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 121a13ea5b7SHugh Dickins /* 122a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 123a13ea5b7SHugh Dickins */ 124a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 125a13ea5b7SHugh Dickins { 126a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 127a13ea5b7SHugh Dickins return 0; 128a13ea5b7SHugh Dickins } 129a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 130a62eaf15SAndi Kleen 131d559db08SKAMEZAWA Hiroyuki 13234e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 13334e55232SKAMEZAWA Hiroyuki 134ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 13534e55232SKAMEZAWA Hiroyuki { 13634e55232SKAMEZAWA Hiroyuki int i; 13734e55232SKAMEZAWA Hiroyuki 13834e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 13905af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 14005af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 14105af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 14234e55232SKAMEZAWA Hiroyuki } 14334e55232SKAMEZAWA Hiroyuki } 14405af2e10SDavid Rientjes current->rss_stat.events = 0; 14534e55232SKAMEZAWA Hiroyuki } 14634e55232SKAMEZAWA Hiroyuki 14734e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 14834e55232SKAMEZAWA Hiroyuki { 14934e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 15034e55232SKAMEZAWA Hiroyuki 15134e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 15234e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 15334e55232SKAMEZAWA Hiroyuki else 15434e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 15534e55232SKAMEZAWA Hiroyuki } 15634e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 15734e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 15834e55232SKAMEZAWA Hiroyuki 15934e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 16034e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 16134e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 16234e55232SKAMEZAWA Hiroyuki { 16334e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 16434e55232SKAMEZAWA Hiroyuki return; 16534e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 166ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 16734e55232SKAMEZAWA Hiroyuki } 1689547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 16934e55232SKAMEZAWA Hiroyuki 17034e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 17134e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 17234e55232SKAMEZAWA Hiroyuki 17334e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 17434e55232SKAMEZAWA Hiroyuki { 17534e55232SKAMEZAWA Hiroyuki } 17634e55232SKAMEZAWA Hiroyuki 1779547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 1789547d01bSPeter Zijlstra 1799547d01bSPeter Zijlstra #ifdef HAVE_GENERIC_MMU_GATHER 1809547d01bSPeter Zijlstra 1819547d01bSPeter Zijlstra static int tlb_next_batch(struct mmu_gather *tlb) 1829547d01bSPeter Zijlstra { 1839547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 1849547d01bSPeter Zijlstra 1859547d01bSPeter Zijlstra batch = tlb->active; 1869547d01bSPeter Zijlstra if (batch->next) { 1879547d01bSPeter Zijlstra tlb->active = batch->next; 1889547d01bSPeter Zijlstra return 1; 1899547d01bSPeter Zijlstra } 1909547d01bSPeter Zijlstra 19153a59fc6SMichal Hocko if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) 19253a59fc6SMichal Hocko return 0; 19353a59fc6SMichal Hocko 1949547d01bSPeter Zijlstra batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 1959547d01bSPeter Zijlstra if (!batch) 1969547d01bSPeter Zijlstra return 0; 1979547d01bSPeter Zijlstra 19853a59fc6SMichal Hocko tlb->batch_count++; 1999547d01bSPeter Zijlstra batch->next = NULL; 2009547d01bSPeter Zijlstra batch->nr = 0; 2019547d01bSPeter Zijlstra batch->max = MAX_GATHER_BATCH; 2029547d01bSPeter Zijlstra 2039547d01bSPeter Zijlstra tlb->active->next = batch; 2049547d01bSPeter Zijlstra tlb->active = batch; 2059547d01bSPeter Zijlstra 2069547d01bSPeter Zijlstra return 1; 2079547d01bSPeter Zijlstra } 2089547d01bSPeter Zijlstra 2099547d01bSPeter Zijlstra /* tlb_gather_mmu 2109547d01bSPeter Zijlstra * Called to initialize an (on-stack) mmu_gather structure for page-table 2119547d01bSPeter Zijlstra * tear-down from @mm. The @fullmm argument is used when @mm is without 2129547d01bSPeter Zijlstra * users and we're going to destroy the full address space (exit/execve). 2139547d01bSPeter Zijlstra */ 2142b047252SLinus Torvalds void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) 2159547d01bSPeter Zijlstra { 2169547d01bSPeter Zijlstra tlb->mm = mm; 2179547d01bSPeter Zijlstra 2182b047252SLinus Torvalds /* Is it from 0 to ~0? */ 2192b047252SLinus Torvalds tlb->fullmm = !(start | (end+1)); 2201de14c3cSDave Hansen tlb->need_flush_all = 0; 2212b047252SLinus Torvalds tlb->start = start; 2222b047252SLinus Torvalds tlb->end = end; 2239547d01bSPeter Zijlstra tlb->need_flush = 0; 2249547d01bSPeter Zijlstra tlb->local.next = NULL; 2259547d01bSPeter Zijlstra tlb->local.nr = 0; 2269547d01bSPeter Zijlstra tlb->local.max = ARRAY_SIZE(tlb->__pages); 2279547d01bSPeter Zijlstra tlb->active = &tlb->local; 22853a59fc6SMichal Hocko tlb->batch_count = 0; 2299547d01bSPeter Zijlstra 2309547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2319547d01bSPeter Zijlstra tlb->batch = NULL; 23234e55232SKAMEZAWA Hiroyuki #endif 2339547d01bSPeter Zijlstra } 2349547d01bSPeter Zijlstra 235*1cf35d47SLinus Torvalds static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) 2369547d01bSPeter Zijlstra { 2379547d01bSPeter Zijlstra tlb->need_flush = 0; 2389547d01bSPeter Zijlstra tlb_flush(tlb); 2399547d01bSPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 2409547d01bSPeter Zijlstra tlb_table_flush(tlb); 2419547d01bSPeter Zijlstra #endif 242*1cf35d47SLinus Torvalds } 243*1cf35d47SLinus Torvalds 244*1cf35d47SLinus Torvalds static void tlb_flush_mmu_free(struct mmu_gather *tlb) 245*1cf35d47SLinus Torvalds { 246*1cf35d47SLinus Torvalds struct mmu_gather_batch *batch; 2479547d01bSPeter Zijlstra 2489547d01bSPeter Zijlstra for (batch = &tlb->local; batch; batch = batch->next) { 2499547d01bSPeter Zijlstra free_pages_and_swap_cache(batch->pages, batch->nr); 2509547d01bSPeter Zijlstra batch->nr = 0; 2519547d01bSPeter Zijlstra } 2529547d01bSPeter Zijlstra tlb->active = &tlb->local; 2539547d01bSPeter Zijlstra } 2549547d01bSPeter Zijlstra 255*1cf35d47SLinus Torvalds void tlb_flush_mmu(struct mmu_gather *tlb) 256*1cf35d47SLinus Torvalds { 257*1cf35d47SLinus Torvalds if (!tlb->need_flush) 258*1cf35d47SLinus Torvalds return; 259*1cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 260*1cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 261*1cf35d47SLinus Torvalds } 262*1cf35d47SLinus Torvalds 2639547d01bSPeter Zijlstra /* tlb_finish_mmu 2649547d01bSPeter Zijlstra * Called at the end of the shootdown operation to free up any resources 2659547d01bSPeter Zijlstra * that were required. 2669547d01bSPeter Zijlstra */ 2679547d01bSPeter Zijlstra void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 2689547d01bSPeter Zijlstra { 2699547d01bSPeter Zijlstra struct mmu_gather_batch *batch, *next; 2709547d01bSPeter Zijlstra 2719547d01bSPeter Zijlstra tlb_flush_mmu(tlb); 2729547d01bSPeter Zijlstra 2739547d01bSPeter Zijlstra /* keep the page table cache within bounds */ 2749547d01bSPeter Zijlstra check_pgt_cache(); 2759547d01bSPeter Zijlstra 2769547d01bSPeter Zijlstra for (batch = tlb->local.next; batch; batch = next) { 2779547d01bSPeter Zijlstra next = batch->next; 2789547d01bSPeter Zijlstra free_pages((unsigned long)batch, 0); 2799547d01bSPeter Zijlstra } 2809547d01bSPeter Zijlstra tlb->local.next = NULL; 2819547d01bSPeter Zijlstra } 2829547d01bSPeter Zijlstra 2839547d01bSPeter Zijlstra /* __tlb_remove_page 2849547d01bSPeter Zijlstra * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while 2859547d01bSPeter Zijlstra * handling the additional races in SMP caused by other CPUs caching valid 2869547d01bSPeter Zijlstra * mappings in their TLBs. Returns the number of free page slots left. 2879547d01bSPeter Zijlstra * When out of page slots we must call tlb_flush_mmu(). 2889547d01bSPeter Zijlstra */ 2899547d01bSPeter Zijlstra int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 2909547d01bSPeter Zijlstra { 2919547d01bSPeter Zijlstra struct mmu_gather_batch *batch; 2929547d01bSPeter Zijlstra 293f21760b1SShaohua Li VM_BUG_ON(!tlb->need_flush); 2949547d01bSPeter Zijlstra 2959547d01bSPeter Zijlstra batch = tlb->active; 2969547d01bSPeter Zijlstra batch->pages[batch->nr++] = page; 2979547d01bSPeter Zijlstra if (batch->nr == batch->max) { 2989547d01bSPeter Zijlstra if (!tlb_next_batch(tlb)) 2999547d01bSPeter Zijlstra return 0; 3000b43c3aaSShaohua Li batch = tlb->active; 3019547d01bSPeter Zijlstra } 302309381feSSasha Levin VM_BUG_ON_PAGE(batch->nr > batch->max, page); 3039547d01bSPeter Zijlstra 3049547d01bSPeter Zijlstra return batch->max - batch->nr; 3059547d01bSPeter Zijlstra } 3069547d01bSPeter Zijlstra 3079547d01bSPeter Zijlstra #endif /* HAVE_GENERIC_MMU_GATHER */ 30834e55232SKAMEZAWA Hiroyuki 30926723911SPeter Zijlstra #ifdef CONFIG_HAVE_RCU_TABLE_FREE 31026723911SPeter Zijlstra 31126723911SPeter Zijlstra /* 31226723911SPeter Zijlstra * See the comment near struct mmu_table_batch. 31326723911SPeter Zijlstra */ 31426723911SPeter Zijlstra 31526723911SPeter Zijlstra static void tlb_remove_table_smp_sync(void *arg) 31626723911SPeter Zijlstra { 31726723911SPeter Zijlstra /* Simply deliver the interrupt */ 31826723911SPeter Zijlstra } 31926723911SPeter Zijlstra 32026723911SPeter Zijlstra static void tlb_remove_table_one(void *table) 32126723911SPeter Zijlstra { 32226723911SPeter Zijlstra /* 32326723911SPeter Zijlstra * This isn't an RCU grace period and hence the page-tables cannot be 32426723911SPeter Zijlstra * assumed to be actually RCU-freed. 32526723911SPeter Zijlstra * 32626723911SPeter Zijlstra * It is however sufficient for software page-table walkers that rely on 32726723911SPeter Zijlstra * IRQ disabling. See the comment near struct mmu_table_batch. 32826723911SPeter Zijlstra */ 32926723911SPeter Zijlstra smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 33026723911SPeter Zijlstra __tlb_remove_table(table); 33126723911SPeter Zijlstra } 33226723911SPeter Zijlstra 33326723911SPeter Zijlstra static void tlb_remove_table_rcu(struct rcu_head *head) 33426723911SPeter Zijlstra { 33526723911SPeter Zijlstra struct mmu_table_batch *batch; 33626723911SPeter Zijlstra int i; 33726723911SPeter Zijlstra 33826723911SPeter Zijlstra batch = container_of(head, struct mmu_table_batch, rcu); 33926723911SPeter Zijlstra 34026723911SPeter Zijlstra for (i = 0; i < batch->nr; i++) 34126723911SPeter Zijlstra __tlb_remove_table(batch->tables[i]); 34226723911SPeter Zijlstra 34326723911SPeter Zijlstra free_page((unsigned long)batch); 34426723911SPeter Zijlstra } 34526723911SPeter Zijlstra 34626723911SPeter Zijlstra void tlb_table_flush(struct mmu_gather *tlb) 34726723911SPeter Zijlstra { 34826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 34926723911SPeter Zijlstra 35026723911SPeter Zijlstra if (*batch) { 35126723911SPeter Zijlstra call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 35226723911SPeter Zijlstra *batch = NULL; 35326723911SPeter Zijlstra } 35426723911SPeter Zijlstra } 35526723911SPeter Zijlstra 35626723911SPeter Zijlstra void tlb_remove_table(struct mmu_gather *tlb, void *table) 35726723911SPeter Zijlstra { 35826723911SPeter Zijlstra struct mmu_table_batch **batch = &tlb->batch; 35926723911SPeter Zijlstra 36026723911SPeter Zijlstra tlb->need_flush = 1; 36126723911SPeter Zijlstra 36226723911SPeter Zijlstra /* 36326723911SPeter Zijlstra * When there's less then two users of this mm there cannot be a 36426723911SPeter Zijlstra * concurrent page-table walk. 36526723911SPeter Zijlstra */ 36626723911SPeter Zijlstra if (atomic_read(&tlb->mm->mm_users) < 2) { 36726723911SPeter Zijlstra __tlb_remove_table(table); 36826723911SPeter Zijlstra return; 36926723911SPeter Zijlstra } 37026723911SPeter Zijlstra 37126723911SPeter Zijlstra if (*batch == NULL) { 37226723911SPeter Zijlstra *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); 37326723911SPeter Zijlstra if (*batch == NULL) { 37426723911SPeter Zijlstra tlb_remove_table_one(table); 37526723911SPeter Zijlstra return; 37626723911SPeter Zijlstra } 37726723911SPeter Zijlstra (*batch)->nr = 0; 37826723911SPeter Zijlstra } 37926723911SPeter Zijlstra (*batch)->tables[(*batch)->nr++] = table; 38026723911SPeter Zijlstra if ((*batch)->nr == MAX_TABLE_BATCH) 38126723911SPeter Zijlstra tlb_table_flush(tlb); 38226723911SPeter Zijlstra } 38326723911SPeter Zijlstra 3849547d01bSPeter Zijlstra #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ 38526723911SPeter Zijlstra 3861da177e4SLinus Torvalds /* 3871da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 3881da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 3891da177e4SLinus Torvalds */ 3909e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 3919e1b32caSBenjamin Herrenschmidt unsigned long addr) 3921da177e4SLinus Torvalds { 3932f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 3941da177e4SLinus Torvalds pmd_clear(pmd); 3959e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 396e1f56c89SKirill A. Shutemov atomic_long_dec(&tlb->mm->nr_ptes); 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds 399e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 400e0da382cSHugh Dickins unsigned long addr, unsigned long end, 401e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4021da177e4SLinus Torvalds { 4031da177e4SLinus Torvalds pmd_t *pmd; 4041da177e4SLinus Torvalds unsigned long next; 405e0da382cSHugh Dickins unsigned long start; 4061da177e4SLinus Torvalds 407e0da382cSHugh Dickins start = addr; 4081da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 4091da177e4SLinus Torvalds do { 4101da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 4111da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 4121da177e4SLinus Torvalds continue; 4139e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 4141da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 4151da177e4SLinus Torvalds 416e0da382cSHugh Dickins start &= PUD_MASK; 417e0da382cSHugh Dickins if (start < floor) 418e0da382cSHugh Dickins return; 419e0da382cSHugh Dickins if (ceiling) { 420e0da382cSHugh Dickins ceiling &= PUD_MASK; 421e0da382cSHugh Dickins if (!ceiling) 422e0da382cSHugh Dickins return; 4231da177e4SLinus Torvalds } 424e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 425e0da382cSHugh Dickins return; 426e0da382cSHugh Dickins 427e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 428e0da382cSHugh Dickins pud_clear(pud); 4299e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds 432e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 433e0da382cSHugh Dickins unsigned long addr, unsigned long end, 434e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4351da177e4SLinus Torvalds { 4361da177e4SLinus Torvalds pud_t *pud; 4371da177e4SLinus Torvalds unsigned long next; 438e0da382cSHugh Dickins unsigned long start; 4391da177e4SLinus Torvalds 440e0da382cSHugh Dickins start = addr; 4411da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 4421da177e4SLinus Torvalds do { 4431da177e4SLinus Torvalds next = pud_addr_end(addr, end); 4441da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 4451da177e4SLinus Torvalds continue; 446e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 4471da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 4481da177e4SLinus Torvalds 449e0da382cSHugh Dickins start &= PGDIR_MASK; 450e0da382cSHugh Dickins if (start < floor) 451e0da382cSHugh Dickins return; 452e0da382cSHugh Dickins if (ceiling) { 453e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 454e0da382cSHugh Dickins if (!ceiling) 455e0da382cSHugh Dickins return; 4561da177e4SLinus Torvalds } 457e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 458e0da382cSHugh Dickins return; 459e0da382cSHugh Dickins 460e0da382cSHugh Dickins pud = pud_offset(pgd, start); 461e0da382cSHugh Dickins pgd_clear(pgd); 4629e1b32caSBenjamin Herrenschmidt pud_free_tlb(tlb, pud, start); 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds 4651da177e4SLinus Torvalds /* 466e0da382cSHugh Dickins * This function frees user-level page tables of a process. 4671da177e4SLinus Torvalds */ 46842b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 469e0da382cSHugh Dickins unsigned long addr, unsigned long end, 470e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 4711da177e4SLinus Torvalds { 4721da177e4SLinus Torvalds pgd_t *pgd; 4731da177e4SLinus Torvalds unsigned long next; 4741da177e4SLinus Torvalds 475e0da382cSHugh Dickins /* 476e0da382cSHugh Dickins * The next few lines have given us lots of grief... 477e0da382cSHugh Dickins * 478e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 479e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 480e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 481e0da382cSHugh Dickins * 482e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 483e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 484e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 485e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 486e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 487e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 488e0da382cSHugh Dickins * that end 0 case should be mythical). 489e0da382cSHugh Dickins * 490e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 491e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 492e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 493e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 494e0da382cSHugh Dickins * 495e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 496e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 497e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 498e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 499e0da382cSHugh Dickins */ 500e0da382cSHugh Dickins 501e0da382cSHugh Dickins addr &= PMD_MASK; 502e0da382cSHugh Dickins if (addr < floor) { 503e0da382cSHugh Dickins addr += PMD_SIZE; 504e0da382cSHugh Dickins if (!addr) 505e0da382cSHugh Dickins return; 506e0da382cSHugh Dickins } 507e0da382cSHugh Dickins if (ceiling) { 508e0da382cSHugh Dickins ceiling &= PMD_MASK; 509e0da382cSHugh Dickins if (!ceiling) 510e0da382cSHugh Dickins return; 511e0da382cSHugh Dickins } 512e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 513e0da382cSHugh Dickins end -= PMD_SIZE; 514e0da382cSHugh Dickins if (addr > end - 1) 515e0da382cSHugh Dickins return; 516e0da382cSHugh Dickins 51742b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 5181da177e4SLinus Torvalds do { 5191da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 5201da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 5211da177e4SLinus Torvalds continue; 52242b77728SJan Beulich free_pud_range(tlb, pgd, addr, next, floor, ceiling); 5231da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 524e0da382cSHugh Dickins } 525e0da382cSHugh Dickins 52642b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 527e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 528e0da382cSHugh Dickins { 529e0da382cSHugh Dickins while (vma) { 530e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 531e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 532e0da382cSHugh Dickins 5338f4f8c16SHugh Dickins /* 53425d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 53525d9e2d1Snpiggin@suse.de * pgtables 5368f4f8c16SHugh Dickins */ 5375beb4930SRik van Riel unlink_anon_vmas(vma); 5388f4f8c16SHugh Dickins unlink_file_vma(vma); 5398f4f8c16SHugh Dickins 5409da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 5413bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 5423bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 5433bf5ee95SHugh Dickins } else { 5443bf5ee95SHugh Dickins /* 5453bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 5463bf5ee95SHugh Dickins */ 5473bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 5484866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 549e0da382cSHugh Dickins vma = next; 550e0da382cSHugh Dickins next = vma->vm_next; 5515beb4930SRik van Riel unlink_anon_vmas(vma); 5528f4f8c16SHugh Dickins unlink_file_vma(vma); 553e0da382cSHugh Dickins } 5543bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 555e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 5563bf5ee95SHugh Dickins } 557e0da382cSHugh Dickins vma = next; 558e0da382cSHugh Dickins } 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 5618ac1f832SAndrea Arcangeli int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, 5628ac1f832SAndrea Arcangeli pmd_t *pmd, unsigned long address) 5631da177e4SLinus Torvalds { 564c4088ebdSKirill A. Shutemov spinlock_t *ptl; 5652f569afdSMartin Schwidefsky pgtable_t new = pte_alloc_one(mm, address); 5668ac1f832SAndrea Arcangeli int wait_split_huge_page; 5671da177e4SLinus Torvalds if (!new) 5681bb3630eSHugh Dickins return -ENOMEM; 5691bb3630eSHugh Dickins 570362a61adSNick Piggin /* 571362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 572362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 573362a61adSNick Piggin * put into page tables. 574362a61adSNick Piggin * 575362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 576362a61adSNick Piggin * table walking code (when walking the page table without locking; 577362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 578362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 579362a61adSNick Piggin * being the notable exception) will already guarantee loads are 580362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 581362a61adSNick Piggin * smp_read_barrier_depends() barriers in page table walking code. 582362a61adSNick Piggin */ 583362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 584362a61adSNick Piggin 585c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 5868ac1f832SAndrea Arcangeli wait_split_huge_page = 0; 5878ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 588e1f56c89SKirill A. Shutemov atomic_long_inc(&mm->nr_ptes); 5891da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 5902f569afdSMartin Schwidefsky new = NULL; 5918ac1f832SAndrea Arcangeli } else if (unlikely(pmd_trans_splitting(*pmd))) 5928ac1f832SAndrea Arcangeli wait_split_huge_page = 1; 593c4088ebdSKirill A. Shutemov spin_unlock(ptl); 5942f569afdSMartin Schwidefsky if (new) 5952f569afdSMartin Schwidefsky pte_free(mm, new); 5968ac1f832SAndrea Arcangeli if (wait_split_huge_page) 5978ac1f832SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 5981bb3630eSHugh Dickins return 0; 5991da177e4SLinus Torvalds } 6001da177e4SLinus Torvalds 6011bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 6021da177e4SLinus Torvalds { 6031bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 6041da177e4SLinus Torvalds if (!new) 6051bb3630eSHugh Dickins return -ENOMEM; 6061da177e4SLinus Torvalds 607362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 608362a61adSNick Piggin 609872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 6108ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 611872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 6122f569afdSMartin Schwidefsky new = NULL; 6138ac1f832SAndrea Arcangeli } else 6148ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_splitting(*pmd)); 615872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 6162f569afdSMartin Schwidefsky if (new) 6172f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 6181bb3630eSHugh Dickins return 0; 6191da177e4SLinus Torvalds } 6201da177e4SLinus Torvalds 621d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 622ae859762SHugh Dickins { 623d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 624d559db08SKAMEZAWA Hiroyuki } 625d559db08SKAMEZAWA Hiroyuki 626d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 627d559db08SKAMEZAWA Hiroyuki { 628d559db08SKAMEZAWA Hiroyuki int i; 629d559db08SKAMEZAWA Hiroyuki 63034e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 63105af2e10SDavid Rientjes sync_mm_rss(mm); 632d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 633d559db08SKAMEZAWA Hiroyuki if (rss[i]) 634d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 635ae859762SHugh Dickins } 636ae859762SHugh Dickins 6371da177e4SLinus Torvalds /* 6386aab341eSLinus Torvalds * This function is called to print an error when a bad pte 6396aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 6406aab341eSLinus Torvalds * a region that doesn't allow it. 641b5810039SNick Piggin * 642b5810039SNick Piggin * The calling function must still handle the error. 643b5810039SNick Piggin */ 6443dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 6453dc14741SHugh Dickins pte_t pte, struct page *page) 646b5810039SNick Piggin { 6473dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 6483dc14741SHugh Dickins pud_t *pud = pud_offset(pgd, addr); 6493dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 6503dc14741SHugh Dickins struct address_space *mapping; 6513dc14741SHugh Dickins pgoff_t index; 652d936cf9bSHugh Dickins static unsigned long resume; 653d936cf9bSHugh Dickins static unsigned long nr_shown; 654d936cf9bSHugh Dickins static unsigned long nr_unshown; 655d936cf9bSHugh Dickins 656d936cf9bSHugh Dickins /* 657d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 658d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 659d936cf9bSHugh Dickins */ 660d936cf9bSHugh Dickins if (nr_shown == 60) { 661d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 662d936cf9bSHugh Dickins nr_unshown++; 663d936cf9bSHugh Dickins return; 664d936cf9bSHugh Dickins } 665d936cf9bSHugh Dickins if (nr_unshown) { 6661e9e6365SHugh Dickins printk(KERN_ALERT 6671e9e6365SHugh Dickins "BUG: Bad page map: %lu messages suppressed\n", 668d936cf9bSHugh Dickins nr_unshown); 669d936cf9bSHugh Dickins nr_unshown = 0; 670d936cf9bSHugh Dickins } 671d936cf9bSHugh Dickins nr_shown = 0; 672d936cf9bSHugh Dickins } 673d936cf9bSHugh Dickins if (nr_shown++ == 0) 674d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 6753dc14741SHugh Dickins 6763dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 6773dc14741SHugh Dickins index = linear_page_index(vma, addr); 6783dc14741SHugh Dickins 6791e9e6365SHugh Dickins printk(KERN_ALERT 6801e9e6365SHugh Dickins "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 6813dc14741SHugh Dickins current->comm, 6823dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 683718a3821SWu Fengguang if (page) 684f0b791a3SDave Hansen dump_page(page, "bad pte"); 6851e9e6365SHugh Dickins printk(KERN_ALERT 6863dc14741SHugh Dickins "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", 6873dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 6883dc14741SHugh Dickins /* 6893dc14741SHugh Dickins * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y 6903dc14741SHugh Dickins */ 6913dc14741SHugh Dickins if (vma->vm_ops) 692071361d3SJoe Perches printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n", 693071361d3SJoe Perches vma->vm_ops->fault); 69472c2d531SAl Viro if (vma->vm_file) 695071361d3SJoe Perches printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n", 696071361d3SJoe Perches vma->vm_file->f_op->mmap); 697b5810039SNick Piggin dump_stack(); 698373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 699b5810039SNick Piggin } 700b5810039SNick Piggin 7012ec74c3eSSagi Grimberg static inline bool is_cow_mapping(vm_flags_t flags) 70267121172SLinus Torvalds { 70367121172SLinus Torvalds return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 70467121172SLinus Torvalds } 70567121172SLinus Torvalds 706b5810039SNick Piggin /* 7077e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 7086aab341eSLinus Torvalds * 7097e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 7107e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 7117e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 712b379d790SJared Hulbert * 7137e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 7147e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 7157e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 7167e675137SNick Piggin * described below. 7177e675137SNick Piggin * 7187e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 7197e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 7207e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 7216aab341eSLinus Torvalds * 722b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 723b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 7247e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 7257e675137SNick Piggin * mapping will always honor the rule 7266aab341eSLinus Torvalds * 7276aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 7286aab341eSLinus Torvalds * 7297e675137SNick Piggin * And for normal mappings this is false. 730b379d790SJared Hulbert * 7317e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 7327e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 7337e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 7347e675137SNick Piggin * special (because none can have been COWed). 735b379d790SJared Hulbert * 736b379d790SJared Hulbert * 7377e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 7387e675137SNick Piggin * 739b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 740b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 741b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 742b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 743b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 744b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 745b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 746b379d790SJared Hulbert * 747ee498ed7SHugh Dickins */ 7487e675137SNick Piggin #ifdef __HAVE_ARCH_PTE_SPECIAL 7497e675137SNick Piggin # define HAVE_PTE_SPECIAL 1 7507e675137SNick Piggin #else 7517e675137SNick Piggin # define HAVE_PTE_SPECIAL 0 7527e675137SNick Piggin #endif 7537e675137SNick Piggin struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 7547e675137SNick Piggin pte_t pte) 755ee498ed7SHugh Dickins { 75622b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 7577e675137SNick Piggin 7587e675137SNick Piggin if (HAVE_PTE_SPECIAL) { 75922b31eecSHugh Dickins if (likely(!pte_special(pte))) 76022b31eecSHugh Dickins goto check_pfn; 761a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 762a13ea5b7SHugh Dickins return NULL; 76362eede62SHugh Dickins if (!is_zero_pfn(pfn)) 76422b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 7657e675137SNick Piggin return NULL; 7667e675137SNick Piggin } 7677e675137SNick Piggin 7687e675137SNick Piggin /* !HAVE_PTE_SPECIAL case follows: */ 7697e675137SNick Piggin 770b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 771b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 772b379d790SJared Hulbert if (!pfn_valid(pfn)) 773b379d790SJared Hulbert return NULL; 774b379d790SJared Hulbert goto out; 775b379d790SJared Hulbert } else { 7767e675137SNick Piggin unsigned long off; 7777e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 7786aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 7796aab341eSLinus Torvalds return NULL; 78067121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 781fb155c16SLinus Torvalds return NULL; 7826aab341eSLinus Torvalds } 783b379d790SJared Hulbert } 7846aab341eSLinus Torvalds 78562eede62SHugh Dickins if (is_zero_pfn(pfn)) 78662eede62SHugh Dickins return NULL; 78722b31eecSHugh Dickins check_pfn: 78822b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 78922b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 79022b31eecSHugh Dickins return NULL; 79122b31eecSHugh Dickins } 7926aab341eSLinus Torvalds 7936aab341eSLinus Torvalds /* 7947e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 7957e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 7966aab341eSLinus Torvalds */ 797b379d790SJared Hulbert out: 7986aab341eSLinus Torvalds return pfn_to_page(pfn); 799ee498ed7SHugh Dickins } 800ee498ed7SHugh Dickins 801ee498ed7SHugh Dickins /* 8021da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 8031da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 8041da177e4SLinus Torvalds * covered by this vma. 8051da177e4SLinus Torvalds */ 8061da177e4SLinus Torvalds 807570a335bSHugh Dickins static inline unsigned long 8081da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 809b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 8108c103762SHugh Dickins unsigned long addr, int *rss) 8111da177e4SLinus Torvalds { 812b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 8131da177e4SLinus Torvalds pte_t pte = *src_pte; 8141da177e4SLinus Torvalds struct page *page; 8151da177e4SLinus Torvalds 8161da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 8171da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 8181da177e4SLinus Torvalds if (!pte_file(pte)) { 8190697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 8200697212aSChristoph Lameter 821570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 822570a335bSHugh Dickins return entry.val; 823570a335bSHugh Dickins 8241da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 8251da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 8261da177e4SLinus Torvalds spin_lock(&mmlist_lock); 827f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 828f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 829f412ac08SHugh Dickins &src_mm->mmlist); 8301da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8311da177e4SLinus Torvalds } 832b084d435SKAMEZAWA Hiroyuki if (likely(!non_swap_entry(entry))) 833b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 8349f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 8359f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 8369f9f1acdSKonstantin Khlebnikov 8379f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 8389f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]++; 8399f9f1acdSKonstantin Khlebnikov else 8409f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]++; 8419f9f1acdSKonstantin Khlebnikov 8429f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 8430697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 8440697212aSChristoph Lameter /* 8459f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 8469f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 8470697212aSChristoph Lameter */ 8480697212aSChristoph Lameter make_migration_entry_read(&entry); 8490697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 850c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 851c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 8520697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 8530697212aSChristoph Lameter } 8541da177e4SLinus Torvalds } 8559f9f1acdSKonstantin Khlebnikov } 856ae859762SHugh Dickins goto out_set_pte; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8611da177e4SLinus Torvalds * in the parent and the child 8621da177e4SLinus Torvalds */ 86367121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 8641da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8653dc90795SZachary Amsden pte = pte_wrprotect(pte); 8661da177e4SLinus Torvalds } 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds /* 8691da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8701da177e4SLinus Torvalds * the child 8711da177e4SLinus Torvalds */ 8721da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8731da177e4SLinus Torvalds pte = pte_mkclean(pte); 8741da177e4SLinus Torvalds pte = pte_mkold(pte); 8756aab341eSLinus Torvalds 8766aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 8776aab341eSLinus Torvalds if (page) { 8781da177e4SLinus Torvalds get_page(page); 87921333b2bSHugh Dickins page_dup_rmap(page); 880d559db08SKAMEZAWA Hiroyuki if (PageAnon(page)) 881d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]++; 882d559db08SKAMEZAWA Hiroyuki else 883d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]++; 8846aab341eSLinus Torvalds } 885ae859762SHugh Dickins 886ae859762SHugh Dickins out_set_pte: 887ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 888570a335bSHugh Dickins return 0; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 89171e3aac0SAndrea Arcangeli int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 8921da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 8931da177e4SLinus Torvalds unsigned long addr, unsigned long end) 8941da177e4SLinus Torvalds { 895c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 8961da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 897c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 898e040f218SHugh Dickins int progress = 0; 899d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 900570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds again: 903d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 904d559db08SKAMEZAWA Hiroyuki 905c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 9061da177e4SLinus Torvalds if (!dst_pte) 9071da177e4SLinus Torvalds return -ENOMEM; 908ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 9094c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 910f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 911c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 912c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9136606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds do { 9161da177e4SLinus Torvalds /* 9171da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9181da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9191da177e4SLinus Torvalds */ 920e040f218SHugh Dickins if (progress >= 32) { 921e040f218SHugh Dickins progress = 0; 922e040f218SHugh Dickins if (need_resched() || 92395c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9241da177e4SLinus Torvalds break; 925e040f218SHugh Dickins } 9261da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9271da177e4SLinus Torvalds progress++; 9281da177e4SLinus Torvalds continue; 9291da177e4SLinus Torvalds } 930570a335bSHugh Dickins entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, 931570a335bSHugh Dickins vma, addr, rss); 932570a335bSHugh Dickins if (entry.val) 933570a335bSHugh Dickins break; 9341da177e4SLinus Torvalds progress += 8; 9351da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9361da177e4SLinus Torvalds 9376606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 938c74df32cSHugh Dickins spin_unlock(src_ptl); 939ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 940d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 941c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 942c74df32cSHugh Dickins cond_resched(); 943570a335bSHugh Dickins 944570a335bSHugh Dickins if (entry.val) { 945570a335bSHugh Dickins if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) 946570a335bSHugh Dickins return -ENOMEM; 947570a335bSHugh Dickins progress = 0; 948570a335bSHugh Dickins } 9491da177e4SLinus Torvalds if (addr != end) 9501da177e4SLinus Torvalds goto again; 9511da177e4SLinus Torvalds return 0; 9521da177e4SLinus Torvalds } 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9551da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 9561da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9571da177e4SLinus Torvalds { 9581da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 9591da177e4SLinus Torvalds unsigned long next; 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 9621da177e4SLinus Torvalds if (!dst_pmd) 9631da177e4SLinus Torvalds return -ENOMEM; 9641da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 9651da177e4SLinus Torvalds do { 9661da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 96771e3aac0SAndrea Arcangeli if (pmd_trans_huge(*src_pmd)) { 96871e3aac0SAndrea Arcangeli int err; 96914d1a55cSAndrea Arcangeli VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); 97071e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 97171e3aac0SAndrea Arcangeli dst_pmd, src_pmd, addr, vma); 97271e3aac0SAndrea Arcangeli if (err == -ENOMEM) 97371e3aac0SAndrea Arcangeli return -ENOMEM; 97471e3aac0SAndrea Arcangeli if (!err) 97571e3aac0SAndrea Arcangeli continue; 97671e3aac0SAndrea Arcangeli /* fall through */ 97771e3aac0SAndrea Arcangeli } 9781da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 9791da177e4SLinus Torvalds continue; 9801da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 9811da177e4SLinus Torvalds vma, addr, next)) 9821da177e4SLinus Torvalds return -ENOMEM; 9831da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 9841da177e4SLinus Torvalds return 0; 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 9881da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 9891da177e4SLinus Torvalds unsigned long addr, unsigned long end) 9901da177e4SLinus Torvalds { 9911da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 9921da177e4SLinus Torvalds unsigned long next; 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 9951da177e4SLinus Torvalds if (!dst_pud) 9961da177e4SLinus Torvalds return -ENOMEM; 9971da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 9981da177e4SLinus Torvalds do { 9991da177e4SLinus Torvalds next = pud_addr_end(addr, end); 10001da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 10011da177e4SLinus Torvalds continue; 10021da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 10031da177e4SLinus Torvalds vma, addr, next)) 10041da177e4SLinus Torvalds return -ENOMEM; 10051da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 10061da177e4SLinus Torvalds return 0; 10071da177e4SLinus Torvalds } 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 10101da177e4SLinus Torvalds struct vm_area_struct *vma) 10111da177e4SLinus Torvalds { 10121da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 10131da177e4SLinus Torvalds unsigned long next; 10141da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 10151da177e4SLinus Torvalds unsigned long end = vma->vm_end; 10162ec74c3eSSagi Grimberg unsigned long mmun_start; /* For mmu_notifiers */ 10172ec74c3eSSagi Grimberg unsigned long mmun_end; /* For mmu_notifiers */ 10182ec74c3eSSagi Grimberg bool is_cow; 1019cddb8a5cSAndrea Arcangeli int ret; 10201da177e4SLinus Torvalds 1021d992895bSNick Piggin /* 1022d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1023d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1024d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1025d992895bSNick Piggin * efficient than faulting. 1026d992895bSNick Piggin */ 10274b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR | 10284b6e1e37SKonstantin Khlebnikov VM_PFNMAP | VM_MIXEDMAP))) { 1029d992895bSNick Piggin if (!vma->anon_vma) 1030d992895bSNick Piggin return 0; 1031d992895bSNick Piggin } 1032d992895bSNick Piggin 10331da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 10341da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 10351da177e4SLinus Torvalds 1036b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) { 10372ab64037Svenkatesh.pallipadi@intel.com /* 10382ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 10392ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 10402ab64037Svenkatesh.pallipadi@intel.com */ 10415180da41SSuresh Siddha ret = track_pfn_copy(vma); 10422ab64037Svenkatesh.pallipadi@intel.com if (ret) 10432ab64037Svenkatesh.pallipadi@intel.com return ret; 10442ab64037Svenkatesh.pallipadi@intel.com } 10452ab64037Svenkatesh.pallipadi@intel.com 1046cddb8a5cSAndrea Arcangeli /* 1047cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1048cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1049cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1050cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1051cddb8a5cSAndrea Arcangeli */ 10522ec74c3eSSagi Grimberg is_cow = is_cow_mapping(vma->vm_flags); 10532ec74c3eSSagi Grimberg mmun_start = addr; 10542ec74c3eSSagi Grimberg mmun_end = end; 10552ec74c3eSSagi Grimberg if (is_cow) 10562ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_start(src_mm, mmun_start, 10572ec74c3eSSagi Grimberg mmun_end); 1058cddb8a5cSAndrea Arcangeli 1059cddb8a5cSAndrea Arcangeli ret = 0; 10601da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 10611da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 10621da177e4SLinus Torvalds do { 10631da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 10641da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 10651da177e4SLinus Torvalds continue; 1066cddb8a5cSAndrea Arcangeli if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 1067cddb8a5cSAndrea Arcangeli vma, addr, next))) { 1068cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1069cddb8a5cSAndrea Arcangeli break; 1070cddb8a5cSAndrea Arcangeli } 10711da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1072cddb8a5cSAndrea Arcangeli 10732ec74c3eSSagi Grimberg if (is_cow) 10742ec74c3eSSagi Grimberg mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); 1075cddb8a5cSAndrea Arcangeli return ret; 10761da177e4SLinus Torvalds } 10771da177e4SLinus Torvalds 107851c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1079b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 10801da177e4SLinus Torvalds unsigned long addr, unsigned long end, 108197a89413SPeter Zijlstra struct zap_details *details) 10821da177e4SLinus Torvalds { 1083b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1084d16dfc55SPeter Zijlstra int force_flush = 0; 1085d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 108697a89413SPeter Zijlstra spinlock_t *ptl; 10875f1a1907SSteven Rostedt pte_t *start_pte; 108897a89413SPeter Zijlstra pte_t *pte; 1089d559db08SKAMEZAWA Hiroyuki 1090d16dfc55SPeter Zijlstra again: 1091e303297eSPeter Zijlstra init_rss_vec(rss); 10925f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 10935f1a1907SSteven Rostedt pte = start_pte; 10946606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 10951da177e4SLinus Torvalds do { 10961da177e4SLinus Torvalds pte_t ptent = *pte; 109751c6f666SRobin Holt if (pte_none(ptent)) { 10981da177e4SLinus Torvalds continue; 109951c6f666SRobin Holt } 110051c6f666SRobin Holt 11016f5e6b9eSHugh Dickins if (pte_present(ptent)) { 11026f5e6b9eSHugh Dickins struct page *page; 11036f5e6b9eSHugh Dickins 11046aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 11051da177e4SLinus Torvalds if (unlikely(details) && page) { 11061da177e4SLinus Torvalds /* 11071da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 11081da177e4SLinus Torvalds * invalidate cache without truncating: 11091da177e4SLinus Torvalds * unmap shared but keep private pages. 11101da177e4SLinus Torvalds */ 11111da177e4SLinus Torvalds if (details->check_mapping && 11121da177e4SLinus Torvalds details->check_mapping != page->mapping) 11131da177e4SLinus Torvalds continue; 11141da177e4SLinus Torvalds /* 11151da177e4SLinus Torvalds * Each page->index must be checked when 11161da177e4SLinus Torvalds * invalidating or truncating nonlinear. 11171da177e4SLinus Torvalds */ 11181da177e4SLinus Torvalds if (details->nonlinear_vma && 11191da177e4SLinus Torvalds (page->index < details->first_index || 11201da177e4SLinus Torvalds page->index > details->last_index)) 11211da177e4SLinus Torvalds continue; 11221da177e4SLinus Torvalds } 1123b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1124a600388dSZachary Amsden tlb->fullmm); 11251da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 11261da177e4SLinus Torvalds if (unlikely(!page)) 11271da177e4SLinus Torvalds continue; 11281da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 11291da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 113041bb3476SCyrill Gorcunov addr) != page->index) { 113141bb3476SCyrill Gorcunov pte_t ptfile = pgoff_to_pte(page->index); 113241bb3476SCyrill Gorcunov if (pte_soft_dirty(ptent)) 113341bb3476SCyrill Gorcunov pte_file_mksoft_dirty(ptfile); 113441bb3476SCyrill Gorcunov set_pte_at(mm, addr, pte, ptfile); 113541bb3476SCyrill Gorcunov } 11361da177e4SLinus Torvalds if (PageAnon(page)) 1137d559db08SKAMEZAWA Hiroyuki rss[MM_ANONPAGES]--; 11386237bcd9SHugh Dickins else { 1139*1cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 1140*1cf35d47SLinus Torvalds force_flush = 1; 11416237bcd9SHugh Dickins set_page_dirty(page); 1142*1cf35d47SLinus Torvalds } 11434917e5d0SJohannes Weiner if (pte_young(ptent) && 114464363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1145bf3f3bc5SNick Piggin mark_page_accessed(page); 1146d559db08SKAMEZAWA Hiroyuki rss[MM_FILEPAGES]--; 11476237bcd9SHugh Dickins } 1148edc315fdSHugh Dickins page_remove_rmap(page); 11493dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 11503dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 1151*1cf35d47SLinus Torvalds if (unlikely(!__tlb_remove_page(tlb, page))) { 1152*1cf35d47SLinus Torvalds force_flush = 1; 1153d16dfc55SPeter Zijlstra break; 1154*1cf35d47SLinus Torvalds } 11551da177e4SLinus Torvalds continue; 11561da177e4SLinus Torvalds } 11571da177e4SLinus Torvalds /* 11581da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 11591da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 11601da177e4SLinus Torvalds */ 11611da177e4SLinus Torvalds if (unlikely(details)) 11621da177e4SLinus Torvalds continue; 11632509ef26SHugh Dickins if (pte_file(ptent)) { 11642509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) 11652509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1166b084d435SKAMEZAWA Hiroyuki } else { 1167b084d435SKAMEZAWA Hiroyuki swp_entry_t entry = pte_to_swp_entry(ptent); 1168b084d435SKAMEZAWA Hiroyuki 1169b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1170b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 11719f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 11729f9f1acdSKonstantin Khlebnikov struct page *page; 11739f9f1acdSKonstantin Khlebnikov 11749f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 11759f9f1acdSKonstantin Khlebnikov 11769f9f1acdSKonstantin Khlebnikov if (PageAnon(page)) 11779f9f1acdSKonstantin Khlebnikov rss[MM_ANONPAGES]--; 11789f9f1acdSKonstantin Khlebnikov else 11799f9f1acdSKonstantin Khlebnikov rss[MM_FILEPAGES]--; 11809f9f1acdSKonstantin Khlebnikov } 1181b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 11822509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 1183b084d435SKAMEZAWA Hiroyuki } 11849888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 118597a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1186ae859762SHugh Dickins 1187d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 11886606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 118951c6f666SRobin Holt 1190*1cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 1191d16dfc55SPeter Zijlstra if (force_flush) { 11922b047252SLinus Torvalds unsigned long old_end; 11932b047252SLinus Torvalds 11942b047252SLinus Torvalds /* 11952b047252SLinus Torvalds * Flush the TLB just for the previous segment, 11962b047252SLinus Torvalds * then update the range to be the remaining 11972b047252SLinus Torvalds * TLB range. 11982b047252SLinus Torvalds */ 11992b047252SLinus Torvalds old_end = tlb->end; 1200e6c495a9SVineet Gupta tlb->end = addr; 1201*1cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 12022b047252SLinus Torvalds tlb->start = addr; 12032b047252SLinus Torvalds tlb->end = old_end; 1204*1cf35d47SLinus Torvalds } 1205*1cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 1206*1cf35d47SLinus Torvalds 1207*1cf35d47SLinus Torvalds /* 1208*1cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 1209*1cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 1210*1cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 1211*1cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 1212*1cf35d47SLinus Torvalds */ 1213*1cf35d47SLinus Torvalds if (force_flush) { 1214*1cf35d47SLinus Torvalds force_flush = 0; 1215*1cf35d47SLinus Torvalds tlb_flush_mmu_free(tlb); 12162b047252SLinus Torvalds 12172b047252SLinus Torvalds if (addr != end) 1218d16dfc55SPeter Zijlstra goto again; 1219d16dfc55SPeter Zijlstra } 1220d16dfc55SPeter Zijlstra 122151c6f666SRobin Holt return addr; 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds 122451c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1225b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 12261da177e4SLinus Torvalds unsigned long addr, unsigned long end, 122797a89413SPeter Zijlstra struct zap_details *details) 12281da177e4SLinus Torvalds { 12291da177e4SLinus Torvalds pmd_t *pmd; 12301da177e4SLinus Torvalds unsigned long next; 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 12331da177e4SLinus Torvalds do { 12341da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 123571e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 123614d1a55cSAndrea Arcangeli if (next - addr != HPAGE_PMD_SIZE) { 1237e0897d75SDavid Rientjes #ifdef CONFIG_DEBUG_VM 1238e0897d75SDavid Rientjes if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { 1239e0897d75SDavid 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", 1240e0897d75SDavid Rientjes __func__, addr, end, 1241e0897d75SDavid Rientjes vma->vm_start, 1242e0897d75SDavid Rientjes vma->vm_end); 1243e0897d75SDavid Rientjes BUG(); 1244e0897d75SDavid Rientjes } 1245e0897d75SDavid Rientjes #endif 1246e180377fSKirill A. Shutemov split_huge_page_pmd(vma, addr, pmd); 1247f21760b1SShaohua Li } else if (zap_huge_pmd(tlb, vma, pmd, addr)) 12481a5a9906SAndrea Arcangeli goto next; 124971e3aac0SAndrea Arcangeli /* fall through */ 125071e3aac0SAndrea Arcangeli } 12511a5a9906SAndrea Arcangeli /* 12521a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 12531a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 12541a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 12551a5a9906SAndrea Arcangeli * because MADV_DONTNEED holds the mmap_sem in read 12561a5a9906SAndrea Arcangeli * mode. 12571a5a9906SAndrea Arcangeli */ 12581a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 12591a5a9906SAndrea Arcangeli goto next; 126097a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 12611a5a9906SAndrea Arcangeli next: 126297a89413SPeter Zijlstra cond_resched(); 126397a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 126451c6f666SRobin Holt 126551c6f666SRobin Holt return addr; 12661da177e4SLinus Torvalds } 12671da177e4SLinus Torvalds 126851c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1269b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 12701da177e4SLinus Torvalds unsigned long addr, unsigned long end, 127197a89413SPeter Zijlstra struct zap_details *details) 12721da177e4SLinus Torvalds { 12731da177e4SLinus Torvalds pud_t *pud; 12741da177e4SLinus Torvalds unsigned long next; 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 12771da177e4SLinus Torvalds do { 12781da177e4SLinus Torvalds next = pud_addr_end(addr, end); 127997a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 12801da177e4SLinus Torvalds continue; 128197a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 128297a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 128351c6f666SRobin Holt 128451c6f666SRobin Holt return addr; 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 1287038c7aa1SAl Viro static void unmap_page_range(struct mmu_gather *tlb, 128851c6f666SRobin Holt struct vm_area_struct *vma, 12891da177e4SLinus Torvalds unsigned long addr, unsigned long end, 129097a89413SPeter Zijlstra struct zap_details *details) 12911da177e4SLinus Torvalds { 12921da177e4SLinus Torvalds pgd_t *pgd; 12931da177e4SLinus Torvalds unsigned long next; 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 12961da177e4SLinus Torvalds details = NULL; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds BUG_ON(addr >= end); 1299569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_start(); 13001da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 13011da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 13021da177e4SLinus Torvalds do { 13031da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 130497a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 13051da177e4SLinus Torvalds continue; 130697a89413SPeter Zijlstra next = zap_pud_range(tlb, vma, pgd, addr, next, details); 130797a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 13081da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 1309569b846dSKAMEZAWA Hiroyuki mem_cgroup_uncharge_end(); 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 1312f5cc4eefSAl Viro 1313f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 13141da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 13154f74d2c8SLinus Torvalds unsigned long end_addr, 13161da177e4SLinus Torvalds struct zap_details *details) 13171da177e4SLinus Torvalds { 1318f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 13191da177e4SLinus Torvalds unsigned long end; 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds if (start >= vma->vm_end) 1322f5cc4eefSAl Viro return; 13231da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 13241da177e4SLinus Torvalds if (end <= vma->vm_start) 1325f5cc4eefSAl Viro return; 13261da177e4SLinus Torvalds 1327cbc91f71SSrikar Dronamraju if (vma->vm_file) 1328cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1329cbc91f71SSrikar Dronamraju 1330b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 13315180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 13322ab64037Svenkatesh.pallipadi@intel.com 13338b2a1238SAl Viro if (start != end) { 133451c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1335a137e1ccSAndi Kleen /* 1336a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1337a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1338a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 13397aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1340a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 13417aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1342a137e1ccSAndi Kleen * before calling this function to clean up. 1343a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1344a137e1ccSAndi Kleen * safe to do nothing in this case. 1345a137e1ccSAndi Kleen */ 134624669e58SAneesh Kumar K.V if (vma->vm_file) { 134724669e58SAneesh Kumar K.V mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); 1348d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 134924669e58SAneesh Kumar K.V mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); 135024669e58SAneesh Kumar K.V } 135151c6f666SRobin Holt } else 1352038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 135397a89413SPeter Zijlstra } 135451c6f666SRobin Holt } 13551da177e4SLinus Torvalds 1356f5cc4eefSAl Viro /** 1357f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1358f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1359f5cc4eefSAl Viro * @vma: the starting vma 1360f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1361f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1362f5cc4eefSAl Viro * 1363f5cc4eefSAl Viro * Unmap all pages in the vma list. 1364f5cc4eefSAl Viro * 1365f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1366f5cc4eefSAl Viro * 1367f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1368f5cc4eefSAl Viro * 1369f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1370f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1371f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1372f5cc4eefSAl Viro * drops the lock and schedules. 1373f5cc4eefSAl Viro */ 1374f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1375f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 13764f74d2c8SLinus Torvalds unsigned long end_addr) 1377f5cc4eefSAl Viro { 1378f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1379f5cc4eefSAl Viro 1380f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); 1381f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 13824f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1383cddb8a5cSAndrea Arcangeli mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds /** 13871da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 13881da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1389eb4546bbSRandy Dunlap * @start: starting address of pages to zap 13901da177e4SLinus Torvalds * @size: number of bytes to zap 13911da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 1392f5cc4eefSAl Viro * 1393f5cc4eefSAl Viro * Caller must protect the VMA list 13941da177e4SLinus Torvalds */ 13957e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 13961da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 13971da177e4SLinus Torvalds { 13981da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1399d16dfc55SPeter Zijlstra struct mmu_gather tlb; 14007e027b14SLinus Torvalds unsigned long end = start + size; 14011da177e4SLinus Torvalds 14021da177e4SLinus Torvalds lru_add_drain(); 14032b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 1404365e9c87SHugh Dickins update_hiwater_rss(mm); 14057e027b14SLinus Torvalds mmu_notifier_invalidate_range_start(mm, start, end); 14067e027b14SLinus Torvalds for ( ; vma && vma->vm_start < end; vma = vma->vm_next) 14074f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, start, end, details); 14087e027b14SLinus Torvalds mmu_notifier_invalidate_range_end(mm, start, end); 14097e027b14SLinus Torvalds tlb_finish_mmu(&tlb, start, end); 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds 1412c627f9ccSJack Steiner /** 1413f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1414f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1415f5cc4eefSAl Viro * @address: starting address of pages to zap 1416f5cc4eefSAl Viro * @size: number of bytes to zap 1417f5cc4eefSAl Viro * @details: details of nonlinear truncation or shared cache invalidation 1418f5cc4eefSAl Viro * 1419f5cc4eefSAl Viro * The range must fit into one VMA. 1420f5cc4eefSAl Viro */ 1421f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1422f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1423f5cc4eefSAl Viro { 1424f5cc4eefSAl Viro struct mm_struct *mm = vma->vm_mm; 1425f5cc4eefSAl Viro struct mmu_gather tlb; 1426f5cc4eefSAl Viro unsigned long end = address + size; 1427f5cc4eefSAl Viro 1428f5cc4eefSAl Viro lru_add_drain(); 14292b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, address, end); 1430f5cc4eefSAl Viro update_hiwater_rss(mm); 1431f5cc4eefSAl Viro mmu_notifier_invalidate_range_start(mm, address, end); 14324f74d2c8SLinus Torvalds unmap_single_vma(&tlb, vma, address, end, details); 1433f5cc4eefSAl Viro mmu_notifier_invalidate_range_end(mm, address, end); 1434f5cc4eefSAl Viro tlb_finish_mmu(&tlb, address, end); 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds 1437c627f9ccSJack Steiner /** 1438c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1439c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1440c627f9ccSJack Steiner * @address: starting address of pages to zap 1441c627f9ccSJack Steiner * @size: number of bytes to zap 1442c627f9ccSJack Steiner * 1443c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1444c627f9ccSJack Steiner * 1445c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1446c627f9ccSJack Steiner * 1447c627f9ccSJack Steiner * Returns 0 if successful. 1448c627f9ccSJack Steiner */ 1449c627f9ccSJack Steiner int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1450c627f9ccSJack Steiner unsigned long size) 1451c627f9ccSJack Steiner { 1452c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1453c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 1454c627f9ccSJack Steiner return -1; 1455f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1456c627f9ccSJack Steiner return 0; 1457c627f9ccSJack Steiner } 1458c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1459c627f9ccSJack Steiner 1460142762bdSJohannes Weiner /** 1461240aadeeSMichel Lespinasse * follow_page_mask - look up a page descriptor from a user-virtual address 1462142762bdSJohannes Weiner * @vma: vm_area_struct mapping @address 1463142762bdSJohannes Weiner * @address: virtual address to look up 1464142762bdSJohannes Weiner * @flags: flags modifying lookup behaviour 1465240aadeeSMichel Lespinasse * @page_mask: on output, *page_mask is set according to the size of the page 1466142762bdSJohannes Weiner * 1467142762bdSJohannes Weiner * @flags can have FOLL_ flags set, defined in <linux/mm.h> 1468142762bdSJohannes Weiner * 1469142762bdSJohannes Weiner * Returns the mapped (struct page *), %NULL if no mapping exists, or 1470142762bdSJohannes Weiner * an error pointer if there is a mapping to something not represented 1471142762bdSJohannes Weiner * by a page descriptor (see also vm_normal_page()). 14721da177e4SLinus Torvalds */ 1473240aadeeSMichel Lespinasse struct page *follow_page_mask(struct vm_area_struct *vma, 1474240aadeeSMichel Lespinasse unsigned long address, unsigned int flags, 1475240aadeeSMichel Lespinasse unsigned int *page_mask) 14761da177e4SLinus Torvalds { 14771da177e4SLinus Torvalds pgd_t *pgd; 14781da177e4SLinus Torvalds pud_t *pud; 14791da177e4SLinus Torvalds pmd_t *pmd; 14801da177e4SLinus Torvalds pte_t *ptep, pte; 1481deceb6cdSHugh Dickins spinlock_t *ptl; 14821da177e4SLinus Torvalds struct page *page; 14836aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 14841da177e4SLinus Torvalds 1485240aadeeSMichel Lespinasse *page_mask = 0; 1486240aadeeSMichel Lespinasse 1487deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 1488deceb6cdSHugh Dickins if (!IS_ERR(page)) { 1489deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 1490deceb6cdSHugh Dickins goto out; 1491deceb6cdSHugh Dickins } 14921da177e4SLinus Torvalds 1493deceb6cdSHugh Dickins page = NULL; 14941da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 14951da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 1496deceb6cdSHugh Dickins goto no_page_table; 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds pud = pud_offset(pgd, address); 1499ceb86879SAndi Kleen if (pud_none(*pud)) 1500ceb86879SAndi Kleen goto no_page_table; 15018a07651eSHugh Dickins if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { 1502e632a938SNaoya Horiguchi if (flags & FOLL_GET) 1503e632a938SNaoya Horiguchi goto out; 1504ceb86879SAndi Kleen page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); 1505ceb86879SAndi Kleen goto out; 1506ceb86879SAndi Kleen } 1507ceb86879SAndi Kleen if (unlikely(pud_bad(*pud))) 1508deceb6cdSHugh Dickins goto no_page_table; 15091da177e4SLinus Torvalds 15101da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 1511aeed5fceSHugh Dickins if (pmd_none(*pmd)) 1512deceb6cdSHugh Dickins goto no_page_table; 151371e3aac0SAndrea Arcangeli if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { 1514deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 1515e632a938SNaoya Horiguchi if (flags & FOLL_GET) { 1516e632a938SNaoya Horiguchi /* 1517e632a938SNaoya Horiguchi * Refcount on tail pages are not well-defined and 1518e632a938SNaoya Horiguchi * shouldn't be taken. The caller should handle a NULL 1519e632a938SNaoya Horiguchi * return when trying to follow tail pages. 1520e632a938SNaoya Horiguchi */ 1521e632a938SNaoya Horiguchi if (PageHead(page)) 1522e632a938SNaoya Horiguchi get_page(page); 1523e632a938SNaoya Horiguchi else { 1524e632a938SNaoya Horiguchi page = NULL; 1525e632a938SNaoya Horiguchi goto out; 1526e632a938SNaoya Horiguchi } 1527e632a938SNaoya Horiguchi } 1528deceb6cdSHugh Dickins goto out; 1529deceb6cdSHugh Dickins } 15300b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pmd_numa(*pmd)) 15310b9d7052SAndrea Arcangeli goto no_page_table; 153271e3aac0SAndrea Arcangeli if (pmd_trans_huge(*pmd)) { 1533500d65d4SAndrea Arcangeli if (flags & FOLL_SPLIT) { 1534e180377fSKirill A. Shutemov split_huge_page_pmd(vma, address, pmd); 1535500d65d4SAndrea Arcangeli goto split_fallthrough; 1536500d65d4SAndrea Arcangeli } 1537c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 153871e3aac0SAndrea Arcangeli if (likely(pmd_trans_huge(*pmd))) { 153971e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_splitting(*pmd))) { 1540c4088ebdSKirill A. Shutemov spin_unlock(ptl); 154171e3aac0SAndrea Arcangeli wait_split_huge_page(vma->anon_vma, pmd); 154271e3aac0SAndrea Arcangeli } else { 1543b676b293SDavid Rientjes page = follow_trans_huge_pmd(vma, address, 154471e3aac0SAndrea Arcangeli pmd, flags); 1545c4088ebdSKirill A. Shutemov spin_unlock(ptl); 1546240aadeeSMichel Lespinasse *page_mask = HPAGE_PMD_NR - 1; 154771e3aac0SAndrea Arcangeli goto out; 154871e3aac0SAndrea Arcangeli } 154971e3aac0SAndrea Arcangeli } else 1550c4088ebdSKirill A. Shutemov spin_unlock(ptl); 155171e3aac0SAndrea Arcangeli /* fall through */ 155271e3aac0SAndrea Arcangeli } 1553500d65d4SAndrea Arcangeli split_fallthrough: 1554aeed5fceSHugh Dickins if (unlikely(pmd_bad(*pmd))) 1555aeed5fceSHugh Dickins goto no_page_table; 1556aeed5fceSHugh Dickins 1557deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 15581da177e4SLinus Torvalds 15591da177e4SLinus Torvalds pte = *ptep; 15605117b3b8SHugh Dickins if (!pte_present(pte)) { 15615117b3b8SHugh Dickins swp_entry_t entry; 15625117b3b8SHugh Dickins /* 15635117b3b8SHugh Dickins * KSM's break_ksm() relies upon recognizing a ksm page 15645117b3b8SHugh Dickins * even while it is being migrated, so for that case we 15655117b3b8SHugh Dickins * need migration_entry_wait(). 15665117b3b8SHugh Dickins */ 15675117b3b8SHugh Dickins if (likely(!(flags & FOLL_MIGRATION))) 156889f5b7daSLinus Torvalds goto no_page; 15695117b3b8SHugh Dickins if (pte_none(pte) || pte_file(pte)) 15705117b3b8SHugh Dickins goto no_page; 15715117b3b8SHugh Dickins entry = pte_to_swp_entry(pte); 15725117b3b8SHugh Dickins if (!is_migration_entry(entry)) 15735117b3b8SHugh Dickins goto no_page; 15745117b3b8SHugh Dickins pte_unmap_unlock(ptep, ptl); 15755117b3b8SHugh Dickins migration_entry_wait(mm, pmd, address); 15765117b3b8SHugh Dickins goto split_fallthrough; 15775117b3b8SHugh Dickins } 15780b9d7052SAndrea Arcangeli if ((flags & FOLL_NUMA) && pte_numa(pte)) 15790b9d7052SAndrea Arcangeli goto no_page; 1580deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 1581deceb6cdSHugh Dickins goto unlock; 1582a13ea5b7SHugh Dickins 15836aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 1584a13ea5b7SHugh Dickins if (unlikely(!page)) { 1585a13ea5b7SHugh Dickins if ((flags & FOLL_DUMP) || 158662eede62SHugh Dickins !is_zero_pfn(pte_pfn(pte))) 158789f5b7daSLinus Torvalds goto bad_page; 1588a13ea5b7SHugh Dickins page = pte_page(pte); 1589a13ea5b7SHugh Dickins } 1590deceb6cdSHugh Dickins 1591deceb6cdSHugh Dickins if (flags & FOLL_GET) 159270b50f94SAndrea Arcangeli get_page_foll(page); 1593deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 1594deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 1595deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 1596f33ea7f4SNick Piggin set_page_dirty(page); 1597bd775c42SKOSAKI Motohiro /* 1598bd775c42SKOSAKI Motohiro * pte_mkyoung() would be more correct here, but atomic care 1599bd775c42SKOSAKI Motohiro * is needed to avoid losing the dirty bit: it is easier to use 1600bd775c42SKOSAKI Motohiro * mark_page_accessed(). 1601bd775c42SKOSAKI Motohiro */ 16021da177e4SLinus Torvalds mark_page_accessed(page); 16031da177e4SLinus Torvalds } 1604a1fde08cSLinus Torvalds if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { 1605110d74a9SMichel Lespinasse /* 1606110d74a9SMichel Lespinasse * The preliminary mapping check is mainly to avoid the 1607110d74a9SMichel Lespinasse * pointless overhead of lock_page on the ZERO_PAGE 1608110d74a9SMichel Lespinasse * which might bounce very badly if there is contention. 1609110d74a9SMichel Lespinasse * 1610110d74a9SMichel Lespinasse * If the page is already locked, we don't need to 1611110d74a9SMichel Lespinasse * handle it now - vmscan will handle it later if and 1612110d74a9SMichel Lespinasse * when it attempts to reclaim the page. 1613110d74a9SMichel Lespinasse */ 1614110d74a9SMichel Lespinasse if (page->mapping && trylock_page(page)) { 1615110d74a9SMichel Lespinasse lru_add_drain(); /* push cached pages to LRU */ 1616110d74a9SMichel Lespinasse /* 1617e6c509f8SHugh Dickins * Because we lock page here, and migration is 1618e6c509f8SHugh Dickins * blocked by the pte's page reference, and we 1619e6c509f8SHugh Dickins * know the page is still mapped, we don't even 1620e6c509f8SHugh Dickins * need to check for file-cache page truncation. 1621110d74a9SMichel Lespinasse */ 1622110d74a9SMichel Lespinasse mlock_vma_page(page); 1623110d74a9SMichel Lespinasse unlock_page(page); 1624110d74a9SMichel Lespinasse } 1625110d74a9SMichel Lespinasse } 1626deceb6cdSHugh Dickins unlock: 1627deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 16281da177e4SLinus Torvalds out: 1629deceb6cdSHugh Dickins return page; 1630deceb6cdSHugh Dickins 163189f5b7daSLinus Torvalds bad_page: 163289f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 163389f5b7daSLinus Torvalds return ERR_PTR(-EFAULT); 163489f5b7daSLinus Torvalds 163589f5b7daSLinus Torvalds no_page: 163689f5b7daSLinus Torvalds pte_unmap_unlock(ptep, ptl); 163789f5b7daSLinus Torvalds if (!pte_none(pte)) 163889f5b7daSLinus Torvalds return page; 16398e4b9a60SHugh Dickins 1640deceb6cdSHugh Dickins no_page_table: 1641deceb6cdSHugh Dickins /* 1642deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 16438e4b9a60SHugh Dickins * has touched so far, we don't want to allocate unnecessary pages or 16448e4b9a60SHugh Dickins * page tables. Return error instead of NULL to skip handle_mm_fault, 16458e4b9a60SHugh Dickins * then get_dump_page() will return NULL to leave a hole in the dump. 16468e4b9a60SHugh Dickins * But we can only make this optimization where a hole would surely 16478e4b9a60SHugh Dickins * be zero-filled if handle_mm_fault() actually did handle it. 1648deceb6cdSHugh Dickins */ 16498e4b9a60SHugh Dickins if ((flags & FOLL_DUMP) && 16508e4b9a60SHugh Dickins (!vma->vm_ops || !vma->vm_ops->fault)) 16518e4b9a60SHugh Dickins return ERR_PTR(-EFAULT); 1652deceb6cdSHugh Dickins return page; 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds 165595042f9eSLinus Torvalds static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) 165695042f9eSLinus Torvalds { 1657a09a79f6SMikulas Patocka return stack_guard_page_start(vma, addr) || 1658a09a79f6SMikulas Patocka stack_guard_page_end(vma, addr+PAGE_SIZE); 165995042f9eSLinus Torvalds } 166095042f9eSLinus Torvalds 16610014bd99SHuang Ying /** 16620014bd99SHuang Ying * __get_user_pages() - pin user pages in memory 16630014bd99SHuang Ying * @tsk: task_struct of target task 16640014bd99SHuang Ying * @mm: mm_struct of target mm 16650014bd99SHuang Ying * @start: starting user address 16660014bd99SHuang Ying * @nr_pages: number of pages from start to pin 16670014bd99SHuang Ying * @gup_flags: flags modifying pin behaviour 16680014bd99SHuang Ying * @pages: array that receives pointers to the pages pinned. 16690014bd99SHuang Ying * Should be at least nr_pages long. Or NULL, if caller 16700014bd99SHuang Ying * only intends to ensure the pages are faulted in. 16710014bd99SHuang Ying * @vmas: array of pointers to vmas corresponding to each page. 16720014bd99SHuang Ying * Or NULL if the caller does not require them. 16730014bd99SHuang Ying * @nonblocking: whether waiting for disk IO or mmap_sem contention 16740014bd99SHuang Ying * 16750014bd99SHuang Ying * Returns number of pages pinned. This may be fewer than the number 16760014bd99SHuang Ying * requested. If nr_pages is 0 or negative, returns 0. If no pages 16770014bd99SHuang Ying * were pinned, returns -errno. Each page returned must be released 16780014bd99SHuang Ying * with a put_page() call when it is finished with. vmas will only 16790014bd99SHuang Ying * remain valid while mmap_sem is held. 16800014bd99SHuang Ying * 16810014bd99SHuang Ying * Must be called with mmap_sem held for read or write. 16820014bd99SHuang Ying * 16830014bd99SHuang Ying * __get_user_pages walks a process's page tables and takes a reference to 16840014bd99SHuang Ying * each struct page that each user address corresponds to at a given 16850014bd99SHuang Ying * instant. That is, it takes the page that would be accessed if a user 16860014bd99SHuang Ying * thread accesses the given user virtual address at that instant. 16870014bd99SHuang Ying * 16880014bd99SHuang Ying * This does not guarantee that the page exists in the user mappings when 16890014bd99SHuang Ying * __get_user_pages returns, and there may even be a completely different 16900014bd99SHuang Ying * page there in some cases (eg. if mmapped pagecache has been invalidated 16910014bd99SHuang Ying * and subsequently re faulted). However it does guarantee that the page 16920014bd99SHuang Ying * won't be freed completely. And mostly callers simply care that the page 16930014bd99SHuang Ying * contains data that was valid *at some point in time*. Typically, an IO 16940014bd99SHuang Ying * or similar operation cannot guarantee anything stronger anyway because 16950014bd99SHuang Ying * locks can't be held over the syscall boundary. 16960014bd99SHuang Ying * 16970014bd99SHuang Ying * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If 16980014bd99SHuang Ying * the page is written to, set_page_dirty (or set_page_dirty_lock, as 16990014bd99SHuang Ying * appropriate) must be called after the page is finished with, and 17000014bd99SHuang Ying * before put_page is called. 17010014bd99SHuang Ying * 17020014bd99SHuang Ying * If @nonblocking != NULL, __get_user_pages will not wait for disk IO 17030014bd99SHuang Ying * or mmap_sem contention, and if waiting is needed to pin all pages, 17040014bd99SHuang Ying * *@nonblocking will be set to 0. 17050014bd99SHuang Ying * 17060014bd99SHuang Ying * In most cases, get_user_pages or get_user_pages_fast should be used 17070014bd99SHuang Ying * instead of __get_user_pages. __get_user_pages should be used only if 17080014bd99SHuang Ying * you need some special @gup_flags. 17090014bd99SHuang Ying */ 171028a35716SMichel Lespinasse long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 171128a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 171228a35716SMichel Lespinasse unsigned int gup_flags, struct page **pages, 171328a35716SMichel Lespinasse struct vm_area_struct **vmas, int *nonblocking) 17141da177e4SLinus Torvalds { 171528a35716SMichel Lespinasse long i; 171658fa879eSHugh Dickins unsigned long vm_flags; 1717240aadeeSMichel Lespinasse unsigned int page_mask; 17181da177e4SLinus Torvalds 171928a35716SMichel Lespinasse if (!nr_pages) 1720900cf086SJonathan Corbet return 0; 172158fa879eSHugh Dickins 172258fa879eSHugh Dickins VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); 172358fa879eSHugh Dickins 17241da177e4SLinus Torvalds /* 17250b9d7052SAndrea Arcangeli * If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault 17260b9d7052SAndrea Arcangeli * would be called on PROT_NONE ranges. We must never invoke 17270b9d7052SAndrea Arcangeli * handle_mm_fault on PROT_NONE ranges or the NUMA hinting 17280b9d7052SAndrea Arcangeli * page faults would unprotect the PROT_NONE ranges if 17290b9d7052SAndrea Arcangeli * _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd 17300b9d7052SAndrea Arcangeli * bitflag. So to avoid that, don't set FOLL_NUMA if 17310b9d7052SAndrea Arcangeli * FOLL_FORCE is set. 17320b9d7052SAndrea Arcangeli */ 17330b9d7052SAndrea Arcangeli if (!(gup_flags & FOLL_FORCE)) 17340b9d7052SAndrea Arcangeli gup_flags |= FOLL_NUMA; 17350b9d7052SAndrea Arcangeli 17361da177e4SLinus Torvalds i = 0; 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds do { 17391da177e4SLinus Torvalds struct vm_area_struct *vma; 17401da177e4SLinus Torvalds 17411da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 1742e7f22e20SStephen Wilson if (!vma && in_gate_area(mm, start)) { 17431da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 17441da177e4SLinus Torvalds pgd_t *pgd; 17451da177e4SLinus Torvalds pud_t *pud; 17461da177e4SLinus Torvalds pmd_t *pmd; 17471da177e4SLinus Torvalds pte_t *pte; 1748b291f000SNick Piggin 1749b291f000SNick Piggin /* user gate pages are read-only */ 175058fa879eSHugh Dickins if (gup_flags & FOLL_WRITE) 1751cda540acSHugh Dickins goto efault; 17521da177e4SLinus Torvalds if (pg > TASK_SIZE) 17531da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 17541da177e4SLinus Torvalds else 17551da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 17561da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 17571da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 17581da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 17591da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1760690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1761cda540acSHugh Dickins goto efault; 1762f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 17631da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1764690dbe1cSHugh Dickins if (pte_none(*pte)) { 1765690dbe1cSHugh Dickins pte_unmap(pte); 1766cda540acSHugh Dickins goto efault; 1767690dbe1cSHugh Dickins } 176895042f9eSLinus Torvalds vma = get_gate_vma(mm); 17691da177e4SLinus Torvalds if (pages) { 1770de51257aSHugh Dickins struct page *page; 1771de51257aSHugh Dickins 177295042f9eSLinus Torvalds page = vm_normal_page(vma, start, *pte); 1773de51257aSHugh Dickins if (!page) { 1774de51257aSHugh Dickins if (!(gup_flags & FOLL_DUMP) && 1775de51257aSHugh Dickins is_zero_pfn(pte_pfn(*pte))) 1776de51257aSHugh Dickins page = pte_page(*pte); 1777de51257aSHugh Dickins else { 1778de51257aSHugh Dickins pte_unmap(pte); 1779cda540acSHugh Dickins goto efault; 1780de51257aSHugh Dickins } 1781de51257aSHugh Dickins } 17826aab341eSLinus Torvalds pages[i] = page; 17836aab341eSLinus Torvalds get_page(page); 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds pte_unmap(pte); 1786240aadeeSMichel Lespinasse page_mask = 0; 178795042f9eSLinus Torvalds goto next_page; 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds 1790cda540acSHugh Dickins if (!vma) 1791cda540acSHugh Dickins goto efault; 1792cda540acSHugh Dickins vm_flags = vma->vm_flags; 1793cda540acSHugh Dickins if (vm_flags & (VM_IO | VM_PFNMAP)) 1794cda540acSHugh Dickins goto efault; 1795cda540acSHugh Dickins 1796cda540acSHugh Dickins if (gup_flags & FOLL_WRITE) { 1797cda540acSHugh Dickins if (!(vm_flags & VM_WRITE)) { 1798cda540acSHugh Dickins if (!(gup_flags & FOLL_FORCE)) 1799cda540acSHugh Dickins goto efault; 1800cda540acSHugh Dickins /* 1801cda540acSHugh Dickins * We used to let the write,force case do COW 1802cda540acSHugh Dickins * in a VM_MAYWRITE VM_SHARED !VM_WRITE vma, so 1803cda540acSHugh Dickins * ptrace could set a breakpoint in a read-only 1804cda540acSHugh Dickins * mapping of an executable, without corrupting 1805cda540acSHugh Dickins * the file (yet only when that file had been 1806cda540acSHugh Dickins * opened for writing!). Anon pages in shared 1807cda540acSHugh Dickins * mappings are surprising: now just reject it. 1808cda540acSHugh Dickins */ 1809cda540acSHugh Dickins if (!is_cow_mapping(vm_flags)) { 1810cda540acSHugh Dickins WARN_ON_ONCE(vm_flags & VM_MAYWRITE); 1811cda540acSHugh Dickins goto efault; 1812cda540acSHugh Dickins } 1813cda540acSHugh Dickins } 1814cda540acSHugh Dickins } else { 1815cda540acSHugh Dickins if (!(vm_flags & VM_READ)) { 1816cda540acSHugh Dickins if (!(gup_flags & FOLL_FORCE)) 1817cda540acSHugh Dickins goto efault; 1818cda540acSHugh Dickins /* 1819cda540acSHugh Dickins * Is there actually any vma we can reach here 1820cda540acSHugh Dickins * which does not have VM_MAYREAD set? 1821cda540acSHugh Dickins */ 1822cda540acSHugh Dickins if (!(vm_flags & VM_MAYREAD)) 1823cda540acSHugh Dickins goto efault; 1824cda540acSHugh Dickins } 1825cda540acSHugh Dickins } 18261da177e4SLinus Torvalds 18272a15efc9SHugh Dickins if (is_vm_hugetlb_page(vma)) { 18282a15efc9SHugh Dickins i = follow_hugetlb_page(mm, vma, pages, vmas, 182958fa879eSHugh Dickins &start, &nr_pages, i, gup_flags); 18302a15efc9SHugh Dickins continue; 18312a15efc9SHugh Dickins } 1832deceb6cdSHugh Dickins 18331da177e4SLinus Torvalds do { 183408ef4729SHugh Dickins struct page *page; 183558fa879eSHugh Dickins unsigned int foll_flags = gup_flags; 1836240aadeeSMichel Lespinasse unsigned int page_increm; 18371da177e4SLinus Torvalds 1838462e00ccSEthan Solomita /* 18394779280dSYing Han * If we have a pending SIGKILL, don't keep faulting 18401c3aff1cSHugh Dickins * pages and potentially allocating memory. 1841462e00ccSEthan Solomita */ 18421c3aff1cSHugh Dickins if (unlikely(fatal_signal_pending(current))) 18434779280dSYing Han return i ? i : -ERESTARTSYS; 1844462e00ccSEthan Solomita 1845deceb6cdSHugh Dickins cond_resched(); 1846240aadeeSMichel Lespinasse while (!(page = follow_page_mask(vma, start, 1847240aadeeSMichel Lespinasse foll_flags, &page_mask))) { 1848a68d2ebcSLinus Torvalds int ret; 184953a7706dSMichel Lespinasse unsigned int fault_flags = 0; 185053a7706dSMichel Lespinasse 1851a09a79f6SMikulas Patocka /* For mlock, just skip the stack guard page. */ 1852a09a79f6SMikulas Patocka if (foll_flags & FOLL_MLOCK) { 1853a09a79f6SMikulas Patocka if (stack_guard_page(vma, start)) 1854a09a79f6SMikulas Patocka goto next_page; 1855a09a79f6SMikulas Patocka } 185653a7706dSMichel Lespinasse if (foll_flags & FOLL_WRITE) 185753a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_WRITE; 185853a7706dSMichel Lespinasse if (nonblocking) 185953a7706dSMichel Lespinasse fault_flags |= FAULT_FLAG_ALLOW_RETRY; 1860318b275fSGleb Natapov if (foll_flags & FOLL_NOWAIT) 1861318b275fSGleb Natapov fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); 1862d06063ccSLinus Torvalds 1863d26ed650SHugh Dickins ret = handle_mm_fault(mm, vma, start, 186453a7706dSMichel Lespinasse fault_flags); 1865d26ed650SHugh Dickins 186683c54070SNick Piggin if (ret & VM_FAULT_ERROR) { 186783c54070SNick Piggin if (ret & VM_FAULT_OOM) 186883c54070SNick Piggin return i ? i : -ENOMEM; 186969ebb83eSHuang Ying if (ret & (VM_FAULT_HWPOISON | 187069ebb83eSHuang Ying VM_FAULT_HWPOISON_LARGE)) { 187169ebb83eSHuang Ying if (i) 187269ebb83eSHuang Ying return i; 187369ebb83eSHuang Ying else if (gup_flags & FOLL_HWPOISON) 187469ebb83eSHuang Ying return -EHWPOISON; 187569ebb83eSHuang Ying else 187669ebb83eSHuang Ying return -EFAULT; 187769ebb83eSHuang Ying } 187869ebb83eSHuang Ying if (ret & VM_FAULT_SIGBUS) 1879cda540acSHugh Dickins goto efault; 188083c54070SNick Piggin BUG(); 188183c54070SNick Piggin } 1882e7f22e20SStephen Wilson 1883e7f22e20SStephen Wilson if (tsk) { 188483c54070SNick Piggin if (ret & VM_FAULT_MAJOR) 188583c54070SNick Piggin tsk->maj_flt++; 188683c54070SNick Piggin else 188783c54070SNick Piggin tsk->min_flt++; 1888e7f22e20SStephen Wilson } 188983c54070SNick Piggin 189053a7706dSMichel Lespinasse if (ret & VM_FAULT_RETRY) { 1891318b275fSGleb Natapov if (nonblocking) 189253a7706dSMichel Lespinasse *nonblocking = 0; 189353a7706dSMichel Lespinasse return i; 189453a7706dSMichel Lespinasse } 189553a7706dSMichel Lespinasse 1896f33ea7f4SNick Piggin /* 189783c54070SNick Piggin * The VM_FAULT_WRITE bit tells us that 189883c54070SNick Piggin * do_wp_page has broken COW when necessary, 189983c54070SNick Piggin * even if maybe_mkwrite decided not to set 190083c54070SNick Piggin * pte_write. We can thus safely do subsequent 1901878b63acSHugh Dickins * page lookups as if they were reads. But only 1902878b63acSHugh Dickins * do so when looping for pte_write is futile: 1903878b63acSHugh Dickins * in some cases userspace may also be wanting 1904878b63acSHugh Dickins * to write to the gotten user page, which a 1905878b63acSHugh Dickins * read fault here might prevent (a readonly 1906878b63acSHugh Dickins * page might get reCOWed by userspace write). 1907f33ea7f4SNick Piggin */ 1908878b63acSHugh Dickins if ((ret & VM_FAULT_WRITE) && 1909878b63acSHugh Dickins !(vma->vm_flags & VM_WRITE)) 1910deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1911a68d2ebcSLinus Torvalds 19127f7bbbe5SBenjamin Herrenschmidt cond_resched(); 19131da177e4SLinus Torvalds } 191489f5b7daSLinus Torvalds if (IS_ERR(page)) 191589f5b7daSLinus Torvalds return i ? i : PTR_ERR(page); 19161da177e4SLinus Torvalds if (pages) { 191708ef4729SHugh Dickins pages[i] = page; 191803beb076SJames Bottomley 1919a6f36be3SRussell King flush_anon_page(vma, page, start); 192008ef4729SHugh Dickins flush_dcache_page(page); 1921240aadeeSMichel Lespinasse page_mask = 0; 19221da177e4SLinus Torvalds } 192395042f9eSLinus Torvalds next_page: 1924240aadeeSMichel Lespinasse if (vmas) { 19251da177e4SLinus Torvalds vmas[i] = vma; 1926240aadeeSMichel Lespinasse page_mask = 0; 1927240aadeeSMichel Lespinasse } 1928240aadeeSMichel Lespinasse page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask); 1929240aadeeSMichel Lespinasse if (page_increm > nr_pages) 1930240aadeeSMichel Lespinasse page_increm = nr_pages; 1931240aadeeSMichel Lespinasse i += page_increm; 1932240aadeeSMichel Lespinasse start += page_increm * PAGE_SIZE; 1933240aadeeSMichel Lespinasse nr_pages -= page_increm; 19349d73777eSPeter Zijlstra } while (nr_pages && start < vma->vm_end); 19359d73777eSPeter Zijlstra } while (nr_pages); 19361da177e4SLinus Torvalds return i; 1937cda540acSHugh Dickins efault: 1938cda540acSHugh Dickins return i ? : -EFAULT; 19391da177e4SLinus Torvalds } 19400014bd99SHuang Ying EXPORT_SYMBOL(__get_user_pages); 1941b291f000SNick Piggin 19422efaca92SBenjamin Herrenschmidt /* 19432efaca92SBenjamin Herrenschmidt * fixup_user_fault() - manually resolve a user page fault 19442efaca92SBenjamin Herrenschmidt * @tsk: the task_struct to use for page fault accounting, or 19452efaca92SBenjamin Herrenschmidt * NULL if faults are not to be recorded. 19462efaca92SBenjamin Herrenschmidt * @mm: mm_struct of target mm 19472efaca92SBenjamin Herrenschmidt * @address: user address 19482efaca92SBenjamin Herrenschmidt * @fault_flags:flags to pass down to handle_mm_fault() 19492efaca92SBenjamin Herrenschmidt * 19502efaca92SBenjamin Herrenschmidt * This is meant to be called in the specific scenario where for locking reasons 19512efaca92SBenjamin Herrenschmidt * we try to access user memory in atomic context (within a pagefault_disable() 19522efaca92SBenjamin Herrenschmidt * section), this returns -EFAULT, and we want to resolve the user fault before 19532efaca92SBenjamin Herrenschmidt * trying again. 19542efaca92SBenjamin Herrenschmidt * 19552efaca92SBenjamin Herrenschmidt * Typically this is meant to be used by the futex code. 19562efaca92SBenjamin Herrenschmidt * 19572efaca92SBenjamin Herrenschmidt * The main difference with get_user_pages() is that this function will 19582efaca92SBenjamin Herrenschmidt * unconditionally call handle_mm_fault() which will in turn perform all the 19592efaca92SBenjamin Herrenschmidt * necessary SW fixup of the dirty and young bits in the PTE, while 19602efaca92SBenjamin Herrenschmidt * handle_mm_fault() only guarantees to update these in the struct page. 19612efaca92SBenjamin Herrenschmidt * 19622efaca92SBenjamin Herrenschmidt * This is important for some architectures where those bits also gate the 19632efaca92SBenjamin Herrenschmidt * access permission to the page because they are maintained in software. On 19642efaca92SBenjamin Herrenschmidt * such architectures, gup() will not be enough to make a subsequent access 19652efaca92SBenjamin Herrenschmidt * succeed. 19662efaca92SBenjamin Herrenschmidt * 19672efaca92SBenjamin Herrenschmidt * This should be called with the mm_sem held for read. 19682efaca92SBenjamin Herrenschmidt */ 19692efaca92SBenjamin Herrenschmidt int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm, 19702efaca92SBenjamin Herrenschmidt unsigned long address, unsigned int fault_flags) 19712efaca92SBenjamin Herrenschmidt { 19722efaca92SBenjamin Herrenschmidt struct vm_area_struct *vma; 19731b17844bSLinus Torvalds vm_flags_t vm_flags; 19742efaca92SBenjamin Herrenschmidt int ret; 19752efaca92SBenjamin Herrenschmidt 19762efaca92SBenjamin Herrenschmidt vma = find_extend_vma(mm, address); 19772efaca92SBenjamin Herrenschmidt if (!vma || address < vma->vm_start) 19782efaca92SBenjamin Herrenschmidt return -EFAULT; 19792efaca92SBenjamin Herrenschmidt 19801b17844bSLinus Torvalds vm_flags = (fault_flags & FAULT_FLAG_WRITE) ? VM_WRITE : VM_READ; 19811b17844bSLinus Torvalds if (!(vm_flags & vma->vm_flags)) 19821b17844bSLinus Torvalds return -EFAULT; 19831b17844bSLinus Torvalds 19842efaca92SBenjamin Herrenschmidt ret = handle_mm_fault(mm, vma, address, fault_flags); 19852efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_ERROR) { 19862efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_OOM) 19872efaca92SBenjamin Herrenschmidt return -ENOMEM; 19882efaca92SBenjamin Herrenschmidt if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) 19892efaca92SBenjamin Herrenschmidt return -EHWPOISON; 19902efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_SIGBUS) 19912efaca92SBenjamin Herrenschmidt return -EFAULT; 19922efaca92SBenjamin Herrenschmidt BUG(); 19932efaca92SBenjamin Herrenschmidt } 19942efaca92SBenjamin Herrenschmidt if (tsk) { 19952efaca92SBenjamin Herrenschmidt if (ret & VM_FAULT_MAJOR) 19962efaca92SBenjamin Herrenschmidt tsk->maj_flt++; 19972efaca92SBenjamin Herrenschmidt else 19982efaca92SBenjamin Herrenschmidt tsk->min_flt++; 19992efaca92SBenjamin Herrenschmidt } 20002efaca92SBenjamin Herrenschmidt return 0; 20012efaca92SBenjamin Herrenschmidt } 20022efaca92SBenjamin Herrenschmidt 20032efaca92SBenjamin Herrenschmidt /* 2004d2bf6be8SNick Piggin * get_user_pages() - pin user pages in memory 2005e7f22e20SStephen Wilson * @tsk: the task_struct to use for page fault accounting, or 2006e7f22e20SStephen Wilson * NULL if faults are not to be recorded. 2007d2bf6be8SNick Piggin * @mm: mm_struct of target mm 2008d2bf6be8SNick Piggin * @start: starting user address 20099d73777eSPeter Zijlstra * @nr_pages: number of pages from start to pin 2010d2bf6be8SNick Piggin * @write: whether pages will be written to by the caller 2011cda540acSHugh Dickins * @force: whether to force access even when user mapping is currently 2012cda540acSHugh Dickins * protected (but never forces write access to shared mapping). 2013d2bf6be8SNick Piggin * @pages: array that receives pointers to the pages pinned. 2014d2bf6be8SNick Piggin * Should be at least nr_pages long. Or NULL, if caller 2015d2bf6be8SNick Piggin * only intends to ensure the pages are faulted in. 2016d2bf6be8SNick Piggin * @vmas: array of pointers to vmas corresponding to each page. 2017d2bf6be8SNick Piggin * Or NULL if the caller does not require them. 2018d2bf6be8SNick Piggin * 2019d2bf6be8SNick Piggin * Returns number of pages pinned. This may be fewer than the number 20209d73777eSPeter Zijlstra * requested. If nr_pages is 0 or negative, returns 0. If no pages 2021d2bf6be8SNick Piggin * were pinned, returns -errno. Each page returned must be released 2022d2bf6be8SNick Piggin * with a put_page() call when it is finished with. vmas will only 2023d2bf6be8SNick Piggin * remain valid while mmap_sem is held. 2024d2bf6be8SNick Piggin * 2025d2bf6be8SNick Piggin * Must be called with mmap_sem held for read or write. 2026d2bf6be8SNick Piggin * 2027d2bf6be8SNick Piggin * get_user_pages walks a process's page tables and takes a reference to 2028d2bf6be8SNick Piggin * each struct page that each user address corresponds to at a given 2029d2bf6be8SNick Piggin * instant. That is, it takes the page that would be accessed if a user 2030d2bf6be8SNick Piggin * thread accesses the given user virtual address at that instant. 2031d2bf6be8SNick Piggin * 2032d2bf6be8SNick Piggin * This does not guarantee that the page exists in the user mappings when 2033d2bf6be8SNick Piggin * get_user_pages returns, and there may even be a completely different 2034d2bf6be8SNick Piggin * page there in some cases (eg. if mmapped pagecache has been invalidated 2035d2bf6be8SNick Piggin * and subsequently re faulted). However it does guarantee that the page 2036d2bf6be8SNick Piggin * won't be freed completely. And mostly callers simply care that the page 2037d2bf6be8SNick Piggin * contains data that was valid *at some point in time*. Typically, an IO 2038d2bf6be8SNick Piggin * or similar operation cannot guarantee anything stronger anyway because 2039d2bf6be8SNick Piggin * locks can't be held over the syscall boundary. 2040d2bf6be8SNick Piggin * 2041d2bf6be8SNick Piggin * If write=0, the page must not be written to. If the page is written to, 2042d2bf6be8SNick Piggin * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called 2043d2bf6be8SNick Piggin * after the page is finished with, and before put_page is called. 2044d2bf6be8SNick Piggin * 2045d2bf6be8SNick Piggin * get_user_pages is typically used for fewer-copy IO operations, to get a 2046d2bf6be8SNick Piggin * handle on the memory by some means other than accesses via the user virtual 2047d2bf6be8SNick Piggin * addresses. The pages may be submitted for DMA to devices or accessed via 2048d2bf6be8SNick Piggin * their kernel linear mapping (via the kmap APIs). Care should be taken to 2049d2bf6be8SNick Piggin * use the correct cache flushing APIs. 2050d2bf6be8SNick Piggin * 2051d2bf6be8SNick Piggin * See also get_user_pages_fast, for performance critical applications. 2052d2bf6be8SNick Piggin */ 205328a35716SMichel Lespinasse long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 205428a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, int write, 205528a35716SMichel Lespinasse int force, struct page **pages, struct vm_area_struct **vmas) 2056b291f000SNick Piggin { 205758fa879eSHugh Dickins int flags = FOLL_TOUCH; 2058b291f000SNick Piggin 205958fa879eSHugh Dickins if (pages) 206058fa879eSHugh Dickins flags |= FOLL_GET; 2061b291f000SNick Piggin if (write) 206258fa879eSHugh Dickins flags |= FOLL_WRITE; 2063b291f000SNick Piggin if (force) 206458fa879eSHugh Dickins flags |= FOLL_FORCE; 2065b291f000SNick Piggin 206653a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 206753a7706dSMichel Lespinasse NULL); 2068b291f000SNick Piggin } 20691da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 20701da177e4SLinus Torvalds 2071f3e8fccdSHugh Dickins /** 2072f3e8fccdSHugh Dickins * get_dump_page() - pin user page in memory while writing it to core dump 2073f3e8fccdSHugh Dickins * @addr: user address 2074f3e8fccdSHugh Dickins * 2075f3e8fccdSHugh Dickins * Returns struct page pointer of user page pinned for dump, 2076f3e8fccdSHugh Dickins * to be freed afterwards by page_cache_release() or put_page(). 2077f3e8fccdSHugh Dickins * 2078f3e8fccdSHugh Dickins * Returns NULL on any kind of failure - a hole must then be inserted into 2079f3e8fccdSHugh Dickins * the corefile, to preserve alignment with its headers; and also returns 2080f3e8fccdSHugh Dickins * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - 2081f3e8fccdSHugh Dickins * allowing a hole to be left in the corefile to save diskspace. 2082f3e8fccdSHugh Dickins * 2083f3e8fccdSHugh Dickins * Called without mmap_sem, but after all other threads have been killed. 2084f3e8fccdSHugh Dickins */ 2085f3e8fccdSHugh Dickins #ifdef CONFIG_ELF_CORE 2086f3e8fccdSHugh Dickins struct page *get_dump_page(unsigned long addr) 2087f3e8fccdSHugh Dickins { 2088f3e8fccdSHugh Dickins struct vm_area_struct *vma; 2089f3e8fccdSHugh Dickins struct page *page; 2090f3e8fccdSHugh Dickins 2091f3e8fccdSHugh Dickins if (__get_user_pages(current, current->mm, addr, 1, 209253a7706dSMichel Lespinasse FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, 209353a7706dSMichel Lespinasse NULL) < 1) 2094f3e8fccdSHugh Dickins return NULL; 2095f3e8fccdSHugh Dickins flush_cache_page(vma, addr, page_to_pfn(page)); 2096f3e8fccdSHugh Dickins return page; 2097f3e8fccdSHugh Dickins } 2098f3e8fccdSHugh Dickins #endif /* CONFIG_ELF_CORE */ 2099f3e8fccdSHugh Dickins 210025ca1d6cSNamhyung Kim pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 2101920c7a5dSHarvey Harrison spinlock_t **ptl) 2102c9cfcddfSLinus Torvalds { 2103c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 2104c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 2105c9cfcddfSLinus Torvalds if (pud) { 210649c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 2107f66055abSAndrea Arcangeli if (pmd) { 2108f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 2109c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 2110c9cfcddfSLinus Torvalds } 2111f66055abSAndrea Arcangeli } 2112c9cfcddfSLinus Torvalds return NULL; 2113c9cfcddfSLinus Torvalds } 2114c9cfcddfSLinus Torvalds 21151da177e4SLinus Torvalds /* 2116238f58d8SLinus Torvalds * This is the old fallback for page remapping. 2117238f58d8SLinus Torvalds * 2118238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 2119238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 2120238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 2121238f58d8SLinus Torvalds */ 2122423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 2123423bad60SNick Piggin struct page *page, pgprot_t prot) 2124238f58d8SLinus Torvalds { 2125423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2126238f58d8SLinus Torvalds int retval; 2127238f58d8SLinus Torvalds pte_t *pte; 2128238f58d8SLinus Torvalds spinlock_t *ptl; 2129238f58d8SLinus Torvalds 2130238f58d8SLinus Torvalds retval = -EINVAL; 2131a145dd41SLinus Torvalds if (PageAnon(page)) 21325b4e655eSKAMEZAWA Hiroyuki goto out; 2133238f58d8SLinus Torvalds retval = -ENOMEM; 2134238f58d8SLinus Torvalds flush_dcache_page(page); 2135c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 2136238f58d8SLinus Torvalds if (!pte) 21375b4e655eSKAMEZAWA Hiroyuki goto out; 2138238f58d8SLinus Torvalds retval = -EBUSY; 2139238f58d8SLinus Torvalds if (!pte_none(*pte)) 2140238f58d8SLinus Torvalds goto out_unlock; 2141238f58d8SLinus Torvalds 2142238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 2143238f58d8SLinus Torvalds get_page(page); 214434e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_FILEPAGES); 2145238f58d8SLinus Torvalds page_add_file_rmap(page); 2146238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 2147238f58d8SLinus Torvalds 2148238f58d8SLinus Torvalds retval = 0; 21498a9f3ccdSBalbir Singh pte_unmap_unlock(pte, ptl); 21508a9f3ccdSBalbir Singh return retval; 2151238f58d8SLinus Torvalds out_unlock: 2152238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 2153238f58d8SLinus Torvalds out: 2154238f58d8SLinus Torvalds return retval; 2155238f58d8SLinus Torvalds } 2156238f58d8SLinus Torvalds 2157bfa5bf6dSRolf Eike Beer /** 2158bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 2159bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2160bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 2161bfa5bf6dSRolf Eike Beer * @page: source kernel page 2162bfa5bf6dSRolf Eike Beer * 2163a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 2164a145dd41SLinus Torvalds * into a user vma. 2165a145dd41SLinus Torvalds * 2166a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 2167a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 2168a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 21698dfcc9baSNick Piggin * (see split_page()). 2170a145dd41SLinus Torvalds * 2171a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 2172a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 2173a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 2174a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 2175a145dd41SLinus Torvalds * ask for a shared writable mapping! 2176a145dd41SLinus Torvalds * 2177a145dd41SLinus Torvalds * The page does not need to be reserved. 21784b6e1e37SKonstantin Khlebnikov * 21794b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 21804b6e1e37SKonstantin Khlebnikov * under mm->mmap_sem write-lock, so it can change vma->vm_flags. 21814b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 21824b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 2183a145dd41SLinus Torvalds */ 2184423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 2185423bad60SNick Piggin struct page *page) 2186a145dd41SLinus Torvalds { 2187a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 2188a145dd41SLinus Torvalds return -EFAULT; 2189a145dd41SLinus Torvalds if (!page_count(page)) 2190a145dd41SLinus Torvalds return -EINVAL; 21914b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 21924b6e1e37SKonstantin Khlebnikov BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); 21934b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 21944b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 21954b6e1e37SKonstantin Khlebnikov } 2196423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2197a145dd41SLinus Torvalds } 2198e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 2199a145dd41SLinus Torvalds 2200423bad60SNick Piggin static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2201423bad60SNick Piggin unsigned long pfn, pgprot_t prot) 2202423bad60SNick Piggin { 2203423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 2204423bad60SNick Piggin int retval; 2205423bad60SNick Piggin pte_t *pte, entry; 2206423bad60SNick Piggin spinlock_t *ptl; 2207423bad60SNick Piggin 2208423bad60SNick Piggin retval = -ENOMEM; 2209423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 2210423bad60SNick Piggin if (!pte) 2211423bad60SNick Piggin goto out; 2212423bad60SNick Piggin retval = -EBUSY; 2213423bad60SNick Piggin if (!pte_none(*pte)) 2214423bad60SNick Piggin goto out_unlock; 2215423bad60SNick Piggin 2216423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 2217423bad60SNick Piggin entry = pte_mkspecial(pfn_pte(pfn, prot)); 2218423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 22194b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 2220423bad60SNick Piggin 2221423bad60SNick Piggin retval = 0; 2222423bad60SNick Piggin out_unlock: 2223423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 2224423bad60SNick Piggin out: 2225423bad60SNick Piggin return retval; 2226423bad60SNick Piggin } 2227423bad60SNick Piggin 2228e0dc0d8fSNick Piggin /** 2229e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 2230e0dc0d8fSNick Piggin * @vma: user vma to map to 2231e0dc0d8fSNick Piggin * @addr: target user address of this page 2232e0dc0d8fSNick Piggin * @pfn: source kernel pfn 2233e0dc0d8fSNick Piggin * 2234c462f179SRobert P. J. Day * Similar to vm_insert_page, this allows drivers to insert individual pages 2235e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 2236e0dc0d8fSNick Piggin * 2237e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 2238e0dc0d8fSNick Piggin * in that case the handler should return NULL. 22390d71d10aSNick Piggin * 22400d71d10aSNick Piggin * vma cannot be a COW mapping. 22410d71d10aSNick Piggin * 22420d71d10aSNick Piggin * As this is called only for pages that do not currently exist, we 22430d71d10aSNick Piggin * do not need to flush old virtual caches or the TLB. 2244e0dc0d8fSNick Piggin */ 2245e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2246e0dc0d8fSNick Piggin unsigned long pfn) 2247e0dc0d8fSNick Piggin { 22482ab64037Svenkatesh.pallipadi@intel.com int ret; 2249e4b866edSvenkatesh.pallipadi@intel.com pgprot_t pgprot = vma->vm_page_prot; 22507e675137SNick Piggin /* 22517e675137SNick Piggin * Technically, architectures with pte_special can avoid all these 22527e675137SNick Piggin * restrictions (same for remap_pfn_range). However we would like 22537e675137SNick Piggin * consistency in testing and feature parity among all, so we should 22547e675137SNick Piggin * try to keep these invariants in place for everybody. 22557e675137SNick Piggin */ 2256b379d790SJared Hulbert BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 2257b379d790SJared Hulbert BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 2258b379d790SJared Hulbert (VM_PFNMAP|VM_MIXEDMAP)); 2259b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 2260b379d790SJared Hulbert BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 2261e0dc0d8fSNick Piggin 2262423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2263423bad60SNick Piggin return -EFAULT; 22645180da41SSuresh Siddha if (track_pfn_insert(vma, &pgprot, pfn)) 22652ab64037Svenkatesh.pallipadi@intel.com return -EINVAL; 22662ab64037Svenkatesh.pallipadi@intel.com 2267e4b866edSvenkatesh.pallipadi@intel.com ret = insert_pfn(vma, addr, pfn, pgprot); 22682ab64037Svenkatesh.pallipadi@intel.com 22692ab64037Svenkatesh.pallipadi@intel.com return ret; 2270e0dc0d8fSNick Piggin } 2271e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 2272e0dc0d8fSNick Piggin 2273423bad60SNick Piggin int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 2274423bad60SNick Piggin unsigned long pfn) 2275423bad60SNick Piggin { 2276423bad60SNick Piggin BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); 2277423bad60SNick Piggin 2278423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 2279423bad60SNick Piggin return -EFAULT; 2280423bad60SNick Piggin 2281423bad60SNick Piggin /* 2282423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2283423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2284423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 228562eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 228662eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2287423bad60SNick Piggin */ 2288423bad60SNick Piggin if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { 2289423bad60SNick Piggin struct page *page; 2290423bad60SNick Piggin 2291423bad60SNick Piggin page = pfn_to_page(pfn); 2292423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 2293423bad60SNick Piggin } 2294423bad60SNick Piggin return insert_pfn(vma, addr, pfn, vma->vm_page_prot); 2295423bad60SNick Piggin } 2296423bad60SNick Piggin EXPORT_SYMBOL(vm_insert_mixed); 2297423bad60SNick Piggin 2298a145dd41SLinus Torvalds /* 22991da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 23001da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 23011da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 23021da177e4SLinus Torvalds */ 23031da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 23041da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23051da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23061da177e4SLinus Torvalds { 23071da177e4SLinus Torvalds pte_t *pte; 2308c74df32cSHugh Dickins spinlock_t *ptl; 23091da177e4SLinus Torvalds 2310c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 23111da177e4SLinus Torvalds if (!pte) 23121da177e4SLinus Torvalds return -ENOMEM; 23136606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 23141da177e4SLinus Torvalds do { 23151da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 23167e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 23171da177e4SLinus Torvalds pfn++; 23181da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 23196606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 2320c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 23211da177e4SLinus Torvalds return 0; 23221da177e4SLinus Torvalds } 23231da177e4SLinus Torvalds 23241da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 23251da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23261da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23271da177e4SLinus Torvalds { 23281da177e4SLinus Torvalds pmd_t *pmd; 23291da177e4SLinus Torvalds unsigned long next; 23301da177e4SLinus Torvalds 23311da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23321da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 23331da177e4SLinus Torvalds if (!pmd) 23341da177e4SLinus Torvalds return -ENOMEM; 2335f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 23361da177e4SLinus Torvalds do { 23371da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 23381da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 23391da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 23401da177e4SLinus Torvalds return -ENOMEM; 23411da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 23421da177e4SLinus Torvalds return 0; 23431da177e4SLinus Torvalds } 23441da177e4SLinus Torvalds 23451da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 23461da177e4SLinus Torvalds unsigned long addr, unsigned long end, 23471da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 23481da177e4SLinus Torvalds { 23491da177e4SLinus Torvalds pud_t *pud; 23501da177e4SLinus Torvalds unsigned long next; 23511da177e4SLinus Torvalds 23521da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23531da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 23541da177e4SLinus Torvalds if (!pud) 23551da177e4SLinus Torvalds return -ENOMEM; 23561da177e4SLinus Torvalds do { 23571da177e4SLinus Torvalds next = pud_addr_end(addr, end); 23581da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 23591da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 23601da177e4SLinus Torvalds return -ENOMEM; 23611da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 23621da177e4SLinus Torvalds return 0; 23631da177e4SLinus Torvalds } 23641da177e4SLinus Torvalds 2365bfa5bf6dSRolf Eike Beer /** 2366bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 2367bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 2368bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 2369bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 2370bfa5bf6dSRolf Eike Beer * @size: size of map area 2371bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 2372bfa5bf6dSRolf Eike Beer * 2373bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 2374bfa5bf6dSRolf Eike Beer */ 23751da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 23761da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 23771da177e4SLinus Torvalds { 23781da177e4SLinus Torvalds pgd_t *pgd; 23791da177e4SLinus Torvalds unsigned long next; 23802d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 23811da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23821da177e4SLinus Torvalds int err; 23831da177e4SLinus Torvalds 23841da177e4SLinus Torvalds /* 23851da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 23861da177e4SLinus Torvalds * rest of the world about it: 23871da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 23881da177e4SLinus Torvalds * (accesses can have side effects). 23896aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 23906aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 23916aab341eSLinus Torvalds * with them. 2392314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2393314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2394314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2395314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2396fb155c16SLinus Torvalds * 2397fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2398fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2399fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2400b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 24011da177e4SLinus Torvalds */ 2402b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2403b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2404b3b9c293SKonstantin Khlebnikov return -EINVAL; 24056aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2406b3b9c293SKonstantin Khlebnikov } 2407b3b9c293SKonstantin Khlebnikov 2408b3b9c293SKonstantin Khlebnikov err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); 2409b3b9c293SKonstantin Khlebnikov if (err) 24103c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 2411fb155c16SLinus Torvalds 2412314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds BUG_ON(addr >= end); 24151da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 24161da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 24171da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 24181da177e4SLinus Torvalds do { 24191da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 24201da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 24211da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 24221da177e4SLinus Torvalds if (err) 24231da177e4SLinus Torvalds break; 24241da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 24252ab64037Svenkatesh.pallipadi@intel.com 24262ab64037Svenkatesh.pallipadi@intel.com if (err) 24275180da41SSuresh Siddha untrack_pfn(vma, pfn, PAGE_ALIGN(size)); 24282ab64037Svenkatesh.pallipadi@intel.com 24291da177e4SLinus Torvalds return err; 24301da177e4SLinus Torvalds } 24311da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 24321da177e4SLinus Torvalds 2433b4cbb197SLinus Torvalds /** 2434b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2435b4cbb197SLinus Torvalds * @vma: user vma to map to 2436b4cbb197SLinus Torvalds * @start: start of area 2437b4cbb197SLinus Torvalds * @len: size of area 2438b4cbb197SLinus Torvalds * 2439b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2440b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2441b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2442b4cbb197SLinus Torvalds * 2443b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2444b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2445b4cbb197SLinus Torvalds */ 2446b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2447b4cbb197SLinus Torvalds { 2448b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2449b4cbb197SLinus Torvalds 2450b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2451b4cbb197SLinus Torvalds if (start + len < start) 2452b4cbb197SLinus Torvalds return -EINVAL; 2453b4cbb197SLinus Torvalds /* 2454b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2455b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2456b4cbb197SLinus Torvalds * just have smaller alignment. 2457b4cbb197SLinus Torvalds */ 2458b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2459b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2460b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2461b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2462b4cbb197SLinus Torvalds return -EINVAL; 2463b4cbb197SLinus Torvalds 2464b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2465b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2466b4cbb197SLinus Torvalds return -EINVAL; 2467b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2468b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2469b4cbb197SLinus Torvalds 2470b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2471b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2472b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2473b4cbb197SLinus Torvalds return -EINVAL; 2474b4cbb197SLinus Torvalds 2475b4cbb197SLinus Torvalds /* Ok, let it rip */ 2476b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2477b4cbb197SLinus Torvalds } 2478b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2479b4cbb197SLinus Torvalds 2480aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2481aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2482aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2483aee16b3cSJeremy Fitzhardinge { 2484aee16b3cSJeremy Fitzhardinge pte_t *pte; 2485aee16b3cSJeremy Fitzhardinge int err; 24862f569afdSMartin Schwidefsky pgtable_t token; 248794909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 2488aee16b3cSJeremy Fitzhardinge 2489aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 2490aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 2491aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2492aee16b3cSJeremy Fitzhardinge if (!pte) 2493aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2494aee16b3cSJeremy Fitzhardinge 2495aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2496aee16b3cSJeremy Fitzhardinge 249738e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 249838e0edb1SJeremy Fitzhardinge 24992f569afdSMartin Schwidefsky token = pmd_pgtable(*pmd); 2500aee16b3cSJeremy Fitzhardinge 2501aee16b3cSJeremy Fitzhardinge do { 2502c36987e2SDaisuke Nishimura err = fn(pte++, token, addr, data); 2503aee16b3cSJeremy Fitzhardinge if (err) 2504aee16b3cSJeremy Fitzhardinge break; 2505c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2506aee16b3cSJeremy Fitzhardinge 250738e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 250838e0edb1SJeremy Fitzhardinge 2509aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 2510aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 2511aee16b3cSJeremy Fitzhardinge return err; 2512aee16b3cSJeremy Fitzhardinge } 2513aee16b3cSJeremy Fitzhardinge 2514aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2515aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2516aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2517aee16b3cSJeremy Fitzhardinge { 2518aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2519aee16b3cSJeremy Fitzhardinge unsigned long next; 2520aee16b3cSJeremy Fitzhardinge int err; 2521aee16b3cSJeremy Fitzhardinge 2522ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2523ceb86879SAndi Kleen 2524aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 2525aee16b3cSJeremy Fitzhardinge if (!pmd) 2526aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2527aee16b3cSJeremy Fitzhardinge do { 2528aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 2529aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 2530aee16b3cSJeremy Fitzhardinge if (err) 2531aee16b3cSJeremy Fitzhardinge break; 2532aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 2533aee16b3cSJeremy Fitzhardinge return err; 2534aee16b3cSJeremy Fitzhardinge } 2535aee16b3cSJeremy Fitzhardinge 2536aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 2537aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2538aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 2539aee16b3cSJeremy Fitzhardinge { 2540aee16b3cSJeremy Fitzhardinge pud_t *pud; 2541aee16b3cSJeremy Fitzhardinge unsigned long next; 2542aee16b3cSJeremy Fitzhardinge int err; 2543aee16b3cSJeremy Fitzhardinge 2544aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 2545aee16b3cSJeremy Fitzhardinge if (!pud) 2546aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2547aee16b3cSJeremy Fitzhardinge do { 2548aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 2549aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 2550aee16b3cSJeremy Fitzhardinge if (err) 2551aee16b3cSJeremy Fitzhardinge break; 2552aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 2553aee16b3cSJeremy Fitzhardinge return err; 2554aee16b3cSJeremy Fitzhardinge } 2555aee16b3cSJeremy Fitzhardinge 2556aee16b3cSJeremy Fitzhardinge /* 2557aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2558aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2559aee16b3cSJeremy Fitzhardinge */ 2560aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2561aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2562aee16b3cSJeremy Fitzhardinge { 2563aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 2564aee16b3cSJeremy Fitzhardinge unsigned long next; 256557250a5bSJeremy Fitzhardinge unsigned long end = addr + size; 2566aee16b3cSJeremy Fitzhardinge int err; 2567aee16b3cSJeremy Fitzhardinge 2568aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 2569aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 2570aee16b3cSJeremy Fitzhardinge do { 2571aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 2572aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 2573aee16b3cSJeremy Fitzhardinge if (err) 2574aee16b3cSJeremy Fitzhardinge break; 2575aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 257657250a5bSJeremy Fitzhardinge 2577aee16b3cSJeremy Fitzhardinge return err; 2578aee16b3cSJeremy Fitzhardinge } 2579aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2580aee16b3cSJeremy Fitzhardinge 25811da177e4SLinus Torvalds /* 25828f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 25838f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 25848f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 2585a335b2e1SRyota Ozaki * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault 25868f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 25878f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 2588a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 25898f4e2101SHugh Dickins */ 25904c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 25918f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 25928f4e2101SHugh Dickins { 25938f4e2101SHugh Dickins int same = 1; 25948f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 25958f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 25964c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 25974c21e2f2SHugh Dickins spin_lock(ptl); 25988f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 25994c21e2f2SHugh Dickins spin_unlock(ptl); 26008f4e2101SHugh Dickins } 26018f4e2101SHugh Dickins #endif 26028f4e2101SHugh Dickins pte_unmap(page_table); 26038f4e2101SHugh Dickins return same; 26048f4e2101SHugh Dickins } 26058f4e2101SHugh Dickins 26069de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 26076aab341eSLinus Torvalds { 26080abdd7a8SDan Williams debug_dma_assert_idle(src); 26090abdd7a8SDan Williams 26106aab341eSLinus Torvalds /* 26116aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 26126aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 26136aab341eSLinus Torvalds * just copying from the original user address. If that 26146aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 26156aab341eSLinus Torvalds */ 26166aab341eSLinus Torvalds if (unlikely(!src)) { 26179b04c5feSCong Wang void *kaddr = kmap_atomic(dst); 26185d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 26195d2a2dbbSLinus Torvalds 26205d2a2dbbSLinus Torvalds /* 26215d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 26225d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 26235d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 26245d2a2dbbSLinus Torvalds * zeroes. 26255d2a2dbbSLinus Torvalds */ 26265d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 26273ecb01dfSJan Beulich clear_page(kaddr); 26289b04c5feSCong Wang kunmap_atomic(kaddr); 2629c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 26300ed361deSNick Piggin } else 26319de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 26326aab341eSLinus Torvalds } 26336aab341eSLinus Torvalds 26341da177e4SLinus Torvalds /* 2635fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2636fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2637fb09a464SKirill A. Shutemov * 2638fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2639fb09a464SKirill A. Shutemov */ 2640fb09a464SKirill A. Shutemov static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, 2641fb09a464SKirill A. Shutemov unsigned long address) 2642fb09a464SKirill A. Shutemov { 2643fb09a464SKirill A. Shutemov struct vm_fault vmf; 2644fb09a464SKirill A. Shutemov int ret; 2645fb09a464SKirill A. Shutemov 2646fb09a464SKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 2647fb09a464SKirill A. Shutemov vmf.pgoff = page->index; 2648fb09a464SKirill A. Shutemov vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2649fb09a464SKirill A. Shutemov vmf.page = page; 2650fb09a464SKirill A. Shutemov 2651fb09a464SKirill A. Shutemov ret = vma->vm_ops->page_mkwrite(vma, &vmf); 2652fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2653fb09a464SKirill A. Shutemov return ret; 2654fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 2655fb09a464SKirill A. Shutemov lock_page(page); 2656fb09a464SKirill A. Shutemov if (!page->mapping) { 2657fb09a464SKirill A. Shutemov unlock_page(page); 2658fb09a464SKirill A. Shutemov return 0; /* retry */ 2659fb09a464SKirill A. Shutemov } 2660fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2661fb09a464SKirill A. Shutemov } else 2662fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 2663fb09a464SKirill A. Shutemov return ret; 2664fb09a464SKirill A. Shutemov } 2665fb09a464SKirill A. Shutemov 2666fb09a464SKirill A. Shutemov /* 26671da177e4SLinus Torvalds * This routine handles present pages, when users try to write 26681da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 26691da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 26701da177e4SLinus Torvalds * 26711da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 26721da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 26731da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 26741da177e4SLinus Torvalds * COW. 26751da177e4SLinus Torvalds * 26761da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 26771da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 26781da177e4SLinus Torvalds * and potentially makes it more efficient. 26791da177e4SLinus Torvalds * 26808f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 26818f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 26828f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 26831da177e4SLinus Torvalds */ 26841da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 268565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 26868f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 2687e6219ec8SNamhyung Kim __releases(ptl) 26881da177e4SLinus Torvalds { 26892ec74c3eSSagi Grimberg struct page *old_page, *new_page = NULL; 26901da177e4SLinus Torvalds pte_t entry; 2691b009c024SMichel Lespinasse int ret = 0; 2692a200ee18SPeter Zijlstra int page_mkwrite = 0; 2693d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 26941756954cSDavid Rientjes unsigned long mmun_start = 0; /* For mmu_notifiers */ 26951756954cSDavid Rientjes unsigned long mmun_end = 0; /* For mmu_notifiers */ 26961da177e4SLinus Torvalds 26976aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 2698251b97f5SPeter Zijlstra if (!old_page) { 2699251b97f5SPeter Zijlstra /* 2700251b97f5SPeter Zijlstra * VM_MIXEDMAP !pfn_valid() case 2701251b97f5SPeter Zijlstra * 2702251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 2703251b97f5SPeter Zijlstra * Just mark the pages writable as we can't do any dirty 2704251b97f5SPeter Zijlstra * accounting on raw pfn maps. 2705251b97f5SPeter Zijlstra */ 2706251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2707251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 2708251b97f5SPeter Zijlstra goto reuse; 2709920fc356SHugh Dickins goto gotten; 2710251b97f5SPeter Zijlstra } 27111da177e4SLinus Torvalds 2712d08b3851SPeter Zijlstra /* 2713ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 2714ee6a6457SPeter Zijlstra * not dirty accountable. 2715d08b3851SPeter Zijlstra */ 27169a840895SHugh Dickins if (PageAnon(old_page) && !PageKsm(old_page)) { 2717ab967d86SHugh Dickins if (!trylock_page(old_page)) { 2718ab967d86SHugh Dickins page_cache_get(old_page); 2719ab967d86SHugh Dickins pte_unmap_unlock(page_table, ptl); 2720ab967d86SHugh Dickins lock_page(old_page); 2721ab967d86SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, 2722ab967d86SHugh Dickins &ptl); 2723ab967d86SHugh Dickins if (!pte_same(*page_table, orig_pte)) { 2724ab967d86SHugh Dickins unlock_page(old_page); 2725ab967d86SHugh Dickins goto unlock; 2726ab967d86SHugh Dickins } 2727ab967d86SHugh Dickins page_cache_release(old_page); 2728ab967d86SHugh Dickins } 2729b009c024SMichel Lespinasse if (reuse_swap_page(old_page)) { 2730c44b6743SRik van Riel /* 2731c44b6743SRik van Riel * The page is all ours. Move it to our anon_vma so 2732c44b6743SRik van Riel * the rmap code will not search our parent or siblings. 2733c44b6743SRik van Riel * Protected against the rmap code by the page lock. 2734c44b6743SRik van Riel */ 2735c44b6743SRik van Riel page_move_anon_rmap(old_page, vma, address); 2736ee6a6457SPeter Zijlstra unlock_page(old_page); 2737b009c024SMichel Lespinasse goto reuse; 2738b009c024SMichel Lespinasse } 2739b009c024SMichel Lespinasse unlock_page(old_page); 2740ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 2741d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 2742ee6a6457SPeter Zijlstra /* 2743ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 2744ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 2745ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 2746ee6a6457SPeter Zijlstra */ 27479637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 2748c2ec175cSNick Piggin int tmp; 27499637a5efSDavid Howells page_cache_get(old_page); 27509637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 2751fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, old_page, address); 2752fb09a464SKirill A. Shutemov if (unlikely(!tmp || (tmp & 2753fb09a464SKirill A. Shutemov (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 2754fb09a464SKirill A. Shutemov page_cache_release(old_page); 2755fb09a464SKirill A. Shutemov return tmp; 2756c2ec175cSNick Piggin } 27579637a5efSDavid Howells /* 27589637a5efSDavid Howells * Since we dropped the lock we need to revalidate 27599637a5efSDavid Howells * the PTE as someone else may have changed it. If 27609637a5efSDavid Howells * they did, we just return, as we can count on the 27619637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 27629637a5efSDavid Howells */ 27639637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 27649637a5efSDavid Howells &ptl); 2765b827e496SNick Piggin if (!pte_same(*page_table, orig_pte)) { 2766b827e496SNick Piggin unlock_page(old_page); 27679637a5efSDavid Howells goto unlock; 2768b827e496SNick Piggin } 2769a200ee18SPeter Zijlstra 2770a200ee18SPeter Zijlstra page_mkwrite = 1; 27719637a5efSDavid Howells } 2772d08b3851SPeter Zijlstra dirty_page = old_page; 2773d08b3851SPeter Zijlstra get_page(dirty_page); 27749637a5efSDavid Howells 2775251b97f5SPeter Zijlstra reuse: 27768c8a743cSPeter Zijlstra /* 27778c8a743cSPeter Zijlstra * Clear the pages cpupid information as the existing 27788c8a743cSPeter Zijlstra * information potentially belongs to a now completely 27798c8a743cSPeter Zijlstra * unrelated process. 27808c8a743cSPeter Zijlstra */ 27818c8a743cSPeter Zijlstra if (old_page) 27828c8a743cSPeter Zijlstra page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1); 27838c8a743cSPeter Zijlstra 2784eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 278565500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 278665500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2787954ffcb3SKAMEZAWA Hiroyuki if (ptep_set_access_flags(vma, address, page_table, entry,1)) 27884b3073e1SRussell King update_mmu_cache(vma, address, page_table); 278972ddc8f7SMichel Lespinasse pte_unmap_unlock(page_table, ptl); 279065500d23SHugh Dickins ret |= VM_FAULT_WRITE; 279172ddc8f7SMichel Lespinasse 279272ddc8f7SMichel Lespinasse if (!dirty_page) 279372ddc8f7SMichel Lespinasse return ret; 279472ddc8f7SMichel Lespinasse 279572ddc8f7SMichel Lespinasse /* 279672ddc8f7SMichel Lespinasse * Yes, Virginia, this is actually required to prevent a race 279772ddc8f7SMichel Lespinasse * with clear_page_dirty_for_io() from clearing the page dirty 279872ddc8f7SMichel Lespinasse * bit after it clear all dirty ptes, but before a racing 279972ddc8f7SMichel Lespinasse * do_wp_page installs a dirty pte. 280072ddc8f7SMichel Lespinasse * 2801f0c6d4d2SKirill A. Shutemov * do_shared_fault is protected similarly. 280272ddc8f7SMichel Lespinasse */ 280372ddc8f7SMichel Lespinasse if (!page_mkwrite) { 280472ddc8f7SMichel Lespinasse wait_on_page_locked(dirty_page); 2805ed6d7c8eSMiklos Szeredi set_page_dirty_balance(dirty_page); 280641c4d25fSJan Kara /* file_update_time outside page_lock */ 280741c4d25fSJan Kara if (vma->vm_file) 280841c4d25fSJan Kara file_update_time(vma->vm_file); 280972ddc8f7SMichel Lespinasse } 281072ddc8f7SMichel Lespinasse put_page(dirty_page); 281172ddc8f7SMichel Lespinasse if (page_mkwrite) { 281272ddc8f7SMichel Lespinasse struct address_space *mapping = dirty_page->mapping; 281372ddc8f7SMichel Lespinasse 281472ddc8f7SMichel Lespinasse set_page_dirty(dirty_page); 281572ddc8f7SMichel Lespinasse unlock_page(dirty_page); 281672ddc8f7SMichel Lespinasse page_cache_release(dirty_page); 281772ddc8f7SMichel Lespinasse if (mapping) { 281872ddc8f7SMichel Lespinasse /* 281972ddc8f7SMichel Lespinasse * Some device drivers do not set page.mapping 282072ddc8f7SMichel Lespinasse * but still dirty their pages 282172ddc8f7SMichel Lespinasse */ 282272ddc8f7SMichel Lespinasse balance_dirty_pages_ratelimited(mapping); 282372ddc8f7SMichel Lespinasse } 282472ddc8f7SMichel Lespinasse } 282572ddc8f7SMichel Lespinasse 282672ddc8f7SMichel Lespinasse return ret; 28271da177e4SLinus Torvalds } 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds /* 28301da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 28311da177e4SLinus Torvalds */ 28321da177e4SLinus Torvalds page_cache_get(old_page); 2833920fc356SHugh Dickins gotten: 28348f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 283765500d23SHugh Dickins goto oom; 2838a13ea5b7SHugh Dickins 283962eede62SHugh Dickins if (is_zero_pfn(pte_pfn(orig_pte))) { 2840a13ea5b7SHugh Dickins new_page = alloc_zeroed_user_highpage_movable(vma, address); 2841a13ea5b7SHugh Dickins if (!new_page) 2842a13ea5b7SHugh Dickins goto oom; 2843a13ea5b7SHugh Dickins } else { 2844769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 28451da177e4SLinus Torvalds if (!new_page) 284665500d23SHugh Dickins goto oom; 2847a13ea5b7SHugh Dickins cow_user_page(new_page, old_page, address, vma); 2848a13ea5b7SHugh Dickins } 2849a13ea5b7SHugh Dickins __SetPageUptodate(new_page); 2850a13ea5b7SHugh Dickins 2851d715ae08SMichal Hocko if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)) 28528a9f3ccdSBalbir Singh goto oom_free_new; 28538a9f3ccdSBalbir Singh 28546bdb913fSHaggai Eran mmun_start = address & PAGE_MASK; 28551756954cSDavid Rientjes mmun_end = mmun_start + PAGE_SIZE; 28566bdb913fSHaggai Eran mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); 28576bdb913fSHaggai Eran 28581da177e4SLinus Torvalds /* 28591da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 28601da177e4SLinus Torvalds */ 28618f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 286265500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 2863920fc356SHugh Dickins if (old_page) { 28644294621fSHugh Dickins if (!PageAnon(old_page)) { 286534e55232SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_FILEPAGES); 286634e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 28674294621fSHugh Dickins } 2868920fc356SHugh Dickins } else 286934e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 2870eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 287165500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 287265500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 28734ce072f1SSiddha, Suresh B /* 28744ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 28754ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 28764ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 28774ce072f1SSiddha, Suresh B * thread doing COW. 28784ce072f1SSiddha, Suresh B */ 2879828502d3SIzik Eidus ptep_clear_flush(vma, address, page_table); 28809617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 2881828502d3SIzik Eidus /* 2882828502d3SIzik Eidus * We call the notify macro here because, when using secondary 2883828502d3SIzik Eidus * mmu page tables (such as kvm shadow page tables), we want the 2884828502d3SIzik Eidus * new page to be mapped directly into the secondary page table. 2885828502d3SIzik Eidus */ 2886828502d3SIzik Eidus set_pte_at_notify(mm, address, page_table, entry); 28874b3073e1SRussell King update_mmu_cache(vma, address, page_table); 2888945754a1SNick Piggin if (old_page) { 2889945754a1SNick Piggin /* 2890945754a1SNick Piggin * Only after switching the pte to the new page may 2891945754a1SNick Piggin * we remove the mapcount here. Otherwise another 2892945754a1SNick Piggin * process may come and find the rmap count decremented 2893945754a1SNick Piggin * before the pte is switched to the new page, and 2894945754a1SNick Piggin * "reuse" the old page writing into it while our pte 2895945754a1SNick Piggin * here still points into it and can be read by other 2896945754a1SNick Piggin * threads. 2897945754a1SNick Piggin * 2898945754a1SNick Piggin * The critical issue is to order this 2899945754a1SNick Piggin * page_remove_rmap with the ptp_clear_flush above. 2900945754a1SNick Piggin * Those stores are ordered by (if nothing else,) 2901945754a1SNick Piggin * the barrier present in the atomic_add_negative 2902945754a1SNick Piggin * in page_remove_rmap. 2903945754a1SNick Piggin * 2904945754a1SNick Piggin * Then the TLB flush in ptep_clear_flush ensures that 2905945754a1SNick Piggin * no process can access the old page before the 2906945754a1SNick Piggin * decremented mapcount is visible. And the old page 2907945754a1SNick Piggin * cannot be reused until after the decremented 2908945754a1SNick Piggin * mapcount is visible. So transitively, TLBs to 2909945754a1SNick Piggin * old page will be flushed before it can be reused. 2910945754a1SNick Piggin */ 2911edc315fdSHugh Dickins page_remove_rmap(old_page); 2912945754a1SNick Piggin } 2913945754a1SNick Piggin 29141da177e4SLinus Torvalds /* Free the old page.. */ 29151da177e4SLinus Torvalds new_page = old_page; 2916f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 29178a9f3ccdSBalbir Singh } else 29188a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(new_page); 29198a9f3ccdSBalbir Singh 29206bdb913fSHaggai Eran if (new_page) 29216bdb913fSHaggai Eran page_cache_release(new_page); 292265500d23SHugh Dickins unlock: 29238f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 29241756954cSDavid Rientjes if (mmun_end > mmun_start) 29256bdb913fSHaggai Eran mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); 2926e15f8c01SMichel Lespinasse if (old_page) { 2927e15f8c01SMichel Lespinasse /* 2928e15f8c01SMichel Lespinasse * Don't let another task, with possibly unlocked vma, 2929e15f8c01SMichel Lespinasse * keep the mlocked page. 2930e15f8c01SMichel Lespinasse */ 2931e15f8c01SMichel Lespinasse if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { 2932e15f8c01SMichel Lespinasse lock_page(old_page); /* LRU manipulation */ 2933e15f8c01SMichel Lespinasse munlock_vma_page(old_page); 2934e15f8c01SMichel Lespinasse unlock_page(old_page); 2935e15f8c01SMichel Lespinasse } 2936e15f8c01SMichel Lespinasse page_cache_release(old_page); 2937e15f8c01SMichel Lespinasse } 2938f33ea7f4SNick Piggin return ret; 29398a9f3ccdSBalbir Singh oom_free_new: 29406dbf6d3bSHugh Dickins page_cache_release(new_page); 294165500d23SHugh Dickins oom: 294266521d5aSDominik Dingel if (old_page) 29431da177e4SLinus Torvalds page_cache_release(old_page); 29441da177e4SLinus Torvalds return VM_FAULT_OOM; 29451da177e4SLinus Torvalds } 29461da177e4SLinus Torvalds 294797a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 29481da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 29491da177e4SLinus Torvalds struct zap_details *details) 29501da177e4SLinus Torvalds { 2951f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 29521da177e4SLinus Torvalds } 29531da177e4SLinus Torvalds 29546b2dbba8SMichel Lespinasse static inline void unmap_mapping_range_tree(struct rb_root *root, 29551da177e4SLinus Torvalds struct zap_details *details) 29561da177e4SLinus Torvalds { 29571da177e4SLinus Torvalds struct vm_area_struct *vma; 29581da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 29591da177e4SLinus Torvalds 29606b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 29611da177e4SLinus Torvalds details->first_index, details->last_index) { 29621da177e4SLinus Torvalds 29631da177e4SLinus Torvalds vba = vma->vm_pgoff; 2964d6e93217SLibin vea = vba + vma_pages(vma) - 1; 29651da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 29661da177e4SLinus Torvalds zba = details->first_index; 29671da177e4SLinus Torvalds if (zba < vba) 29681da177e4SLinus Torvalds zba = vba; 29691da177e4SLinus Torvalds zea = details->last_index; 29701da177e4SLinus Torvalds if (zea > vea) 29711da177e4SLinus Torvalds zea = vea; 29721da177e4SLinus Torvalds 297397a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 29741da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 29751da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 297697a89413SPeter Zijlstra details); 29771da177e4SLinus Torvalds } 29781da177e4SLinus Torvalds } 29791da177e4SLinus Torvalds 29801da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 29811da177e4SLinus Torvalds struct zap_details *details) 29821da177e4SLinus Torvalds { 29831da177e4SLinus Torvalds struct vm_area_struct *vma; 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /* 29861da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 29871da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 29881da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 29891da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 29901da177e4SLinus Torvalds */ 29916b2dbba8SMichel Lespinasse list_for_each_entry(vma, head, shared.nonlinear) { 29921da177e4SLinus Torvalds details->nonlinear_vma = vma; 299397a89413SPeter Zijlstra unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); 29941da177e4SLinus Torvalds } 29951da177e4SLinus Torvalds } 29961da177e4SLinus Torvalds 29971da177e4SLinus Torvalds /** 299872fd4a35SRobert P. J. Day * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. 29993d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 30001da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 30011da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 300225d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 30031da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 30041da177e4SLinus Torvalds * partial pages. 30051da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 30061da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 30071da177e4SLinus Torvalds * end of the file. 30081da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 30091da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 30101da177e4SLinus Torvalds */ 30111da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 30121da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 30131da177e4SLinus Torvalds { 30141da177e4SLinus Torvalds struct zap_details details; 30151da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 30161da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 30171da177e4SLinus Torvalds 30181da177e4SLinus Torvalds /* Check for overflow. */ 30191da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 30201da177e4SLinus Torvalds long long holeend = 30211da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 30221da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 30231da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 30241da177e4SLinus Torvalds } 30251da177e4SLinus Torvalds 30261da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 30271da177e4SLinus Torvalds details.nonlinear_vma = NULL; 30281da177e4SLinus Torvalds details.first_index = hba; 30291da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 30301da177e4SLinus Torvalds if (details.last_index < details.first_index) 30311da177e4SLinus Torvalds details.last_index = ULONG_MAX; 30321da177e4SLinus Torvalds 303397a89413SPeter Zijlstra 30343d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 30356b2dbba8SMichel Lespinasse if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) 30361da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 30371da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 30381da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 30393d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 30401da177e4SLinus Torvalds } 30411da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 30421da177e4SLinus Torvalds 30431da177e4SLinus Torvalds /* 30448f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 30458f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 30468f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 30471da177e4SLinus Torvalds */ 304865500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 304965500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 305030c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 30511da177e4SLinus Torvalds { 30528f4e2101SHugh Dickins spinlock_t *ptl; 305356f31801SHugh Dickins struct page *page, *swapcache; 305465500d23SHugh Dickins swp_entry_t entry; 30551da177e4SLinus Torvalds pte_t pte; 3056d065bd81SMichel Lespinasse int locked; 305756039efaSKAMEZAWA Hiroyuki struct mem_cgroup *ptr; 3058ad8c2ee8SRik van Riel int exclusive = 0; 305983c54070SNick Piggin int ret = 0; 30601da177e4SLinus Torvalds 30614c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 30628f4e2101SHugh Dickins goto out; 306365500d23SHugh Dickins 306465500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 3065d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 30660697212aSChristoph Lameter if (is_migration_entry(entry)) { 30670697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 3068d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3069d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3070d1737fdbSAndi Kleen } else { 3071d1737fdbSAndi Kleen print_bad_pte(vma, address, orig_pte, NULL); 3072d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3073d1737fdbSAndi Kleen } 30740697212aSChristoph Lameter goto out; 30750697212aSChristoph Lameter } 30760ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 30771da177e4SLinus Torvalds page = lookup_swap_cache(entry); 30781da177e4SLinus Torvalds if (!page) { 307902098feaSHugh Dickins page = swapin_readahead(entry, 308002098feaSHugh Dickins GFP_HIGHUSER_MOVABLE, vma, address); 30811da177e4SLinus Torvalds if (!page) { 30821da177e4SLinus Torvalds /* 30838f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 30848f4e2101SHugh Dickins * while we released the pte lock. 30851da177e4SLinus Torvalds */ 30868f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 30871da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 30881da177e4SLinus Torvalds ret = VM_FAULT_OOM; 30890ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 309065500d23SHugh Dickins goto unlock; 30911da177e4SLinus Torvalds } 30921da177e4SLinus Torvalds 30931da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 30941da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3095f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 3096456f998eSYing Han mem_cgroup_count_vm_event(mm, PGMAJFAULT); 3097d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 309871f72525SWu Fengguang /* 309971f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 310071f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 310171f72525SWu Fengguang */ 3102d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3103d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 310456f31801SHugh Dickins swapcache = page; 31054779cb31SAndi Kleen goto out_release; 31061da177e4SLinus Torvalds } 31071da177e4SLinus Torvalds 310856f31801SHugh Dickins swapcache = page; 3109d065bd81SMichel Lespinasse locked = lock_page_or_retry(page, mm, flags); 3110e709ffd6SRik van Riel 311120a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 3112d065bd81SMichel Lespinasse if (!locked) { 3113d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3114d065bd81SMichel Lespinasse goto out_release; 3115d065bd81SMichel Lespinasse } 31161da177e4SLinus Torvalds 31174969c119SAndrea Arcangeli /* 311831c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 311931c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 312031c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 312131c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 31224969c119SAndrea Arcangeli */ 312331c4a3d3SHugh Dickins if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) 31244969c119SAndrea Arcangeli goto out_page; 31254969c119SAndrea Arcangeli 3126cbf86cfeSHugh Dickins page = ksm_might_need_to_copy(page, vma, address); 31274969c119SAndrea Arcangeli if (unlikely(!page)) { 31285ad64688SHugh Dickins ret = VM_FAULT_OOM; 31294969c119SAndrea Arcangeli page = swapcache; 31304969c119SAndrea Arcangeli goto out_page; 31314969c119SAndrea Arcangeli } 31325ad64688SHugh Dickins 31332c26fdd7SKAMEZAWA Hiroyuki if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { 3134073e587eSKAMEZAWA Hiroyuki ret = VM_FAULT_OOM; 3135bc43f75cSJohannes Weiner goto out_page; 3136073e587eSKAMEZAWA Hiroyuki } 3137073e587eSKAMEZAWA Hiroyuki 31381da177e4SLinus Torvalds /* 31398f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 31401da177e4SLinus Torvalds */ 31418f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 31429e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 3143b8107480SKirill Korotaev goto out_nomap; 3144b8107480SKirill Korotaev 3145b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 3146b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3147b8107480SKirill Korotaev goto out_nomap; 31481da177e4SLinus Torvalds } 31491da177e4SLinus Torvalds 31508c7c6e34SKAMEZAWA Hiroyuki /* 31518c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 31528c7c6e34SKAMEZAWA Hiroyuki * 31538c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 31548c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 31558c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 31568c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 31578c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 315803f3c433SKAMEZAWA Hiroyuki * Because delete_from_swap_page() may be called by reuse_swap_page(), 315903f3c433SKAMEZAWA Hiroyuki * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry 316003f3c433SKAMEZAWA Hiroyuki * in page->private. In this case, a record in swap_cgroup is silently 316103f3c433SKAMEZAWA Hiroyuki * discarded at swap_free(). 31628c7c6e34SKAMEZAWA Hiroyuki */ 31631da177e4SLinus Torvalds 316434e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 3165b084d435SKAMEZAWA Hiroyuki dec_mm_counter_fast(mm, MM_SWAPENTS); 31661da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 316730c9f3a9SLinus Torvalds if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { 31681da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 316930c9f3a9SLinus Torvalds flags &= ~FAULT_FLAG_WRITE; 31709a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 3171ad8c2ee8SRik van Riel exclusive = 1; 31721da177e4SLinus Torvalds } 31731da177e4SLinus Torvalds flush_icache_page(vma, page); 3174179ef71cSCyrill Gorcunov if (pte_swp_soft_dirty(orig_pte)) 3175179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 31761da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 317756f31801SHugh Dickins if (page == swapcache) 3178ad8c2ee8SRik van Riel do_page_add_anon_rmap(page, vma, address, exclusive); 317956f31801SHugh Dickins else /* ksm created a completely new copy */ 318056f31801SHugh Dickins page_add_new_anon_rmap(page, vma, address); 318103f3c433SKAMEZAWA Hiroyuki /* It's better to call commit-charge after rmap is established */ 318203f3c433SKAMEZAWA Hiroyuki mem_cgroup_commit_charge_swapin(page, ptr); 31831da177e4SLinus Torvalds 3184c475a8abSHugh Dickins swap_free(entry); 3185b291f000SNick Piggin if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 3186a2c43eedSHugh Dickins try_to_free_swap(page); 3187c475a8abSHugh Dickins unlock_page(page); 318856f31801SHugh Dickins if (page != swapcache) { 31894969c119SAndrea Arcangeli /* 31904969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 31914969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 31924969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 31934969c119SAndrea Arcangeli * further safety release the lock after the swap_free 31944969c119SAndrea Arcangeli * so that the swap count won't change under a 31954969c119SAndrea Arcangeli * parallel locked swapcache. 31964969c119SAndrea Arcangeli */ 31974969c119SAndrea Arcangeli unlock_page(swapcache); 31984969c119SAndrea Arcangeli page_cache_release(swapcache); 31994969c119SAndrea Arcangeli } 3200c475a8abSHugh Dickins 320130c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 320261469f1dSHugh Dickins ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); 320361469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 320461469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 32051da177e4SLinus Torvalds goto out; 32061da177e4SLinus Torvalds } 32071da177e4SLinus Torvalds 32081da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 32094b3073e1SRussell King update_mmu_cache(vma, address, page_table); 321065500d23SHugh Dickins unlock: 32118f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 32121da177e4SLinus Torvalds out: 32131da177e4SLinus Torvalds return ret; 3214b8107480SKirill Korotaev out_nomap: 32157a81b88cSKAMEZAWA Hiroyuki mem_cgroup_cancel_charge_swapin(ptr); 32168f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 3217bc43f75cSJohannes Weiner out_page: 3218b8107480SKirill Korotaev unlock_page(page); 32194779cb31SAndi Kleen out_release: 3220b8107480SKirill Korotaev page_cache_release(page); 322156f31801SHugh Dickins if (page != swapcache) { 32224969c119SAndrea Arcangeli unlock_page(swapcache); 32234969c119SAndrea Arcangeli page_cache_release(swapcache); 32244969c119SAndrea Arcangeli } 322565500d23SHugh Dickins return ret; 32261da177e4SLinus Torvalds } 32271da177e4SLinus Torvalds 32281da177e4SLinus Torvalds /* 32298ca3eb08SLuck, Tony * This is like a special single-page "expand_{down|up}wards()", 32308ca3eb08SLuck, Tony * except we must first make sure that 'address{-|+}PAGE_SIZE' 3231320b2b8dSLinus Torvalds * doesn't hit another vma. 3232320b2b8dSLinus Torvalds */ 3233320b2b8dSLinus Torvalds static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) 3234320b2b8dSLinus Torvalds { 3235320b2b8dSLinus Torvalds address &= PAGE_MASK; 3236320b2b8dSLinus Torvalds if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { 32370e8e50e2SLinus Torvalds struct vm_area_struct *prev = vma->vm_prev; 3238320b2b8dSLinus Torvalds 32390e8e50e2SLinus Torvalds /* 32400e8e50e2SLinus Torvalds * Is there a mapping abutting this one below? 32410e8e50e2SLinus Torvalds * 32420e8e50e2SLinus Torvalds * That's only ok if it's the same stack mapping 32430e8e50e2SLinus Torvalds * that has gotten split.. 32440e8e50e2SLinus Torvalds */ 32450e8e50e2SLinus Torvalds if (prev && prev->vm_end == address) 32460e8e50e2SLinus Torvalds return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; 32470e8e50e2SLinus Torvalds 3248d05f3169SMichal Hocko expand_downwards(vma, address - PAGE_SIZE); 3249320b2b8dSLinus Torvalds } 32508ca3eb08SLuck, Tony if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { 32518ca3eb08SLuck, Tony struct vm_area_struct *next = vma->vm_next; 32528ca3eb08SLuck, Tony 32538ca3eb08SLuck, Tony /* As VM_GROWSDOWN but s/below/above/ */ 32548ca3eb08SLuck, Tony if (next && next->vm_start == address + PAGE_SIZE) 32558ca3eb08SLuck, Tony return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; 32568ca3eb08SLuck, Tony 32578ca3eb08SLuck, Tony expand_upwards(vma, address + PAGE_SIZE); 32588ca3eb08SLuck, Tony } 3259320b2b8dSLinus Torvalds return 0; 3260320b2b8dSLinus Torvalds } 3261320b2b8dSLinus Torvalds 3262320b2b8dSLinus Torvalds /* 32638f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 32648f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 32658f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 32661da177e4SLinus Torvalds */ 326765500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 326865500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 326930c9f3a9SLinus Torvalds unsigned int flags) 32701da177e4SLinus Torvalds { 32718f4e2101SHugh Dickins struct page *page; 32728f4e2101SHugh Dickins spinlock_t *ptl; 32731da177e4SLinus Torvalds pte_t entry; 32741da177e4SLinus Torvalds 32755528f913SLinus Torvalds pte_unmap(page_table); 3276320b2b8dSLinus Torvalds 327711ac5524SLinus Torvalds /* Check if we need to add a guard page to the stack */ 327811ac5524SLinus Torvalds if (check_stack_guard_page(vma, address) < 0) 327911ac5524SLinus Torvalds return VM_FAULT_SIGBUS; 328011ac5524SLinus Torvalds 328111ac5524SLinus Torvalds /* Use the zero-page for reads */ 328262eede62SHugh Dickins if (!(flags & FAULT_FLAG_WRITE)) { 328362eede62SHugh Dickins entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), 328462eede62SHugh Dickins vma->vm_page_prot)); 328511ac5524SLinus Torvalds page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 3286a13ea5b7SHugh Dickins if (!pte_none(*page_table)) 3287a13ea5b7SHugh Dickins goto unlock; 3288a13ea5b7SHugh Dickins goto setpte; 3289a13ea5b7SHugh Dickins } 3290a13ea5b7SHugh Dickins 32911da177e4SLinus Torvalds /* Allocate our own private page. */ 32921da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 329365500d23SHugh Dickins goto oom; 3294769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 32951da177e4SLinus Torvalds if (!page) 329665500d23SHugh Dickins goto oom; 329752f37629SMinchan Kim /* 329852f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 329952f37629SMinchan Kim * preceeding stores to the page contents become visible before 330052f37629SMinchan Kim * the set_pte_at() write. 330152f37629SMinchan Kim */ 33020ed361deSNick Piggin __SetPageUptodate(page); 33031da177e4SLinus Torvalds 3304d715ae08SMichal Hocko if (mem_cgroup_charge_anon(page, mm, GFP_KERNEL)) 33058a9f3ccdSBalbir Singh goto oom_free_page; 33068a9f3ccdSBalbir Singh 330765500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 33081ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 33091ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 33108f4e2101SHugh Dickins 33118f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 33121c2fb7a4SAndrea Arcangeli if (!pte_none(*page_table)) 33138f4e2101SHugh Dickins goto release; 33149ba69294SHugh Dickins 331534e55232SKAMEZAWA Hiroyuki inc_mm_counter_fast(mm, MM_ANONPAGES); 33169617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 3317a13ea5b7SHugh Dickins setpte: 331865500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 33214b3073e1SRussell King update_mmu_cache(vma, address, page_table); 332265500d23SHugh Dickins unlock: 33238f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 332483c54070SNick Piggin return 0; 33258f4e2101SHugh Dickins release: 33268a9f3ccdSBalbir Singh mem_cgroup_uncharge_page(page); 33278f4e2101SHugh Dickins page_cache_release(page); 33288f4e2101SHugh Dickins goto unlock; 33298a9f3ccdSBalbir Singh oom_free_page: 33306dbf6d3bSHugh Dickins page_cache_release(page); 333165500d23SHugh Dickins oom: 33321da177e4SLinus Torvalds return VM_FAULT_OOM; 33331da177e4SLinus Torvalds } 33341da177e4SLinus Torvalds 33357eae74afSKirill A. Shutemov static int __do_fault(struct vm_area_struct *vma, unsigned long address, 33367eae74afSKirill A. Shutemov pgoff_t pgoff, unsigned int flags, struct page **page) 33377eae74afSKirill A. Shutemov { 33387eae74afSKirill A. Shutemov struct vm_fault vmf; 33397eae74afSKirill A. Shutemov int ret; 33407eae74afSKirill A. Shutemov 33417eae74afSKirill A. Shutemov vmf.virtual_address = (void __user *)(address & PAGE_MASK); 33427eae74afSKirill A. Shutemov vmf.pgoff = pgoff; 33437eae74afSKirill A. Shutemov vmf.flags = flags; 33447eae74afSKirill A. Shutemov vmf.page = NULL; 33457eae74afSKirill A. Shutemov 33467eae74afSKirill A. Shutemov ret = vma->vm_ops->fault(vma, &vmf); 33477eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 33487eae74afSKirill A. Shutemov return ret; 33497eae74afSKirill A. Shutemov 33507eae74afSKirill A. Shutemov if (unlikely(PageHWPoison(vmf.page))) { 33517eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 33527eae74afSKirill A. Shutemov unlock_page(vmf.page); 33537eae74afSKirill A. Shutemov page_cache_release(vmf.page); 33547eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 33557eae74afSKirill A. Shutemov } 33567eae74afSKirill A. Shutemov 33577eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 33587eae74afSKirill A. Shutemov lock_page(vmf.page); 33597eae74afSKirill A. Shutemov else 33607eae74afSKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); 33617eae74afSKirill A. Shutemov 33627eae74afSKirill A. Shutemov *page = vmf.page; 33637eae74afSKirill A. Shutemov return ret; 33647eae74afSKirill A. Shutemov } 33657eae74afSKirill A. Shutemov 33668c6e50b0SKirill A. Shutemov /** 33678c6e50b0SKirill A. Shutemov * do_set_pte - setup new PTE entry for given page and add reverse page mapping. 33688c6e50b0SKirill A. Shutemov * 33698c6e50b0SKirill A. Shutemov * @vma: virtual memory area 33708c6e50b0SKirill A. Shutemov * @address: user virtual address 33718c6e50b0SKirill A. Shutemov * @page: page to map 33728c6e50b0SKirill A. Shutemov * @pte: pointer to target page table entry 33738c6e50b0SKirill A. Shutemov * @write: true, if new entry is writable 33748c6e50b0SKirill A. Shutemov * @anon: true, if it's anonymous page 33758c6e50b0SKirill A. Shutemov * 33768c6e50b0SKirill A. Shutemov * Caller must hold page table lock relevant for @pte. 33778c6e50b0SKirill A. Shutemov * 33788c6e50b0SKirill A. Shutemov * Target users are page handler itself and implementations of 33798c6e50b0SKirill A. Shutemov * vm_ops->map_pages. 33808c6e50b0SKirill A. Shutemov */ 33818c6e50b0SKirill A. Shutemov void do_set_pte(struct vm_area_struct *vma, unsigned long address, 33823bb97794SKirill A. Shutemov struct page *page, pte_t *pte, bool write, bool anon) 33833bb97794SKirill A. Shutemov { 33843bb97794SKirill A. Shutemov pte_t entry; 33853bb97794SKirill A. Shutemov 33863bb97794SKirill A. Shutemov flush_icache_page(vma, page); 33873bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 33883bb97794SKirill A. Shutemov if (write) 33893bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 33903bb97794SKirill A. Shutemov else if (pte_file(*pte) && pte_file_soft_dirty(*pte)) 33913bb97794SKirill A. Shutemov pte_mksoft_dirty(entry); 33923bb97794SKirill A. Shutemov if (anon) { 33933bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 33943bb97794SKirill A. Shutemov page_add_new_anon_rmap(page, vma, address); 33953bb97794SKirill A. Shutemov } else { 33963bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); 33973bb97794SKirill A. Shutemov page_add_file_rmap(page); 33983bb97794SKirill A. Shutemov } 33993bb97794SKirill A. Shutemov set_pte_at(vma->vm_mm, address, pte, entry); 34003bb97794SKirill A. Shutemov 34013bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 34023bb97794SKirill A. Shutemov update_mmu_cache(vma, address, pte); 34033bb97794SKirill A. Shutemov } 34043bb97794SKirill A. Shutemov 34058c6e50b0SKirill A. Shutemov #define FAULT_AROUND_ORDER 4 34061592eef0SKirill A. Shutemov 34071592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 34081592eef0SKirill A. Shutemov static unsigned int fault_around_order = FAULT_AROUND_ORDER; 34091592eef0SKirill A. Shutemov 34101592eef0SKirill A. Shutemov static int fault_around_order_get(void *data, u64 *val) 34111592eef0SKirill A. Shutemov { 34121592eef0SKirill A. Shutemov *val = fault_around_order; 34131592eef0SKirill A. Shutemov return 0; 34141592eef0SKirill A. Shutemov } 34151592eef0SKirill A. Shutemov 34161592eef0SKirill A. Shutemov static int fault_around_order_set(void *data, u64 val) 34171592eef0SKirill A. Shutemov { 34181592eef0SKirill A. Shutemov BUILD_BUG_ON((1UL << FAULT_AROUND_ORDER) > PTRS_PER_PTE); 34191592eef0SKirill A. Shutemov if (1UL << val > PTRS_PER_PTE) 34201592eef0SKirill A. Shutemov return -EINVAL; 34211592eef0SKirill A. Shutemov fault_around_order = val; 34221592eef0SKirill A. Shutemov return 0; 34231592eef0SKirill A. Shutemov } 34241592eef0SKirill A. Shutemov DEFINE_SIMPLE_ATTRIBUTE(fault_around_order_fops, 34251592eef0SKirill A. Shutemov fault_around_order_get, fault_around_order_set, "%llu\n"); 34261592eef0SKirill A. Shutemov 34271592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 34281592eef0SKirill A. Shutemov { 34291592eef0SKirill A. Shutemov void *ret; 34301592eef0SKirill A. Shutemov 34311592eef0SKirill A. Shutemov ret = debugfs_create_file("fault_around_order", 0644, NULL, NULL, 34321592eef0SKirill A. Shutemov &fault_around_order_fops); 34331592eef0SKirill A. Shutemov if (!ret) 34341592eef0SKirill A. Shutemov pr_warn("Failed to create fault_around_order in debugfs"); 34351592eef0SKirill A. Shutemov return 0; 34361592eef0SKirill A. Shutemov } 34371592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 34381592eef0SKirill A. Shutemov 34391592eef0SKirill A. Shutemov static inline unsigned long fault_around_pages(void) 34401592eef0SKirill A. Shutemov { 34411592eef0SKirill A. Shutemov return 1UL << fault_around_order; 34421592eef0SKirill A. Shutemov } 34431592eef0SKirill A. Shutemov 34441592eef0SKirill A. Shutemov static inline unsigned long fault_around_mask(void) 34451592eef0SKirill A. Shutemov { 34461592eef0SKirill A. Shutemov return ~((1UL << (PAGE_SHIFT + fault_around_order)) - 1); 34471592eef0SKirill A. Shutemov } 34481592eef0SKirill A. Shutemov #else 34491592eef0SKirill A. Shutemov static inline unsigned long fault_around_pages(void) 34501592eef0SKirill A. Shutemov { 34511592eef0SKirill A. Shutemov unsigned long nr_pages; 34521592eef0SKirill A. Shutemov 34531592eef0SKirill A. Shutemov nr_pages = 1UL << FAULT_AROUND_ORDER; 34541592eef0SKirill A. Shutemov BUILD_BUG_ON(nr_pages > PTRS_PER_PTE); 34551592eef0SKirill A. Shutemov return nr_pages; 34561592eef0SKirill A. Shutemov } 34571592eef0SKirill A. Shutemov 34581592eef0SKirill A. Shutemov static inline unsigned long fault_around_mask(void) 34591592eef0SKirill A. Shutemov { 34601592eef0SKirill A. Shutemov return ~((1UL << (PAGE_SHIFT + FAULT_AROUND_ORDER)) - 1); 34611592eef0SKirill A. Shutemov } 34621592eef0SKirill A. Shutemov #endif 34638c6e50b0SKirill A. Shutemov 34648c6e50b0SKirill A. Shutemov static void do_fault_around(struct vm_area_struct *vma, unsigned long address, 34658c6e50b0SKirill A. Shutemov pte_t *pte, pgoff_t pgoff, unsigned int flags) 34668c6e50b0SKirill A. Shutemov { 34678c6e50b0SKirill A. Shutemov unsigned long start_addr; 34688c6e50b0SKirill A. Shutemov pgoff_t max_pgoff; 34698c6e50b0SKirill A. Shutemov struct vm_fault vmf; 34708c6e50b0SKirill A. Shutemov int off; 34718c6e50b0SKirill A. Shutemov 34721592eef0SKirill A. Shutemov start_addr = max(address & fault_around_mask(), vma->vm_start); 34738c6e50b0SKirill A. Shutemov off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 34748c6e50b0SKirill A. Shutemov pte -= off; 34758c6e50b0SKirill A. Shutemov pgoff -= off; 34768c6e50b0SKirill A. Shutemov 34778c6e50b0SKirill A. Shutemov /* 34788c6e50b0SKirill A. Shutemov * max_pgoff is either end of page table or end of vma 34791592eef0SKirill A. Shutemov * or fault_around_pages() from pgoff, depending what is neast. 34808c6e50b0SKirill A. Shutemov */ 34818c6e50b0SKirill A. Shutemov max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + 34828c6e50b0SKirill A. Shutemov PTRS_PER_PTE - 1; 34838c6e50b0SKirill A. Shutemov max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, 34841592eef0SKirill A. Shutemov pgoff + fault_around_pages() - 1); 34858c6e50b0SKirill A. Shutemov 34868c6e50b0SKirill A. Shutemov /* Check if it makes any sense to call ->map_pages */ 34878c6e50b0SKirill A. Shutemov while (!pte_none(*pte)) { 34888c6e50b0SKirill A. Shutemov if (++pgoff > max_pgoff) 34898c6e50b0SKirill A. Shutemov return; 34908c6e50b0SKirill A. Shutemov start_addr += PAGE_SIZE; 34918c6e50b0SKirill A. Shutemov if (start_addr >= vma->vm_end) 34928c6e50b0SKirill A. Shutemov return; 34938c6e50b0SKirill A. Shutemov pte++; 34948c6e50b0SKirill A. Shutemov } 34958c6e50b0SKirill A. Shutemov 34968c6e50b0SKirill A. Shutemov vmf.virtual_address = (void __user *) start_addr; 34978c6e50b0SKirill A. Shutemov vmf.pte = pte; 34988c6e50b0SKirill A. Shutemov vmf.pgoff = pgoff; 34998c6e50b0SKirill A. Shutemov vmf.max_pgoff = max_pgoff; 35008c6e50b0SKirill A. Shutemov vmf.flags = flags; 35018c6e50b0SKirill A. Shutemov vma->vm_ops->map_pages(vma, &vmf); 35028c6e50b0SKirill A. Shutemov } 35038c6e50b0SKirill A. Shutemov 3504e655fb29SKirill A. Shutemov static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3505e655fb29SKirill A. Shutemov unsigned long address, pmd_t *pmd, 3506e655fb29SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 3507e655fb29SKirill A. Shutemov { 3508e655fb29SKirill A. Shutemov struct page *fault_page; 3509e655fb29SKirill A. Shutemov spinlock_t *ptl; 35103bb97794SKirill A. Shutemov pte_t *pte; 35118c6e50b0SKirill A. Shutemov int ret = 0; 35128c6e50b0SKirill A. Shutemov 35138c6e50b0SKirill A. Shutemov /* 35148c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 35158c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 35168c6e50b0SKirill A. Shutemov * something). 35178c6e50b0SKirill A. Shutemov */ 35188c6e50b0SKirill A. Shutemov if (vma->vm_ops->map_pages) { 35198c6e50b0SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 35208c6e50b0SKirill A. Shutemov do_fault_around(vma, address, pte, pgoff, flags); 35218c6e50b0SKirill A. Shutemov if (!pte_same(*pte, orig_pte)) 35228c6e50b0SKirill A. Shutemov goto unlock_out; 35238c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 35248c6e50b0SKirill A. Shutemov } 3525e655fb29SKirill A. Shutemov 3526e655fb29SKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 3527e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3528e655fb29SKirill A. Shutemov return ret; 3529e655fb29SKirill A. Shutemov 3530e655fb29SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3531e655fb29SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3532e655fb29SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3533e655fb29SKirill A. Shutemov unlock_page(fault_page); 3534e655fb29SKirill A. Shutemov page_cache_release(fault_page); 3535e655fb29SKirill A. Shutemov return ret; 3536e655fb29SKirill A. Shutemov } 35373bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, false, false); 3538e655fb29SKirill A. Shutemov unlock_page(fault_page); 35398c6e50b0SKirill A. Shutemov unlock_out: 35408c6e50b0SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3541e655fb29SKirill A. Shutemov return ret; 3542e655fb29SKirill A. Shutemov } 3543e655fb29SKirill A. Shutemov 3544ec47c3b9SKirill A. Shutemov static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3545ec47c3b9SKirill A. Shutemov unsigned long address, pmd_t *pmd, 3546ec47c3b9SKirill A. Shutemov pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 3547ec47c3b9SKirill A. Shutemov { 3548ec47c3b9SKirill A. Shutemov struct page *fault_page, *new_page; 3549ec47c3b9SKirill A. Shutemov spinlock_t *ptl; 35503bb97794SKirill A. Shutemov pte_t *pte; 3551ec47c3b9SKirill A. Shutemov int ret; 3552ec47c3b9SKirill A. Shutemov 3553ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 3554ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3555ec47c3b9SKirill A. Shutemov 3556ec47c3b9SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 3557ec47c3b9SKirill A. Shutemov if (!new_page) 3558ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3559ec47c3b9SKirill A. Shutemov 3560d715ae08SMichal Hocko if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)) { 3561ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3562ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 3563ec47c3b9SKirill A. Shutemov } 3564ec47c3b9SKirill A. Shutemov 3565ec47c3b9SKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 3566ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3567ec47c3b9SKirill A. Shutemov goto uncharge_out; 3568ec47c3b9SKirill A. Shutemov 3569ec47c3b9SKirill A. Shutemov copy_user_highpage(new_page, fault_page, address, vma); 3570ec47c3b9SKirill A. Shutemov __SetPageUptodate(new_page); 3571ec47c3b9SKirill A. Shutemov 3572ec47c3b9SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3573ec47c3b9SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3574ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3575ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3576ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 3577ec47c3b9SKirill A. Shutemov goto uncharge_out; 3578ec47c3b9SKirill A. Shutemov } 35793bb97794SKirill A. Shutemov do_set_pte(vma, address, new_page, pte, true, true); 3580ec47c3b9SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3581ec47c3b9SKirill A. Shutemov unlock_page(fault_page); 3582ec47c3b9SKirill A. Shutemov page_cache_release(fault_page); 3583ec47c3b9SKirill A. Shutemov return ret; 3584ec47c3b9SKirill A. Shutemov uncharge_out: 3585ec47c3b9SKirill A. Shutemov mem_cgroup_uncharge_page(new_page); 3586ec47c3b9SKirill A. Shutemov page_cache_release(new_page); 3587ec47c3b9SKirill A. Shutemov return ret; 3588ec47c3b9SKirill A. Shutemov } 3589ec47c3b9SKirill A. Shutemov 3590f0c6d4d2SKirill A. Shutemov static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, 359116abfa08SHugh Dickins unsigned long address, pmd_t *pmd, 359254cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 35931da177e4SLinus Torvalds { 3594f0c6d4d2SKirill A. Shutemov struct page *fault_page; 3595f0c6d4d2SKirill A. Shutemov struct address_space *mapping; 35968f4e2101SHugh Dickins spinlock_t *ptl; 35973bb97794SKirill A. Shutemov pte_t *pte; 3598f0c6d4d2SKirill A. Shutemov int dirtied = 0; 3599f0c6d4d2SKirill A. Shutemov int ret, tmp; 36001d65f86dSKAMEZAWA Hiroyuki 36017eae74afSKirill A. Shutemov ret = __do_fault(vma, address, pgoff, flags, &fault_page); 36027eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 3603f0c6d4d2SKirill A. Shutemov return ret; 36041da177e4SLinus Torvalds 36051da177e4SLinus Torvalds /* 3606f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 3607f0c6d4d2SKirill A. Shutemov * about to become writable 36081da177e4SLinus Torvalds */ 3609fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 3610f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3611fb09a464SKirill A. Shutemov tmp = do_page_mkwrite(vma, fault_page, address); 3612fb09a464SKirill A. Shutemov if (unlikely(!tmp || 3613fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3614f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3615f0c6d4d2SKirill A. Shutemov return tmp; 361669676147SMark Fasheh } 3617d0217ac0SNick Piggin } 3618fb09a464SKirill A. Shutemov 3619f0c6d4d2SKirill A. Shutemov pte = pte_offset_map_lock(mm, pmd, address, &ptl); 3620f0c6d4d2SKirill A. Shutemov if (unlikely(!pte_same(*pte, orig_pte))) { 3621f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3622f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3623f0c6d4d2SKirill A. Shutemov page_cache_release(fault_page); 3624f0c6d4d2SKirill A. Shutemov return ret; 36259637a5efSDavid Howells } 36263bb97794SKirill A. Shutemov do_set_pte(vma, address, fault_page, pte, true, false); 3627f0c6d4d2SKirill A. Shutemov pte_unmap_unlock(pte, ptl); 3628d00806b1SNick Piggin 3629f0c6d4d2SKirill A. Shutemov if (set_page_dirty(fault_page)) 363041c4d25fSJan Kara dirtied = 1; 3631f0c6d4d2SKirill A. Shutemov mapping = fault_page->mapping; 3632f0c6d4d2SKirill A. Shutemov unlock_page(fault_page); 3633f0c6d4d2SKirill A. Shutemov if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { 3634b827e496SNick Piggin /* 3635b827e496SNick Piggin * Some device drivers do not set page.mapping but still 3636b827e496SNick Piggin * dirty their pages 3637b827e496SNick Piggin */ 3638b827e496SNick Piggin balance_dirty_pages_ratelimited(mapping); 3639d08b3851SPeter Zijlstra } 3640d00806b1SNick Piggin 3641b827e496SNick Piggin /* file_update_time outside page_lock */ 3642f0c6d4d2SKirill A. Shutemov if (vma->vm_file && !vma->vm_ops->page_mkwrite) 3643b827e496SNick Piggin file_update_time(vma->vm_file); 3644b827e496SNick Piggin 3645b827e496SNick Piggin return ret; 364654cb8821SNick Piggin } 3647d00806b1SNick Piggin 364854cb8821SNick Piggin static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 364954cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 365030c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 365154cb8821SNick Piggin { 365254cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 36530da7e01fSDean Nelson - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 365454cb8821SNick Piggin 365516abfa08SHugh Dickins pte_unmap(page_table); 3656e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3657e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3658e655fb29SKirill A. Shutemov orig_pte); 3659ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3660ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3661ec47c3b9SKirill A. Shutemov orig_pte); 3662f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 366354cb8821SNick Piggin } 366454cb8821SNick Piggin 3665f4b81804SJes Sorensen /* 36661da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 36671da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 36681da177e4SLinus Torvalds * nonlinear vmas. 36698f4e2101SHugh Dickins * 36708f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 36718f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 36728f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 36731da177e4SLinus Torvalds */ 3674d0217ac0SNick Piggin static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 367565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 367630c9f3a9SLinus Torvalds unsigned int flags, pte_t orig_pte) 36771da177e4SLinus Torvalds { 367865500d23SHugh Dickins pgoff_t pgoff; 36791da177e4SLinus Torvalds 368030c9f3a9SLinus Torvalds flags |= FAULT_FLAG_NONLINEAR; 368130c9f3a9SLinus Torvalds 36824c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 368383c54070SNick Piggin return 0; 36841da177e4SLinus Torvalds 36852509ef26SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { 368665500d23SHugh Dickins /* 368765500d23SHugh Dickins * Page table corrupted: show pte and kill process. 368865500d23SHugh Dickins */ 36893dc14741SHugh Dickins print_bad_pte(vma, address, orig_pte, NULL); 3690d99be1a8SHugh Dickins return VM_FAULT_SIGBUS; 369165500d23SHugh Dickins } 369265500d23SHugh Dickins 369365500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 3694e655fb29SKirill A. Shutemov if (!(flags & FAULT_FLAG_WRITE)) 3695e655fb29SKirill A. Shutemov return do_read_fault(mm, vma, address, pmd, pgoff, flags, 3696e655fb29SKirill A. Shutemov orig_pte); 3697ec47c3b9SKirill A. Shutemov if (!(vma->vm_flags & VM_SHARED)) 3698ec47c3b9SKirill A. Shutemov return do_cow_fault(mm, vma, address, pmd, pgoff, flags, 3699ec47c3b9SKirill A. Shutemov orig_pte); 3700f0c6d4d2SKirill A. Shutemov return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); 37011da177e4SLinus Torvalds } 37021da177e4SLinus Torvalds 3703b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 370404bb2f94SRik van Riel unsigned long addr, int page_nid, 370504bb2f94SRik van Riel int *flags) 37069532fec1SMel Gorman { 37079532fec1SMel Gorman get_page(page); 37089532fec1SMel Gorman 37099532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 371004bb2f94SRik van Riel if (page_nid == numa_node_id()) { 37119532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 371204bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 371304bb2f94SRik van Riel } 37149532fec1SMel Gorman 37159532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 37169532fec1SMel Gorman } 37179532fec1SMel Gorman 3718b19a9939SRashika Kheria static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, 3719d10e63f2SMel Gorman unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) 3720d10e63f2SMel Gorman { 37214daae3b4SMel Gorman struct page *page = NULL; 3722d10e63f2SMel Gorman spinlock_t *ptl; 37238191acbdSMel Gorman int page_nid = -1; 372490572890SPeter Zijlstra int last_cpupid; 3725cbee9f88SPeter Zijlstra int target_nid; 3726b8593bfdSMel Gorman bool migrated = false; 37276688cc05SPeter Zijlstra int flags = 0; 3728d10e63f2SMel Gorman 3729d10e63f2SMel Gorman /* 3730d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 3731d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 3732d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 3733d10e63f2SMel Gorman * 3734d10e63f2SMel Gorman * ptep_modify_prot_start is not called as this is clearing 3735d10e63f2SMel Gorman * the _PAGE_NUMA bit and it is not really expected that there 3736d10e63f2SMel Gorman * would be concurrent hardware modifications to the PTE. 3737d10e63f2SMel Gorman */ 3738d10e63f2SMel Gorman ptl = pte_lockptr(mm, pmd); 3739d10e63f2SMel Gorman spin_lock(ptl); 37404daae3b4SMel Gorman if (unlikely(!pte_same(*ptep, pte))) { 37414daae3b4SMel Gorman pte_unmap_unlock(ptep, ptl); 37424daae3b4SMel Gorman goto out; 37434daae3b4SMel Gorman } 37444daae3b4SMel Gorman 3745d10e63f2SMel Gorman pte = pte_mknonnuma(pte); 3746d10e63f2SMel Gorman set_pte_at(mm, addr, ptep, pte); 3747d10e63f2SMel Gorman update_mmu_cache(vma, addr, ptep); 3748d10e63f2SMel Gorman 3749d10e63f2SMel Gorman page = vm_normal_page(vma, addr, pte); 3750d10e63f2SMel Gorman if (!page) { 3751d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 3752d10e63f2SMel Gorman return 0; 3753d10e63f2SMel Gorman } 3754a1a46184SMel Gorman BUG_ON(is_zero_pfn(page_to_pfn(page))); 3755d10e63f2SMel Gorman 37566688cc05SPeter Zijlstra /* 37576688cc05SPeter Zijlstra * Avoid grouping on DSO/COW pages in specific and RO pages 37586688cc05SPeter Zijlstra * in general, RO pages shouldn't hurt as much anyway since 37596688cc05SPeter Zijlstra * they can be in shared cache state. 37606688cc05SPeter Zijlstra */ 37616688cc05SPeter Zijlstra if (!pte_write(pte)) 37626688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 37636688cc05SPeter Zijlstra 3764dabe1d99SRik van Riel /* 3765dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 3766dabe1d99SRik van Riel * is later used when determining whether to group tasks together 3767dabe1d99SRik van Riel */ 3768dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 3769dabe1d99SRik van Riel flags |= TNF_SHARED; 3770dabe1d99SRik van Riel 377190572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 37728191acbdSMel Gorman page_nid = page_to_nid(page); 377304bb2f94SRik van Riel target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); 3774d10e63f2SMel Gorman pte_unmap_unlock(ptep, ptl); 37754daae3b4SMel Gorman if (target_nid == -1) { 37764daae3b4SMel Gorman put_page(page); 37774daae3b4SMel Gorman goto out; 37784daae3b4SMel Gorman } 37794daae3b4SMel Gorman 37804daae3b4SMel Gorman /* Migrate to the requested node */ 37811bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 37826688cc05SPeter Zijlstra if (migrated) { 37838191acbdSMel Gorman page_nid = target_nid; 37846688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 37856688cc05SPeter Zijlstra } 37864daae3b4SMel Gorman 37874daae3b4SMel Gorman out: 37888191acbdSMel Gorman if (page_nid != -1) 37896688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 3790d10e63f2SMel Gorman return 0; 3791d10e63f2SMel Gorman } 3792d10e63f2SMel Gorman 37931da177e4SLinus Torvalds /* 37941da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 37951da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 37961da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 37971da177e4SLinus Torvalds * 37981da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 37991da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 38001da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 38011da177e4SLinus Torvalds * 3802c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 3803c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 3804c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 38051da177e4SLinus Torvalds */ 3806c0292554SKirill A. Shutemov static int handle_pte_fault(struct mm_struct *mm, 38071da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 380830c9f3a9SLinus Torvalds pte_t *pte, pmd_t *pmd, unsigned int flags) 38091da177e4SLinus Torvalds { 38101da177e4SLinus Torvalds pte_t entry; 38118f4e2101SHugh Dickins spinlock_t *ptl; 38121da177e4SLinus Torvalds 38138dab5241SBenjamin Herrenschmidt entry = *pte; 38141da177e4SLinus Torvalds if (!pte_present(entry)) { 381565500d23SHugh Dickins if (pte_none(entry)) { 3816f4b81804SJes Sorensen if (vma->vm_ops) { 38173c18ddd1SNick Piggin if (likely(vma->vm_ops->fault)) 381854cb8821SNick Piggin return do_linear_fault(mm, vma, address, 381930c9f3a9SLinus Torvalds pte, pmd, flags, entry); 3820f4b81804SJes Sorensen } 3821f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 382230c9f3a9SLinus Torvalds pte, pmd, flags); 382365500d23SHugh Dickins } 38241da177e4SLinus Torvalds if (pte_file(entry)) 3825d0217ac0SNick Piggin return do_nonlinear_fault(mm, vma, address, 382630c9f3a9SLinus Torvalds pte, pmd, flags, entry); 382765500d23SHugh Dickins return do_swap_page(mm, vma, address, 382830c9f3a9SLinus Torvalds pte, pmd, flags, entry); 38291da177e4SLinus Torvalds } 38301da177e4SLinus Torvalds 3831d10e63f2SMel Gorman if (pte_numa(entry)) 3832d10e63f2SMel Gorman return do_numa_page(mm, vma, address, entry, pte, pmd); 3833d10e63f2SMel Gorman 38344c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 38358f4e2101SHugh Dickins spin_lock(ptl); 38368f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 38378f4e2101SHugh Dickins goto unlock; 383830c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) { 38391da177e4SLinus Torvalds if (!pte_write(entry)) 38408f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 38418f4e2101SHugh Dickins pte, pmd, ptl, entry); 38421da177e4SLinus Torvalds entry = pte_mkdirty(entry); 38431da177e4SLinus Torvalds } 38441da177e4SLinus Torvalds entry = pte_mkyoung(entry); 384530c9f3a9SLinus Torvalds if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { 38464b3073e1SRussell King update_mmu_cache(vma, address, pte); 38471a44e149SAndrea Arcangeli } else { 38481a44e149SAndrea Arcangeli /* 38491a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 38501a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 38511a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 38521a44e149SAndrea Arcangeli * with threads. 38531a44e149SAndrea Arcangeli */ 385430c9f3a9SLinus Torvalds if (flags & FAULT_FLAG_WRITE) 385561c77326SShaohua Li flush_tlb_fix_spurious_fault(vma, address); 38561a44e149SAndrea Arcangeli } 38578f4e2101SHugh Dickins unlock: 38588f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 385983c54070SNick Piggin return 0; 38601da177e4SLinus Torvalds } 38611da177e4SLinus Torvalds 38621da177e4SLinus Torvalds /* 38631da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 38641da177e4SLinus Torvalds */ 3865519e5247SJohannes Weiner static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3866d06063ccSLinus Torvalds unsigned long address, unsigned int flags) 38671da177e4SLinus Torvalds { 38681da177e4SLinus Torvalds pgd_t *pgd; 38691da177e4SLinus Torvalds pud_t *pud; 38701da177e4SLinus Torvalds pmd_t *pmd; 38711da177e4SLinus Torvalds pte_t *pte; 38721da177e4SLinus Torvalds 3873ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 387430c9f3a9SLinus Torvalds return hugetlb_fault(mm, vma, address, flags); 38751da177e4SLinus Torvalds 38761da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 38771da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 38781da177e4SLinus Torvalds if (!pud) 3879c74df32cSHugh Dickins return VM_FAULT_OOM; 38801da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 38811da177e4SLinus Torvalds if (!pmd) 3882c74df32cSHugh Dickins return VM_FAULT_OOM; 388371e3aac0SAndrea Arcangeli if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { 3884c0292554SKirill A. Shutemov int ret = VM_FAULT_FALLBACK; 388571e3aac0SAndrea Arcangeli if (!vma->vm_ops) 3886c0292554SKirill A. Shutemov ret = do_huge_pmd_anonymous_page(mm, vma, address, 388771e3aac0SAndrea Arcangeli pmd, flags); 3888c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 3889c0292554SKirill A. Shutemov return ret; 389071e3aac0SAndrea Arcangeli } else { 389171e3aac0SAndrea Arcangeli pmd_t orig_pmd = *pmd; 38921f1d06c3SDavid Rientjes int ret; 38931f1d06c3SDavid Rientjes 389471e3aac0SAndrea Arcangeli barrier(); 389571e3aac0SAndrea Arcangeli if (pmd_trans_huge(orig_pmd)) { 3896a1dd450bSWill Deacon unsigned int dirty = flags & FAULT_FLAG_WRITE; 3897a1dd450bSWill Deacon 3898e53289c0SLinus Torvalds /* 3899e53289c0SLinus Torvalds * If the pmd is splitting, return and retry the 3900e53289c0SLinus Torvalds * the fault. Alternative: wait until the split 3901e53289c0SLinus Torvalds * is done, and goto retry. 3902e53289c0SLinus Torvalds */ 3903e53289c0SLinus Torvalds if (pmd_trans_splitting(orig_pmd)) 3904e53289c0SLinus Torvalds return 0; 3905e53289c0SLinus Torvalds 39063d59eebcSLinus Torvalds if (pmd_numa(orig_pmd)) 39074daae3b4SMel Gorman return do_huge_pmd_numa_page(mm, vma, address, 3908d10e63f2SMel Gorman orig_pmd, pmd); 3909d10e63f2SMel Gorman 39103d59eebcSLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 39111f1d06c3SDavid Rientjes ret = do_huge_pmd_wp_page(mm, vma, address, pmd, 39121f1d06c3SDavid Rientjes orig_pmd); 39139845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 39141f1d06c3SDavid Rientjes return ret; 3915a1dd450bSWill Deacon } else { 3916a1dd450bSWill Deacon huge_pmd_set_accessed(mm, vma, address, pmd, 3917a1dd450bSWill Deacon orig_pmd, dirty); 391871e3aac0SAndrea Arcangeli return 0; 391971e3aac0SAndrea Arcangeli } 392071e3aac0SAndrea Arcangeli } 39219845cbbdSKirill A. Shutemov } 392271e3aac0SAndrea Arcangeli 39230f19c179SMel Gorman /* THP should already have been handled */ 39240f19c179SMel Gorman BUG_ON(pmd_numa(*pmd)); 3925d10e63f2SMel Gorman 392671e3aac0SAndrea Arcangeli /* 392771e3aac0SAndrea Arcangeli * Use __pte_alloc instead of pte_alloc_map, because we can't 392871e3aac0SAndrea Arcangeli * run pte_offset_map on the pmd, if an huge pmd could 392971e3aac0SAndrea Arcangeli * materialize from under us from a different thread. 393071e3aac0SAndrea Arcangeli */ 39314fd01770SMel Gorman if (unlikely(pmd_none(*pmd)) && 39324fd01770SMel Gorman unlikely(__pte_alloc(mm, vma, pmd, address))) 3933c74df32cSHugh Dickins return VM_FAULT_OOM; 393471e3aac0SAndrea Arcangeli /* if an huge pmd materialized from under us just retry later */ 393571e3aac0SAndrea Arcangeli if (unlikely(pmd_trans_huge(*pmd))) 393671e3aac0SAndrea Arcangeli return 0; 393771e3aac0SAndrea Arcangeli /* 393871e3aac0SAndrea Arcangeli * A regular pmd is established and it can't morph into a huge pmd 393971e3aac0SAndrea Arcangeli * from under us anymore at this point because we hold the mmap_sem 394071e3aac0SAndrea Arcangeli * read mode and khugepaged takes it in write mode. So now it's 394171e3aac0SAndrea Arcangeli * safe to run pte_offset_map(). 394271e3aac0SAndrea Arcangeli */ 394371e3aac0SAndrea Arcangeli pte = pte_offset_map(pmd, address); 39441da177e4SLinus Torvalds 394530c9f3a9SLinus Torvalds return handle_pte_fault(mm, vma, address, pte, pmd, flags); 39461da177e4SLinus Torvalds } 39471da177e4SLinus Torvalds 3948519e5247SJohannes Weiner int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 3949519e5247SJohannes Weiner unsigned long address, unsigned int flags) 3950519e5247SJohannes Weiner { 3951519e5247SJohannes Weiner int ret; 3952519e5247SJohannes Weiner 3953519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 3954519e5247SJohannes Weiner 3955519e5247SJohannes Weiner count_vm_event(PGFAULT); 3956519e5247SJohannes Weiner mem_cgroup_count_vm_event(mm, PGFAULT); 3957519e5247SJohannes Weiner 3958519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 3959519e5247SJohannes Weiner check_sync_rss_stat(current); 3960519e5247SJohannes Weiner 3961519e5247SJohannes Weiner /* 3962519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 3963519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 3964519e5247SJohannes Weiner */ 3965519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 396649426420SJohannes Weiner mem_cgroup_oom_enable(); 3967519e5247SJohannes Weiner 3968519e5247SJohannes Weiner ret = __handle_mm_fault(mm, vma, address, flags); 3969519e5247SJohannes Weiner 397049426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 397149426420SJohannes Weiner mem_cgroup_oom_disable(); 397249426420SJohannes Weiner /* 397349426420SJohannes Weiner * The task may have entered a memcg OOM situation but 397449426420SJohannes Weiner * if the allocation error was handled gracefully (no 397549426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 397649426420SJohannes Weiner * Just clean up the OOM state peacefully. 397749426420SJohannes Weiner */ 397849426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 397949426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 398049426420SJohannes Weiner } 39813812c8c8SJohannes Weiner 3982519e5247SJohannes Weiner return ret; 3983519e5247SJohannes Weiner } 3984519e5247SJohannes Weiner 39851da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 39861da177e4SLinus Torvalds /* 39871da177e4SLinus Torvalds * Allocate page upper directory. 3988872fec16SHugh Dickins * We've already handled the fast-path in-line. 39891da177e4SLinus Torvalds */ 39901bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 39911da177e4SLinus Torvalds { 3992c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 3993c74df32cSHugh Dickins if (!new) 39941bb3630eSHugh Dickins return -ENOMEM; 39951da177e4SLinus Torvalds 3996362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 3997362a61adSNick Piggin 3998872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 39991bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 40005e541973SBenjamin Herrenschmidt pud_free(mm, new); 40011bb3630eSHugh Dickins else 40021da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 4003872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 40041bb3630eSHugh Dickins return 0; 40051da177e4SLinus Torvalds } 40061da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 40071da177e4SLinus Torvalds 40081da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 40091da177e4SLinus Torvalds /* 40101da177e4SLinus Torvalds * Allocate page middle directory. 4011872fec16SHugh Dickins * We've already handled the fast-path in-line. 40121da177e4SLinus Torvalds */ 40131bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 40141da177e4SLinus Torvalds { 4015c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 4016c74df32cSHugh Dickins if (!new) 40171bb3630eSHugh Dickins return -ENOMEM; 40181da177e4SLinus Torvalds 4019362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 4020362a61adSNick Piggin 4021872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 40221da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 40231bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 40245e541973SBenjamin Herrenschmidt pmd_free(mm, new); 40251bb3630eSHugh Dickins else 40261da177e4SLinus Torvalds pud_populate(mm, pud, new); 40271da177e4SLinus Torvalds #else 40281bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 40295e541973SBenjamin Herrenschmidt pmd_free(mm, new); 40301bb3630eSHugh Dickins else 40311da177e4SLinus Torvalds pgd_populate(mm, pud, new); 40321da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 4033872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 40341bb3630eSHugh Dickins return 0; 40351da177e4SLinus Torvalds } 40361da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 40371da177e4SLinus Torvalds 40381da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 40391da177e4SLinus Torvalds 40401da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 40415ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 40421da177e4SLinus Torvalds 40431da177e4SLinus Torvalds static int __init gate_vma_init(void) 40441da177e4SLinus Torvalds { 40451da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 40461da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 40471da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 4048b6558c4aSRoland McGrath gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; 4049b6558c4aSRoland McGrath gate_vma.vm_page_prot = __P101; 4050909af768SJason Baron 40511da177e4SLinus Torvalds return 0; 40521da177e4SLinus Torvalds } 40531da177e4SLinus Torvalds __initcall(gate_vma_init); 40541da177e4SLinus Torvalds #endif 40551da177e4SLinus Torvalds 405631db58b3SStephen Wilson struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 40571da177e4SLinus Torvalds { 40581da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 40591da177e4SLinus Torvalds return &gate_vma; 40601da177e4SLinus Torvalds #else 40611da177e4SLinus Torvalds return NULL; 40621da177e4SLinus Torvalds #endif 40631da177e4SLinus Torvalds } 40641da177e4SLinus Torvalds 4065cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 40661da177e4SLinus Torvalds { 40671da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 40681da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 40691da177e4SLinus Torvalds return 1; 40701da177e4SLinus Torvalds #endif 40711da177e4SLinus Torvalds return 0; 40721da177e4SLinus Torvalds } 40731da177e4SLinus Torvalds 40741da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 40750ec76a11SDavid Howells 40761b36ba81SNamhyung Kim static int __follow_pte(struct mm_struct *mm, unsigned long address, 4077f8ad0f49SJohannes Weiner pte_t **ptepp, spinlock_t **ptlp) 4078f8ad0f49SJohannes Weiner { 4079f8ad0f49SJohannes Weiner pgd_t *pgd; 4080f8ad0f49SJohannes Weiner pud_t *pud; 4081f8ad0f49SJohannes Weiner pmd_t *pmd; 4082f8ad0f49SJohannes Weiner pte_t *ptep; 4083f8ad0f49SJohannes Weiner 4084f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 4085f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 4086f8ad0f49SJohannes Weiner goto out; 4087f8ad0f49SJohannes Weiner 4088f8ad0f49SJohannes Weiner pud = pud_offset(pgd, address); 4089f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 4090f8ad0f49SJohannes Weiner goto out; 4091f8ad0f49SJohannes Weiner 4092f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 4093f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 4094f8ad0f49SJohannes Weiner if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 4095f8ad0f49SJohannes Weiner goto out; 4096f8ad0f49SJohannes Weiner 4097f8ad0f49SJohannes Weiner /* We cannot handle huge page PFN maps. Luckily they don't exist. */ 4098f8ad0f49SJohannes Weiner if (pmd_huge(*pmd)) 4099f8ad0f49SJohannes Weiner goto out; 4100f8ad0f49SJohannes Weiner 4101f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 4102f8ad0f49SJohannes Weiner if (!ptep) 4103f8ad0f49SJohannes Weiner goto out; 4104f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 4105f8ad0f49SJohannes Weiner goto unlock; 4106f8ad0f49SJohannes Weiner *ptepp = ptep; 4107f8ad0f49SJohannes Weiner return 0; 4108f8ad0f49SJohannes Weiner unlock: 4109f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 4110f8ad0f49SJohannes Weiner out: 4111f8ad0f49SJohannes Weiner return -EINVAL; 4112f8ad0f49SJohannes Weiner } 4113f8ad0f49SJohannes Weiner 41141b36ba81SNamhyung Kim static inline int follow_pte(struct mm_struct *mm, unsigned long address, 41151b36ba81SNamhyung Kim pte_t **ptepp, spinlock_t **ptlp) 41161b36ba81SNamhyung Kim { 41171b36ba81SNamhyung Kim int res; 41181b36ba81SNamhyung Kim 41191b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 41201b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 41211b36ba81SNamhyung Kim !(res = __follow_pte(mm, address, ptepp, ptlp))); 41221b36ba81SNamhyung Kim return res; 41231b36ba81SNamhyung Kim } 41241b36ba81SNamhyung Kim 41253b6748e2SJohannes Weiner /** 41263b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 41273b6748e2SJohannes Weiner * @vma: memory mapping 41283b6748e2SJohannes Weiner * @address: user virtual address 41293b6748e2SJohannes Weiner * @pfn: location to store found PFN 41303b6748e2SJohannes Weiner * 41313b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 41323b6748e2SJohannes Weiner * 41333b6748e2SJohannes Weiner * Returns zero and the pfn at @pfn on success, -ve otherwise. 41343b6748e2SJohannes Weiner */ 41353b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 41363b6748e2SJohannes Weiner unsigned long *pfn) 41373b6748e2SJohannes Weiner { 41383b6748e2SJohannes Weiner int ret = -EINVAL; 41393b6748e2SJohannes Weiner spinlock_t *ptl; 41403b6748e2SJohannes Weiner pte_t *ptep; 41413b6748e2SJohannes Weiner 41423b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 41433b6748e2SJohannes Weiner return ret; 41443b6748e2SJohannes Weiner 41453b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 41463b6748e2SJohannes Weiner if (ret) 41473b6748e2SJohannes Weiner return ret; 41483b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 41493b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 41503b6748e2SJohannes Weiner return 0; 41513b6748e2SJohannes Weiner } 41523b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 41533b6748e2SJohannes Weiner 415428b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 4155d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 415628b2ee20SRik van Riel unsigned long address, unsigned int flags, 4157d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 415828b2ee20SRik van Riel { 415903668a4dSJohannes Weiner int ret = -EINVAL; 416028b2ee20SRik van Riel pte_t *ptep, pte; 416128b2ee20SRik van Riel spinlock_t *ptl; 416228b2ee20SRik van Riel 4163d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 4164d87fe660Svenkatesh.pallipadi@intel.com goto out; 416528b2ee20SRik van Riel 416603668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 4167d87fe660Svenkatesh.pallipadi@intel.com goto out; 416828b2ee20SRik van Riel pte = *ptep; 416903668a4dSJohannes Weiner 417028b2ee20SRik van Riel if ((flags & FOLL_WRITE) && !pte_write(pte)) 417128b2ee20SRik van Riel goto unlock; 417228b2ee20SRik van Riel 417328b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 417403668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 417528b2ee20SRik van Riel 417603668a4dSJohannes Weiner ret = 0; 417728b2ee20SRik van Riel unlock: 417828b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 417928b2ee20SRik van Riel out: 4180d87fe660Svenkatesh.pallipadi@intel.com return ret; 418128b2ee20SRik van Riel } 418228b2ee20SRik van Riel 418328b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 418428b2ee20SRik van Riel void *buf, int len, int write) 418528b2ee20SRik van Riel { 418628b2ee20SRik van Riel resource_size_t phys_addr; 418728b2ee20SRik van Riel unsigned long prot = 0; 41882bc7273bSKOSAKI Motohiro void __iomem *maddr; 418928b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 419028b2ee20SRik van Riel 4191d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 419228b2ee20SRik van Riel return -EINVAL; 419328b2ee20SRik van Riel 419428b2ee20SRik van Riel maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); 419528b2ee20SRik van Riel if (write) 419628b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 419728b2ee20SRik van Riel else 419828b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 419928b2ee20SRik van Riel iounmap(maddr); 420028b2ee20SRik van Riel 420128b2ee20SRik van Riel return len; 420228b2ee20SRik van Riel } 42035a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 420428b2ee20SRik van Riel #endif 420528b2ee20SRik van Riel 42060ec76a11SDavid Howells /* 4207206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 4208206cb636SStephen Wilson * given task for page fault accounting. 42090ec76a11SDavid Howells */ 4210206cb636SStephen Wilson static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 4211206cb636SStephen Wilson unsigned long addr, void *buf, int len, int write) 42120ec76a11SDavid Howells { 42130ec76a11SDavid Howells struct vm_area_struct *vma; 42140ec76a11SDavid Howells void *old_buf = buf; 42150ec76a11SDavid Howells 42160ec76a11SDavid Howells down_read(&mm->mmap_sem); 4217183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 42180ec76a11SDavid Howells while (len) { 42190ec76a11SDavid Howells int bytes, ret, offset; 42200ec76a11SDavid Howells void *maddr; 422128b2ee20SRik van Riel struct page *page = NULL; 42220ec76a11SDavid Howells 42230ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 42240ec76a11SDavid Howells write, 1, &page, &vma); 422528b2ee20SRik van Riel if (ret <= 0) { 422628b2ee20SRik van Riel /* 422728b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 422828b2ee20SRik van Riel * we can access using slightly different code. 422928b2ee20SRik van Riel */ 423028b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 423128b2ee20SRik van Riel vma = find_vma(mm, addr); 4232fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 42330ec76a11SDavid Howells break; 423428b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 423528b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 423628b2ee20SRik van Riel len, write); 423728b2ee20SRik van Riel if (ret <= 0) 423828b2ee20SRik van Riel #endif 423928b2ee20SRik van Riel break; 424028b2ee20SRik van Riel bytes = ret; 424128b2ee20SRik van Riel } else { 42420ec76a11SDavid Howells bytes = len; 42430ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 42440ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 42450ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 42460ec76a11SDavid Howells 42470ec76a11SDavid Howells maddr = kmap(page); 42480ec76a11SDavid Howells if (write) { 42490ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 42500ec76a11SDavid Howells maddr + offset, buf, bytes); 42510ec76a11SDavid Howells set_page_dirty_lock(page); 42520ec76a11SDavid Howells } else { 42530ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 42540ec76a11SDavid Howells buf, maddr + offset, bytes); 42550ec76a11SDavid Howells } 42560ec76a11SDavid Howells kunmap(page); 42570ec76a11SDavid Howells page_cache_release(page); 425828b2ee20SRik van Riel } 42590ec76a11SDavid Howells len -= bytes; 42600ec76a11SDavid Howells buf += bytes; 42610ec76a11SDavid Howells addr += bytes; 42620ec76a11SDavid Howells } 42630ec76a11SDavid Howells up_read(&mm->mmap_sem); 42640ec76a11SDavid Howells 42650ec76a11SDavid Howells return buf - old_buf; 42660ec76a11SDavid Howells } 426703252919SAndi Kleen 42685ddd36b9SStephen Wilson /** 4269ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 42705ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 42715ddd36b9SStephen Wilson * @addr: start address to access 42725ddd36b9SStephen Wilson * @buf: source or destination buffer 42735ddd36b9SStephen Wilson * @len: number of bytes to transfer 42745ddd36b9SStephen Wilson * @write: whether the access is a write 42755ddd36b9SStephen Wilson * 42765ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 42775ddd36b9SStephen Wilson */ 42785ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 42795ddd36b9SStephen Wilson void *buf, int len, int write) 42805ddd36b9SStephen Wilson { 42815ddd36b9SStephen Wilson return __access_remote_vm(NULL, mm, addr, buf, len, write); 42825ddd36b9SStephen Wilson } 42835ddd36b9SStephen Wilson 428403252919SAndi Kleen /* 4285206cb636SStephen Wilson * Access another process' address space. 4286206cb636SStephen Wilson * Source/target buffer must be kernel space, 4287206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 4288206cb636SStephen Wilson */ 4289206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 4290206cb636SStephen Wilson void *buf, int len, int write) 4291206cb636SStephen Wilson { 4292206cb636SStephen Wilson struct mm_struct *mm; 4293206cb636SStephen Wilson int ret; 4294206cb636SStephen Wilson 4295206cb636SStephen Wilson mm = get_task_mm(tsk); 4296206cb636SStephen Wilson if (!mm) 4297206cb636SStephen Wilson return 0; 4298206cb636SStephen Wilson 4299206cb636SStephen Wilson ret = __access_remote_vm(tsk, mm, addr, buf, len, write); 4300206cb636SStephen Wilson mmput(mm); 4301206cb636SStephen Wilson 4302206cb636SStephen Wilson return ret; 4303206cb636SStephen Wilson } 4304206cb636SStephen Wilson 430503252919SAndi Kleen /* 430603252919SAndi Kleen * Print the name of a VMA. 430703252919SAndi Kleen */ 430803252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 430903252919SAndi Kleen { 431003252919SAndi Kleen struct mm_struct *mm = current->mm; 431103252919SAndi Kleen struct vm_area_struct *vma; 431203252919SAndi Kleen 4313e8bff74aSIngo Molnar /* 4314e8bff74aSIngo Molnar * Do not print if we are in atomic 4315e8bff74aSIngo Molnar * contexts (in exception stacks, etc.): 4316e8bff74aSIngo Molnar */ 4317e8bff74aSIngo Molnar if (preempt_count()) 4318e8bff74aSIngo Molnar return; 4319e8bff74aSIngo Molnar 432003252919SAndi Kleen down_read(&mm->mmap_sem); 432103252919SAndi Kleen vma = find_vma(mm, ip); 432203252919SAndi Kleen if (vma && vma->vm_file) { 432303252919SAndi Kleen struct file *f = vma->vm_file; 432403252919SAndi Kleen char *buf = (char *)__get_free_page(GFP_KERNEL); 432503252919SAndi Kleen if (buf) { 43262fbc57c5SAndy Shevchenko char *p; 432703252919SAndi Kleen 4328cf28b486SJan Blunck p = d_path(&f->f_path, buf, PAGE_SIZE); 432903252919SAndi Kleen if (IS_ERR(p)) 433003252919SAndi Kleen p = "?"; 43312fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 433203252919SAndi Kleen vma->vm_start, 433303252919SAndi Kleen vma->vm_end - vma->vm_start); 433403252919SAndi Kleen free_page((unsigned long)buf); 433503252919SAndi Kleen } 433603252919SAndi Kleen } 433751a07e50SJeff Liu up_read(&mm->mmap_sem); 433803252919SAndi Kleen } 43393ee1afa3SNick Piggin 4340662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 43413ee1afa3SNick Piggin void might_fault(void) 43423ee1afa3SNick Piggin { 434395156f00SPeter Zijlstra /* 434495156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 434595156f00SPeter Zijlstra * holding the mmap_sem, this is safe because kernel memory doesn't 434695156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 434795156f00SPeter Zijlstra * below annotations will generate false positives. 434895156f00SPeter Zijlstra */ 434995156f00SPeter Zijlstra if (segment_eq(get_fs(), KERNEL_DS)) 435095156f00SPeter Zijlstra return; 435195156f00SPeter Zijlstra 43523ee1afa3SNick Piggin /* 43533ee1afa3SNick Piggin * it would be nicer only to annotate paths which are not under 43543ee1afa3SNick Piggin * pagefault_disable, however that requires a larger audit and 43553ee1afa3SNick Piggin * providing helpers like get_user_atomic. 43563ee1afa3SNick Piggin */ 4357662bbcb2SMichael S. Tsirkin if (in_atomic()) 4358662bbcb2SMichael S. Tsirkin return; 4359662bbcb2SMichael S. Tsirkin 4360662bbcb2SMichael S. Tsirkin __might_sleep(__FILE__, __LINE__, 0); 4361662bbcb2SMichael S. Tsirkin 4362662bbcb2SMichael S. Tsirkin if (current->mm) 43633ee1afa3SNick Piggin might_lock_read(¤t->mm->mmap_sem); 43643ee1afa3SNick Piggin } 43653ee1afa3SNick Piggin EXPORT_SYMBOL(might_fault); 43663ee1afa3SNick Piggin #endif 436747ad8475SAndrea Arcangeli 436847ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 436947ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 437047ad8475SAndrea Arcangeli unsigned long addr, 437147ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 437247ad8475SAndrea Arcangeli { 437347ad8475SAndrea Arcangeli int i; 437447ad8475SAndrea Arcangeli struct page *p = page; 437547ad8475SAndrea Arcangeli 437647ad8475SAndrea Arcangeli might_sleep(); 437747ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 437847ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 437947ad8475SAndrea Arcangeli cond_resched(); 438047ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 438147ad8475SAndrea Arcangeli } 438247ad8475SAndrea Arcangeli } 438347ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 438447ad8475SAndrea Arcangeli unsigned long addr, unsigned int pages_per_huge_page) 438547ad8475SAndrea Arcangeli { 438647ad8475SAndrea Arcangeli int i; 438747ad8475SAndrea Arcangeli 438847ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 438947ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 439047ad8475SAndrea Arcangeli return; 439147ad8475SAndrea Arcangeli } 439247ad8475SAndrea Arcangeli 439347ad8475SAndrea Arcangeli might_sleep(); 439447ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 439547ad8475SAndrea Arcangeli cond_resched(); 439647ad8475SAndrea Arcangeli clear_user_highpage(page + i, addr + i * PAGE_SIZE); 439747ad8475SAndrea Arcangeli } 439847ad8475SAndrea Arcangeli } 439947ad8475SAndrea Arcangeli 440047ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 440147ad8475SAndrea Arcangeli unsigned long addr, 440247ad8475SAndrea Arcangeli struct vm_area_struct *vma, 440347ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 440447ad8475SAndrea Arcangeli { 440547ad8475SAndrea Arcangeli int i; 440647ad8475SAndrea Arcangeli struct page *dst_base = dst; 440747ad8475SAndrea Arcangeli struct page *src_base = src; 440847ad8475SAndrea Arcangeli 440947ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 441047ad8475SAndrea Arcangeli cond_resched(); 441147ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 441247ad8475SAndrea Arcangeli 441347ad8475SAndrea Arcangeli i++; 441447ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 441547ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 441647ad8475SAndrea Arcangeli } 441747ad8475SAndrea Arcangeli } 441847ad8475SAndrea Arcangeli 441947ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 442047ad8475SAndrea Arcangeli unsigned long addr, struct vm_area_struct *vma, 442147ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 442247ad8475SAndrea Arcangeli { 442347ad8475SAndrea Arcangeli int i; 442447ad8475SAndrea Arcangeli 442547ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 442647ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 442747ad8475SAndrea Arcangeli pages_per_huge_page); 442847ad8475SAndrea Arcangeli return; 442947ad8475SAndrea Arcangeli } 443047ad8475SAndrea Arcangeli 443147ad8475SAndrea Arcangeli might_sleep(); 443247ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; i++) { 443347ad8475SAndrea Arcangeli cond_resched(); 443447ad8475SAndrea Arcangeli copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); 443547ad8475SAndrea Arcangeli } 443647ad8475SAndrea Arcangeli } 443747ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 443849076ec2SKirill A. Shutemov 443940b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 4440b35f1819SKirill A. Shutemov 4441b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 4442b35f1819SKirill A. Shutemov 4443b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 4444b35f1819SKirill A. Shutemov { 4445b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 4446b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 4447b35f1819SKirill A. Shutemov } 4448b35f1819SKirill A. Shutemov 4449539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 445049076ec2SKirill A. Shutemov { 445149076ec2SKirill A. Shutemov spinlock_t *ptl; 445249076ec2SKirill A. Shutemov 4453b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 445449076ec2SKirill A. Shutemov if (!ptl) 445549076ec2SKirill A. Shutemov return false; 4456539edb58SPeter Zijlstra page->ptl = ptl; 445749076ec2SKirill A. Shutemov return true; 445849076ec2SKirill A. Shutemov } 445949076ec2SKirill A. Shutemov 4460539edb58SPeter Zijlstra void ptlock_free(struct page *page) 446149076ec2SKirill A. Shutemov { 4462b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 446349076ec2SKirill A. Shutemov } 446449076ec2SKirill A. Shutemov #endif 4465