11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/memory.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 91da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* 131da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 141da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 171da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 181da177e4SLinus Torvalds * far as I could see. 191da177e4SLinus Torvalds * 201da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 211da177e4SLinus Torvalds */ 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds /* 241da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 251da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 261da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 271da177e4SLinus Torvalds * Found it. Everything seems to work now. 281da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 331da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 341da177e4SLinus Torvalds * 351da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 361da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 371da177e4SLinus Torvalds * 381da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 391da177e4SLinus Torvalds */ 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/kernel_stat.h> 421da177e4SLinus Torvalds #include <linux/mm.h> 431da177e4SLinus Torvalds #include <linux/hugetlb.h> 441da177e4SLinus Torvalds #include <linux/mman.h> 451da177e4SLinus Torvalds #include <linux/swap.h> 461da177e4SLinus Torvalds #include <linux/highmem.h> 471da177e4SLinus Torvalds #include <linux/pagemap.h> 489a840895SHugh Dickins #include <linux/ksm.h> 491da177e4SLinus Torvalds #include <linux/rmap.h> 50b95f1b31SPaul Gortmaker #include <linux/export.h> 510ff92245SShailabh Nagar #include <linux/delayacct.h> 521da177e4SLinus Torvalds #include <linux/init.h> 53edc79b2aSPeter Zijlstra #include <linux/writeback.h> 548a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 55cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 563dc14741SHugh Dickins #include <linux/kallsyms.h> 573dc14741SHugh Dickins #include <linux/swapops.h> 583dc14741SHugh Dickins #include <linux/elf.h> 595a0e3ad6STejun Heo #include <linux/gfp.h> 604daae3b4SMel Gorman #include <linux/migrate.h> 612fbc57c5SAndy Shevchenko #include <linux/string.h> 620abdd7a8SDan Williams #include <linux/dma-debug.h> 631592eef0SKirill A. Shutemov #include <linux/debugfs.h> 641da177e4SLinus Torvalds 656952b61dSAlexey Dobriyan #include <asm/io.h> 661da177e4SLinus Torvalds #include <asm/pgalloc.h> 671da177e4SLinus Torvalds #include <asm/uaccess.h> 681da177e4SLinus Torvalds #include <asm/tlb.h> 691da177e4SLinus Torvalds #include <asm/tlbflush.h> 701da177e4SLinus Torvalds #include <asm/pgtable.h> 711da177e4SLinus Torvalds 7242b77728SJan Beulich #include "internal.h" 7342b77728SJan Beulich 7490572890SPeter Zijlstra #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS 7590572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 7675980e97SPeter Zijlstra #endif 7775980e97SPeter Zijlstra 78d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 791da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 801da177e4SLinus Torvalds unsigned long max_mapnr; 811da177e4SLinus Torvalds struct page *mem_map; 821da177e4SLinus Torvalds 831da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 841da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 851da177e4SLinus Torvalds #endif 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds /* 881da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 891da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 901da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 911da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 921da177e4SLinus Torvalds * and ZONE_HIGHMEM. 931da177e4SLinus Torvalds */ 941da177e4SLinus Torvalds void * high_memory; 951da177e4SLinus Torvalds 961da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 971da177e4SLinus Torvalds 9832a93233SIngo Molnar /* 9932a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 10032a93233SIngo Molnar * 10132a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 10232a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 10332a93233SIngo Molnar */ 10432a93233SIngo Molnar int randomize_va_space __read_mostly = 10532a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 10632a93233SIngo Molnar 1; 10732a93233SIngo Molnar #else 10832a93233SIngo Molnar 2; 10932a93233SIngo Molnar #endif 110a62eaf15SAndi Kleen 111a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 112a62eaf15SAndi Kleen { 113a62eaf15SAndi Kleen randomize_va_space = 0; 1149b41046cSOGAWA Hirofumi return 1; 115a62eaf15SAndi Kleen } 116a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 117a62eaf15SAndi Kleen 11862eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 11903f6462aSHugh Dickins unsigned long highest_memmap_pfn __read_mostly; 120a13ea5b7SHugh Dickins 1210b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1220b70068eSArd Biesheuvel 123a13ea5b7SHugh Dickins /* 124a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 125a13ea5b7SHugh Dickins */ 126a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 127a13ea5b7SHugh Dickins { 128a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 129a13ea5b7SHugh Dickins return 0; 130a13ea5b7SHugh Dickins } 131a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 132a62eaf15SAndi Kleen 133d559db08SKAMEZAWA Hiroyuki 13434e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 13534e55232SKAMEZAWA Hiroyuki 136ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 13734e55232SKAMEZAWA Hiroyuki { 13834e55232SKAMEZAWA Hiroyuki int i; 13934e55232SKAMEZAWA Hiroyuki 14034e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 14105af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 14205af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 14305af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 14434e55232SKAMEZAWA Hiroyuki } 14534e55232SKAMEZAWA Hiroyuki } 14605af2e10SDavid Rientjes current->rss_stat.events = 0; 14734e55232SKAMEZAWA Hiroyuki } 14834e55232SKAMEZAWA Hiroyuki 14934e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 15034e55232SKAMEZAWA Hiroyuki { 15134e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 15234e55232SKAMEZAWA Hiroyuki 15334e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 15434e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 15534e55232SKAMEZAWA Hiroyuki else 15634e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 15734e55232SKAMEZAWA Hiroyuki } 15834e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 15934e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 16034e55232SKAMEZAWA Hiroyuki 16134e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 16234e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 16334e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 16434e55232SKAMEZAWA Hiroyuki { 16534e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 16634e55232SKAMEZAWA Hiroyuki return; 16734e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 168ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 16934e55232SKAMEZAWA Hiroyuki } 1709547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 17134e55232SKAMEZAWA Hiroyuki 17234e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 17334e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 17434e55232SKAMEZAWA Hiroyuki 17534e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 17634e55232SKAMEZAWA Hiroyuki { 17734e55232SKAMEZAWA Hiroyuki } 17834e55232SKAMEZAWA Hiroyuki 1799547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 1809547d01bSPeter Zijlstra 1819547d01bSPeter Zijlstra #ifdef HAVE_GENERIC_MMU_GATHER 1829547d01bSPeter Zijlstra 1839547d01bSPeter Zijlstra static int tlb_next_batch(struct mmu_gather *tlb) 1849547d01bSPeter Zijlstra { 1859547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 1869547d01bSPeter Zijlstra 1879547d01bSPeter Zijlstra batch = tlb->active; 1889547d01bSPeter Zijlstra if (batch->next) { 1899547d01bSPeter Zijlstra tlb->active = batch->next; 1909547d01bSPeter Zijlstra return 1; 1919547d01bSPeter Zijlstra } 1929547d01bSPeter Zijlstra 19353a59fc6SMichal Hocko if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) 19453a59fc6SMichal Hocko return 0; 19553a59fc6SMichal Hocko 1969547d01bSPeter Zijlstra batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 1979547d01bSPeter Zijlstra if (!batch) 1989547d01bSPeter Zijlstra return 0; 1999547d01bSPeter Zijlstra 20053a59fc6SMichal Hocko tlb->batch_count++; 2019547d01bSPeter Zijlstra batch->next = NULL; 2029547d01bSPeter Zijlstra batch->nr = 0; 2039547d01bSPeter Zijlstra batch->max = MAX_GATHER_BATCH; 2049547d01bSPeter Zijlstra 2059547d01bSPeter Zijlstra tlb->active->next = batch; 2069547d01bSPeter Zijlstra tlb->active = batch; 2079547d01bSPeter Zijlstra 2089547d01bSPeter Zijlstra return 1; 2099547d01bSPeter Zijlstra } 2109547d01bSPeter Zijlstra 2119547d01bSPeter Zijlstra /* tlb_gather_mmu 2129547d01bSPeter Zijlstra * Called to initialize an (on-stack) mmu_gather structure for page-table 2139547d01bSPeter Zijlstra * tear-down from @mm. The @fullmm argument is used when @mm is without 2149547d01bSPeter Zijlstra * users and we're going to destroy the full address space (exit/execve). 2159547d01bSPeter Zijlstra */ 2162b047252SLinus Torvalds void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) 2179547d01bSPeter Zijlstra { 2189547d01bSPeter Zijlstra tlb->mm = mm; 2199547d01bSPeter Zijlstra 2202b047252SLinus Torvalds /* Is it from 0 to ~0? */ 2212b047252SLinus Torvalds tlb->fullmm = !(start | (end+1)); 2221de14c3cSDave Hansen tlb->need_flush_all = 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 232fb7332a9SWill Deacon 233fb7332a9SWill Deacon __tlb_reset_range(tlb); 2349547d01bSPeter Zijlstra } 2359547d01bSPeter Zijlstra 2361cf35d47SLinus Torvalds static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) 2379547d01bSPeter Zijlstra { 238721c21c1SWill Deacon if (!tlb->end) 239721c21c1SWill Deacon return; 240721c21c1SWill Deacon 2419547d01bSPeter Zijlstra tlb_flush(tlb); 24234ee645eSJoerg Roedel mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); 2439547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2449547d01bSPeter Zijlstra tlb_table_flush(tlb); 2459547d01bSPeter Zijlstra #endif 246fb7332a9SWill Deacon __tlb_reset_range(tlb); 2471cf35d47SLinus Torvalds } 2481cf35d47SLinus Torvalds 2491cf35d47SLinus Torvalds static void tlb_flush_mmu_free(struct mmu_gather *tlb) 2501cf35d47SLinus Torvalds { 2511cf35d47SLinus Torvalds struct mmu_gather_batch *batch; 2529547d01bSPeter Zijlstra 253721c21c1SWill Deacon for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { 2549547d01bSPeter Zijlstra free_pages_and_swap_cache(batch->pages, batch->nr); 2559547d01bSPeter Zijlstra batch->nr = 0; 2569547d01bSPeter Zijlstra } 2579547d01bSPeter Zijlstra tlb->active = &tlb->local; 2589547d01bSPeter Zijlstra } 2599547d01bSPeter Zijlstra 2601cf35d47SLinus Torvalds void tlb_flush_mmu(struct mmu_gather *tlb) 2611cf35d47SLinus Torvalds { 2621cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 2631cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 2641cf35d47SLinus Torvalds } 2651cf35d47SLinus Torvalds 2669547d01bSPeter Zijlstra /* tlb_finish_mmu 2679547d01bSPeter Zijlstra * Called at the end of the shootdown operation to free up any resources 2689547d01bSPeter Zijlstra * that were required. 2699547d01bSPeter Zijlstra */ 2709547d01bSPeter Zijlstra void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 2719547d01bSPeter Zijlstra { 2729547d01bSPeter Zijlstra struct mmu_gather_batch *batch, *next; 2739547d01bSPeter Zijlstra 2749547d01bSPeter Zijlstra tlb_flush_mmu(tlb); 2759547d01bSPeter Zijlstra 2769547d01bSPeter Zijlstra /* keep the page table cache within bounds */ 2779547d01bSPeter Zijlstra check_pgt_cache(); 2789547d01bSPeter Zijlstra 2799547d01bSPeter Zijlstra for (batch = tlb->local.next; batch; batch = next) { 2809547d01bSPeter Zijlstra next = batch->next; 2819547d01bSPeter Zijlstra free_pages((unsigned long)batch, 0); 2829547d01bSPeter Zijlstra } 2839547d01bSPeter Zijlstra tlb->local.next = NULL; 2849547d01bSPeter Zijlstra } 2859547d01bSPeter Zijlstra 2869547d01bSPeter Zijlstra /* __tlb_remove_page 2879547d01bSPeter Zijlstra * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while 2889547d01bSPeter Zijlstra * handling the additional races in SMP caused by other CPUs caching valid 2899547d01bSPeter Zijlstra * mappings in their TLBs. Returns the number of free page slots left. 2909547d01bSPeter Zijlstra * When out of page slots we must call tlb_flush_mmu(). 2919547d01bSPeter Zijlstra */ 2929547d01bSPeter Zijlstra int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 2939547d01bSPeter Zijlstra { 2949547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2959547d01bSPeter Zijlstra 296fb7332a9SWill Deacon VM_BUG_ON(!tlb->end); 2979547d01bSPeter Zijlstra 2989547d01bSPeter Zijlstra batch = tlb->active; 2999547d01bSPeter Zijlstra batch->pages[batch->nr++] = page; 3009547d01bSPeter Zijlstra if (batch->nr == batch->max) { 3019547d01bSPeter Zijlstra if (!tlb_next_batch(tlb)) 3029547d01bSPeter Zijlstra return 0; 3030b43c3aaSShaohua Li batch = tlb->active; 3049547d01bSPeter Zijlstra } 305309381feSSasha Levin VM_BUG_ON_PAGE(batch->nr > batch->max, page); 3069547d01bSPeter Zijlstra 3079547d01bSPeter Zijlstra return batch->max - batch->nr; 3089547d01bSPeter Zijlstra } 3099547d01bSPeter Zijlstra 3109547d01bSPeter Zijlstra #endif /* HAVE_GENERIC_MMU_GATHER */ 31134e55232SKAMEZAWA Hiroyuki 31226723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 31326723911SPeter Zijlstra 31426723911SPeter Zijlstra /* 31526723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 31626723911SPeter Zijlstra */ 31726723911SPeter Zijlstra 31826723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 31926723911SPeter Zijlstra { 32026723911SPeter Zijlstra /* Simply deliver the interrupt */ 32126723911SPeter Zijlstra } 32226723911SPeter Zijlstra 32326723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 32426723911SPeter Zijlstra { 32526723911SPeter Zijlstra /* 32626723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 32726723911SPeter Zijlstra * assumed to be actually RCU-freed. 32826723911SPeter Zijlstra * 32926723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 33026723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 33126723911SPeter Zijlstra */ 33226723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 33326723911SPeter Zijlstra __tlb_remove_table(table); 33426723911SPeter Zijlstra } 33526723911SPeter Zijlstra 33626723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 33726723911SPeter Zijlstra { 33826723911SPeter Zijlstra struct mmu_table_batch *batch; 33926723911SPeter Zijlstra int i; 34026723911SPeter Zijlstra 34126723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 34226723911SPeter Zijlstra 34326723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 34426723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 34526723911SPeter Zijlstra 34626723911SPeter Zijlstra free_page((unsigned long)batch); 34726723911SPeter Zijlstra } 34826723911SPeter Zijlstra 34926723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 35026723911SPeter Zijlstra { 35126723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 35226723911SPeter Zijlstra 35326723911SPeter Zijlstra if (*batch) { 35426723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 35526723911SPeter Zijlstra *batch = NULL; 35626723911SPeter Zijlstra } 35726723911SPeter Zijlstra } 35826723911SPeter Zijlstra 35926723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 36026723911SPeter Zijlstra { 36126723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 36226723911SPeter Zijlstra 36326723911SPeter Zijlstra /* 36426723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 36526723911SPeter Zijlstra * concurrent page-table walk. 36626723911SPeter Zijlstra */ 36726723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 36826723911SPeter Zijlstra __tlb_remove_table(table); 36926723911SPeter Zijlstra return; 37026723911SPeter Zijlstra } 37126723911SPeter Zijlstra 37226723911SPeter Zijlstra if (*batch == NULL) { 37326723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 37426723911SPeter Zijlstra if (*batch == NULL) { 37526723911SPeter Zijlstra tlb_remove_table_one(table); 37626723911SPeter Zijlstra return; 37726723911SPeter Zijlstra } 37826723911SPeter Zijlstra (*batch)->nr = 0; 37926723911SPeter Zijlstra } 38026723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 38126723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 38226723911SPeter Zijlstra tlb_table_flush(tlb); 38326723911SPeter Zijlstra } 38426723911SPeter Zijlstra 3859547d01bSPeter Zijlstra #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ 38626723911SPeter Zijlstra 3871da177e4SLinus Torvalds /* 3881da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 3891da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 3901da177e4SLinus Torvalds */ 3919e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 3929e1b32caSBenjamin Herrenschmidt unsigned long addr) 3931da177e4SLinus Torvalds { 3942f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 3951da177e4SLinus Torvalds pmd_clear(pmd); 3969e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 397e1f56c89SKirill A. Shutemov atomic_long_dec(&tlb->mm->nr_ptes); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 400e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 401e0da382cSHugh Dickins unsigned long addr, unsigned long end, 402e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4031da177e4SLinus Torvalds { 4041da177e4SLinus Torvalds pmd_t *pmd; 4051da177e4SLinus Torvalds unsigned long next; 406e0da382cSHugh Dickins unsigned long start; 4071da177e4SLinus Torvalds 408e0da382cSHugh Dickins start = addr; 4091da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 4101da177e4SLinus Torvalds do { 4111da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 4121da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 4131da177e4SLinus Torvalds continue; 4149e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 4151da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 4161da177e4SLinus Torvalds 417e0da382cSHugh Dickins start &= PUD_MASK; 418e0da382cSHugh Dickins if (start < floor) 419e0da382cSHugh Dickins return; 420e0da382cSHugh Dickins if (ceiling) { 421e0da382cSHugh Dickins ceiling &= PUD_MASK; 422e0da382cSHugh Dickins if (!ceiling) 423e0da382cSHugh Dickins return; 4241da177e4SLinus Torvalds } 425e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 426e0da382cSHugh Dickins return; 427e0da382cSHugh Dickins 428e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 429e0da382cSHugh Dickins pud_clear(pud); 4309e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 431dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 4321da177e4SLinus Torvalds } 4331da177e4SLinus Torvalds 434e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 435e0da382cSHugh Dickins unsigned long addr, unsigned long end, 436e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4371da177e4SLinus Torvalds { 4381da177e4SLinus Torvalds pud_t *pud; 4391da177e4SLinus Torvalds unsigned long next; 440e0da382cSHugh Dickins unsigned long start; 4411da177e4SLinus Torvalds 442e0da382cSHugh Dickins start = addr; 4431da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 4441da177e4SLinus Torvalds do { 4451da177e4SLinus Torvalds next = pud_addr_end(addr, end); 4461da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 4471da177e4SLinus Torvalds continue; 448e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 4491da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 4501da177e4SLinus Torvalds 451e0da382cSHugh Dickins start &= PGDIR_MASK; 452e0da382cSHugh Dickins if (start < floor) 453e0da382cSHugh Dickins return; 454e0da382cSHugh Dickins if (ceiling) { 455e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 456e0da382cSHugh Dickins if (!ceiling) 457e0da382cSHugh Dickins return; 4581da177e4SLinus Torvalds } 459e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 460e0da382cSHugh Dickins return; 461e0da382cSHugh Dickins 462e0da382cSHugh Dickins pud = pud_offset(pgd, start); 463e0da382cSHugh Dickins pgd_clear(pgd); 4649e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 4651da177e4SLinus Torvalds } 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds /* 468e0da382cSHugh Dickins * This function frees user-level page tables of a process. 4691da177e4SLinus Torvalds */ 47042b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 471e0da382cSHugh Dickins unsigned long addr, unsigned long end, 472e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4731da177e4SLinus Torvalds { 4741da177e4SLinus Torvalds pgd_t *pgd; 4751da177e4SLinus Torvalds unsigned long next; 4761da177e4SLinus Torvalds 477e0da382cSHugh Dickins /* 478e0da382cSHugh Dickins * The next few lines have given us lots of grief... 479e0da382cSHugh Dickins * 480e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 481e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 482e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 483e0da382cSHugh Dickins * 484e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 485e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 486e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 487e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 488e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 489e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 490e0da382cSHugh Dickins * that end 0 case should be mythical). 491e0da382cSHugh Dickins * 492e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 493e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 494e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 495e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 496e0da382cSHugh Dickins * 497e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 498e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 499e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 500e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 501e0da382cSHugh Dickins */ 502e0da382cSHugh Dickins 503e0da382cSHugh Dickins addr &= PMD_MASK; 504e0da382cSHugh Dickins if (addr < floor) { 505e0da382cSHugh Dickins addr += PMD_SIZE; 506e0da382cSHugh Dickins if (!addr) 507e0da382cSHugh Dickins return; 508e0da382cSHugh Dickins } 509e0da382cSHugh Dickins if (ceiling) { 510e0da382cSHugh Dickins ceiling &= PMD_MASK; 511e0da382cSHugh Dickins if (!ceiling) 512e0da382cSHugh Dickins return; 513e0da382cSHugh Dickins } 514e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 515e0da382cSHugh Dickins end -= PMD_SIZE; 516e0da382cSHugh Dickins if (addr > end - 1) 517e0da382cSHugh Dickins return; 518e0da382cSHugh Dickins 51942b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 5201da177e4SLinus Torvalds do { 5211da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 5221da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 5231da177e4SLinus Torvalds continue; 52442b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 5251da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 526e0da382cSHugh Dickins } 527e0da382cSHugh Dickins 52842b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 529e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 530e0da382cSHugh Dickins { 531e0da382cSHugh Dickins while (vma) { 532e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 533e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 534e0da382cSHugh Dickins 5358f4f8c16SHugh Dickins /* 53625d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 53725d9e2d1Snpiggin@suse.de * pgtables 5388f4f8c16SHugh Dickins */ 5395beb4930SRik van Riel unlink_anon_vmas(vma); 5408f4f8c16SHugh Dickins unlink_file_vma(vma); 5418f4f8c16SHugh Dickins 5429da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 5433bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 5443bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 5453bf5ee95SHugh Dickins } else { 5463bf5ee95SHugh Dickins /* 5473bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 5483bf5ee95SHugh Dickins */ 5493bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 5504866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 551e0da382cSHugh Dickins vma = next; 552e0da382cSHugh Dickins next = vma->vm_next; 5535beb4930SRik van Riel unlink_anon_vmas(vma); 5548f4f8c16SHugh Dickins unlink_file_vma(vma); 555e0da382cSHugh Dickins } 5563bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 557e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 5583bf5ee95SHugh Dickins } 559e0da382cSHugh Dickins vma = next; 560e0da382cSHugh Dickins } 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 5638ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 5648ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 5651da177e4SLinus Torvalds { 566c4088ebdSKirill A. Shutemov spinlock_t *ptl; 5672f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 5688ac1f832SAndrea Arcangeli int wait_split_huge_page; 5691da177e4SLinus Torvalds if (!new) 5701bb3630eSHugh Dickins return -ENOMEM; 5711bb3630eSHugh Dickins 572362a61adSNick Piggin /* 573362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 574362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 575362a61adSNick Piggin * put into page tables. 576362a61adSNick Piggin * 577362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 578362a61adSNick Piggin * table walking code (when walking the page table without locking; 579362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 580362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 581362a61adSNick Piggin * being the notable exception) will already guarantee loads are 582362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 583362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 584362a61adSNick Piggin */ 585362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 586362a61adSNick Piggin 587c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 5888ac1f832SAndrea Arcangeli wait_split_huge_page = 0; 5898ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 590e1f56c89SKirill A. Shutemov atomic_long_inc(&mm->nr_ptes); 5911da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 5922f569afdSMartin Schwidefsky new = NULL; 5938ac1f832SAndrea Arcangeli } else if (unlikely(pmd_trans_splitting(*pmd))) 5948ac1f832SAndrea Arcangeli wait_split_huge_page = 1; 595c4088ebdSKirill A. Shutemov spin_unlock(ptl); 5962f569afdSMartin Schwidefsky if (new) 5972f569afdSMartin Schwidefsky pte_free(mm, new); 5988ac1f832SAndrea Arcangeli if (wait_split_huge_page) 5998ac1f832SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 6001bb3630eSHugh Dickins return 0; 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds 6031bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 6041da177e4SLinus Torvalds { 6051bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 6061da177e4SLinus Torvalds if (!new) 6071bb3630eSHugh Dickins return -ENOMEM; 6081da177e4SLinus Torvalds 609362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 610362a61adSNick Piggin 611872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 6128ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 613872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 6142f569afdSMartin Schwidefsky new = NULL; 6158ac1f832SAndrea Arcangeli } else 6168ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_splitting(*pmd)); 617872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 6182f569afdSMartin Schwidefsky if (new) 6192f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 6201bb3630eSHugh Dickins return 0; 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds 623d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 624ae859762SHugh Dickins { 625d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 626d559db08SKAMEZAWA Hiroyuki } 627d559db08SKAMEZAWA Hiroyuki 628d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 629d559db08SKAMEZAWA Hiroyuki { 630d559db08SKAMEZAWA Hiroyuki int i; 631d559db08SKAMEZAWA Hiroyuki 63234e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 63305af2e10SDavid Rientjes sync_mm_rss(mm); 634d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 635d559db08SKAMEZAWA Hiroyuki if (rss[i]) 636d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 637ae859762SHugh Dickins } 638ae859762SHugh Dickins 6391da177e4SLinus Torvalds /* 6406aab341eSLinus Torvalds * This function is called to print an error when a bad pte 6416aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 6426aab341eSLinus Torvalds * a region that doesn't allow it. 643b5810039SNick Piggin * 644b5810039SNick Piggin * The calling function must still handle the error. 645b5810039SNick Piggin */ 6463dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 6473dc14741SHugh Dickins pte_t pte, struct page *page) 648b5810039SNick Piggin { 6493dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 6503dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 6513dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 6523dc14741SHugh Dickins struct address_space *mapping; 6533dc14741SHugh Dickins pgoff_t index; 654d936cf9bSHugh Dickins static unsigned long resume; 655d936cf9bSHugh Dickins static unsigned long nr_shown; 656d936cf9bSHugh Dickins static unsigned long nr_unshown; 657d936cf9bSHugh Dickins 658d936cf9bSHugh Dickins /* 659d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 660d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 661d936cf9bSHugh Dickins */ 662d936cf9bSHugh Dickins if (nr_shown == 60) { 663d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 664d936cf9bSHugh Dickins nr_unshown++; 665d936cf9bSHugh Dickins return; 666d936cf9bSHugh Dickins } 667d936cf9bSHugh Dickins if (nr_unshown) { 6681e9e6365SHugh Dickins printk(KERN_ALERT 6691e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 670d936cf9bSHugh Dickins nr_unshown); 671d936cf9bSHugh Dickins nr_unshown = 0; 672d936cf9bSHugh Dickins } 673d936cf9bSHugh Dickins nr_shown = 0; 674d936cf9bSHugh Dickins } 675d936cf9bSHugh Dickins if (nr_shown++ == 0) 676d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 6773dc14741SHugh Dickins 6783dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 6793dc14741SHugh Dickins index = linear_page_index(vma, addr); 6803dc14741SHugh Dickins 6811e9e6365SHugh Dickins printk(KERN_ALERT 6821e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 6833dc14741SHugh Dickins current->comm, 6843dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 685718a3821SWu Fengguang if (page) 686f0b791a3SDave Hansen dump_page(page, "bad pte"); 6871e9e6365SHugh Dickins printk(KERN_ALERT 6883dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 6893dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 6903dc14741SHugh Dickins /* 6913dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 6923dc14741SHugh Dickins */ 6932682582aSKonstantin Khlebnikov pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", 6942682582aSKonstantin Khlebnikov vma->vm_file, 6952682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 6962682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 6972682582aSKonstantin Khlebnikov mapping ? mapping->a_ops->readpage : NULL); 698b5810039SNick Piggin dump_stack(); 699373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 700b5810039SNick Piggin } 701b5810039SNick Piggin 702b5810039SNick Piggin /* 7037e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 7046aab341eSLinus Torvalds * 7057e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7067e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7077e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 708b379d790SJared Hulbert * 7097e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7107e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7117e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7127e675137SNick Piggin * described below. 7137e675137SNick Piggin * 7147e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7157e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7167e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7176aab341eSLinus Torvalds * 718b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 719b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7207e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7217e675137SNick Piggin * mapping will always honor the rule 7226aab341eSLinus Torvalds * 7236aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7246aab341eSLinus Torvalds * 7257e675137SNick Piggin * And for normal mappings this is false. 726b379d790SJared Hulbert * 7277e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7287e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7297e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7307e675137SNick Piggin * special (because none can have been COWed). 731b379d790SJared Hulbert * 732b379d790SJared Hulbert * 7337e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7347e675137SNick Piggin * 735b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 736b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 737b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 738b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 739b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 740b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 741b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 742b379d790SJared Hulbert * 743ee498ed7SHugh Dickins */ 7447e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7457e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7467e675137SNick Piggin #else 7477e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7487e675137SNick Piggin #endif 7497e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7507e675137SNick Piggin pte_t pte) 751ee498ed7SHugh Dickins { 75222b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7537e675137SNick Piggin 7547e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 755b38af472SHugh Dickins if (likely(!pte_special(pte))) 75622b31eecSHugh Dickins goto check_pfn; 757667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 758667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 759a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 760a13ea5b7SHugh Dickins return NULL; 76162eede62SHugh Dickins if (!is_zero_pfn(pfn)) 76222b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7637e675137SNick Piggin return NULL; 7647e675137SNick Piggin } 7657e675137SNick Piggin 7667e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7677e675137SNick Piggin 768b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 769b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 770b379d790SJared Hulbert if (!pfn_valid(pfn)) 771b379d790SJared Hulbert return NULL; 772b379d790SJared Hulbert goto out; 773b379d790SJared Hulbert } else { 7747e675137SNick Piggin unsigned long off; 7757e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7766aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7776aab341eSLinus Torvalds return NULL; 77867121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 779fb155c16SLinus Torvalds return NULL; 7806aab341eSLinus Torvalds } 781b379d790SJared Hulbert } 7826aab341eSLinus Torvalds 783b38af472SHugh Dickins if (is_zero_pfn(pfn)) 784b38af472SHugh Dickins return NULL; 78522b31eecSHugh Dickins check_pfn: 78622b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 78722b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 78822b31eecSHugh Dickins return NULL; 78922b31eecSHugh Dickins } 7906aab341eSLinus Torvalds 7916aab341eSLinus Torvalds /* 7927e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7937e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7946aab341eSLinus Torvalds */ 795b379d790SJared Hulbert out: 7966aab341eSLinus Torvalds return pfn_to_page(pfn); 797ee498ed7SHugh Dickins } 798ee498ed7SHugh Dickins 799ee498ed7SHugh Dickins /* 8001da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 8011da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 8021da177e4SLinus Torvalds * covered by this vma. 8031da177e4SLinus Torvalds */ 8041da177e4SLinus Torvalds 805570a335bSHugh Dickins static inline unsigned long 8061da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 807b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8088c103762SHugh Dickins unsigned long addr, int *rss) 8091da177e4SLinus Torvalds { 810b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8111da177e4SLinus Torvalds pte_t pte = *src_pte; 8121da177e4SLinus Torvalds struct page *page; 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8151da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8160697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8170697212aSChristoph Lameter 8182022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 819570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 820570a335bSHugh Dickins return entry.val; 821570a335bSHugh Dickins 8221da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8231da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8241da177e4SLinus Torvalds spin_lock(&mmlist_lock); 825f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 826f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 827f412ac08SHugh Dickins &src_mm->mmlist); 8281da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8291da177e4SLinus Torvalds } 830b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8312022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 8329f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8339f9f1acdSKonstantin Khlebnikov 8349f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 8359f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]++; 8369f9f1acdSKonstantin Khlebnikov else 8379f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]++; 8389f9f1acdSKonstantin Khlebnikov 8399f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8400697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8410697212aSChristoph Lameter /* 8429f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8439f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8440697212aSChristoph Lameter */ 8450697212aSChristoph Lameter make_migration_entry_read(&entry); 8460697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 847c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 848c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 8490697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8500697212aSChristoph Lameter } 8511da177e4SLinus Torvalds } 852ae859762SHugh Dickins goto out_set_pte; 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds /* 8561da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8571da177e4SLinus Torvalds * in the parent and the child 8581da177e4SLinus Torvalds */ 85967121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8601da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8613dc90795SZachary Amsden pte = pte_wrprotect(pte); 8621da177e4SLinus Torvalds } 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds /* 8651da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8661da177e4SLinus Torvalds * the child 8671da177e4SLinus Torvalds */ 8681da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8691da177e4SLinus Torvalds pte = pte_mkclean(pte); 8701da177e4SLinus Torvalds pte = pte_mkold(pte); 8716aab341eSLinus Torvalds 8726aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 8736aab341eSLinus Torvalds if (page) { 8741da177e4SLinus Torvalds get_page(page); 87521333b2bSHugh Dickins page_dup_rmap(page); 876d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 877d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 878d559db08SKAMEZAWA Hiroyuki else 879d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 8806aab341eSLinus Torvalds } 881ae859762SHugh Dickins 882ae859762SHugh Dickins out_set_pte: 883ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 884570a335bSHugh Dickins return 0; 8851da177e4SLinus Torvalds } 8861da177e4SLinus Torvalds 88721bda264SJerome Marchand static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8881da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8891da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8901da177e4SLinus Torvalds { 891c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8921da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 893c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 894e040f218SHugh Dickins int progress = 0; 895d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 896570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 8971da177e4SLinus Torvalds 8981da177e4SLinus Torvalds again: 899d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 900d559db08SKAMEZAWA Hiroyuki 901c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 9021da177e4SLinus Torvalds if (!dst_pte) 9031da177e4SLinus Torvalds return -ENOMEM; 904ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 9054c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 906f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 907c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 908c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9096606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds do { 9121da177e4SLinus Torvalds /* 9131da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9141da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9151da177e4SLinus Torvalds */ 916e040f218SHugh Dickins if (progress >= 32) { 917e040f218SHugh Dickins progress = 0; 918e040f218SHugh Dickins if (need_resched() || 91995c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9201da177e4SLinus Torvalds break; 921e040f218SHugh Dickins } 9221da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9231da177e4SLinus Torvalds progress++; 9241da177e4SLinus Torvalds continue; 9251da177e4SLinus Torvalds } 926570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 927570a335bSHugh Dickins vma, addr, rss); 928570a335bSHugh Dickins if (entry.val) 929570a335bSHugh Dickins break; 9301da177e4SLinus Torvalds progress += 8; 9311da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9321da177e4SLinus Torvalds 9336606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 934c74df32cSHugh Dickins spin_unlock(src_ptl); 935ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 936d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 937c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 938c74df32cSHugh Dickins cond_resched(); 939570a335bSHugh Dickins 940570a335bSHugh Dickins if (entry.val) { 941570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 942570a335bSHugh Dickins return -ENOMEM; 943570a335bSHugh Dickins progress = 0; 944570a335bSHugh Dickins } 9451da177e4SLinus Torvalds if (addr != end) 9461da177e4SLinus Torvalds goto again; 9471da177e4SLinus Torvalds return 0; 9481da177e4SLinus Torvalds } 9491da177e4SLinus Torvalds 9501da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9511da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9521da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9531da177e4SLinus Torvalds { 9541da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9551da177e4SLinus Torvalds unsigned long next; 9561da177e4SLinus Torvalds 9571da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9581da177e4SLinus Torvalds if (!dst_pmd) 9591da177e4SLinus Torvalds return -ENOMEM; 9601da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9611da177e4SLinus Torvalds do { 9621da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 96371e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 96471e3aac0SAndrea Arcangeli int err; 96514d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 96671e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 96771e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 96871e3aac0SAndrea Arcangeli if (err == -ENOMEM) 96971e3aac0SAndrea Arcangeli return -ENOMEM; 97071e3aac0SAndrea Arcangeli if (!err) 97171e3aac0SAndrea Arcangeli continue; 97271e3aac0SAndrea Arcangeli /* fall through */ 97371e3aac0SAndrea Arcangeli } 9741da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 9751da177e4SLinus Torvalds continue; 9761da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 9771da177e4SLinus Torvalds vma, addr, next)) 9781da177e4SLinus Torvalds return -ENOMEM; 9791da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 9801da177e4SLinus Torvalds return 0; 9811da177e4SLinus Torvalds } 9821da177e4SLinus Torvalds 9831da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9841da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9851da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9861da177e4SLinus Torvalds { 9871da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9881da177e4SLinus Torvalds unsigned long next; 9891da177e4SLinus Torvalds 9901da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9911da177e4SLinus Torvalds if (!dst_pud) 9921da177e4SLinus Torvalds return -ENOMEM; 9931da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9941da177e4SLinus Torvalds do { 9951da177e4SLinus Torvalds next = pud_addr_end(addr, end); 9961da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 9971da177e4SLinus Torvalds continue; 9981da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 9991da177e4SLinus Torvalds vma, addr, next)) 10001da177e4SLinus Torvalds return -ENOMEM; 10011da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 10021da177e4SLinus Torvalds return 0; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10061da177e4SLinus Torvalds struct vm_area_struct *vma) 10071da177e4SLinus Torvalds { 10081da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 10091da177e4SLinus Torvalds unsigned long next; 10101da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10111da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10122ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10132ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10142ec74c3eSSagi Grimberg bool is_cow; 1015cddb8a5cSAndrea Arcangeli int ret; 10161da177e4SLinus Torvalds 1017d992895bSNick Piggin /* 1018d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1019d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1020d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1021d992895bSNick Piggin * efficient than faulting. 1022d992895bSNick Piggin */ 10230661a336SKirill A. Shutemov if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && 10240661a336SKirill A. Shutemov !vma->anon_vma) 1025d992895bSNick Piggin return 0; 1026d992895bSNick Piggin 10271da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10281da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10291da177e4SLinus Torvalds 1030b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10312ab64037Svenkatesh.pallipadi@intel.com /* 10322ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10332ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10342ab64037Svenkatesh.pallipadi@intel.com */ 10355180da41SSuresh Siddha ret = track_pfn_copy(vma); 10362ab64037Svenkatesh.pallipadi@intel.com if (ret) 10372ab64037Svenkatesh.pallipadi@intel.com return ret; 10382ab64037Svenkatesh.pallipadi@intel.com } 10392ab64037Svenkatesh.pallipadi@intel.com 1040cddb8a5cSAndrea Arcangeli /* 1041cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1042cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1043cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1044cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1045cddb8a5cSAndrea Arcangeli */ 10462ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10472ec74c3eSSagi Grimberg mmun_start = addr; 10482ec74c3eSSagi Grimberg mmun_end = end; 10492ec74c3eSSagi Grimberg if (is_cow) 10502ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10512ec74c3eSSagi Grimberg mmun_end); 1052cddb8a5cSAndrea Arcangeli 1053cddb8a5cSAndrea Arcangeli ret = 0; 10541da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10551da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10561da177e4SLinus Torvalds do { 10571da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10581da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10591da177e4SLinus Torvalds continue; 1060cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1061cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1062cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1063cddb8a5cSAndrea Arcangeli break; 1064cddb8a5cSAndrea Arcangeli } 10651da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1066cddb8a5cSAndrea Arcangeli 10672ec74c3eSSagi Grimberg if (is_cow) 10682ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1069cddb8a5cSAndrea Arcangeli return ret; 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 107251c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1073b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 10741da177e4SLinus Torvalds unsigned long addr, unsigned long end, 107597a89413SPeter Zijlstra struct zap_details *details) 10761da177e4SLinus Torvalds { 1077b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1078d16dfc55SPeter Zijlstra int force_flush = 0; 1079d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 108097a89413SPeter Zijlstra spinlock_t *ptl; 10815f1a1907SSteven Rostedt pte_t *start_pte; 108297a89413SPeter Zijlstra pte_t *pte; 10838a5f14a2SKirill A. Shutemov swp_entry_t entry; 1084d559db08SKAMEZAWA Hiroyuki 1085d16dfc55SPeter Zijlstra again: 1086e303297eSPeter Zijlstra init_rss_vec(rss); 10875f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10885f1a1907SSteven Rostedt pte = start_pte; 10896606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10901da177e4SLinus Torvalds do { 10911da177e4SLinus Torvalds pte_t ptent = *pte; 109251c6f666SRobin Holt if (pte_none(ptent)) { 10931da177e4SLinus Torvalds continue; 109451c6f666SRobin Holt } 109551c6f666SRobin Holt 10966f5e6b9eSHugh Dickins if (pte_present(ptent)) { 10976f5e6b9eSHugh Dickins struct page *page; 10986f5e6b9eSHugh Dickins 10996aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 11001da177e4SLinus Torvalds if (unlikely(details) && page) { 11011da177e4SLinus Torvalds /* 11021da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 11031da177e4SLinus Torvalds * invalidate cache without truncating: 11041da177e4SLinus Torvalds * unmap shared but keep private pages. 11051da177e4SLinus Torvalds */ 11061da177e4SLinus Torvalds if (details->check_mapping && 11071da177e4SLinus Torvalds details->check_mapping != page->mapping) 11081da177e4SLinus Torvalds continue; 11091da177e4SLinus Torvalds } 1110b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1111a600388dSZachary Amsden tlb->fullmm); 11121da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11131da177e4SLinus Torvalds if (unlikely(!page)) 11141da177e4SLinus Torvalds continue; 11151da177e4SLinus Torvalds if (PageAnon(page)) 1116d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 11176237bcd9SHugh Dickins else { 11181cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 11191cf35d47SLinus Torvalds force_flush = 1; 11206237bcd9SHugh Dickins set_page_dirty(page); 11211cf35d47SLinus Torvalds } 11224917e5d0SJohannes Weiner if (pte_young(ptent) && 112364363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1124bf3f3bc5SNick Piggin mark_page_accessed(page); 1125d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 11266237bcd9SHugh Dickins } 1127edc315fdSHugh Dickins page_remove_rmap(page); 11283dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11293dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 11301cf35d47SLinus Torvalds if (unlikely(!__tlb_remove_page(tlb, page))) { 11311cf35d47SLinus Torvalds force_flush = 1; 1132ce9ec37bSWill Deacon addr += PAGE_SIZE; 1133d16dfc55SPeter Zijlstra break; 11341cf35d47SLinus Torvalds } 11351da177e4SLinus Torvalds continue; 11361da177e4SLinus Torvalds } 11378a5f14a2SKirill A. Shutemov /* If details->check_mapping, we leave swap entries. */ 11381da177e4SLinus Torvalds if (unlikely(details)) 11391da177e4SLinus Torvalds continue; 1140b084d435SKAMEZAWA Hiroyuki 11418a5f14a2SKirill A. Shutemov entry = pte_to_swp_entry(ptent); 1142b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1143b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11449f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11459f9f1acdSKonstantin Khlebnikov struct page *page; 11469f9f1acdSKonstantin Khlebnikov 11479f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 11489f9f1acdSKonstantin Khlebnikov 11499f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 11509f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]--; 11519f9f1acdSKonstantin Khlebnikov else 11529f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]--; 11539f9f1acdSKonstantin Khlebnikov } 1154b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 11552509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 11569888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 115797a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1158ae859762SHugh Dickins 1159d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 11606606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 116151c6f666SRobin Holt 11621cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 1163fb7332a9SWill Deacon if (force_flush) 11641cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 11651cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 11661cf35d47SLinus Torvalds 11671cf35d47SLinus Torvalds /* 11681cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 11691cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 11701cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 11711cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 11721cf35d47SLinus Torvalds */ 11731cf35d47SLinus Torvalds if (force_flush) { 11741cf35d47SLinus Torvalds force_flush = 0; 11751cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 11762b047252SLinus Torvalds 11772b047252SLinus Torvalds if (addr != end) 1178d16dfc55SPeter Zijlstra goto again; 1179d16dfc55SPeter Zijlstra } 1180d16dfc55SPeter Zijlstra 118151c6f666SRobin Holt return addr; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 118451c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1185b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 11861da177e4SLinus Torvalds unsigned long addr, unsigned long end, 118797a89413SPeter Zijlstra struct zap_details *details) 11881da177e4SLinus Torvalds { 11891da177e4SLinus Torvalds pmd_t *pmd; 11901da177e4SLinus Torvalds unsigned long next; 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 11931da177e4SLinus Torvalds do { 11941da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 119571e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 119614d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1197e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1198e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1199e0897d75SDavid 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", 1200e0897d75SDavid Rientjes __func__, addr, end, 1201e0897d75SDavid Rientjes vma->vm_start, 1202e0897d75SDavid Rientjes vma->vm_end); 1203e0897d75SDavid Rientjes BUG(); 1204e0897d75SDavid Rientjes } 1205e0897d75SDavid Rientjes #endif 1206e180377fSKirill A. Shutemov split_huge_page_pmd(vma, addr, pmd); 1207f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 12081a5a9906SAndrea Arcangeli goto next; 120971e3aac0SAndrea Arcangeli /* fall through */ 121071e3aac0SAndrea Arcangeli } 12111a5a9906SAndrea Arcangeli /* 12121a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 12131a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 12141a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 12151a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12161a5a9906SAndrea Arcangeli * mode. 12171a5a9906SAndrea Arcangeli */ 12181a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12191a5a9906SAndrea Arcangeli goto next; 122097a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12211a5a9906SAndrea Arcangeli next: 122297a89413SPeter Zijlstra cond_resched(); 122397a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 122451c6f666SRobin Holt 122551c6f666SRobin Holt return addr; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds 122851c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1229b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12301da177e4SLinus Torvalds unsigned long addr, unsigned long end, 123197a89413SPeter Zijlstra struct zap_details *details) 12321da177e4SLinus Torvalds { 12331da177e4SLinus Torvalds pud_t *pud; 12341da177e4SLinus Torvalds unsigned long next; 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12371da177e4SLinus Torvalds do { 12381da177e4SLinus Torvalds next = pud_addr_end(addr, end); 123997a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12401da177e4SLinus Torvalds continue; 124197a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 124297a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 124351c6f666SRobin Holt 124451c6f666SRobin Holt return addr; 12451da177e4SLinus Torvalds } 12461da177e4SLinus Torvalds 1247038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 124851c6f666SRobin Holt struct vm_area_struct *vma, 12491da177e4SLinus Torvalds unsigned long addr, unsigned long end, 125097a89413SPeter Zijlstra struct zap_details *details) 12511da177e4SLinus Torvalds { 12521da177e4SLinus Torvalds pgd_t *pgd; 12531da177e4SLinus Torvalds unsigned long next; 12541da177e4SLinus Torvalds 12558a5f14a2SKirill A. Shutemov if (details && !details->check_mapping) 12561da177e4SLinus Torvalds details = NULL; 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds BUG_ON(addr >= end); 12591da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 12601da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 12611da177e4SLinus Torvalds do { 12621da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 126397a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 12641da177e4SLinus Torvalds continue; 126597a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 126697a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 12671da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 12681da177e4SLinus Torvalds } 12691da177e4SLinus Torvalds 1270f5cc4eefSAl Viro 1271f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 12721da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 12734f74d2c8SLinus Torvalds unsigned long end_addr, 12741da177e4SLinus Torvalds struct zap_details *details) 12751da177e4SLinus Torvalds { 1276f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 12771da177e4SLinus Torvalds unsigned long end; 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds if (start >= vma->vm_end) 1280f5cc4eefSAl Viro return; 12811da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 12821da177e4SLinus Torvalds if (end <= vma->vm_start) 1283f5cc4eefSAl Viro return; 12841da177e4SLinus Torvalds 1285cbc91f71SSrikar Dronamraju if (vma->vm_file) 1286cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1287cbc91f71SSrikar Dronamraju 1288b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 12895180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 12902ab64037Svenkatesh.pallipadi@intel.com 12918b2a1238SAl Viro if (start != end) { 129251c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1293a137e1ccSAndi Kleen /* 1294a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1295a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1296a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 12977aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1298a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 12997aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1300a137e1ccSAndi Kleen * before calling this function to clean up. 1301a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1302a137e1ccSAndi Kleen * safe to do nothing in this case. 1303a137e1ccSAndi Kleen */ 130424669e58SAneesh Kumar K.V if (vma->vm_file) { 130583cde9e8SDavidlohr Bueso i_mmap_lock_write(vma->vm_file->f_mapping); 1306d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 130783cde9e8SDavidlohr Bueso i_mmap_unlock_write(vma->vm_file->f_mapping); 130824669e58SAneesh Kumar K.V } 130951c6f666SRobin Holt } else 1310038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 131197a89413SPeter Zijlstra } 131251c6f666SRobin Holt } 13131da177e4SLinus Torvalds 1314f5cc4eefSAl Viro /** 1315f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1316f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1317f5cc4eefSAl Viro * @vma: the starting vma 1318f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1319f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1320f5cc4eefSAl Viro * 1321f5cc4eefSAl Viro * Unmap all pages in the vma list. 1322f5cc4eefSAl Viro * 1323f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1324f5cc4eefSAl Viro * 1325f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1326f5cc4eefSAl Viro * 1327f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1328f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1329f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1330f5cc4eefSAl Viro * drops the lock and schedules. 1331f5cc4eefSAl Viro */ 1332f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1333f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13344f74d2c8SLinus Torvalds unsigned long end_addr) 1335f5cc4eefSAl Viro { 1336f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1337f5cc4eefSAl Viro 1338f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1339f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13404f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1341cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds /** 13451da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13461da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1347eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13481da177e4SLinus Torvalds * @size: number of bytes to zap 13498a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1350f5cc4eefSAl Viro * 1351f5cc4eefSAl Viro * Caller must protect the VMA list 13521da177e4SLinus Torvalds */ 13537e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 13541da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 13551da177e4SLinus Torvalds { 13561da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1357d16dfc55SPeter Zijlstra struct mmu_gather tlb; 13587e027b14SLinus Torvalds unsigned long end = start + size; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds lru_add_drain(); 13612b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 1362365e9c87SHugh Dickins update_hiwater_rss(mm); 13637e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 13647e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 13654f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 13667e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 13677e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 1370c627f9ccSJack Steiner /** 1371f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1372f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1373f5cc4eefSAl Viro * @address: starting address of pages to zap 1374f5cc4eefSAl Viro * @size: number of bytes to zap 13758a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1376f5cc4eefSAl Viro * 1377f5cc4eefSAl Viro * The range must fit into one VMA. 1378f5cc4eefSAl Viro */ 1379f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1380f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1381f5cc4eefSAl Viro { 1382f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1383f5cc4eefSAl Viro struct mmu_gather tlb; 1384f5cc4eefSAl Viro unsigned long end = address + size; 1385f5cc4eefSAl Viro 1386f5cc4eefSAl Viro lru_add_drain(); 13872b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, address, end); 1388f5cc4eefSAl Viro update_hiwater_rss(mm); 1389f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 13904f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1391f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1392f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds 1395c627f9ccSJack Steiner /** 1396c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1397c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1398c627f9ccSJack Steiner * @address: starting address of pages to zap 1399c627f9ccSJack Steiner * @size: number of bytes to zap 1400c627f9ccSJack Steiner * 1401c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1402c627f9ccSJack Steiner * 1403c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1404c627f9ccSJack Steiner * 1405c627f9ccSJack Steiner * Returns 0 if successful. 1406c627f9ccSJack Steiner */ 1407c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1408c627f9ccSJack Steiner unsigned long size) 1409c627f9ccSJack Steiner { 1410c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1411c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1412c627f9ccSJack Steiner return -1; 1413f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1414c627f9ccSJack Steiner return 0; 1415c627f9ccSJack Steiner } 1416c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1417c627f9ccSJack Steiner 141825ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 1419920c7a5dSHarvey Harrison spinlock_t **ptl) 1420c9cfcddfSLinus Torvalds { 1421c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1422c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1423c9cfcddfSLinus Torvalds if (pud) { 142449c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1425f66055abSAndrea Arcangeli if (pmd) { 1426f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 1427c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1428c9cfcddfSLinus Torvalds } 1429f66055abSAndrea Arcangeli } 1430c9cfcddfSLinus Torvalds return NULL; 1431c9cfcddfSLinus Torvalds } 1432c9cfcddfSLinus Torvalds 14331da177e4SLinus Torvalds /* 1434238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1435238f58d8SLinus Torvalds * 1436238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1437238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1438238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1439238f58d8SLinus Torvalds */ 1440423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1441423bad60SNick Piggin struct page *page, pgprot_t prot) 1442238f58d8SLinus Torvalds { 1443423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1444238f58d8SLinus Torvalds int retval; 1445238f58d8SLinus Torvalds pte_t *pte; 1446238f58d8SLinus Torvalds spinlock_t *ptl; 1447238f58d8SLinus Torvalds 1448238f58d8SLinus Torvalds retval = -EINVAL; 1449a145dd41SLinus Torvalds if (PageAnon(page)) 14505b4e655eSKAMEZAWA Hiroyuki goto out; 1451238f58d8SLinus Torvalds retval = -ENOMEM; 1452238f58d8SLinus Torvalds flush_dcache_page(page); 1453c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1454238f58d8SLinus Torvalds if (!pte) 14555b4e655eSKAMEZAWA Hiroyuki goto out; 1456238f58d8SLinus Torvalds retval = -EBUSY; 1457238f58d8SLinus Torvalds if (!pte_none(*pte)) 1458238f58d8SLinus Torvalds goto out_unlock; 1459238f58d8SLinus Torvalds 1460238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1461238f58d8SLinus Torvalds get_page(page); 146234e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 1463238f58d8SLinus Torvalds page_add_file_rmap(page); 1464238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1465238f58d8SLinus Torvalds 1466238f58d8SLinus Torvalds retval = 0; 14678a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 14688a9f3ccdSBalbir Singh return retval; 1469238f58d8SLinus Torvalds out_unlock: 1470238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1471238f58d8SLinus Torvalds out: 1472238f58d8SLinus Torvalds return retval; 1473238f58d8SLinus Torvalds } 1474238f58d8SLinus Torvalds 1475bfa5bf6dSRolf Eike Beer /** 1476bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1477bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1478bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1479bfa5bf6dSRolf Eike Beer * @page: source kernel page 1480bfa5bf6dSRolf Eike Beer * 1481a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1482a145dd41SLinus Torvalds * into a user vma. 1483a145dd41SLinus Torvalds * 1484a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1485a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1486a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 14878dfcc9baSNick Piggin * (see split_page()). 1488a145dd41SLinus Torvalds * 1489a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1490a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1491a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1492a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1493a145dd41SLinus Torvalds * ask for a shared writable mapping! 1494a145dd41SLinus Torvalds * 1495a145dd41SLinus Torvalds * The page does not need to be reserved. 14964b6e1e37SKonstantin Khlebnikov * 14974b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 14984b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 14994b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 15004b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 1501a145dd41SLinus Torvalds */ 1502423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 1503423bad60SNick Piggin struct page *page) 1504a145dd41SLinus Torvalds { 1505a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1506a145dd41SLinus Torvalds return -EFAULT; 1507a145dd41SLinus Torvalds if (!page_count(page)) 1508a145dd41SLinus Torvalds return -EINVAL; 15094b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 15104b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 15114b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 15124b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 15134b6e1e37SKonstantin Khlebnikov } 1514423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1515a145dd41SLinus Torvalds } 1516e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1517a145dd41SLinus Torvalds 1518423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1519423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 1520423bad60SNick Piggin { 1521423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1522423bad60SNick Piggin int retval; 1523423bad60SNick Piggin pte_t *pte, entry; 1524423bad60SNick Piggin spinlock_t *ptl; 1525423bad60SNick Piggin 1526423bad60SNick Piggin retval = -ENOMEM; 1527423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1528423bad60SNick Piggin if (!pte) 1529423bad60SNick Piggin goto out; 1530423bad60SNick Piggin retval = -EBUSY; 1531423bad60SNick Piggin if (!pte_none(*pte)) 1532423bad60SNick Piggin goto out_unlock; 1533423bad60SNick Piggin 1534423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 1535423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 1536423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 15374b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 1538423bad60SNick Piggin 1539423bad60SNick Piggin retval = 0; 1540423bad60SNick Piggin out_unlock: 1541423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 1542423bad60SNick Piggin out: 1543423bad60SNick Piggin return retval; 1544423bad60SNick Piggin } 1545423bad60SNick Piggin 1546e0dc0d8fSNick Piggin /** 1547e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 1548e0dc0d8fSNick Piggin * @vma: user vma to map to 1549e0dc0d8fSNick Piggin * @addr: target user address of this page 1550e0dc0d8fSNick Piggin * @pfn: source kernel pfn 1551e0dc0d8fSNick Piggin * 1552c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 1553e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 1554e0dc0d8fSNick Piggin * 1555e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 1556e0dc0d8fSNick Piggin * in that case the handler should return NULL. 15570d71d10aSNick Piggin * 15580d71d10aSNick Piggin * vma cannot be a COW mapping. 15590d71d10aSNick Piggin * 15600d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 15610d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 1562e0dc0d8fSNick Piggin */ 1563e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1564e0dc0d8fSNick Piggin unsigned long pfn) 1565e0dc0d8fSNick Piggin { 15662ab64037Svenkatesh.pallipadi@intel.com int ret; 1567e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 15687e675137SNick Piggin /* 15697e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 15707e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 15717e675137SNick Piggin * consistency in testing and feature parity among all, so we should 15727e675137SNick Piggin * try to keep these invariants in place for everybody. 15737e675137SNick Piggin */ 1574b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 1575b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 1576b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 1577b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 1578b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 1579e0dc0d8fSNick Piggin 1580423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1581423bad60SNick Piggin return -EFAULT; 15825180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 15832ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 15842ab64037Svenkatesh.pallipadi@intel.com 1585e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 15862ab64037Svenkatesh.pallipadi@intel.com 15872ab64037Svenkatesh.pallipadi@intel.com return ret; 1588e0dc0d8fSNick Piggin } 1589e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 1590e0dc0d8fSNick Piggin 1591423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 1592423bad60SNick Piggin unsigned long pfn) 1593423bad60SNick Piggin { 1594423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 1595423bad60SNick Piggin 1596423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1597423bad60SNick Piggin return -EFAULT; 1598423bad60SNick Piggin 1599423bad60SNick Piggin /* 1600423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 1601423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 1602423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 160362eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 160462eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 1605423bad60SNick Piggin */ 1606423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 1607423bad60SNick Piggin struct page *page; 1608423bad60SNick Piggin 1609423bad60SNick Piggin page = pfn_to_page(pfn); 1610423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1611423bad60SNick Piggin } 1612423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 1613423bad60SNick Piggin } 1614423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 1615423bad60SNick Piggin 1616a145dd41SLinus Torvalds /* 16171da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 16181da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 16191da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 16201da177e4SLinus Torvalds */ 16211da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 16221da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16231da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16241da177e4SLinus Torvalds { 16251da177e4SLinus Torvalds pte_t *pte; 1626c74df32cSHugh Dickins spinlock_t *ptl; 16271da177e4SLinus Torvalds 1628c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 16291da177e4SLinus Torvalds if (!pte) 16301da177e4SLinus Torvalds return -ENOMEM; 16316606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 16321da177e4SLinus Torvalds do { 16331da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 16347e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 16351da177e4SLinus Torvalds pfn++; 16361da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 16376606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1638c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 16391da177e4SLinus Torvalds return 0; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 16431da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16441da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16451da177e4SLinus Torvalds { 16461da177e4SLinus Torvalds pmd_t *pmd; 16471da177e4SLinus Torvalds unsigned long next; 16481da177e4SLinus Torvalds 16491da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16501da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 16511da177e4SLinus Torvalds if (!pmd) 16521da177e4SLinus Torvalds return -ENOMEM; 1653f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 16541da177e4SLinus Torvalds do { 16551da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 16561da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 16571da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16581da177e4SLinus Torvalds return -ENOMEM; 16591da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 16601da177e4SLinus Torvalds return 0; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 16641da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16651da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16661da177e4SLinus Torvalds { 16671da177e4SLinus Torvalds pud_t *pud; 16681da177e4SLinus Torvalds unsigned long next; 16691da177e4SLinus Torvalds 16701da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16711da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 16721da177e4SLinus Torvalds if (!pud) 16731da177e4SLinus Torvalds return -ENOMEM; 16741da177e4SLinus Torvalds do { 16751da177e4SLinus Torvalds next = pud_addr_end(addr, end); 16761da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 16771da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16781da177e4SLinus Torvalds return -ENOMEM; 16791da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 16801da177e4SLinus Torvalds return 0; 16811da177e4SLinus Torvalds } 16821da177e4SLinus Torvalds 1683bfa5bf6dSRolf Eike Beer /** 1684bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 1685bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1686bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 1687bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 1688bfa5bf6dSRolf Eike Beer * @size: size of map area 1689bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 1690bfa5bf6dSRolf Eike Beer * 1691bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 1692bfa5bf6dSRolf Eike Beer */ 16931da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 16941da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 16951da177e4SLinus Torvalds { 16961da177e4SLinus Torvalds pgd_t *pgd; 16971da177e4SLinus Torvalds unsigned long next; 16982d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 16991da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 17001da177e4SLinus Torvalds int err; 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds /* 17031da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 17041da177e4SLinus Torvalds * rest of the world about it: 17051da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 17061da177e4SLinus Torvalds * (accesses can have side effects). 17076aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 17086aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 17096aab341eSLinus Torvalds * with them. 1710314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 1711314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 1712314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 1713314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 1714fb155c16SLinus Torvalds * 1715fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 1716fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 1717fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 1718b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 17191da177e4SLinus Torvalds */ 1720b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 1721b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 1722b3b9c293SKonstantin Khlebnikov return -EINVAL; 17236aab341eSLinus Torvalds vma->vm_pgoff = pfn; 1724b3b9c293SKonstantin Khlebnikov } 1725b3b9c293SKonstantin Khlebnikov 1726b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 1727b3b9c293SKonstantin Khlebnikov if (err) 17283c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 1729fb155c16SLinus Torvalds 1730314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds BUG_ON(addr >= end); 17331da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 17341da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 17351da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 17361da177e4SLinus Torvalds do { 17371da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 17381da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 17391da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 17401da177e4SLinus Torvalds if (err) 17411da177e4SLinus Torvalds break; 17421da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 17432ab64037Svenkatesh.pallipadi@intel.com 17442ab64037Svenkatesh.pallipadi@intel.com if (err) 17455180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 17462ab64037Svenkatesh.pallipadi@intel.com 17471da177e4SLinus Torvalds return err; 17481da177e4SLinus Torvalds } 17491da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 17501da177e4SLinus Torvalds 1751b4cbb197SLinus Torvalds /** 1752b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 1753b4cbb197SLinus Torvalds * @vma: user vma to map to 1754b4cbb197SLinus Torvalds * @start: start of area 1755b4cbb197SLinus Torvalds * @len: size of area 1756b4cbb197SLinus Torvalds * 1757b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 1758b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 1759b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 1760b4cbb197SLinus Torvalds * 1761b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 1762b4cbb197SLinus Torvalds * whatever write-combining details or similar. 1763b4cbb197SLinus Torvalds */ 1764b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 1765b4cbb197SLinus Torvalds { 1766b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 1767b4cbb197SLinus Torvalds 1768b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 1769b4cbb197SLinus Torvalds if (start + len < start) 1770b4cbb197SLinus Torvalds return -EINVAL; 1771b4cbb197SLinus Torvalds /* 1772b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 1773b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 1774b4cbb197SLinus Torvalds * just have smaller alignment. 1775b4cbb197SLinus Torvalds */ 1776b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 1777b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 1778b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 1779b4cbb197SLinus Torvalds if (pfn + pages < pfn) 1780b4cbb197SLinus Torvalds return -EINVAL; 1781b4cbb197SLinus Torvalds 1782b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 1783b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 1784b4cbb197SLinus Torvalds return -EINVAL; 1785b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 1786b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 1787b4cbb197SLinus Torvalds 1788b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 1789b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 1790b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 1791b4cbb197SLinus Torvalds return -EINVAL; 1792b4cbb197SLinus Torvalds 1793b4cbb197SLinus Torvalds /* Ok, let it rip */ 1794b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 1795b4cbb197SLinus Torvalds } 1796b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 1797b4cbb197SLinus Torvalds 1798aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 1799aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1800aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1801aee16b3cSJeremy Fitzhardinge { 1802aee16b3cSJeremy Fitzhardinge pte_t *pte; 1803aee16b3cSJeremy Fitzhardinge int err; 18042f569afdSMartin Schwidefsky pgtable_t token; 180594909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 1806aee16b3cSJeremy Fitzhardinge 1807aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 1808aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 1809aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 1810aee16b3cSJeremy Fitzhardinge if (!pte) 1811aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1812aee16b3cSJeremy Fitzhardinge 1813aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 1814aee16b3cSJeremy Fitzhardinge 181538e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 181638e0edb1SJeremy Fitzhardinge 18172f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 1818aee16b3cSJeremy Fitzhardinge 1819aee16b3cSJeremy Fitzhardinge do { 1820c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 1821aee16b3cSJeremy Fitzhardinge if (err) 1822aee16b3cSJeremy Fitzhardinge break; 1823c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 1824aee16b3cSJeremy Fitzhardinge 182538e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 182638e0edb1SJeremy Fitzhardinge 1827aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 1828aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 1829aee16b3cSJeremy Fitzhardinge return err; 1830aee16b3cSJeremy Fitzhardinge } 1831aee16b3cSJeremy Fitzhardinge 1832aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 1833aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1834aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1835aee16b3cSJeremy Fitzhardinge { 1836aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 1837aee16b3cSJeremy Fitzhardinge unsigned long next; 1838aee16b3cSJeremy Fitzhardinge int err; 1839aee16b3cSJeremy Fitzhardinge 1840ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 1841ceb86879SAndi Kleen 1842aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 1843aee16b3cSJeremy Fitzhardinge if (!pmd) 1844aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1845aee16b3cSJeremy Fitzhardinge do { 1846aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 1847aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 1848aee16b3cSJeremy Fitzhardinge if (err) 1849aee16b3cSJeremy Fitzhardinge break; 1850aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 1851aee16b3cSJeremy Fitzhardinge return err; 1852aee16b3cSJeremy Fitzhardinge } 1853aee16b3cSJeremy Fitzhardinge 1854aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 1855aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1856aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1857aee16b3cSJeremy Fitzhardinge { 1858aee16b3cSJeremy Fitzhardinge pud_t *pud; 1859aee16b3cSJeremy Fitzhardinge unsigned long next; 1860aee16b3cSJeremy Fitzhardinge int err; 1861aee16b3cSJeremy Fitzhardinge 1862aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 1863aee16b3cSJeremy Fitzhardinge if (!pud) 1864aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1865aee16b3cSJeremy Fitzhardinge do { 1866aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 1867aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 1868aee16b3cSJeremy Fitzhardinge if (err) 1869aee16b3cSJeremy Fitzhardinge break; 1870aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 1871aee16b3cSJeremy Fitzhardinge return err; 1872aee16b3cSJeremy Fitzhardinge } 1873aee16b3cSJeremy Fitzhardinge 1874aee16b3cSJeremy Fitzhardinge /* 1875aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 1876aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 1877aee16b3cSJeremy Fitzhardinge */ 1878aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 1879aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 1880aee16b3cSJeremy Fitzhardinge { 1881aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 1882aee16b3cSJeremy Fitzhardinge unsigned long next; 188357250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 1884aee16b3cSJeremy Fitzhardinge int err; 1885aee16b3cSJeremy Fitzhardinge 1886aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 1887aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 1888aee16b3cSJeremy Fitzhardinge do { 1889aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 1890aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 1891aee16b3cSJeremy Fitzhardinge if (err) 1892aee16b3cSJeremy Fitzhardinge break; 1893aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 189457250a5bSJeremy Fitzhardinge 1895aee16b3cSJeremy Fitzhardinge return err; 1896aee16b3cSJeremy Fitzhardinge } 1897aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 1898aee16b3cSJeremy Fitzhardinge 18991da177e4SLinus Torvalds /* 19009b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 19019b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 19029b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 19039b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 19049b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 1905a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 19068f4e2101SHugh Dickins */ 19074c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 19088f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 19098f4e2101SHugh Dickins { 19108f4e2101SHugh Dickins int same = 1; 19118f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 19128f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 19134c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 19144c21e2f2SHugh Dickins spin_lock(ptl); 19158f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 19164c21e2f2SHugh Dickins spin_unlock(ptl); 19178f4e2101SHugh Dickins } 19188f4e2101SHugh Dickins #endif 19198f4e2101SHugh Dickins pte_unmap(page_table); 19208f4e2101SHugh Dickins return same; 19218f4e2101SHugh Dickins } 19228f4e2101SHugh Dickins 19239de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 19246aab341eSLinus Torvalds { 19250abdd7a8SDan Williams debug_dma_assert_idle(src); 19260abdd7a8SDan Williams 19276aab341eSLinus Torvalds /* 19286aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 19296aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 19306aab341eSLinus Torvalds * just copying from the original user address. If that 19316aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 19326aab341eSLinus Torvalds */ 19336aab341eSLinus Torvalds if (unlikely(!src)) { 19349b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 19355d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 19365d2a2dbbSLinus Torvalds 19375d2a2dbbSLinus Torvalds /* 19385d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 19395d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 19405d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 19415d2a2dbbSLinus Torvalds * zeroes. 19425d2a2dbbSLinus Torvalds */ 19435d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 19443ecb01dfSJan Beulich clear_page(kaddr); 19459b04c5feSCong Wang kunmap_atomic(kaddr); 1946c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 19470ed361deSNick Piggin } else 19489de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 19496aab341eSLinus Torvalds } 19506aab341eSLinus Torvalds 19511da177e4SLinus Torvalds /* 1952fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 1953fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 1954fb09a464SKirill A. Shutemov * 1955fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 1956fb09a464SKirill A. Shutemov */ 1957fb09a464SKirill A. Shutemov static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, 1958fb09a464SKirill A. Shutemov unsigned long address) 1959fb09a464SKirill A. Shutemov { 1960fb09a464SKirill A. Shutemov struct vm_fault vmf; 1961fb09a464SKirill A. Shutemov int ret; 1962fb09a464SKirill A. Shutemov 1963fb09a464SKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 1964fb09a464SKirill A. Shutemov vmf.pgoff = page->index; 1965fb09a464SKirill A. Shutemov vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 1966fb09a464SKirill A. Shutemov vmf.page = page; 19672e4cdab0SMatthew Wilcox vmf.cow_page = NULL; 1968fb09a464SKirill A. Shutemov 1969fb09a464SKirill A. Shutemov ret = vma->vm_ops->page_mkwrite(vma, &vmf); 1970fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 1971fb09a464SKirill A. Shutemov return ret; 1972fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 1973fb09a464SKirill A. Shutemov lock_page(page); 1974fb09a464SKirill A. Shutemov if (!page->mapping) { 1975fb09a464SKirill A. Shutemov unlock_page(page); 1976fb09a464SKirill A. Shutemov return 0; /* retry */ 1977fb09a464SKirill A. Shutemov } 1978fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 1979fb09a464SKirill A. Shutemov } else 1980fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 1981fb09a464SKirill A. Shutemov return ret; 1982fb09a464SKirill A. Shutemov } 1983fb09a464SKirill A. Shutemov 1984fb09a464SKirill A. Shutemov /* 19854e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 19864e047f89SShachar Raindel * 19874e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 19884e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 19894e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 19904e047f89SShachar Raindel * any related book-keeping. 19914e047f89SShachar Raindel */ 19924e047f89SShachar Raindel static inline int wp_page_reuse(struct mm_struct *mm, 19934e047f89SShachar Raindel struct vm_area_struct *vma, unsigned long address, 19944e047f89SShachar Raindel pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, 19954e047f89SShachar Raindel struct page *page, int page_mkwrite, 19964e047f89SShachar Raindel int dirty_shared) 19974e047f89SShachar Raindel __releases(ptl) 19984e047f89SShachar Raindel { 19994e047f89SShachar Raindel pte_t entry; 20004e047f89SShachar Raindel /* 20014e047f89SShachar Raindel * Clear the pages cpupid information as the existing 20024e047f89SShachar Raindel * information potentially belongs to a now completely 20034e047f89SShachar Raindel * unrelated process. 20044e047f89SShachar Raindel */ 20054e047f89SShachar Raindel if (page) 20064e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 20074e047f89SShachar Raindel 20084e047f89SShachar Raindel flush_cache_page(vma, address, pte_pfn(orig_pte)); 20094e047f89SShachar Raindel entry = pte_mkyoung(orig_pte); 20104e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 20114e047f89SShachar Raindel if (ptep_set_access_flags(vma, address, page_table, entry, 1)) 20124e047f89SShachar Raindel update_mmu_cache(vma, address, page_table); 20134e047f89SShachar Raindel pte_unmap_unlock(page_table, ptl); 20144e047f89SShachar Raindel 20154e047f89SShachar Raindel if (dirty_shared) { 20164e047f89SShachar Raindel struct address_space *mapping; 20174e047f89SShachar Raindel int dirtied; 20184e047f89SShachar Raindel 20194e047f89SShachar Raindel if (!page_mkwrite) 20204e047f89SShachar Raindel lock_page(page); 20214e047f89SShachar Raindel 20224e047f89SShachar Raindel dirtied = set_page_dirty(page); 20234e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 20244e047f89SShachar Raindel mapping = page->mapping; 20254e047f89SShachar Raindel unlock_page(page); 20264e047f89SShachar Raindel page_cache_release(page); 20274e047f89SShachar Raindel 20284e047f89SShachar Raindel if ((dirtied || page_mkwrite) && mapping) { 20294e047f89SShachar Raindel /* 20304e047f89SShachar Raindel * Some device drivers do not set page.mapping 20314e047f89SShachar Raindel * but still dirty their pages 20324e047f89SShachar Raindel */ 20334e047f89SShachar Raindel balance_dirty_pages_ratelimited(mapping); 20344e047f89SShachar Raindel } 20354e047f89SShachar Raindel 20364e047f89SShachar Raindel if (!page_mkwrite) 20374e047f89SShachar Raindel file_update_time(vma->vm_file); 20384e047f89SShachar Raindel } 20394e047f89SShachar Raindel 20404e047f89SShachar Raindel return VM_FAULT_WRITE; 20414e047f89SShachar Raindel } 20424e047f89SShachar Raindel 20434e047f89SShachar Raindel /* 20442f38ab2cSShachar Raindel * Handle the case of a page which we actually need to copy to a new page. 20452f38ab2cSShachar Raindel * 20462f38ab2cSShachar Raindel * Called with mmap_sem locked and the old page referenced, but 20472f38ab2cSShachar Raindel * without the ptl held. 20482f38ab2cSShachar Raindel * 20492f38ab2cSShachar Raindel * High level logic flow: 20502f38ab2cSShachar Raindel * 20512f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 20522f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 20532f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 20542f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 20552f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 20562f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 20572f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 20582f38ab2cSShachar Raindel */ 20592f38ab2cSShachar Raindel static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma, 20602f38ab2cSShachar Raindel unsigned long address, pte_t *page_table, pmd_t *pmd, 20612f38ab2cSShachar Raindel pte_t orig_pte, struct page *old_page) 20622f38ab2cSShachar Raindel { 20632f38ab2cSShachar Raindel struct page *new_page = NULL; 20642f38ab2cSShachar Raindel spinlock_t *ptl = NULL; 20652f38ab2cSShachar Raindel pte_t entry; 20662f38ab2cSShachar Raindel int page_copied = 0; 20672f38ab2cSShachar Raindel const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */ 20682f38ab2cSShachar Raindel const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */ 20692f38ab2cSShachar Raindel struct mem_cgroup *memcg; 20702f38ab2cSShachar Raindel 20712f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 20722f38ab2cSShachar Raindel goto oom; 20732f38ab2cSShachar Raindel 20742f38ab2cSShachar Raindel if (is_zero_pfn(pte_pfn(orig_pte))) { 20752f38ab2cSShachar Raindel new_page = alloc_zeroed_user_highpage_movable(vma, address); 20762f38ab2cSShachar Raindel if (!new_page) 20772f38ab2cSShachar Raindel goto oom; 20782f38ab2cSShachar Raindel } else { 20792f38ab2cSShachar Raindel new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 20802f38ab2cSShachar Raindel if (!new_page) 20812f38ab2cSShachar Raindel goto oom; 20822f38ab2cSShachar Raindel cow_user_page(new_page, old_page, address, vma); 20832f38ab2cSShachar Raindel } 20842f38ab2cSShachar Raindel __SetPageUptodate(new_page); 20852f38ab2cSShachar Raindel 20862f38ab2cSShachar Raindel if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) 20872f38ab2cSShachar Raindel goto oom_free_new; 20882f38ab2cSShachar Raindel 20892f38ab2cSShachar Raindel mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 20902f38ab2cSShachar Raindel 20912f38ab2cSShachar Raindel /* 20922f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 20932f38ab2cSShachar Raindel */ 20942f38ab2cSShachar Raindel page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 20952f38ab2cSShachar Raindel if (likely(pte_same(*page_table, orig_pte))) { 20962f38ab2cSShachar Raindel if (old_page) { 20972f38ab2cSShachar Raindel if (!PageAnon(old_page)) { 20982f38ab2cSShachar Raindel dec_mm_counter_fast(mm, MM_FILEPAGES); 20992f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 21002f38ab2cSShachar Raindel } 21012f38ab2cSShachar Raindel } else { 21022f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 21032f38ab2cSShachar Raindel } 21042f38ab2cSShachar Raindel flush_cache_page(vma, address, pte_pfn(orig_pte)); 21052f38ab2cSShachar Raindel entry = mk_pte(new_page, vma->vm_page_prot); 21062f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 21072f38ab2cSShachar Raindel /* 21082f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 21092f38ab2cSShachar Raindel * pte with the new entry. This will avoid a race condition 21102f38ab2cSShachar Raindel * seen in the presence of one thread doing SMC and another 21112f38ab2cSShachar Raindel * thread doing COW. 21122f38ab2cSShachar Raindel */ 21132f38ab2cSShachar Raindel ptep_clear_flush_notify(vma, address, page_table); 21142f38ab2cSShachar Raindel page_add_new_anon_rmap(new_page, vma, address); 21152f38ab2cSShachar Raindel mem_cgroup_commit_charge(new_page, memcg, false); 21162f38ab2cSShachar Raindel lru_cache_add_active_or_unevictable(new_page, vma); 21172f38ab2cSShachar Raindel /* 21182f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 21192f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 21202f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 21212f38ab2cSShachar Raindel */ 21222f38ab2cSShachar Raindel set_pte_at_notify(mm, address, page_table, entry); 21232f38ab2cSShachar Raindel update_mmu_cache(vma, address, page_table); 21242f38ab2cSShachar Raindel if (old_page) { 21252f38ab2cSShachar Raindel /* 21262f38ab2cSShachar Raindel * Only after switching the pte to the new page may 21272f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 21282f38ab2cSShachar Raindel * process may come and find the rmap count decremented 21292f38ab2cSShachar Raindel * before the pte is switched to the new page, and 21302f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 21312f38ab2cSShachar Raindel * here still points into it and can be read by other 21322f38ab2cSShachar Raindel * threads. 21332f38ab2cSShachar Raindel * 21342f38ab2cSShachar Raindel * The critical issue is to order this 21352f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 21362f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 21372f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 21382f38ab2cSShachar Raindel * in page_remove_rmap. 21392f38ab2cSShachar Raindel * 21402f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 21412f38ab2cSShachar Raindel * no process can access the old page before the 21422f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 21432f38ab2cSShachar Raindel * cannot be reused until after the decremented 21442f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 21452f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 21462f38ab2cSShachar Raindel */ 21472f38ab2cSShachar Raindel page_remove_rmap(old_page); 21482f38ab2cSShachar Raindel } 21492f38ab2cSShachar Raindel 21502f38ab2cSShachar Raindel /* Free the old page.. */ 21512f38ab2cSShachar Raindel new_page = old_page; 21522f38ab2cSShachar Raindel page_copied = 1; 21532f38ab2cSShachar Raindel } else { 21542f38ab2cSShachar Raindel mem_cgroup_cancel_charge(new_page, memcg); 21552f38ab2cSShachar Raindel } 21562f38ab2cSShachar Raindel 21572f38ab2cSShachar Raindel if (new_page) 21582f38ab2cSShachar Raindel page_cache_release(new_page); 21592f38ab2cSShachar Raindel 21602f38ab2cSShachar Raindel pte_unmap_unlock(page_table, ptl); 21612f38ab2cSShachar Raindel mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 21622f38ab2cSShachar Raindel if (old_page) { 21632f38ab2cSShachar Raindel /* 21642f38ab2cSShachar Raindel * Don't let another task, with possibly unlocked vma, 21652f38ab2cSShachar Raindel * keep the mlocked page. 21662f38ab2cSShachar Raindel */ 21672f38ab2cSShachar Raindel if (page_copied && (vma->vm_flags & VM_LOCKED)) { 21682f38ab2cSShachar Raindel lock_page(old_page); /* LRU manipulation */ 21692f38ab2cSShachar Raindel munlock_vma_page(old_page); 21702f38ab2cSShachar Raindel unlock_page(old_page); 21712f38ab2cSShachar Raindel } 21722f38ab2cSShachar Raindel page_cache_release(old_page); 21732f38ab2cSShachar Raindel } 21742f38ab2cSShachar Raindel return page_copied ? VM_FAULT_WRITE : 0; 21752f38ab2cSShachar Raindel oom_free_new: 21762f38ab2cSShachar Raindel page_cache_release(new_page); 21772f38ab2cSShachar Raindel oom: 21782f38ab2cSShachar Raindel if (old_page) 21792f38ab2cSShachar Raindel page_cache_release(old_page); 21802f38ab2cSShachar Raindel return VM_FAULT_OOM; 21812f38ab2cSShachar Raindel } 21822f38ab2cSShachar Raindel 2183dd906184SBoaz Harrosh /* 2184dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 2185dd906184SBoaz Harrosh * mapping 2186dd906184SBoaz Harrosh */ 2187dd906184SBoaz Harrosh static int wp_pfn_shared(struct mm_struct *mm, 2188dd906184SBoaz Harrosh struct vm_area_struct *vma, unsigned long address, 2189dd906184SBoaz Harrosh pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, 2190dd906184SBoaz Harrosh pmd_t *pmd) 2191dd906184SBoaz Harrosh { 2192dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 2193dd906184SBoaz Harrosh struct vm_fault vmf = { 2194dd906184SBoaz Harrosh .page = NULL, 2195dd906184SBoaz Harrosh .pgoff = linear_page_index(vma, address), 2196dd906184SBoaz Harrosh .virtual_address = (void __user *)(address & PAGE_MASK), 2197dd906184SBoaz Harrosh .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE, 2198dd906184SBoaz Harrosh }; 2199dd906184SBoaz Harrosh int ret; 2200dd906184SBoaz Harrosh 2201dd906184SBoaz Harrosh pte_unmap_unlock(page_table, ptl); 2202dd906184SBoaz Harrosh ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); 2203dd906184SBoaz Harrosh if (ret & VM_FAULT_ERROR) 2204dd906184SBoaz Harrosh return ret; 2205dd906184SBoaz Harrosh page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2206dd906184SBoaz Harrosh /* 2207dd906184SBoaz Harrosh * We might have raced with another page fault while we 2208dd906184SBoaz Harrosh * released the pte_offset_map_lock. 2209dd906184SBoaz Harrosh */ 2210dd906184SBoaz Harrosh if (!pte_same(*page_table, orig_pte)) { 2211dd906184SBoaz Harrosh pte_unmap_unlock(page_table, ptl); 2212dd906184SBoaz Harrosh return 0; 2213dd906184SBoaz Harrosh } 2214dd906184SBoaz Harrosh } 2215dd906184SBoaz Harrosh return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte, 2216dd906184SBoaz Harrosh NULL, 0, 0); 2217dd906184SBoaz Harrosh } 2218dd906184SBoaz Harrosh 221993e478d4SShachar Raindel static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma, 222093e478d4SShachar Raindel unsigned long address, pte_t *page_table, 222193e478d4SShachar Raindel pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte, 222293e478d4SShachar Raindel struct page *old_page) 222393e478d4SShachar Raindel __releases(ptl) 222493e478d4SShachar Raindel { 222593e478d4SShachar Raindel int page_mkwrite = 0; 222693e478d4SShachar Raindel 222793e478d4SShachar Raindel page_cache_get(old_page); 222893e478d4SShachar Raindel 222993e478d4SShachar Raindel /* 223093e478d4SShachar Raindel * Only catch write-faults on shared writable pages, 223193e478d4SShachar Raindel * read-only shared pages can get COWed by 223293e478d4SShachar Raindel * get_user_pages(.write=1, .force=1). 223393e478d4SShachar Raindel */ 223493e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 223593e478d4SShachar Raindel int tmp; 223693e478d4SShachar Raindel 223793e478d4SShachar Raindel pte_unmap_unlock(page_table, ptl); 223893e478d4SShachar Raindel tmp = do_page_mkwrite(vma, old_page, address); 223993e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 224093e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 224193e478d4SShachar Raindel page_cache_release(old_page); 224293e478d4SShachar Raindel return tmp; 224393e478d4SShachar Raindel } 224493e478d4SShachar Raindel /* 224593e478d4SShachar Raindel * Since we dropped the lock we need to revalidate 224693e478d4SShachar Raindel * the PTE as someone else may have changed it. If 224793e478d4SShachar Raindel * they did, we just return, as we can count on the 224893e478d4SShachar Raindel * MMU to tell us if they didn't also make it writable. 224993e478d4SShachar Raindel */ 225093e478d4SShachar Raindel page_table = pte_offset_map_lock(mm, pmd, address, 225193e478d4SShachar Raindel &ptl); 225293e478d4SShachar Raindel if (!pte_same(*page_table, orig_pte)) { 225393e478d4SShachar Raindel unlock_page(old_page); 225493e478d4SShachar Raindel pte_unmap_unlock(page_table, ptl); 225593e478d4SShachar Raindel page_cache_release(old_page); 225693e478d4SShachar Raindel return 0; 225793e478d4SShachar Raindel } 225893e478d4SShachar Raindel page_mkwrite = 1; 225993e478d4SShachar Raindel } 226093e478d4SShachar Raindel 226193e478d4SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 226293e478d4SShachar Raindel orig_pte, old_page, page_mkwrite, 1); 226393e478d4SShachar Raindel } 226493e478d4SShachar Raindel 22652f38ab2cSShachar Raindel /* 22661da177e4SLinus Torvalds * This routine handles present pages, when users try to write 22671da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 22681da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 22691da177e4SLinus Torvalds * 22701da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 22711da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 22721da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 22731da177e4SLinus Torvalds * COW. 22741da177e4SLinus Torvalds * 22751da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 22761da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 22771da177e4SLinus Torvalds * and potentially makes it more efficient. 22781da177e4SLinus Torvalds * 22798f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 22808f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 22818f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 22821da177e4SLinus Torvalds */ 22831da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 228465500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 22858f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2286e6219ec8SNamhyung Kim __releases(ptl) 22871da177e4SLinus Torvalds { 22882f38ab2cSShachar Raindel struct page *old_page; 22891da177e4SLinus Torvalds 22906aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2291251b97f5SPeter Zijlstra if (!old_page) { 2292251b97f5SPeter Zijlstra /* 229364e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 229464e45507SPeter Feiner * VM_PFNMAP VMA. 2295251b97f5SPeter Zijlstra * 2296251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2297dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 2298251b97f5SPeter Zijlstra */ 2299251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2300251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2301dd906184SBoaz Harrosh return wp_pfn_shared(mm, vma, address, page_table, ptl, 2302dd906184SBoaz Harrosh orig_pte, pmd); 23032f38ab2cSShachar Raindel 23042f38ab2cSShachar Raindel pte_unmap_unlock(page_table, ptl); 23052f38ab2cSShachar Raindel return wp_page_copy(mm, vma, address, page_table, pmd, 23062f38ab2cSShachar Raindel orig_pte, old_page); 2307251b97f5SPeter Zijlstra } 23081da177e4SLinus Torvalds 2309d08b3851SPeter Zijlstra /* 2310ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2311ee6a6457SPeter Zijlstra * not dirty accountable. 2312d08b3851SPeter Zijlstra */ 23139a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2314ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2315ab967d86SHugh Dickins page_cache_get(old_page); 2316ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2317ab967d86SHugh Dickins lock_page(old_page); 2318ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2319ab967d86SHugh Dickins &ptl); 2320ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2321ab967d86SHugh Dickins unlock_page(old_page); 232228766805SShachar Raindel pte_unmap_unlock(page_table, ptl); 232328766805SShachar Raindel page_cache_release(old_page); 232428766805SShachar Raindel return 0; 2325ab967d86SHugh Dickins } 2326ab967d86SHugh Dickins page_cache_release(old_page); 2327ab967d86SHugh Dickins } 2328b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2329c44b6743SRik van Riel /* 2330c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2331c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2332c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2333c44b6743SRik van Riel */ 2334c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2335ee6a6457SPeter Zijlstra unlock_page(old_page); 23364e047f89SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 23374e047f89SShachar Raindel orig_pte, old_page, 0, 0); 2338b009c024SMichel Lespinasse } 2339b009c024SMichel Lespinasse unlock_page(old_page); 2340ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2341d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 234293e478d4SShachar Raindel return wp_page_shared(mm, vma, address, page_table, pmd, 234393e478d4SShachar Raindel ptl, orig_pte, old_page); 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds /* 23471da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 23481da177e4SLinus Torvalds */ 23491da177e4SLinus Torvalds page_cache_get(old_page); 235028766805SShachar Raindel 23518f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 23522f38ab2cSShachar Raindel return wp_page_copy(mm, vma, address, page_table, pmd, 23532f38ab2cSShachar Raindel orig_pte, old_page); 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 235697a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 23571da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 23581da177e4SLinus Torvalds struct zap_details *details) 23591da177e4SLinus Torvalds { 2360f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 23611da177e4SLinus Torvalds } 23621da177e4SLinus Torvalds 23636b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 23641da177e4SLinus Torvalds struct zap_details *details) 23651da177e4SLinus Torvalds { 23661da177e4SLinus Torvalds struct vm_area_struct *vma; 23671da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 23681da177e4SLinus Torvalds 23696b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 23701da177e4SLinus Torvalds details->first_index, details->last_index) { 23711da177e4SLinus Torvalds 23721da177e4SLinus Torvalds vba = vma->vm_pgoff; 2373d6e93217SLibin vea = vba + vma_pages(vma) - 1; 23741da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 23751da177e4SLinus Torvalds zba = details->first_index; 23761da177e4SLinus Torvalds if (zba < vba) 23771da177e4SLinus Torvalds zba = vba; 23781da177e4SLinus Torvalds zea = details->last_index; 23791da177e4SLinus Torvalds if (zea > vea) 23801da177e4SLinus Torvalds zea = vea; 23811da177e4SLinus Torvalds 238297a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 23831da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 23841da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 238597a89413SPeter Zijlstra details); 23861da177e4SLinus Torvalds } 23871da177e4SLinus Torvalds } 23881da177e4SLinus Torvalds 23891da177e4SLinus Torvalds /** 23908a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 23918a5f14a2SKirill A. Shutemov * address_space corresponding to the specified page range in the underlying 23928a5f14a2SKirill A. Shutemov * file. 23938a5f14a2SKirill A. Shutemov * 23943d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 23951da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 23961da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 239725d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 23981da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 23991da177e4SLinus Torvalds * partial pages. 24001da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 24011da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 24021da177e4SLinus Torvalds * end of the file. 24031da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 24041da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 24051da177e4SLinus Torvalds */ 24061da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 24071da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 24081da177e4SLinus Torvalds { 24091da177e4SLinus Torvalds struct zap_details details; 24101da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 24111da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds /* Check for overflow. */ 24141da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 24151da177e4SLinus Torvalds long long holeend = 24161da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 24171da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 24181da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 24191da177e4SLinus Torvalds } 24201da177e4SLinus Torvalds 24211da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 24221da177e4SLinus Torvalds details.first_index = hba; 24231da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 24241da177e4SLinus Torvalds if (details.last_index < details.first_index) 24251da177e4SLinus Torvalds details.last_index = ULONG_MAX; 24261da177e4SLinus Torvalds 242797a89413SPeter Zijlstra 2428283307c7SMatthew Wilcox /* DAX uses i_mmap_lock to serialise file truncate vs page fault */ 242948ec833bSKirill A. Shutemov i_mmap_lock_write(mapping); 24306b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 24311da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 243248ec833bSKirill A. Shutemov i_mmap_unlock_write(mapping); 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 24351da177e4SLinus Torvalds 24361da177e4SLinus Torvalds /* 24378f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 24388f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 24399a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 24409a95f3cfSPaul Cassella * 24419a95f3cfSPaul Cassella * We return with the mmap_sem locked or unlocked in the same cases 24429a95f3cfSPaul Cassella * as does filemap_fault(). 24431da177e4SLinus Torvalds */ 244465500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 244565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 244630c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 24471da177e4SLinus Torvalds { 24488f4e2101SHugh Dickins spinlock_t *ptl; 244956f31801SHugh Dickins struct page *page, *swapcache; 245000501b53SJohannes Weiner struct mem_cgroup *memcg; 245165500d23SHugh Dickins swp_entry_t entry; 24521da177e4SLinus Torvalds pte_t pte; 2453d065bd81SMichel Lespinasse int locked; 2454ad8c2ee8SRik van Riel int exclusive = 0; 245583c54070SNick Piggin int ret = 0; 24561da177e4SLinus Torvalds 24574c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 24588f4e2101SHugh Dickins goto out; 245965500d23SHugh Dickins 246065500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 2461d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 24620697212aSChristoph Lameter if (is_migration_entry(entry)) { 24630697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 2464d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 2465d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2466d1737fdbSAndi Kleen } else { 2467d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 2468d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 2469d1737fdbSAndi Kleen } 24700697212aSChristoph Lameter goto out; 24710697212aSChristoph Lameter } 24720ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 24731da177e4SLinus Torvalds page = lookup_swap_cache(entry); 24741da177e4SLinus Torvalds if (!page) { 247502098feaSHugh Dickins page = swapin_readahead(entry, 247602098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 24771da177e4SLinus Torvalds if (!page) { 24781da177e4SLinus Torvalds /* 24798f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 24808f4e2101SHugh Dickins * while we released the pte lock. 24811da177e4SLinus Torvalds */ 24828f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 24831da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 24841da177e4SLinus Torvalds ret = VM_FAULT_OOM; 24850ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 248665500d23SHugh Dickins goto unlock; 24871da177e4SLinus Torvalds } 24881da177e4SLinus Torvalds 24891da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 24901da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2491f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 2492456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 2493d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 249471f72525SWu Fengguang /* 249571f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 249671f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 249771f72525SWu Fengguang */ 2498d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2499d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 250056f31801SHugh Dickins swapcache = page; 25014779cb31SAndi Kleen goto out_release; 25021da177e4SLinus Torvalds } 25031da177e4SLinus Torvalds 250456f31801SHugh Dickins swapcache = page; 2505d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 2506e709ffd6SRik van Riel 250720a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 2508d065bd81SMichel Lespinasse if (!locked) { 2509d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 2510d065bd81SMichel Lespinasse goto out_release; 2511d065bd81SMichel Lespinasse } 25121da177e4SLinus Torvalds 25134969c119SAndrea Arcangeli /* 251431c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 251531c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 251631c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 251731c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 25184969c119SAndrea Arcangeli */ 251931c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 25204969c119SAndrea Arcangeli goto out_page; 25214969c119SAndrea Arcangeli 2522cbf86cfeSHugh Dickins page = ksm_might_need_to_copy(page, vma, address); 25234969c119SAndrea Arcangeli if (unlikely(!page)) { 25245ad64688SHugh Dickins ret = VM_FAULT_OOM; 25254969c119SAndrea Arcangeli page = swapcache; 25264969c119SAndrea Arcangeli goto out_page; 25274969c119SAndrea Arcangeli } 25285ad64688SHugh Dickins 252900501b53SJohannes Weiner if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) { 2530073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 2531bc43f75cSJohannes Weiner goto out_page; 2532073e587eSKAMEZAWA Hiroyuki } 2533073e587eSKAMEZAWA Hiroyuki 25341da177e4SLinus Torvalds /* 25358f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 25361da177e4SLinus Torvalds */ 25378f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 25389e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 2539b8107480SKirill Korotaev goto out_nomap; 2540b8107480SKirill Korotaev 2541b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 2542b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 2543b8107480SKirill Korotaev goto out_nomap; 25441da177e4SLinus Torvalds } 25451da177e4SLinus Torvalds 25468c7c6e34SKAMEZAWA Hiroyuki /* 25478c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 25488c7c6e34SKAMEZAWA Hiroyuki * 25498c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 25508c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 25518c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 25528c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 25538c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 25548c7c6e34SKAMEZAWA Hiroyuki */ 25551da177e4SLinus Torvalds 255634e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2557b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 25581da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 255930c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 25601da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 256130c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 25629a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 2563ad8c2ee8SRik van Riel exclusive = 1; 25641da177e4SLinus Torvalds } 25651da177e4SLinus Torvalds flush_icache_page(vma, page); 2566179ef71cSCyrill Gorcunov if (pte_swp_soft_dirty(orig_pte)) 2567179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 25681da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 256900501b53SJohannes Weiner if (page == swapcache) { 2570ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 257100501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, true); 257200501b53SJohannes Weiner } else { /* ksm created a completely new copy */ 257356f31801SHugh Dickins page_add_new_anon_rmap(page, vma, address); 257400501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, false); 257500501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 257600501b53SJohannes Weiner } 25771da177e4SLinus Torvalds 2578c475a8abSHugh Dickins swap_free(entry); 2579b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 2580a2c43eedSHugh Dickins try_to_free_swap(page); 2581c475a8abSHugh Dickins unlock_page(page); 258256f31801SHugh Dickins if (page != swapcache) { 25834969c119SAndrea Arcangeli /* 25844969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 25854969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 25864969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 25874969c119SAndrea Arcangeli * further safety release the lock after the swap_free 25884969c119SAndrea Arcangeli * so that the swap count won't change under a 25894969c119SAndrea Arcangeli * parallel locked swapcache. 25904969c119SAndrea Arcangeli */ 25914969c119SAndrea Arcangeli unlock_page(swapcache); 25924969c119SAndrea Arcangeli page_cache_release(swapcache); 25934969c119SAndrea Arcangeli } 2594c475a8abSHugh Dickins 259530c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 259661469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 259761469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 259861469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 25991da177e4SLinus Torvalds goto out; 26001da177e4SLinus Torvalds } 26011da177e4SLinus Torvalds 26021da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 26034b3073e1SRussell King update_mmu_cache(vma, address, page_table); 260465500d23SHugh Dickins unlock: 26058f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 26061da177e4SLinus Torvalds out: 26071da177e4SLinus Torvalds return ret; 2608b8107480SKirill Korotaev out_nomap: 260900501b53SJohannes Weiner mem_cgroup_cancel_charge(page, memcg); 26108f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2611bc43f75cSJohannes Weiner out_page: 2612b8107480SKirill Korotaev unlock_page(page); 26134779cb31SAndi Kleen out_release: 2614b8107480SKirill Korotaev page_cache_release(page); 261556f31801SHugh Dickins if (page != swapcache) { 26164969c119SAndrea Arcangeli unlock_page(swapcache); 26174969c119SAndrea Arcangeli page_cache_release(swapcache); 26184969c119SAndrea Arcangeli } 261965500d23SHugh Dickins return ret; 26201da177e4SLinus Torvalds } 26211da177e4SLinus Torvalds 26221da177e4SLinus Torvalds /* 26238ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 26248ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 2625320b2b8dSLinus Torvalds * doesn't hit another vma. 2626320b2b8dSLinus Torvalds */ 2627320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 2628320b2b8dSLinus Torvalds { 2629320b2b8dSLinus Torvalds address &= PAGE_MASK; 2630320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 26310e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 2632320b2b8dSLinus Torvalds 26330e8e50e2SLinus Torvalds /* 26340e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 26350e8e50e2SLinus Torvalds * 26360e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 26370e8e50e2SLinus Torvalds * that has gotten split.. 26380e8e50e2SLinus Torvalds */ 26390e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 26400e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 26410e8e50e2SLinus Torvalds 2642fee7e49dSLinus Torvalds return expand_downwards(vma, address - PAGE_SIZE); 2643320b2b8dSLinus Torvalds } 26448ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 26458ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 26468ca3eb08SLuck, Tony 26478ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 26488ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 26498ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 26508ca3eb08SLuck, Tony 2651fee7e49dSLinus Torvalds return expand_upwards(vma, address + PAGE_SIZE); 26528ca3eb08SLuck, Tony } 2653320b2b8dSLinus Torvalds return 0; 2654320b2b8dSLinus Torvalds } 2655320b2b8dSLinus Torvalds 2656320b2b8dSLinus Torvalds /* 26578f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 26588f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 26598f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 26601da177e4SLinus Torvalds */ 266165500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 266265500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 266330c9f3a9SLinus Torvalds unsigned int flags) 26641da177e4SLinus Torvalds { 266500501b53SJohannes Weiner struct mem_cgroup *memcg; 26668f4e2101SHugh Dickins struct page *page; 26678f4e2101SHugh Dickins spinlock_t *ptl; 26681da177e4SLinus Torvalds pte_t entry; 26691da177e4SLinus Torvalds 26705528f913SLinus Torvalds pte_unmap(page_table); 2671320b2b8dSLinus Torvalds 267211ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 267311ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 26749c145c56SLinus Torvalds return VM_FAULT_SIGSEGV; 267511ac5524SLinus Torvalds 267611ac5524SLinus Torvalds /* Use the zero-page for reads */ 2677593befa6SDominik Dingel if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) { 267862eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 267962eede62SHugh Dickins vma->vm_page_prot)); 268011ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2681a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 2682a13ea5b7SHugh Dickins goto unlock; 2683a13ea5b7SHugh Dickins goto setpte; 2684a13ea5b7SHugh Dickins } 2685a13ea5b7SHugh Dickins 26861da177e4SLinus Torvalds /* Allocate our own private page. */ 26871da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 268865500d23SHugh Dickins goto oom; 2689769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 26901da177e4SLinus Torvalds if (!page) 269165500d23SHugh Dickins goto oom; 269252f37629SMinchan Kim /* 269352f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 269452f37629SMinchan Kim * preceeding stores to the page contents become visible before 269552f37629SMinchan Kim * the set_pte_at() write. 269652f37629SMinchan Kim */ 26970ed361deSNick Piggin __SetPageUptodate(page); 26981da177e4SLinus Torvalds 269900501b53SJohannes Weiner if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) 27008a9f3ccdSBalbir Singh goto oom_free_page; 27018a9f3ccdSBalbir Singh 270265500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 27031ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 27041ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 27058f4e2101SHugh Dickins 27068f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 27071c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 27088f4e2101SHugh Dickins goto release; 27099ba69294SHugh Dickins 271034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 27119617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 271200501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, false); 271300501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 2714a13ea5b7SHugh Dickins setpte: 271565500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 27161da177e4SLinus Torvalds 27171da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 27184b3073e1SRussell King update_mmu_cache(vma, address, page_table); 271965500d23SHugh Dickins unlock: 27208f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 272183c54070SNick Piggin return 0; 27228f4e2101SHugh Dickins release: 272300501b53SJohannes Weiner mem_cgroup_cancel_charge(page, memcg); 27248f4e2101SHugh Dickins page_cache_release(page); 27258f4e2101SHugh Dickins goto unlock; 27268a9f3ccdSBalbir Singh oom_free_page: 27276dbf6d3bSHugh Dickins page_cache_release(page); 272865500d23SHugh Dickins oom: 27291da177e4SLinus Torvalds return VM_FAULT_OOM; 27301da177e4SLinus Torvalds } 27311da177e4SLinus Torvalds 27329a95f3cfSPaul Cassella /* 27339a95f3cfSPaul Cassella * The mmap_sem must have been held on entry, and may have been 27349a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 27359a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 27369a95f3cfSPaul Cassella */ 27377eae74afSKirill A. Shutemov static int __do_fault(struct vm_area_struct *vma, unsigned long address, 27382e4cdab0SMatthew Wilcox pgoff_t pgoff, unsigned int flags, 27392e4cdab0SMatthew Wilcox struct page *cow_page, struct page **page) 27407eae74afSKirill A. Shutemov { 27417eae74afSKirill A. Shutemov struct vm_fault vmf; 27427eae74afSKirill A. Shutemov int ret; 27437eae74afSKirill A. Shutemov 27447eae74afSKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 27457eae74afSKirill A. Shutemov vmf.pgoff = pgoff; 27467eae74afSKirill A. Shutemov vmf.flags = flags; 27477eae74afSKirill A. Shutemov vmf.page = NULL; 27482e4cdab0SMatthew Wilcox vmf.cow_page = cow_page; 27497eae74afSKirill A. Shutemov 27507eae74afSKirill A. Shutemov ret = vma->vm_ops->fault(vma, &vmf); 27517eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 27527eae74afSKirill A. Shutemov return ret; 27532e4cdab0SMatthew Wilcox if (!vmf.page) 27542e4cdab0SMatthew Wilcox goto out; 27557eae74afSKirill A. Shutemov 27567eae74afSKirill A. Shutemov if (unlikely(PageHWPoison(vmf.page))) { 27577eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 27587eae74afSKirill A. Shutemov unlock_page(vmf.page); 27597eae74afSKirill A. Shutemov page_cache_release(vmf.page); 27607eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 27617eae74afSKirill A. Shutemov } 27627eae74afSKirill A. Shutemov 27637eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 27647eae74afSKirill A. Shutemov lock_page(vmf.page); 27657eae74afSKirill A. Shutemov else 27667eae74afSKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); 27677eae74afSKirill A. Shutemov 27682e4cdab0SMatthew Wilcox out: 27697eae74afSKirill A. Shutemov *page = vmf.page; 27707eae74afSKirill A. Shutemov return ret; 27717eae74afSKirill A. Shutemov } 27727eae74afSKirill A. Shutemov 27738c6e50b0SKirill A. Shutemov /** 27748c6e50b0SKirill A. Shutemov * do_set_pte - setup new PTE entry for given page and add reverse page mapping. 27758c6e50b0SKirill A. Shutemov * 27768c6e50b0SKirill A. Shutemov * @vma: virtual memory area 27778c6e50b0SKirill A. Shutemov * @address: user virtual address 27788c6e50b0SKirill A. Shutemov * @page: page to map 27798c6e50b0SKirill A. Shutemov * @pte: pointer to target page table entry 27808c6e50b0SKirill A. Shutemov * @write: true, if new entry is writable 27818c6e50b0SKirill A. Shutemov * @anon: true, if it's anonymous page 27828c6e50b0SKirill A. Shutemov * 27838c6e50b0SKirill A. Shutemov * Caller must hold page table lock relevant for @pte. 27848c6e50b0SKirill A. Shutemov * 27858c6e50b0SKirill A. Shutemov * Target users are page handler itself and implementations of 27868c6e50b0SKirill A. Shutemov * vm_ops->map_pages. 27878c6e50b0SKirill A. Shutemov */ 27888c6e50b0SKirill A. Shutemov void do_set_pte(struct vm_area_struct *vma, unsigned long address, 27893bb97794SKirill A. Shutemov struct page *page, pte_t *pte, bool write, bool anon) 27903bb97794SKirill A. Shutemov { 27913bb97794SKirill A. Shutemov pte_t entry; 27923bb97794SKirill A. Shutemov 27933bb97794SKirill A. Shutemov flush_icache_page(vma, page); 27943bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 27953bb97794SKirill A. Shutemov if (write) 27963bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 27973bb97794SKirill A. Shutemov if (anon) { 27983bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 27993bb97794SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address); 28003bb97794SKirill A. Shutemov } else { 28013bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); 28023bb97794SKirill A. Shutemov page_add_file_rmap(page); 28033bb97794SKirill A. Shutemov } 28043bb97794SKirill A. Shutemov set_pte_at(vma->vm_mm, address, pte, entry); 28053bb97794SKirill A. Shutemov 28063bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 28073bb97794SKirill A. Shutemov update_mmu_cache(vma, address, pte); 28083bb97794SKirill A. Shutemov } 28093bb97794SKirill A. Shutemov 28103a91053aSKirill A. Shutemov static unsigned long fault_around_bytes __read_mostly = 28113a91053aSKirill A. Shutemov rounddown_pow_of_two(65536); 2812a9b0f861SKirill A. Shutemov 28131592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 2814a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 28151592eef0SKirill A. Shutemov { 2816a9b0f861SKirill A. Shutemov *val = fault_around_bytes; 28171592eef0SKirill A. Shutemov return 0; 28181592eef0SKirill A. Shutemov } 28191592eef0SKirill A. Shutemov 2820b4903d6eSAndrey Ryabinin /* 2821b4903d6eSAndrey Ryabinin * fault_around_pages() and fault_around_mask() expects fault_around_bytes 2822b4903d6eSAndrey Ryabinin * rounded down to nearest page order. It's what do_fault_around() expects to 2823b4903d6eSAndrey Ryabinin * see. 2824b4903d6eSAndrey Ryabinin */ 2825a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 28261592eef0SKirill A. Shutemov { 2827a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 28281592eef0SKirill A. Shutemov return -EINVAL; 2829b4903d6eSAndrey Ryabinin if (val > PAGE_SIZE) 2830b4903d6eSAndrey Ryabinin fault_around_bytes = rounddown_pow_of_two(val); 2831b4903d6eSAndrey Ryabinin else 2832b4903d6eSAndrey Ryabinin fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ 28331592eef0SKirill A. Shutemov return 0; 28341592eef0SKirill A. Shutemov } 2835a9b0f861SKirill A. Shutemov DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, 2836a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 28371592eef0SKirill A. Shutemov 28381592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 28391592eef0SKirill A. Shutemov { 28401592eef0SKirill A. Shutemov void *ret; 28411592eef0SKirill A. Shutemov 2842a9b0f861SKirill A. Shutemov ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, 2843a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 28441592eef0SKirill A. Shutemov if (!ret) 2845a9b0f861SKirill A. Shutemov pr_warn("Failed to create fault_around_bytes in debugfs"); 28461592eef0SKirill A. Shutemov return 0; 28471592eef0SKirill A. Shutemov } 28481592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 28491592eef0SKirill A. Shutemov #endif 28508c6e50b0SKirill A. Shutemov 28511fdb412bSKirill A. Shutemov /* 28521fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 28531fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 28541fdb412bSKirill A. Shutemov * faults to handle. 28551fdb412bSKirill A. Shutemov * 28561fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 28571fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 28581fdb412bSKirill A. Shutemov * 28591fdb412bSKirill A. Shutemov * This function is called with the page table lock taken. In the split ptlock 28601fdb412bSKirill A. Shutemov * case the page table lock only protects only those entries which belong to 28611fdb412bSKirill A. Shutemov * the page table corresponding to the fault address. 28621fdb412bSKirill A. Shutemov * 28631fdb412bSKirill A. Shutemov * This function doesn't cross the VMA boundaries, in order to call map_pages() 28641fdb412bSKirill A. Shutemov * only once. 28651fdb412bSKirill A. Shutemov * 28661fdb412bSKirill A. Shutemov * fault_around_pages() defines how many pages we'll try to map. 28671fdb412bSKirill A. Shutemov * do_fault_around() expects it to return a power of two less than or equal to 28681fdb412bSKirill A. Shutemov * PTRS_PER_PTE. 28691fdb412bSKirill A. Shutemov * 28701fdb412bSKirill A. Shutemov * The virtual address of the area that we map is naturally aligned to the 28711fdb412bSKirill A. Shutemov * fault_around_pages() value (and therefore to page order). This way it's 28721fdb412bSKirill A. Shutemov * easier to guarantee that we don't cross page table boundaries. 28731fdb412bSKirill A. Shutemov */ 28748c6e50b0SKirill A. Shutemov static void do_fault_around(struct vm_area_struct *vma, unsigned long address, 28758c6e50b0SKirill A. Shutemov pte_t *pte, pgoff_t pgoff, unsigned int flags) 28768c6e50b0SKirill A. Shutemov { 2877aecd6f44SKirill A. Shutemov unsigned long start_addr, nr_pages, mask; 28788c6e50b0SKirill A. Shutemov pgoff_t max_pgoff; 28798c6e50b0SKirill A. Shutemov struct vm_fault vmf; 28808c6e50b0SKirill A. Shutemov int off; 28818c6e50b0SKirill A. Shutemov 28824db0c3c2SJason Low nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; 2883aecd6f44SKirill A. Shutemov mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; 2884aecd6f44SKirill A. Shutemov 2885aecd6f44SKirill A. Shutemov start_addr = max(address & mask, vma->vm_start); 28868c6e50b0SKirill A. Shutemov off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 28878c6e50b0SKirill A. Shutemov pte -= off; 28888c6e50b0SKirill A. Shutemov pgoff -= off; 28898c6e50b0SKirill A. Shutemov 28908c6e50b0SKirill A. Shutemov /* 28918c6e50b0SKirill A. Shutemov * max_pgoff is either end of page table or end of vma 2892850e9c69SKirill A. Shutemov * or fault_around_pages() from pgoff, depending what is nearest. 28938c6e50b0SKirill A. Shutemov */ 28948c6e50b0SKirill A. Shutemov max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + 28958c6e50b0SKirill A. Shutemov PTRS_PER_PTE - 1; 28968c6e50b0SKirill A. Shutemov max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, 2897aecd6f44SKirill A. Shutemov pgoff + nr_pages - 1); 28988c6e50b0SKirill A. Shutemov 28998c6e50b0SKirill A. Shutemov /* Check if it makes any sense to call ->map_pages */ 29008c6e50b0SKirill A. Shutemov while (!pte_none(*pte)) { 29018c6e50b0SKirill A. Shutemov if (++pgoff > max_pgoff) 29028c6e50b0SKirill A. Shutemov return; 29038c6e50b0SKirill A. Shutemov start_addr += PAGE_SIZE; 29048c6e50b0SKirill A. Shutemov if (start_addr >= vma->vm_end) 29058c6e50b0SKirill A. Shutemov return; 29068c6e50b0SKirill A. Shutemov pte++; 29078c6e50b0SKirill A. Shutemov } 29088c6e50b0SKirill A. Shutemov 29098c6e50b0SKirill A. Shutemov vmf.virtual_address = (void __user *) start_addr; 29108c6e50b0SKirill A. Shutemov vmf.pte = pte; 29118c6e50b0SKirill A. Shutemov vmf.pgoff = pgoff; 29128c6e50b0SKirill A. Shutemov vmf.max_pgoff = max_pgoff; 29138c6e50b0SKirill A. Shutemov vmf.flags = flags; 29148c6e50b0SKirill A. Shutemov vma->vm_ops->map_pages(vma, &vmf); 29158c6e50b0SKirill A. Shutemov } 29168c6e50b0SKirill A. Shutemov 2917e655fb29SKirill A. Shutemov static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2918e655fb29SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2919e655fb29SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2920e655fb29SKirill A. Shutemov { 2921e655fb29SKirill A. Shutemov struct page *fault_page; 2922e655fb29SKirill A. Shutemov spinlock_t *ptl; 29233bb97794SKirill A. Shutemov pte_t *pte; 29248c6e50b0SKirill A. Shutemov int ret = 0; 29258c6e50b0SKirill A. Shutemov 29268c6e50b0SKirill A. Shutemov /* 29278c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 29288c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 29298c6e50b0SKirill A. Shutemov * something). 29308c6e50b0SKirill A. Shutemov */ 29319b4bdd2fSKirill A. Shutemov if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { 29328c6e50b0SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 29338c6e50b0SKirill A. Shutemov do_fault_around(vma, address, pte, pgoff, flags); 29348c6e50b0SKirill A. Shutemov if (!pte_same(*pte, orig_pte)) 29358c6e50b0SKirill A. Shutemov goto unlock_out; 29368c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 29378c6e50b0SKirill A. Shutemov } 2938e655fb29SKirill A. Shutemov 29392e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 2940e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2941e655fb29SKirill A. Shutemov return ret; 2942e655fb29SKirill A. Shutemov 2943e655fb29SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2944e655fb29SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2945e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2946e655fb29SKirill A. Shutemov unlock_page(fault_page); 2947e655fb29SKirill A. Shutemov page_cache_release(fault_page); 2948e655fb29SKirill A. Shutemov return ret; 2949e655fb29SKirill A. Shutemov } 29503bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, false, false); 2951e655fb29SKirill A. Shutemov unlock_page(fault_page); 29528c6e50b0SKirill A. Shutemov unlock_out: 29538c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2954e655fb29SKirill A. Shutemov return ret; 2955e655fb29SKirill A. Shutemov } 2956e655fb29SKirill A. Shutemov 2957ec47c3b9SKirill A. Shutemov static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2958ec47c3b9SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2959ec47c3b9SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2960ec47c3b9SKirill A. Shutemov { 2961ec47c3b9SKirill A. Shutemov struct page *fault_page, *new_page; 296200501b53SJohannes Weiner struct mem_cgroup *memcg; 2963ec47c3b9SKirill A. Shutemov spinlock_t *ptl; 29643bb97794SKirill A. Shutemov pte_t *pte; 2965ec47c3b9SKirill A. Shutemov int ret; 2966ec47c3b9SKirill A. Shutemov 2967ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 2968ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2969ec47c3b9SKirill A. Shutemov 2970ec47c3b9SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 2971ec47c3b9SKirill A. Shutemov if (!new_page) 2972ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2973ec47c3b9SKirill A. Shutemov 297400501b53SJohannes Weiner if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) { 2975ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 2976ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2977ec47c3b9SKirill A. Shutemov } 2978ec47c3b9SKirill A. Shutemov 29792e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page); 2980ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2981ec47c3b9SKirill A. Shutemov goto uncharge_out; 2982ec47c3b9SKirill A. Shutemov 29832e4cdab0SMatthew Wilcox if (fault_page) 2984ec47c3b9SKirill A. Shutemov copy_user_highpage(new_page, fault_page, address, vma); 2985ec47c3b9SKirill A. Shutemov __SetPageUptodate(new_page); 2986ec47c3b9SKirill A. Shutemov 2987ec47c3b9SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2988ec47c3b9SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2989ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 29902e4cdab0SMatthew Wilcox if (fault_page) { 2991ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 2992ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 29932e4cdab0SMatthew Wilcox } else { 29942e4cdab0SMatthew Wilcox /* 29952e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 29962e4cdab0SMatthew Wilcox * i_mmap_lock for read to protect against truncate. 29972e4cdab0SMatthew Wilcox */ 29982e4cdab0SMatthew Wilcox i_mmap_unlock_read(vma->vm_file->f_mapping); 29992e4cdab0SMatthew Wilcox } 3000ec47c3b9SKirill A. Shutemov goto uncharge_out; 3001ec47c3b9SKirill A. Shutemov } 30023bb97794SKirill A. Shutemov do_set_pte(vma, address, new_page, pte, true, true); 300300501b53SJohannes Weiner mem_cgroup_commit_charge(new_page, memcg, false); 300400501b53SJohannes Weiner lru_cache_add_active_or_unevictable(new_page, vma); 3005ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 30062e4cdab0SMatthew Wilcox if (fault_page) { 3007ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3008ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 30092e4cdab0SMatthew Wilcox } else { 30102e4cdab0SMatthew Wilcox /* 30112e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 30122e4cdab0SMatthew Wilcox * i_mmap_lock for read to protect against truncate. 30132e4cdab0SMatthew Wilcox */ 30142e4cdab0SMatthew Wilcox i_mmap_unlock_read(vma->vm_file->f_mapping); 30152e4cdab0SMatthew Wilcox } 3016ec47c3b9SKirill A. Shutemov return ret; 3017ec47c3b9SKirill A. Shutemov uncharge_out: 301800501b53SJohannes Weiner mem_cgroup_cancel_charge(new_page, memcg); 3019ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3020ec47c3b9SKirill A. Shutemov return ret; 3021ec47c3b9SKirill A. Shutemov } 3022ec47c3b9SKirill A. Shutemov 3023f0c6d4d2SKirill A. Shutemov static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, 302416abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 302554cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 30261da177e4SLinus Torvalds { 3027f0c6d4d2SKirill A. Shutemov struct page *fault_page; 3028f0c6d4d2SKirill A. Shutemov struct address_space *mapping; 30298f4e2101SHugh Dickins spinlock_t *ptl; 30303bb97794SKirill A. Shutemov pte_t *pte; 3031f0c6d4d2SKirill A. Shutemov int dirtied = 0; 3032f0c6d4d2SKirill A. Shutemov int ret, tmp; 30331d65f86dSKAMEZAWA Hiroyuki 30342e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 30357eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3036f0c6d4d2SKirill A. Shutemov return ret; 30371da177e4SLinus Torvalds 30381da177e4SLinus Torvalds /* 3039f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 3040f0c6d4d2SKirill A. Shutemov * about to become writable 30411da177e4SLinus Torvalds */ 3042fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 3043f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3044fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, fault_page, address); 3045fb09a464SKirill A. Shutemov if (unlikely(!tmp || 3046fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3047f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3048f0c6d4d2SKirill A. Shutemov return tmp; 304969676147SMark Fasheh } 3050d0217ac0SNick Piggin } 3051fb09a464SKirill A. Shutemov 3052f0c6d4d2SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3053f0c6d4d2SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3054f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3055f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3056f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3057f0c6d4d2SKirill A. Shutemov return ret; 30589637a5efSDavid Howells } 30593bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, true, false); 3060f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3061d00806b1SNick Piggin 3062f0c6d4d2SKirill A. Shutemov if (set_page_dirty(fault_page)) 306341c4d25fSJan Kara dirtied = 1; 3064d82fa87dSAndrew Morton /* 3065d82fa87dSAndrew Morton * Take a local copy of the address_space - page.mapping may be zeroed 3066d82fa87dSAndrew Morton * by truncate after unlock_page(). The address_space itself remains 3067d82fa87dSAndrew Morton * pinned by vma->vm_file's reference. We rely on unlock_page()'s 3068d82fa87dSAndrew Morton * release semantics to prevent the compiler from undoing this copying. 3069d82fa87dSAndrew Morton */ 3070f0c6d4d2SKirill A. Shutemov mapping = fault_page->mapping; 3071f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3072f0c6d4d2SKirill A. Shutemov if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { 3073b827e496SNick Piggin /* 3074b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3075b827e496SNick Piggin * dirty their pages 3076b827e496SNick Piggin */ 3077b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3078d08b3851SPeter Zijlstra } 3079d00806b1SNick Piggin 308074ec6751SJohannes Weiner if (!vma->vm_ops->page_mkwrite) 3081b827e496SNick Piggin file_update_time(vma->vm_file); 3082b827e496SNick Piggin 3083b827e496SNick Piggin return ret; 308454cb8821SNick Piggin } 3085d00806b1SNick Piggin 30869a95f3cfSPaul Cassella /* 30879a95f3cfSPaul Cassella * We enter with non-exclusive mmap_sem (to exclude vma changes, 30889a95f3cfSPaul Cassella * but allow concurrent faults). 30899a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 30909a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 30919a95f3cfSPaul Cassella */ 30929b4bdd2fSKirill A. Shutemov static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 309354cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 309430c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 309554cb8821SNick Piggin { 309654cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 30970da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 309854cb8821SNick Piggin 309916abfa08SHugh Dickins pte_unmap(page_table); 3100e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3101e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3102e655fb29SKirill A. Shutemov orig_pte); 3103ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3104ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3105ec47c3b9SKirill A. Shutemov orig_pte); 3106f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 310754cb8821SNick Piggin } 310854cb8821SNick Piggin 3109b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 311004bb2f94SRik van Riel unsigned long addr, int page_nid, 311104bb2f94SRik van Riel int *flags) 31129532fec1SMel Gorman { 31139532fec1SMel Gorman get_page(page); 31149532fec1SMel Gorman 31159532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 311604bb2f94SRik van Riel if (page_nid == numa_node_id()) { 31179532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 311804bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 311904bb2f94SRik van Riel } 31209532fec1SMel Gorman 31219532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 31229532fec1SMel Gorman } 31239532fec1SMel Gorman 3124b19a9939SRashika Kheria static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3125d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3126d10e63f2SMel Gorman { 31274daae3b4SMel Gorman struct page *page = NULL; 3128d10e63f2SMel Gorman spinlock_t *ptl; 31298191acbdSMel Gorman int page_nid = -1; 313090572890SPeter Zijlstra int last_cpupid; 3131cbee9f88SPeter Zijlstra int target_nid; 3132b8593bfdSMel Gorman bool migrated = false; 3133b191f9b1SMel Gorman bool was_writable = pte_write(pte); 31346688cc05SPeter Zijlstra int flags = 0; 3135d10e63f2SMel Gorman 3136c0e7cad9SMel Gorman /* A PROT_NONE fault should not end up here */ 3137c0e7cad9SMel Gorman BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))); 3138c0e7cad9SMel Gorman 3139d10e63f2SMel Gorman /* 3140d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3141d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3142d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3143d10e63f2SMel Gorman * 31444d942466SMel Gorman * We can safely just do a "set_pte_at()", because the old 31454d942466SMel Gorman * page table entry is not accessible, so there would be no 31464d942466SMel Gorman * concurrent hardware modifications to the PTE. 3147d10e63f2SMel Gorman */ 3148d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3149d10e63f2SMel Gorman spin_lock(ptl); 31504daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 31514daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 31524daae3b4SMel Gorman goto out; 31534daae3b4SMel Gorman } 31544daae3b4SMel Gorman 31554d942466SMel Gorman /* Make it present again */ 31564d942466SMel Gorman pte = pte_modify(pte, vma->vm_page_prot); 31574d942466SMel Gorman pte = pte_mkyoung(pte); 3158b191f9b1SMel Gorman if (was_writable) 3159b191f9b1SMel Gorman pte = pte_mkwrite(pte); 3160d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3161d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3162d10e63f2SMel Gorman 3163d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3164d10e63f2SMel Gorman if (!page) { 3165d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3166d10e63f2SMel Gorman return 0; 3167d10e63f2SMel Gorman } 3168d10e63f2SMel Gorman 31696688cc05SPeter Zijlstra /* 3170bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 3171bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 3172bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 3173bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 3174bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 3175bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 31766688cc05SPeter Zijlstra */ 3177bea66fbdSMel Gorman if (!(vma->vm_flags & VM_WRITE)) 31786688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 31796688cc05SPeter Zijlstra 3180dabe1d99SRik van Riel /* 3181dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 3182dabe1d99SRik van Riel * is later used when determining whether to group tasks together 3183dabe1d99SRik van Riel */ 3184dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 3185dabe1d99SRik van Riel flags |= TNF_SHARED; 3186dabe1d99SRik van Riel 318790572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 31888191acbdSMel Gorman page_nid = page_to_nid(page); 318904bb2f94SRik van Riel target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); 3190d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 31914daae3b4SMel Gorman if (target_nid == -1) { 31924daae3b4SMel Gorman put_page(page); 31934daae3b4SMel Gorman goto out; 31944daae3b4SMel Gorman } 31954daae3b4SMel Gorman 31964daae3b4SMel Gorman /* Migrate to the requested node */ 31971bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 31986688cc05SPeter Zijlstra if (migrated) { 31998191acbdSMel Gorman page_nid = target_nid; 32006688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 3201074c2381SMel Gorman } else 3202074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 32034daae3b4SMel Gorman 32044daae3b4SMel Gorman out: 32058191acbdSMel Gorman if (page_nid != -1) 32066688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 3207d10e63f2SMel Gorman return 0; 3208d10e63f2SMel Gorman } 3209d10e63f2SMel Gorman 32101da177e4SLinus Torvalds /* 32111da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 32121da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 32131da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 32141da177e4SLinus Torvalds * 32151da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 32161da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 32171da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 32181da177e4SLinus Torvalds * 3219c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3220c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 32219a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 32229a95f3cfSPaul Cassella * 32239a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 32249a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 32251da177e4SLinus Torvalds */ 3226c0292554SKirill A. Shutemov static int handle_pte_fault(struct mm_struct *mm, 32271da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 322830c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 32291da177e4SLinus Torvalds { 32301da177e4SLinus Torvalds pte_t entry; 32318f4e2101SHugh Dickins spinlock_t *ptl; 32321da177e4SLinus Torvalds 3233e37c6982SChristian Borntraeger /* 3234e37c6982SChristian Borntraeger * some architectures can have larger ptes than wordsize, 3235e37c6982SChristian Borntraeger * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y, 3236e37c6982SChristian Borntraeger * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses. 3237e37c6982SChristian Borntraeger * The code below just needs a consistent view for the ifs and 3238e37c6982SChristian Borntraeger * we later double check anyway with the ptl lock held. So here 3239e37c6982SChristian Borntraeger * a barrier will do. 3240e37c6982SChristian Borntraeger */ 3241e37c6982SChristian Borntraeger entry = *pte; 3242e37c6982SChristian Borntraeger barrier(); 32431da177e4SLinus Torvalds if (!pte_present(entry)) { 324465500d23SHugh Dickins if (pte_none(entry)) { 3245f4b81804SJes Sorensen if (vma->vm_ops) { 32463c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 32479b4bdd2fSKirill A. Shutemov return do_fault(mm, vma, address, pte, 32489b4bdd2fSKirill A. Shutemov pmd, flags, entry); 3249f4b81804SJes Sorensen } 3250f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 325130c9f3a9SLinus Torvalds pte, pmd, flags); 325265500d23SHugh Dickins } 325365500d23SHugh Dickins return do_swap_page(mm, vma, address, 325430c9f3a9SLinus Torvalds pte, pmd, flags, entry); 32551da177e4SLinus Torvalds } 32561da177e4SLinus Torvalds 32578a0516edSMel Gorman if (pte_protnone(entry)) 3258d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3259d10e63f2SMel Gorman 32604c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 32618f4e2101SHugh Dickins spin_lock(ptl); 32628f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 32638f4e2101SHugh Dickins goto unlock; 326430c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 32651da177e4SLinus Torvalds if (!pte_write(entry)) 32668f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 32678f4e2101SHugh Dickins pte, pmd, ptl, entry); 32681da177e4SLinus Torvalds entry = pte_mkdirty(entry); 32691da177e4SLinus Torvalds } 32701da177e4SLinus Torvalds entry = pte_mkyoung(entry); 327130c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 32724b3073e1SRussell King update_mmu_cache(vma, address, pte); 32731a44e149SAndrea Arcangeli } else { 32741a44e149SAndrea Arcangeli /* 32751a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 32761a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 32771a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 32781a44e149SAndrea Arcangeli * with threads. 32791a44e149SAndrea Arcangeli */ 328030c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 328161c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 32821a44e149SAndrea Arcangeli } 32838f4e2101SHugh Dickins unlock: 32848f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 328583c54070SNick Piggin return 0; 32861da177e4SLinus Torvalds } 32871da177e4SLinus Torvalds 32881da177e4SLinus Torvalds /* 32891da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 32909a95f3cfSPaul Cassella * 32919a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 32929a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 32931da177e4SLinus Torvalds */ 3294519e5247SJohannes Weiner static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3295d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 32961da177e4SLinus Torvalds { 32971da177e4SLinus Torvalds pgd_t *pgd; 32981da177e4SLinus Torvalds pud_t *pud; 32991da177e4SLinus Torvalds pmd_t *pmd; 33001da177e4SLinus Torvalds pte_t *pte; 33011da177e4SLinus Torvalds 3302ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 330330c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 33041da177e4SLinus Torvalds 33051da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 33061da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 33071da177e4SLinus Torvalds if (!pud) 3308c74df32cSHugh Dickins return VM_FAULT_OOM; 33091da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 33101da177e4SLinus Torvalds if (!pmd) 3311c74df32cSHugh Dickins return VM_FAULT_OOM; 331271e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 3313c0292554SKirill A. Shutemov int ret = VM_FAULT_FALLBACK; 331471e3aac0SAndrea Arcangeli if (!vma->vm_ops) 3315c0292554SKirill A. Shutemov ret = do_huge_pmd_anonymous_page(mm, vma, address, 331671e3aac0SAndrea Arcangeli pmd, flags); 3317c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 3318c0292554SKirill A. Shutemov return ret; 331971e3aac0SAndrea Arcangeli } else { 332071e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 33211f1d06c3SDavid Rientjes int ret; 33221f1d06c3SDavid Rientjes 332371e3aac0SAndrea Arcangeli barrier(); 332471e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3325a1dd450bSWill Deacon unsigned int dirty = flags & FAULT_FLAG_WRITE; 3326a1dd450bSWill Deacon 3327e53289c0SLinus Torvalds /* 3328e53289c0SLinus Torvalds * If the pmd is splitting, return and retry the 3329e53289c0SLinus Torvalds * the fault. Alternative: wait until the split 3330e53289c0SLinus Torvalds * is done, and goto retry. 3331e53289c0SLinus Torvalds */ 3332e53289c0SLinus Torvalds if (pmd_trans_splitting(orig_pmd)) 3333e53289c0SLinus Torvalds return 0; 3334e53289c0SLinus Torvalds 33358a0516edSMel Gorman if (pmd_protnone(orig_pmd)) 33364daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3337d10e63f2SMel Gorman orig_pmd, pmd); 3338d10e63f2SMel Gorman 33393d59eebcSLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 33401f1d06c3SDavid Rientjes ret = do_huge_pmd_wp_page(mm, vma, address, pmd, 33411f1d06c3SDavid Rientjes orig_pmd); 33429845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 33431f1d06c3SDavid Rientjes return ret; 3344a1dd450bSWill Deacon } else { 3345a1dd450bSWill Deacon huge_pmd_set_accessed(mm, vma, address, pmd, 3346a1dd450bSWill Deacon orig_pmd, dirty); 334771e3aac0SAndrea Arcangeli return 0; 334871e3aac0SAndrea Arcangeli } 334971e3aac0SAndrea Arcangeli } 33509845cbbdSKirill A. Shutemov } 335171e3aac0SAndrea Arcangeli 335271e3aac0SAndrea Arcangeli /* 335371e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 335471e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 335571e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 335671e3aac0SAndrea Arcangeli */ 33574fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 33584fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3359c74df32cSHugh Dickins return VM_FAULT_OOM; 336071e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 336171e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 336271e3aac0SAndrea Arcangeli return 0; 336371e3aac0SAndrea Arcangeli /* 336471e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 336571e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 336671e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 336771e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 336871e3aac0SAndrea Arcangeli */ 336971e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 33701da177e4SLinus Torvalds 337130c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 33721da177e4SLinus Torvalds } 33731da177e4SLinus Torvalds 33749a95f3cfSPaul Cassella /* 33759a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 33769a95f3cfSPaul Cassella * 33779a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 33789a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 33799a95f3cfSPaul Cassella */ 3380519e5247SJohannes Weiner int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3381519e5247SJohannes Weiner unsigned long address, unsigned int flags) 3382519e5247SJohannes Weiner { 3383519e5247SJohannes Weiner int ret; 3384519e5247SJohannes Weiner 3385519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 3386519e5247SJohannes Weiner 3387519e5247SJohannes Weiner count_vm_event(PGFAULT); 3388519e5247SJohannes Weiner mem_cgroup_count_vm_event(mm, PGFAULT); 3389519e5247SJohannes Weiner 3390519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 3391519e5247SJohannes Weiner check_sync_rss_stat(current); 3392519e5247SJohannes Weiner 3393519e5247SJohannes Weiner /* 3394519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 3395519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 3396519e5247SJohannes Weiner */ 3397519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 339849426420SJohannes Weiner mem_cgroup_oom_enable(); 3399519e5247SJohannes Weiner 3400519e5247SJohannes Weiner ret = __handle_mm_fault(mm, vma, address, flags); 3401519e5247SJohannes Weiner 340249426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 340349426420SJohannes Weiner mem_cgroup_oom_disable(); 340449426420SJohannes Weiner /* 340549426420SJohannes Weiner * The task may have entered a memcg OOM situation but 340649426420SJohannes Weiner * if the allocation error was handled gracefully (no 340749426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 340849426420SJohannes Weiner * Just clean up the OOM state peacefully. 340949426420SJohannes Weiner */ 341049426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 341149426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 341249426420SJohannes Weiner } 34133812c8c8SJohannes Weiner 3414519e5247SJohannes Weiner return ret; 3415519e5247SJohannes Weiner } 3416e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 3417519e5247SJohannes Weiner 34181da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 34191da177e4SLinus Torvalds /* 34201da177e4SLinus Torvalds * Allocate page upper directory. 3421872fec16SHugh Dickins * We've already handled the fast-path in-line. 34221da177e4SLinus Torvalds */ 34231bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 34241da177e4SLinus Torvalds { 3425c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3426c74df32cSHugh Dickins if (!new) 34271bb3630eSHugh Dickins return -ENOMEM; 34281da177e4SLinus Torvalds 3429362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3430362a61adSNick Piggin 3431872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 34321bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 34335e541973SBenjamin Herrenschmidt pud_free(mm, new); 34341bb3630eSHugh Dickins else 34351da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3436872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 34371bb3630eSHugh Dickins return 0; 34381da177e4SLinus Torvalds } 34391da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 34401da177e4SLinus Torvalds 34411da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 34421da177e4SLinus Torvalds /* 34431da177e4SLinus Torvalds * Allocate page middle directory. 3444872fec16SHugh Dickins * We've already handled the fast-path in-line. 34451da177e4SLinus Torvalds */ 34461bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 34471da177e4SLinus Torvalds { 3448c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3449c74df32cSHugh Dickins if (!new) 34501bb3630eSHugh Dickins return -ENOMEM; 34511da177e4SLinus Torvalds 3452362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3453362a61adSNick Piggin 3454872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 34551da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 3456dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 3457dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 34581da177e4SLinus Torvalds pud_populate(mm, pud, new); 3459dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 34605e541973SBenjamin Herrenschmidt pmd_free(mm, new); 3461dc6c9a35SKirill A. Shutemov #else 3462dc6c9a35SKirill A. Shutemov if (!pgd_present(*pud)) { 3463dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 34641da177e4SLinus Torvalds pgd_populate(mm, pud, new); 3465dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 3466dc6c9a35SKirill A. Shutemov pmd_free(mm, new); 34671da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3468872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 34691bb3630eSHugh Dickins return 0; 34701da177e4SLinus Torvalds } 34711da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 34721da177e4SLinus Torvalds 34731b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3474f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3475f8ad0f49SJohannes Weiner { 3476f8ad0f49SJohannes Weiner pgd_t *pgd; 3477f8ad0f49SJohannes Weiner pud_t *pud; 3478f8ad0f49SJohannes Weiner pmd_t *pmd; 3479f8ad0f49SJohannes Weiner pte_t *ptep; 3480f8ad0f49SJohannes Weiner 3481f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3482f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3483f8ad0f49SJohannes Weiner goto out; 3484f8ad0f49SJohannes Weiner 3485f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3486f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3487f8ad0f49SJohannes Weiner goto out; 3488f8ad0f49SJohannes Weiner 3489f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3490f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3491f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3492f8ad0f49SJohannes Weiner goto out; 3493f8ad0f49SJohannes Weiner 3494f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3495f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3496f8ad0f49SJohannes Weiner goto out; 3497f8ad0f49SJohannes Weiner 3498f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3499f8ad0f49SJohannes Weiner if (!ptep) 3500f8ad0f49SJohannes Weiner goto out; 3501f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3502f8ad0f49SJohannes Weiner goto unlock; 3503f8ad0f49SJohannes Weiner *ptepp = ptep; 3504f8ad0f49SJohannes Weiner return 0; 3505f8ad0f49SJohannes Weiner unlock: 3506f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3507f8ad0f49SJohannes Weiner out: 3508f8ad0f49SJohannes Weiner return -EINVAL; 3509f8ad0f49SJohannes Weiner } 3510f8ad0f49SJohannes Weiner 35111b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 35121b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 35131b36ba81SNamhyung Kim { 35141b36ba81SNamhyung Kim int res; 35151b36ba81SNamhyung Kim 35161b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 35171b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 35181b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 35191b36ba81SNamhyung Kim return res; 35201b36ba81SNamhyung Kim } 35211b36ba81SNamhyung Kim 35223b6748e2SJohannes Weiner /** 35233b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 35243b6748e2SJohannes Weiner * @vma: memory mapping 35253b6748e2SJohannes Weiner * @address: user virtual address 35263b6748e2SJohannes Weiner * @pfn: location to store found PFN 35273b6748e2SJohannes Weiner * 35283b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 35293b6748e2SJohannes Weiner * 35303b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 35313b6748e2SJohannes Weiner */ 35323b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 35333b6748e2SJohannes Weiner unsigned long *pfn) 35343b6748e2SJohannes Weiner { 35353b6748e2SJohannes Weiner int ret = -EINVAL; 35363b6748e2SJohannes Weiner spinlock_t *ptl; 35373b6748e2SJohannes Weiner pte_t *ptep; 35383b6748e2SJohannes Weiner 35393b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 35403b6748e2SJohannes Weiner return ret; 35413b6748e2SJohannes Weiner 35423b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 35433b6748e2SJohannes Weiner if (ret) 35443b6748e2SJohannes Weiner return ret; 35453b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 35463b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 35473b6748e2SJohannes Weiner return 0; 35483b6748e2SJohannes Weiner } 35493b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 35503b6748e2SJohannes Weiner 355128b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 3552d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 355328b2ee20SRik van Riel unsigned long address, unsigned int flags, 3554d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 355528b2ee20SRik van Riel { 355603668a4dSJohannes Weiner int ret = -EINVAL; 355728b2ee20SRik van Riel pte_t *ptep, pte; 355828b2ee20SRik van Riel spinlock_t *ptl; 355928b2ee20SRik van Riel 3560d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 3561d87fe660Svenkatesh.pallipadi@intel.com goto out; 356228b2ee20SRik van Riel 356303668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 3564d87fe660Svenkatesh.pallipadi@intel.com goto out; 356528b2ee20SRik van Riel pte = *ptep; 356603668a4dSJohannes Weiner 356728b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 356828b2ee20SRik van Riel goto unlock; 356928b2ee20SRik van Riel 357028b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 357103668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 357228b2ee20SRik van Riel 357303668a4dSJohannes Weiner ret = 0; 357428b2ee20SRik van Riel unlock: 357528b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 357628b2ee20SRik van Riel out: 3577d87fe660Svenkatesh.pallipadi@intel.com return ret; 357828b2ee20SRik van Riel } 357928b2ee20SRik van Riel 358028b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 358128b2ee20SRik van Riel void *buf, int len, int write) 358228b2ee20SRik van Riel { 358328b2ee20SRik van Riel resource_size_t phys_addr; 358428b2ee20SRik van Riel unsigned long prot = 0; 35852bc7273bSKOSAKI Motohiro void __iomem *maddr; 358628b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 358728b2ee20SRik van Riel 3588d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 358928b2ee20SRik van Riel return -EINVAL; 359028b2ee20SRik van Riel 35919cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 359228b2ee20SRik van Riel if (write) 359328b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 359428b2ee20SRik van Riel else 359528b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 359628b2ee20SRik van Riel iounmap(maddr); 359728b2ee20SRik van Riel 359828b2ee20SRik van Riel return len; 359928b2ee20SRik van Riel } 36005a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 360128b2ee20SRik van Riel #endif 360228b2ee20SRik van Riel 36030ec76a11SDavid Howells /* 3604206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 3605206cb636SStephen Wilson * given task for page fault accounting. 36060ec76a11SDavid Howells */ 3607206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 3608206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 36090ec76a11SDavid Howells { 36100ec76a11SDavid Howells struct vm_area_struct *vma; 36110ec76a11SDavid Howells void *old_buf = buf; 36120ec76a11SDavid Howells 36130ec76a11SDavid Howells down_read(&mm->mmap_sem); 3614183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 36150ec76a11SDavid Howells while (len) { 36160ec76a11SDavid Howells int bytes, ret, offset; 36170ec76a11SDavid Howells void *maddr; 361828b2ee20SRik van Riel struct page *page = NULL; 36190ec76a11SDavid Howells 36200ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 36210ec76a11SDavid Howells write, 1, &page, &vma); 362228b2ee20SRik van Riel if (ret <= 0) { 3623dbffcd03SRik van Riel #ifndef CONFIG_HAVE_IOREMAP_PROT 3624dbffcd03SRik van Riel break; 3625dbffcd03SRik van Riel #else 362628b2ee20SRik van Riel /* 362728b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 362828b2ee20SRik van Riel * we can access using slightly different code. 362928b2ee20SRik van Riel */ 363028b2ee20SRik van Riel vma = find_vma(mm, addr); 3631fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 36320ec76a11SDavid Howells break; 363328b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 363428b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 363528b2ee20SRik van Riel len, write); 363628b2ee20SRik van Riel if (ret <= 0) 363728b2ee20SRik van Riel break; 363828b2ee20SRik van Riel bytes = ret; 3639dbffcd03SRik van Riel #endif 364028b2ee20SRik van Riel } else { 36410ec76a11SDavid Howells bytes = len; 36420ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 36430ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 36440ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 36450ec76a11SDavid Howells 36460ec76a11SDavid Howells maddr = kmap(page); 36470ec76a11SDavid Howells if (write) { 36480ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 36490ec76a11SDavid Howells maddr + offset, buf, bytes); 36500ec76a11SDavid Howells set_page_dirty_lock(page); 36510ec76a11SDavid Howells } else { 36520ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 36530ec76a11SDavid Howells buf, maddr + offset, bytes); 36540ec76a11SDavid Howells } 36550ec76a11SDavid Howells kunmap(page); 36560ec76a11SDavid Howells page_cache_release(page); 365728b2ee20SRik van Riel } 36580ec76a11SDavid Howells len -= bytes; 36590ec76a11SDavid Howells buf += bytes; 36600ec76a11SDavid Howells addr += bytes; 36610ec76a11SDavid Howells } 36620ec76a11SDavid Howells up_read(&mm->mmap_sem); 36630ec76a11SDavid Howells 36640ec76a11SDavid Howells return buf - old_buf; 36650ec76a11SDavid Howells } 366603252919SAndi Kleen 36675ddd36b9SStephen Wilson /** 3668ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 36695ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 36705ddd36b9SStephen Wilson * @addr: start address to access 36715ddd36b9SStephen Wilson * @buf: source or destination buffer 36725ddd36b9SStephen Wilson * @len: number of bytes to transfer 36735ddd36b9SStephen Wilson * @write: whether the access is a write 36745ddd36b9SStephen Wilson * 36755ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 36765ddd36b9SStephen Wilson */ 36775ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 36785ddd36b9SStephen Wilson void *buf, int len, int write) 36795ddd36b9SStephen Wilson { 36805ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 36815ddd36b9SStephen Wilson } 36825ddd36b9SStephen Wilson 368303252919SAndi Kleen /* 3684206cb636SStephen Wilson * Access another process' address space. 3685206cb636SStephen Wilson * Source/target buffer must be kernel space, 3686206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 3687206cb636SStephen Wilson */ 3688206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 3689206cb636SStephen Wilson void *buf, int len, int write) 3690206cb636SStephen Wilson { 3691206cb636SStephen Wilson struct mm_struct *mm; 3692206cb636SStephen Wilson int ret; 3693206cb636SStephen Wilson 3694206cb636SStephen Wilson mm = get_task_mm(tsk); 3695206cb636SStephen Wilson if (!mm) 3696206cb636SStephen Wilson return 0; 3697206cb636SStephen Wilson 3698206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 3699206cb636SStephen Wilson mmput(mm); 3700206cb636SStephen Wilson 3701206cb636SStephen Wilson return ret; 3702206cb636SStephen Wilson } 3703206cb636SStephen Wilson 370403252919SAndi Kleen /* 370503252919SAndi Kleen * Print the name of a VMA. 370603252919SAndi Kleen */ 370703252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 370803252919SAndi Kleen { 370903252919SAndi Kleen struct mm_struct *mm = current->mm; 371003252919SAndi Kleen struct vm_area_struct *vma; 371103252919SAndi Kleen 3712e8bff74aSIngo Molnar /* 3713e8bff74aSIngo Molnar * Do not print if we are in atomic 3714e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 3715e8bff74aSIngo Molnar */ 3716e8bff74aSIngo Molnar if (preempt_count()) 3717e8bff74aSIngo Molnar return; 3718e8bff74aSIngo Molnar 371903252919SAndi Kleen down_read(&mm->mmap_sem); 372003252919SAndi Kleen vma = find_vma(mm, ip); 372103252919SAndi Kleen if (vma && vma->vm_file) { 372203252919SAndi Kleen struct file *f = vma->vm_file; 372303252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 372403252919SAndi Kleen if (buf) { 37252fbc57c5SAndy Shevchenko char *p; 372603252919SAndi Kleen 3727cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 372803252919SAndi Kleen if (IS_ERR(p)) 372903252919SAndi Kleen p = "?"; 37302fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 373103252919SAndi Kleen vma->vm_start, 373203252919SAndi Kleen vma->vm_end - vma->vm_start); 373303252919SAndi Kleen free_page((unsigned long)buf); 373403252919SAndi Kleen } 373503252919SAndi Kleen } 373651a07e50SJeff Liu up_read(&mm->mmap_sem); 373703252919SAndi Kleen } 37383ee1afa3SNick Piggin 3739662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 3740*9ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 37413ee1afa3SNick Piggin { 374295156f00SPeter Zijlstra /* 374395156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 374495156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 374595156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 374695156f00SPeter Zijlstra * below annotations will generate false positives. 374795156f00SPeter Zijlstra */ 374895156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 374995156f00SPeter Zijlstra return; 3750*9ec23531SDavid Hildenbrand if (pagefault_disabled()) 3751662bbcb2SMichael S. Tsirkin return; 3752*9ec23531SDavid Hildenbrand __might_sleep(file, line, 0); 3753*9ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 3754662bbcb2SMichael S. Tsirkin if (current->mm) 37553ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 3756*9ec23531SDavid Hildenbrand #endif 37573ee1afa3SNick Piggin } 3758*9ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 37593ee1afa3SNick Piggin #endif 376047ad8475SAndrea Arcangeli 376147ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 376247ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 376347ad8475SAndrea Arcangeli unsigned long addr, 376447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 376547ad8475SAndrea Arcangeli { 376647ad8475SAndrea Arcangeli int i; 376747ad8475SAndrea Arcangeli struct page *p = page; 376847ad8475SAndrea Arcangeli 376947ad8475SAndrea Arcangeli might_sleep(); 377047ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 377147ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 377247ad8475SAndrea Arcangeli cond_resched(); 377347ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 377447ad8475SAndrea Arcangeli } 377547ad8475SAndrea Arcangeli } 377647ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 377747ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 377847ad8475SAndrea Arcangeli { 377947ad8475SAndrea Arcangeli int i; 378047ad8475SAndrea Arcangeli 378147ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 378247ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 378347ad8475SAndrea Arcangeli return; 378447ad8475SAndrea Arcangeli } 378547ad8475SAndrea Arcangeli 378647ad8475SAndrea Arcangeli might_sleep(); 378747ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 378847ad8475SAndrea Arcangeli cond_resched(); 378947ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 379047ad8475SAndrea Arcangeli } 379147ad8475SAndrea Arcangeli } 379247ad8475SAndrea Arcangeli 379347ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 379447ad8475SAndrea Arcangeli unsigned long addr, 379547ad8475SAndrea Arcangeli struct vm_area_struct *vma, 379647ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 379747ad8475SAndrea Arcangeli { 379847ad8475SAndrea Arcangeli int i; 379947ad8475SAndrea Arcangeli struct page *dst_base = dst; 380047ad8475SAndrea Arcangeli struct page *src_base = src; 380147ad8475SAndrea Arcangeli 380247ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 380347ad8475SAndrea Arcangeli cond_resched(); 380447ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 380547ad8475SAndrea Arcangeli 380647ad8475SAndrea Arcangeli i++; 380747ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 380847ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 380947ad8475SAndrea Arcangeli } 381047ad8475SAndrea Arcangeli } 381147ad8475SAndrea Arcangeli 381247ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 381347ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 381447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 381547ad8475SAndrea Arcangeli { 381647ad8475SAndrea Arcangeli int i; 381747ad8475SAndrea Arcangeli 381847ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 381947ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 382047ad8475SAndrea Arcangeli pages_per_huge_page); 382147ad8475SAndrea Arcangeli return; 382247ad8475SAndrea Arcangeli } 382347ad8475SAndrea Arcangeli 382447ad8475SAndrea Arcangeli might_sleep(); 382547ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 382647ad8475SAndrea Arcangeli cond_resched(); 382747ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 382847ad8475SAndrea Arcangeli } 382947ad8475SAndrea Arcangeli } 383047ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 383149076ec2SKirill A. Shutemov 383240b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 3833b35f1819SKirill A. Shutemov 3834b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 3835b35f1819SKirill A. Shutemov 3836b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 3837b35f1819SKirill A. Shutemov { 3838b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 3839b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 3840b35f1819SKirill A. Shutemov } 3841b35f1819SKirill A. Shutemov 3842539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 384349076ec2SKirill A. Shutemov { 384449076ec2SKirill A. Shutemov spinlock_t *ptl; 384549076ec2SKirill A. Shutemov 3846b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 384749076ec2SKirill A. Shutemov if (!ptl) 384849076ec2SKirill A. Shutemov return false; 3849539edb58SPeter Zijlstra page->ptl = ptl; 385049076ec2SKirill A. Shutemov return true; 385149076ec2SKirill A. Shutemov } 385249076ec2SKirill A. Shutemov 3853539edb58SPeter Zijlstra void ptlock_free(struct page *page) 385449076ec2SKirill A. Shutemov { 3855b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 385649076ec2SKirill A. Shutemov } 385749076ec2SKirill A. Shutemov #endif 3858