched(); 124797a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 124851c6f666SRobin Holt 124951c6f666SRobin Holt return addr; 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds 125251c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1253b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12541da177e4SLinus Torvalds unsigned long addr, unsigned long end, 125597a89413SPeter Zijlstra struct zap_details *details) 12561da177e4SLinus Torvalds { 12571da177e4SLinus Torvalds pud_t *pud; 12581da177e4SLinus Torvalds unsigned long next; 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12611da177e4SLinus Torvalds do { 12621da177e4SLinus Torvalds next = pud_addr_end(addr, end); 126397a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12641da177e4SLinus Torvalds continue; 126597a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 126697a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 126751c6f666SRobin Holt 126851c6f666SRobin Holt return addr; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 1271038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 127251c6f666SRobin Holt struct vm_area_struct *vma, 12731da177e4SLinus Torvalds unsigned long addr, unsigned long end, 127497a89413SPeter Zijlstra struct zap_details *details) 12751da177e4SLinus Torvalds { 12761da177e4SLinus Torvalds pgd_t *pgd; 12771da177e4SLinus Torvalds unsigned long next; 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 12801da177e4SLinus Torvalds details = NULL; 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds BUG_ON(addr >= end); 1283569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_start(); 12841da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 12851da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 12861da177e4SLinus Torvalds do { 12871da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 128897a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 12891da177e4SLinus Torvalds continue; 129097a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 129197a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 12921da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 1293569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_end(); 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds 1296f5cc4eefSAl Viro 1297f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 12981da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 12994f74d2c8SLinus Torvalds unsigned long end_addr, 13001da177e4SLinus Torvalds struct zap_details *details) 13011da177e4SLinus Torvalds { 1302f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 13031da177e4SLinus Torvalds unsigned long end; 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds if (start >= vma->vm_end) 1306f5cc4eefSAl Viro return; 13071da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 13081da177e4SLinus Torvalds if (end <= vma->vm_start) 1309f5cc4eefSAl Viro return; 13101da177e4SLinus Torvalds 1311cbc91f71SSrikar Dronamraju if (vma->vm_file) 1312cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1313cbc91f71SSrikar Dronamraju 1314b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 13155180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 13162ab64037Svenkatesh.pallipadi@intel.com 13178b2a1238SAl Viro if (start != end) { 131851c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1319a137e1ccSAndi Kleen /* 1320a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1321a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1322a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 1323*7aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1324a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 1325*7aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1326a137e1ccSAndi Kleen * before calling this function to clean up. 1327a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1328a137e1ccSAndi Kleen * safe to do nothing in this case. 1329a137e1ccSAndi Kleen */ 133024669e58SAneesh Kumar K.V if (vma->vm_file) { 133124669e58SAneesh Kumar K.V mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); 1332d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 133324669e58SAneesh Kumar K.V mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); 133424669e58SAneesh Kumar K.V } 133551c6f666SRobin Holt } else 1336038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 133797a89413SPeter Zijlstra } 133851c6f666SRobin Holt } 13391da177e4SLinus Torvalds 1340f5cc4eefSAl Viro /** 1341f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1342f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1343f5cc4eefSAl Viro * @vma: the starting vma 1344f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1345f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1346f5cc4eefSAl Viro * 1347f5cc4eefSAl Viro * Unmap all pages in the vma list. 1348f5cc4eefSAl Viro * 1349f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1350f5cc4eefSAl Viro * 1351f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1352f5cc4eefSAl Viro * 1353f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1354f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1355f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1356f5cc4eefSAl Viro * drops the lock and schedules. 1357f5cc4eefSAl Viro */ 1358f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1359f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13604f74d2c8SLinus Torvalds unsigned long end_addr) 1361f5cc4eefSAl Viro { 1362f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1363f5cc4eefSAl Viro 1364f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1365f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13664f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1367cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds /** 13711da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13721da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1373eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13741da177e4SLinus Torvalds * @size: number of bytes to zap 13751da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 1376f5cc4eefSAl Viro * 1377f5cc4eefSAl Viro * Caller must protect the VMA list 13781da177e4SLinus Torvalds */ 13797e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 13801da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 13811da177e4SLinus Torvalds { 13821da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1383d16dfc55SPeter Zijlstra struct mmu_gather tlb; 13847e027b14SLinus Torvalds unsigned long end = start + size; 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds lru_add_drain(); 13872b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 1388365e9c87SHugh Dickins update_hiwater_rss(mm); 13897e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 13907e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 13914f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 13927e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 13937e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds 1396c627f9ccSJack Steiner /** 1397f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1398f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1399f5cc4eefSAl Viro * @address: starting address of pages to zap 1400f5cc4eefSAl Viro * @size: number of bytes to zap 1401f5cc4eefSAl Viro * @details: details of nonlinear truncation or shared cache invalidation 1402f5cc4eefSAl Viro * 1403f5cc4eefSAl Viro * The range must fit into one VMA. 1404f5cc4eefSAl Viro */ 1405f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1406f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1407f5cc4eefSAl Viro { 1408f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1409f5cc4eefSAl Viro struct mmu_gather tlb; 1410f5cc4eefSAl Viro unsigned long end = address + size; 1411f5cc4eefSAl Viro 1412f5cc4eefSAl Viro lru_add_drain(); 14132b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, address, end); 1414f5cc4eefSAl Viro update_hiwater_rss(mm); 1415f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 14164f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1417f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1418f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds 1421c627f9ccSJack Steiner /** 1422c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1423c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1424c627f9ccSJack Steiner * @address: starting address of pages to zap 1425c627f9ccSJack Steiner * @size: number of bytes to zap 1426c627f9ccSJack Steiner * 1427c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1428c627f9ccSJack Steiner * 1429c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1430c627f9ccSJack Steiner * 1431c627f9ccSJack Steiner * Returns 0 if successful. 1432c627f9ccSJack Steiner */ 1433c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1434c627f9ccSJack Steiner unsigned long size) 1435c627f9ccSJack Steiner { 1436c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1437c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1438c627f9ccSJack Steiner return -1; 1439f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1440c627f9ccSJack Steiner return 0; 1441c627f9ccSJack Steiner } 1442c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1443c627f9ccSJack Steiner 1444142762bdSJohannes Weiner /** 1445240aadeeSMichel Lespinasse * follow_page_mask - look up a page descriptor from a user-virtual address 1446142762bdSJohannes Weiner * @vma: vm_area_struct mapping @address 1447142762bdSJohannes Weiner * @address: virtual address to look up 1448142762bdSJohannes Weiner * @flags: flags modifying lookup behaviour 1449240aadeeSMichel Lespinasse * @page_mask: on output, *page_mask is set according to the size of the page 1450142762bdSJohannes Weiner * 1451142762bdSJohannes Weiner * @flags can have FOLL_ flags set, defined in <linux/mm.h> 1452142762bdSJohannes Weiner * 1453142762bdSJohannes Weiner * Returns the mapped (struct page *), %NULL if no mapping exists, or 1454142762bdSJohannes Weiner * an error pointer if there is a mapping to something not represented 1455142762bdSJohannes Weiner * by a page descriptor (see also vm_normal_page()). 14561da177e4SLinus Torvalds */ 1457240aadeeSMichel Lespinasse struct page *follow_page_mask(struct vm_area_struct *vma, 1458240aadeeSMichel Lespinasse unsigned long address, unsigned int flags, 1459240aadeeSMichel Lespinasse unsigned int *page_mask) 14601da177e4SLinus Torvalds { 14611da177e4SLinus Torvalds pgd_t *pgd; 14621da177e4SLinus Torvalds pud_t *pud; 14631da177e4SLinus Torvalds pmd_t *pmd; 14641da177e4SLinus Torvalds pte_t *ptep, pte; 1465deceb6cdSHugh Dickins spinlock_t *ptl; 14661da177e4SLinus Torvalds struct page *page; 14676aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 14681da177e4SLinus Torvalds 1469240aadeeSMichel Lespinasse *page_mask = 0; 1470240aadeeSMichel Lespinasse 1471deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 1472deceb6cdSHugh Dickins if (!IS_ERR(page)) { 1473deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1474deceb6cdSHugh Dickins goto out; 1475deceb6cdSHugh Dickins } 14761da177e4SLinus Torvalds 1477deceb6cdSHugh Dickins page = NULL; 14781da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 14791da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 1480deceb6cdSHugh Dickins goto no_page_table; 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds pud = pud_offset(pgd, address); 1483ceb86879SAndi Kleen if (pud_none(*pud)) 1484ceb86879SAndi Kleen goto no_page_table; 14858a07651eSHugh Dickins if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { 1486e632a938SNaoya Horiguchi if (flags & FOLL_GET) 1487e632a938SNaoya Horiguchi goto out; 1488ceb86879SAndi Kleen page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); 1489ceb86879SAndi Kleen goto out; 1490ceb86879SAndi Kleen } 1491ceb86879SAndi Kleen if (unlikely(pud_bad(*pud))) 1492deceb6cdSHugh Dickins goto no_page_table; 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 1495aeed5fceSHugh Dickins if (pmd_none(*pmd)) 1496deceb6cdSHugh Dickins goto no_page_table; 149771e3aac0SAndrea Arcangeli if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { 1498deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 1499e632a938SNaoya Horiguchi if (flags & FOLL_GET) { 1500e632a938SNaoya Horiguchi /* 1501e632a938SNaoya Horiguchi * Refcount on tail pages are not well-defined and 1502e632a938SNaoya Horiguchi * shouldn't be taken. The caller should handle a NULL 1503e632a938SNaoya Horiguchi * return when trying to follow tail pages. 1504e632a938SNaoya Horiguchi */ 1505e632a938SNaoya Horiguchi if (PageHead(page)) 1506e632a938SNaoya Horiguchi get_page(page); 1507e632a938SNaoya Horiguchi else { 1508e632a938SNaoya Horiguchi page = NULL; 1509e632a938SNaoya Horiguchi goto out; 1510e632a938SNaoya Horiguchi } 1511e632a938SNaoya Horiguchi } 1512deceb6cdSHugh Dickins goto out; 1513deceb6cdSHugh Dickins } 15140b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pmd_numa(*pmd)) 15150b9d7052SAndrea Arcangeli goto no_page_table; 151671e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 1517500d65d4SAndrea Arcangeli if (flags & FOLL_SPLIT) { 1518e180377fSKirill A. Shutemov split_huge_page_pmd(vma, address, pmd); 1519500d65d4SAndrea Arcangeli goto split_fallthrough; 1520500d65d4SAndrea Arcangeli } 1521c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 152271e3aac0SAndrea Arcangeli if (likely(pmd_trans_huge(*pmd))) { 152371e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_splitting(*pmd))) { 1524c4088ebdSKirill A. Shutemov spin_unlock(ptl); 152571e3aac0SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 152671e3aac0SAndrea Arcangeli } else { 1527b676b293SDavid Rientjes page = follow_trans_huge_pmd(vma, address, 152871e3aac0SAndrea Arcangeli pmd, flags); 1529c4088ebdSKirill A. Shutemov spin_unlock(ptl); 1530240aadeeSMichel Lespinasse *page_mask = HPAGE_PMD_NR - 1; 153171e3aac0SAndrea Arcangeli goto out; 153271e3aac0SAndrea Arcangeli } 153371e3aac0SAndrea Arcangeli } else 1534c4088ebdSKirill A. Shutemov spin_unlock(ptl); 153571e3aac0SAndrea Arcangeli /* fall through */ 153671e3aac0SAndrea Arcangeli } 1537500d65d4SAndrea Arcangeli split_fallthrough: 1538aeed5fceSHugh Dickins if (unlikely(pmd_bad(*pmd))) 1539aeed5fceSHugh Dickins goto no_page_table; 1540aeed5fceSHugh Dickins 1541deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 15421da177e4SLinus Torvalds 15431da177e4SLinus Torvalds pte = *ptep; 15445117b3b8SHugh Dickins if (!pte_present(pte)) { 15455117b3b8SHugh Dickins swp_entry_t entry; 15465117b3b8SHugh Dickins /* 15475117b3b8SHugh Dickins * KSM's break_ksm() relies upon recognizing a ksm page 15485117b3b8SHugh Dickins * even while it is being migrated, so for that case we 15495117b3b8SHugh Dickins * need migration_entry_wait(). 15505117b3b8SHugh Dickins */ 15515117b3b8SHugh Dickins if (likely(!(flags & FOLL_MIGRATION))) 155289f5b7daSLinus Torvalds goto no_page; 15535117b3b8SHugh Dickins if (pte_none(pte) || pte_file(pte)) 15545117b3b8SHugh Dickins goto no_page; 15555117b3b8SHugh Dickins entry = pte_to_swp_entry(pte); 15565117b3b8SHugh Dickins if (!is_migration_entry(entry)) 15575117b3b8SHugh Dickins goto no_page; 15585117b3b8SHugh Dickins pte_unmap_unlock(ptep, ptl); 15595117b3b8SHugh Dickins migration_entry_wait(mm, pmd, address); 15605117b3b8SHugh Dickins goto split_fallthrough; 15615117b3b8SHugh Dickins } 15620b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pte_numa(pte)) 15630b9d7052SAndrea Arcangeli goto no_page; 1564deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 1565deceb6cdSHugh Dickins goto unlock; 1566a13ea5b7SHugh Dickins 15676aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 1568a13ea5b7SHugh Dickins if (unlikely(!page)) { 1569a13ea5b7SHugh Dickins if ((flags & FOLL_DUMP) || 157062eede62SHugh Dickins !is_zero_pfn(pte_pfn(pte))) 157189f5b7daSLinus Torvalds goto bad_page; 1572a13ea5b7SHugh Dickins page = pte_page(pte); 1573a13ea5b7SHugh Dickins } 1574deceb6cdSHugh Dickins 1575deceb6cdSHugh Dickins if (flags & FOLL_GET) 157670b50f94SAndrea Arcangeli get_page_foll(page); 1577deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 1578deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 1579deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 1580f33ea7f4SNick Piggin set_page_dirty(page); 1581bd775c42SKOSAKI Motohiro /* 1582bd775c42SKOSAKI Motohiro * pte_mkyoung() would be more correct here, but atomic care 1583bd775c42SKOSAKI Motohiro * is needed to avoid losing the dirty bit: it is easier to use 1584bd775c42SKOSAKI Motohiro * mark_page_accessed(). 1585bd775c42SKOSAKI Motohiro */ 15861da177e4SLinus Torvalds mark_page_accessed(page); 15871da177e4SLinus Torvalds } 1588a1fde08cSLinus Torvalds if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { 1589110d74a9SMichel Lespinasse /* 1590110d74a9SMichel Lespinasse * The preliminary mapping check is mainly to avoid the 1591110d74a9SMichel Lespinasse * pointless overhead of lock_page on the ZERO_PAGE 1592110d74a9SMichel Lespinasse * which might bounce very badly if there is contention. 1593110d74a9SMichel Lespinasse * 1594110d74a9SMichel Lespinasse * If the page is already locked, we don't need to 1595110d74a9SMichel Lespinasse * handle it now - vmscan will handle it later if and 1596110d74a9SMichel Lespinasse * when it attempts to reclaim the page. 1597110d74a9SMichel Lespinasse */ 1598110d74a9SMichel Lespinasse if (page->mapping && trylock_page(page)) { 1599110d74a9SMichel Lespinasse lru_add_drain(); /* push cached pages to LRU */ 1600110d74a9SMichel Lespinasse /* 1601e6c509f8SHugh Dickins * Because we lock page here, and migration is 1602e6c509f8SHugh Dickins * blocked by the pte's page reference, and we 1603e6c509f8SHugh Dickins * know the page is still mapped, we don't even 1604e6c509f8SHugh Dickins * need to check for file-cache page truncation. 1605110d74a9SMichel Lespinasse */ 1606110d74a9SMichel Lespinasse mlock_vma_page(page); 1607110d74a9SMichel Lespinasse unlock_page(page); 1608110d74a9SMichel Lespinasse } 1609110d74a9SMichel Lespinasse } 1610deceb6cdSHugh Dickins unlock: 1611deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 16121da177e4SLinus Torvalds out: 1613deceb6cdSHugh Dickins return page; 1614deceb6cdSHugh Dickins 161589f5b7daSLinus Torvalds bad_page: 161689f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 161789f5b7daSLinus Torvalds return ERR_PTR(-EFAULT); 161889f5b7daSLinus Torvalds 161989f5b7daSLinus Torvalds no_page: 162089f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 162189f5b7daSLinus Torvalds if (!pte_none(pte)) 162289f5b7daSLinus Torvalds return page; 16238e4b9a60SHugh Dickins 1624deceb6cdSHugh Dickins no_page_table: 1625deceb6cdSHugh Dickins /* 1626deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 16278e4b9a60SHugh Dickins * has touched so far, we don't want to allocate unnecessary pages or 16288e4b9a60SHugh Dickins * page tables. Return error instead of NULL to skip handle_mm_fault, 16298e4b9a60SHugh Dickins * then get_dump_page() will return NULL to leave a hole in the dump. 16308e4b9a60SHugh Dickins * But we can only make this optimization where a hole would surely 16318e4b9a60SHugh Dickins * be zero-filled if handle_mm_fault() actually did handle it. 1632deceb6cdSHugh Dickins */ 16338e4b9a60SHugh Dickins if ((flags & FOLL_DUMP) && 16348e4b9a60SHugh Dickins (!vma->vm_ops || !vma->vm_ops->fault)) 16358e4b9a60SHugh Dickins return ERR_PTR(-EFAULT); 1636deceb6cdSHugh Dickins return page; 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds 163995042f9eSLinus Torvalds static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) 164095042f9eSLinus Torvalds { 1641a09a79f6SMikulas Patocka return stack_guard_page_start(vma, addr) || 1642a09a79f6SMikulas Patocka stack_guard_page_end(vma, addr+PAGE_SIZE); 164395042f9eSLinus Torvalds } 164495042f9eSLinus Torvalds 16450014bd99SHuang Ying /** 16460014bd99SHuang Ying * __get_user_pages() - pin user pages in memory 16470014bd99SHuang Ying * @tsk: task_struct of target task 16480014bd99SHuang Ying * @mm: mm_struct of target mm 16490014bd99SHuang Ying * @start: starting user address 16500014bd99SHuang Ying * @nr_pages: number of pages from start to pin 16510014bd99SHuang Ying * @gup_flags: flags modifying pin behaviour 16520014bd99SHuang Ying * @pages: array that receives pointers to the pages pinned. 16530014bd99SHuang Ying * Should be at least nr_pages long. Or NULL, if caller 16540014bd99SHuang Ying * only intends to ensure the pages are faulted in. 16550014bd99SHuang Ying * @vmas: array of pointers to vmas corresponding to each page. 16560014bd99SHuang Ying * Or NULL if the caller does not require them. 16570014bd99SHuang Ying * @nonblocking: whether waiting for disk IO or mmap_sem contention 16580014bd99SHuang Ying * 16590014bd99SHuang Ying * Returns number of pages pinned. This may be fewer than the number 16600014bd99SHuang Ying * requested. If nr_pages is 0 or negative, returns 0. If no pages 16610014bd99SHuang Ying * were pinned, returns -errno. Each page returned must be released 16620014bd99SHuang Ying * with a put_page() call when it is finished with. vmas will only 16630014bd99SHuang Ying * remain valid while mmap_sem is held. 16640014bd99SHuang Ying * 16650014bd99SHuang Ying * Must be called with mmap_sem held for read or write. 16660014bd99SHuang Ying * 16670014bd99SHuang Ying * __get_user_pages walks a process's page tables and takes a reference to 16680014bd99SHuang Ying * each struct page that each user address corresponds to at a given 16690014bd99SHuang Ying * instant. That is, it takes the page that would be accessed if a user 16700014bd99SHuang Ying * thread accesses the given user virtual address at that instant. 16710014bd99SHuang Ying * 16720014bd99SHuang Ying * This does not guarantee that the page exists in the user mappings when 16730014bd99SHuang Ying * __get_user_pages returns, and there may even be a completely different 16740014bd99SHuang Ying * page there in some cases (eg. if mmapped pagecache has been invalidated 16750014bd99SHuang Ying * and subsequently re faulted). However it does guarantee that the page 16760014bd99SHuang Ying * won't be freed completely. And mostly callers simply care that the page 16770014bd99SHuang Ying * contains data that was valid *at some point in time*. Typically, an IO 16780014bd99SHuang Ying * or similar operation cannot guarantee anything stronger anyway because 16790014bd99SHuang Ying * locks can't be held over the syscall boundary. 16800014bd99SHuang Ying * 16810014bd99SHuang Ying * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If 16820014bd99SHuang Ying * the page is written to, set_page_dirty (or set_page_dirty_lock, as 16830014bd99SHuang Ying * appropriate) must be called after the page is finished with, and 16840014bd99SHuang Ying * before put_page is called. 16850014bd99SHuang Ying * 16860014bd99SHuang Ying * If @nonblocking != NULL, __get_user_pages will not wait for disk IO 16870014bd99SHuang Ying * or mmap_sem contention, and if waiting is needed to pin all pages, 16880014bd99SHuang Ying * *@nonblocking will be set to 0. 16890014bd99SHuang Ying * 16900014bd99SHuang Ying * In most cases, get_user_pages or get_user_pages_fast should be used 16910014bd99SHuang Ying * instead of __get_user_pages. __get_user_pages should be used only if 16920014bd99SHuang Ying * you need some special @gup_flags. 16930014bd99SHuang Ying */ 169428a35716SMichel Lespinasse long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 169528a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 169628a35716SMichel Lespinasse unsigned int gup_flags, struct page **pages, 169728a35716SMichel Lespinasse struct vm_area_struct **vmas, int *nonblocking) 16981da177e4SLinus Torvalds { 169928a35716SMichel Lespinasse long i; 170058fa879eSHugh Dickins unsigned long vm_flags; 1701240aadeeSMichel Lespinasse unsigned int page_mask; 17021da177e4SLinus Torvalds 170328a35716SMichel Lespinasse if (!nr_pages) 1704900cf086SJonathan Corbet return 0; 170558fa879eSHugh Dickins 170658fa879eSHugh Dickins VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); 170758fa879eSHugh Dickins 17081da177e4SLinus Torvalds /* 17090b9d7052SAndrea Arcangeli * If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault 17100b9d7052SAndrea Arcangeli * would be called on PROT_NONE ranges. We must never invoke 17110b9d7052SAndrea Arcangeli * handle_mm_fault on PROT_NONE ranges or the NUMA hinting 17120b9d7052SAndrea Arcangeli * page faults would unprotect the PROT_NONE ranges if 17130b9d7052SAndrea Arcangeli * _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd 17140b9d7052SAndrea Arcangeli * bitflag. So to avoid that, don't set FOLL_NUMA if 17150b9d7052SAndrea Arcangeli * FOLL_FORCE is set. 17160b9d7052SAndrea Arcangeli */ 17170b9d7052SAndrea Arcangeli if (!(gup_flags & FOLL_FORCE)) 17180b9d7052SAndrea Arcangeli gup_flags |= FOLL_NUMA; 17190b9d7052SAndrea Arcangeli 17201da177e4SLinus Torvalds i = 0; 17211da177e4SLinus Torvalds 17221da177e4SLinus Torvalds do { 17231da177e4SLinus Torvalds struct vm_area_struct *vma; 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 1726e7f22e20SStephen Wilson if (!vma && in_gate_area(mm, start)) { 17271da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 17281da177e4SLinus Torvalds pgd_t *pgd; 17291da177e4SLinus Torvalds pud_t *pud; 17301da177e4SLinus Torvalds pmd_t *pmd; 17311da177e4SLinus Torvalds pte_t *pte; 1732b291f000SNick Piggin 1733b291f000SNick Piggin /* user gate pages are read-only */ 173458fa879eSHugh Dickins if (gup_flags & FOLL_WRITE) 1735cda540acSHugh Dickins goto efault; 17361da177e4SLinus Torvalds if (pg > TASK_SIZE) 17371da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 17381da177e4SLinus Torvalds else 17391da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 17401da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 17411da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 17421da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 17431da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1744690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1745cda540acSHugh Dickins goto efault; 1746f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 17471da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1748690dbe1cSHugh Dickins if (pte_none(*pte)) { 1749690dbe1cSHugh Dickins pte_unmap(pte); 1750cda540acSHugh Dickins goto efault; 1751690dbe1cSHugh Dickins } 175295042f9eSLinus Torvalds vma = get_gate_vma(mm); 17531da177e4SLinus Torvalds if (pages) { 1754de51257aSHugh Dickins struct page *page; 1755de51257aSHugh Dickins 175695042f9eSLinus Torvalds page = vm_normal_page(vma, start, *pte); 1757de51257aSHugh Dickins if (!page) { 1758de51257aSHugh Dickins if (!(gup_flags & FOLL_DUMP) && 1759de51257aSHugh Dickins is_zero_pfn(pte_pfn(*pte))) 1760de51257aSHugh Dickins page = pte_page(*pte); 1761de51257aSHugh Dickins else { 1762de51257aSHugh Dickins pte_unmap(pte); 1763cda540acSHugh Dickins goto efault; 1764de51257aSHugh Dickins } 1765de51257aSHugh Dickins } 17666aab341eSLinus Torvalds pages[i] = page; 17676aab341eSLinus Torvalds get_page(page); 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds pte_unmap(pte); 1770240aadeeSMichel Lespinasse page_mask = 0; 177195042f9eSLinus Torvalds goto next_page; 17721da177e4SLinus Torvalds } 17731da177e4SLinus Torvalds 1774cda540acSHugh Dickins if (!vma) 1775cda540acSHugh Dickins goto efault; 1776cda540acSHugh Dickins vm_flags = vma->vm_flags; 1777cda540acSHugh Dickins if (vm_flags & (VM_IO | VM_PFNMAP)) 1778cda540acSHugh Dickins goto efault; 1779cda540acSHugh Dickins 1780cda540acSHugh Dickins if (gup_flags & FOLL_WRITE) { 1781cda540acSHugh Dickins if (!(vm_flags & VM_WRITE)) { 1782cda540acSHugh Dickins if (!(gup_flags & FOLL_FORCE)) 1783cda540acSHugh Dickins goto efault; 1784cda540acSHugh Dickins /* 1785cda540acSHugh Dickins * We used to let the write,force case do COW 1786cda540acSHugh Dickins * in a VM_MAYWRITE VM_SHARED !VM_WRITE vma, so 1787cda540acSHugh Dickins * ptrace could set a breakpoint in a read-only 1788cda540acSHugh Dickins * mapping of an executable, without corrupting 1789cda540acSHugh Dickins * the file (yet only when that file had been 1790cda540acSHugh Dickins * opened for writing!). Anon pages in shared 1791cda540acSHugh Dickins * mappings are surprising: now just reject it. 1792cda540acSHugh Dickins */ 1793cda540acSHugh Dickins if (!is_cow_mapping(vm_flags)) { 1794cda540acSHugh Dickins WARN_ON_ONCE(vm_flags & VM_MAYWRITE); 1795cda540acSHugh Dickins goto efault; 1796cda540acSHugh Dickins } 1797cda540acSHugh Dickins } 1798cda540acSHugh Dickins } else { 1799cda540acSHugh Dickins if (!(vm_flags & VM_READ)) { 1800cda540acSHugh Dickins if (!(gup_flags & FOLL_FORCE)) 1801cda540acSHugh Dickins goto efault; 1802cda540acSHugh Dickins /* 1803cda540acSHugh Dickins * Is there actually any vma we can reach here 1804cda540acSHugh Dickins * which does not have VM_MAYREAD set? 1805cda540acSHugh Dickins */ 1806cda540acSHugh Dickins if (!(vm_flags & VM_MAYREAD)) 1807cda540acSHugh Dickins goto efault; 1808cda540acSHugh Dickins } 1809cda540acSHugh Dickins } 18101da177e4SLinus Torvalds 18112a15efc9SHugh Dickins if (is_vm_hugetlb_page(vma)) { 18122a15efc9SHugh Dickins i = follow_hugetlb_page(mm, vma, pages, vmas, 181358fa879eSHugh Dickins &start, &nr_pages, i, gup_flags); 18142a15efc9SHugh Dickins continue; 18152a15efc9SHugh Dickins } 1816deceb6cdSHugh Dickins 18171da177e4SLinus Torvalds do { 181808ef4729SHugh Dickins struct page *page; 181958fa879eSHugh Dickins unsigned int foll_flags = gup_flags; 1820240aadeeSMichel Lespinasse unsigned int page_increm; 18211da177e4SLinus Torvalds 1822462e00ccSEthan Solomita /* 18234779280dSYing Han * If we have a pending SIGKILL, don't keep faulting 18241c3aff1cSHugh Dickins * pages and potentially allocating memory. 1825462e00ccSEthan Solomita */ 18261c3aff1cSHugh Dickins if (unlikely(fatal_signal_pending(current))) 18274779280dSYing Han return i ? i : -ERESTARTSYS; 1828462e00ccSEthan Solomita 1829deceb6cdSHugh Dickins cond_resched(); 1830240aadeeSMichel Lespinasse while (!(page = follow_page_mask(vma, start, 1831240aadeeSMichel Lespinasse foll_flags, &page_mask))) { 1832a68d2ebcSLinus Torvalds int ret; 183353a7706dSMichel Lespinasse unsigned int fault_flags = 0; 183453a7706dSMichel Lespinasse 1835a09a79f6SMikulas Patocka /* For mlock, just skip the stack guard page. */ 1836a09a79f6SMikulas Patocka if (foll_flags & FOLL_MLOCK) { 1837a09a79f6SMikulas Patocka if (stack_guard_page(vma, start)) 1838a09a79f6SMikulas Patocka goto next_page; 1839a09a79f6SMikulas Patocka } 184053a7706dSMichel Lespinasse if (foll_flags & FOLL_WRITE) 184153a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_WRITE; 184253a7706dSMichel Lespinasse if (nonblocking) 184353a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_ALLOW_RETRY; 1844318b275fSGleb Natapov if (foll_flags & FOLL_NOWAIT) 1845318b275fSGleb Natapov fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); 1846d06063ccSLinus Torvalds 1847d26ed650SHugh Dickins ret = handle_mm_fault(mm, vma, start, 184853a7706dSMichel Lespinasse fault_flags); 1849d26ed650SHugh Dickins 185083c54070SNick Piggin if (ret & VM_FAULT_ERROR) { 185183c54070SNick Piggin if (ret & VM_FAULT_OOM) 185283c54070SNick Piggin return i ? i : -ENOMEM; 185369ebb83eSHuang Ying if (ret & (VM_FAULT_HWPOISON | 185469ebb83eSHuang Ying VM_FAULT_HWPOISON_LARGE)) { 185569ebb83eSHuang Ying if (i) 185669ebb83eSHuang Ying return i; 185769ebb83eSHuang Ying else if (gup_flags & FOLL_HWPOISON) 185869ebb83eSHuang Ying return -EHWPOISON; 185969ebb83eSHuang Ying else 186069ebb83eSHuang Ying return -EFAULT; 186169ebb83eSHuang Ying } 186269ebb83eSHuang Ying if (ret & VM_FAULT_SIGBUS) 1863cda540acSHugh Dickins goto efault; 186483c54070SNick Piggin BUG(); 186583c54070SNick Piggin } 1866e7f22e20SStephen Wilson 1867e7f22e20SStephen Wilson if (tsk) { 186883c54070SNick Piggin if (ret & VM_FAULT_MAJOR) 186983c54070SNick Piggin tsk->maj_flt++; 187083c54070SNick Piggin else 187183c54070SNick Piggin tsk->min_flt++; 1872e7f22e20SStephen Wilson } 187383c54070SNick Piggin 187453a7706dSMichel Lespinasse if (ret & VM_FAULT_RETRY) { 1875318b275fSGleb Natapov if (nonblocking) 187653a7706dSMichel Lespinasse *nonblocking = 0; 187753a7706dSMichel Lespinasse return i; 187853a7706dSMichel Lespinasse } 187953a7706dSMichel Lespinasse 1880f33ea7f4SNick Piggin /* 188183c54070SNick Piggin * The VM_FAULT_WRITE bit tells us that 188283c54070SNick Piggin * do_wp_page has broken COW when necessary, 188383c54070SNick Piggin * even if maybe_mkwrite decided not to set 188483c54070SNick Piggin * pte_write. We can thus safely do subsequent 1885878b63acSHugh Dickins * page lookups as if they were reads. But only 1886878b63acSHugh Dickins * do so when looping for pte_write is futile: 1887878b63acSHugh Dickins * in some cases userspace may also be wanting 1888878b63acSHugh Dickins * to write to the gotten user page, which a 1889878b63acSHugh Dickins * read fault here might prevent (a readonly 1890878b63acSHugh Dickins * page might get reCOWed by userspace write). 1891f33ea7f4SNick Piggin */ 1892878b63acSHugh Dickins if ((ret & VM_FAULT_WRITE) && 1893878b63acSHugh Dickins !(vma->vm_flags & VM_WRITE)) 1894deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1895a68d2ebcSLinus Torvalds 18967f7bbbe5SBenjamin Herrenschmidt cond_resched(); 18971da177e4SLinus Torvalds } 189889f5b7daSLinus Torvalds if (IS_ERR(page)) 189989f5b7daSLinus Torvalds return i ? i : PTR_ERR(page); 19001da177e4SLinus Torvalds if (pages) { 190108ef4729SHugh Dickins pages[i] = page; 190203beb076SJames Bottomley 1903a6f36be3SRussell King flush_anon_page(vma, page, start); 190408ef4729SHugh Dickins flush_dcache_page(page); 1905240aadeeSMichel Lespinasse page_mask = 0; 19061da177e4SLinus Torvalds } 190795042f9eSLinus Torvalds next_page: 1908240aadeeSMichel Lespinasse if (vmas) { 19091da177e4SLinus Torvalds vmas[i] = vma; 1910240aadeeSMichel Lespinasse page_mask = 0; 1911240aadeeSMichel Lespinasse } 1912240aadeeSMichel Lespinasse page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask); 1913240aadeeSMichel Lespinasse if (page_increm > nr_pages) 1914240aadeeSMichel Lespinasse page_increm = nr_pages; 1915240aadeeSMichel Lespinasse i += page_increm; 1916240aadeeSMichel Lespinasse start += page_increm * PAGE_SIZE; 1917240aadeeSMichel Lespinasse nr_pages -= page_increm; 19189d73777eSPeter Zijlstra } while (nr_pages && start < vma->vm_end); 19199d73777eSPeter Zijlstra } while (nr_pages); 19201da177e4SLinus Torvalds return i; 1921cda540acSHugh Dickins efault: 1922cda540acSHugh Dickins return i ? : -EFAULT; 19231da177e4SLinus Torvalds } 19240014bd99SHuang Ying EXPORT_SYMBOL(__get_user_pages); 1925b291f000SNick Piggin 19262efaca92SBenjamin Herrenschmidt /* 19272efaca92SBenjamin Herrenschmidt * fixup_user_fault() - manually resolve a user page fault 19282efaca92SBenjamin Herrenschmidt * @tsk: the task_struct to use for page fault accounting, or 19292efaca92SBenjamin Herrenschmidt * NULL if faults are not to be recorded. 19302efaca92SBenjamin Herrenschmidt * @mm: mm_struct of target mm 19312efaca92SBenjamin Herrenschmidt * @address: user address 19322efaca92SBenjamin Herrenschmidt * @fault_flags:flags to pass down to handle_mm_fault() 19332efaca92SBenjamin Herrenschmidt * 19342efaca92SBenjamin Herrenschmidt * This is meant to be called in the specific scenario where for locking reasons 19352efaca92SBenjamin Herrenschmidt * we try to access user memory in atomic context (within a pagefault_disable() 19362efaca92SBenjamin Herrenschmidt * section), this returns -EFAULT, and we want to resolve the user fault before 19372efaca92SBenjamin Herrenschmidt * trying again. 19382efaca92SBenjamin Herrenschmidt * 19392efaca92SBenjamin Herrenschmidt * Typically this is meant to be used by the futex code. 19402efaca92SBenjamin Herrenschmidt * 19412efaca92SBenjamin Herrenschmidt * The main difference with get_user_pages() is that this function will 19422efaca92SBenjamin Herrenschmidt * unconditionally call handle_mm_fault() which will in turn perform all the 19432efaca92SBenjamin Herrenschmidt * necessary SW fixup of the dirty and young bits in the PTE, while 19442efaca92SBenjamin Herrenschmidt * handle_mm_fault() only guarantees to update these in the struct page. 19452efaca92SBenjamin Herrenschmidt * 19462efaca92SBenjamin Herrenschmidt * This is important for some architectures where those bits also gate the 19472efaca92SBenjamin Herrenschmidt * access permission to the page because they are maintained in software. On 19482efaca92SBenjamin Herrenschmidt * such architectures, gup() will not be enough to make a subsequent access 19492efaca92SBenjamin Herrenschmidt * succeed. 19502efaca92SBenjamin Herrenschmidt * 19512efaca92SBenjamin Herrenschmidt * This should be called with the mm_sem held for read. 19522efaca92SBenjamin Herrenschmidt */ 19532efaca92SBenjamin Herrenschmidt int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm, 19542efaca92SBenjamin Herrenschmidt unsigned long address, unsigned int fault_flags) 19552efaca92SBenjamin Herrenschmidt { 19562efaca92SBenjamin Herrenschmidt struct vm_area_struct *vma; 19572efaca92SBenjamin Herrenschmidt int ret; 19582efaca92SBenjamin Herrenschmidt 19592efaca92SBenjamin Herrenschmidt vma = find_extend_vma(mm, address); 19602efaca92SBenjamin Herrenschmidt if (!vma || address < vma->vm_start) 19612efaca92SBenjamin Herrenschmidt return -EFAULT; 19622efaca92SBenjamin Herrenschmidt 19632efaca92SBenjamin Herrenschmidt ret = handle_mm_fault(mm, vma, address, fault_flags); 19642efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_ERROR) { 19652efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_OOM) 19662efaca92SBenjamin Herrenschmidt return -ENOMEM; 19672efaca92SBenjamin Herrenschmidt if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) 19682efaca92SBenjamin Herrenschmidt return -EHWPOISON; 19692efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_SIGBUS) 19702efaca92SBenjamin Herrenschmidt return -EFAULT; 19712efaca92SBenjamin Herrenschmidt BUG(); 19722efaca92SBenjamin Herrenschmidt } 19732efaca92SBenjamin Herrenschmidt if (tsk) { 19742efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_MAJOR) 19752efaca92SBenjamin Herrenschmidt tsk->maj_flt++; 19762efaca92SBenjamin Herrenschmidt else 19772efaca92SBenjamin Herrenschmidt tsk->min_flt++; 19782efaca92SBenjamin Herrenschmidt } 19792efaca92SBenjamin Herrenschmidt return 0; 19802efaca92SBenjamin Herrenschmidt } 19812efaca92SBenjamin Herrenschmidt 19822efaca92SBenjamin Herrenschmidt /* 1983d2bf6be8SNick Piggin * get_user_pages() - pin user pages in memory 1984e7f22e20SStephen Wilson * @tsk: the task_struct to use for page fault accounting, or 1985e7f22e20SStephen Wilson * NULL if faults are not to be recorded. 1986d2bf6be8SNick Piggin * @mm: mm_struct of target mm 1987d2bf6be8SNick Piggin * @start: starting user address 19889d73777eSPeter Zijlstra * @nr_pages: number of pages from start to pin 1989d2bf6be8SNick Piggin * @write: whether pages will be written to by the caller 1990cda540acSHugh Dickins * @force: whether to force access even when user mapping is currently 1991cda540acSHugh Dickins * protected (but never forces write access to shared mapping). 1992d2bf6be8SNick Piggin * @pages: array that receives pointers to the pages pinned. 1993d2bf6be8SNick Piggin * Should be at least nr_pages long. Or NULL, if caller 1994d2bf6be8SNick Piggin * only intends to ensure the pages are faulted in. 1995d2bf6be8SNick Piggin * @vmas: array of pointers to vmas corresponding to each page. 1996d2bf6be8SNick Piggin * Or NULL if the caller does not require them. 1997d2bf6be8SNick Piggin * 1998d2bf6be8SNick Piggin * Returns number of pages pinned. This may be fewer than the number 19999d73777eSPeter Zijlstra * requested. If nr_pages is 0 or negative, returns 0. If no pages 2000d2bf6be8SNick Piggin * were pinned, returns -errno. Each page returned must be released 2001d2bf6be8SNick Piggin * with a put_page() call when it is finished with. vmas will only 2002d2bf6be8SNick Piggin * remain valid while mmap_sem is held. 2003d2bf6be8SNick Piggin * 2004d2bf6be8SNick Piggin * Must be called with mmap_sem held for read or write. 2005d2bf6be8SNick Piggin * 2006d2bf6be8SNick Piggin * get_user_pages walks a process's page tables and takes a reference to 2007d2bf6be8SNick Piggin * each struct page that each user address corresponds to at a given 2008d2bf6be8SNick Piggin * instant. That is, it takes the page that would be accessed if a user 2009d2bf6be8SNick Piggin * thread accesses the given user virtual address at that instant. 2010d2bf6be8SNick Piggin * 2011d2bf6be8SNick Piggin * This does not guarantee that the page exists in the user mappings when 2012d2bf6be8SNick Piggin * get_user_pages returns, and there may even be a completely different 2013d2bf6be8SNick Piggin * page there in some cases (eg. if mmapped pagecache has been invalidated 2014d2bf6be8SNick Piggin * and subsequently re faulted). However it does guarantee that the page 2015d2bf6be8SNick Piggin * won't be freed completely. And mostly callers simply care that the page 2016d2bf6be8SNick Piggin * contains data that was valid *at some point in time*. Typically, an IO 2017d2bf6be8SNick Piggin * or similar operation cannot guarantee anything stronger anyway because 2018d2bf6be8SNick Piggin * locks can't be held over the syscall boundary. 2019d2bf6be8SNick Piggin * 2020d2bf6be8SNick Piggin * If write=0, the page must not be written to. If the page is written to, 2021d2bf6be8SNick Piggin * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called 2022d2bf6be8SNick Piggin * after the page is finished with, and before put_page is called. 2023d2bf6be8SNick Piggin * 2024d2bf6be8SNick Piggin * get_user_pages is typically used for fewer-copy IO operations, to get a 2025d2bf6be8SNick Piggin * handle on the memory by some means other than accesses via the user virtual 2026d2bf6be8SNick Piggin * addresses. The pages may be submitted for DMA to devices or accessed via 2027d2bf6be8SNick Piggin * their kernel linear mapping (via the kmap APIs). Care should be taken to 2028d2bf6be8SNick Piggin * use the correct cache flushing APIs. 2029d2bf6be8SNick Piggin * 2030d2bf6be8SNick Piggin * See also get_user_pages_fast, for performance critical applications. 2031d2bf6be8SNick Piggin */ 203228a35716SMichel Lespinasse long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 203328a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, int write, 203428a35716SMichel Lespinasse int force, struct page **pages, struct vm_area_struct **vmas) 2035b291f000SNick Piggin { 203658fa879eSHugh Dickins int flags = FOLL_TOUCH; 2037b291f000SNick Piggin 203858fa879eSHugh Dickins if (pages) 203958fa879eSHugh Dickins flags |= FOLL_GET; 2040b291f000SNick Piggin if (write) 204158fa879eSHugh Dickins flags |= FOLL_WRITE; 2042b291f000SNick Piggin if (force) 204358fa879eSHugh Dickins flags |= FOLL_FORCE; 2044b291f000SNick Piggin 204553a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 204653a7706dSMichel Lespinasse NULL); 2047b291f000SNick Piggin } 20481da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 20491da177e4SLinus Torvalds 2050f3e8fccdSHugh Dickins /** 2051f3e8fccdSHugh Dickins * get_dump_page() - pin user page in memory while writing it to core dump 2052f3e8fccdSHugh Dickins * @addr: user address 2053f3e8fccdSHugh Dickins * 2054f3e8fccdSHugh Dickins * Returns struct page pointer of user page pinned for dump, 2055f3e8fccdSHugh Dickins * to be freed afterwards by page_cache_release() or put_page(). 2056f3e8fccdSHugh Dickins * 2057f3e8fccdSHugh Dickins * Returns NULL on any kind of failure - a hole must then be inserted into 2058f3e8fccdSHugh Dickins * the corefile, to preserve alignment with its headers; and also returns 2059f3e8fccdSHugh Dickins * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - 2060f3e8fccdSHugh Dickins * allowing a hole to be left in the corefile to save diskspace. 2061f3e8fccdSHugh Dickins * 2062f3e8fccdSHugh Dickins * Called without mmap_sem, but after all other threads have been killed. 2063f3e8fccdSHugh Dickins */ 2064f3e8fccdSHugh Dickins #ifdef CONFIG_ELF_CORE 2065f3e8fccdSHugh Dickins struct page *get_dump_page(unsigned long addr) 2066f3e8fccdSHugh Dickins { 2067f3e8fccdSHugh Dickins struct vm_area_struct *vma; 2068f3e8fccdSHugh Dickins struct page *page; 2069f3e8fccdSHugh Dickins 2070f3e8fccdSHugh Dickins if (__get_user_pages(current, current->mm, addr, 1, 207153a7706dSMichel Lespinasse FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, 207253a7706dSMichel Lespinasse NULL) < 1) 2073f3e8fccdSHugh Dickins return NULL; 2074f3e8fccdSHugh Dickins flush_cache_page(vma, addr, page_to_pfn(page)); 2075f3e8fccdSHugh Dickins return page; 2076f3e8fccdSHugh Dickins } 2077f3e8fccdSHugh Dickins #endif /* CONFIG_ELF_CORE */ 2078f3e8fccdSHugh Dickins 207925ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 2080920c7a5dSHarvey Harrison spinlock_t **ptl) 2081c9cfcddfSLinus Torvalds { 2082c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 2083c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 2084c9cfcddfSLinus Torvalds if (pud) { 208549c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 2086f66055abSAndrea Arcangeli if (pmd) { 2087f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 2088c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 2089c9cfcddfSLinus Torvalds } 2090f66055abSAndrea Arcangeli } 2091c9cfcddfSLinus Torvalds return NULL; 2092c9cfcddfSLinus Torvalds } 2093c9cfcddfSLinus Torvalds 20941da177e4SLinus Torvalds /* 2095238f58d8SLinus Torvalds * This is the old fallback for page remapping. 2096238f58d8SLinus Torvalds * 2097238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 2098238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 2099238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 2100238f58d8SLinus Torvalds */ 2101423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 2102423bad60SNick Piggin struct page *page, pgprot_t prot) 2103238f58d8SLinus Torvalds { 2104423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2105238f58d8SLinus Torvalds int retval; 2106238f58d8SLinus Torvalds pte_t *pte; 2107238f58d8SLinus Torvalds spinlock_t *ptl; 2108238f58d8SLinus Torvalds 2109238f58d8SLinus Torvalds retval = -EINVAL; 2110a145dd41SLinus Torvalds if (PageAnon(page)) 21115b4e655eSKAMEZAWA Hiroyuki goto out; 2112238f58d8SLinus Torvalds retval = -ENOMEM; 2113238f58d8SLinus Torvalds flush_dcache_page(page); 2114c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 2115238f58d8SLinus Torvalds if (!pte) 21165b4e655eSKAMEZAWA Hiroyuki goto out; 2117238f58d8SLinus Torvalds retval = -EBUSY; 2118238f58d8SLinus Torvalds if (!pte_none(*pte)) 2119238f58d8SLinus Torvalds goto out_unlock; 2120238f58d8SLinus Torvalds 2121238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 2122238f58d8SLinus Torvalds get_page(page); 212334e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 2124238f58d8SLinus Torvalds page_add_file_rmap(page); 2125238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 2126238f58d8SLinus Torvalds 2127238f58d8SLinus Torvalds retval = 0; 21288a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 21298a9f3ccdSBalbir Singh return retval; 2130238f58d8SLinus Torvalds out_unlock: 2131238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 2132238f58d8SLinus Torvalds out: 2133238f58d8SLinus Torvalds return retval; 2134238f58d8SLinus Torvalds } 2135238f58d8SLinus Torvalds 2136bfa5bf6dSRolf Eike Beer /** 2137bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 2138bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2139bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 2140bfa5bf6dSRolf Eike Beer * @page: source kernel page 2141bfa5bf6dSRolf Eike Beer * 2142a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 2143a145dd41SLinus Torvalds * into a user vma. 2144a145dd41SLinus Torvalds * 2145a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2146a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2147a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 21488dfcc9baSNick Piggin * (see split_page()). 2149a145dd41SLinus Torvalds * 2150a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2151a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2152a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2153a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2154a145dd41SLinus Torvalds * ask for a shared writable mapping! 2155a145dd41SLinus Torvalds * 2156a145dd41SLinus Torvalds * The page does not need to be reserved. 21574b6e1e37SKonstantin Khlebnikov * 21584b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 21594b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 21604b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 21614b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2162a145dd41SLinus Torvalds */ 2163423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2164423bad60SNick Piggin struct page *page) 2165a145dd41SLinus Torvalds { 2166a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2167a145dd41SLinus Torvalds return -EFAULT; 2168a145dd41SLinus Torvalds if (!page_count(page)) 2169a145dd41SLinus Torvalds return -EINVAL; 21704b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 21714b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 21724b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 21734b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 21744b6e1e37SKonstantin Khlebnikov } 2175423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2176a145dd41SLinus Torvalds } 2177e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2178a145dd41SLinus Torvalds 2179423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2180423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 2181423bad60SNick Piggin { 2182423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2183423bad60SNick Piggin int retval; 2184423bad60SNick Piggin pte_t *pte, entry; 2185423bad60SNick Piggin spinlock_t *ptl; 2186423bad60SNick Piggin 2187423bad60SNick Piggin retval = -ENOMEM; 2188423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2189423bad60SNick Piggin if (!pte) 2190423bad60SNick Piggin goto out; 2191423bad60SNick Piggin retval = -EBUSY; 2192423bad60SNick Piggin if (!pte_none(*pte)) 2193423bad60SNick Piggin goto out_unlock; 2194423bad60SNick Piggin 2195423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 2196423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 2197423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 21984b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2199423bad60SNick Piggin 2200423bad60SNick Piggin retval = 0; 2201423bad60SNick Piggin out_unlock: 2202423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 2203423bad60SNick Piggin out: 2204423bad60SNick Piggin return retval; 2205423bad60SNick Piggin } 2206423bad60SNick Piggin 2207e0dc0d8fSNick Piggin /** 2208e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 2209e0dc0d8fSNick Piggin * @vma: user vma to map to 2210e0dc0d8fSNick Piggin * @addr: target user address of this page 2211e0dc0d8fSNick Piggin * @pfn: source kernel pfn 2212e0dc0d8fSNick Piggin * 2213c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 2214e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 2215e0dc0d8fSNick Piggin * 2216e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 2217e0dc0d8fSNick Piggin * in that case the handler should return NULL. 22180d71d10aSNick Piggin * 22190d71d10aSNick Piggin * vma cannot be a COW mapping. 22200d71d10aSNick Piggin * 22210d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 22220d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 2223e0dc0d8fSNick Piggin */ 2224e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2225e0dc0d8fSNick Piggin unsigned long pfn) 2226e0dc0d8fSNick Piggin { 22272ab64037Svenkatesh.pallipadi@intel.com int ret; 2228e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 22297e675137SNick Piggin /* 22307e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 22317e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 22327e675137SNick Piggin * consistency in testing and feature parity among all, so we should 22337e675137SNick Piggin * try to keep these invariants in place for everybody. 22347e675137SNick Piggin */ 2235b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 2236b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 2237b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 2238b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 2239b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 2240e0dc0d8fSNick Piggin 2241423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2242423bad60SNick Piggin return -EFAULT; 22435180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 22442ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 22452ab64037Svenkatesh.pallipadi@intel.com 2246e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 22472ab64037Svenkatesh.pallipadi@intel.com 22482ab64037Svenkatesh.pallipadi@intel.com return ret; 2249e0dc0d8fSNick Piggin } 2250e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 2251e0dc0d8fSNick Piggin 2252423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 2253423bad60SNick Piggin unsigned long pfn) 2254423bad60SNick Piggin { 2255423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 2256423bad60SNick Piggin 2257423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2258423bad60SNick Piggin return -EFAULT; 2259423bad60SNick Piggin 2260423bad60SNick Piggin /* 2261423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2262423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2263423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 226462eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 226562eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2266423bad60SNick Piggin */ 2267423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 2268423bad60SNick Piggin struct page *page; 2269423bad60SNick Piggin 2270423bad60SNick Piggin page = pfn_to_page(pfn); 2271423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2272423bad60SNick Piggin } 2273423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 2274423bad60SNick Piggin } 2275423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 2276423bad60SNick Piggin 2277a145dd41SLinus Torvalds /* 22781da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 22791da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 22801da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 22811da177e4SLinus Torvalds */ 22821da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 22831da177e4SLinus Torvalds unsigned long addr, unsigned long end, 22841da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 22851da177e4SLinus Torvalds { 22861da177e4SLinus Torvalds pte_t *pte; 2287c74df32cSHugh Dickins spinlock_t *ptl; 22881da177e4SLinus Torvalds 2289c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 22901da177e4SLinus Torvalds if (!pte) 22911da177e4SLinus Torvalds return -ENOMEM; 22926606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 22931da177e4SLinus Torvalds do { 22941da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 22957e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 22961da177e4SLinus Torvalds pfn++; 22971da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 22986606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 2299c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 23001da177e4SLinus Torvalds return 0; 23011da177e4SLinus Torvalds } 23021da177e4SLinus Torvalds 23031da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23041da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23051da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23061da177e4SLinus Torvalds { 23071da177e4SLinus Torvalds pmd_t *pmd; 23081da177e4SLinus Torvalds unsigned long next; 23091da177e4SLinus Torvalds 23101da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23111da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 23121da177e4SLinus Torvalds if (!pmd) 23131da177e4SLinus Torvalds return -ENOMEM; 2314f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 23151da177e4SLinus Torvalds do { 23161da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 23171da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 23181da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 23191da177e4SLinus Torvalds return -ENOMEM; 23201da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 23211da177e4SLinus Torvalds return 0; 23221da177e4SLinus Torvalds } 23231da177e4SLinus Torvalds 23241da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 23251da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23261da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23271da177e4SLinus Torvalds { 23281da177e4SLinus Torvalds pud_t *pud; 23291da177e4SLinus Torvalds unsigned long next; 23301da177e4SLinus Torvalds 23311da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23321da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 23331da177e4SLinus Torvalds if (!pud) 23341da177e4SLinus Torvalds return -ENOMEM; 23351da177e4SLinus Torvalds do { 23361da177e4SLinus Torvalds next = pud_addr_end(addr, end); 23371da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 23381da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 23391da177e4SLinus Torvalds return -ENOMEM; 23401da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 23411da177e4SLinus Torvalds return 0; 23421da177e4SLinus Torvalds } 23431da177e4SLinus Torvalds 2344bfa5bf6dSRolf Eike Beer /** 2345bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 2346bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2347bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 2348bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 2349bfa5bf6dSRolf Eike Beer * @size: size of map area 2350bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 2351bfa5bf6dSRolf Eike Beer * 2352bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 2353bfa5bf6dSRolf Eike Beer */ 23541da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 23551da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 23561da177e4SLinus Torvalds { 23571da177e4SLinus Torvalds pgd_t *pgd; 23581da177e4SLinus Torvalds unsigned long next; 23592d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 23601da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23611da177e4SLinus Torvalds int err; 23621da177e4SLinus Torvalds 23631da177e4SLinus Torvalds /* 23641da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 23651da177e4SLinus Torvalds * rest of the world about it: 23661da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 23671da177e4SLinus Torvalds * (accesses can have side effects). 23686aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 23696aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 23706aab341eSLinus Torvalds * with them. 2371314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2372314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2373314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2374314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2375fb155c16SLinus Torvalds * 2376fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2377fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2378fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2379b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 23801da177e4SLinus Torvalds */ 2381b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2382b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2383b3b9c293SKonstantin Khlebnikov return -EINVAL; 23846aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2385b3b9c293SKonstantin Khlebnikov } 2386b3b9c293SKonstantin Khlebnikov 2387b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 2388b3b9c293SKonstantin Khlebnikov if (err) 23893c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 2390fb155c16SLinus Torvalds 2391314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 23921da177e4SLinus Torvalds 23931da177e4SLinus Torvalds BUG_ON(addr >= end); 23941da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23951da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 23961da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 23971da177e4SLinus Torvalds do { 23981da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 23991da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 24001da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 24011da177e4SLinus Torvalds if (err) 24021da177e4SLinus Torvalds break; 24031da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 24042ab64037Svenkatesh.pallipadi@intel.com 24052ab64037Svenkatesh.pallipadi@intel.com if (err) 24065180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 24072ab64037Svenkatesh.pallipadi@intel.com 24081da177e4SLinus Torvalds return err; 24091da177e4SLinus Torvalds } 24101da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 24111da177e4SLinus Torvalds 2412b4cbb197SLinus Torvalds /** 2413b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2414b4cbb197SLinus Torvalds * @vma: user vma to map to 2415b4cbb197SLinus Torvalds * @start: start of area 2416b4cbb197SLinus Torvalds * @len: size of area 2417b4cbb197SLinus Torvalds * 2418b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2419b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2420b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2421b4cbb197SLinus Torvalds * 2422b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2423b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2424b4cbb197SLinus Torvalds */ 2425b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2426b4cbb197SLinus Torvalds { 2427b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2428b4cbb197SLinus Torvalds 2429b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2430b4cbb197SLinus Torvalds if (start + len < start) 2431b4cbb197SLinus Torvalds return -EINVAL; 2432b4cbb197SLinus Torvalds /* 2433b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2434b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2435b4cbb197SLinus Torvalds * just have smaller alignment. 2436b4cbb197SLinus Torvalds */ 2437b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2438b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2439b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2440b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2441b4cbb197SLinus Torvalds return -EINVAL; 2442b4cbb197SLinus Torvalds 2443b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2444b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2445b4cbb197SLinus Torvalds return -EINVAL; 2446b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2447b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2448b4cbb197SLinus Torvalds 2449b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2450b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2451b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2452b4cbb197SLinus Torvalds return -EINVAL; 2453b4cbb197SLinus Torvalds 2454b4cbb197SLinus Torvalds /* Ok, let it rip */ 2455b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2456b4cbb197SLinus Torvalds } 2457b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2458b4cbb197SLinus Torvalds 2459aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2460aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2461aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2462aee16b3cSJeremy Fitzhardinge { 2463aee16b3cSJeremy Fitzhardinge pte_t *pte; 2464aee16b3cSJeremy Fitzhardinge int err; 24652f569afdSMartin Schwidefsky pgtable_t token; 246694909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 2467aee16b3cSJeremy Fitzhardinge 2468aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 2469aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 2470aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2471aee16b3cSJeremy Fitzhardinge if (!pte) 2472aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2473aee16b3cSJeremy Fitzhardinge 2474aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2475aee16b3cSJeremy Fitzhardinge 247638e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 247738e0edb1SJeremy Fitzhardinge 24782f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 2479aee16b3cSJeremy Fitzhardinge 2480aee16b3cSJeremy Fitzhardinge do { 2481c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 2482aee16b3cSJeremy Fitzhardinge if (err) 2483aee16b3cSJeremy Fitzhardinge break; 2484c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2485aee16b3cSJeremy Fitzhardinge 248638e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 248738e0edb1SJeremy Fitzhardinge 2488aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 2489aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 2490aee16b3cSJeremy Fitzhardinge return err; 2491aee16b3cSJeremy Fitzhardinge } 2492aee16b3cSJeremy Fitzhardinge 2493aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2494aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2495aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2496aee16b3cSJeremy Fitzhardinge { 2497aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2498aee16b3cSJeremy Fitzhardinge unsigned long next; 2499aee16b3cSJeremy Fitzhardinge int err; 2500aee16b3cSJeremy Fitzhardinge 2501ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2502ceb86879SAndi Kleen 2503aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 2504aee16b3cSJeremy Fitzhardinge if (!pmd) 2505aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2506aee16b3cSJeremy Fitzhardinge do { 2507aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 2508aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 2509aee16b3cSJeremy Fitzhardinge if (err) 2510aee16b3cSJeremy Fitzhardinge break; 2511aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 2512aee16b3cSJeremy Fitzhardinge return err; 2513aee16b3cSJeremy Fitzhardinge } 2514aee16b3cSJeremy Fitzhardinge 2515aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 2516aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2517aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2518aee16b3cSJeremy Fitzhardinge { 2519aee16b3cSJeremy Fitzhardinge pud_t *pud; 2520aee16b3cSJeremy Fitzhardinge unsigned long next; 2521aee16b3cSJeremy Fitzhardinge int err; 2522aee16b3cSJeremy Fitzhardinge 2523aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 2524aee16b3cSJeremy Fitzhardinge if (!pud) 2525aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2526aee16b3cSJeremy Fitzhardinge do { 2527aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 2528aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 2529aee16b3cSJeremy Fitzhardinge if (err) 2530aee16b3cSJeremy Fitzhardinge break; 2531aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 2532aee16b3cSJeremy Fitzhardinge return err; 2533aee16b3cSJeremy Fitzhardinge } 2534aee16b3cSJeremy Fitzhardinge 2535aee16b3cSJeremy Fitzhardinge /* 2536aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2537aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2538aee16b3cSJeremy Fitzhardinge */ 2539aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2540aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2541aee16b3cSJeremy Fitzhardinge { 2542aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 2543aee16b3cSJeremy Fitzhardinge unsigned long next; 254457250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 2545aee16b3cSJeremy Fitzhardinge int err; 2546aee16b3cSJeremy Fitzhardinge 2547aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 2548aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 2549aee16b3cSJeremy Fitzhardinge do { 2550aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 2551aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 2552aee16b3cSJeremy Fitzhardinge if (err) 2553aee16b3cSJeremy Fitzhardinge break; 2554aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 255557250a5bSJeremy Fitzhardinge 2556aee16b3cSJeremy Fitzhardinge return err; 2557aee16b3cSJeremy Fitzhardinge } 2558aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2559aee16b3cSJeremy Fitzhardinge 25601da177e4SLinus Torvalds /* 25618f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 25628f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 25638f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 2564a335b2e1SRyota Ozaki * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault 25658f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 25668f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 2567a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 25688f4e2101SHugh Dickins */ 25694c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 25708f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 25718f4e2101SHugh Dickins { 25728f4e2101SHugh Dickins int same = 1; 25738f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 25748f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 25754c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 25764c21e2f2SHugh Dickins spin_lock(ptl); 25778f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 25784c21e2f2SHugh Dickins spin_unlock(ptl); 25798f4e2101SHugh Dickins } 25808f4e2101SHugh Dickins #endif 25818f4e2101SHugh Dickins pte_unmap(page_table); 25828f4e2101SHugh Dickins return same; 25838f4e2101SHugh Dickins } 25848f4e2101SHugh Dickins 25859de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 25866aab341eSLinus Torvalds { 25870abdd7a8SDan Williams debug_dma_assert_idle(src); 25880abdd7a8SDan Williams 25896aab341eSLinus Torvalds /* 25906aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 25916aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 25926aab341eSLinus Torvalds * just copying from the original user address. If that 25936aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 25946aab341eSLinus Torvalds */ 25956aab341eSLinus Torvalds if (unlikely(!src)) { 25969b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 25975d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 25985d2a2dbbSLinus Torvalds 25995d2a2dbbSLinus Torvalds /* 26005d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 26015d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 26025d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 26035d2a2dbbSLinus Torvalds * zeroes. 26045d2a2dbbSLinus Torvalds */ 26055d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 26063ecb01dfSJan Beulich clear_page(kaddr); 26079b04c5feSCong Wang kunmap_atomic(kaddr); 2608c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 26090ed361deSNick Piggin } else 26109de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 26116aab341eSLinus Torvalds } 26126aab341eSLinus Torvalds 26131da177e4SLinus Torvalds /* 2614fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2615fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2616fb09a464SKirill A. Shutemov * 2617fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2618fb09a464SKirill A. Shutemov */ 2619fb09a464SKirill A. Shutemov static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, 2620fb09a464SKirill A. Shutemov unsigned long address) 2621fb09a464SKirill A. Shutemov { 2622fb09a464SKirill A. Shutemov struct vm_fault vmf; 2623fb09a464SKirill A. Shutemov int ret; 2624fb09a464SKirill A. Shutemov 2625fb09a464SKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 2626fb09a464SKirill A. Shutemov vmf.pgoff = page->index; 2627fb09a464SKirill A. Shutemov vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2628fb09a464SKirill A. Shutemov vmf.page = page; 2629fb09a464SKirill A. Shutemov 2630fb09a464SKirill A. Shutemov ret = vma->vm_ops->page_mkwrite(vma, &vmf); 2631fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2632fb09a464SKirill A. Shutemov return ret; 2633fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 2634fb09a464SKirill A. Shutemov lock_page(page); 2635fb09a464SKirill A. Shutemov if (!page->mapping) { 2636fb09a464SKirill A. Shutemov unlock_page(page); 2637fb09a464SKirill A. Shutemov return 0; /* retry */ 2638fb09a464SKirill A. Shutemov } 2639fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2640fb09a464SKirill A. Shutemov } else 2641fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 2642fb09a464SKirill A. Shutemov return ret; 2643fb09a464SKirill A. Shutemov } 2644fb09a464SKirill A. Shutemov 2645fb09a464SKirill A. Shutemov /* 26461da177e4SLinus Torvalds * This routine handles present pages, when users try to write 26471da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 26481da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 26491da177e4SLinus Torvalds * 26501da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 26511da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 26521da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 26531da177e4SLinus Torvalds * COW. 26541da177e4SLinus Torvalds * 26551da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 26561da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 26571da177e4SLinus Torvalds * and potentially makes it more efficient. 26581da177e4SLinus Torvalds * 26598f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 26608f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 26618f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 26621da177e4SLinus Torvalds */ 26631da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 266465500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 26658f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2666e6219ec8SNamhyung Kim __releases(ptl) 26671da177e4SLinus Torvalds { 26682ec74c3eSSagi Grimberg struct page *old_page, *new_page = NULL; 26691da177e4SLinus Torvalds pte_t entry; 2670b009c024SMichel Lespinasse int ret = 0; 2671a200ee18SPeter Zijlstra int page_mkwrite = 0; 2672d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 26731756954cSDavid Rientjes unsigned long mmun_start = 0; /* For mmu_notifiers */ 26741756954cSDavid Rientjes unsigned long mmun_end = 0; /* For mmu_notifiers */ 26751da177e4SLinus Torvalds 26766aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2677251b97f5SPeter Zijlstra if (!old_page) { 2678251b97f5SPeter Zijlstra /* 2679251b97f5SPeter Zijlstra * VM_MIXEDMAP !pfn_valid() case 2680251b97f5SPeter Zijlstra * 2681251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2682251b97f5SPeter Zijlstra * Just mark the pages writable as we can't do any dirty 2683251b97f5SPeter Zijlstra * accounting on raw pfn maps. 2684251b97f5SPeter Zijlstra */ 2685251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2686251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2687251b97f5SPeter Zijlstra goto reuse; 2688920fc356SHugh Dickins goto gotten; 2689251b97f5SPeter Zijlstra } 26901da177e4SLinus Torvalds 2691d08b3851SPeter Zijlstra /* 2692ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2693ee6a6457SPeter Zijlstra * not dirty accountable. 2694d08b3851SPeter Zijlstra */ 26959a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2696ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2697ab967d86SHugh Dickins page_cache_get(old_page); 2698ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2699ab967d86SHugh Dickins lock_page(old_page); 2700ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2701ab967d86SHugh Dickins &ptl); 2702ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2703ab967d86SHugh Dickins unlock_page(old_page); 2704ab967d86SHugh Dickins goto unlock; 2705ab967d86SHugh Dickins } 2706ab967d86SHugh Dickins page_cache_release(old_page); 2707ab967d86SHugh Dickins } 2708b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2709c44b6743SRik van Riel /* 2710c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2711c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2712c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2713c44b6743SRik van Riel */ 2714c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2715ee6a6457SPeter Zijlstra unlock_page(old_page); 2716b009c024SMichel Lespinasse goto reuse; 2717b009c024SMichel Lespinasse } 2718b009c024SMichel Lespinasse unlock_page(old_page); 2719ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2720d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 2721ee6a6457SPeter Zijlstra /* 2722ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 2723ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 2724ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 2725ee6a6457SPeter Zijlstra */ 27269637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 2727c2ec175cSNick Piggin int tmp; 27289637a5efSDavid Howells page_cache_get(old_page); 27299637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 2730fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, old_page, address); 2731fb09a464SKirill A. Shutemov if (unlikely(!tmp || (tmp & 2732fb09a464SKirill A. Shutemov (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 2733fb09a464SKirill A. Shutemov page_cache_release(old_page); 2734fb09a464SKirill A. Shutemov return tmp; 2735c2ec175cSNick Piggin } 27369637a5efSDavid Howells /* 27379637a5efSDavid Howells * Since we dropped the lock we need to revalidate 27389637a5efSDavid Howells * the PTE as someone else may have changed it. If 27399637a5efSDavid Howells * they did, we just return, as we can count on the 27409637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 27419637a5efSDavid Howells */ 27429637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 27439637a5efSDavid Howells &ptl); 2744b827e496SNick Piggin if (!pte_same(*page_table, orig_pte)) { 2745b827e496SNick Piggin unlock_page(old_page); 27469637a5efSDavid Howells goto unlock; 2747b827e496SNick Piggin } 2748a200ee18SPeter Zijlstra 2749a200ee18SPeter Zijlstra page_mkwrite = 1; 27509637a5efSDavid Howells } 2751d08b3851SPeter Zijlstra dirty_page = old_page; 2752d08b3851SPeter Zijlstra get_page(dirty_page); 27539637a5efSDavid Howells 2754251b97f5SPeter Zijlstra reuse: 27558c8a743cSPeter Zijlstra /* 27568c8a743cSPeter Zijlstra * Clear the pages cpupid information as the existing 27578c8a743cSPeter Zijlstra * information potentially belongs to a now completely 27588c8a743cSPeter Zijlstra * unrelated process. 27598c8a743cSPeter Zijlstra */ 27608c8a743cSPeter Zijlstra if (old_page) 27618c8a743cSPeter Zijlstra page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1); 27628c8a743cSPeter Zijlstra 2763eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 276465500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 276565500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2766954ffcb3SKAMEZAWA Hiroyuki if (ptep_set_access_flags(vma, address, page_table, entry,1)) 27674b3073e1SRussell King update_mmu_cache(vma, address, page_table); 276872ddc8f7SMichel Lespinasse pte_unmap_unlock(page_table, ptl); 276965500d23SHugh Dickins ret |= VM_FAULT_WRITE; 277072ddc8f7SMichel Lespinasse 277172ddc8f7SMichel Lespinasse if (!dirty_page) 277272ddc8f7SMichel Lespinasse return ret; 277372ddc8f7SMichel Lespinasse 277472ddc8f7SMichel Lespinasse /* 277572ddc8f7SMichel Lespinasse * Yes, Virginia, this is actually required to prevent a race 277672ddc8f7SMichel Lespinasse * with clear_page_dirty_for_io() from clearing the page dirty 277772ddc8f7SMichel Lespinasse * bit after it clear all dirty ptes, but before a racing 277872ddc8f7SMichel Lespinasse * do_wp_page installs a dirty pte. 277972ddc8f7SMichel Lespinasse * 2780f0c6d4d2SKirill A. Shutemov * do_shared_fault is protected similarly. 278172ddc8f7SMichel Lespinasse */ 278272ddc8f7SMichel Lespinasse if (!page_mkwrite) { 278372ddc8f7SMichel Lespinasse wait_on_page_locked(dirty_page); 278472ddc8f7SMichel Lespinasse set_page_dirty_balance(dirty_page, page_mkwrite); 278541c4d25fSJan Kara /* file_update_time outside page_lock */ 278641c4d25fSJan Kara if (vma->vm_file) 278741c4d25fSJan Kara file_update_time(vma->vm_file); 278872ddc8f7SMichel Lespinasse } 278972ddc8f7SMichel Lespinasse put_page(dirty_page); 279072ddc8f7SMichel Lespinasse if (page_mkwrite) { 279172ddc8f7SMichel Lespinasse struct address_space *mapping = dirty_page->mapping; 279272ddc8f7SMichel Lespinasse 279372ddc8f7SMichel Lespinasse set_page_dirty(dirty_page); 279472ddc8f7SMichel Lespinasse unlock_page(dirty_page); 279572ddc8f7SMichel Lespinasse page_cache_release(dirty_page); 279672ddc8f7SMichel Lespinasse if (mapping) { 279772ddc8f7SMichel Lespinasse /* 279872ddc8f7SMichel Lespinasse * Some device drivers do not set page.mapping 279972ddc8f7SMichel Lespinasse * but still dirty their pages 280072ddc8f7SMichel Lespinasse */ 280172ddc8f7SMichel Lespinasse balance_dirty_pages_ratelimited(mapping); 280272ddc8f7SMichel Lespinasse } 280372ddc8f7SMichel Lespinasse } 280472ddc8f7SMichel Lespinasse 280572ddc8f7SMichel Lespinasse return ret; 28061da177e4SLinus Torvalds } 28071da177e4SLinus Torvalds 28081da177e4SLinus Torvalds /* 28091da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 28101da177e4SLinus Torvalds */ 28111da177e4SLinus Torvalds page_cache_get(old_page); 2812920fc356SHugh Dickins gotten: 28138f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 28141da177e4SLinus Torvalds 28151da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 281665500d23SHugh Dickins goto oom; 2817a13ea5b7SHugh Dickins 281862eede62SHugh Dickins if (is_zero_pfn(pte_pfn(orig_pte))) { 2819a13ea5b7SHugh Dickins new_page = alloc_zeroed_user_highpage_movable(vma, address); 2820a13ea5b7SHugh Dickins if (!new_page) 2821a13ea5b7SHugh Dickins goto oom; 2822a13ea5b7SHugh Dickins } else { 2823769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 28241da177e4SLinus Torvalds if (!new_page) 282565500d23SHugh Dickins goto oom; 2826a13ea5b7SHugh Dickins cow_user_page(new_page, old_page, address, vma); 2827a13ea5b7SHugh Dickins } 2828a13ea5b7SHugh Dickins __SetPageUptodate(new_page); 2829a13ea5b7SHugh Dickins 28302c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) 28318a9f3ccdSBalbir Singh goto oom_free_new; 28328a9f3ccdSBalbir Singh 28336bdb913fSHaggai Eran mmun_start = address & PAGE_MASK; 28341756954cSDavid Rientjes mmun_end = mmun_start + PAGE_SIZE; 28356bdb913fSHaggai Eran mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 28366bdb913fSHaggai Eran 28371da177e4SLinus Torvalds /* 28381da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 28391da177e4SLinus Torvalds */ 28408f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 284165500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 2842920fc356SHugh Dickins if (old_page) { 28434294621fSHugh Dickins if (!PageAnon(old_page)) { 284434e55232SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_FILEPAGES); 284534e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 28464294621fSHugh Dickins } 2847920fc356SHugh Dickins } else 284834e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2849eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 285065500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 285165500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 28524ce072f1SSiddha, Suresh B /* 28534ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 28544ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 28554ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 28564ce072f1SSiddha, Suresh B * thread doing COW. 28574ce072f1SSiddha, Suresh B */ 2858828502d3SIzik Eidus ptep_clear_flush(vma, address, page_table); 28599617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 2860828502d3SIzik Eidus /* 2861828502d3SIzik Eidus * We call the notify macro here because, when using secondary 2862828502d3SIzik Eidus * mmu page tables (such as kvm shadow page tables), we want the 2863828502d3SIzik Eidus * new page to be mapped directly into the secondary page table. 2864828502d3SIzik Eidus */ 2865828502d3SIzik Eidus set_pte_at_notify(mm, address, page_table, entry); 28664b3073e1SRussell King update_mmu_cache(vma, address, page_table); 2867945754a1SNick Piggin if (old_page) { 2868945754a1SNick Piggin /* 2869945754a1SNick Piggin * Only after switching the pte to the new page may 2870945754a1SNick Piggin * we remove the mapcount here. Otherwise another 2871945754a1SNick Piggin * process may come and find the rmap count decremented 2872945754a1SNick Piggin * before the pte is switched to the new page, and 2873945754a1SNick Piggin * "reuse" the old page writing into it while our pte 2874945754a1SNick Piggin * here still points into it and can be read by other 2875945754a1SNick Piggin * threads. 2876945754a1SNick Piggin * 2877945754a1SNick Piggin * The critical issue is to order this 2878945754a1SNick Piggin * page_remove_rmap with the ptp_clear_flush above. 2879945754a1SNick Piggin * Those stores are ordered by (if nothing else,) 2880945754a1SNick Piggin * the barrier present in the atomic_add_negative 2881945754a1SNick Piggin * in page_remove_rmap. 2882945754a1SNick Piggin * 2883945754a1SNick Piggin * Then the TLB flush in ptep_clear_flush ensures that 2884945754a1SNick Piggin * no process can access the old page before the 2885945754a1SNick Piggin * decremented mapcount is visible. And the old page 2886945754a1SNick Piggin * cannot be reused until after the decremented 2887945754a1SNick Piggin * mapcount is visible. So transitively, TLBs to 2888945754a1SNick Piggin * old page will be flushed before it can be reused. 2889945754a1SNick Piggin */ 2890edc315fdSHugh Dickins page_remove_rmap(old_page); 2891945754a1SNick Piggin } 2892945754a1SNick Piggin 28931da177e4SLinus Torvalds /* Free the old page.. */ 28941da177e4SLinus Torvalds new_page = old_page; 2895f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 28968a9f3ccdSBalbir Singh } else 28978a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(new_page); 28988a9f3ccdSBalbir Singh 28996bdb913fSHaggai Eran if (new_page) 29006bdb913fSHaggai Eran page_cache_release(new_page); 290165500d23SHugh Dickins unlock: 29028f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 29031756954cSDavid Rientjes if (mmun_end > mmun_start) 29046bdb913fSHaggai Eran mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 2905e15f8c01SMichel Lespinasse if (old_page) { 2906e15f8c01SMichel Lespinasse /* 2907e15f8c01SMichel Lespinasse * Don't let another task, with possibly unlocked vma, 2908e15f8c01SMichel Lespinasse * keep the mlocked page. 2909e15f8c01SMichel Lespinasse */ 2910e15f8c01SMichel Lespinasse if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { 2911e15f8c01SMichel Lespinasse lock_page(old_page); /* LRU manipulation */ 2912e15f8c01SMichel Lespinasse munlock_vma_page(old_page); 2913e15f8c01SMichel Lespinasse unlock_page(old_page); 2914e15f8c01SMichel Lespinasse } 2915e15f8c01SMichel Lespinasse page_cache_release(old_page); 2916e15f8c01SMichel Lespinasse } 2917f33ea7f4SNick Piggin return ret; 29188a9f3ccdSBalbir Singh oom_free_new: 29196dbf6d3bSHugh Dickins page_cache_release(new_page); 292065500d23SHugh Dickins oom: 292166521d5aSDominik Dingel if (old_page) 29221da177e4SLinus Torvalds page_cache_release(old_page); 29231da177e4SLinus Torvalds return VM_FAULT_OOM; 29241da177e4SLinus Torvalds } 29251da177e4SLinus Torvalds 292697a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 29271da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 29281da177e4SLinus Torvalds struct zap_details *details) 29291da177e4SLinus Torvalds { 2930f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 29311da177e4SLinus Torvalds } 29321da177e4SLinus Torvalds 29336b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 29341da177e4SLinus Torvalds struct zap_details *details) 29351da177e4SLinus Torvalds { 29361da177e4SLinus Torvalds struct vm_area_struct *vma; 29371da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 29381da177e4SLinus Torvalds 29396b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 29401da177e4SLinus Torvalds details->first_index, details->last_index) { 29411da177e4SLinus Torvalds 29421da177e4SLinus Torvalds vba = vma->vm_pgoff; 2943d6e93217SLibin vea = vba + vma_pages(vma) - 1; 29441da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 29451da177e4SLinus Torvalds zba = details->first_index; 29461da177e4SLinus Torvalds if (zba < vba) 29471da177e4SLinus Torvalds zba = vba; 29481da177e4SLinus Torvalds zea = details->last_index; 29491da177e4SLinus Torvalds if (zea > vea) 29501da177e4SLinus Torvalds zea = vea; 29511da177e4SLinus Torvalds 295297a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 29531da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 29541da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 295597a89413SPeter Zijlstra details); 29561da177e4SLinus Torvalds } 29571da177e4SLinus Torvalds } 29581da177e4SLinus Torvalds 29591da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 29601da177e4SLinus Torvalds struct zap_details *details) 29611da177e4SLinus Torvalds { 29621da177e4SLinus Torvalds struct vm_area_struct *vma; 29631da177e4SLinus Torvalds 29641da177e4SLinus Torvalds /* 29651da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 29661da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 29671da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 29681da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 29691da177e4SLinus Torvalds */ 29706b2dbba8SMichel Lespinasse list_for_each_entry(vma, head, shared.nonlinear) { 29711da177e4SLinus Torvalds details->nonlinear_vma = vma; 297297a89413SPeter Zijlstra unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); 29731da177e4SLinus Torvalds } 29741da177e4SLinus Torvalds } 29751da177e4SLinus Torvalds 29761da177e4SLinus Torvalds /** 297772fd4a35SRobert P. J. Day * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. 29783d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 29791da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 29801da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 298125d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 29821da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 29831da177e4SLinus Torvalds * partial pages. 29841da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 29851da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 29861da177e4SLinus Torvalds * end of the file. 29871da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 29881da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 29891da177e4SLinus Torvalds */ 29901da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 29911da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 29921da177e4SLinus Torvalds { 29931da177e4SLinus Torvalds struct zap_details details; 29941da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 29951da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 29961da177e4SLinus Torvalds 29971da177e4SLinus Torvalds /* Check for overflow. */ 29981da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 29991da177e4SLinus Torvalds long long holeend = 30001da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 30011da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 30021da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 30031da177e4SLinus Torvalds } 30041da177e4SLinus Torvalds 30051da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 30061da177e4SLinus Torvalds details.nonlinear_vma = NULL; 30071da177e4SLinus Torvalds details.first_index = hba; 30081da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 30091da177e4SLinus Torvalds if (details.last_index < details.first_index) 30101da177e4SLinus Torvalds details.last_index = ULONG_MAX; 30111da177e4SLinus Torvalds 301297a89413SPeter Zijlstra 30133d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 30146b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 30151da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 30161da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 30171da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 30183d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 30191da177e4SLinus Torvalds } 30201da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 30211da177e4SLinus Torvalds 30221da177e4SLinus Torvalds /* 30238f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 30248f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 30258f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 30261da177e4SLinus Torvalds */ 302765500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 302865500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 302930c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 30301da177e4SLinus Torvalds { 30318f4e2101SHugh Dickins spinlock_t *ptl; 303256f31801SHugh Dickins struct page *page, *swapcache; 303365500d23SHugh Dickins swp_entry_t entry; 30341da177e4SLinus Torvalds pte_t pte; 3035d065bd81SMichel Lespinasse int locked; 303656039efaSKAMEZAWA Hiroyuki struct mem_cgroup *ptr; 3037ad8c2ee8SRik van Riel int exclusive = 0; 303883c54070SNick Piggin int ret = 0; 30391da177e4SLinus Torvalds 30404c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 30418f4e2101SHugh Dickins goto out; 304265500d23SHugh Dickins 304365500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 3044d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 30450697212aSChristoph Lameter if (is_migration_entry(entry)) { 30460697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 3047d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3048d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3049d1737fdbSAndi Kleen } else { 3050d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 3051d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3052d1737fdbSAndi Kleen } 30530697212aSChristoph Lameter goto out; 30540697212aSChristoph Lameter } 30550ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 30561da177e4SLinus Torvalds page = lookup_swap_cache(entry); 30571da177e4SLinus Torvalds if (!page) { 305802098feaSHugh Dickins page = swapin_readahead(entry, 305902098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 30601da177e4SLinus Torvalds if (!page) { 30611da177e4SLinus Torvalds /* 30628f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 30638f4e2101SHugh Dickins * while we released the pte lock. 30641da177e4SLinus Torvalds */ 30658f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 30661da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 30671da177e4SLinus Torvalds ret = VM_FAULT_OOM; 30680ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 306965500d23SHugh Dickins goto unlock; 30701da177e4SLinus Torvalds } 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 30731da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3074f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 3075456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 3076d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 307771f72525SWu Fengguang /* 307871f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 307971f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 308071f72525SWu Fengguang */ 3081d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3082d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 308356f31801SHugh Dickins swapcache = page; 30844779cb31SAndi Kleen goto out_release; 30851da177e4SLinus Torvalds } 30861da177e4SLinus Torvalds 308756f31801SHugh Dickins swapcache = page; 3088d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 3089e709ffd6SRik van Riel 309020a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 3091d065bd81SMichel Lespinasse if (!locked) { 3092d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3093d065bd81SMichel Lespinasse goto out_release; 3094d065bd81SMichel Lespinasse } 30951da177e4SLinus Torvalds 30964969c119SAndrea Arcangeli /* 309731c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 309831c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 309931c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 310031c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 31014969c119SAndrea Arcangeli */ 310231c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 31034969c119SAndrea Arcangeli goto out_page; 31044969c119SAndrea Arcangeli 3105cbf86cfeSHugh Dickins page = ksm_might_need_to_copy(page, vma, address); 31064969c119SAndrea Arcangeli if (unlikely(!page)) { 31075ad64688SHugh Dickins ret = VM_FAULT_OOM; 31084969c119SAndrea Arcangeli page = swapcache; 31094969c119SAndrea Arcangeli goto out_page; 31104969c119SAndrea Arcangeli } 31115ad64688SHugh Dickins 31122c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { 3113073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 3114bc43f75cSJohannes Weiner goto out_page; 3115073e587eSKAMEZAWA Hiroyuki } 3116073e587eSKAMEZAWA Hiroyuki 31171da177e4SLinus Torvalds /* 31188f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 31191da177e4SLinus Torvalds */ 31208f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 31219e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 3122b8107480SKirill Korotaev goto out_nomap; 3123b8107480SKirill Korotaev 3124b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 3125b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3126b8107480SKirill Korotaev goto out_nomap; 31271da177e4SLinus Torvalds } 31281da177e4SLinus Torvalds 31298c7c6e34SKAMEZAWA Hiroyuki /* 31308c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 31318c7c6e34SKAMEZAWA Hiroyuki * 31328c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 31338c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 31348c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 31358c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 31368c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 313703f3c433SKAMEZAWA Hiroyuki * Because delete_from_swap_page() may be called by reuse_swap_page(), 313803f3c433SKAMEZAWA Hiroyuki * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry 313903f3c433SKAMEZAWA Hiroyuki * in page->private. In this case, a record in swap_cgroup is silently 314003f3c433SKAMEZAWA Hiroyuki * discarded at swap_free(). 31418c7c6e34SKAMEZAWA Hiroyuki */ 31421da177e4SLinus Torvalds 314334e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 3144b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 31451da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 314630c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 31471da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 314830c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 31499a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 3150ad8c2ee8SRik van Riel exclusive = 1; 31511da177e4SLinus Torvalds } 31521da177e4SLinus Torvalds flush_icache_page(vma, page); 3153179ef71cSCyrill Gorcunov if (pte_swp_soft_dirty(orig_pte)) 3154179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 31551da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 315656f31801SHugh Dickins if (page == swapcache) 3157ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 315856f31801SHugh Dickins else /* ksm created a completely new copy */ 315956f31801SHugh Dickins page_add_new_anon_rmap(page, vma, address); 316003f3c433SKAMEZAWA Hiroyuki /* It's better to call commit-charge after rmap is established */ 316103f3c433SKAMEZAWA Hiroyuki mem_cgroup_commit_charge_swapin(page, ptr); 31621da177e4SLinus Torvalds 3163c475a8abSHugh Dickins swap_free(entry); 3164b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 3165a2c43eedSHugh Dickins try_to_free_swap(page); 3166c475a8abSHugh Dickins unlock_page(page); 316756f31801SHugh Dickins if (page != swapcache) { 31684969c119SAndrea Arcangeli /* 31694969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 31704969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 31714969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 31724969c119SAndrea Arcangeli * further safety release the lock after the swap_free 31734969c119SAndrea Arcangeli * so that the swap count won't change under a 31744969c119SAndrea Arcangeli * parallel locked swapcache. 31754969c119SAndrea Arcangeli */ 31764969c119SAndrea Arcangeli unlock_page(swapcache); 31774969c119SAndrea Arcangeli page_cache_release(swapcache); 31784969c119SAndrea Arcangeli } 3179c475a8abSHugh Dickins 318030c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 318161469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 318261469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 318361469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 31841da177e4SLinus Torvalds goto out; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds 31871da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 31884b3073e1SRussell King update_mmu_cache(vma, address, page_table); 318965500d23SHugh Dickins unlock: 31908f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 31911da177e4SLinus Torvalds out: 31921da177e4SLinus Torvalds return ret; 3193b8107480SKirill Korotaev out_nomap: 31947a81b88cSKAMEZAWA Hiroyuki mem_cgroup_cancel_charge_swapin(ptr); 31958f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3196bc43f75cSJohannes Weiner out_page: 3197b8107480SKirill Korotaev unlock_page(page); 31984779cb31SAndi Kleen out_release: 3199b8107480SKirill Korotaev page_cache_release(page); 320056f31801SHugh Dickins if (page != swapcache) { 32014969c119SAndrea Arcangeli unlock_page(swapcache); 32024969c119SAndrea Arcangeli page_cache_release(swapcache); 32034969c119SAndrea Arcangeli } 320465500d23SHugh Dickins return ret; 32051da177e4SLinus Torvalds } 32061da177e4SLinus Torvalds 32071da177e4SLinus Torvalds /* 32088ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 32098ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 3210320b2b8dSLinus Torvalds * doesn't hit another vma. 3211320b2b8dSLinus Torvalds */ 3212320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 3213320b2b8dSLinus Torvalds { 3214320b2b8dSLinus Torvalds address &= PAGE_MASK; 3215320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 32160e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 3217320b2b8dSLinus Torvalds 32180e8e50e2SLinus Torvalds /* 32190e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 32200e8e50e2SLinus Torvalds * 32210e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 32220e8e50e2SLinus Torvalds * that has gotten split.. 32230e8e50e2SLinus Torvalds */ 32240e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 32250e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 32260e8e50e2SLinus Torvalds 3227d05f3169SMichal Hocko expand_downwards(vma, address - PAGE_SIZE); 3228320b2b8dSLinus Torvalds } 32298ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 32308ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 32318ca3eb08SLuck, Tony 32328ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 32338ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 32348ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 32358ca3eb08SLuck, Tony 32368ca3eb08SLuck, Tony expand_upwards(vma, address + PAGE_SIZE); 32378ca3eb08SLuck, Tony } 3238320b2b8dSLinus Torvalds return 0; 3239320b2b8dSLinus Torvalds } 3240320b2b8dSLinus Torvalds 3241320b2b8dSLinus Torvalds /* 32428f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 32438f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 32448f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 32451da177e4SLinus Torvalds */ 324665500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 324765500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 324830c9f3a9SLinus Torvalds unsigned int flags) 32491da177e4SLinus Torvalds { 32508f4e2101SHugh Dickins struct page *page; 32518f4e2101SHugh Dickins spinlock_t *ptl; 32521da177e4SLinus Torvalds pte_t entry; 32531da177e4SLinus Torvalds 32545528f913SLinus Torvalds pte_unmap(page_table); 3255320b2b8dSLinus Torvalds 325611ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 325711ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 325811ac5524SLinus Torvalds return VM_FAULT_SIGBUS; 325911ac5524SLinus Torvalds 326011ac5524SLinus Torvalds /* Use the zero-page for reads */ 326162eede62SHugh Dickins if (!(flags & FAULT_FLAG_WRITE)) { 326262eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 326362eede62SHugh Dickins vma->vm_page_prot)); 326411ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 3265a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 3266a13ea5b7SHugh Dickins goto unlock; 3267a13ea5b7SHugh Dickins goto setpte; 3268a13ea5b7SHugh Dickins } 3269a13ea5b7SHugh Dickins 32701da177e4SLinus Torvalds /* Allocate our own private page. */ 32711da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 327265500d23SHugh Dickins goto oom; 3273769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 32741da177e4SLinus Torvalds if (!page) 327565500d23SHugh Dickins goto oom; 327652f37629SMinchan Kim /* 327752f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 327852f37629SMinchan Kim * preceeding stores to the page contents become visible before 327952f37629SMinchan Kim * the set_pte_at() write. 328052f37629SMinchan Kim */ 32810ed361deSNick Piggin __SetPageUptodate(page); 32821da177e4SLinus Torvalds 32832c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) 32848a9f3ccdSBalbir Singh goto oom_free_page; 32858a9f3ccdSBalbir Singh 328665500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 32871ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 32881ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 32898f4e2101SHugh Dickins 32908f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 32911c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 32928f4e2101SHugh Dickins goto release; 32939ba69294SHugh Dickins 329434e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 32959617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 3296a13ea5b7SHugh Dickins setpte: 329765500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 32981da177e4SLinus Torvalds 32991da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 33004b3073e1SRussell King update_mmu_cache(vma, address, page_table); 330165500d23SHugh Dickins unlock: 33028f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 330383c54070SNick Piggin return 0; 33048f4e2101SHugh Dickins release: 33058a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(page); 33068f4e2101SHugh Dickins page_cache_release(page); 33078f4e2101SHugh Dickins goto unlock; 33088a9f3ccdSBalbir Singh oom_free_page: 33096dbf6d3bSHugh Dickins page_cache_release(page); 331065500d23SHugh Dickins oom: 33111da177e4SLinus Torvalds return VM_FAULT_OOM; 33121da177e4SLinus Torvalds } 33131da177e4SLinus Torvalds 33147eae74afSKirill A. Shutemov static int __do_fault(struct vm_area_struct *vma, unsigned long address, 33157eae74afSKirill A. Shutemov pgoff_t pgoff, unsigned int flags, struct page **page) 33167eae74afSKirill A. Shutemov { 33177eae74afSKirill A. Shutemov struct vm_fault vmf; 33187eae74afSKirill A. Shutemov int ret; 33197eae74afSKirill A. Shutemov 33207eae74afSKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 33217eae74afSKirill A. Shutemov vmf.pgoff = pgoff; 33227eae74afSKirill A. Shutemov vmf.flags = flags; 33237eae74afSKirill A. Shutemov vmf.page = NULL; 33247eae74afSKirill A. Shutemov 33257eae74afSKirill A. Shutemov ret = vma->vm_ops->fault(vma, &vmf); 33267eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 33277eae74afSKirill A. Shutemov return ret; 33287eae74afSKirill A. Shutemov 33297eae74afSKirill A. Shutemov if (unlikely(PageHWPoison(vmf.page))) { 33307eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 33317eae74afSKirill A. Shutemov unlock_page(vmf.page); 33327eae74afSKirill A. Shutemov page_cache_release(vmf.page); 33337eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 33347eae74afSKirill A. Shutemov } 33357eae74afSKirill A. Shutemov 33367eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 33377eae74afSKirill A. Shutemov lock_page(vmf.page); 33387eae74afSKirill A. Shutemov else 33397eae74afSKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); 33407eae74afSKirill A. Shutemov 33417eae74afSKirill A. Shutemov *page = vmf.page; 33427eae74afSKirill A. Shutemov return ret; 33437eae74afSKirill A. Shutemov } 33447eae74afSKirill A. Shutemov 33453bb97794SKirill A. Shutemov static void do_set_pte(struct vm_area_struct *vma, unsigned long address, 33463bb97794SKirill A. Shutemov struct page *page, pte_t *pte, bool write, bool anon) 33473bb97794SKirill A. Shutemov { 33483bb97794SKirill A. Shutemov pte_t entry; 33493bb97794SKirill A. Shutemov 33503bb97794SKirill A. Shutemov flush_icache_page(vma, page); 33513bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 33523bb97794SKirill A. Shutemov if (write) 33533bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 33543bb97794SKirill A. Shutemov else if (pte_file(*pte) && pte_file_soft_dirty(*pte)) 33553bb97794SKirill A. Shutemov pte_mksoft_dirty(entry); 33563bb97794SKirill A. Shutemov if (anon) { 33573bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 33583bb97794SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address); 33593bb97794SKirill A. Shutemov } else { 33603bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); 33613bb97794SKirill A. Shutemov page_add_file_rmap(page); 33623bb97794SKirill A. Shutemov } 33633bb97794SKirill A. Shutemov set_pte_at(vma->vm_mm, address, pte, entry); 33643bb97794SKirill A. Shutemov 33653bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 33663bb97794SKirill A. Shutemov update_mmu_cache(vma, address, pte); 33673bb97794SKirill A. Shutemov } 33683bb97794SKirill A. Shutemov 3369e655fb29SKirill A. Shutemov static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3370e655fb29SKirill A. Shutemov unsigned long address, pmd_t *pmd, 3371e655fb29SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 3372e655fb29SKirill A. Shutemov { 3373e655fb29SKirill A. Shutemov struct page *fault_page; 3374e655fb29SKirill A. Shutemov spinlock_t *ptl; 33753bb97794SKirill A. Shutemov pte_t *pte; 3376e655fb29SKirill A. Shutemov int ret; 3377e655fb29SKirill A. Shutemov 3378e655fb29SKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 3379e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3380e655fb29SKirill A. Shutemov return ret; 3381e655fb29SKirill A. Shutemov 3382e655fb29SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3383e655fb29SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3384e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3385e655fb29SKirill A. Shutemov unlock_page(fault_page); 3386e655fb29SKirill A. Shutemov page_cache_release(fault_page); 3387e655fb29SKirill A. Shutemov return ret; 3388e655fb29SKirill A. Shutemov } 33893bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, false, false); 3390e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3391e655fb29SKirill A. Shutemov unlock_page(fault_page); 3392e655fb29SKirill A. Shutemov return ret; 3393e655fb29SKirill A. Shutemov } 3394e655fb29SKirill A. Shutemov 3395ec47c3b9SKirill A. Shutemov static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3396ec47c3b9SKirill A. Shutemov unsigned long address, pmd_t *pmd, 3397ec47c3b9SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 3398ec47c3b9SKirill A. Shutemov { 3399ec47c3b9SKirill A. Shutemov struct page *fault_page, *new_page; 3400ec47c3b9SKirill A. Shutemov spinlock_t *ptl; 34013bb97794SKirill A. Shutemov pte_t *pte; 3402ec47c3b9SKirill A. Shutemov int ret; 3403ec47c3b9SKirill A. Shutemov 3404ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 3405ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3406ec47c3b9SKirill A. Shutemov 3407ec47c3b9SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 3408ec47c3b9SKirill A. Shutemov if (!new_page) 3409ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3410ec47c3b9SKirill A. Shutemov 3411ec47c3b9SKirill A. Shutemov if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) { 3412ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3413ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3414ec47c3b9SKirill A. Shutemov } 3415ec47c3b9SKirill A. Shutemov 3416ec47c3b9SKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 3417ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3418ec47c3b9SKirill A. Shutemov goto uncharge_out; 3419ec47c3b9SKirill A. Shutemov 3420ec47c3b9SKirill A. Shutemov copy_user_highpage(new_page, fault_page, address, vma); 3421ec47c3b9SKirill A. Shutemov __SetPageUptodate(new_page); 3422ec47c3b9SKirill A. Shutemov 3423ec47c3b9SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3424ec47c3b9SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3425ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3426ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3427ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 3428ec47c3b9SKirill A. Shutemov goto uncharge_out; 3429ec47c3b9SKirill A. Shutemov } 34303bb97794SKirill A. Shutemov do_set_pte(vma, address, new_page, pte, true, true); 3431ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3432ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3433ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 3434ec47c3b9SKirill A. Shutemov return ret; 3435ec47c3b9SKirill A. Shutemov uncharge_out: 3436ec47c3b9SKirill A. Shutemov mem_cgroup_uncharge_page(new_page); 3437ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3438ec47c3b9SKirill A. Shutemov return ret; 3439ec47c3b9SKirill A. Shutemov } 3440ec47c3b9SKirill A. Shutemov 3441f0c6d4d2SKirill A. Shutemov static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, 344216abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 344354cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 34441da177e4SLinus Torvalds { 3445f0c6d4d2SKirill A. Shutemov struct page *fault_page; 3446f0c6d4d2SKirill A. Shutemov struct address_space *mapping; 34478f4e2101SHugh Dickins spinlock_t *ptl; 34483bb97794SKirill A. Shutemov pte_t *pte; 3449f0c6d4d2SKirill A. Shutemov int dirtied = 0; 3450f0c6d4d2SKirill A. Shutemov int ret, tmp; 34511d65f86dSKAMEZAWA Hiroyuki 34527eae74afSKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 34537eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3454f0c6d4d2SKirill A. Shutemov return ret; 34551da177e4SLinus Torvalds 34561da177e4SLinus Torvalds /* 3457f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 3458f0c6d4d2SKirill A. Shutemov * about to become writable 34591da177e4SLinus Torvalds */ 3460fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 3461f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3462fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, fault_page, address); 3463fb09a464SKirill A. Shutemov if (unlikely(!tmp || 3464fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3465f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3466f0c6d4d2SKirill A. Shutemov return tmp; 346769676147SMark Fasheh } 3468d0217ac0SNick Piggin } 3469fb09a464SKirill A. Shutemov 3470f0c6d4d2SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3471f0c6d4d2SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3472f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3473f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3474f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3475f0c6d4d2SKirill A. Shutemov return ret; 34769637a5efSDavid Howells } 34773bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, true, false); 3478f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3479d00806b1SNick Piggin 3480f0c6d4d2SKirill A. Shutemov if (set_page_dirty(fault_page)) 348141c4d25fSJan Kara dirtied = 1; 3482f0c6d4d2SKirill A. Shutemov mapping = fault_page->mapping; 3483f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3484f0c6d4d2SKirill A. Shutemov if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { 3485b827e496SNick Piggin /* 3486b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3487b827e496SNick Piggin * dirty their pages 3488b827e496SNick Piggin */ 3489b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3490d08b3851SPeter Zijlstra } 3491d00806b1SNick Piggin 3492b827e496SNick Piggin /* file_update_time outside page_lock */ 3493f0c6d4d2SKirill A. Shutemov if (vma->vm_file && !vma->vm_ops->page_mkwrite) 3494b827e496SNick Piggin file_update_time(vma->vm_file); 3495b827e496SNick Piggin 3496b827e496SNick Piggin return ret; 349754cb8821SNick Piggin } 3498d00806b1SNick Piggin 349954cb8821SNick Piggin static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 350054cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 350130c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 350254cb8821SNick Piggin { 350354cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 35040da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 350554cb8821SNick Piggin 350616abfa08SHugh Dickins pte_unmap(page_table); 3507e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3508e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3509e655fb29SKirill A. Shutemov orig_pte); 3510ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3511ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3512ec47c3b9SKirill A. Shutemov orig_pte); 3513f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 351454cb8821SNick Piggin } 351554cb8821SNick Piggin 3516f4b81804SJes Sorensen /* 35171da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 35181da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 35191da177e4SLinus Torvalds * nonlinear vmas. 35208f4e2101SHugh Dickins * 35218f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 35228f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 35238f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 35241da177e4SLinus Torvalds */ 3525d0217ac0SNick Piggin static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 352665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 352730c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 35281da177e4SLinus Torvalds { 352965500d23SHugh Dickins pgoff_t pgoff; 35301da177e4SLinus Torvalds 353130c9f3a9SLinus Torvalds flags |= FAULT_FLAG_NONLINEAR; 353230c9f3a9SLinus Torvalds 35334c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 353483c54070SNick Piggin return 0; 35351da177e4SLinus Torvalds 35362509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { 353765500d23SHugh Dickins /* 353865500d23SHugh Dickins * Page table corrupted: show pte and kill process. 353965500d23SHugh Dickins */ 35403dc14741SHugh Dickins print_bad_pte(vma, address, orig_pte, NULL); 3541d99be1a8SHugh Dickins return VM_FAULT_SIGBUS; 354265500d23SHugh Dickins } 354365500d23SHugh Dickins 354465500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 3545e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3546e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3547e655fb29SKirill A. Shutemov orig_pte); 3548ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3549ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3550ec47c3b9SKirill A. Shutemov orig_pte); 3551f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 35521da177e4SLinus Torvalds } 35531da177e4SLinus Torvalds 3554b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 355504bb2f94SRik van Riel unsigned long addr, int page_nid, 355604bb2f94SRik van Riel int *flags) 35579532fec1SMel Gorman { 35589532fec1SMel Gorman get_page(page); 35599532fec1SMel Gorman 35609532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 356104bb2f94SRik van Riel if (page_nid == numa_node_id()) { 35629532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 356304bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 356404bb2f94SRik van Riel } 35659532fec1SMel Gorman 35669532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 35679532fec1SMel Gorman } 35689532fec1SMel Gorman 3569b19a9939SRashika Kheria static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3570d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3571d10e63f2SMel Gorman { 35724daae3b4SMel Gorman struct page *page = NULL; 3573d10e63f2SMel Gorman spinlock_t *ptl; 35748191acbdSMel Gorman int page_nid = -1; 357590572890SPeter Zijlstra int last_cpupid; 3576cbee9f88SPeter Zijlstra int target_nid; 3577b8593bfdSMel Gorman bool migrated = false; 35786688cc05SPeter Zijlstra int flags = 0; 3579d10e63f2SMel Gorman 3580d10e63f2SMel Gorman /* 3581d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3582d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3583d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3584d10e63f2SMel Gorman * 3585d10e63f2SMel Gorman * ptep_modify_prot_start is not called as this is clearing 3586d10e63f2SMel Gorman * the _PAGE_NUMA bit and it is not really expected that there 3587d10e63f2SMel Gorman * would be concurrent hardware modifications to the PTE. 3588d10e63f2SMel Gorman */ 3589d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3590d10e63f2SMel Gorman spin_lock(ptl); 35914daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 35924daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 35934daae3b4SMel Gorman goto out; 35944daae3b4SMel Gorman } 35954daae3b4SMel Gorman 3596d10e63f2SMel Gorman pte = pte_mknonnuma(pte); 3597d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3598d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3599d10e63f2SMel Gorman 3600d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3601d10e63f2SMel Gorman if (!page) { 3602d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3603d10e63f2SMel Gorman return 0; 3604d10e63f2SMel Gorman } 3605a1a46184SMel Gorman BUG_ON(is_zero_pfn(page_to_pfn(page))); 3606d10e63f2SMel Gorman 36076688cc05SPeter Zijlstra /* 36086688cc05SPeter Zijlstra * Avoid grouping on DSO/COW pages in specific and RO pages 36096688cc05SPeter Zijlstra * in general, RO pages shouldn't hurt as much anyway since 36106688cc05SPeter Zijlstra * they can be in shared cache state. 36116688cc05SPeter Zijlstra */ 36126688cc05SPeter Zijlstra if (!pte_write(pte)) 36136688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 36146688cc05SPeter Zijlstra 3615dabe1d99SRik van Riel /* 3616dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 3617dabe1d99SRik van Riel * is later used when determining whether to group tasks together 3618dabe1d99SRik van Riel */ 3619dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 3620dabe1d99SRik van Riel flags |= TNF_SHARED; 3621dabe1d99SRik van Riel 362290572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 36238191acbdSMel Gorman page_nid = page_to_nid(page); 362404bb2f94SRik van Riel target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); 3625d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 36264daae3b4SMel Gorman if (target_nid == -1) { 36274daae3b4SMel Gorman put_page(page); 36284daae3b4SMel Gorman goto out; 36294daae3b4SMel Gorman } 36304daae3b4SMel Gorman 36314daae3b4SMel Gorman /* Migrate to the requested node */ 36321bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 36336688cc05SPeter Zijlstra if (migrated) { 36348191acbdSMel Gorman page_nid = target_nid; 36356688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 36366688cc05SPeter Zijlstra } 36374daae3b4SMel Gorman 36384daae3b4SMel Gorman out: 36398191acbdSMel Gorman if (page_nid != -1) 36406688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 3641d10e63f2SMel Gorman return 0; 3642d10e63f2SMel Gorman } 3643d10e63f2SMel Gorman 36441da177e4SLinus Torvalds /* 36451da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 36461da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 36471da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 36481da177e4SLinus Torvalds * 36491da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 36501da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 36511da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 36521da177e4SLinus Torvalds * 3653c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3654c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 3655c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 36561da177e4SLinus Torvalds */ 3657c0292554SKirill A. Shutemov static int handle_pte_fault(struct mm_struct *mm, 36581da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 365930c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 36601da177e4SLinus Torvalds { 36611da177e4SLinus Torvalds pte_t entry; 36628f4e2101SHugh Dickins spinlock_t *ptl; 36631da177e4SLinus Torvalds 36648dab5241SBenjamin Herrenschmidt entry = *pte; 36651da177e4SLinus Torvalds if (!pte_present(entry)) { 366665500d23SHugh Dickins if (pte_none(entry)) { 3667f4b81804SJes Sorensen if (vma->vm_ops) { 36683c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 366954cb8821SNick Piggin return do_linear_fault(mm, vma, address, 367030c9f3a9SLinus Torvalds pte, pmd, flags, entry); 3671f4b81804SJes Sorensen } 3672f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 367330c9f3a9SLinus Torvalds pte, pmd, flags); 367465500d23SHugh Dickins } 36751da177e4SLinus Torvalds if (pte_file(entry)) 3676d0217ac0SNick Piggin return do_nonlinear_fault(mm, vma, address, 367730c9f3a9SLinus Torvalds pte, pmd, flags, entry); 367865500d23SHugh Dickins return do_swap_page(mm, vma, address, 367930c9f3a9SLinus Torvalds pte, pmd, flags, entry); 36801da177e4SLinus Torvalds } 36811da177e4SLinus Torvalds 3682d10e63f2SMel Gorman if (pte_numa(entry)) 3683d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3684d10e63f2SMel Gorman 36854c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 36868f4e2101SHugh Dickins spin_lock(ptl); 36878f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 36888f4e2101SHugh Dickins goto unlock; 368930c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 36901da177e4SLinus Torvalds if (!pte_write(entry)) 36918f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 36928f4e2101SHugh Dickins pte, pmd, ptl, entry); 36931da177e4SLinus Torvalds entry = pte_mkdirty(entry); 36941da177e4SLinus Torvalds } 36951da177e4SLinus Torvalds entry = pte_mkyoung(entry); 369630c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 36974b3073e1SRussell King update_mmu_cache(vma, address, pte); 36981a44e149SAndrea Arcangeli } else { 36991a44e149SAndrea Arcangeli /* 37001a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 37011a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 37021a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 37031a44e149SAndrea Arcangeli * with threads. 37041a44e149SAndrea Arcangeli */ 370530c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 370661c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 37071a44e149SAndrea Arcangeli } 37088f4e2101SHugh Dickins unlock: 37098f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 371083c54070SNick Piggin return 0; 37111da177e4SLinus Torvalds } 37121da177e4SLinus Torvalds 37131da177e4SLinus Torvalds /* 37141da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 37151da177e4SLinus Torvalds */ 3716519e5247SJohannes Weiner static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3717d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 37181da177e4SLinus Torvalds { 37191da177e4SLinus Torvalds pgd_t *pgd; 37201da177e4SLinus Torvalds pud_t *pud; 37211da177e4SLinus Torvalds pmd_t *pmd; 37221da177e4SLinus Torvalds pte_t *pte; 37231da177e4SLinus Torvalds 3724ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 372530c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 37261da177e4SLinus Torvalds 37271da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 37281da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 37291da177e4SLinus Torvalds if (!pud) 3730c74df32cSHugh Dickins return VM_FAULT_OOM; 37311da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 37321da177e4SLinus Torvalds if (!pmd) 3733c74df32cSHugh Dickins return VM_FAULT_OOM; 373471e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 3735c0292554SKirill A. Shutemov int ret = VM_FAULT_FALLBACK; 373671e3aac0SAndrea Arcangeli if (!vma->vm_ops) 3737c0292554SKirill A. Shutemov ret = do_huge_pmd_anonymous_page(mm, vma, address, 373871e3aac0SAndrea Arcangeli pmd, flags); 3739c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 3740c0292554SKirill A. Shutemov return ret; 374171e3aac0SAndrea Arcangeli } else { 374271e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 37431f1d06c3SDavid Rientjes int ret; 37441f1d06c3SDavid Rientjes 374571e3aac0SAndrea Arcangeli barrier(); 374671e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3747a1dd450bSWill Deacon unsigned int dirty = flags & FAULT_FLAG_WRITE; 3748a1dd450bSWill Deacon 3749e53289c0SLinus Torvalds /* 3750e53289c0SLinus Torvalds * If the pmd is splitting, return and retry the 3751e53289c0SLinus Torvalds * the fault. Alternative: wait until the split 3752e53289c0SLinus Torvalds * is done, and goto retry. 3753e53289c0SLinus Torvalds */ 3754e53289c0SLinus Torvalds if (pmd_trans_splitting(orig_pmd)) 3755e53289c0SLinus Torvalds return 0; 3756e53289c0SLinus Torvalds 37573d59eebcSLinus Torvalds if (pmd_numa(orig_pmd)) 37584daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3759d10e63f2SMel Gorman orig_pmd, pmd); 3760d10e63f2SMel Gorman 37613d59eebcSLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 37621f1d06c3SDavid Rientjes ret = do_huge_pmd_wp_page(mm, vma, address, pmd, 37631f1d06c3SDavid Rientjes orig_pmd); 37649845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 37651f1d06c3SDavid Rientjes return ret; 3766a1dd450bSWill Deacon } else { 3767a1dd450bSWill Deacon huge_pmd_set_accessed(mm, vma, address, pmd, 3768a1dd450bSWill Deacon orig_pmd, dirty); 376971e3aac0SAndrea Arcangeli return 0; 377071e3aac0SAndrea Arcangeli } 377171e3aac0SAndrea Arcangeli } 37729845cbbdSKirill A. Shutemov } 377371e3aac0SAndrea Arcangeli 37740f19c179SMel Gorman /* THP should already have been handled */ 37750f19c179SMel Gorman BUG_ON(pmd_numa(*pmd)); 3776d10e63f2SMel Gorman 377771e3aac0SAndrea Arcangeli /* 377871e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 377971e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 378071e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 378171e3aac0SAndrea Arcangeli */ 37824fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 37834fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3784c74df32cSHugh Dickins return VM_FAULT_OOM; 378571e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 378671e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 378771e3aac0SAndrea Arcangeli return 0; 378871e3aac0SAndrea Arcangeli /* 378971e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 379071e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 379171e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 379271e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 379371e3aac0SAndrea Arcangeli */ 379471e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 37951da177e4SLinus Torvalds 379630c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 37971da177e4SLinus Torvalds } 37981da177e4SLinus Torvalds 3799519e5247SJohannes Weiner int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3800519e5247SJohannes Weiner unsigned long address, unsigned int flags) 3801519e5247SJohannes Weiner { 3802519e5247SJohannes Weiner int ret; 3803519e5247SJohannes Weiner 3804519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 3805519e5247SJohannes Weiner 3806519e5247SJohannes Weiner count_vm_event(PGFAULT); 3807519e5247SJohannes Weiner mem_cgroup_count_vm_event(mm, PGFAULT); 3808519e5247SJohannes Weiner 3809519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 3810519e5247SJohannes Weiner check_sync_rss_stat(current); 3811519e5247SJohannes Weiner 3812519e5247SJohannes Weiner /* 3813519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 3814519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 3815519e5247SJohannes Weiner */ 3816519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 381749426420SJohannes Weiner mem_cgroup_oom_enable(); 3818519e5247SJohannes Weiner 3819519e5247SJohannes Weiner ret = __handle_mm_fault(mm, vma, address, flags); 3820519e5247SJohannes Weiner 382149426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 382249426420SJohannes Weiner mem_cgroup_oom_disable(); 382349426420SJohannes Weiner /* 382449426420SJohannes Weiner * The task may have entered a memcg OOM situation but 382549426420SJohannes Weiner * if the allocation error was handled gracefully (no 382649426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 382749426420SJohannes Weiner * Just clean up the OOM state peacefully. 382849426420SJohannes Weiner */ 382949426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 383049426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 383149426420SJohannes Weiner } 38323812c8c8SJohannes Weiner 3833519e5247SJohannes Weiner return ret; 3834519e5247SJohannes Weiner } 3835519e5247SJohannes Weiner 38361da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 38371da177e4SLinus Torvalds /* 38381da177e4SLinus Torvalds * Allocate page upper directory. 3839872fec16SHugh Dickins * We've already handled the fast-path in-line. 38401da177e4SLinus Torvalds */ 38411bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 38421da177e4SLinus Torvalds { 3843c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3844c74df32cSHugh Dickins if (!new) 38451bb3630eSHugh Dickins return -ENOMEM; 38461da177e4SLinus Torvalds 3847362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3848362a61adSNick Piggin 3849872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 38501bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 38515e541973SBenjamin Herrenschmidt pud_free(mm, new); 38521bb3630eSHugh Dickins else 38531da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3854872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 38551bb3630eSHugh Dickins return 0; 38561da177e4SLinus Torvalds } 38571da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 38581da177e4SLinus Torvalds 38591da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 38601da177e4SLinus Torvalds /* 38611da177e4SLinus Torvalds * Allocate page middle directory. 3862872fec16SHugh Dickins * We've already handled the fast-path in-line. 38631da177e4SLinus Torvalds */ 38641bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 38651da177e4SLinus Torvalds { 3866c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3867c74df32cSHugh Dickins if (!new) 38681bb3630eSHugh Dickins return -ENOMEM; 38691da177e4SLinus Torvalds 3870362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3871362a61adSNick Piggin 3872872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 38731da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 38741bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 38755e541973SBenjamin Herrenschmidt pmd_free(mm, new); 38761bb3630eSHugh Dickins else 38771da177e4SLinus Torvalds pud_populate(mm, pud, new); 38781da177e4SLinus Torvalds #else 38791bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 38805e541973SBenjamin Herrenschmidt pmd_free(mm, new); 38811bb3630eSHugh Dickins else 38821da177e4SLinus Torvalds pgd_populate(mm, pud, new); 38831da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3884872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 38851bb3630eSHugh Dickins return 0; 38861da177e4SLinus Torvalds } 38871da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 38881da177e4SLinus Torvalds 38891da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 38901da177e4SLinus Torvalds 38911da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 38925ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 38931da177e4SLinus Torvalds 38941da177e4SLinus Torvalds static int __init gate_vma_init(void) 38951da177e4SLinus Torvalds { 38961da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 38971da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 38981da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 3899b6558c4aSRoland McGrath gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; 3900b6558c4aSRoland McGrath gate_vma.vm_page_prot = __P101; 3901909af768SJason Baron 39021da177e4SLinus Torvalds return 0; 39031da177e4SLinus Torvalds } 39041da177e4SLinus Torvalds __initcall(gate_vma_init); 39051da177e4SLinus Torvalds #endif 39061da177e4SLinus Torvalds 390731db58b3SStephen Wilson struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 39081da177e4SLinus Torvalds { 39091da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 39101da177e4SLinus Torvalds return &gate_vma; 39111da177e4SLinus Torvalds #else 39121da177e4SLinus Torvalds return NULL; 39131da177e4SLinus Torvalds #endif 39141da177e4SLinus Torvalds } 39151da177e4SLinus Torvalds 3916cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 39171da177e4SLinus Torvalds { 39181da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 39191da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 39201da177e4SLinus Torvalds return 1; 39211da177e4SLinus Torvalds #endif 39221da177e4SLinus Torvalds return 0; 39231da177e4SLinus Torvalds } 39241da177e4SLinus Torvalds 39251da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 39260ec76a11SDavid Howells 39271b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3928f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3929f8ad0f49SJohannes Weiner { 3930f8ad0f49SJohannes Weiner pgd_t *pgd; 3931f8ad0f49SJohannes Weiner pud_t *pud; 3932f8ad0f49SJohannes Weiner pmd_t *pmd; 3933f8ad0f49SJohannes Weiner pte_t *ptep; 3934f8ad0f49SJohannes Weiner 3935f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3936f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3937f8ad0f49SJohannes Weiner goto out; 3938f8ad0f49SJohannes Weiner 3939f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3940f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3941f8ad0f49SJohannes Weiner goto out; 3942f8ad0f49SJohannes Weiner 3943f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3944f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3945f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3946f8ad0f49SJohannes Weiner goto out; 3947f8ad0f49SJohannes Weiner 3948f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3949f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3950f8ad0f49SJohannes Weiner goto out; 3951f8ad0f49SJohannes Weiner 3952f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3953f8ad0f49SJohannes Weiner if (!ptep) 3954f8ad0f49SJohannes Weiner goto out; 3955f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3956f8ad0f49SJohannes Weiner goto unlock; 3957f8ad0f49SJohannes Weiner *ptepp = ptep; 3958f8ad0f49SJohannes Weiner return 0; 3959f8ad0f49SJohannes Weiner unlock: 3960f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3961f8ad0f49SJohannes Weiner out: 3962f8ad0f49SJohannes Weiner return -EINVAL; 3963f8ad0f49SJohannes Weiner } 3964f8ad0f49SJohannes Weiner 39651b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 39661b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 39671b36ba81SNamhyung Kim { 39681b36ba81SNamhyung Kim int res; 39691b36ba81SNamhyung Kim 39701b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 39711b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 39721b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 39731b36ba81SNamhyung Kim return res; 39741b36ba81SNamhyung Kim } 39751b36ba81SNamhyung Kim 39763b6748e2SJohannes Weiner /** 39773b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 39783b6748e2SJohannes Weiner * @vma: memory mapping 39793b6748e2SJohannes Weiner * @address: user virtual address 39803b6748e2SJohannes Weiner * @pfn: location to store found PFN 39813b6748e2SJohannes Weiner * 39823b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 39833b6748e2SJohannes Weiner * 39843b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 39853b6748e2SJohannes Weiner */ 39863b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 39873b6748e2SJohannes Weiner unsigned long *pfn) 39883b6748e2SJohannes Weiner { 39893b6748e2SJohannes Weiner int ret = -EINVAL; 39903b6748e2SJohannes Weiner spinlock_t *ptl; 39913b6748e2SJohannes Weiner pte_t *ptep; 39923b6748e2SJohannes Weiner 39933b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 39943b6748e2SJohannes Weiner return ret; 39953b6748e2SJohannes Weiner 39963b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 39973b6748e2SJohannes Weiner if (ret) 39983b6748e2SJohannes Weiner return ret; 39993b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 40003b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 40013b6748e2SJohannes Weiner return 0; 40023b6748e2SJohannes Weiner } 40033b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 40043b6748e2SJohannes Weiner 400528b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 4006d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 400728b2ee20SRik van Riel unsigned long address, unsigned int flags, 4008d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 400928b2ee20SRik van Riel { 401003668a4dSJohannes Weiner int ret = -EINVAL; 401128b2ee20SRik van Riel pte_t *ptep, pte; 401228b2ee20SRik van Riel spinlock_t *ptl; 401328b2ee20SRik van Riel 4014d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 4015d87fe660Svenkatesh.pallipadi@intel.com goto out; 401628b2ee20SRik van Riel 401703668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 4018d87fe660Svenkatesh.pallipadi@intel.com goto out; 401928b2ee20SRik van Riel pte = *ptep; 402003668a4dSJohannes Weiner 402128b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 402228b2ee20SRik van Riel goto unlock; 402328b2ee20SRik van Riel 402428b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 402503668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 402628b2ee20SRik van Riel 402703668a4dSJohannes Weiner ret = 0; 402828b2ee20SRik van Riel unlock: 402928b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 403028b2ee20SRik van Riel out: 4031d87fe660Svenkatesh.pallipadi@intel.com return ret; 403228b2ee20SRik van Riel } 403328b2ee20SRik van Riel 403428b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 403528b2ee20SRik van Riel void *buf, int len, int write) 403628b2ee20SRik van Riel { 403728b2ee20SRik van Riel resource_size_t phys_addr; 403828b2ee20SRik van Riel unsigned long prot = 0; 40392bc7273bSKOSAKI Motohiro void __iomem *maddr; 404028b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 404128b2ee20SRik van Riel 4042d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 404328b2ee20SRik van Riel return -EINVAL; 404428b2ee20SRik van Riel 404528b2ee20SRik van Riel maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); 404628b2ee20SRik van Riel if (write) 404728b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 404828b2ee20SRik van Riel else 404928b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 405028b2ee20SRik van Riel iounmap(maddr); 405128b2ee20SRik van Riel 405228b2ee20SRik van Riel return len; 405328b2ee20SRik van Riel } 40545a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 405528b2ee20SRik van Riel #endif 405628b2ee20SRik van Riel 40570ec76a11SDavid Howells /* 4058206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 4059206cb636SStephen Wilson * given task for page fault accounting. 40600ec76a11SDavid Howells */ 4061206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 4062206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 40630ec76a11SDavid Howells { 40640ec76a11SDavid Howells struct vm_area_struct *vma; 40650ec76a11SDavid Howells void *old_buf = buf; 40660ec76a11SDavid Howells 40670ec76a11SDavid Howells down_read(&mm->mmap_sem); 4068183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 40690ec76a11SDavid Howells while (len) { 40700ec76a11SDavid Howells int bytes, ret, offset; 40710ec76a11SDavid Howells void *maddr; 407228b2ee20SRik van Riel struct page *page = NULL; 40730ec76a11SDavid Howells 40740ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 40750ec76a11SDavid Howells write, 1, &page, &vma); 407628b2ee20SRik van Riel if (ret <= 0) { 407728b2ee20SRik van Riel /* 407828b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 407928b2ee20SRik van Riel * we can access using slightly different code. 408028b2ee20SRik van Riel */ 408128b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 408228b2ee20SRik van Riel vma = find_vma(mm, addr); 4083fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 40840ec76a11SDavid Howells break; 408528b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 408628b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 408728b2ee20SRik van Riel len, write); 408828b2ee20SRik van Riel if (ret <= 0) 408928b2ee20SRik van Riel #endif 409028b2ee20SRik van Riel break; 409128b2ee20SRik van Riel bytes = ret; 409228b2ee20SRik van Riel } else { 40930ec76a11SDavid Howells bytes = len; 40940ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 40950ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 40960ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 40970ec76a11SDavid Howells 40980ec76a11SDavid Howells maddr = kmap(page); 40990ec76a11SDavid Howells if (write) { 41000ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 41010ec76a11SDavid Howells maddr + offset, buf, bytes); 41020ec76a11SDavid Howells set_page_dirty_lock(page); 41030ec76a11SDavid Howells } else { 41040ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 41050ec76a11SDavid Howells buf, maddr + offset, bytes); 41060ec76a11SDavid Howells } 41070ec76a11SDavid Howells kunmap(page); 41080ec76a11SDavid Howells page_cache_release(page); 410928b2ee20SRik van Riel } 41100ec76a11SDavid Howells len -= bytes; 41110ec76a11SDavid Howells buf += bytes; 41120ec76a11SDavid Howells addr += bytes; 41130ec76a11SDavid Howells } 41140ec76a11SDavid Howells up_read(&mm->mmap_sem); 41150ec76a11SDavid Howells 41160ec76a11SDavid Howells return buf - old_buf; 41170ec76a11SDavid Howells } 411803252919SAndi Kleen 41195ddd36b9SStephen Wilson /** 4120ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 41215ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 41225ddd36b9SStephen Wilson * @addr: start address to access 41235ddd36b9SStephen Wilson * @buf: source or destination buffer 41245ddd36b9SStephen Wilson * @len: number of bytes to transfer 41255ddd36b9SStephen Wilson * @write: whether the access is a write 41265ddd36b9SStephen Wilson * 41275ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 41285ddd36b9SStephen Wilson */ 41295ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 41305ddd36b9SStephen Wilson void *buf, int len, int write) 41315ddd36b9SStephen Wilson { 41325ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 41335ddd36b9SStephen Wilson } 41345ddd36b9SStephen Wilson 413503252919SAndi Kleen /* 4136206cb636SStephen Wilson * Access another process' address space. 4137206cb636SStephen Wilson * Source/target buffer must be kernel space, 4138206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 4139206cb636SStephen Wilson */ 4140206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 4141206cb636SStephen Wilson void *buf, int len, int write) 4142206cb636SStephen Wilson { 4143206cb636SStephen Wilson struct mm_struct *mm; 4144206cb636SStephen Wilson int ret; 4145206cb636SStephen Wilson 4146206cb636SStephen Wilson mm = get_task_mm(tsk); 4147206cb636SStephen Wilson if (!mm) 4148206cb636SStephen Wilson return 0; 4149206cb636SStephen Wilson 4150206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 4151206cb636SStephen Wilson mmput(mm); 4152206cb636SStephen Wilson 4153206cb636SStephen Wilson return ret; 4154206cb636SStephen Wilson } 4155206cb636SStephen Wilson 415603252919SAndi Kleen /* 415703252919SAndi Kleen * Print the name of a VMA. 415803252919SAndi Kleen */ 415903252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 416003252919SAndi Kleen { 416103252919SAndi Kleen struct mm_struct *mm = current->mm; 416203252919SAndi Kleen struct vm_area_struct *vma; 416303252919SAndi Kleen 4164e8bff74aSIngo Molnar /* 4165e8bff74aSIngo Molnar * Do not print if we are in atomic 4166e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 4167e8bff74aSIngo Molnar */ 4168e8bff74aSIngo Molnar if (preempt_count()) 4169e8bff74aSIngo Molnar return; 4170e8bff74aSIngo Molnar 417103252919SAndi Kleen down_read(&mm->mmap_sem); 417203252919SAndi Kleen vma = find_vma(mm, ip); 417303252919SAndi Kleen if (vma && vma->vm_file) { 417403252919SAndi Kleen struct file *f = vma->vm_file; 417503252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 417603252919SAndi Kleen if (buf) { 41772fbc57c5SAndy Shevchenko char *p; 417803252919SAndi Kleen 4179cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 418003252919SAndi Kleen if (IS_ERR(p)) 418103252919SAndi Kleen p = "?"; 41822fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 418303252919SAndi Kleen vma->vm_start, 418403252919SAndi Kleen vma->vm_end - vma->vm_start); 418503252919SAndi Kleen free_page((unsigned long)buf); 418603252919SAndi Kleen } 418703252919SAndi Kleen } 418851a07e50SJeff Liu up_read(&mm->mmap_sem); 418903252919SAndi Kleen } 41903ee1afa3SNick Piggin 4191662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 41923ee1afa3SNick Piggin void might_fault(void) 41933ee1afa3SNick Piggin { 419495156f00SPeter Zijlstra /* 419595156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 419695156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 419795156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 419895156f00SPeter Zijlstra * below annotations will generate false positives. 419995156f00SPeter Zijlstra */ 420095156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 420195156f00SPeter Zijlstra return; 420295156f00SPeter Zijlstra 42033ee1afa3SNick Piggin /* 42043ee1afa3SNick Piggin * it would be nicer only to annotate paths which are not under 42053ee1afa3SNick Piggin * pagefault_disable, however that requires a larger audit and 42063ee1afa3SNick Piggin * providing helpers like get_user_atomic. 42073ee1afa3SNick Piggin */ 4208662bbcb2SMichael S. Tsirkin if (in_atomic()) 4209662bbcb2SMichael S. Tsirkin return; 4210662bbcb2SMichael S. Tsirkin 4211662bbcb2SMichael S. Tsirkin __might_sleep(__FILE__, __LINE__, 0); 4212662bbcb2SMichael S. Tsirkin 4213662bbcb2SMichael S. Tsirkin if (current->mm) 42143ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 42153ee1afa3SNick Piggin } 42163ee1afa3SNick Piggin EXPORT_SYMBOL(might_fault); 42173ee1afa3SNick Piggin #endif 421847ad8475SAndrea Arcangeli 421947ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 422047ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 422147ad8475SAndrea Arcangeli unsigned long addr, 422247ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 422347ad8475SAndrea Arcangeli { 422447ad8475SAndrea Arcangeli int i; 422547ad8475SAndrea Arcangeli struct page *p = page; 422647ad8475SAndrea Arcangeli 422747ad8475SAndrea Arcangeli might_sleep(); 422847ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 422947ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 423047ad8475SAndrea Arcangeli cond_resched(); 423147ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 423247ad8475SAndrea Arcangeli } 423347ad8475SAndrea Arcangeli } 423447ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 423547ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 423647ad8475SAndrea Arcangeli { 423747ad8475SAndrea Arcangeli int i; 423847ad8475SAndrea Arcangeli 423947ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 424047ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 424147ad8475SAndrea Arcangeli return; 424247ad8475SAndrea Arcangeli } 424347ad8475SAndrea Arcangeli 424447ad8475SAndrea Arcangeli might_sleep(); 424547ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 424647ad8475SAndrea Arcangeli cond_resched(); 424747ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 424847ad8475SAndrea Arcangeli } 424947ad8475SAndrea Arcangeli } 425047ad8475SAndrea Arcangeli 425147ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 425247ad8475SAndrea Arcangeli unsigned long addr, 425347ad8475SAndrea Arcangeli struct vm_area_struct *vma, 425447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 425547ad8475SAndrea Arcangeli { 425647ad8475SAndrea Arcangeli int i; 425747ad8475SAndrea Arcangeli struct page *dst_base = dst; 425847ad8475SAndrea Arcangeli struct page *src_base = src; 425947ad8475SAndrea Arcangeli 426047ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 426147ad8475SAndrea Arcangeli cond_resched(); 426247ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 426347ad8475SAndrea Arcangeli 426447ad8475SAndrea Arcangeli i++; 426547ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 426647ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 426747ad8475SAndrea Arcangeli } 426847ad8475SAndrea Arcangeli } 426947ad8475SAndrea Arcangeli 427047ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 427147ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 427247ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 427347ad8475SAndrea Arcangeli { 427447ad8475SAndrea Arcangeli int i; 427547ad8475SAndrea Arcangeli 427647ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 427747ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 427847ad8475SAndrea Arcangeli pages_per_huge_page); 427947ad8475SAndrea Arcangeli return; 428047ad8475SAndrea Arcangeli } 428147ad8475SAndrea Arcangeli 428247ad8475SAndrea Arcangeli might_sleep(); 428347ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 428447ad8475SAndrea Arcangeli cond_resched(); 428547ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 428647ad8475SAndrea Arcangeli } 428747ad8475SAndrea Arcangeli } 428847ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 428949076ec2SKirill A. Shutemov 429040b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 4291b35f1819SKirill A. Shutemov 4292b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 4293b35f1819SKirill A. Shutemov 4294b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 4295b35f1819SKirill A. Shutemov { 4296b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 4297b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 4298b35f1819SKirill A. Shutemov } 4299b35f1819SKirill A. Shutemov 4300539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 430149076ec2SKirill A. Shutemov { 430249076ec2SKirill A. Shutemov spinlock_t *ptl; 430349076ec2SKirill A. Shutemov 4304b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 430549076ec2SKirill A. Shutemov if (!ptl) 430649076ec2SKirill A. Shutemov return false; 4307539edb58SPeter Zijlstra page->ptl = ptl; 430849076ec2SKirill A. Shutemov return true; 430949076ec2SKirill A. Shutemov } 431049076ec2SKirill A. Shutemov 4311539edb58SPeter Zijlstra void ptlock_free(struct page *page) 431249076ec2SKirill A. Shutemov { 4313b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 431449076ec2SKirill A. Shutemov } 431549076ec2SKirill A. Shutemov #endif 4316