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 */ 6933dc14741SHugh Dickins if (vma->vm_ops) 694071361d3SJoe Perches printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n", 695071361d3SJoe Perches vma->vm_ops->fault); 69672c2d531SAl Viro if (vma->vm_file) 697071361d3SJoe Perches printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n", 698071361d3SJoe Perches vma->vm_file->f_op->mmap); 699b5810039SNick Piggin dump_stack(); 700373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 701b5810039SNick Piggin } 702b5810039SNick Piggin 703b5810039SNick Piggin /* 7047e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 7056aab341eSLinus Torvalds * 7067e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7077e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7087e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 709b379d790SJared Hulbert * 7107e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7117e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7127e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7137e675137SNick Piggin * described below. 7147e675137SNick Piggin * 7157e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7167e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7177e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7186aab341eSLinus Torvalds * 719b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 720b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7217e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7227e675137SNick Piggin * mapping will always honor the rule 7236aab341eSLinus Torvalds * 7246aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7256aab341eSLinus Torvalds * 7267e675137SNick Piggin * And for normal mappings this is false. 727b379d790SJared Hulbert * 7287e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7297e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7307e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7317e675137SNick Piggin * special (because none can have been COWed). 732b379d790SJared Hulbert * 733b379d790SJared Hulbert * 7347e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7357e675137SNick Piggin * 736b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 737b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 738b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 739b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 740b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 741b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 742b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 743b379d790SJared Hulbert * 744ee498ed7SHugh Dickins */ 7457e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7467e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7477e675137SNick Piggin #else 7487e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7497e675137SNick Piggin #endif 7507e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7517e675137SNick Piggin pte_t pte) 752ee498ed7SHugh Dickins { 75322b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7547e675137SNick Piggin 7557e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 756b38af472SHugh Dickins if (likely(!pte_special(pte))) 75722b31eecSHugh Dickins goto check_pfn; 758667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 759667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 760a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 761a13ea5b7SHugh Dickins return NULL; 76262eede62SHugh Dickins if (!is_zero_pfn(pfn)) 76322b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7647e675137SNick Piggin return NULL; 7657e675137SNick Piggin } 7667e675137SNick Piggin 7677e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7687e675137SNick Piggin 769b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 770b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 771b379d790SJared Hulbert if (!pfn_valid(pfn)) 772b379d790SJared Hulbert return NULL; 773b379d790SJared Hulbert goto out; 774b379d790SJared Hulbert } else { 7757e675137SNick Piggin unsigned long off; 7767e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7776aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7786aab341eSLinus Torvalds return NULL; 77967121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 780fb155c16SLinus Torvalds return NULL; 7816aab341eSLinus Torvalds } 782b379d790SJared Hulbert } 7836aab341eSLinus Torvalds 784b38af472SHugh Dickins if (is_zero_pfn(pfn)) 785b38af472SHugh Dickins return NULL; 78622b31eecSHugh Dickins check_pfn: 78722b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 78822b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 78922b31eecSHugh Dickins return NULL; 79022b31eecSHugh Dickins } 7916aab341eSLinus Torvalds 7926aab341eSLinus Torvalds /* 7937e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7947e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7956aab341eSLinus Torvalds */ 796b379d790SJared Hulbert out: 7976aab341eSLinus Torvalds return pfn_to_page(pfn); 798ee498ed7SHugh Dickins } 799ee498ed7SHugh Dickins 800ee498ed7SHugh Dickins /* 8011da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 8021da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 8031da177e4SLinus Torvalds * covered by this vma. 8041da177e4SLinus Torvalds */ 8051da177e4SLinus Torvalds 806570a335bSHugh Dickins static inline unsigned long 8071da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 808b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8098c103762SHugh Dickins unsigned long addr, int *rss) 8101da177e4SLinus Torvalds { 811b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8121da177e4SLinus Torvalds pte_t pte = *src_pte; 8131da177e4SLinus Torvalds struct page *page; 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8161da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8170697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8180697212aSChristoph Lameter 8192022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 820570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 821570a335bSHugh Dickins return entry.val; 822570a335bSHugh Dickins 8231da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8241da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8251da177e4SLinus Torvalds spin_lock(&mmlist_lock); 826f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 827f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 828f412ac08SHugh Dickins &src_mm->mmlist); 8291da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8301da177e4SLinus Torvalds } 831b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8322022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 8339f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8349f9f1acdSKonstantin Khlebnikov 8359f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 8369f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]++; 8379f9f1acdSKonstantin Khlebnikov else 8389f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]++; 8399f9f1acdSKonstantin Khlebnikov 8409f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8410697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8420697212aSChristoph Lameter /* 8439f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8449f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8450697212aSChristoph Lameter */ 8460697212aSChristoph Lameter make_migration_entry_read(&entry); 8470697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 848c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 849c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 8500697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8510697212aSChristoph Lameter } 8521da177e4SLinus Torvalds } 853ae859762SHugh Dickins goto out_set_pte; 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds /* 8571da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8581da177e4SLinus Torvalds * in the parent and the child 8591da177e4SLinus Torvalds */ 86067121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8611da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8623dc90795SZachary Amsden pte = pte_wrprotect(pte); 8631da177e4SLinus Torvalds } 8641da177e4SLinus Torvalds 8651da177e4SLinus Torvalds /* 8661da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8671da177e4SLinus Torvalds * the child 8681da177e4SLinus Torvalds */ 8691da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8701da177e4SLinus Torvalds pte = pte_mkclean(pte); 8711da177e4SLinus Torvalds pte = pte_mkold(pte); 8726aab341eSLinus Torvalds 8736aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 8746aab341eSLinus Torvalds if (page) { 8751da177e4SLinus Torvalds get_page(page); 87621333b2bSHugh Dickins page_dup_rmap(page); 877d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 878d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 879d559db08SKAMEZAWA Hiroyuki else 880d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 8816aab341eSLinus Torvalds } 882ae859762SHugh Dickins 883ae859762SHugh Dickins out_set_pte: 884ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 885570a335bSHugh Dickins return 0; 8861da177e4SLinus Torvalds } 8871da177e4SLinus Torvalds 88821bda264SJerome Marchand static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8891da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8901da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8911da177e4SLinus Torvalds { 892c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8931da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 894c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 895e040f218SHugh Dickins int progress = 0; 896d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 897570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 8981da177e4SLinus Torvalds 8991da177e4SLinus Torvalds again: 900d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 901d559db08SKAMEZAWA Hiroyuki 902c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 9031da177e4SLinus Torvalds if (!dst_pte) 9041da177e4SLinus Torvalds return -ENOMEM; 905ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 9064c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 907f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 908c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 909c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9106606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds do { 9131da177e4SLinus Torvalds /* 9141da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9151da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9161da177e4SLinus Torvalds */ 917e040f218SHugh Dickins if (progress >= 32) { 918e040f218SHugh Dickins progress = 0; 919e040f218SHugh Dickins if (need_resched() || 92095c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9211da177e4SLinus Torvalds break; 922e040f218SHugh Dickins } 9231da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9241da177e4SLinus Torvalds progress++; 9251da177e4SLinus Torvalds continue; 9261da177e4SLinus Torvalds } 927570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 928570a335bSHugh Dickins vma, addr, rss); 929570a335bSHugh Dickins if (entry.val) 930570a335bSHugh Dickins break; 9311da177e4SLinus Torvalds progress += 8; 9321da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9331da177e4SLinus Torvalds 9346606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 935c74df32cSHugh Dickins spin_unlock(src_ptl); 936ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 937d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 938c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 939c74df32cSHugh Dickins cond_resched(); 940570a335bSHugh Dickins 941570a335bSHugh Dickins if (entry.val) { 942570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 943570a335bSHugh Dickins return -ENOMEM; 944570a335bSHugh Dickins progress = 0; 945570a335bSHugh Dickins } 9461da177e4SLinus Torvalds if (addr != end) 9471da177e4SLinus Torvalds goto again; 9481da177e4SLinus Torvalds return 0; 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9521da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9531da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9541da177e4SLinus Torvalds { 9551da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9561da177e4SLinus Torvalds unsigned long next; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9591da177e4SLinus Torvalds if (!dst_pmd) 9601da177e4SLinus Torvalds return -ENOMEM; 9611da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9621da177e4SLinus Torvalds do { 9631da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 96471e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 96571e3aac0SAndrea Arcangeli int err; 96614d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 96771e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 96871e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 96971e3aac0SAndrea Arcangeli if (err == -ENOMEM) 97071e3aac0SAndrea Arcangeli return -ENOMEM; 97171e3aac0SAndrea Arcangeli if (!err) 97271e3aac0SAndrea Arcangeli continue; 97371e3aac0SAndrea Arcangeli /* fall through */ 97471e3aac0SAndrea Arcangeli } 9751da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 9761da177e4SLinus Torvalds continue; 9771da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 9781da177e4SLinus Torvalds vma, addr, next)) 9791da177e4SLinus Torvalds return -ENOMEM; 9801da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 9811da177e4SLinus Torvalds return 0; 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9851da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9861da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9871da177e4SLinus Torvalds { 9881da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9891da177e4SLinus Torvalds unsigned long next; 9901da177e4SLinus Torvalds 9911da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9921da177e4SLinus Torvalds if (!dst_pud) 9931da177e4SLinus Torvalds return -ENOMEM; 9941da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9951da177e4SLinus Torvalds do { 9961da177e4SLinus Torvalds next = pud_addr_end(addr, end); 9971da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 9981da177e4SLinus Torvalds continue; 9991da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 10001da177e4SLinus Torvalds vma, addr, next)) 10011da177e4SLinus Torvalds return -ENOMEM; 10021da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 10031da177e4SLinus Torvalds return 0; 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10071da177e4SLinus Torvalds struct vm_area_struct *vma) 10081da177e4SLinus Torvalds { 10091da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 10101da177e4SLinus Torvalds unsigned long next; 10111da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10121da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10132ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10142ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10152ec74c3eSSagi Grimberg bool is_cow; 1016cddb8a5cSAndrea Arcangeli int ret; 10171da177e4SLinus Torvalds 1018d992895bSNick Piggin /* 1019d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1020d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1021d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1022d992895bSNick Piggin * efficient than faulting. 1023d992895bSNick Piggin */ 10240661a336SKirill A. Shutemov if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && 10250661a336SKirill A. Shutemov !vma->anon_vma) 1026d992895bSNick Piggin return 0; 1027d992895bSNick Piggin 10281da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10291da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10301da177e4SLinus Torvalds 1031b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10322ab64037Svenkatesh.pallipadi@intel.com /* 10332ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10342ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10352ab64037Svenkatesh.pallipadi@intel.com */ 10365180da41SSuresh Siddha ret = track_pfn_copy(vma); 10372ab64037Svenkatesh.pallipadi@intel.com if (ret) 10382ab64037Svenkatesh.pallipadi@intel.com return ret; 10392ab64037Svenkatesh.pallipadi@intel.com } 10402ab64037Svenkatesh.pallipadi@intel.com 1041cddb8a5cSAndrea Arcangeli /* 1042cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1043cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1044cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1045cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1046cddb8a5cSAndrea Arcangeli */ 10472ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10482ec74c3eSSagi Grimberg mmun_start = addr; 10492ec74c3eSSagi Grimberg mmun_end = end; 10502ec74c3eSSagi Grimberg if (is_cow) 10512ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10522ec74c3eSSagi Grimberg mmun_end); 1053cddb8a5cSAndrea Arcangeli 1054cddb8a5cSAndrea Arcangeli ret = 0; 10551da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10561da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10571da177e4SLinus Torvalds do { 10581da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10591da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10601da177e4SLinus Torvalds continue; 1061cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1062cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1063cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1064cddb8a5cSAndrea Arcangeli break; 1065cddb8a5cSAndrea Arcangeli } 10661da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1067cddb8a5cSAndrea Arcangeli 10682ec74c3eSSagi Grimberg if (is_cow) 10692ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1070cddb8a5cSAndrea Arcangeli return ret; 10711da177e4SLinus Torvalds } 10721da177e4SLinus Torvalds 107351c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1074b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 10751da177e4SLinus Torvalds unsigned long addr, unsigned long end, 107697a89413SPeter Zijlstra struct zap_details *details) 10771da177e4SLinus Torvalds { 1078b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1079d16dfc55SPeter Zijlstra int force_flush = 0; 1080d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 108197a89413SPeter Zijlstra spinlock_t *ptl; 10825f1a1907SSteven Rostedt pte_t *start_pte; 108397a89413SPeter Zijlstra pte_t *pte; 10848a5f14a2SKirill A. Shutemov swp_entry_t entry; 1085d559db08SKAMEZAWA Hiroyuki 1086d16dfc55SPeter Zijlstra again: 1087e303297eSPeter Zijlstra init_rss_vec(rss); 10885f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10895f1a1907SSteven Rostedt pte = start_pte; 10906606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10911da177e4SLinus Torvalds do { 10921da177e4SLinus Torvalds pte_t ptent = *pte; 109351c6f666SRobin Holt if (pte_none(ptent)) { 10941da177e4SLinus Torvalds continue; 109551c6f666SRobin Holt } 109651c6f666SRobin Holt 10976f5e6b9eSHugh Dickins if (pte_present(ptent)) { 10986f5e6b9eSHugh Dickins struct page *page; 10996f5e6b9eSHugh Dickins 11006aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 11011da177e4SLinus Torvalds if (unlikely(details) && page) { 11021da177e4SLinus Torvalds /* 11031da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 11041da177e4SLinus Torvalds * invalidate cache without truncating: 11051da177e4SLinus Torvalds * unmap shared but keep private pages. 11061da177e4SLinus Torvalds */ 11071da177e4SLinus Torvalds if (details->check_mapping && 11081da177e4SLinus Torvalds details->check_mapping != page->mapping) 11091da177e4SLinus Torvalds continue; 11101da177e4SLinus Torvalds } 1111b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1112a600388dSZachary Amsden tlb->fullmm); 11131da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11141da177e4SLinus Torvalds if (unlikely(!page)) 11151da177e4SLinus Torvalds continue; 11161da177e4SLinus Torvalds if (PageAnon(page)) 1117d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 11186237bcd9SHugh Dickins else { 11191cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 11201cf35d47SLinus Torvalds force_flush = 1; 11216237bcd9SHugh Dickins set_page_dirty(page); 11221cf35d47SLinus Torvalds } 11234917e5d0SJohannes Weiner if (pte_young(ptent) && 112464363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1125bf3f3bc5SNick Piggin mark_page_accessed(page); 1126d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 11276237bcd9SHugh Dickins } 1128edc315fdSHugh Dickins page_remove_rmap(page); 11293dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11303dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 11311cf35d47SLinus Torvalds if (unlikely(!__tlb_remove_page(tlb, page))) { 11321cf35d47SLinus Torvalds force_flush = 1; 1133ce9ec37bSWill Deacon addr += PAGE_SIZE; 1134d16dfc55SPeter Zijlstra break; 11351cf35d47SLinus Torvalds } 11361da177e4SLinus Torvalds continue; 11371da177e4SLinus Torvalds } 11388a5f14a2SKirill A. Shutemov /* If details->check_mapping, we leave swap entries. */ 11391da177e4SLinus Torvalds if (unlikely(details)) 11401da177e4SLinus Torvalds continue; 1141b084d435SKAMEZAWA Hiroyuki 11428a5f14a2SKirill A. Shutemov entry = pte_to_swp_entry(ptent); 1143b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1144b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11459f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11469f9f1acdSKonstantin Khlebnikov struct page *page; 11479f9f1acdSKonstantin Khlebnikov 11489f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 11499f9f1acdSKonstantin Khlebnikov 11509f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 11519f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]--; 11529f9f1acdSKonstantin Khlebnikov else 11539f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]--; 11549f9f1acdSKonstantin Khlebnikov } 1155b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 11562509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 11579888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 115897a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1159ae859762SHugh Dickins 1160d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 11616606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 116251c6f666SRobin Holt 11631cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 1164fb7332a9SWill Deacon if (force_flush) 11651cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 11661cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 11671cf35d47SLinus Torvalds 11681cf35d47SLinus Torvalds /* 11691cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 11701cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 11711cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 11721cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 11731cf35d47SLinus Torvalds */ 11741cf35d47SLinus Torvalds if (force_flush) { 11751cf35d47SLinus Torvalds force_flush = 0; 11761cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 11772b047252SLinus Torvalds 11782b047252SLinus Torvalds if (addr != end) 1179d16dfc55SPeter Zijlstra goto again; 1180d16dfc55SPeter Zijlstra } 1181d16dfc55SPeter Zijlstra 118251c6f666SRobin Holt return addr; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds 118551c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1186b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 11871da177e4SLinus Torvalds unsigned long addr, unsigned long end, 118897a89413SPeter Zijlstra struct zap_details *details) 11891da177e4SLinus Torvalds { 11901da177e4SLinus Torvalds pmd_t *pmd; 11911da177e4SLinus Torvalds unsigned long next; 11921da177e4SLinus Torvalds 11931da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 11941da177e4SLinus Torvalds do { 11951da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 119671e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 119714d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1198e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1199e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1200e0897d75SDavid 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", 1201e0897d75SDavid Rientjes __func__, addr, end, 1202e0897d75SDavid Rientjes vma->vm_start, 1203e0897d75SDavid Rientjes vma->vm_end); 1204e0897d75SDavid Rientjes BUG(); 1205e0897d75SDavid Rientjes } 1206e0897d75SDavid Rientjes #endif 1207e180377fSKirill A. Shutemov split_huge_page_pmd(vma, addr, pmd); 1208f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 12091a5a9906SAndrea Arcangeli goto next; 121071e3aac0SAndrea Arcangeli /* fall through */ 121171e3aac0SAndrea Arcangeli } 12121a5a9906SAndrea Arcangeli /* 12131a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 12141a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 12151a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 12161a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12171a5a9906SAndrea Arcangeli * mode. 12181a5a9906SAndrea Arcangeli */ 12191a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12201a5a9906SAndrea Arcangeli goto next; 122197a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12221a5a9906SAndrea Arcangeli next: 122397a89413SPeter Zijlstra cond_resched(); 122497a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 122551c6f666SRobin Holt 122651c6f666SRobin Holt return addr; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 122951c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1230b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12311da177e4SLinus Torvalds unsigned long addr, unsigned long end, 123297a89413SPeter Zijlstra struct zap_details *details) 12331da177e4SLinus Torvalds { 12341da177e4SLinus Torvalds pud_t *pud; 12351da177e4SLinus Torvalds unsigned long next; 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12381da177e4SLinus Torvalds do { 12391da177e4SLinus Torvalds next = pud_addr_end(addr, end); 124097a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12411da177e4SLinus Torvalds continue; 124297a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 124397a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 124451c6f666SRobin Holt 124551c6f666SRobin Holt return addr; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 1248038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 124951c6f666SRobin Holt struct vm_area_struct *vma, 12501da177e4SLinus Torvalds unsigned long addr, unsigned long end, 125197a89413SPeter Zijlstra struct zap_details *details) 12521da177e4SLinus Torvalds { 12531da177e4SLinus Torvalds pgd_t *pgd; 12541da177e4SLinus Torvalds unsigned long next; 12551da177e4SLinus Torvalds 12568a5f14a2SKirill A. Shutemov if (details && !details->check_mapping) 12571da177e4SLinus Torvalds details = NULL; 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds BUG_ON(addr >= end); 12601da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 12611da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 12621da177e4SLinus Torvalds do { 12631da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 126497a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 12651da177e4SLinus Torvalds continue; 126697a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 126797a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 12681da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 1271f5cc4eefSAl Viro 1272f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 12731da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 12744f74d2c8SLinus Torvalds unsigned long end_addr, 12751da177e4SLinus Torvalds struct zap_details *details) 12761da177e4SLinus Torvalds { 1277f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 12781da177e4SLinus Torvalds unsigned long end; 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds if (start >= vma->vm_end) 1281f5cc4eefSAl Viro return; 12821da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 12831da177e4SLinus Torvalds if (end <= vma->vm_start) 1284f5cc4eefSAl Viro return; 12851da177e4SLinus Torvalds 1286cbc91f71SSrikar Dronamraju if (vma->vm_file) 1287cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1288cbc91f71SSrikar Dronamraju 1289b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 12905180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 12912ab64037Svenkatesh.pallipadi@intel.com 12928b2a1238SAl Viro if (start != end) { 129351c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1294a137e1ccSAndi Kleen /* 1295a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1296a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1297a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 12987aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1299a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 13007aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1301a137e1ccSAndi Kleen * before calling this function to clean up. 1302a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1303a137e1ccSAndi Kleen * safe to do nothing in this case. 1304a137e1ccSAndi Kleen */ 130524669e58SAneesh Kumar K.V if (vma->vm_file) { 130683cde9e8SDavidlohr Bueso i_mmap_lock_write(vma->vm_file->f_mapping); 1307d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 130883cde9e8SDavidlohr Bueso i_mmap_unlock_write(vma->vm_file->f_mapping); 130924669e58SAneesh Kumar K.V } 131051c6f666SRobin Holt } else 1311038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 131297a89413SPeter Zijlstra } 131351c6f666SRobin Holt } 13141da177e4SLinus Torvalds 1315f5cc4eefSAl Viro /** 1316f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1317f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1318f5cc4eefSAl Viro * @vma: the starting vma 1319f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1320f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1321f5cc4eefSAl Viro * 1322f5cc4eefSAl Viro * Unmap all pages in the vma list. 1323f5cc4eefSAl Viro * 1324f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1325f5cc4eefSAl Viro * 1326f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1327f5cc4eefSAl Viro * 1328f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1329f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1330f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1331f5cc4eefSAl Viro * drops the lock and schedules. 1332f5cc4eefSAl Viro */ 1333f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1334f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13354f74d2c8SLinus Torvalds unsigned long end_addr) 1336f5cc4eefSAl Viro { 1337f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1338f5cc4eefSAl Viro 1339f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1340f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13414f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1342cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13431da177e4SLinus Torvalds } 13441da177e4SLinus Torvalds 13451da177e4SLinus Torvalds /** 13461da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13471da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1348eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13491da177e4SLinus Torvalds * @size: number of bytes to zap 13508a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1351f5cc4eefSAl Viro * 1352f5cc4eefSAl Viro * Caller must protect the VMA list 13531da177e4SLinus Torvalds */ 13547e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 13551da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 13561da177e4SLinus Torvalds { 13571da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1358d16dfc55SPeter Zijlstra struct mmu_gather tlb; 13597e027b14SLinus Torvalds unsigned long end = start + size; 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds lru_add_drain(); 13622b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 1363365e9c87SHugh Dickins update_hiwater_rss(mm); 13647e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 13657e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 13664f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 13677e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 13687e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 13691da177e4SLinus Torvalds } 13701da177e4SLinus Torvalds 1371c627f9ccSJack Steiner /** 1372f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1373f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1374f5cc4eefSAl Viro * @address: starting address of pages to zap 1375f5cc4eefSAl Viro * @size: number of bytes to zap 13768a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1377f5cc4eefSAl Viro * 1378f5cc4eefSAl Viro * The range must fit into one VMA. 1379f5cc4eefSAl Viro */ 1380f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1381f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1382f5cc4eefSAl Viro { 1383f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1384f5cc4eefSAl Viro struct mmu_gather tlb; 1385f5cc4eefSAl Viro unsigned long end = address + size; 1386f5cc4eefSAl Viro 1387f5cc4eefSAl Viro lru_add_drain(); 13882b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, address, end); 1389f5cc4eefSAl Viro update_hiwater_rss(mm); 1390f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 13914f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1392f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1393f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds 1396c627f9ccSJack Steiner /** 1397c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1398c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1399c627f9ccSJack Steiner * @address: starting address of pages to zap 1400c627f9ccSJack Steiner * @size: number of bytes to zap 1401c627f9ccSJack Steiner * 1402c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1403c627f9ccSJack Steiner * 1404c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1405c627f9ccSJack Steiner * 1406c627f9ccSJack Steiner * Returns 0 if successful. 1407c627f9ccSJack Steiner */ 1408c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1409c627f9ccSJack Steiner unsigned long size) 1410c627f9ccSJack Steiner { 1411c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1412c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1413c627f9ccSJack Steiner return -1; 1414f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1415c627f9ccSJack Steiner return 0; 1416c627f9ccSJack Steiner } 1417c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1418c627f9ccSJack Steiner 141925ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 1420920c7a5dSHarvey Harrison spinlock_t **ptl) 1421c9cfcddfSLinus Torvalds { 1422c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1423c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1424c9cfcddfSLinus Torvalds if (pud) { 142549c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1426f66055abSAndrea Arcangeli if (pmd) { 1427f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 1428c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1429c9cfcddfSLinus Torvalds } 1430f66055abSAndrea Arcangeli } 1431c9cfcddfSLinus Torvalds return NULL; 1432c9cfcddfSLinus Torvalds } 1433c9cfcddfSLinus Torvalds 14341da177e4SLinus Torvalds /* 1435238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1436238f58d8SLinus Torvalds * 1437238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1438238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1439238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1440238f58d8SLinus Torvalds */ 1441423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1442423bad60SNick Piggin struct page *page, pgprot_t prot) 1443238f58d8SLinus Torvalds { 1444423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1445238f58d8SLinus Torvalds int retval; 1446238f58d8SLinus Torvalds pte_t *pte; 1447238f58d8SLinus Torvalds spinlock_t *ptl; 1448238f58d8SLinus Torvalds 1449238f58d8SLinus Torvalds retval = -EINVAL; 1450a145dd41SLinus Torvalds if (PageAnon(page)) 14515b4e655eSKAMEZAWA Hiroyuki goto out; 1452238f58d8SLinus Torvalds retval = -ENOMEM; 1453238f58d8SLinus Torvalds flush_dcache_page(page); 1454c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1455238f58d8SLinus Torvalds if (!pte) 14565b4e655eSKAMEZAWA Hiroyuki goto out; 1457238f58d8SLinus Torvalds retval = -EBUSY; 1458238f58d8SLinus Torvalds if (!pte_none(*pte)) 1459238f58d8SLinus Torvalds goto out_unlock; 1460238f58d8SLinus Torvalds 1461238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1462238f58d8SLinus Torvalds get_page(page); 146334e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 1464238f58d8SLinus Torvalds page_add_file_rmap(page); 1465238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1466238f58d8SLinus Torvalds 1467238f58d8SLinus Torvalds retval = 0; 14688a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 14698a9f3ccdSBalbir Singh return retval; 1470238f58d8SLinus Torvalds out_unlock: 1471238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1472238f58d8SLinus Torvalds out: 1473238f58d8SLinus Torvalds return retval; 1474238f58d8SLinus Torvalds } 1475238f58d8SLinus Torvalds 1476bfa5bf6dSRolf Eike Beer /** 1477bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1478bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1479bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1480bfa5bf6dSRolf Eike Beer * @page: source kernel page 1481bfa5bf6dSRolf Eike Beer * 1482a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1483a145dd41SLinus Torvalds * into a user vma. 1484a145dd41SLinus Torvalds * 1485a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1486a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1487a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 14888dfcc9baSNick Piggin * (see split_page()). 1489a145dd41SLinus Torvalds * 1490a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1491a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1492a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1493a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1494a145dd41SLinus Torvalds * ask for a shared writable mapping! 1495a145dd41SLinus Torvalds * 1496a145dd41SLinus Torvalds * The page does not need to be reserved. 14974b6e1e37SKonstantin Khlebnikov * 14984b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 14994b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 15004b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 15014b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 1502a145dd41SLinus Torvalds */ 1503423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 1504423bad60SNick Piggin struct page *page) 1505a145dd41SLinus Torvalds { 1506a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1507a145dd41SLinus Torvalds return -EFAULT; 1508a145dd41SLinus Torvalds if (!page_count(page)) 1509a145dd41SLinus Torvalds return -EINVAL; 15104b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 15114b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 15124b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 15134b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 15144b6e1e37SKonstantin Khlebnikov } 1515423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1516a145dd41SLinus Torvalds } 1517e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1518a145dd41SLinus Torvalds 1519423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1520423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 1521423bad60SNick Piggin { 1522423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1523423bad60SNick Piggin int retval; 1524423bad60SNick Piggin pte_t *pte, entry; 1525423bad60SNick Piggin spinlock_t *ptl; 1526423bad60SNick Piggin 1527423bad60SNick Piggin retval = -ENOMEM; 1528423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1529423bad60SNick Piggin if (!pte) 1530423bad60SNick Piggin goto out; 1531423bad60SNick Piggin retval = -EBUSY; 1532423bad60SNick Piggin if (!pte_none(*pte)) 1533423bad60SNick Piggin goto out_unlock; 1534423bad60SNick Piggin 1535423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 1536423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 1537423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 15384b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 1539423bad60SNick Piggin 1540423bad60SNick Piggin retval = 0; 1541423bad60SNick Piggin out_unlock: 1542423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 1543423bad60SNick Piggin out: 1544423bad60SNick Piggin return retval; 1545423bad60SNick Piggin } 1546423bad60SNick Piggin 1547e0dc0d8fSNick Piggin /** 1548e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 1549e0dc0d8fSNick Piggin * @vma: user vma to map to 1550e0dc0d8fSNick Piggin * @addr: target user address of this page 1551e0dc0d8fSNick Piggin * @pfn: source kernel pfn 1552e0dc0d8fSNick Piggin * 1553c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 1554e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 1555e0dc0d8fSNick Piggin * 1556e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 1557e0dc0d8fSNick Piggin * in that case the handler should return NULL. 15580d71d10aSNick Piggin * 15590d71d10aSNick Piggin * vma cannot be a COW mapping. 15600d71d10aSNick Piggin * 15610d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 15620d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 1563e0dc0d8fSNick Piggin */ 1564e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1565e0dc0d8fSNick Piggin unsigned long pfn) 1566e0dc0d8fSNick Piggin { 15672ab64037Svenkatesh.pallipadi@intel.com int ret; 1568e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 15697e675137SNick Piggin /* 15707e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 15717e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 15727e675137SNick Piggin * consistency in testing and feature parity among all, so we should 15737e675137SNick Piggin * try to keep these invariants in place for everybody. 15747e675137SNick Piggin */ 1575b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 1576b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 1577b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 1578b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 1579b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 1580e0dc0d8fSNick Piggin 1581423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1582423bad60SNick Piggin return -EFAULT; 15835180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 15842ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 15852ab64037Svenkatesh.pallipadi@intel.com 1586e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 15872ab64037Svenkatesh.pallipadi@intel.com 15882ab64037Svenkatesh.pallipadi@intel.com return ret; 1589e0dc0d8fSNick Piggin } 1590e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 1591e0dc0d8fSNick Piggin 1592423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 1593423bad60SNick Piggin unsigned long pfn) 1594423bad60SNick Piggin { 1595423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 1596423bad60SNick Piggin 1597423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 1598423bad60SNick Piggin return -EFAULT; 1599423bad60SNick Piggin 1600423bad60SNick Piggin /* 1601423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 1602423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 1603423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 160462eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 160562eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 1606423bad60SNick Piggin */ 1607423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 1608423bad60SNick Piggin struct page *page; 1609423bad60SNick Piggin 1610423bad60SNick Piggin page = pfn_to_page(pfn); 1611423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1612423bad60SNick Piggin } 1613423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 1614423bad60SNick Piggin } 1615423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 1616423bad60SNick Piggin 1617a145dd41SLinus Torvalds /* 16181da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 16191da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 16201da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 16211da177e4SLinus Torvalds */ 16221da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 16231da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16241da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16251da177e4SLinus Torvalds { 16261da177e4SLinus Torvalds pte_t *pte; 1627c74df32cSHugh Dickins spinlock_t *ptl; 16281da177e4SLinus Torvalds 1629c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 16301da177e4SLinus Torvalds if (!pte) 16311da177e4SLinus Torvalds return -ENOMEM; 16326606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 16331da177e4SLinus Torvalds do { 16341da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 16357e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 16361da177e4SLinus Torvalds pfn++; 16371da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 16386606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1639c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 16401da177e4SLinus Torvalds return 0; 16411da177e4SLinus Torvalds } 16421da177e4SLinus Torvalds 16431da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 16441da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16451da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16461da177e4SLinus Torvalds { 16471da177e4SLinus Torvalds pmd_t *pmd; 16481da177e4SLinus Torvalds unsigned long next; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16511da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 16521da177e4SLinus Torvalds if (!pmd) 16531da177e4SLinus Torvalds return -ENOMEM; 1654f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 16551da177e4SLinus Torvalds do { 16561da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 16571da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 16581da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16591da177e4SLinus Torvalds return -ENOMEM; 16601da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 16611da177e4SLinus Torvalds return 0; 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 16651da177e4SLinus Torvalds unsigned long addr, unsigned long end, 16661da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 16671da177e4SLinus Torvalds { 16681da177e4SLinus Torvalds pud_t *pud; 16691da177e4SLinus Torvalds unsigned long next; 16701da177e4SLinus Torvalds 16711da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 16721da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 16731da177e4SLinus Torvalds if (!pud) 16741da177e4SLinus Torvalds return -ENOMEM; 16751da177e4SLinus Torvalds do { 16761da177e4SLinus Torvalds next = pud_addr_end(addr, end); 16771da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 16781da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 16791da177e4SLinus Torvalds return -ENOMEM; 16801da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 16811da177e4SLinus Torvalds return 0; 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds 1684bfa5bf6dSRolf Eike Beer /** 1685bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 1686bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1687bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 1688bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 1689bfa5bf6dSRolf Eike Beer * @size: size of map area 1690bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 1691bfa5bf6dSRolf Eike Beer * 1692bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 1693bfa5bf6dSRolf Eike Beer */ 16941da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 16951da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 16961da177e4SLinus Torvalds { 16971da177e4SLinus Torvalds pgd_t *pgd; 16981da177e4SLinus Torvalds unsigned long next; 16992d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 17001da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 17011da177e4SLinus Torvalds int err; 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds /* 17041da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 17051da177e4SLinus Torvalds * rest of the world about it: 17061da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 17071da177e4SLinus Torvalds * (accesses can have side effects). 17086aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 17096aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 17106aab341eSLinus Torvalds * with them. 1711314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 1712314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 1713314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 1714314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 1715fb155c16SLinus Torvalds * 1716fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 1717fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 1718fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 1719b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 17201da177e4SLinus Torvalds */ 1721b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 1722b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 1723b3b9c293SKonstantin Khlebnikov return -EINVAL; 17246aab341eSLinus Torvalds vma->vm_pgoff = pfn; 1725b3b9c293SKonstantin Khlebnikov } 1726b3b9c293SKonstantin Khlebnikov 1727b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 1728b3b9c293SKonstantin Khlebnikov if (err) 17293c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 1730fb155c16SLinus Torvalds 1731314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 17321da177e4SLinus Torvalds 17331da177e4SLinus Torvalds BUG_ON(addr >= end); 17341da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 17351da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 17361da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 17371da177e4SLinus Torvalds do { 17381da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 17391da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 17401da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 17411da177e4SLinus Torvalds if (err) 17421da177e4SLinus Torvalds break; 17431da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 17442ab64037Svenkatesh.pallipadi@intel.com 17452ab64037Svenkatesh.pallipadi@intel.com if (err) 17465180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 17472ab64037Svenkatesh.pallipadi@intel.com 17481da177e4SLinus Torvalds return err; 17491da177e4SLinus Torvalds } 17501da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 17511da177e4SLinus Torvalds 1752b4cbb197SLinus Torvalds /** 1753b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 1754b4cbb197SLinus Torvalds * @vma: user vma to map to 1755b4cbb197SLinus Torvalds * @start: start of area 1756b4cbb197SLinus Torvalds * @len: size of area 1757b4cbb197SLinus Torvalds * 1758b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 1759b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 1760b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 1761b4cbb197SLinus Torvalds * 1762b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 1763b4cbb197SLinus Torvalds * whatever write-combining details or similar. 1764b4cbb197SLinus Torvalds */ 1765b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 1766b4cbb197SLinus Torvalds { 1767b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 1768b4cbb197SLinus Torvalds 1769b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 1770b4cbb197SLinus Torvalds if (start + len < start) 1771b4cbb197SLinus Torvalds return -EINVAL; 1772b4cbb197SLinus Torvalds /* 1773b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 1774b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 1775b4cbb197SLinus Torvalds * just have smaller alignment. 1776b4cbb197SLinus Torvalds */ 1777b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 1778b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 1779b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 1780b4cbb197SLinus Torvalds if (pfn + pages < pfn) 1781b4cbb197SLinus Torvalds return -EINVAL; 1782b4cbb197SLinus Torvalds 1783b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 1784b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 1785b4cbb197SLinus Torvalds return -EINVAL; 1786b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 1787b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 1788b4cbb197SLinus Torvalds 1789b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 1790b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 1791b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 1792b4cbb197SLinus Torvalds return -EINVAL; 1793b4cbb197SLinus Torvalds 1794b4cbb197SLinus Torvalds /* Ok, let it rip */ 1795b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 1796b4cbb197SLinus Torvalds } 1797b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 1798b4cbb197SLinus Torvalds 1799aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 1800aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1801aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1802aee16b3cSJeremy Fitzhardinge { 1803aee16b3cSJeremy Fitzhardinge pte_t *pte; 1804aee16b3cSJeremy Fitzhardinge int err; 18052f569afdSMartin Schwidefsky pgtable_t token; 180694909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 1807aee16b3cSJeremy Fitzhardinge 1808aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 1809aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 1810aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 1811aee16b3cSJeremy Fitzhardinge if (!pte) 1812aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1813aee16b3cSJeremy Fitzhardinge 1814aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 1815aee16b3cSJeremy Fitzhardinge 181638e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 181738e0edb1SJeremy Fitzhardinge 18182f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 1819aee16b3cSJeremy Fitzhardinge 1820aee16b3cSJeremy Fitzhardinge do { 1821c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 1822aee16b3cSJeremy Fitzhardinge if (err) 1823aee16b3cSJeremy Fitzhardinge break; 1824c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 1825aee16b3cSJeremy Fitzhardinge 182638e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 182738e0edb1SJeremy Fitzhardinge 1828aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 1829aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 1830aee16b3cSJeremy Fitzhardinge return err; 1831aee16b3cSJeremy Fitzhardinge } 1832aee16b3cSJeremy Fitzhardinge 1833aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 1834aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1835aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1836aee16b3cSJeremy Fitzhardinge { 1837aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 1838aee16b3cSJeremy Fitzhardinge unsigned long next; 1839aee16b3cSJeremy Fitzhardinge int err; 1840aee16b3cSJeremy Fitzhardinge 1841ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 1842ceb86879SAndi Kleen 1843aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 1844aee16b3cSJeremy Fitzhardinge if (!pmd) 1845aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1846aee16b3cSJeremy Fitzhardinge do { 1847aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 1848aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 1849aee16b3cSJeremy Fitzhardinge if (err) 1850aee16b3cSJeremy Fitzhardinge break; 1851aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 1852aee16b3cSJeremy Fitzhardinge return err; 1853aee16b3cSJeremy Fitzhardinge } 1854aee16b3cSJeremy Fitzhardinge 1855aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 1856aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1857aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1858aee16b3cSJeremy Fitzhardinge { 1859aee16b3cSJeremy Fitzhardinge pud_t *pud; 1860aee16b3cSJeremy Fitzhardinge unsigned long next; 1861aee16b3cSJeremy Fitzhardinge int err; 1862aee16b3cSJeremy Fitzhardinge 1863aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 1864aee16b3cSJeremy Fitzhardinge if (!pud) 1865aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1866aee16b3cSJeremy Fitzhardinge do { 1867aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 1868aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 1869aee16b3cSJeremy Fitzhardinge if (err) 1870aee16b3cSJeremy Fitzhardinge break; 1871aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 1872aee16b3cSJeremy Fitzhardinge return err; 1873aee16b3cSJeremy Fitzhardinge } 1874aee16b3cSJeremy Fitzhardinge 1875aee16b3cSJeremy Fitzhardinge /* 1876aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 1877aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 1878aee16b3cSJeremy Fitzhardinge */ 1879aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 1880aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 1881aee16b3cSJeremy Fitzhardinge { 1882aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 1883aee16b3cSJeremy Fitzhardinge unsigned long next; 188457250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 1885aee16b3cSJeremy Fitzhardinge int err; 1886aee16b3cSJeremy Fitzhardinge 1887aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 1888aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 1889aee16b3cSJeremy Fitzhardinge do { 1890aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 1891aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 1892aee16b3cSJeremy Fitzhardinge if (err) 1893aee16b3cSJeremy Fitzhardinge break; 1894aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 189557250a5bSJeremy Fitzhardinge 1896aee16b3cSJeremy Fitzhardinge return err; 1897aee16b3cSJeremy Fitzhardinge } 1898aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 1899aee16b3cSJeremy Fitzhardinge 19001da177e4SLinus Torvalds /* 19019b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 19029b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 19039b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 19049b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 19059b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 1906a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 19078f4e2101SHugh Dickins */ 19084c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 19098f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 19108f4e2101SHugh Dickins { 19118f4e2101SHugh Dickins int same = 1; 19128f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 19138f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 19144c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 19154c21e2f2SHugh Dickins spin_lock(ptl); 19168f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 19174c21e2f2SHugh Dickins spin_unlock(ptl); 19188f4e2101SHugh Dickins } 19198f4e2101SHugh Dickins #endif 19208f4e2101SHugh Dickins pte_unmap(page_table); 19218f4e2101SHugh Dickins return same; 19228f4e2101SHugh Dickins } 19238f4e2101SHugh Dickins 19249de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 19256aab341eSLinus Torvalds { 19260abdd7a8SDan Williams debug_dma_assert_idle(src); 19270abdd7a8SDan Williams 19286aab341eSLinus Torvalds /* 19296aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 19306aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 19316aab341eSLinus Torvalds * just copying from the original user address. If that 19326aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 19336aab341eSLinus Torvalds */ 19346aab341eSLinus Torvalds if (unlikely(!src)) { 19359b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 19365d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 19375d2a2dbbSLinus Torvalds 19385d2a2dbbSLinus Torvalds /* 19395d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 19405d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 19415d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 19425d2a2dbbSLinus Torvalds * zeroes. 19435d2a2dbbSLinus Torvalds */ 19445d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 19453ecb01dfSJan Beulich clear_page(kaddr); 19469b04c5feSCong Wang kunmap_atomic(kaddr); 1947c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 19480ed361deSNick Piggin } else 19499de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 19506aab341eSLinus Torvalds } 19516aab341eSLinus Torvalds 19521da177e4SLinus Torvalds /* 1953fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 1954fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 1955fb09a464SKirill A. Shutemov * 1956fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 1957fb09a464SKirill A. Shutemov */ 1958fb09a464SKirill A. Shutemov static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, 1959fb09a464SKirill A. Shutemov unsigned long address) 1960fb09a464SKirill A. Shutemov { 1961fb09a464SKirill A. Shutemov struct vm_fault vmf; 1962fb09a464SKirill A. Shutemov int ret; 1963fb09a464SKirill A. Shutemov 1964fb09a464SKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 1965fb09a464SKirill A. Shutemov vmf.pgoff = page->index; 1966fb09a464SKirill A. Shutemov vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 1967fb09a464SKirill A. Shutemov vmf.page = page; 19682e4cdab0SMatthew Wilcox vmf.cow_page = NULL; 1969fb09a464SKirill A. Shutemov 1970fb09a464SKirill A. Shutemov ret = vma->vm_ops->page_mkwrite(vma, &vmf); 1971fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 1972fb09a464SKirill A. Shutemov return ret; 1973fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 1974fb09a464SKirill A. Shutemov lock_page(page); 1975fb09a464SKirill A. Shutemov if (!page->mapping) { 1976fb09a464SKirill A. Shutemov unlock_page(page); 1977fb09a464SKirill A. Shutemov return 0; /* retry */ 1978fb09a464SKirill A. Shutemov } 1979fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 1980fb09a464SKirill A. Shutemov } else 1981fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 1982fb09a464SKirill A. Shutemov return ret; 1983fb09a464SKirill A. Shutemov } 1984fb09a464SKirill A. Shutemov 1985fb09a464SKirill A. Shutemov /* 1986*4e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 1987*4e047f89SShachar Raindel * 1988*4e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 1989*4e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 1990*4e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 1991*4e047f89SShachar Raindel * any related book-keeping. 1992*4e047f89SShachar Raindel */ 1993*4e047f89SShachar Raindel static inline int wp_page_reuse(struct mm_struct *mm, 1994*4e047f89SShachar Raindel struct vm_area_struct *vma, unsigned long address, 1995*4e047f89SShachar Raindel pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, 1996*4e047f89SShachar Raindel struct page *page, int page_mkwrite, 1997*4e047f89SShachar Raindel int dirty_shared) 1998*4e047f89SShachar Raindel __releases(ptl) 1999*4e047f89SShachar Raindel { 2000*4e047f89SShachar Raindel pte_t entry; 2001*4e047f89SShachar Raindel /* 2002*4e047f89SShachar Raindel * Clear the pages cpupid information as the existing 2003*4e047f89SShachar Raindel * information potentially belongs to a now completely 2004*4e047f89SShachar Raindel * unrelated process. 2005*4e047f89SShachar Raindel */ 2006*4e047f89SShachar Raindel if (page) 2007*4e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 2008*4e047f89SShachar Raindel 2009*4e047f89SShachar Raindel flush_cache_page(vma, address, pte_pfn(orig_pte)); 2010*4e047f89SShachar Raindel entry = pte_mkyoung(orig_pte); 2011*4e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2012*4e047f89SShachar Raindel if (ptep_set_access_flags(vma, address, page_table, entry, 1)) 2013*4e047f89SShachar Raindel update_mmu_cache(vma, address, page_table); 2014*4e047f89SShachar Raindel pte_unmap_unlock(page_table, ptl); 2015*4e047f89SShachar Raindel 2016*4e047f89SShachar Raindel if (dirty_shared) { 2017*4e047f89SShachar Raindel struct address_space *mapping; 2018*4e047f89SShachar Raindel int dirtied; 2019*4e047f89SShachar Raindel 2020*4e047f89SShachar Raindel if (!page_mkwrite) 2021*4e047f89SShachar Raindel lock_page(page); 2022*4e047f89SShachar Raindel 2023*4e047f89SShachar Raindel dirtied = set_page_dirty(page); 2024*4e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 2025*4e047f89SShachar Raindel mapping = page->mapping; 2026*4e047f89SShachar Raindel unlock_page(page); 2027*4e047f89SShachar Raindel page_cache_release(page); 2028*4e047f89SShachar Raindel 2029*4e047f89SShachar Raindel if ((dirtied || page_mkwrite) && mapping) { 2030*4e047f89SShachar Raindel /* 2031*4e047f89SShachar Raindel * Some device drivers do not set page.mapping 2032*4e047f89SShachar Raindel * but still dirty their pages 2033*4e047f89SShachar Raindel */ 2034*4e047f89SShachar Raindel balance_dirty_pages_ratelimited(mapping); 2035*4e047f89SShachar Raindel } 2036*4e047f89SShachar Raindel 2037*4e047f89SShachar Raindel if (!page_mkwrite) 2038*4e047f89SShachar Raindel file_update_time(vma->vm_file); 2039*4e047f89SShachar Raindel } 2040*4e047f89SShachar Raindel 2041*4e047f89SShachar Raindel return VM_FAULT_WRITE; 2042*4e047f89SShachar Raindel } 2043*4e047f89SShachar Raindel 2044*4e047f89SShachar Raindel /* 20451da177e4SLinus Torvalds * This routine handles present pages, when users try to write 20461da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 20471da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 20481da177e4SLinus Torvalds * 20491da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 20501da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 20511da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 20521da177e4SLinus Torvalds * COW. 20531da177e4SLinus Torvalds * 20541da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 20551da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 20561da177e4SLinus Torvalds * and potentially makes it more efficient. 20571da177e4SLinus Torvalds * 20588f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 20598f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 20608f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 20611da177e4SLinus Torvalds */ 20621da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 206365500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 20648f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2065e6219ec8SNamhyung Kim __releases(ptl) 20661da177e4SLinus Torvalds { 20672ec74c3eSSagi Grimberg struct page *old_page, *new_page = NULL; 20681da177e4SLinus Torvalds pte_t entry; 2069b009c024SMichel Lespinasse int ret = 0; 20701756954cSDavid Rientjes unsigned long mmun_start = 0; /* For mmu_notifiers */ 20711756954cSDavid Rientjes unsigned long mmun_end = 0; /* For mmu_notifiers */ 207200501b53SJohannes Weiner struct mem_cgroup *memcg; 20731da177e4SLinus Torvalds 20746aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2075251b97f5SPeter Zijlstra if (!old_page) { 2076251b97f5SPeter Zijlstra /* 207764e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 207864e45507SPeter Feiner * VM_PFNMAP VMA. 2079251b97f5SPeter Zijlstra * 2080251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2081251b97f5SPeter Zijlstra * Just mark the pages writable as we can't do any dirty 2082251b97f5SPeter Zijlstra * accounting on raw pfn maps. 2083251b97f5SPeter Zijlstra */ 2084251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2085251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2086*4e047f89SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 2087*4e047f89SShachar Raindel orig_pte, old_page, 0, 0); 2088920fc356SHugh Dickins goto gotten; 2089251b97f5SPeter Zijlstra } 20901da177e4SLinus Torvalds 2091d08b3851SPeter Zijlstra /* 2092ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2093ee6a6457SPeter Zijlstra * not dirty accountable. 2094d08b3851SPeter Zijlstra */ 20959a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2096ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2097ab967d86SHugh Dickins page_cache_get(old_page); 2098ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2099ab967d86SHugh Dickins lock_page(old_page); 2100ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2101ab967d86SHugh Dickins &ptl); 2102ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2103ab967d86SHugh Dickins unlock_page(old_page); 2104ab967d86SHugh Dickins goto unlock; 2105ab967d86SHugh Dickins } 2106ab967d86SHugh Dickins page_cache_release(old_page); 2107ab967d86SHugh Dickins } 2108b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2109c44b6743SRik van Riel /* 2110c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2111c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2112c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2113c44b6743SRik van Riel */ 2114c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2115ee6a6457SPeter Zijlstra unlock_page(old_page); 2116*4e047f89SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 2117*4e047f89SShachar Raindel orig_pte, old_page, 0, 0); 2118b009c024SMichel Lespinasse } 2119b009c024SMichel Lespinasse unlock_page(old_page); 2120ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2121d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 2122*4e047f89SShachar Raindel int page_mkwrite = 0; 2123*4e047f89SShachar Raindel 2124f38b4b31SJohannes Weiner page_cache_get(old_page); 2125*4e047f89SShachar Raindel 2126ee6a6457SPeter Zijlstra /* 2127ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 2128ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 2129ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 2130ee6a6457SPeter Zijlstra */ 21319637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 2132c2ec175cSNick Piggin int tmp; 2133f38b4b31SJohannes Weiner 21349637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 2135fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, old_page, address); 2136fb09a464SKirill A. Shutemov if (unlikely(!tmp || (tmp & 2137fb09a464SKirill A. Shutemov (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 2138fb09a464SKirill A. Shutemov page_cache_release(old_page); 2139fb09a464SKirill A. Shutemov return tmp; 2140c2ec175cSNick Piggin } 21419637a5efSDavid Howells /* 21429637a5efSDavid Howells * Since we dropped the lock we need to revalidate 21439637a5efSDavid Howells * the PTE as someone else may have changed it. If 21449637a5efSDavid Howells * they did, we just return, as we can count on the 21459637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 21469637a5efSDavid Howells */ 21479637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 21489637a5efSDavid Howells &ptl); 2149b827e496SNick Piggin if (!pte_same(*page_table, orig_pte)) { 2150b827e496SNick Piggin unlock_page(old_page); 21519637a5efSDavid Howells goto unlock; 2152b827e496SNick Piggin } 2153a200ee18SPeter Zijlstra page_mkwrite = 1; 21549637a5efSDavid Howells } 2155f38b4b31SJohannes Weiner 2156*4e047f89SShachar Raindel return wp_page_reuse(mm, vma, address, page_table, ptl, 2157*4e047f89SShachar Raindel orig_pte, old_page, page_mkwrite, 1); 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds 21601da177e4SLinus Torvalds /* 21611da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 21621da177e4SLinus Torvalds */ 21631da177e4SLinus Torvalds page_cache_get(old_page); 2164920fc356SHugh Dickins gotten: 21658f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 21661da177e4SLinus Torvalds 21671da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 216865500d23SHugh Dickins goto oom; 2169a13ea5b7SHugh Dickins 217062eede62SHugh Dickins if (is_zero_pfn(pte_pfn(orig_pte))) { 2171a13ea5b7SHugh Dickins new_page = alloc_zeroed_user_highpage_movable(vma, address); 2172a13ea5b7SHugh Dickins if (!new_page) 2173a13ea5b7SHugh Dickins goto oom; 2174a13ea5b7SHugh Dickins } else { 2175769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 21761da177e4SLinus Torvalds if (!new_page) 217765500d23SHugh Dickins goto oom; 2178a13ea5b7SHugh Dickins cow_user_page(new_page, old_page, address, vma); 2179a13ea5b7SHugh Dickins } 2180a13ea5b7SHugh Dickins __SetPageUptodate(new_page); 2181a13ea5b7SHugh Dickins 218200501b53SJohannes Weiner if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) 21838a9f3ccdSBalbir Singh goto oom_free_new; 21848a9f3ccdSBalbir Singh 21856bdb913fSHaggai Eran mmun_start = address & PAGE_MASK; 21861756954cSDavid Rientjes mmun_end = mmun_start + PAGE_SIZE; 21876bdb913fSHaggai Eran mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 21886bdb913fSHaggai Eran 21891da177e4SLinus Torvalds /* 21901da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 21911da177e4SLinus Torvalds */ 21928f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 219365500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 2194920fc356SHugh Dickins if (old_page) { 21954294621fSHugh Dickins if (!PageAnon(old_page)) { 219634e55232SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_FILEPAGES); 219734e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 21984294621fSHugh Dickins } 2199920fc356SHugh Dickins } else 220034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2201eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 220265500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 220365500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 22044ce072f1SSiddha, Suresh B /* 22054ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 22064ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 22074ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 22084ce072f1SSiddha, Suresh B * thread doing COW. 22094ce072f1SSiddha, Suresh B */ 221034ee645eSJoerg Roedel ptep_clear_flush_notify(vma, address, page_table); 22119617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 221200501b53SJohannes Weiner mem_cgroup_commit_charge(new_page, memcg, false); 221300501b53SJohannes Weiner lru_cache_add_active_or_unevictable(new_page, vma); 2214828502d3SIzik Eidus /* 2215828502d3SIzik Eidus * We call the notify macro here because, when using secondary 2216828502d3SIzik Eidus * mmu page tables (such as kvm shadow page tables), we want the 2217828502d3SIzik Eidus * new page to be mapped directly into the secondary page table. 2218828502d3SIzik Eidus */ 2219828502d3SIzik Eidus set_pte_at_notify(mm, address, page_table, entry); 22204b3073e1SRussell King update_mmu_cache(vma, address, page_table); 2221945754a1SNick Piggin if (old_page) { 2222945754a1SNick Piggin /* 2223945754a1SNick Piggin * Only after switching the pte to the new page may 2224945754a1SNick Piggin * we remove the mapcount here. Otherwise another 2225945754a1SNick Piggin * process may come and find the rmap count decremented 2226945754a1SNick Piggin * before the pte is switched to the new page, and 2227945754a1SNick Piggin * "reuse" the old page writing into it while our pte 2228945754a1SNick Piggin * here still points into it and can be read by other 2229945754a1SNick Piggin * threads. 2230945754a1SNick Piggin * 2231945754a1SNick Piggin * The critical issue is to order this 2232945754a1SNick Piggin * page_remove_rmap with the ptp_clear_flush above. 2233945754a1SNick Piggin * Those stores are ordered by (if nothing else,) 2234945754a1SNick Piggin * the barrier present in the atomic_add_negative 2235945754a1SNick Piggin * in page_remove_rmap. 2236945754a1SNick Piggin * 2237945754a1SNick Piggin * Then the TLB flush in ptep_clear_flush ensures that 2238945754a1SNick Piggin * no process can access the old page before the 2239945754a1SNick Piggin * decremented mapcount is visible. And the old page 2240945754a1SNick Piggin * cannot be reused until after the decremented 2241945754a1SNick Piggin * mapcount is visible. So transitively, TLBs to 2242945754a1SNick Piggin * old page will be flushed before it can be reused. 2243945754a1SNick Piggin */ 2244edc315fdSHugh Dickins page_remove_rmap(old_page); 2245945754a1SNick Piggin } 2246945754a1SNick Piggin 22471da177e4SLinus Torvalds /* Free the old page.. */ 22481da177e4SLinus Torvalds new_page = old_page; 2249f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 22508a9f3ccdSBalbir Singh } else 225100501b53SJohannes Weiner mem_cgroup_cancel_charge(new_page, memcg); 22528a9f3ccdSBalbir Singh 22536bdb913fSHaggai Eran if (new_page) 22546bdb913fSHaggai Eran page_cache_release(new_page); 225565500d23SHugh Dickins unlock: 22568f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 22571756954cSDavid Rientjes if (mmun_end > mmun_start) 22586bdb913fSHaggai Eran mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 2259e15f8c01SMichel Lespinasse if (old_page) { 2260e15f8c01SMichel Lespinasse /* 2261e15f8c01SMichel Lespinasse * Don't let another task, with possibly unlocked vma, 2262e15f8c01SMichel Lespinasse * keep the mlocked page. 2263e15f8c01SMichel Lespinasse */ 2264e15f8c01SMichel Lespinasse if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { 2265e15f8c01SMichel Lespinasse lock_page(old_page); /* LRU manipulation */ 2266e15f8c01SMichel Lespinasse munlock_vma_page(old_page); 2267e15f8c01SMichel Lespinasse unlock_page(old_page); 2268e15f8c01SMichel Lespinasse } 2269e15f8c01SMichel Lespinasse page_cache_release(old_page); 2270e15f8c01SMichel Lespinasse } 2271f33ea7f4SNick Piggin return ret; 22728a9f3ccdSBalbir Singh oom_free_new: 22736dbf6d3bSHugh Dickins page_cache_release(new_page); 227465500d23SHugh Dickins oom: 227566521d5aSDominik Dingel if (old_page) 22761da177e4SLinus Torvalds page_cache_release(old_page); 22771da177e4SLinus Torvalds return VM_FAULT_OOM; 22781da177e4SLinus Torvalds } 22791da177e4SLinus Torvalds 228097a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 22811da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 22821da177e4SLinus Torvalds struct zap_details *details) 22831da177e4SLinus Torvalds { 2284f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 22851da177e4SLinus Torvalds } 22861da177e4SLinus Torvalds 22876b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 22881da177e4SLinus Torvalds struct zap_details *details) 22891da177e4SLinus Torvalds { 22901da177e4SLinus Torvalds struct vm_area_struct *vma; 22911da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 22921da177e4SLinus Torvalds 22936b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 22941da177e4SLinus Torvalds details->first_index, details->last_index) { 22951da177e4SLinus Torvalds 22961da177e4SLinus Torvalds vba = vma->vm_pgoff; 2297d6e93217SLibin vea = vba + vma_pages(vma) - 1; 22981da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 22991da177e4SLinus Torvalds zba = details->first_index; 23001da177e4SLinus Torvalds if (zba < vba) 23011da177e4SLinus Torvalds zba = vba; 23021da177e4SLinus Torvalds zea = details->last_index; 23031da177e4SLinus Torvalds if (zea > vea) 23041da177e4SLinus Torvalds zea = vea; 23051da177e4SLinus Torvalds 230697a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 23071da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 23081da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 230997a89413SPeter Zijlstra details); 23101da177e4SLinus Torvalds } 23111da177e4SLinus Torvalds } 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds /** 23148a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 23158a5f14a2SKirill A. Shutemov * address_space corresponding to the specified page range in the underlying 23168a5f14a2SKirill A. Shutemov * file. 23178a5f14a2SKirill A. Shutemov * 23183d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 23191da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 23201da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 232125d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 23221da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 23231da177e4SLinus Torvalds * partial pages. 23241da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 23251da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 23261da177e4SLinus Torvalds * end of the file. 23271da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 23281da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 23291da177e4SLinus Torvalds */ 23301da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 23311da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 23321da177e4SLinus Torvalds { 23331da177e4SLinus Torvalds struct zap_details details; 23341da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 23351da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds /* Check for overflow. */ 23381da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 23391da177e4SLinus Torvalds long long holeend = 23401da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 23411da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 23421da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 23431da177e4SLinus Torvalds } 23441da177e4SLinus Torvalds 23451da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 23461da177e4SLinus Torvalds details.first_index = hba; 23471da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 23481da177e4SLinus Torvalds if (details.last_index < details.first_index) 23491da177e4SLinus Torvalds details.last_index = ULONG_MAX; 23501da177e4SLinus Torvalds 235197a89413SPeter Zijlstra 2352283307c7SMatthew Wilcox /* DAX uses i_mmap_lock to serialise file truncate vs page fault */ 235348ec833bSKirill A. Shutemov i_mmap_lock_write(mapping); 23546b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 23551da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 235648ec833bSKirill A. Shutemov i_mmap_unlock_write(mapping); 23571da177e4SLinus Torvalds } 23581da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 23591da177e4SLinus Torvalds 23601da177e4SLinus Torvalds /* 23618f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 23628f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 23639a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 23649a95f3cfSPaul Cassella * 23659a95f3cfSPaul Cassella * We return with the mmap_sem locked or unlocked in the same cases 23669a95f3cfSPaul Cassella * as does filemap_fault(). 23671da177e4SLinus Torvalds */ 236865500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 236965500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 237030c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 23711da177e4SLinus Torvalds { 23728f4e2101SHugh Dickins spinlock_t *ptl; 237356f31801SHugh Dickins struct page *page, *swapcache; 237400501b53SJohannes Weiner struct mem_cgroup *memcg; 237565500d23SHugh Dickins swp_entry_t entry; 23761da177e4SLinus Torvalds pte_t pte; 2377d065bd81SMichel Lespinasse int locked; 2378ad8c2ee8SRik van Riel int exclusive = 0; 237983c54070SNick Piggin int ret = 0; 23801da177e4SLinus Torvalds 23814c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 23828f4e2101SHugh Dickins goto out; 238365500d23SHugh Dickins 238465500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 2385d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 23860697212aSChristoph Lameter if (is_migration_entry(entry)) { 23870697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 2388d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 2389d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2390d1737fdbSAndi Kleen } else { 2391d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 2392d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 2393d1737fdbSAndi Kleen } 23940697212aSChristoph Lameter goto out; 23950697212aSChristoph Lameter } 23960ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 23971da177e4SLinus Torvalds page = lookup_swap_cache(entry); 23981da177e4SLinus Torvalds if (!page) { 239902098feaSHugh Dickins page = swapin_readahead(entry, 240002098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 24011da177e4SLinus Torvalds if (!page) { 24021da177e4SLinus Torvalds /* 24038f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 24048f4e2101SHugh Dickins * while we released the pte lock. 24051da177e4SLinus Torvalds */ 24068f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 24071da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 24081da177e4SLinus Torvalds ret = VM_FAULT_OOM; 24090ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 241065500d23SHugh Dickins goto unlock; 24111da177e4SLinus Torvalds } 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 24141da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2415f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 2416456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 2417d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 241871f72525SWu Fengguang /* 241971f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 242071f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 242171f72525SWu Fengguang */ 2422d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 2423d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 242456f31801SHugh Dickins swapcache = page; 24254779cb31SAndi Kleen goto out_release; 24261da177e4SLinus Torvalds } 24271da177e4SLinus Torvalds 242856f31801SHugh Dickins swapcache = page; 2429d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 2430e709ffd6SRik van Riel 243120a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 2432d065bd81SMichel Lespinasse if (!locked) { 2433d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 2434d065bd81SMichel Lespinasse goto out_release; 2435d065bd81SMichel Lespinasse } 24361da177e4SLinus Torvalds 24374969c119SAndrea Arcangeli /* 243831c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 243931c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 244031c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 244131c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 24424969c119SAndrea Arcangeli */ 244331c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 24444969c119SAndrea Arcangeli goto out_page; 24454969c119SAndrea Arcangeli 2446cbf86cfeSHugh Dickins page = ksm_might_need_to_copy(page, vma, address); 24474969c119SAndrea Arcangeli if (unlikely(!page)) { 24485ad64688SHugh Dickins ret = VM_FAULT_OOM; 24494969c119SAndrea Arcangeli page = swapcache; 24504969c119SAndrea Arcangeli goto out_page; 24514969c119SAndrea Arcangeli } 24525ad64688SHugh Dickins 245300501b53SJohannes Weiner if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) { 2454073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 2455bc43f75cSJohannes Weiner goto out_page; 2456073e587eSKAMEZAWA Hiroyuki } 2457073e587eSKAMEZAWA Hiroyuki 24581da177e4SLinus Torvalds /* 24598f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 24601da177e4SLinus Torvalds */ 24618f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 24629e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 2463b8107480SKirill Korotaev goto out_nomap; 2464b8107480SKirill Korotaev 2465b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 2466b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 2467b8107480SKirill Korotaev goto out_nomap; 24681da177e4SLinus Torvalds } 24691da177e4SLinus Torvalds 24708c7c6e34SKAMEZAWA Hiroyuki /* 24718c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 24728c7c6e34SKAMEZAWA Hiroyuki * 24738c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 24748c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 24758c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 24768c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 24778c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 24788c7c6e34SKAMEZAWA Hiroyuki */ 24791da177e4SLinus Torvalds 248034e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2481b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 24821da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 248330c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 24841da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 248530c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 24869a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 2487ad8c2ee8SRik van Riel exclusive = 1; 24881da177e4SLinus Torvalds } 24891da177e4SLinus Torvalds flush_icache_page(vma, page); 2490179ef71cSCyrill Gorcunov if (pte_swp_soft_dirty(orig_pte)) 2491179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 24921da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 249300501b53SJohannes Weiner if (page == swapcache) { 2494ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 249500501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, true); 249600501b53SJohannes Weiner } else { /* ksm created a completely new copy */ 249756f31801SHugh Dickins page_add_new_anon_rmap(page, vma, address); 249800501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, false); 249900501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 250000501b53SJohannes Weiner } 25011da177e4SLinus Torvalds 2502c475a8abSHugh Dickins swap_free(entry); 2503b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 2504a2c43eedSHugh Dickins try_to_free_swap(page); 2505c475a8abSHugh Dickins unlock_page(page); 250656f31801SHugh Dickins if (page != swapcache) { 25074969c119SAndrea Arcangeli /* 25084969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 25094969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 25104969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 25114969c119SAndrea Arcangeli * further safety release the lock after the swap_free 25124969c119SAndrea Arcangeli * so that the swap count won't change under a 25134969c119SAndrea Arcangeli * parallel locked swapcache. 25144969c119SAndrea Arcangeli */ 25154969c119SAndrea Arcangeli unlock_page(swapcache); 25164969c119SAndrea Arcangeli page_cache_release(swapcache); 25174969c119SAndrea Arcangeli } 2518c475a8abSHugh Dickins 251930c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 252061469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 252161469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 252261469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 25231da177e4SLinus Torvalds goto out; 25241da177e4SLinus Torvalds } 25251da177e4SLinus Torvalds 25261da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 25274b3073e1SRussell King update_mmu_cache(vma, address, page_table); 252865500d23SHugh Dickins unlock: 25298f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 25301da177e4SLinus Torvalds out: 25311da177e4SLinus Torvalds return ret; 2532b8107480SKirill Korotaev out_nomap: 253300501b53SJohannes Weiner mem_cgroup_cancel_charge(page, memcg); 25348f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2535bc43f75cSJohannes Weiner out_page: 2536b8107480SKirill Korotaev unlock_page(page); 25374779cb31SAndi Kleen out_release: 2538b8107480SKirill Korotaev page_cache_release(page); 253956f31801SHugh Dickins if (page != swapcache) { 25404969c119SAndrea Arcangeli unlock_page(swapcache); 25414969c119SAndrea Arcangeli page_cache_release(swapcache); 25424969c119SAndrea Arcangeli } 254365500d23SHugh Dickins return ret; 25441da177e4SLinus Torvalds } 25451da177e4SLinus Torvalds 25461da177e4SLinus Torvalds /* 25478ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 25488ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 2549320b2b8dSLinus Torvalds * doesn't hit another vma. 2550320b2b8dSLinus Torvalds */ 2551320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 2552320b2b8dSLinus Torvalds { 2553320b2b8dSLinus Torvalds address &= PAGE_MASK; 2554320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 25550e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 2556320b2b8dSLinus Torvalds 25570e8e50e2SLinus Torvalds /* 25580e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 25590e8e50e2SLinus Torvalds * 25600e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 25610e8e50e2SLinus Torvalds * that has gotten split.. 25620e8e50e2SLinus Torvalds */ 25630e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 25640e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 25650e8e50e2SLinus Torvalds 2566fee7e49dSLinus Torvalds return expand_downwards(vma, address - PAGE_SIZE); 2567320b2b8dSLinus Torvalds } 25688ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 25698ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 25708ca3eb08SLuck, Tony 25718ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 25728ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 25738ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 25748ca3eb08SLuck, Tony 2575fee7e49dSLinus Torvalds return expand_upwards(vma, address + PAGE_SIZE); 25768ca3eb08SLuck, Tony } 2577320b2b8dSLinus Torvalds return 0; 2578320b2b8dSLinus Torvalds } 2579320b2b8dSLinus Torvalds 2580320b2b8dSLinus Torvalds /* 25818f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 25828f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 25838f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 25841da177e4SLinus Torvalds */ 258565500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 258665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 258730c9f3a9SLinus Torvalds unsigned int flags) 25881da177e4SLinus Torvalds { 258900501b53SJohannes Weiner struct mem_cgroup *memcg; 25908f4e2101SHugh Dickins struct page *page; 25918f4e2101SHugh Dickins spinlock_t *ptl; 25921da177e4SLinus Torvalds pte_t entry; 25931da177e4SLinus Torvalds 25945528f913SLinus Torvalds pte_unmap(page_table); 2595320b2b8dSLinus Torvalds 259611ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 259711ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 25989c145c56SLinus Torvalds return VM_FAULT_SIGSEGV; 259911ac5524SLinus Torvalds 260011ac5524SLinus Torvalds /* Use the zero-page for reads */ 2601593befa6SDominik Dingel if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) { 260262eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 260362eede62SHugh Dickins vma->vm_page_prot)); 260411ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2605a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 2606a13ea5b7SHugh Dickins goto unlock; 2607a13ea5b7SHugh Dickins goto setpte; 2608a13ea5b7SHugh Dickins } 2609a13ea5b7SHugh Dickins 26101da177e4SLinus Torvalds /* Allocate our own private page. */ 26111da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 261265500d23SHugh Dickins goto oom; 2613769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 26141da177e4SLinus Torvalds if (!page) 261565500d23SHugh Dickins goto oom; 261652f37629SMinchan Kim /* 261752f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 261852f37629SMinchan Kim * preceeding stores to the page contents become visible before 261952f37629SMinchan Kim * the set_pte_at() write. 262052f37629SMinchan Kim */ 26210ed361deSNick Piggin __SetPageUptodate(page); 26221da177e4SLinus Torvalds 262300501b53SJohannes Weiner if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) 26248a9f3ccdSBalbir Singh goto oom_free_page; 26258a9f3ccdSBalbir Singh 262665500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 26271ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 26281ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 26298f4e2101SHugh Dickins 26308f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 26311c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 26328f4e2101SHugh Dickins goto release; 26339ba69294SHugh Dickins 263434e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 26359617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 263600501b53SJohannes Weiner mem_cgroup_commit_charge(page, memcg, false); 263700501b53SJohannes Weiner lru_cache_add_active_or_unevictable(page, vma); 2638a13ea5b7SHugh Dickins setpte: 263965500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 26401da177e4SLinus Torvalds 26411da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 26424b3073e1SRussell King update_mmu_cache(vma, address, page_table); 264365500d23SHugh Dickins unlock: 26448f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 264583c54070SNick Piggin return 0; 26468f4e2101SHugh Dickins release: 264700501b53SJohannes Weiner mem_cgroup_cancel_charge(page, memcg); 26488f4e2101SHugh Dickins page_cache_release(page); 26498f4e2101SHugh Dickins goto unlock; 26508a9f3ccdSBalbir Singh oom_free_page: 26516dbf6d3bSHugh Dickins page_cache_release(page); 265265500d23SHugh Dickins oom: 26531da177e4SLinus Torvalds return VM_FAULT_OOM; 26541da177e4SLinus Torvalds } 26551da177e4SLinus Torvalds 26569a95f3cfSPaul Cassella /* 26579a95f3cfSPaul Cassella * The mmap_sem must have been held on entry, and may have been 26589a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 26599a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 26609a95f3cfSPaul Cassella */ 26617eae74afSKirill A. Shutemov static int __do_fault(struct vm_area_struct *vma, unsigned long address, 26622e4cdab0SMatthew Wilcox pgoff_t pgoff, unsigned int flags, 26632e4cdab0SMatthew Wilcox struct page *cow_page, struct page **page) 26647eae74afSKirill A. Shutemov { 26657eae74afSKirill A. Shutemov struct vm_fault vmf; 26667eae74afSKirill A. Shutemov int ret; 26677eae74afSKirill A. Shutemov 26687eae74afSKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 26697eae74afSKirill A. Shutemov vmf.pgoff = pgoff; 26707eae74afSKirill A. Shutemov vmf.flags = flags; 26717eae74afSKirill A. Shutemov vmf.page = NULL; 26722e4cdab0SMatthew Wilcox vmf.cow_page = cow_page; 26737eae74afSKirill A. Shutemov 26747eae74afSKirill A. Shutemov ret = vma->vm_ops->fault(vma, &vmf); 26757eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 26767eae74afSKirill A. Shutemov return ret; 26772e4cdab0SMatthew Wilcox if (!vmf.page) 26782e4cdab0SMatthew Wilcox goto out; 26797eae74afSKirill A. Shutemov 26807eae74afSKirill A. Shutemov if (unlikely(PageHWPoison(vmf.page))) { 26817eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 26827eae74afSKirill A. Shutemov unlock_page(vmf.page); 26837eae74afSKirill A. Shutemov page_cache_release(vmf.page); 26847eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 26857eae74afSKirill A. Shutemov } 26867eae74afSKirill A. Shutemov 26877eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 26887eae74afSKirill A. Shutemov lock_page(vmf.page); 26897eae74afSKirill A. Shutemov else 26907eae74afSKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); 26917eae74afSKirill A. Shutemov 26922e4cdab0SMatthew Wilcox out: 26937eae74afSKirill A. Shutemov *page = vmf.page; 26947eae74afSKirill A. Shutemov return ret; 26957eae74afSKirill A. Shutemov } 26967eae74afSKirill A. Shutemov 26978c6e50b0SKirill A. Shutemov /** 26988c6e50b0SKirill A. Shutemov * do_set_pte - setup new PTE entry for given page and add reverse page mapping. 26998c6e50b0SKirill A. Shutemov * 27008c6e50b0SKirill A. Shutemov * @vma: virtual memory area 27018c6e50b0SKirill A. Shutemov * @address: user virtual address 27028c6e50b0SKirill A. Shutemov * @page: page to map 27038c6e50b0SKirill A. Shutemov * @pte: pointer to target page table entry 27048c6e50b0SKirill A. Shutemov * @write: true, if new entry is writable 27058c6e50b0SKirill A. Shutemov * @anon: true, if it's anonymous page 27068c6e50b0SKirill A. Shutemov * 27078c6e50b0SKirill A. Shutemov * Caller must hold page table lock relevant for @pte. 27088c6e50b0SKirill A. Shutemov * 27098c6e50b0SKirill A. Shutemov * Target users are page handler itself and implementations of 27108c6e50b0SKirill A. Shutemov * vm_ops->map_pages. 27118c6e50b0SKirill A. Shutemov */ 27128c6e50b0SKirill A. Shutemov void do_set_pte(struct vm_area_struct *vma, unsigned long address, 27133bb97794SKirill A. Shutemov struct page *page, pte_t *pte, bool write, bool anon) 27143bb97794SKirill A. Shutemov { 27153bb97794SKirill A. Shutemov pte_t entry; 27163bb97794SKirill A. Shutemov 27173bb97794SKirill A. Shutemov flush_icache_page(vma, page); 27183bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 27193bb97794SKirill A. Shutemov if (write) 27203bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 27213bb97794SKirill A. Shutemov if (anon) { 27223bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 27233bb97794SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address); 27243bb97794SKirill A. Shutemov } else { 27253bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); 27263bb97794SKirill A. Shutemov page_add_file_rmap(page); 27273bb97794SKirill A. Shutemov } 27283bb97794SKirill A. Shutemov set_pte_at(vma->vm_mm, address, pte, entry); 27293bb97794SKirill A. Shutemov 27303bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 27313bb97794SKirill A. Shutemov update_mmu_cache(vma, address, pte); 27323bb97794SKirill A. Shutemov } 27333bb97794SKirill A. Shutemov 27343a91053aSKirill A. Shutemov static unsigned long fault_around_bytes __read_mostly = 27353a91053aSKirill A. Shutemov rounddown_pow_of_two(65536); 2736a9b0f861SKirill A. Shutemov 27371592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 2738a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 27391592eef0SKirill A. Shutemov { 2740a9b0f861SKirill A. Shutemov *val = fault_around_bytes; 27411592eef0SKirill A. Shutemov return 0; 27421592eef0SKirill A. Shutemov } 27431592eef0SKirill A. Shutemov 2744b4903d6eSAndrey Ryabinin /* 2745b4903d6eSAndrey Ryabinin * fault_around_pages() and fault_around_mask() expects fault_around_bytes 2746b4903d6eSAndrey Ryabinin * rounded down to nearest page order. It's what do_fault_around() expects to 2747b4903d6eSAndrey Ryabinin * see. 2748b4903d6eSAndrey Ryabinin */ 2749a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 27501592eef0SKirill A. Shutemov { 2751a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 27521592eef0SKirill A. Shutemov return -EINVAL; 2753b4903d6eSAndrey Ryabinin if (val > PAGE_SIZE) 2754b4903d6eSAndrey Ryabinin fault_around_bytes = rounddown_pow_of_two(val); 2755b4903d6eSAndrey Ryabinin else 2756b4903d6eSAndrey Ryabinin fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ 27571592eef0SKirill A. Shutemov return 0; 27581592eef0SKirill A. Shutemov } 2759a9b0f861SKirill A. Shutemov DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, 2760a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 27611592eef0SKirill A. Shutemov 27621592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 27631592eef0SKirill A. Shutemov { 27641592eef0SKirill A. Shutemov void *ret; 27651592eef0SKirill A. Shutemov 2766a9b0f861SKirill A. Shutemov ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, 2767a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 27681592eef0SKirill A. Shutemov if (!ret) 2769a9b0f861SKirill A. Shutemov pr_warn("Failed to create fault_around_bytes in debugfs"); 27701592eef0SKirill A. Shutemov return 0; 27711592eef0SKirill A. Shutemov } 27721592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 27731592eef0SKirill A. Shutemov #endif 27748c6e50b0SKirill A. Shutemov 27751fdb412bSKirill A. Shutemov /* 27761fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 27771fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 27781fdb412bSKirill A. Shutemov * faults to handle. 27791fdb412bSKirill A. Shutemov * 27801fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 27811fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 27821fdb412bSKirill A. Shutemov * 27831fdb412bSKirill A. Shutemov * This function is called with the page table lock taken. In the split ptlock 27841fdb412bSKirill A. Shutemov * case the page table lock only protects only those entries which belong to 27851fdb412bSKirill A. Shutemov * the page table corresponding to the fault address. 27861fdb412bSKirill A. Shutemov * 27871fdb412bSKirill A. Shutemov * This function doesn't cross the VMA boundaries, in order to call map_pages() 27881fdb412bSKirill A. Shutemov * only once. 27891fdb412bSKirill A. Shutemov * 27901fdb412bSKirill A. Shutemov * fault_around_pages() defines how many pages we'll try to map. 27911fdb412bSKirill A. Shutemov * do_fault_around() expects it to return a power of two less than or equal to 27921fdb412bSKirill A. Shutemov * PTRS_PER_PTE. 27931fdb412bSKirill A. Shutemov * 27941fdb412bSKirill A. Shutemov * The virtual address of the area that we map is naturally aligned to the 27951fdb412bSKirill A. Shutemov * fault_around_pages() value (and therefore to page order). This way it's 27961fdb412bSKirill A. Shutemov * easier to guarantee that we don't cross page table boundaries. 27971fdb412bSKirill A. Shutemov */ 27988c6e50b0SKirill A. Shutemov static void do_fault_around(struct vm_area_struct *vma, unsigned long address, 27998c6e50b0SKirill A. Shutemov pte_t *pte, pgoff_t pgoff, unsigned int flags) 28008c6e50b0SKirill A. Shutemov { 2801aecd6f44SKirill A. Shutemov unsigned long start_addr, nr_pages, mask; 28028c6e50b0SKirill A. Shutemov pgoff_t max_pgoff; 28038c6e50b0SKirill A. Shutemov struct vm_fault vmf; 28048c6e50b0SKirill A. Shutemov int off; 28058c6e50b0SKirill A. Shutemov 2806aecd6f44SKirill A. Shutemov nr_pages = ACCESS_ONCE(fault_around_bytes) >> PAGE_SHIFT; 2807aecd6f44SKirill A. Shutemov mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; 2808aecd6f44SKirill A. Shutemov 2809aecd6f44SKirill A. Shutemov start_addr = max(address & mask, vma->vm_start); 28108c6e50b0SKirill A. Shutemov off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 28118c6e50b0SKirill A. Shutemov pte -= off; 28128c6e50b0SKirill A. Shutemov pgoff -= off; 28138c6e50b0SKirill A. Shutemov 28148c6e50b0SKirill A. Shutemov /* 28158c6e50b0SKirill A. Shutemov * max_pgoff is either end of page table or end of vma 2816850e9c69SKirill A. Shutemov * or fault_around_pages() from pgoff, depending what is nearest. 28178c6e50b0SKirill A. Shutemov */ 28188c6e50b0SKirill A. Shutemov max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + 28198c6e50b0SKirill A. Shutemov PTRS_PER_PTE - 1; 28208c6e50b0SKirill A. Shutemov max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, 2821aecd6f44SKirill A. Shutemov pgoff + nr_pages - 1); 28228c6e50b0SKirill A. Shutemov 28238c6e50b0SKirill A. Shutemov /* Check if it makes any sense to call ->map_pages */ 28248c6e50b0SKirill A. Shutemov while (!pte_none(*pte)) { 28258c6e50b0SKirill A. Shutemov if (++pgoff > max_pgoff) 28268c6e50b0SKirill A. Shutemov return; 28278c6e50b0SKirill A. Shutemov start_addr += PAGE_SIZE; 28288c6e50b0SKirill A. Shutemov if (start_addr >= vma->vm_end) 28298c6e50b0SKirill A. Shutemov return; 28308c6e50b0SKirill A. Shutemov pte++; 28318c6e50b0SKirill A. Shutemov } 28328c6e50b0SKirill A. Shutemov 28338c6e50b0SKirill A. Shutemov vmf.virtual_address = (void __user *) start_addr; 28348c6e50b0SKirill A. Shutemov vmf.pte = pte; 28358c6e50b0SKirill A. Shutemov vmf.pgoff = pgoff; 28368c6e50b0SKirill A. Shutemov vmf.max_pgoff = max_pgoff; 28378c6e50b0SKirill A. Shutemov vmf.flags = flags; 28388c6e50b0SKirill A. Shutemov vma->vm_ops->map_pages(vma, &vmf); 28398c6e50b0SKirill A. Shutemov } 28408c6e50b0SKirill A. Shutemov 2841e655fb29SKirill A. Shutemov static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2842e655fb29SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2843e655fb29SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2844e655fb29SKirill A. Shutemov { 2845e655fb29SKirill A. Shutemov struct page *fault_page; 2846e655fb29SKirill A. Shutemov spinlock_t *ptl; 28473bb97794SKirill A. Shutemov pte_t *pte; 28488c6e50b0SKirill A. Shutemov int ret = 0; 28498c6e50b0SKirill A. Shutemov 28508c6e50b0SKirill A. Shutemov /* 28518c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 28528c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 28538c6e50b0SKirill A. Shutemov * something). 28548c6e50b0SKirill A. Shutemov */ 28559b4bdd2fSKirill A. Shutemov if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { 28568c6e50b0SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 28578c6e50b0SKirill A. Shutemov do_fault_around(vma, address, pte, pgoff, flags); 28588c6e50b0SKirill A. Shutemov if (!pte_same(*pte, orig_pte)) 28598c6e50b0SKirill A. Shutemov goto unlock_out; 28608c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 28618c6e50b0SKirill A. Shutemov } 2862e655fb29SKirill A. Shutemov 28632e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 2864e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2865e655fb29SKirill A. Shutemov return ret; 2866e655fb29SKirill A. Shutemov 2867e655fb29SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2868e655fb29SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2869e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2870e655fb29SKirill A. Shutemov unlock_page(fault_page); 2871e655fb29SKirill A. Shutemov page_cache_release(fault_page); 2872e655fb29SKirill A. Shutemov return ret; 2873e655fb29SKirill A. Shutemov } 28743bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, false, false); 2875e655fb29SKirill A. Shutemov unlock_page(fault_page); 28768c6e50b0SKirill A. Shutemov unlock_out: 28778c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2878e655fb29SKirill A. Shutemov return ret; 2879e655fb29SKirill A. Shutemov } 2880e655fb29SKirill A. Shutemov 2881ec47c3b9SKirill A. Shutemov static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, 2882ec47c3b9SKirill A. Shutemov unsigned long address, pmd_t *pmd, 2883ec47c3b9SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 2884ec47c3b9SKirill A. Shutemov { 2885ec47c3b9SKirill A. Shutemov struct page *fault_page, *new_page; 288600501b53SJohannes Weiner struct mem_cgroup *memcg; 2887ec47c3b9SKirill A. Shutemov spinlock_t *ptl; 28883bb97794SKirill A. Shutemov pte_t *pte; 2889ec47c3b9SKirill A. Shutemov int ret; 2890ec47c3b9SKirill A. Shutemov 2891ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 2892ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2893ec47c3b9SKirill A. Shutemov 2894ec47c3b9SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 2895ec47c3b9SKirill A. Shutemov if (!new_page) 2896ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2897ec47c3b9SKirill A. Shutemov 289800501b53SJohannes Weiner if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) { 2899ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 2900ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 2901ec47c3b9SKirill A. Shutemov } 2902ec47c3b9SKirill A. Shutemov 29032e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page); 2904ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2905ec47c3b9SKirill A. Shutemov goto uncharge_out; 2906ec47c3b9SKirill A. Shutemov 29072e4cdab0SMatthew Wilcox if (fault_page) 2908ec47c3b9SKirill A. Shutemov copy_user_highpage(new_page, fault_page, address, vma); 2909ec47c3b9SKirill A. Shutemov __SetPageUptodate(new_page); 2910ec47c3b9SKirill A. Shutemov 2911ec47c3b9SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2912ec47c3b9SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2913ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 29142e4cdab0SMatthew Wilcox if (fault_page) { 2915ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 2916ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 29172e4cdab0SMatthew Wilcox } else { 29182e4cdab0SMatthew Wilcox /* 29192e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 29202e4cdab0SMatthew Wilcox * i_mmap_lock for read to protect against truncate. 29212e4cdab0SMatthew Wilcox */ 29222e4cdab0SMatthew Wilcox i_mmap_unlock_read(vma->vm_file->f_mapping); 29232e4cdab0SMatthew Wilcox } 2924ec47c3b9SKirill A. Shutemov goto uncharge_out; 2925ec47c3b9SKirill A. Shutemov } 29263bb97794SKirill A. Shutemov do_set_pte(vma, address, new_page, pte, true, true); 292700501b53SJohannes Weiner mem_cgroup_commit_charge(new_page, memcg, false); 292800501b53SJohannes Weiner lru_cache_add_active_or_unevictable(new_page, vma); 2929ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 29302e4cdab0SMatthew Wilcox if (fault_page) { 2931ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 2932ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 29332e4cdab0SMatthew Wilcox } else { 29342e4cdab0SMatthew Wilcox /* 29352e4cdab0SMatthew Wilcox * The fault handler has no page to lock, so it holds 29362e4cdab0SMatthew Wilcox * i_mmap_lock for read to protect against truncate. 29372e4cdab0SMatthew Wilcox */ 29382e4cdab0SMatthew Wilcox i_mmap_unlock_read(vma->vm_file->f_mapping); 29392e4cdab0SMatthew Wilcox } 2940ec47c3b9SKirill A. Shutemov return ret; 2941ec47c3b9SKirill A. Shutemov uncharge_out: 294200501b53SJohannes Weiner mem_cgroup_cancel_charge(new_page, memcg); 2943ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 2944ec47c3b9SKirill A. Shutemov return ret; 2945ec47c3b9SKirill A. Shutemov } 2946ec47c3b9SKirill A. Shutemov 2947f0c6d4d2SKirill A. Shutemov static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, 294816abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 294954cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 29501da177e4SLinus Torvalds { 2951f0c6d4d2SKirill A. Shutemov struct page *fault_page; 2952f0c6d4d2SKirill A. Shutemov struct address_space *mapping; 29538f4e2101SHugh Dickins spinlock_t *ptl; 29543bb97794SKirill A. Shutemov pte_t *pte; 2955f0c6d4d2SKirill A. Shutemov int dirtied = 0; 2956f0c6d4d2SKirill A. Shutemov int ret, tmp; 29571d65f86dSKAMEZAWA Hiroyuki 29582e4cdab0SMatthew Wilcox ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); 29597eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 2960f0c6d4d2SKirill A. Shutemov return ret; 29611da177e4SLinus Torvalds 29621da177e4SLinus Torvalds /* 2963f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 2964f0c6d4d2SKirill A. Shutemov * about to become writable 29651da177e4SLinus Torvalds */ 2966fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 2967f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 2968fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, fault_page, address); 2969fb09a464SKirill A. Shutemov if (unlikely(!tmp || 2970fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 2971f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 2972f0c6d4d2SKirill A. Shutemov return tmp; 297369676147SMark Fasheh } 2974d0217ac0SNick Piggin } 2975fb09a464SKirill A. Shutemov 2976f0c6d4d2SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 2977f0c6d4d2SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 2978f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2979f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 2980f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 2981f0c6d4d2SKirill A. Shutemov return ret; 29829637a5efSDavid Howells } 29833bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, true, false); 2984f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 2985d00806b1SNick Piggin 2986f0c6d4d2SKirill A. Shutemov if (set_page_dirty(fault_page)) 298741c4d25fSJan Kara dirtied = 1; 2988d82fa87dSAndrew Morton /* 2989d82fa87dSAndrew Morton * Take a local copy of the address_space - page.mapping may be zeroed 2990d82fa87dSAndrew Morton * by truncate after unlock_page(). The address_space itself remains 2991d82fa87dSAndrew Morton * pinned by vma->vm_file's reference. We rely on unlock_page()'s 2992d82fa87dSAndrew Morton * release semantics to prevent the compiler from undoing this copying. 2993d82fa87dSAndrew Morton */ 2994f0c6d4d2SKirill A. Shutemov mapping = fault_page->mapping; 2995f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 2996f0c6d4d2SKirill A. Shutemov if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { 2997b827e496SNick Piggin /* 2998b827e496SNick Piggin * Some device drivers do not set page.mapping but still 2999b827e496SNick Piggin * dirty their pages 3000b827e496SNick Piggin */ 3001b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3002d08b3851SPeter Zijlstra } 3003d00806b1SNick Piggin 300474ec6751SJohannes Weiner if (!vma->vm_ops->page_mkwrite) 3005b827e496SNick Piggin file_update_time(vma->vm_file); 3006b827e496SNick Piggin 3007b827e496SNick Piggin return ret; 300854cb8821SNick Piggin } 3009d00806b1SNick Piggin 30109a95f3cfSPaul Cassella /* 30119a95f3cfSPaul Cassella * We enter with non-exclusive mmap_sem (to exclude vma changes, 30129a95f3cfSPaul Cassella * but allow concurrent faults). 30139a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 30149a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 30159a95f3cfSPaul Cassella */ 30169b4bdd2fSKirill A. Shutemov static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 301754cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 301830c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 301954cb8821SNick Piggin { 302054cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 30210da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 302254cb8821SNick Piggin 302316abfa08SHugh Dickins pte_unmap(page_table); 3024e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3025e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3026e655fb29SKirill A. Shutemov orig_pte); 3027ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3028ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3029ec47c3b9SKirill A. Shutemov orig_pte); 3030f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 303154cb8821SNick Piggin } 303254cb8821SNick Piggin 3033b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 303404bb2f94SRik van Riel unsigned long addr, int page_nid, 303504bb2f94SRik van Riel int *flags) 30369532fec1SMel Gorman { 30379532fec1SMel Gorman get_page(page); 30389532fec1SMel Gorman 30399532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 304004bb2f94SRik van Riel if (page_nid == numa_node_id()) { 30419532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 304204bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 304304bb2f94SRik van Riel } 30449532fec1SMel Gorman 30459532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 30469532fec1SMel Gorman } 30479532fec1SMel Gorman 3048b19a9939SRashika Kheria static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3049d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3050d10e63f2SMel Gorman { 30514daae3b4SMel Gorman struct page *page = NULL; 3052d10e63f2SMel Gorman spinlock_t *ptl; 30538191acbdSMel Gorman int page_nid = -1; 305490572890SPeter Zijlstra int last_cpupid; 3055cbee9f88SPeter Zijlstra int target_nid; 3056b8593bfdSMel Gorman bool migrated = false; 3057b191f9b1SMel Gorman bool was_writable = pte_write(pte); 30586688cc05SPeter Zijlstra int flags = 0; 3059d10e63f2SMel Gorman 3060c0e7cad9SMel Gorman /* A PROT_NONE fault should not end up here */ 3061c0e7cad9SMel Gorman BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))); 3062c0e7cad9SMel Gorman 3063d10e63f2SMel Gorman /* 3064d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3065d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3066d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3067d10e63f2SMel Gorman * 30684d942466SMel Gorman * We can safely just do a "set_pte_at()", because the old 30694d942466SMel Gorman * page table entry is not accessible, so there would be no 30704d942466SMel Gorman * concurrent hardware modifications to the PTE. 3071d10e63f2SMel Gorman */ 3072d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3073d10e63f2SMel Gorman spin_lock(ptl); 30744daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 30754daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 30764daae3b4SMel Gorman goto out; 30774daae3b4SMel Gorman } 30784daae3b4SMel Gorman 30794d942466SMel Gorman /* Make it present again */ 30804d942466SMel Gorman pte = pte_modify(pte, vma->vm_page_prot); 30814d942466SMel Gorman pte = pte_mkyoung(pte); 3082b191f9b1SMel Gorman if (was_writable) 3083b191f9b1SMel Gorman pte = pte_mkwrite(pte); 3084d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3085d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3086d10e63f2SMel Gorman 3087d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3088d10e63f2SMel Gorman if (!page) { 3089d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3090d10e63f2SMel Gorman return 0; 3091d10e63f2SMel Gorman } 3092d10e63f2SMel Gorman 30936688cc05SPeter Zijlstra /* 3094bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 3095bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 3096bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 3097bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 3098bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 3099bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 31006688cc05SPeter Zijlstra */ 3101bea66fbdSMel Gorman if (!(vma->vm_flags & VM_WRITE)) 31026688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 31036688cc05SPeter Zijlstra 3104dabe1d99SRik van Riel /* 3105dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 3106dabe1d99SRik van Riel * is later used when determining whether to group tasks together 3107dabe1d99SRik van Riel */ 3108dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 3109dabe1d99SRik van Riel flags |= TNF_SHARED; 3110dabe1d99SRik van Riel 311190572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 31128191acbdSMel Gorman page_nid = page_to_nid(page); 311304bb2f94SRik van Riel target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); 3114d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 31154daae3b4SMel Gorman if (target_nid == -1) { 31164daae3b4SMel Gorman put_page(page); 31174daae3b4SMel Gorman goto out; 31184daae3b4SMel Gorman } 31194daae3b4SMel Gorman 31204daae3b4SMel Gorman /* Migrate to the requested node */ 31211bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 31226688cc05SPeter Zijlstra if (migrated) { 31238191acbdSMel Gorman page_nid = target_nid; 31246688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 3125074c2381SMel Gorman } else 3126074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 31274daae3b4SMel Gorman 31284daae3b4SMel Gorman out: 31298191acbdSMel Gorman if (page_nid != -1) 31306688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 3131d10e63f2SMel Gorman return 0; 3132d10e63f2SMel Gorman } 3133d10e63f2SMel Gorman 31341da177e4SLinus Torvalds /* 31351da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 31361da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 31371da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 31381da177e4SLinus Torvalds * 31391da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 31401da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 31411da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 31421da177e4SLinus Torvalds * 3143c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3144c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 31459a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 31469a95f3cfSPaul Cassella * 31479a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 31489a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 31491da177e4SLinus Torvalds */ 3150c0292554SKirill A. Shutemov static int handle_pte_fault(struct mm_struct *mm, 31511da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 315230c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 31531da177e4SLinus Torvalds { 31541da177e4SLinus Torvalds pte_t entry; 31558f4e2101SHugh Dickins spinlock_t *ptl; 31561da177e4SLinus Torvalds 3157e37c6982SChristian Borntraeger /* 3158e37c6982SChristian Borntraeger * some architectures can have larger ptes than wordsize, 3159e37c6982SChristian Borntraeger * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y, 3160e37c6982SChristian Borntraeger * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses. 3161e37c6982SChristian Borntraeger * The code below just needs a consistent view for the ifs and 3162e37c6982SChristian Borntraeger * we later double check anyway with the ptl lock held. So here 3163e37c6982SChristian Borntraeger * a barrier will do. 3164e37c6982SChristian Borntraeger */ 3165e37c6982SChristian Borntraeger entry = *pte; 3166e37c6982SChristian Borntraeger barrier(); 31671da177e4SLinus Torvalds if (!pte_present(entry)) { 316865500d23SHugh Dickins if (pte_none(entry)) { 3169f4b81804SJes Sorensen if (vma->vm_ops) { 31703c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 31719b4bdd2fSKirill A. Shutemov return do_fault(mm, vma, address, pte, 31729b4bdd2fSKirill A. Shutemov pmd, flags, entry); 3173f4b81804SJes Sorensen } 3174f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 317530c9f3a9SLinus Torvalds pte, pmd, flags); 317665500d23SHugh Dickins } 317765500d23SHugh Dickins return do_swap_page(mm, vma, address, 317830c9f3a9SLinus Torvalds pte, pmd, flags, entry); 31791da177e4SLinus Torvalds } 31801da177e4SLinus Torvalds 31818a0516edSMel Gorman if (pte_protnone(entry)) 3182d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3183d10e63f2SMel Gorman 31844c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 31858f4e2101SHugh Dickins spin_lock(ptl); 31868f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 31878f4e2101SHugh Dickins goto unlock; 318830c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 31891da177e4SLinus Torvalds if (!pte_write(entry)) 31908f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 31918f4e2101SHugh Dickins pte, pmd, ptl, entry); 31921da177e4SLinus Torvalds entry = pte_mkdirty(entry); 31931da177e4SLinus Torvalds } 31941da177e4SLinus Torvalds entry = pte_mkyoung(entry); 319530c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 31964b3073e1SRussell King update_mmu_cache(vma, address, pte); 31971a44e149SAndrea Arcangeli } else { 31981a44e149SAndrea Arcangeli /* 31991a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 32001a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 32011a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 32021a44e149SAndrea Arcangeli * with threads. 32031a44e149SAndrea Arcangeli */ 320430c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 320561c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 32061a44e149SAndrea Arcangeli } 32078f4e2101SHugh Dickins unlock: 32088f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 320983c54070SNick Piggin return 0; 32101da177e4SLinus Torvalds } 32111da177e4SLinus Torvalds 32121da177e4SLinus Torvalds /* 32131da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 32149a95f3cfSPaul Cassella * 32159a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 32169a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 32171da177e4SLinus Torvalds */ 3218519e5247SJohannes Weiner static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3219d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 32201da177e4SLinus Torvalds { 32211da177e4SLinus Torvalds pgd_t *pgd; 32221da177e4SLinus Torvalds pud_t *pud; 32231da177e4SLinus Torvalds pmd_t *pmd; 32241da177e4SLinus Torvalds pte_t *pte; 32251da177e4SLinus Torvalds 3226ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 322730c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 32281da177e4SLinus Torvalds 32291da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 32301da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 32311da177e4SLinus Torvalds if (!pud) 3232c74df32cSHugh Dickins return VM_FAULT_OOM; 32331da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 32341da177e4SLinus Torvalds if (!pmd) 3235c74df32cSHugh Dickins return VM_FAULT_OOM; 323671e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 3237c0292554SKirill A. Shutemov int ret = VM_FAULT_FALLBACK; 323871e3aac0SAndrea Arcangeli if (!vma->vm_ops) 3239c0292554SKirill A. Shutemov ret = do_huge_pmd_anonymous_page(mm, vma, address, 324071e3aac0SAndrea Arcangeli pmd, flags); 3241c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 3242c0292554SKirill A. Shutemov return ret; 324371e3aac0SAndrea Arcangeli } else { 324471e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 32451f1d06c3SDavid Rientjes int ret; 32461f1d06c3SDavid Rientjes 324771e3aac0SAndrea Arcangeli barrier(); 324871e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3249a1dd450bSWill Deacon unsigned int dirty = flags & FAULT_FLAG_WRITE; 3250a1dd450bSWill Deacon 3251e53289c0SLinus Torvalds /* 3252e53289c0SLinus Torvalds * If the pmd is splitting, return and retry the 3253e53289c0SLinus Torvalds * the fault. Alternative: wait until the split 3254e53289c0SLinus Torvalds * is done, and goto retry. 3255e53289c0SLinus Torvalds */ 3256e53289c0SLinus Torvalds if (pmd_trans_splitting(orig_pmd)) 3257e53289c0SLinus Torvalds return 0; 3258e53289c0SLinus Torvalds 32598a0516edSMel Gorman if (pmd_protnone(orig_pmd)) 32604daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3261d10e63f2SMel Gorman orig_pmd, pmd); 3262d10e63f2SMel Gorman 32633d59eebcSLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 32641f1d06c3SDavid Rientjes ret = do_huge_pmd_wp_page(mm, vma, address, pmd, 32651f1d06c3SDavid Rientjes orig_pmd); 32669845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 32671f1d06c3SDavid Rientjes return ret; 3268a1dd450bSWill Deacon } else { 3269a1dd450bSWill Deacon huge_pmd_set_accessed(mm, vma, address, pmd, 3270a1dd450bSWill Deacon orig_pmd, dirty); 327171e3aac0SAndrea Arcangeli return 0; 327271e3aac0SAndrea Arcangeli } 327371e3aac0SAndrea Arcangeli } 32749845cbbdSKirill A. Shutemov } 327571e3aac0SAndrea Arcangeli 327671e3aac0SAndrea Arcangeli /* 327771e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 327871e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 327971e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 328071e3aac0SAndrea Arcangeli */ 32814fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 32824fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3283c74df32cSHugh Dickins return VM_FAULT_OOM; 328471e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 328571e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 328671e3aac0SAndrea Arcangeli return 0; 328771e3aac0SAndrea Arcangeli /* 328871e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 328971e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 329071e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 329171e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 329271e3aac0SAndrea Arcangeli */ 329371e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 32941da177e4SLinus Torvalds 329530c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 32961da177e4SLinus Torvalds } 32971da177e4SLinus Torvalds 32989a95f3cfSPaul Cassella /* 32999a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 33009a95f3cfSPaul Cassella * 33019a95f3cfSPaul Cassella * The mmap_sem may have been released depending on flags and our 33029a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 33039a95f3cfSPaul Cassella */ 3304519e5247SJohannes Weiner int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3305519e5247SJohannes Weiner unsigned long address, unsigned int flags) 3306519e5247SJohannes Weiner { 3307519e5247SJohannes Weiner int ret; 3308519e5247SJohannes Weiner 3309519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 3310519e5247SJohannes Weiner 3311519e5247SJohannes Weiner count_vm_event(PGFAULT); 3312519e5247SJohannes Weiner mem_cgroup_count_vm_event(mm, PGFAULT); 3313519e5247SJohannes Weiner 3314519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 3315519e5247SJohannes Weiner check_sync_rss_stat(current); 3316519e5247SJohannes Weiner 3317519e5247SJohannes Weiner /* 3318519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 3319519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 3320519e5247SJohannes Weiner */ 3321519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 332249426420SJohannes Weiner mem_cgroup_oom_enable(); 3323519e5247SJohannes Weiner 3324519e5247SJohannes Weiner ret = __handle_mm_fault(mm, vma, address, flags); 3325519e5247SJohannes Weiner 332649426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 332749426420SJohannes Weiner mem_cgroup_oom_disable(); 332849426420SJohannes Weiner /* 332949426420SJohannes Weiner * The task may have entered a memcg OOM situation but 333049426420SJohannes Weiner * if the allocation error was handled gracefully (no 333149426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 333249426420SJohannes Weiner * Just clean up the OOM state peacefully. 333349426420SJohannes Weiner */ 333449426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 333549426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 333649426420SJohannes Weiner } 33373812c8c8SJohannes Weiner 3338519e5247SJohannes Weiner return ret; 3339519e5247SJohannes Weiner } 3340e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 3341519e5247SJohannes Weiner 33421da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 33431da177e4SLinus Torvalds /* 33441da177e4SLinus Torvalds * Allocate page upper directory. 3345872fec16SHugh Dickins * We've already handled the fast-path in-line. 33461da177e4SLinus Torvalds */ 33471bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 33481da177e4SLinus Torvalds { 3349c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3350c74df32cSHugh Dickins if (!new) 33511bb3630eSHugh Dickins return -ENOMEM; 33521da177e4SLinus Torvalds 3353362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3354362a61adSNick Piggin 3355872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 33561bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 33575e541973SBenjamin Herrenschmidt pud_free(mm, new); 33581bb3630eSHugh Dickins else 33591da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 3360872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 33611bb3630eSHugh Dickins return 0; 33621da177e4SLinus Torvalds } 33631da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 33641da177e4SLinus Torvalds 33651da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 33661da177e4SLinus Torvalds /* 33671da177e4SLinus Torvalds * Allocate page middle directory. 3368872fec16SHugh Dickins * We've already handled the fast-path in-line. 33691da177e4SLinus Torvalds */ 33701bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 33711da177e4SLinus Torvalds { 3372c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 3373c74df32cSHugh Dickins if (!new) 33741bb3630eSHugh Dickins return -ENOMEM; 33751da177e4SLinus Torvalds 3376362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3377362a61adSNick Piggin 3378872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 33791da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 3380dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 3381dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 33821da177e4SLinus Torvalds pud_populate(mm, pud, new); 3383dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 33845e541973SBenjamin Herrenschmidt pmd_free(mm, new); 3385dc6c9a35SKirill A. Shutemov #else 3386dc6c9a35SKirill A. Shutemov if (!pgd_present(*pud)) { 3387dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 33881da177e4SLinus Torvalds pgd_populate(mm, pud, new); 3389dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 3390dc6c9a35SKirill A. Shutemov pmd_free(mm, new); 33911da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 3392872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 33931bb3630eSHugh Dickins return 0; 33941da177e4SLinus Torvalds } 33951da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 33961da177e4SLinus Torvalds 33971b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 3398f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 3399f8ad0f49SJohannes Weiner { 3400f8ad0f49SJohannes Weiner pgd_t *pgd; 3401f8ad0f49SJohannes Weiner pud_t *pud; 3402f8ad0f49SJohannes Weiner pmd_t *pmd; 3403f8ad0f49SJohannes Weiner pte_t *ptep; 3404f8ad0f49SJohannes Weiner 3405f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 3406f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 3407f8ad0f49SJohannes Weiner goto out; 3408f8ad0f49SJohannes Weiner 3409f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 3410f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 3411f8ad0f49SJohannes Weiner goto out; 3412f8ad0f49SJohannes Weiner 3413f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 3414f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 3415f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 3416f8ad0f49SJohannes Weiner goto out; 3417f8ad0f49SJohannes Weiner 3418f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 3419f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 3420f8ad0f49SJohannes Weiner goto out; 3421f8ad0f49SJohannes Weiner 3422f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 3423f8ad0f49SJohannes Weiner if (!ptep) 3424f8ad0f49SJohannes Weiner goto out; 3425f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 3426f8ad0f49SJohannes Weiner goto unlock; 3427f8ad0f49SJohannes Weiner *ptepp = ptep; 3428f8ad0f49SJohannes Weiner return 0; 3429f8ad0f49SJohannes Weiner unlock: 3430f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 3431f8ad0f49SJohannes Weiner out: 3432f8ad0f49SJohannes Weiner return -EINVAL; 3433f8ad0f49SJohannes Weiner } 3434f8ad0f49SJohannes Weiner 34351b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 34361b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 34371b36ba81SNamhyung Kim { 34381b36ba81SNamhyung Kim int res; 34391b36ba81SNamhyung Kim 34401b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 34411b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 34421b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 34431b36ba81SNamhyung Kim return res; 34441b36ba81SNamhyung Kim } 34451b36ba81SNamhyung Kim 34463b6748e2SJohannes Weiner /** 34473b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 34483b6748e2SJohannes Weiner * @vma: memory mapping 34493b6748e2SJohannes Weiner * @address: user virtual address 34503b6748e2SJohannes Weiner * @pfn: location to store found PFN 34513b6748e2SJohannes Weiner * 34523b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 34533b6748e2SJohannes Weiner * 34543b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 34553b6748e2SJohannes Weiner */ 34563b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 34573b6748e2SJohannes Weiner unsigned long *pfn) 34583b6748e2SJohannes Weiner { 34593b6748e2SJohannes Weiner int ret = -EINVAL; 34603b6748e2SJohannes Weiner spinlock_t *ptl; 34613b6748e2SJohannes Weiner pte_t *ptep; 34623b6748e2SJohannes Weiner 34633b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 34643b6748e2SJohannes Weiner return ret; 34653b6748e2SJohannes Weiner 34663b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 34673b6748e2SJohannes Weiner if (ret) 34683b6748e2SJohannes Weiner return ret; 34693b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 34703b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 34713b6748e2SJohannes Weiner return 0; 34723b6748e2SJohannes Weiner } 34733b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 34743b6748e2SJohannes Weiner 347528b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 3476d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 347728b2ee20SRik van Riel unsigned long address, unsigned int flags, 3478d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 347928b2ee20SRik van Riel { 348003668a4dSJohannes Weiner int ret = -EINVAL; 348128b2ee20SRik van Riel pte_t *ptep, pte; 348228b2ee20SRik van Riel spinlock_t *ptl; 348328b2ee20SRik van Riel 3484d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 3485d87fe660Svenkatesh.pallipadi@intel.com goto out; 348628b2ee20SRik van Riel 348703668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 3488d87fe660Svenkatesh.pallipadi@intel.com goto out; 348928b2ee20SRik van Riel pte = *ptep; 349003668a4dSJohannes Weiner 349128b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 349228b2ee20SRik van Riel goto unlock; 349328b2ee20SRik van Riel 349428b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 349503668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 349628b2ee20SRik van Riel 349703668a4dSJohannes Weiner ret = 0; 349828b2ee20SRik van Riel unlock: 349928b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 350028b2ee20SRik van Riel out: 3501d87fe660Svenkatesh.pallipadi@intel.com return ret; 350228b2ee20SRik van Riel } 350328b2ee20SRik van Riel 350428b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 350528b2ee20SRik van Riel void *buf, int len, int write) 350628b2ee20SRik van Riel { 350728b2ee20SRik van Riel resource_size_t phys_addr; 350828b2ee20SRik van Riel unsigned long prot = 0; 35092bc7273bSKOSAKI Motohiro void __iomem *maddr; 351028b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 351128b2ee20SRik van Riel 3512d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 351328b2ee20SRik van Riel return -EINVAL; 351428b2ee20SRik van Riel 35159cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 351628b2ee20SRik van Riel if (write) 351728b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 351828b2ee20SRik van Riel else 351928b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 352028b2ee20SRik van Riel iounmap(maddr); 352128b2ee20SRik van Riel 352228b2ee20SRik van Riel return len; 352328b2ee20SRik van Riel } 35245a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 352528b2ee20SRik van Riel #endif 352628b2ee20SRik van Riel 35270ec76a11SDavid Howells /* 3528206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 3529206cb636SStephen Wilson * given task for page fault accounting. 35300ec76a11SDavid Howells */ 3531206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 3532206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 35330ec76a11SDavid Howells { 35340ec76a11SDavid Howells struct vm_area_struct *vma; 35350ec76a11SDavid Howells void *old_buf = buf; 35360ec76a11SDavid Howells 35370ec76a11SDavid Howells down_read(&mm->mmap_sem); 3538183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 35390ec76a11SDavid Howells while (len) { 35400ec76a11SDavid Howells int bytes, ret, offset; 35410ec76a11SDavid Howells void *maddr; 354228b2ee20SRik van Riel struct page *page = NULL; 35430ec76a11SDavid Howells 35440ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 35450ec76a11SDavid Howells write, 1, &page, &vma); 354628b2ee20SRik van Riel if (ret <= 0) { 3547dbffcd03SRik van Riel #ifndef CONFIG_HAVE_IOREMAP_PROT 3548dbffcd03SRik van Riel break; 3549dbffcd03SRik van Riel #else 355028b2ee20SRik van Riel /* 355128b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 355228b2ee20SRik van Riel * we can access using slightly different code. 355328b2ee20SRik van Riel */ 355428b2ee20SRik van Riel vma = find_vma(mm, addr); 3555fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 35560ec76a11SDavid Howells break; 355728b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 355828b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 355928b2ee20SRik van Riel len, write); 356028b2ee20SRik van Riel if (ret <= 0) 356128b2ee20SRik van Riel break; 356228b2ee20SRik van Riel bytes = ret; 3563dbffcd03SRik van Riel #endif 356428b2ee20SRik van Riel } else { 35650ec76a11SDavid Howells bytes = len; 35660ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 35670ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 35680ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 35690ec76a11SDavid Howells 35700ec76a11SDavid Howells maddr = kmap(page); 35710ec76a11SDavid Howells if (write) { 35720ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 35730ec76a11SDavid Howells maddr + offset, buf, bytes); 35740ec76a11SDavid Howells set_page_dirty_lock(page); 35750ec76a11SDavid Howells } else { 35760ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 35770ec76a11SDavid Howells buf, maddr + offset, bytes); 35780ec76a11SDavid Howells } 35790ec76a11SDavid Howells kunmap(page); 35800ec76a11SDavid Howells page_cache_release(page); 358128b2ee20SRik van Riel } 35820ec76a11SDavid Howells len -= bytes; 35830ec76a11SDavid Howells buf += bytes; 35840ec76a11SDavid Howells addr += bytes; 35850ec76a11SDavid Howells } 35860ec76a11SDavid Howells up_read(&mm->mmap_sem); 35870ec76a11SDavid Howells 35880ec76a11SDavid Howells return buf - old_buf; 35890ec76a11SDavid Howells } 359003252919SAndi Kleen 35915ddd36b9SStephen Wilson /** 3592ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 35935ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 35945ddd36b9SStephen Wilson * @addr: start address to access 35955ddd36b9SStephen Wilson * @buf: source or destination buffer 35965ddd36b9SStephen Wilson * @len: number of bytes to transfer 35975ddd36b9SStephen Wilson * @write: whether the access is a write 35985ddd36b9SStephen Wilson * 35995ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 36005ddd36b9SStephen Wilson */ 36015ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 36025ddd36b9SStephen Wilson void *buf, int len, int write) 36035ddd36b9SStephen Wilson { 36045ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 36055ddd36b9SStephen Wilson } 36065ddd36b9SStephen Wilson 360703252919SAndi Kleen /* 3608206cb636SStephen Wilson * Access another process' address space. 3609206cb636SStephen Wilson * Source/target buffer must be kernel space, 3610206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 3611206cb636SStephen Wilson */ 3612206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 3613206cb636SStephen Wilson void *buf, int len, int write) 3614206cb636SStephen Wilson { 3615206cb636SStephen Wilson struct mm_struct *mm; 3616206cb636SStephen Wilson int ret; 3617206cb636SStephen Wilson 3618206cb636SStephen Wilson mm = get_task_mm(tsk); 3619206cb636SStephen Wilson if (!mm) 3620206cb636SStephen Wilson return 0; 3621206cb636SStephen Wilson 3622206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 3623206cb636SStephen Wilson mmput(mm); 3624206cb636SStephen Wilson 3625206cb636SStephen Wilson return ret; 3626206cb636SStephen Wilson } 3627206cb636SStephen Wilson 362803252919SAndi Kleen /* 362903252919SAndi Kleen * Print the name of a VMA. 363003252919SAndi Kleen */ 363103252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 363203252919SAndi Kleen { 363303252919SAndi Kleen struct mm_struct *mm = current->mm; 363403252919SAndi Kleen struct vm_area_struct *vma; 363503252919SAndi Kleen 3636e8bff74aSIngo Molnar /* 3637e8bff74aSIngo Molnar * Do not print if we are in atomic 3638e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 3639e8bff74aSIngo Molnar */ 3640e8bff74aSIngo Molnar if (preempt_count()) 3641e8bff74aSIngo Molnar return; 3642e8bff74aSIngo Molnar 364303252919SAndi Kleen down_read(&mm->mmap_sem); 364403252919SAndi Kleen vma = find_vma(mm, ip); 364503252919SAndi Kleen if (vma && vma->vm_file) { 364603252919SAndi Kleen struct file *f = vma->vm_file; 364703252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 364803252919SAndi Kleen if (buf) { 36492fbc57c5SAndy Shevchenko char *p; 365003252919SAndi Kleen 3651cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 365203252919SAndi Kleen if (IS_ERR(p)) 365303252919SAndi Kleen p = "?"; 36542fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 365503252919SAndi Kleen vma->vm_start, 365603252919SAndi Kleen vma->vm_end - vma->vm_start); 365703252919SAndi Kleen free_page((unsigned long)buf); 365803252919SAndi Kleen } 365903252919SAndi Kleen } 366051a07e50SJeff Liu up_read(&mm->mmap_sem); 366103252919SAndi Kleen } 36623ee1afa3SNick Piggin 3663662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 36643ee1afa3SNick Piggin void might_fault(void) 36653ee1afa3SNick Piggin { 366695156f00SPeter Zijlstra /* 366795156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 366895156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 366995156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 367095156f00SPeter Zijlstra * below annotations will generate false positives. 367195156f00SPeter Zijlstra */ 367295156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 367395156f00SPeter Zijlstra return; 367495156f00SPeter Zijlstra 36753ee1afa3SNick Piggin /* 36763ee1afa3SNick Piggin * it would be nicer only to annotate paths which are not under 36773ee1afa3SNick Piggin * pagefault_disable, however that requires a larger audit and 36783ee1afa3SNick Piggin * providing helpers like get_user_atomic. 36793ee1afa3SNick Piggin */ 3680662bbcb2SMichael S. Tsirkin if (in_atomic()) 3681662bbcb2SMichael S. Tsirkin return; 3682662bbcb2SMichael S. Tsirkin 3683662bbcb2SMichael S. Tsirkin __might_sleep(__FILE__, __LINE__, 0); 3684662bbcb2SMichael S. Tsirkin 3685662bbcb2SMichael S. Tsirkin if (current->mm) 36863ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 36873ee1afa3SNick Piggin } 36883ee1afa3SNick Piggin EXPORT_SYMBOL(might_fault); 36893ee1afa3SNick Piggin #endif 369047ad8475SAndrea Arcangeli 369147ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 369247ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 369347ad8475SAndrea Arcangeli unsigned long addr, 369447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 369547ad8475SAndrea Arcangeli { 369647ad8475SAndrea Arcangeli int i; 369747ad8475SAndrea Arcangeli struct page *p = page; 369847ad8475SAndrea Arcangeli 369947ad8475SAndrea Arcangeli might_sleep(); 370047ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 370147ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 370247ad8475SAndrea Arcangeli cond_resched(); 370347ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 370447ad8475SAndrea Arcangeli } 370547ad8475SAndrea Arcangeli } 370647ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 370747ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 370847ad8475SAndrea Arcangeli { 370947ad8475SAndrea Arcangeli int i; 371047ad8475SAndrea Arcangeli 371147ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 371247ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 371347ad8475SAndrea Arcangeli return; 371447ad8475SAndrea Arcangeli } 371547ad8475SAndrea Arcangeli 371647ad8475SAndrea Arcangeli might_sleep(); 371747ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 371847ad8475SAndrea Arcangeli cond_resched(); 371947ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 372047ad8475SAndrea Arcangeli } 372147ad8475SAndrea Arcangeli } 372247ad8475SAndrea Arcangeli 372347ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 372447ad8475SAndrea Arcangeli unsigned long addr, 372547ad8475SAndrea Arcangeli struct vm_area_struct *vma, 372647ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 372747ad8475SAndrea Arcangeli { 372847ad8475SAndrea Arcangeli int i; 372947ad8475SAndrea Arcangeli struct page *dst_base = dst; 373047ad8475SAndrea Arcangeli struct page *src_base = src; 373147ad8475SAndrea Arcangeli 373247ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 373347ad8475SAndrea Arcangeli cond_resched(); 373447ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 373547ad8475SAndrea Arcangeli 373647ad8475SAndrea Arcangeli i++; 373747ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 373847ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 373947ad8475SAndrea Arcangeli } 374047ad8475SAndrea Arcangeli } 374147ad8475SAndrea Arcangeli 374247ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 374347ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 374447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 374547ad8475SAndrea Arcangeli { 374647ad8475SAndrea Arcangeli int i; 374747ad8475SAndrea Arcangeli 374847ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 374947ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 375047ad8475SAndrea Arcangeli pages_per_huge_page); 375147ad8475SAndrea Arcangeli return; 375247ad8475SAndrea Arcangeli } 375347ad8475SAndrea Arcangeli 375447ad8475SAndrea Arcangeli might_sleep(); 375547ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 375647ad8475SAndrea Arcangeli cond_resched(); 375747ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 375847ad8475SAndrea Arcangeli } 375947ad8475SAndrea Arcangeli } 376047ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 376149076ec2SKirill A. Shutemov 376240b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 3763b35f1819SKirill A. Shutemov 3764b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 3765b35f1819SKirill A. Shutemov 3766b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 3767b35f1819SKirill A. Shutemov { 3768b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 3769b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 3770b35f1819SKirill A. Shutemov } 3771b35f1819SKirill A. Shutemov 3772539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 377349076ec2SKirill A. Shutemov { 377449076ec2SKirill A. Shutemov spinlock_t *ptl; 377549076ec2SKirill A. Shutemov 3776b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 377749076ec2SKirill A. Shutemov if (!ptl) 377849076ec2SKirill A. Shutemov return false; 3779539edb58SPeter Zijlstra page->ptl = ptl; 378049076ec2SKirill A. Shutemov return true; 378149076ec2SKirill A. Shutemov } 378249076ec2SKirill A. Shutemov 3783539edb58SPeter Zijlstra void ptlock_free(struct page *page) 378449076ec2SKirill A. Shutemov { 3785b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 378649076ec2SKirill A. Shutemov } 378749076ec2SKirill A. Shutemov #endif 3788