1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/memory.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 101da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 111da177e4SLinus Torvalds */ 121da177e4SLinus Torvalds 131da177e4SLinus Torvalds /* 141da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 151da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 161da177e4SLinus Torvalds * 171da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 181da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 191da177e4SLinus Torvalds * far as I could see. 201da177e4SLinus Torvalds * 211da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 221da177e4SLinus Torvalds */ 231da177e4SLinus Torvalds 241da177e4SLinus Torvalds /* 251da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 261da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 271da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 281da177e4SLinus Torvalds * Found it. Everything seems to work now. 291da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 301da177e4SLinus Torvalds */ 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds /* 331da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 341da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 351da177e4SLinus Torvalds * 361da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 371da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 381da177e4SLinus Torvalds * 391da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 401da177e4SLinus Torvalds */ 411da177e4SLinus Torvalds 421da177e4SLinus Torvalds #include <linux/kernel_stat.h> 431da177e4SLinus Torvalds #include <linux/mm.h> 446e84f315SIngo Molnar #include <linux/sched/mm.h> 45f7ccbae4SIngo Molnar #include <linux/sched/coredump.h> 466a3827d7SIngo Molnar #include <linux/sched/numa_balancing.h> 4729930025SIngo Molnar #include <linux/sched/task.h> 481da177e4SLinus Torvalds #include <linux/hugetlb.h> 491da177e4SLinus Torvalds #include <linux/mman.h> 501da177e4SLinus Torvalds #include <linux/swap.h> 511da177e4SLinus Torvalds #include <linux/highmem.h> 521da177e4SLinus Torvalds #include <linux/pagemap.h> 535042db43SJérôme Glisse #include <linux/memremap.h> 549a840895SHugh Dickins #include <linux/ksm.h> 551da177e4SLinus Torvalds #include <linux/rmap.h> 56b95f1b31SPaul Gortmaker #include <linux/export.h> 570ff92245SShailabh Nagar #include <linux/delayacct.h> 581da177e4SLinus Torvalds #include <linux/init.h> 5901c8f1c4SDan Williams #include <linux/pfn_t.h> 60edc79b2aSPeter Zijlstra #include <linux/writeback.h> 618a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 62cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 633dc14741SHugh Dickins #include <linux/swapops.h> 643dc14741SHugh Dickins #include <linux/elf.h> 655a0e3ad6STejun Heo #include <linux/gfp.h> 664daae3b4SMel Gorman #include <linux/migrate.h> 672fbc57c5SAndy Shevchenko #include <linux/string.h> 681592eef0SKirill A. Shutemov #include <linux/debugfs.h> 696b251fc9SAndrea Arcangeli #include <linux/userfaultfd_k.h> 70bc2466e4SJan Kara #include <linux/dax.h> 716b31d595SMichal Hocko #include <linux/oom.h> 7298fa15f3SAnshuman Khandual #include <linux/numa.h> 73bce617edSPeter Xu #include <linux/perf_event.h> 74bce617edSPeter Xu #include <linux/ptrace.h> 75e80d3909SJoerg Roedel #include <linux/vmalloc.h> 761da177e4SLinus Torvalds 77b3d1411bSJoel Fernandes (Google) #include <trace/events/kmem.h> 78b3d1411bSJoel Fernandes (Google) 796952b61dSAlexey Dobriyan #include <asm/io.h> 8033a709b2SDave Hansen #include <asm/mmu_context.h> 811da177e4SLinus Torvalds #include <asm/pgalloc.h> 827c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 831da177e4SLinus Torvalds #include <asm/tlb.h> 841da177e4SLinus Torvalds #include <asm/tlbflush.h> 851da177e4SLinus Torvalds 86e80d3909SJoerg Roedel #include "pgalloc-track.h" 8742b77728SJan Beulich #include "internal.h" 8842b77728SJan Beulich 89af27d940SArnd Bergmann #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) 9090572890SPeter Zijlstra #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. 9175980e97SPeter Zijlstra #endif 9275980e97SPeter Zijlstra 93d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 941da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 951da177e4SLinus Torvalds unsigned long max_mapnr; 961da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 97166f61b9STobin C Harding 98166f61b9STobin C Harding struct page *mem_map; 991da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 1001da177e4SLinus Torvalds #endif 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds /* 1031da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 1041da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 1051da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 1061da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 1071da177e4SLinus Torvalds * and ZONE_HIGHMEM. 1081da177e4SLinus Torvalds */ 1091da177e4SLinus Torvalds void *high_memory; 1101da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 1111da177e4SLinus Torvalds 11232a93233SIngo Molnar /* 11332a93233SIngo Molnar * Randomize the address space (stacks, mmaps, brk, etc.). 11432a93233SIngo Molnar * 11532a93233SIngo Molnar * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, 11632a93233SIngo Molnar * as ancient (libc5 based) binaries can segfault. ) 11732a93233SIngo Molnar */ 11832a93233SIngo Molnar int randomize_va_space __read_mostly = 11932a93233SIngo Molnar #ifdef CONFIG_COMPAT_BRK 12032a93233SIngo Molnar 1; 12132a93233SIngo Molnar #else 12232a93233SIngo Molnar 2; 12332a93233SIngo Molnar #endif 124a62eaf15SAndi Kleen 12583d116c5SJia He #ifndef arch_faults_on_old_pte 12683d116c5SJia He static inline bool arch_faults_on_old_pte(void) 12783d116c5SJia He { 12883d116c5SJia He /* 12983d116c5SJia He * Those arches which don't have hw access flag feature need to 13083d116c5SJia He * implement their own helper. By default, "true" means pagefault 13183d116c5SJia He * will be hit on old pte. 13283d116c5SJia He */ 13383d116c5SJia He return true; 13483d116c5SJia He } 13583d116c5SJia He #endif 13683d116c5SJia He 137a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 138a62eaf15SAndi Kleen { 139a62eaf15SAndi Kleen randomize_va_space = 0; 1409b41046cSOGAWA Hirofumi return 1; 141a62eaf15SAndi Kleen } 142a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 143a62eaf15SAndi Kleen 14462eede62SHugh Dickins unsigned long zero_pfn __read_mostly; 1450b70068eSArd Biesheuvel EXPORT_SYMBOL(zero_pfn); 1460b70068eSArd Biesheuvel 147166f61b9STobin C Harding unsigned long highest_memmap_pfn __read_mostly; 148166f61b9STobin C Harding 149a13ea5b7SHugh Dickins /* 150a13ea5b7SHugh Dickins * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() 151a13ea5b7SHugh Dickins */ 152a13ea5b7SHugh Dickins static int __init init_zero_pfn(void) 153a13ea5b7SHugh Dickins { 154a13ea5b7SHugh Dickins zero_pfn = page_to_pfn(ZERO_PAGE(0)); 155a13ea5b7SHugh Dickins return 0; 156a13ea5b7SHugh Dickins } 157a13ea5b7SHugh Dickins core_initcall(init_zero_pfn); 158a62eaf15SAndi Kleen 159e4dcad20SJoel Fernandes (Google) void mm_trace_rss_stat(struct mm_struct *mm, int member, long count) 160b3d1411bSJoel Fernandes (Google) { 161e4dcad20SJoel Fernandes (Google) trace_rss_stat(mm, member, count); 162b3d1411bSJoel Fernandes (Google) } 163d559db08SKAMEZAWA Hiroyuki 16434e55232SKAMEZAWA Hiroyuki #if defined(SPLIT_RSS_COUNTING) 16534e55232SKAMEZAWA Hiroyuki 166ea48cf78SDavid Rientjes void sync_mm_rss(struct mm_struct *mm) 16734e55232SKAMEZAWA Hiroyuki { 16834e55232SKAMEZAWA Hiroyuki int i; 16934e55232SKAMEZAWA Hiroyuki 17034e55232SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) { 17105af2e10SDavid Rientjes if (current->rss_stat.count[i]) { 17205af2e10SDavid Rientjes add_mm_counter(mm, i, current->rss_stat.count[i]); 17305af2e10SDavid Rientjes current->rss_stat.count[i] = 0; 17434e55232SKAMEZAWA Hiroyuki } 17534e55232SKAMEZAWA Hiroyuki } 17605af2e10SDavid Rientjes current->rss_stat.events = 0; 17734e55232SKAMEZAWA Hiroyuki } 17834e55232SKAMEZAWA Hiroyuki 17934e55232SKAMEZAWA Hiroyuki static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) 18034e55232SKAMEZAWA Hiroyuki { 18134e55232SKAMEZAWA Hiroyuki struct task_struct *task = current; 18234e55232SKAMEZAWA Hiroyuki 18334e55232SKAMEZAWA Hiroyuki if (likely(task->mm == mm)) 18434e55232SKAMEZAWA Hiroyuki task->rss_stat.count[member] += val; 18534e55232SKAMEZAWA Hiroyuki else 18634e55232SKAMEZAWA Hiroyuki add_mm_counter(mm, member, val); 18734e55232SKAMEZAWA Hiroyuki } 18834e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) 18934e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) 19034e55232SKAMEZAWA Hiroyuki 19134e55232SKAMEZAWA Hiroyuki /* sync counter once per 64 page faults */ 19234e55232SKAMEZAWA Hiroyuki #define TASK_RSS_EVENTS_THRESH (64) 19334e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 19434e55232SKAMEZAWA Hiroyuki { 19534e55232SKAMEZAWA Hiroyuki if (unlikely(task != current)) 19634e55232SKAMEZAWA Hiroyuki return; 19734e55232SKAMEZAWA Hiroyuki if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) 198ea48cf78SDavid Rientjes sync_mm_rss(task->mm); 19934e55232SKAMEZAWA Hiroyuki } 2009547d01bSPeter Zijlstra #else /* SPLIT_RSS_COUNTING */ 20134e55232SKAMEZAWA Hiroyuki 20234e55232SKAMEZAWA Hiroyuki #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) 20334e55232SKAMEZAWA Hiroyuki #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) 20434e55232SKAMEZAWA Hiroyuki 20534e55232SKAMEZAWA Hiroyuki static void check_sync_rss_stat(struct task_struct *task) 20634e55232SKAMEZAWA Hiroyuki { 20734e55232SKAMEZAWA Hiroyuki } 20834e55232SKAMEZAWA Hiroyuki 2099547d01bSPeter Zijlstra #endif /* SPLIT_RSS_COUNTING */ 2109547d01bSPeter Zijlstra 2111da177e4SLinus Torvalds /* 2121da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 2131da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 2141da177e4SLinus Torvalds */ 2159e1b32caSBenjamin Herrenschmidt static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, 2169e1b32caSBenjamin Herrenschmidt unsigned long addr) 2171da177e4SLinus Torvalds { 2182f569afdSMartin Schwidefsky pgtable_t token = pmd_pgtable(*pmd); 2191da177e4SLinus Torvalds pmd_clear(pmd); 2209e1b32caSBenjamin Herrenschmidt pte_free_tlb(tlb, token, addr); 221c4812909SKirill A. Shutemov mm_dec_nr_ptes(tlb->mm); 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds 224e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 225e0da382cSHugh Dickins unsigned long addr, unsigned long end, 226e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2271da177e4SLinus Torvalds { 2281da177e4SLinus Torvalds pmd_t *pmd; 2291da177e4SLinus Torvalds unsigned long next; 230e0da382cSHugh Dickins unsigned long start; 2311da177e4SLinus Torvalds 232e0da382cSHugh Dickins start = addr; 2331da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 2341da177e4SLinus Torvalds do { 2351da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 2361da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 2371da177e4SLinus Torvalds continue; 2389e1b32caSBenjamin Herrenschmidt free_pte_range(tlb, pmd, addr); 2391da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 2401da177e4SLinus Torvalds 241e0da382cSHugh Dickins start &= PUD_MASK; 242e0da382cSHugh Dickins if (start < floor) 243e0da382cSHugh Dickins return; 244e0da382cSHugh Dickins if (ceiling) { 245e0da382cSHugh Dickins ceiling &= PUD_MASK; 246e0da382cSHugh Dickins if (!ceiling) 247e0da382cSHugh Dickins return; 2481da177e4SLinus Torvalds } 249e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 250e0da382cSHugh Dickins return; 251e0da382cSHugh Dickins 252e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 253e0da382cSHugh Dickins pud_clear(pud); 2549e1b32caSBenjamin Herrenschmidt pmd_free_tlb(tlb, pmd, start); 255dc6c9a35SKirill A. Shutemov mm_dec_nr_pmds(tlb->mm); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258c2febafcSKirill A. Shutemov static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, 259e0da382cSHugh Dickins unsigned long addr, unsigned long end, 260e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2611da177e4SLinus Torvalds { 2621da177e4SLinus Torvalds pud_t *pud; 2631da177e4SLinus Torvalds unsigned long next; 264e0da382cSHugh Dickins unsigned long start; 2651da177e4SLinus Torvalds 266e0da382cSHugh Dickins start = addr; 267c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 2681da177e4SLinus Torvalds do { 2691da177e4SLinus Torvalds next = pud_addr_end(addr, end); 2701da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 2711da177e4SLinus Torvalds continue; 272e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 2731da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 2741da177e4SLinus Torvalds 275c2febafcSKirill A. Shutemov start &= P4D_MASK; 276c2febafcSKirill A. Shutemov if (start < floor) 277c2febafcSKirill A. Shutemov return; 278c2febafcSKirill A. Shutemov if (ceiling) { 279c2febafcSKirill A. Shutemov ceiling &= P4D_MASK; 280c2febafcSKirill A. Shutemov if (!ceiling) 281c2febafcSKirill A. Shutemov return; 282c2febafcSKirill A. Shutemov } 283c2febafcSKirill A. Shutemov if (end - 1 > ceiling - 1) 284c2febafcSKirill A. Shutemov return; 285c2febafcSKirill A. Shutemov 286c2febafcSKirill A. Shutemov pud = pud_offset(p4d, start); 287c2febafcSKirill A. Shutemov p4d_clear(p4d); 288c2febafcSKirill A. Shutemov pud_free_tlb(tlb, pud, start); 289b4e98d9aSKirill A. Shutemov mm_dec_nr_puds(tlb->mm); 290c2febafcSKirill A. Shutemov } 291c2febafcSKirill A. Shutemov 292c2febafcSKirill A. Shutemov static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, 293c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 294c2febafcSKirill A. Shutemov unsigned long floor, unsigned long ceiling) 295c2febafcSKirill A. Shutemov { 296c2febafcSKirill A. Shutemov p4d_t *p4d; 297c2febafcSKirill A. Shutemov unsigned long next; 298c2febafcSKirill A. Shutemov unsigned long start; 299c2febafcSKirill A. Shutemov 300c2febafcSKirill A. Shutemov start = addr; 301c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 302c2febafcSKirill A. Shutemov do { 303c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 304c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 305c2febafcSKirill A. Shutemov continue; 306c2febafcSKirill A. Shutemov free_pud_range(tlb, p4d, addr, next, floor, ceiling); 307c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 308c2febafcSKirill A. Shutemov 309e0da382cSHugh Dickins start &= PGDIR_MASK; 310e0da382cSHugh Dickins if (start < floor) 311e0da382cSHugh Dickins return; 312e0da382cSHugh Dickins if (ceiling) { 313e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 314e0da382cSHugh Dickins if (!ceiling) 315e0da382cSHugh Dickins return; 3161da177e4SLinus Torvalds } 317e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 318e0da382cSHugh Dickins return; 319e0da382cSHugh Dickins 320c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, start); 321e0da382cSHugh Dickins pgd_clear(pgd); 322c2febafcSKirill A. Shutemov p4d_free_tlb(tlb, p4d, start); 3231da177e4SLinus Torvalds } 3241da177e4SLinus Torvalds 3251da177e4SLinus Torvalds /* 326e0da382cSHugh Dickins * This function frees user-level page tables of a process. 3271da177e4SLinus Torvalds */ 32842b77728SJan Beulich void free_pgd_range(struct mmu_gather *tlb, 329e0da382cSHugh Dickins unsigned long addr, unsigned long end, 330e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds pgd_t *pgd; 3331da177e4SLinus Torvalds unsigned long next; 3341da177e4SLinus Torvalds 335e0da382cSHugh Dickins /* 336e0da382cSHugh Dickins * The next few lines have given us lots of grief... 337e0da382cSHugh Dickins * 338e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 339e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 340e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 341e0da382cSHugh Dickins * 342e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 343e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 344e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 345e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 346e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 347e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 348e0da382cSHugh Dickins * that end 0 case should be mythical). 349e0da382cSHugh Dickins * 350e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 351e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 352e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 353e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 354e0da382cSHugh Dickins * 355e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 356e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 357e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 358e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 359e0da382cSHugh Dickins */ 360e0da382cSHugh Dickins 361e0da382cSHugh Dickins addr &= PMD_MASK; 362e0da382cSHugh Dickins if (addr < floor) { 363e0da382cSHugh Dickins addr += PMD_SIZE; 364e0da382cSHugh Dickins if (!addr) 365e0da382cSHugh Dickins return; 366e0da382cSHugh Dickins } 367e0da382cSHugh Dickins if (ceiling) { 368e0da382cSHugh Dickins ceiling &= PMD_MASK; 369e0da382cSHugh Dickins if (!ceiling) 370e0da382cSHugh Dickins return; 371e0da382cSHugh Dickins } 372e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 373e0da382cSHugh Dickins end -= PMD_SIZE; 374e0da382cSHugh Dickins if (addr > end - 1) 375e0da382cSHugh Dickins return; 37607e32661SAneesh Kumar K.V /* 37707e32661SAneesh Kumar K.V * We add page table cache pages with PAGE_SIZE, 37807e32661SAneesh Kumar K.V * (see pte_free_tlb()), flush the tlb if we need 37907e32661SAneesh Kumar K.V */ 380ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 38142b77728SJan Beulich pgd = pgd_offset(tlb->mm, addr); 3821da177e4SLinus Torvalds do { 3831da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 3841da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 3851da177e4SLinus Torvalds continue; 386c2febafcSKirill A. Shutemov free_p4d_range(tlb, pgd, addr, next, floor, ceiling); 3871da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 388e0da382cSHugh Dickins } 389e0da382cSHugh Dickins 39042b77728SJan Beulich void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, 391e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 392e0da382cSHugh Dickins { 393e0da382cSHugh Dickins while (vma) { 394e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 395e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 396e0da382cSHugh Dickins 3978f4f8c16SHugh Dickins /* 39825d9e2d1Snpiggin@suse.de * Hide vma from rmap and truncate_pagecache before freeing 39925d9e2d1Snpiggin@suse.de * pgtables 4008f4f8c16SHugh Dickins */ 4015beb4930SRik van Riel unlink_anon_vmas(vma); 4028f4f8c16SHugh Dickins unlink_file_vma(vma); 4038f4f8c16SHugh Dickins 4049da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 4053bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 4063bf5ee95SHugh Dickins floor, next ? next->vm_start : ceiling); 4073bf5ee95SHugh Dickins } else { 4083bf5ee95SHugh Dickins /* 4093bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 4103bf5ee95SHugh Dickins */ 4113bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 4124866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 413e0da382cSHugh Dickins vma = next; 414e0da382cSHugh Dickins next = vma->vm_next; 4155beb4930SRik van Riel unlink_anon_vmas(vma); 4168f4f8c16SHugh Dickins unlink_file_vma(vma); 417e0da382cSHugh Dickins } 4183bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 419e0da382cSHugh Dickins floor, next ? next->vm_start : ceiling); 4203bf5ee95SHugh Dickins } 421e0da382cSHugh Dickins vma = next; 422e0da382cSHugh Dickins } 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds 4254cf58924SJoel Fernandes (Google) int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) 4261da177e4SLinus Torvalds { 427c4088ebdSKirill A. Shutemov spinlock_t *ptl; 4284cf58924SJoel Fernandes (Google) pgtable_t new = pte_alloc_one(mm); 4291da177e4SLinus Torvalds if (!new) 4301bb3630eSHugh Dickins return -ENOMEM; 4311bb3630eSHugh Dickins 432362a61adSNick Piggin /* 433362a61adSNick Piggin * Ensure all pte setup (eg. pte page lock and page clearing) are 434362a61adSNick Piggin * visible before the pte is made visible to other CPUs by being 435362a61adSNick Piggin * put into page tables. 436362a61adSNick Piggin * 437362a61adSNick Piggin * The other side of the story is the pointer chasing in the page 438362a61adSNick Piggin * table walking code (when walking the page table without locking; 439362a61adSNick Piggin * ie. most of the time). Fortunately, these data accesses consist 440362a61adSNick Piggin * of a chain of data-dependent loads, meaning most CPUs (alpha 441362a61adSNick Piggin * being the notable exception) will already guarantee loads are 442362a61adSNick Piggin * seen in-order. See the alpha page table accessors for the 443bb7cdd38SWill Deacon * smp_rmb() barriers in page table walking code. 444362a61adSNick Piggin */ 445362a61adSNick Piggin smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ 446362a61adSNick Piggin 447c4088ebdSKirill A. Shutemov ptl = pmd_lock(mm, pmd); 4488ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 449c4812909SKirill A. Shutemov mm_inc_nr_ptes(mm); 4501da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 4512f569afdSMartin Schwidefsky new = NULL; 4524b471e88SKirill A. Shutemov } 453c4088ebdSKirill A. Shutemov spin_unlock(ptl); 4542f569afdSMartin Schwidefsky if (new) 4552f569afdSMartin Schwidefsky pte_free(mm, new); 4561bb3630eSHugh Dickins return 0; 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds 4594cf58924SJoel Fernandes (Google) int __pte_alloc_kernel(pmd_t *pmd) 4601da177e4SLinus Torvalds { 4614cf58924SJoel Fernandes (Google) pte_t *new = pte_alloc_one_kernel(&init_mm); 4621da177e4SLinus Torvalds if (!new) 4631bb3630eSHugh Dickins return -ENOMEM; 4641da177e4SLinus Torvalds 465362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 466362a61adSNick Piggin 467872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 4688ac1f832SAndrea Arcangeli if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ 469872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 4702f569afdSMartin Schwidefsky new = NULL; 4714b471e88SKirill A. Shutemov } 472872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 4732f569afdSMartin Schwidefsky if (new) 4742f569afdSMartin Schwidefsky pte_free_kernel(&init_mm, new); 4751bb3630eSHugh Dickins return 0; 4761da177e4SLinus Torvalds } 4771da177e4SLinus Torvalds 478d559db08SKAMEZAWA Hiroyuki static inline void init_rss_vec(int *rss) 479ae859762SHugh Dickins { 480d559db08SKAMEZAWA Hiroyuki memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); 481d559db08SKAMEZAWA Hiroyuki } 482d559db08SKAMEZAWA Hiroyuki 483d559db08SKAMEZAWA Hiroyuki static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) 484d559db08SKAMEZAWA Hiroyuki { 485d559db08SKAMEZAWA Hiroyuki int i; 486d559db08SKAMEZAWA Hiroyuki 48734e55232SKAMEZAWA Hiroyuki if (current->mm == mm) 48805af2e10SDavid Rientjes sync_mm_rss(mm); 489d559db08SKAMEZAWA Hiroyuki for (i = 0; i < NR_MM_COUNTERS; i++) 490d559db08SKAMEZAWA Hiroyuki if (rss[i]) 491d559db08SKAMEZAWA Hiroyuki add_mm_counter(mm, i, rss[i]); 492ae859762SHugh Dickins } 493ae859762SHugh Dickins 4941da177e4SLinus Torvalds /* 4956aab341eSLinus Torvalds * This function is called to print an error when a bad pte 4966aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 4976aab341eSLinus Torvalds * a region that doesn't allow it. 498b5810039SNick Piggin * 499b5810039SNick Piggin * The calling function must still handle the error. 500b5810039SNick Piggin */ 5013dc14741SHugh Dickins static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, 5023dc14741SHugh Dickins pte_t pte, struct page *page) 503b5810039SNick Piggin { 5043dc14741SHugh Dickins pgd_t *pgd = pgd_offset(vma->vm_mm, addr); 505c2febafcSKirill A. Shutemov p4d_t *p4d = p4d_offset(pgd, addr); 506c2febafcSKirill A. Shutemov pud_t *pud = pud_offset(p4d, addr); 5073dc14741SHugh Dickins pmd_t *pmd = pmd_offset(pud, addr); 5083dc14741SHugh Dickins struct address_space *mapping; 5093dc14741SHugh Dickins pgoff_t index; 510d936cf9bSHugh Dickins static unsigned long resume; 511d936cf9bSHugh Dickins static unsigned long nr_shown; 512d936cf9bSHugh Dickins static unsigned long nr_unshown; 513d936cf9bSHugh Dickins 514d936cf9bSHugh Dickins /* 515d936cf9bSHugh Dickins * Allow a burst of 60 reports, then keep quiet for that minute; 516d936cf9bSHugh Dickins * or allow a steady drip of one report per second. 517d936cf9bSHugh Dickins */ 518d936cf9bSHugh Dickins if (nr_shown == 60) { 519d936cf9bSHugh Dickins if (time_before(jiffies, resume)) { 520d936cf9bSHugh Dickins nr_unshown++; 521d936cf9bSHugh Dickins return; 522d936cf9bSHugh Dickins } 523d936cf9bSHugh Dickins if (nr_unshown) { 5241170532bSJoe Perches pr_alert("BUG: Bad page map: %lu messages suppressed\n", 525d936cf9bSHugh Dickins nr_unshown); 526d936cf9bSHugh Dickins nr_unshown = 0; 527d936cf9bSHugh Dickins } 528d936cf9bSHugh Dickins nr_shown = 0; 529d936cf9bSHugh Dickins } 530d936cf9bSHugh Dickins if (nr_shown++ == 0) 531d936cf9bSHugh Dickins resume = jiffies + 60 * HZ; 5323dc14741SHugh Dickins 5333dc14741SHugh Dickins mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; 5343dc14741SHugh Dickins index = linear_page_index(vma, addr); 5353dc14741SHugh Dickins 5361170532bSJoe Perches pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", 5373dc14741SHugh Dickins current->comm, 5383dc14741SHugh Dickins (long long)pte_val(pte), (long long)pmd_val(*pmd)); 539718a3821SWu Fengguang if (page) 540f0b791a3SDave Hansen dump_page(page, "bad pte"); 5416aa9b8b2SKefeng Wang pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n", 5423dc14741SHugh Dickins (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); 543d75f773cSSakari Ailus pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n", 5442682582aSKonstantin Khlebnikov vma->vm_file, 5452682582aSKonstantin Khlebnikov vma->vm_ops ? vma->vm_ops->fault : NULL, 5462682582aSKonstantin Khlebnikov vma->vm_file ? vma->vm_file->f_op->mmap : NULL, 5472682582aSKonstantin Khlebnikov mapping ? mapping->a_ops->readpage : NULL); 548b5810039SNick Piggin dump_stack(); 549373d4d09SRusty Russell add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); 550b5810039SNick Piggin } 551b5810039SNick Piggin 552b5810039SNick Piggin /* 5537e675137SNick Piggin * vm_normal_page -- This function gets the "struct page" associated with a pte. 5546aab341eSLinus Torvalds * 5557e675137SNick Piggin * "Special" mappings do not wish to be associated with a "struct page" (either 5567e675137SNick Piggin * it doesn't exist, or it exists but they don't want to touch it). In this 5577e675137SNick Piggin * case, NULL is returned here. "Normal" mappings do have a struct page. 558b379d790SJared Hulbert * 5597e675137SNick Piggin * There are 2 broad cases. Firstly, an architecture may define a pte_special() 5607e675137SNick Piggin * pte bit, in which case this function is trivial. Secondly, an architecture 5617e675137SNick Piggin * may not have a spare pte bit, which requires a more complicated scheme, 5627e675137SNick Piggin * described below. 5637e675137SNick Piggin * 5647e675137SNick Piggin * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a 5657e675137SNick Piggin * special mapping (even if there are underlying and valid "struct pages"). 5667e675137SNick Piggin * COWed pages of a VM_PFNMAP are always normal. 5676aab341eSLinus Torvalds * 568b379d790SJared Hulbert * The way we recognize COWed pages within VM_PFNMAP mappings is through the 569b379d790SJared Hulbert * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit 5707e675137SNick Piggin * set, and the vm_pgoff will point to the first PFN mapped: thus every special 5717e675137SNick Piggin * mapping will always honor the rule 5726aab341eSLinus Torvalds * 5736aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 5746aab341eSLinus Torvalds * 5757e675137SNick Piggin * And for normal mappings this is false. 576b379d790SJared Hulbert * 5777e675137SNick Piggin * This restricts such mappings to be a linear translation from virtual address 5787e675137SNick Piggin * to pfn. To get around this restriction, we allow arbitrary mappings so long 5797e675137SNick Piggin * as the vma is not a COW mapping; in that case, we know that all ptes are 5807e675137SNick Piggin * special (because none can have been COWed). 581b379d790SJared Hulbert * 582b379d790SJared Hulbert * 5837e675137SNick Piggin * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. 5847e675137SNick Piggin * 585b379d790SJared Hulbert * VM_MIXEDMAP mappings can likewise contain memory with or without "struct 586b379d790SJared Hulbert * page" backing, however the difference is that _all_ pages with a struct 587b379d790SJared Hulbert * page (that is, those where pfn_valid is true) are refcounted and considered 588b379d790SJared Hulbert * normal pages by the VM. The disadvantage is that pages are refcounted 589b379d790SJared Hulbert * (which can be slower and simply not an option for some PFNMAP users). The 590b379d790SJared Hulbert * advantage is that we don't have to follow the strict linearity rule of 591b379d790SJared Hulbert * PFNMAP mappings in order to support COWable mappings. 592b379d790SJared Hulbert * 593ee498ed7SHugh Dickins */ 59425b2995aSChristoph Hellwig struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, 59525b2995aSChristoph Hellwig pte_t pte) 596ee498ed7SHugh Dickins { 59722b31eecSHugh Dickins unsigned long pfn = pte_pfn(pte); 5987e675137SNick Piggin 59900b3a331SLaurent Dufour if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { 600b38af472SHugh Dickins if (likely(!pte_special(pte))) 60122b31eecSHugh Dickins goto check_pfn; 602667a0a06SDavid Vrabel if (vma->vm_ops && vma->vm_ops->find_special_page) 603667a0a06SDavid Vrabel return vma->vm_ops->find_special_page(vma, addr); 604a13ea5b7SHugh Dickins if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) 605a13ea5b7SHugh Dickins return NULL; 606df6ad698SJérôme Glisse if (is_zero_pfn(pfn)) 607df6ad698SJérôme Glisse return NULL; 608e1fb4a08SDave Jiang if (pte_devmap(pte)) 609e1fb4a08SDave Jiang return NULL; 610e1fb4a08SDave Jiang 61122b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 6127e675137SNick Piggin return NULL; 6137e675137SNick Piggin } 6147e675137SNick Piggin 61500b3a331SLaurent Dufour /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ 6167e675137SNick Piggin 617b379d790SJared Hulbert if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 618b379d790SJared Hulbert if (vma->vm_flags & VM_MIXEDMAP) { 619b379d790SJared Hulbert if (!pfn_valid(pfn)) 620b379d790SJared Hulbert return NULL; 621b379d790SJared Hulbert goto out; 622b379d790SJared Hulbert } else { 6237e675137SNick Piggin unsigned long off; 6247e675137SNick Piggin off = (addr - vma->vm_start) >> PAGE_SHIFT; 6256aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 6266aab341eSLinus Torvalds return NULL; 62767121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 628fb155c16SLinus Torvalds return NULL; 6296aab341eSLinus Torvalds } 630b379d790SJared Hulbert } 6316aab341eSLinus Torvalds 632b38af472SHugh Dickins if (is_zero_pfn(pfn)) 633b38af472SHugh Dickins return NULL; 63400b3a331SLaurent Dufour 63522b31eecSHugh Dickins check_pfn: 63622b31eecSHugh Dickins if (unlikely(pfn > highest_memmap_pfn)) { 63722b31eecSHugh Dickins print_bad_pte(vma, addr, pte, NULL); 63822b31eecSHugh Dickins return NULL; 63922b31eecSHugh Dickins } 6406aab341eSLinus Torvalds 6416aab341eSLinus Torvalds /* 6427e675137SNick Piggin * NOTE! We still have PageReserved() pages in the page tables. 6437e675137SNick Piggin * eg. VDSO mappings can cause them to exist. 6446aab341eSLinus Torvalds */ 645b379d790SJared Hulbert out: 6466aab341eSLinus Torvalds return pfn_to_page(pfn); 647ee498ed7SHugh Dickins } 648ee498ed7SHugh Dickins 64928093f9fSGerald Schaefer #ifdef CONFIG_TRANSPARENT_HUGEPAGE 65028093f9fSGerald Schaefer struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, 65128093f9fSGerald Schaefer pmd_t pmd) 65228093f9fSGerald Schaefer { 65328093f9fSGerald Schaefer unsigned long pfn = pmd_pfn(pmd); 65428093f9fSGerald Schaefer 65528093f9fSGerald Schaefer /* 65628093f9fSGerald Schaefer * There is no pmd_special() but there may be special pmds, e.g. 65728093f9fSGerald Schaefer * in a direct-access (dax) mapping, so let's just replicate the 65800b3a331SLaurent Dufour * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. 65928093f9fSGerald Schaefer */ 66028093f9fSGerald Schaefer if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { 66128093f9fSGerald Schaefer if (vma->vm_flags & VM_MIXEDMAP) { 66228093f9fSGerald Schaefer if (!pfn_valid(pfn)) 66328093f9fSGerald Schaefer return NULL; 66428093f9fSGerald Schaefer goto out; 66528093f9fSGerald Schaefer } else { 66628093f9fSGerald Schaefer unsigned long off; 66728093f9fSGerald Schaefer off = (addr - vma->vm_start) >> PAGE_SHIFT; 66828093f9fSGerald Schaefer if (pfn == vma->vm_pgoff + off) 66928093f9fSGerald Schaefer return NULL; 67028093f9fSGerald Schaefer if (!is_cow_mapping(vma->vm_flags)) 67128093f9fSGerald Schaefer return NULL; 67228093f9fSGerald Schaefer } 67328093f9fSGerald Schaefer } 67428093f9fSGerald Schaefer 675e1fb4a08SDave Jiang if (pmd_devmap(pmd)) 676e1fb4a08SDave Jiang return NULL; 6773cde287bSYu Zhao if (is_huge_zero_pmd(pmd)) 67828093f9fSGerald Schaefer return NULL; 67928093f9fSGerald Schaefer if (unlikely(pfn > highest_memmap_pfn)) 68028093f9fSGerald Schaefer return NULL; 68128093f9fSGerald Schaefer 68228093f9fSGerald Schaefer /* 68328093f9fSGerald Schaefer * NOTE! We still have PageReserved() pages in the page tables. 68428093f9fSGerald Schaefer * eg. VDSO mappings can cause them to exist. 68528093f9fSGerald Schaefer */ 68628093f9fSGerald Schaefer out: 68728093f9fSGerald Schaefer return pfn_to_page(pfn); 68828093f9fSGerald Schaefer } 68928093f9fSGerald Schaefer #endif 69028093f9fSGerald Schaefer 691ee498ed7SHugh Dickins /* 6921da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 6931da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 6941da177e4SLinus Torvalds * covered by this vma. 6951da177e4SLinus Torvalds */ 6961da177e4SLinus Torvalds 697df3a57d1SLinus Torvalds static unsigned long 698df3a57d1SLinus Torvalds copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 699b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 7008c103762SHugh Dickins unsigned long addr, int *rss) 7011da177e4SLinus Torvalds { 702b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 7031da177e4SLinus Torvalds pte_t pte = *src_pte; 7041da177e4SLinus Torvalds struct page *page; 7050697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 7060697212aSChristoph Lameter 7072022b4d1SHugh Dickins if (likely(!non_swap_entry(entry))) { 708570a335bSHugh Dickins if (swap_duplicate(entry) < 0) 709570a335bSHugh Dickins return entry.val; 710570a335bSHugh Dickins 7111da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 7121da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 7131da177e4SLinus Torvalds spin_lock(&mmlist_lock); 714f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 715f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 716f412ac08SHugh Dickins &src_mm->mmlist); 7171da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 7181da177e4SLinus Torvalds } 719b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]++; 7202022b4d1SHugh Dickins } else if (is_migration_entry(entry)) { 7219f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 7229f9f1acdSKonstantin Khlebnikov 723eca56ff9SJerome Marchand rss[mm_counter(page)]++; 7249f9f1acdSKonstantin Khlebnikov 7259f9f1acdSKonstantin Khlebnikov if (is_write_migration_entry(entry) && 7260697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 7270697212aSChristoph Lameter /* 7289f9f1acdSKonstantin Khlebnikov * COW mappings require pages in both 7299f9f1acdSKonstantin Khlebnikov * parent and child to be set to read. 7300697212aSChristoph Lameter */ 7310697212aSChristoph Lameter make_migration_entry_read(&entry); 7320697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 733c3d16e16SCyrill Gorcunov if (pte_swp_soft_dirty(*src_pte)) 734c3d16e16SCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 735f45ec5ffSPeter Xu if (pte_swp_uffd_wp(*src_pte)) 736f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 7370697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 7380697212aSChristoph Lameter } 7395042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 7405042db43SJérôme Glisse page = device_private_entry_to_page(entry); 7415042db43SJérôme Glisse 7425042db43SJérôme Glisse /* 7435042db43SJérôme Glisse * Update rss count even for unaddressable pages, as 7445042db43SJérôme Glisse * they should treated just like normal pages in this 7455042db43SJérôme Glisse * respect. 7465042db43SJérôme Glisse * 7475042db43SJérôme Glisse * We will likely want to have some new rss counters 7485042db43SJérôme Glisse * for unaddressable pages, at some point. But for now 7495042db43SJérôme Glisse * keep things as they are. 7505042db43SJérôme Glisse */ 7515042db43SJérôme Glisse get_page(page); 7525042db43SJérôme Glisse rss[mm_counter(page)]++; 7535042db43SJérôme Glisse page_dup_rmap(page, false); 7545042db43SJérôme Glisse 7555042db43SJérôme Glisse /* 7565042db43SJérôme Glisse * We do not preserve soft-dirty information, because so 7575042db43SJérôme Glisse * far, checkpoint/restore is the only feature that 7585042db43SJérôme Glisse * requires that. And checkpoint/restore does not work 7595042db43SJérôme Glisse * when a device driver is involved (you cannot easily 7605042db43SJérôme Glisse * save and restore device driver state). 7615042db43SJérôme Glisse */ 7625042db43SJérôme Glisse if (is_write_device_private_entry(entry) && 7635042db43SJérôme Glisse is_cow_mapping(vm_flags)) { 7645042db43SJérôme Glisse make_device_private_entry_read(&entry); 7655042db43SJérôme Glisse pte = swp_entry_to_pte(entry); 766f45ec5ffSPeter Xu if (pte_swp_uffd_wp(*src_pte)) 767f45ec5ffSPeter Xu pte = pte_swp_mkuffd_wp(pte); 7685042db43SJérôme Glisse set_pte_at(src_mm, addr, src_pte, pte); 7695042db43SJérôme Glisse } 7701da177e4SLinus Torvalds } 771df3a57d1SLinus Torvalds set_pte_at(dst_mm, addr, dst_pte, pte); 772df3a57d1SLinus Torvalds return 0; 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds 77570e806e4SPeter Xu /* 77670e806e4SPeter Xu * Copy a present and normal page if necessary. 77770e806e4SPeter Xu * 77870e806e4SPeter Xu * NOTE! The usual case is that this doesn't need to do 77970e806e4SPeter Xu * anything, and can just return a positive value. That 78070e806e4SPeter Xu * will let the caller know that it can just increase 78170e806e4SPeter Xu * the page refcount and re-use the pte the traditional 78270e806e4SPeter Xu * way. 78370e806e4SPeter Xu * 78470e806e4SPeter Xu * But _if_ we need to copy it because it needs to be 78570e806e4SPeter Xu * pinned in the parent (and the child should get its own 78670e806e4SPeter Xu * copy rather than just a reference to the same page), 78770e806e4SPeter Xu * we'll do that here and return zero to let the caller 78870e806e4SPeter Xu * know we're done. 78970e806e4SPeter Xu * 79070e806e4SPeter Xu * And if we need a pre-allocated page but don't yet have 79170e806e4SPeter Xu * one, return a negative error to let the preallocation 79270e806e4SPeter Xu * code know so that it can do so outside the page table 79370e806e4SPeter Xu * lock. 79470e806e4SPeter Xu */ 79570e806e4SPeter Xu static inline int 796c78f4636SPeter Xu copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 797c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 798c78f4636SPeter Xu struct page **prealloc, pte_t pte, struct page *page) 79970e806e4SPeter Xu { 800c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 80170e806e4SPeter Xu struct page *new_page; 80270e806e4SPeter Xu 803c78f4636SPeter Xu if (!is_cow_mapping(src_vma->vm_flags)) 80470e806e4SPeter Xu return 1; 80570e806e4SPeter Xu 80670e806e4SPeter Xu /* 80770e806e4SPeter Xu * What we want to do is to check whether this page may 80870e806e4SPeter Xu * have been pinned by the parent process. If so, 80970e806e4SPeter Xu * instead of wrprotect the pte on both sides, we copy 81070e806e4SPeter Xu * the page immediately so that we'll always guarantee 81170e806e4SPeter Xu * the pinned page won't be randomly replaced in the 81270e806e4SPeter Xu * future. 81370e806e4SPeter Xu * 814f3c64edaSLinus Torvalds * The page pinning checks are just "has this mm ever 815f3c64edaSLinus Torvalds * seen pinning", along with the (inexact) check of 816f3c64edaSLinus Torvalds * the page count. That might give false positives for 817f3c64edaSLinus Torvalds * for pinning, but it will work correctly. 81870e806e4SPeter Xu */ 81970e806e4SPeter Xu if (likely(!atomic_read(&src_mm->has_pinned))) 82070e806e4SPeter Xu return 1; 82170e806e4SPeter Xu if (likely(!page_maybe_dma_pinned(page))) 82270e806e4SPeter Xu return 1; 82370e806e4SPeter Xu 82470e806e4SPeter Xu new_page = *prealloc; 82570e806e4SPeter Xu if (!new_page) 82670e806e4SPeter Xu return -EAGAIN; 82770e806e4SPeter Xu 82870e806e4SPeter Xu /* 82970e806e4SPeter Xu * We have a prealloc page, all good! Take it 83070e806e4SPeter Xu * over and copy the page & arm it. 83170e806e4SPeter Xu */ 83270e806e4SPeter Xu *prealloc = NULL; 833c78f4636SPeter Xu copy_user_highpage(new_page, page, addr, src_vma); 83470e806e4SPeter Xu __SetPageUptodate(new_page); 835c78f4636SPeter Xu page_add_new_anon_rmap(new_page, dst_vma, addr, false); 836c78f4636SPeter Xu lru_cache_add_inactive_or_unevictable(new_page, dst_vma); 83770e806e4SPeter Xu rss[mm_counter(new_page)]++; 83870e806e4SPeter Xu 83970e806e4SPeter Xu /* All done, just insert the new page copy in the child */ 840c78f4636SPeter Xu pte = mk_pte(new_page, dst_vma->vm_page_prot); 841c78f4636SPeter Xu pte = maybe_mkwrite(pte_mkdirty(pte), dst_vma); 842c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 84370e806e4SPeter Xu return 0; 84470e806e4SPeter Xu } 84570e806e4SPeter Xu 84670e806e4SPeter Xu /* 84770e806e4SPeter Xu * Copy one pte. Returns 0 if succeeded, or -EAGAIN if one preallocated page 84870e806e4SPeter Xu * is required to copy this pte. 84970e806e4SPeter Xu */ 85070e806e4SPeter Xu static inline int 851c78f4636SPeter Xu copy_present_pte(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 852c78f4636SPeter Xu pte_t *dst_pte, pte_t *src_pte, unsigned long addr, int *rss, 853c78f4636SPeter Xu struct page **prealloc) 854df3a57d1SLinus Torvalds { 855c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 856c78f4636SPeter Xu unsigned long vm_flags = src_vma->vm_flags; 857df3a57d1SLinus Torvalds pte_t pte = *src_pte; 858df3a57d1SLinus Torvalds struct page *page; 859df3a57d1SLinus Torvalds 860c78f4636SPeter Xu page = vm_normal_page(src_vma, addr, pte); 86170e806e4SPeter Xu if (page) { 86270e806e4SPeter Xu int retval; 86370e806e4SPeter Xu 864c78f4636SPeter Xu retval = copy_present_page(dst_vma, src_vma, dst_pte, src_pte, 865c78f4636SPeter Xu addr, rss, prealloc, pte, page); 86670e806e4SPeter Xu if (retval <= 0) 86770e806e4SPeter Xu return retval; 86870e806e4SPeter Xu 86970e806e4SPeter Xu get_page(page); 87070e806e4SPeter Xu page_dup_rmap(page, false); 87170e806e4SPeter Xu rss[mm_counter(page)]++; 87270e806e4SPeter Xu } 87370e806e4SPeter Xu 8741da177e4SLinus Torvalds /* 8751da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 8761da177e4SLinus Torvalds * in the parent and the child 8771da177e4SLinus Torvalds */ 8781b2de5d0SLinus Torvalds if (is_cow_mapping(vm_flags) && pte_write(pte)) { 8791da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 8803dc90795SZachary Amsden pte = pte_wrprotect(pte); 8811da177e4SLinus Torvalds } 8821da177e4SLinus Torvalds 8831da177e4SLinus Torvalds /* 8841da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 8851da177e4SLinus Torvalds * the child 8861da177e4SLinus Torvalds */ 8871da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 8881da177e4SLinus Torvalds pte = pte_mkclean(pte); 8891da177e4SLinus Torvalds pte = pte_mkold(pte); 8906aab341eSLinus Torvalds 891b569a176SPeter Xu /* 892b569a176SPeter Xu * Make sure the _PAGE_UFFD_WP bit is cleared if the new VMA 893b569a176SPeter Xu * does not have the VM_UFFD_WP, which means that the uffd 894b569a176SPeter Xu * fork event is not enabled. 895b569a176SPeter Xu */ 896b569a176SPeter Xu if (!(vm_flags & VM_UFFD_WP)) 897b569a176SPeter Xu pte = pte_clear_uffd_wp(pte); 898b569a176SPeter Xu 899c78f4636SPeter Xu set_pte_at(dst_vma->vm_mm, addr, dst_pte, pte); 90070e806e4SPeter Xu return 0; 9016aab341eSLinus Torvalds } 902ae859762SHugh Dickins 90370e806e4SPeter Xu static inline struct page * 90470e806e4SPeter Xu page_copy_prealloc(struct mm_struct *src_mm, struct vm_area_struct *vma, 90570e806e4SPeter Xu unsigned long addr) 90670e806e4SPeter Xu { 90770e806e4SPeter Xu struct page *new_page; 90870e806e4SPeter Xu 90970e806e4SPeter Xu new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, addr); 91070e806e4SPeter Xu if (!new_page) 91170e806e4SPeter Xu return NULL; 91270e806e4SPeter Xu 91370e806e4SPeter Xu if (mem_cgroup_charge(new_page, src_mm, GFP_KERNEL)) { 91470e806e4SPeter Xu put_page(new_page); 91570e806e4SPeter Xu return NULL; 91670e806e4SPeter Xu } 91770e806e4SPeter Xu cgroup_throttle_swaprate(new_page, GFP_KERNEL); 91870e806e4SPeter Xu 91970e806e4SPeter Xu return new_page; 9201da177e4SLinus Torvalds } 9211da177e4SLinus Torvalds 922c78f4636SPeter Xu static int 923c78f4636SPeter Xu copy_pte_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 924c78f4636SPeter Xu pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, 925c78f4636SPeter Xu unsigned long end) 9261da177e4SLinus Torvalds { 927c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 928c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 929c36987e2SDaisuke Nishimura pte_t *orig_src_pte, *orig_dst_pte; 9301da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 931c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 93270e806e4SPeter Xu int progress, ret = 0; 933d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 934570a335bSHugh Dickins swp_entry_t entry = (swp_entry_t){0}; 93570e806e4SPeter Xu struct page *prealloc = NULL; 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds again: 93870e806e4SPeter Xu progress = 0; 939d559db08SKAMEZAWA Hiroyuki init_rss_vec(rss); 940d559db08SKAMEZAWA Hiroyuki 941c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 94270e806e4SPeter Xu if (!dst_pte) { 94370e806e4SPeter Xu ret = -ENOMEM; 94470e806e4SPeter Xu goto out; 94570e806e4SPeter Xu } 946ece0e2b6SPeter Zijlstra src_pte = pte_offset_map(src_pmd, addr); 9474c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 948f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 949c36987e2SDaisuke Nishimura orig_src_pte = src_pte; 950c36987e2SDaisuke Nishimura orig_dst_pte = dst_pte; 9516606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 9521da177e4SLinus Torvalds 9531da177e4SLinus Torvalds do { 9541da177e4SLinus Torvalds /* 9551da177e4SLinus Torvalds * We are holding two locks at this point - either of them 9561da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 9571da177e4SLinus Torvalds */ 958e040f218SHugh Dickins if (progress >= 32) { 959e040f218SHugh Dickins progress = 0; 960e040f218SHugh Dickins if (need_resched() || 96195c354feSNick Piggin spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) 9621da177e4SLinus Torvalds break; 963e040f218SHugh Dickins } 9641da177e4SLinus Torvalds if (pte_none(*src_pte)) { 9651da177e4SLinus Torvalds progress++; 9661da177e4SLinus Torvalds continue; 9671da177e4SLinus Torvalds } 96879a1971cSLinus Torvalds if (unlikely(!pte_present(*src_pte))) { 96979a1971cSLinus Torvalds entry.val = copy_nonpresent_pte(dst_mm, src_mm, 97079a1971cSLinus Torvalds dst_pte, src_pte, 971c78f4636SPeter Xu src_vma, addr, rss); 972570a335bSHugh Dickins if (entry.val) 973570a335bSHugh Dickins break; 9741da177e4SLinus Torvalds progress += 8; 97579a1971cSLinus Torvalds continue; 97679a1971cSLinus Torvalds } 97770e806e4SPeter Xu /* copy_present_pte() will clear `*prealloc' if consumed */ 978c78f4636SPeter Xu ret = copy_present_pte(dst_vma, src_vma, dst_pte, src_pte, 979c78f4636SPeter Xu addr, rss, &prealloc); 98070e806e4SPeter Xu /* 98170e806e4SPeter Xu * If we need a pre-allocated page for this pte, drop the 98270e806e4SPeter Xu * locks, allocate, and try again. 98370e806e4SPeter Xu */ 98470e806e4SPeter Xu if (unlikely(ret == -EAGAIN)) 98570e806e4SPeter Xu break; 98670e806e4SPeter Xu if (unlikely(prealloc)) { 98770e806e4SPeter Xu /* 98870e806e4SPeter Xu * pre-alloc page cannot be reused by next time so as 98970e806e4SPeter Xu * to strictly follow mempolicy (e.g., alloc_page_vma() 99070e806e4SPeter Xu * will allocate page according to address). This 99170e806e4SPeter Xu * could only happen if one pinned pte changed. 99270e806e4SPeter Xu */ 99370e806e4SPeter Xu put_page(prealloc); 99470e806e4SPeter Xu prealloc = NULL; 99570e806e4SPeter Xu } 99679a1971cSLinus Torvalds progress += 8; 9971da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 9981da177e4SLinus Torvalds 9996606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1000c74df32cSHugh Dickins spin_unlock(src_ptl); 1001ece0e2b6SPeter Zijlstra pte_unmap(orig_src_pte); 1002d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(dst_mm, rss); 1003c36987e2SDaisuke Nishimura pte_unmap_unlock(orig_dst_pte, dst_ptl); 1004c74df32cSHugh Dickins cond_resched(); 1005570a335bSHugh Dickins 1006570a335bSHugh Dickins if (entry.val) { 100770e806e4SPeter Xu if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) { 100870e806e4SPeter Xu ret = -ENOMEM; 100970e806e4SPeter Xu goto out; 101070e806e4SPeter Xu } 101170e806e4SPeter Xu entry.val = 0; 101270e806e4SPeter Xu } else if (ret) { 101370e806e4SPeter Xu WARN_ON_ONCE(ret != -EAGAIN); 1014c78f4636SPeter Xu prealloc = page_copy_prealloc(src_mm, src_vma, addr); 101570e806e4SPeter Xu if (!prealloc) 1016570a335bSHugh Dickins return -ENOMEM; 101770e806e4SPeter Xu /* We've captured and resolved the error. Reset, try again. */ 101870e806e4SPeter Xu ret = 0; 1019570a335bSHugh Dickins } 10201da177e4SLinus Torvalds if (addr != end) 10211da177e4SLinus Torvalds goto again; 102270e806e4SPeter Xu out: 102370e806e4SPeter Xu if (unlikely(prealloc)) 102470e806e4SPeter Xu put_page(prealloc); 102570e806e4SPeter Xu return ret; 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds 1028c78f4636SPeter Xu static inline int 1029c78f4636SPeter Xu copy_pmd_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1030c78f4636SPeter Xu pud_t *dst_pud, pud_t *src_pud, unsigned long addr, 1031c78f4636SPeter Xu unsigned long end) 10321da177e4SLinus Torvalds { 1033c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1034c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 10351da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 10361da177e4SLinus Torvalds unsigned long next; 10371da177e4SLinus Torvalds 10381da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 10391da177e4SLinus Torvalds if (!dst_pmd) 10401da177e4SLinus Torvalds return -ENOMEM; 10411da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 10421da177e4SLinus Torvalds do { 10431da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 104484c3fc4eSZi Yan if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) 104584c3fc4eSZi Yan || pmd_devmap(*src_pmd)) { 104671e3aac0SAndrea Arcangeli int err; 1047c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, src_vma); 104871e3aac0SAndrea Arcangeli err = copy_huge_pmd(dst_mm, src_mm, 1049c78f4636SPeter Xu dst_pmd, src_pmd, addr, src_vma); 105071e3aac0SAndrea Arcangeli if (err == -ENOMEM) 105171e3aac0SAndrea Arcangeli return -ENOMEM; 105271e3aac0SAndrea Arcangeli if (!err) 105371e3aac0SAndrea Arcangeli continue; 105471e3aac0SAndrea Arcangeli /* fall through */ 105571e3aac0SAndrea Arcangeli } 10561da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 10571da177e4SLinus Torvalds continue; 1058c78f4636SPeter Xu if (copy_pte_range(dst_vma, src_vma, dst_pmd, src_pmd, 1059c78f4636SPeter Xu addr, next)) 10601da177e4SLinus Torvalds return -ENOMEM; 10611da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 10621da177e4SLinus Torvalds return 0; 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds 1065c78f4636SPeter Xu static inline int 1066c78f4636SPeter Xu copy_pud_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1067c78f4636SPeter Xu p4d_t *dst_p4d, p4d_t *src_p4d, unsigned long addr, 1068c78f4636SPeter Xu unsigned long end) 10691da177e4SLinus Torvalds { 1070c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1071c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 10721da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 10731da177e4SLinus Torvalds unsigned long next; 10741da177e4SLinus Torvalds 1075c2febafcSKirill A. Shutemov dst_pud = pud_alloc(dst_mm, dst_p4d, addr); 10761da177e4SLinus Torvalds if (!dst_pud) 10771da177e4SLinus Torvalds return -ENOMEM; 1078c2febafcSKirill A. Shutemov src_pud = pud_offset(src_p4d, addr); 10791da177e4SLinus Torvalds do { 10801da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1081a00cc7d9SMatthew Wilcox if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { 1082a00cc7d9SMatthew Wilcox int err; 1083a00cc7d9SMatthew Wilcox 1084c78f4636SPeter Xu VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma); 1085a00cc7d9SMatthew Wilcox err = copy_huge_pud(dst_mm, src_mm, 1086c78f4636SPeter Xu dst_pud, src_pud, addr, src_vma); 1087a00cc7d9SMatthew Wilcox if (err == -ENOMEM) 1088a00cc7d9SMatthew Wilcox return -ENOMEM; 1089a00cc7d9SMatthew Wilcox if (!err) 1090a00cc7d9SMatthew Wilcox continue; 1091a00cc7d9SMatthew Wilcox /* fall through */ 1092a00cc7d9SMatthew Wilcox } 10931da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 10941da177e4SLinus Torvalds continue; 1095c78f4636SPeter Xu if (copy_pmd_range(dst_vma, src_vma, dst_pud, src_pud, 1096c78f4636SPeter Xu addr, next)) 10971da177e4SLinus Torvalds return -ENOMEM; 10981da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 10991da177e4SLinus Torvalds return 0; 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 1102c78f4636SPeter Xu static inline int 1103c78f4636SPeter Xu copy_p4d_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma, 1104c78f4636SPeter Xu pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long addr, 1105c78f4636SPeter Xu unsigned long end) 1106c2febafcSKirill A. Shutemov { 1107c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1108c2febafcSKirill A. Shutemov p4d_t *src_p4d, *dst_p4d; 1109c2febafcSKirill A. Shutemov unsigned long next; 1110c2febafcSKirill A. Shutemov 1111c2febafcSKirill A. Shutemov dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); 1112c2febafcSKirill A. Shutemov if (!dst_p4d) 1113c2febafcSKirill A. Shutemov return -ENOMEM; 1114c2febafcSKirill A. Shutemov src_p4d = p4d_offset(src_pgd, addr); 1115c2febafcSKirill A. Shutemov do { 1116c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1117c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(src_p4d)) 1118c2febafcSKirill A. Shutemov continue; 1119c78f4636SPeter Xu if (copy_pud_range(dst_vma, src_vma, dst_p4d, src_p4d, 1120c78f4636SPeter Xu addr, next)) 1121c2febafcSKirill A. Shutemov return -ENOMEM; 1122c2febafcSKirill A. Shutemov } while (dst_p4d++, src_p4d++, addr = next, addr != end); 1123c2febafcSKirill A. Shutemov return 0; 1124c2febafcSKirill A. Shutemov } 1125c2febafcSKirill A. Shutemov 1126c78f4636SPeter Xu int 1127c78f4636SPeter Xu copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma) 11281da177e4SLinus Torvalds { 11291da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 11301da177e4SLinus Torvalds unsigned long next; 1131c78f4636SPeter Xu unsigned long addr = src_vma->vm_start; 1132c78f4636SPeter Xu unsigned long end = src_vma->vm_end; 1133c78f4636SPeter Xu struct mm_struct *dst_mm = dst_vma->vm_mm; 1134c78f4636SPeter Xu struct mm_struct *src_mm = src_vma->vm_mm; 1135ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 11362ec74c3eSSagi Grimberg bool is_cow; 1137cddb8a5cSAndrea Arcangeli int ret; 11381da177e4SLinus Torvalds 1139d992895bSNick Piggin /* 1140d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 1141d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 1142d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 1143d992895bSNick Piggin * efficient than faulting. 1144d992895bSNick Piggin */ 1145c78f4636SPeter Xu if (!(src_vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && 1146c78f4636SPeter Xu !src_vma->anon_vma) 1147d992895bSNick Piggin return 0; 1148d992895bSNick Piggin 1149c78f4636SPeter Xu if (is_vm_hugetlb_page(src_vma)) 1150c78f4636SPeter Xu return copy_hugetlb_page_range(dst_mm, src_mm, src_vma); 11511da177e4SLinus Torvalds 1152c78f4636SPeter Xu if (unlikely(src_vma->vm_flags & VM_PFNMAP)) { 11532ab64037Svenkatesh.pallipadi@intel.com /* 11542ab64037Svenkatesh.pallipadi@intel.com * We do not free on error cases below as remove_vma 11552ab64037Svenkatesh.pallipadi@intel.com * gets called on error from higher level routine 11562ab64037Svenkatesh.pallipadi@intel.com */ 1157c78f4636SPeter Xu ret = track_pfn_copy(src_vma); 11582ab64037Svenkatesh.pallipadi@intel.com if (ret) 11592ab64037Svenkatesh.pallipadi@intel.com return ret; 11602ab64037Svenkatesh.pallipadi@intel.com } 11612ab64037Svenkatesh.pallipadi@intel.com 1162cddb8a5cSAndrea Arcangeli /* 1163cddb8a5cSAndrea Arcangeli * We need to invalidate the secondary MMU mappings only when 1164cddb8a5cSAndrea Arcangeli * there could be a permission downgrade on the ptes of the 1165cddb8a5cSAndrea Arcangeli * parent mm. And a permission downgrade will only happen if 1166cddb8a5cSAndrea Arcangeli * is_cow_mapping() returns true. 1167cddb8a5cSAndrea Arcangeli */ 1168c78f4636SPeter Xu is_cow = is_cow_mapping(src_vma->vm_flags); 1169ac46d4f3SJérôme Glisse 1170ac46d4f3SJérôme Glisse if (is_cow) { 11717269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, 1172c78f4636SPeter Xu 0, src_vma, src_mm, addr, end); 1173ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1174*57efa1feSJason Gunthorpe /* 1175*57efa1feSJason Gunthorpe * Disabling preemption is not needed for the write side, as 1176*57efa1feSJason Gunthorpe * the read side doesn't spin, but goes to the mmap_lock. 1177*57efa1feSJason Gunthorpe * 1178*57efa1feSJason Gunthorpe * Use the raw variant of the seqcount_t write API to avoid 1179*57efa1feSJason Gunthorpe * lockdep complaining about preemptibility. 1180*57efa1feSJason Gunthorpe */ 1181*57efa1feSJason Gunthorpe mmap_assert_write_locked(src_mm); 1182*57efa1feSJason Gunthorpe raw_write_seqcount_begin(&src_mm->write_protect_seq); 1183ac46d4f3SJérôme Glisse } 1184cddb8a5cSAndrea Arcangeli 1185cddb8a5cSAndrea Arcangeli ret = 0; 11861da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 11871da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 11881da177e4SLinus Torvalds do { 11891da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 11901da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 11911da177e4SLinus Torvalds continue; 1192c78f4636SPeter Xu if (unlikely(copy_p4d_range(dst_vma, src_vma, dst_pgd, src_pgd, 1193c78f4636SPeter Xu addr, next))) { 1194cddb8a5cSAndrea Arcangeli ret = -ENOMEM; 1195cddb8a5cSAndrea Arcangeli break; 1196cddb8a5cSAndrea Arcangeli } 11971da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 1198cddb8a5cSAndrea Arcangeli 1199*57efa1feSJason Gunthorpe if (is_cow) { 1200*57efa1feSJason Gunthorpe raw_write_seqcount_end(&src_mm->write_protect_seq); 1201ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1202*57efa1feSJason Gunthorpe } 1203cddb8a5cSAndrea Arcangeli return ret; 12041da177e4SLinus Torvalds } 12051da177e4SLinus Torvalds 120651c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 1207b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 12081da177e4SLinus Torvalds unsigned long addr, unsigned long end, 120997a89413SPeter Zijlstra struct zap_details *details) 12101da177e4SLinus Torvalds { 1211b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 1212d16dfc55SPeter Zijlstra int force_flush = 0; 1213d559db08SKAMEZAWA Hiroyuki int rss[NR_MM_COUNTERS]; 121497a89413SPeter Zijlstra spinlock_t *ptl; 12155f1a1907SSteven Rostedt pte_t *start_pte; 121697a89413SPeter Zijlstra pte_t *pte; 12178a5f14a2SKirill A. Shutemov swp_entry_t entry; 1218d559db08SKAMEZAWA Hiroyuki 1219ed6a7935SPeter Zijlstra tlb_change_page_size(tlb, PAGE_SIZE); 1220d16dfc55SPeter Zijlstra again: 1221e303297eSPeter Zijlstra init_rss_vec(rss); 12225f1a1907SSteven Rostedt start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 12235f1a1907SSteven Rostedt pte = start_pte; 12243ea27719SMel Gorman flush_tlb_batched_pending(mm); 12256606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 12261da177e4SLinus Torvalds do { 12271da177e4SLinus Torvalds pte_t ptent = *pte; 1228166f61b9STobin C Harding if (pte_none(ptent)) 12291da177e4SLinus Torvalds continue; 123051c6f666SRobin Holt 12317b167b68SMinchan Kim if (need_resched()) 12327b167b68SMinchan Kim break; 12337b167b68SMinchan Kim 12346f5e6b9eSHugh Dickins if (pte_present(ptent)) { 12356f5e6b9eSHugh Dickins struct page *page; 12366f5e6b9eSHugh Dickins 123725b2995aSChristoph Hellwig page = vm_normal_page(vma, addr, ptent); 12381da177e4SLinus Torvalds if (unlikely(details) && page) { 12391da177e4SLinus Torvalds /* 12401da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 12411da177e4SLinus Torvalds * invalidate cache without truncating: 12421da177e4SLinus Torvalds * unmap shared but keep private pages. 12431da177e4SLinus Torvalds */ 12441da177e4SLinus Torvalds if (details->check_mapping && 1245800d8c63SKirill A. Shutemov details->check_mapping != page_rmapping(page)) 12461da177e4SLinus Torvalds continue; 12471da177e4SLinus Torvalds } 1248b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 1249a600388dSZachary Amsden tlb->fullmm); 12501da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 12511da177e4SLinus Torvalds if (unlikely(!page)) 12521da177e4SLinus Torvalds continue; 1253eca56ff9SJerome Marchand 1254eca56ff9SJerome Marchand if (!PageAnon(page)) { 12551cf35d47SLinus Torvalds if (pte_dirty(ptent)) { 12561cf35d47SLinus Torvalds force_flush = 1; 12576237bcd9SHugh Dickins set_page_dirty(page); 12581cf35d47SLinus Torvalds } 12594917e5d0SJohannes Weiner if (pte_young(ptent) && 126064363aadSJoe Perches likely(!(vma->vm_flags & VM_SEQ_READ))) 1261bf3f3bc5SNick Piggin mark_page_accessed(page); 12626237bcd9SHugh Dickins } 1263eca56ff9SJerome Marchand rss[mm_counter(page)]--; 1264d281ee61SKirill A. Shutemov page_remove_rmap(page, false); 12653dc14741SHugh Dickins if (unlikely(page_mapcount(page) < 0)) 12663dc14741SHugh Dickins print_bad_pte(vma, addr, ptent, page); 1267e9d55e15SAneesh Kumar K.V if (unlikely(__tlb_remove_page(tlb, page))) { 12681cf35d47SLinus Torvalds force_flush = 1; 1269ce9ec37bSWill Deacon addr += PAGE_SIZE; 1270d16dfc55SPeter Zijlstra break; 12711cf35d47SLinus Torvalds } 12721da177e4SLinus Torvalds continue; 12731da177e4SLinus Torvalds } 12745042db43SJérôme Glisse 12755042db43SJérôme Glisse entry = pte_to_swp_entry(ptent); 1276463b7a17SRalph Campbell if (is_device_private_entry(entry)) { 12775042db43SJérôme Glisse struct page *page = device_private_entry_to_page(entry); 12785042db43SJérôme Glisse 12795042db43SJérôme Glisse if (unlikely(details && details->check_mapping)) { 12805042db43SJérôme Glisse /* 12815042db43SJérôme Glisse * unmap_shared_mapping_pages() wants to 12825042db43SJérôme Glisse * invalidate cache without truncating: 12835042db43SJérôme Glisse * unmap shared but keep private pages. 12845042db43SJérôme Glisse */ 12855042db43SJérôme Glisse if (details->check_mapping != 12865042db43SJérôme Glisse page_rmapping(page)) 12875042db43SJérôme Glisse continue; 12885042db43SJérôme Glisse } 12895042db43SJérôme Glisse 12905042db43SJérôme Glisse pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 12915042db43SJérôme Glisse rss[mm_counter(page)]--; 12925042db43SJérôme Glisse page_remove_rmap(page, false); 12935042db43SJérôme Glisse put_page(page); 12945042db43SJérôme Glisse continue; 12955042db43SJérôme Glisse } 12965042db43SJérôme Glisse 12973e8715fdSKirill A. Shutemov /* If details->check_mapping, we leave swap entries. */ 12983e8715fdSKirill A. Shutemov if (unlikely(details)) 12991da177e4SLinus Torvalds continue; 1300b084d435SKAMEZAWA Hiroyuki 1301b084d435SKAMEZAWA Hiroyuki if (!non_swap_entry(entry)) 1302b084d435SKAMEZAWA Hiroyuki rss[MM_SWAPENTS]--; 13039f9f1acdSKonstantin Khlebnikov else if (is_migration_entry(entry)) { 13049f9f1acdSKonstantin Khlebnikov struct page *page; 13059f9f1acdSKonstantin Khlebnikov 13069f9f1acdSKonstantin Khlebnikov page = migration_entry_to_page(entry); 1307eca56ff9SJerome Marchand rss[mm_counter(page)]--; 13089f9f1acdSKonstantin Khlebnikov } 1309b084d435SKAMEZAWA Hiroyuki if (unlikely(!free_swap_and_cache(entry))) 13102509ef26SHugh Dickins print_bad_pte(vma, addr, ptent, NULL); 13119888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 131297a89413SPeter Zijlstra } while (pte++, addr += PAGE_SIZE, addr != end); 1313ae859762SHugh Dickins 1314d559db08SKAMEZAWA Hiroyuki add_mm_rss_vec(mm, rss); 13156606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 131651c6f666SRobin Holt 13171cf35d47SLinus Torvalds /* Do the actual TLB flush before dropping ptl */ 1318fb7332a9SWill Deacon if (force_flush) 13191cf35d47SLinus Torvalds tlb_flush_mmu_tlbonly(tlb); 13201cf35d47SLinus Torvalds pte_unmap_unlock(start_pte, ptl); 13211cf35d47SLinus Torvalds 13221cf35d47SLinus Torvalds /* 13231cf35d47SLinus Torvalds * If we forced a TLB flush (either due to running out of 13241cf35d47SLinus Torvalds * batch buffers or because we needed to flush dirty TLB 13251cf35d47SLinus Torvalds * entries before releasing the ptl), free the batched 13261cf35d47SLinus Torvalds * memory too. Restart if we didn't do everything. 13271cf35d47SLinus Torvalds */ 13281cf35d47SLinus Torvalds if (force_flush) { 13291cf35d47SLinus Torvalds force_flush = 0; 1330fa0aafb8SPeter Zijlstra tlb_flush_mmu(tlb); 13317b167b68SMinchan Kim } 13327b167b68SMinchan Kim 13337b167b68SMinchan Kim if (addr != end) { 13347b167b68SMinchan Kim cond_resched(); 1335d16dfc55SPeter Zijlstra goto again; 1336d16dfc55SPeter Zijlstra } 1337d16dfc55SPeter Zijlstra 133851c6f666SRobin Holt return addr; 13391da177e4SLinus Torvalds } 13401da177e4SLinus Torvalds 134151c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 1342b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 13431da177e4SLinus Torvalds unsigned long addr, unsigned long end, 134497a89413SPeter Zijlstra struct zap_details *details) 13451da177e4SLinus Torvalds { 13461da177e4SLinus Torvalds pmd_t *pmd; 13471da177e4SLinus Torvalds unsigned long next; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 13501da177e4SLinus Torvalds do { 13511da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 135284c3fc4eSZi Yan if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { 135353406ed1SHugh Dickins if (next - addr != HPAGE_PMD_SIZE) 1354fd60775aSDavid Rientjes __split_huge_pmd(vma, pmd, addr, false, NULL); 135553406ed1SHugh Dickins else if (zap_huge_pmd(tlb, vma, pmd, addr)) 13561a5a9906SAndrea Arcangeli goto next; 135771e3aac0SAndrea Arcangeli /* fall through */ 135871e3aac0SAndrea Arcangeli } 13591a5a9906SAndrea Arcangeli /* 13601a5a9906SAndrea Arcangeli * Here there can be other concurrent MADV_DONTNEED or 13611a5a9906SAndrea Arcangeli * trans huge page faults running, and if the pmd is 13621a5a9906SAndrea Arcangeli * none or trans huge it can change under us. This is 1363c1e8d7c6SMichel Lespinasse * because MADV_DONTNEED holds the mmap_lock in read 13641a5a9906SAndrea Arcangeli * mode. 13651a5a9906SAndrea Arcangeli */ 13661a5a9906SAndrea Arcangeli if (pmd_none_or_trans_huge_or_clear_bad(pmd)) 13671a5a9906SAndrea Arcangeli goto next; 136897a89413SPeter Zijlstra next = zap_pte_range(tlb, vma, pmd, addr, next, details); 13691a5a9906SAndrea Arcangeli next: 137097a89413SPeter Zijlstra cond_resched(); 137197a89413SPeter Zijlstra } while (pmd++, addr = next, addr != end); 137251c6f666SRobin Holt 137351c6f666SRobin Holt return addr; 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 137651c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 1377c2febafcSKirill A. Shutemov struct vm_area_struct *vma, p4d_t *p4d, 13781da177e4SLinus Torvalds unsigned long addr, unsigned long end, 137997a89413SPeter Zijlstra struct zap_details *details) 13801da177e4SLinus Torvalds { 13811da177e4SLinus Torvalds pud_t *pud; 13821da177e4SLinus Torvalds unsigned long next; 13831da177e4SLinus Torvalds 1384c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 13851da177e4SLinus Torvalds do { 13861da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1387a00cc7d9SMatthew Wilcox if (pud_trans_huge(*pud) || pud_devmap(*pud)) { 1388a00cc7d9SMatthew Wilcox if (next - addr != HPAGE_PUD_SIZE) { 138942fc5414SMichel Lespinasse mmap_assert_locked(tlb->mm); 1390a00cc7d9SMatthew Wilcox split_huge_pud(vma, pud, addr); 1391a00cc7d9SMatthew Wilcox } else if (zap_huge_pud(tlb, vma, pud, addr)) 1392a00cc7d9SMatthew Wilcox goto next; 1393a00cc7d9SMatthew Wilcox /* fall through */ 1394a00cc7d9SMatthew Wilcox } 139597a89413SPeter Zijlstra if (pud_none_or_clear_bad(pud)) 13961da177e4SLinus Torvalds continue; 139797a89413SPeter Zijlstra next = zap_pmd_range(tlb, vma, pud, addr, next, details); 1398a00cc7d9SMatthew Wilcox next: 1399a00cc7d9SMatthew Wilcox cond_resched(); 140097a89413SPeter Zijlstra } while (pud++, addr = next, addr != end); 140151c6f666SRobin Holt 140251c6f666SRobin Holt return addr; 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 1405c2febafcSKirill A. Shutemov static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, 1406c2febafcSKirill A. Shutemov struct vm_area_struct *vma, pgd_t *pgd, 1407c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 1408c2febafcSKirill A. Shutemov struct zap_details *details) 1409c2febafcSKirill A. Shutemov { 1410c2febafcSKirill A. Shutemov p4d_t *p4d; 1411c2febafcSKirill A. Shutemov unsigned long next; 1412c2febafcSKirill A. Shutemov 1413c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 1414c2febafcSKirill A. Shutemov do { 1415c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 1416c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 1417c2febafcSKirill A. Shutemov continue; 1418c2febafcSKirill A. Shutemov next = zap_pud_range(tlb, vma, p4d, addr, next, details); 1419c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 1420c2febafcSKirill A. Shutemov 1421c2febafcSKirill A. Shutemov return addr; 1422c2febafcSKirill A. Shutemov } 1423c2febafcSKirill A. Shutemov 1424aac45363SMichal Hocko void unmap_page_range(struct mmu_gather *tlb, 142551c6f666SRobin Holt struct vm_area_struct *vma, 14261da177e4SLinus Torvalds unsigned long addr, unsigned long end, 142797a89413SPeter Zijlstra struct zap_details *details) 14281da177e4SLinus Torvalds { 14291da177e4SLinus Torvalds pgd_t *pgd; 14301da177e4SLinus Torvalds unsigned long next; 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds BUG_ON(addr >= end); 14331da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 14341da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 14351da177e4SLinus Torvalds do { 14361da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 143797a89413SPeter Zijlstra if (pgd_none_or_clear_bad(pgd)) 14381da177e4SLinus Torvalds continue; 1439c2febafcSKirill A. Shutemov next = zap_p4d_range(tlb, vma, pgd, addr, next, details); 144097a89413SPeter Zijlstra } while (pgd++, addr = next, addr != end); 14411da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds 1444f5cc4eefSAl Viro 1445f5cc4eefSAl Viro static void unmap_single_vma(struct mmu_gather *tlb, 14461da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 14474f74d2c8SLinus Torvalds unsigned long end_addr, 14481da177e4SLinus Torvalds struct zap_details *details) 14491da177e4SLinus Torvalds { 1450f5cc4eefSAl Viro unsigned long start = max(vma->vm_start, start_addr); 14511da177e4SLinus Torvalds unsigned long end; 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds if (start >= vma->vm_end) 1454f5cc4eefSAl Viro return; 14551da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 14561da177e4SLinus Torvalds if (end <= vma->vm_start) 1457f5cc4eefSAl Viro return; 14581da177e4SLinus Torvalds 1459cbc91f71SSrikar Dronamraju if (vma->vm_file) 1460cbc91f71SSrikar Dronamraju uprobe_munmap(vma, start, end); 1461cbc91f71SSrikar Dronamraju 1462b3b9c293SKonstantin Khlebnikov if (unlikely(vma->vm_flags & VM_PFNMAP)) 14635180da41SSuresh Siddha untrack_pfn(vma, 0, 0); 14642ab64037Svenkatesh.pallipadi@intel.com 14658b2a1238SAl Viro if (start != end) { 146651c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 1467a137e1ccSAndi Kleen /* 1468a137e1ccSAndi Kleen * It is undesirable to test vma->vm_file as it 1469a137e1ccSAndi Kleen * should be non-null for valid hugetlb area. 1470a137e1ccSAndi Kleen * However, vm_file will be NULL in the error 14717aa6b4adSDavidlohr Bueso * cleanup path of mmap_region. When 1472a137e1ccSAndi Kleen * hugetlbfs ->mmap method fails, 14737aa6b4adSDavidlohr Bueso * mmap_region() nullifies vma->vm_file 1474a137e1ccSAndi Kleen * before calling this function to clean up. 1475a137e1ccSAndi Kleen * Since no pte has actually been setup, it is 1476a137e1ccSAndi Kleen * safe to do nothing in this case. 1477a137e1ccSAndi Kleen */ 147824669e58SAneesh Kumar K.V if (vma->vm_file) { 147983cde9e8SDavidlohr Bueso i_mmap_lock_write(vma->vm_file->f_mapping); 1480d833352aSMel Gorman __unmap_hugepage_range_final(tlb, vma, start, end, NULL); 148183cde9e8SDavidlohr Bueso i_mmap_unlock_write(vma->vm_file->f_mapping); 148224669e58SAneesh Kumar K.V } 148351c6f666SRobin Holt } else 1484038c7aa1SAl Viro unmap_page_range(tlb, vma, start, end, details); 148597a89413SPeter Zijlstra } 148651c6f666SRobin Holt } 14871da177e4SLinus Torvalds 1488f5cc4eefSAl Viro /** 1489f5cc4eefSAl Viro * unmap_vmas - unmap a range of memory covered by a list of vma's 1490f5cc4eefSAl Viro * @tlb: address of the caller's struct mmu_gather 1491f5cc4eefSAl Viro * @vma: the starting vma 1492f5cc4eefSAl Viro * @start_addr: virtual address at which to start unmapping 1493f5cc4eefSAl Viro * @end_addr: virtual address at which to end unmapping 1494f5cc4eefSAl Viro * 1495f5cc4eefSAl Viro * Unmap all pages in the vma list. 1496f5cc4eefSAl Viro * 1497f5cc4eefSAl Viro * Only addresses between `start' and `end' will be unmapped. 1498f5cc4eefSAl Viro * 1499f5cc4eefSAl Viro * The VMA list must be sorted in ascending virtual address order. 1500f5cc4eefSAl Viro * 1501f5cc4eefSAl Viro * unmap_vmas() assumes that the caller will flush the whole unmapped address 1502f5cc4eefSAl Viro * range after unmap_vmas() returns. So the only responsibility here is to 1503f5cc4eefSAl Viro * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 1504f5cc4eefSAl Viro * drops the lock and schedules. 1505f5cc4eefSAl Viro */ 1506f5cc4eefSAl Viro void unmap_vmas(struct mmu_gather *tlb, 1507f5cc4eefSAl Viro struct vm_area_struct *vma, unsigned long start_addr, 15084f74d2c8SLinus Torvalds unsigned long end_addr) 1509f5cc4eefSAl Viro { 1510ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1511f5cc4eefSAl Viro 15126f4f13e8SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, 15136f4f13e8SJérôme Glisse start_addr, end_addr); 1514ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1515f5cc4eefSAl Viro for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) 15164f74d2c8SLinus Torvalds unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); 1517ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds 15201da177e4SLinus Torvalds /** 15211da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 15221da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 1523eb4546bbSRandy Dunlap * @start: starting address of pages to zap 15241da177e4SLinus Torvalds * @size: number of bytes to zap 1525f5cc4eefSAl Viro * 1526f5cc4eefSAl Viro * Caller must protect the VMA list 15271da177e4SLinus Torvalds */ 15287e027b14SLinus Torvalds void zap_page_range(struct vm_area_struct *vma, unsigned long start, 1529ecf1385dSKirill A. Shutemov unsigned long size) 15301da177e4SLinus Torvalds { 1531ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1532d16dfc55SPeter Zijlstra struct mmu_gather tlb; 15331da177e4SLinus Torvalds 15341da177e4SLinus Torvalds lru_add_drain(); 15357269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, 15366f4f13e8SJérôme Glisse start, start + size); 1537ac46d4f3SJérôme Glisse tlb_gather_mmu(&tlb, vma->vm_mm, start, range.end); 1538ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1539ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1540ac46d4f3SJérôme Glisse for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next) 1541ac46d4f3SJérôme Glisse unmap_single_vma(&tlb, vma, start, range.end, NULL); 1542ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1543ac46d4f3SJérôme Glisse tlb_finish_mmu(&tlb, start, range.end); 15441da177e4SLinus Torvalds } 15451da177e4SLinus Torvalds 1546c627f9ccSJack Steiner /** 1547f5cc4eefSAl Viro * zap_page_range_single - remove user pages in a given range 1548f5cc4eefSAl Viro * @vma: vm_area_struct holding the applicable pages 1549f5cc4eefSAl Viro * @address: starting address of pages to zap 1550f5cc4eefSAl Viro * @size: number of bytes to zap 15518a5f14a2SKirill A. Shutemov * @details: details of shared cache invalidation 1552f5cc4eefSAl Viro * 1553f5cc4eefSAl Viro * The range must fit into one VMA. 1554f5cc4eefSAl Viro */ 1555f5cc4eefSAl Viro static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, 1556f5cc4eefSAl Viro unsigned long size, struct zap_details *details) 1557f5cc4eefSAl Viro { 1558ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1559f5cc4eefSAl Viro struct mmu_gather tlb; 1560f5cc4eefSAl Viro 1561f5cc4eefSAl Viro lru_add_drain(); 15627269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, 15636f4f13e8SJérôme Glisse address, address + size); 1564ac46d4f3SJérôme Glisse tlb_gather_mmu(&tlb, vma->vm_mm, address, range.end); 1565ac46d4f3SJérôme Glisse update_hiwater_rss(vma->vm_mm); 1566ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1567ac46d4f3SJérôme Glisse unmap_single_vma(&tlb, vma, address, range.end, details); 1568ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1569ac46d4f3SJérôme Glisse tlb_finish_mmu(&tlb, address, range.end); 15701da177e4SLinus Torvalds } 15711da177e4SLinus Torvalds 1572c627f9ccSJack Steiner /** 1573c627f9ccSJack Steiner * zap_vma_ptes - remove ptes mapping the vma 1574c627f9ccSJack Steiner * @vma: vm_area_struct holding ptes to be zapped 1575c627f9ccSJack Steiner * @address: starting address of pages to zap 1576c627f9ccSJack Steiner * @size: number of bytes to zap 1577c627f9ccSJack Steiner * 1578c627f9ccSJack Steiner * This function only unmaps ptes assigned to VM_PFNMAP vmas. 1579c627f9ccSJack Steiner * 1580c627f9ccSJack Steiner * The entire address range must be fully contained within the vma. 1581c627f9ccSJack Steiner * 1582c627f9ccSJack Steiner */ 158327d036e3SLeon Romanovsky void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 1584c627f9ccSJack Steiner unsigned long size) 1585c627f9ccSJack Steiner { 1586c627f9ccSJack Steiner if (address < vma->vm_start || address + size > vma->vm_end || 1587c627f9ccSJack Steiner !(vma->vm_flags & VM_PFNMAP)) 158827d036e3SLeon Romanovsky return; 158927d036e3SLeon Romanovsky 1590f5cc4eefSAl Viro zap_page_range_single(vma, address, size, NULL); 1591c627f9ccSJack Steiner } 1592c627f9ccSJack Steiner EXPORT_SYMBOL_GPL(zap_vma_ptes); 1593c627f9ccSJack Steiner 15948cd3984dSArjun Roy static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr) 1595c9cfcddfSLinus Torvalds { 1596c2febafcSKirill A. Shutemov pgd_t *pgd; 1597c2febafcSKirill A. Shutemov p4d_t *p4d; 1598c2febafcSKirill A. Shutemov pud_t *pud; 1599c2febafcSKirill A. Shutemov pmd_t *pmd; 1600c2febafcSKirill A. Shutemov 1601c2febafcSKirill A. Shutemov pgd = pgd_offset(mm, addr); 1602c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 1603c2febafcSKirill A. Shutemov if (!p4d) 1604c2febafcSKirill A. Shutemov return NULL; 1605c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 1606c2febafcSKirill A. Shutemov if (!pud) 1607c2febafcSKirill A. Shutemov return NULL; 1608c2febafcSKirill A. Shutemov pmd = pmd_alloc(mm, pud, addr); 1609c2febafcSKirill A. Shutemov if (!pmd) 1610c2febafcSKirill A. Shutemov return NULL; 1611c2febafcSKirill A. Shutemov 1612f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 16138cd3984dSArjun Roy return pmd; 16148cd3984dSArjun Roy } 16158cd3984dSArjun Roy 16168cd3984dSArjun Roy pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, 16178cd3984dSArjun Roy spinlock_t **ptl) 16188cd3984dSArjun Roy { 16198cd3984dSArjun Roy pmd_t *pmd = walk_to_pmd(mm, addr); 16208cd3984dSArjun Roy 16218cd3984dSArjun Roy if (!pmd) 16228cd3984dSArjun Roy return NULL; 1623c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1624c9cfcddfSLinus Torvalds } 1625c9cfcddfSLinus Torvalds 16268efd6f5bSArjun Roy static int validate_page_before_insert(struct page *page) 16278efd6f5bSArjun Roy { 16288efd6f5bSArjun Roy if (PageAnon(page) || PageSlab(page) || page_has_type(page)) 16298efd6f5bSArjun Roy return -EINVAL; 16308efd6f5bSArjun Roy flush_dcache_page(page); 16318efd6f5bSArjun Roy return 0; 16328efd6f5bSArjun Roy } 16338efd6f5bSArjun Roy 16348efd6f5bSArjun Roy static int insert_page_into_pte_locked(struct mm_struct *mm, pte_t *pte, 16358efd6f5bSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 16368efd6f5bSArjun Roy { 16378efd6f5bSArjun Roy if (!pte_none(*pte)) 16388efd6f5bSArjun Roy return -EBUSY; 16398efd6f5bSArjun Roy /* Ok, finally just insert the thing.. */ 16408efd6f5bSArjun Roy get_page(page); 16418efd6f5bSArjun Roy inc_mm_counter_fast(mm, mm_counter_file(page)); 16428efd6f5bSArjun Roy page_add_file_rmap(page, false); 16438efd6f5bSArjun Roy set_pte_at(mm, addr, pte, mk_pte(page, prot)); 16448efd6f5bSArjun Roy return 0; 16458efd6f5bSArjun Roy } 16468efd6f5bSArjun Roy 16471da177e4SLinus Torvalds /* 1648238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1649238f58d8SLinus Torvalds * 1650238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1651238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1652238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1653238f58d8SLinus Torvalds */ 1654423bad60SNick Piggin static int insert_page(struct vm_area_struct *vma, unsigned long addr, 1655423bad60SNick Piggin struct page *page, pgprot_t prot) 1656238f58d8SLinus Torvalds { 1657423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1658238f58d8SLinus Torvalds int retval; 1659238f58d8SLinus Torvalds pte_t *pte; 1660238f58d8SLinus Torvalds spinlock_t *ptl; 1661238f58d8SLinus Torvalds 16628efd6f5bSArjun Roy retval = validate_page_before_insert(page); 16638efd6f5bSArjun Roy if (retval) 16645b4e655eSKAMEZAWA Hiroyuki goto out; 1665238f58d8SLinus Torvalds retval = -ENOMEM; 1666c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1667238f58d8SLinus Torvalds if (!pte) 16685b4e655eSKAMEZAWA Hiroyuki goto out; 16698efd6f5bSArjun Roy retval = insert_page_into_pte_locked(mm, pte, addr, page, prot); 1670238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1671238f58d8SLinus Torvalds out: 1672238f58d8SLinus Torvalds return retval; 1673238f58d8SLinus Torvalds } 1674238f58d8SLinus Torvalds 16758cd3984dSArjun Roy #ifdef pte_index 16767f70c2a6SArjun Roy static int insert_page_in_batch_locked(struct mm_struct *mm, pte_t *pte, 16778cd3984dSArjun Roy unsigned long addr, struct page *page, pgprot_t prot) 16788cd3984dSArjun Roy { 16798cd3984dSArjun Roy int err; 16808cd3984dSArjun Roy 16818cd3984dSArjun Roy if (!page_count(page)) 16828cd3984dSArjun Roy return -EINVAL; 16838cd3984dSArjun Roy err = validate_page_before_insert(page); 16847f70c2a6SArjun Roy if (err) 16857f70c2a6SArjun Roy return err; 16867f70c2a6SArjun Roy return insert_page_into_pte_locked(mm, pte, addr, page, prot); 16878cd3984dSArjun Roy } 16888cd3984dSArjun Roy 16898cd3984dSArjun Roy /* insert_pages() amortizes the cost of spinlock operations 16908cd3984dSArjun Roy * when inserting pages in a loop. Arch *must* define pte_index. 16918cd3984dSArjun Roy */ 16928cd3984dSArjun Roy static int insert_pages(struct vm_area_struct *vma, unsigned long addr, 16938cd3984dSArjun Roy struct page **pages, unsigned long *num, pgprot_t prot) 16948cd3984dSArjun Roy { 16958cd3984dSArjun Roy pmd_t *pmd = NULL; 16967f70c2a6SArjun Roy pte_t *start_pte, *pte; 16977f70c2a6SArjun Roy spinlock_t *pte_lock; 16988cd3984dSArjun Roy struct mm_struct *const mm = vma->vm_mm; 16998cd3984dSArjun Roy unsigned long curr_page_idx = 0; 17008cd3984dSArjun Roy unsigned long remaining_pages_total = *num; 17018cd3984dSArjun Roy unsigned long pages_to_write_in_pmd; 17028cd3984dSArjun Roy int ret; 17038cd3984dSArjun Roy more: 17048cd3984dSArjun Roy ret = -EFAULT; 17058cd3984dSArjun Roy pmd = walk_to_pmd(mm, addr); 17068cd3984dSArjun Roy if (!pmd) 17078cd3984dSArjun Roy goto out; 17088cd3984dSArjun Roy 17098cd3984dSArjun Roy pages_to_write_in_pmd = min_t(unsigned long, 17108cd3984dSArjun Roy remaining_pages_total, PTRS_PER_PTE - pte_index(addr)); 17118cd3984dSArjun Roy 17128cd3984dSArjun Roy /* Allocate the PTE if necessary; takes PMD lock once only. */ 17138cd3984dSArjun Roy ret = -ENOMEM; 17148cd3984dSArjun Roy if (pte_alloc(mm, pmd)) 17158cd3984dSArjun Roy goto out; 17168cd3984dSArjun Roy 17178cd3984dSArjun Roy while (pages_to_write_in_pmd) { 17188cd3984dSArjun Roy int pte_idx = 0; 17198cd3984dSArjun Roy const int batch_size = min_t(int, pages_to_write_in_pmd, 8); 17208cd3984dSArjun Roy 17217f70c2a6SArjun Roy start_pte = pte_offset_map_lock(mm, pmd, addr, &pte_lock); 17227f70c2a6SArjun Roy for (pte = start_pte; pte_idx < batch_size; ++pte, ++pte_idx) { 17237f70c2a6SArjun Roy int err = insert_page_in_batch_locked(mm, pte, 17248cd3984dSArjun Roy addr, pages[curr_page_idx], prot); 17258cd3984dSArjun Roy if (unlikely(err)) { 17267f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 17278cd3984dSArjun Roy ret = err; 17288cd3984dSArjun Roy remaining_pages_total -= pte_idx; 17298cd3984dSArjun Roy goto out; 17308cd3984dSArjun Roy } 17318cd3984dSArjun Roy addr += PAGE_SIZE; 17328cd3984dSArjun Roy ++curr_page_idx; 17338cd3984dSArjun Roy } 17347f70c2a6SArjun Roy pte_unmap_unlock(start_pte, pte_lock); 17358cd3984dSArjun Roy pages_to_write_in_pmd -= batch_size; 17368cd3984dSArjun Roy remaining_pages_total -= batch_size; 17378cd3984dSArjun Roy } 17388cd3984dSArjun Roy if (remaining_pages_total) 17398cd3984dSArjun Roy goto more; 17408cd3984dSArjun Roy ret = 0; 17418cd3984dSArjun Roy out: 17428cd3984dSArjun Roy *num = remaining_pages_total; 17438cd3984dSArjun Roy return ret; 17448cd3984dSArjun Roy } 17458cd3984dSArjun Roy #endif /* ifdef pte_index */ 17468cd3984dSArjun Roy 17478cd3984dSArjun Roy /** 17488cd3984dSArjun Roy * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock. 17498cd3984dSArjun Roy * @vma: user vma to map to 17508cd3984dSArjun Roy * @addr: target start user address of these pages 17518cd3984dSArjun Roy * @pages: source kernel pages 17528cd3984dSArjun Roy * @num: in: number of pages to map. out: number of pages that were *not* 17538cd3984dSArjun Roy * mapped. (0 means all pages were successfully mapped). 17548cd3984dSArjun Roy * 17558cd3984dSArjun Roy * Preferred over vm_insert_page() when inserting multiple pages. 17568cd3984dSArjun Roy * 17578cd3984dSArjun Roy * In case of error, we may have mapped a subset of the provided 17588cd3984dSArjun Roy * pages. It is the caller's responsibility to account for this case. 17598cd3984dSArjun Roy * 17608cd3984dSArjun Roy * The same restrictions apply as in vm_insert_page(). 17618cd3984dSArjun Roy */ 17628cd3984dSArjun Roy int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr, 17638cd3984dSArjun Roy struct page **pages, unsigned long *num) 17648cd3984dSArjun Roy { 17658cd3984dSArjun Roy #ifdef pte_index 17668cd3984dSArjun Roy const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1; 17678cd3984dSArjun Roy 17688cd3984dSArjun Roy if (addr < vma->vm_start || end_addr >= vma->vm_end) 17698cd3984dSArjun Roy return -EFAULT; 17708cd3984dSArjun Roy if (!(vma->vm_flags & VM_MIXEDMAP)) { 1771d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 17728cd3984dSArjun Roy BUG_ON(vma->vm_flags & VM_PFNMAP); 17738cd3984dSArjun Roy vma->vm_flags |= VM_MIXEDMAP; 17748cd3984dSArjun Roy } 17758cd3984dSArjun Roy /* Defer page refcount checking till we're about to map that page. */ 17768cd3984dSArjun Roy return insert_pages(vma, addr, pages, num, vma->vm_page_prot); 17778cd3984dSArjun Roy #else 17788cd3984dSArjun Roy unsigned long idx = 0, pgcount = *num; 177945779b03STom Rix int err = -EINVAL; 17808cd3984dSArjun Roy 17818cd3984dSArjun Roy for (; idx < pgcount; ++idx) { 17828cd3984dSArjun Roy err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]); 17838cd3984dSArjun Roy if (err) 17848cd3984dSArjun Roy break; 17858cd3984dSArjun Roy } 17868cd3984dSArjun Roy *num = pgcount - idx; 17878cd3984dSArjun Roy return err; 17888cd3984dSArjun Roy #endif /* ifdef pte_index */ 17898cd3984dSArjun Roy } 17908cd3984dSArjun Roy EXPORT_SYMBOL(vm_insert_pages); 17918cd3984dSArjun Roy 1792bfa5bf6dSRolf Eike Beer /** 1793bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1794bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1795bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1796bfa5bf6dSRolf Eike Beer * @page: source kernel page 1797bfa5bf6dSRolf Eike Beer * 1798a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1799a145dd41SLinus Torvalds * into a user vma. 1800a145dd41SLinus Torvalds * 1801a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1802a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1803a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 18048dfcc9baSNick Piggin * (see split_page()). 1805a145dd41SLinus Torvalds * 1806a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1807a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1808a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1809a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1810a145dd41SLinus Torvalds * ask for a shared writable mapping! 1811a145dd41SLinus Torvalds * 1812a145dd41SLinus Torvalds * The page does not need to be reserved. 18134b6e1e37SKonstantin Khlebnikov * 18144b6e1e37SKonstantin Khlebnikov * Usually this function is called from f_op->mmap() handler 1815c1e8d7c6SMichel Lespinasse * under mm->mmap_lock write-lock, so it can change vma->vm_flags. 18164b6e1e37SKonstantin Khlebnikov * Caller must set VM_MIXEDMAP on vma if it wants to call this 18174b6e1e37SKonstantin Khlebnikov * function from other places, for example from page-fault handler. 1818a862f68aSMike Rapoport * 1819a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 1820a145dd41SLinus Torvalds */ 1821423bad60SNick Piggin int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 1822423bad60SNick Piggin struct page *page) 1823a145dd41SLinus Torvalds { 1824a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1825a145dd41SLinus Torvalds return -EFAULT; 1826a145dd41SLinus Torvalds if (!page_count(page)) 1827a145dd41SLinus Torvalds return -EINVAL; 18284b6e1e37SKonstantin Khlebnikov if (!(vma->vm_flags & VM_MIXEDMAP)) { 1829d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(vma->vm_mm)); 18304b6e1e37SKonstantin Khlebnikov BUG_ON(vma->vm_flags & VM_PFNMAP); 18314b6e1e37SKonstantin Khlebnikov vma->vm_flags |= VM_MIXEDMAP; 18324b6e1e37SKonstantin Khlebnikov } 1833423bad60SNick Piggin return insert_page(vma, addr, page, vma->vm_page_prot); 1834a145dd41SLinus Torvalds } 1835e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1836a145dd41SLinus Torvalds 1837a667d745SSouptick Joarder /* 1838a667d745SSouptick Joarder * __vm_map_pages - maps range of kernel pages into user vma 1839a667d745SSouptick Joarder * @vma: user vma to map to 1840a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 1841a667d745SSouptick Joarder * @num: number of pages in page array 1842a667d745SSouptick Joarder * @offset: user's requested vm_pgoff 1843a667d745SSouptick Joarder * 1844a667d745SSouptick Joarder * This allows drivers to map range of kernel pages into a user vma. 1845a667d745SSouptick Joarder * 1846a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 1847a667d745SSouptick Joarder */ 1848a667d745SSouptick Joarder static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, 1849a667d745SSouptick Joarder unsigned long num, unsigned long offset) 1850a667d745SSouptick Joarder { 1851a667d745SSouptick Joarder unsigned long count = vma_pages(vma); 1852a667d745SSouptick Joarder unsigned long uaddr = vma->vm_start; 1853a667d745SSouptick Joarder int ret, i; 1854a667d745SSouptick Joarder 1855a667d745SSouptick Joarder /* Fail if the user requested offset is beyond the end of the object */ 185696756fcbSMiguel Ojeda if (offset >= num) 1857a667d745SSouptick Joarder return -ENXIO; 1858a667d745SSouptick Joarder 1859a667d745SSouptick Joarder /* Fail if the user requested size exceeds available object size */ 1860a667d745SSouptick Joarder if (count > num - offset) 1861a667d745SSouptick Joarder return -ENXIO; 1862a667d745SSouptick Joarder 1863a667d745SSouptick Joarder for (i = 0; i < count; i++) { 1864a667d745SSouptick Joarder ret = vm_insert_page(vma, uaddr, pages[offset + i]); 1865a667d745SSouptick Joarder if (ret < 0) 1866a667d745SSouptick Joarder return ret; 1867a667d745SSouptick Joarder uaddr += PAGE_SIZE; 1868a667d745SSouptick Joarder } 1869a667d745SSouptick Joarder 1870a667d745SSouptick Joarder return 0; 1871a667d745SSouptick Joarder } 1872a667d745SSouptick Joarder 1873a667d745SSouptick Joarder /** 1874a667d745SSouptick Joarder * vm_map_pages - maps range of kernel pages starts with non zero offset 1875a667d745SSouptick Joarder * @vma: user vma to map to 1876a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 1877a667d745SSouptick Joarder * @num: number of pages in page array 1878a667d745SSouptick Joarder * 1879a667d745SSouptick Joarder * Maps an object consisting of @num pages, catering for the user's 1880a667d745SSouptick Joarder * requested vm_pgoff 1881a667d745SSouptick Joarder * 1882a667d745SSouptick Joarder * If we fail to insert any page into the vma, the function will return 1883a667d745SSouptick Joarder * immediately leaving any previously inserted pages present. Callers 1884a667d745SSouptick Joarder * from the mmap handler may immediately return the error as their caller 1885a667d745SSouptick Joarder * will destroy the vma, removing any successfully inserted pages. Other 1886a667d745SSouptick Joarder * callers should make their own arrangements for calling unmap_region(). 1887a667d745SSouptick Joarder * 1888a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 1889a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 1890a667d745SSouptick Joarder */ 1891a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 1892a667d745SSouptick Joarder unsigned long num) 1893a667d745SSouptick Joarder { 1894a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, vma->vm_pgoff); 1895a667d745SSouptick Joarder } 1896a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 1897a667d745SSouptick Joarder 1898a667d745SSouptick Joarder /** 1899a667d745SSouptick Joarder * vm_map_pages_zero - map range of kernel pages starts with zero offset 1900a667d745SSouptick Joarder * @vma: user vma to map to 1901a667d745SSouptick Joarder * @pages: pointer to array of source kernel pages 1902a667d745SSouptick Joarder * @num: number of pages in page array 1903a667d745SSouptick Joarder * 1904a667d745SSouptick Joarder * Similar to vm_map_pages(), except that it explicitly sets the offset 1905a667d745SSouptick Joarder * to 0. This function is intended for the drivers that did not consider 1906a667d745SSouptick Joarder * vm_pgoff. 1907a667d745SSouptick Joarder * 1908a667d745SSouptick Joarder * Context: Process context. Called by mmap handlers. 1909a667d745SSouptick Joarder * Return: 0 on success and error code otherwise. 1910a667d745SSouptick Joarder */ 1911a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 1912a667d745SSouptick Joarder unsigned long num) 1913a667d745SSouptick Joarder { 1914a667d745SSouptick Joarder return __vm_map_pages(vma, pages, num, 0); 1915a667d745SSouptick Joarder } 1916a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 1917a667d745SSouptick Joarder 19189b5a8e00SMatthew Wilcox static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1919b2770da6SRoss Zwisler pfn_t pfn, pgprot_t prot, bool mkwrite) 1920423bad60SNick Piggin { 1921423bad60SNick Piggin struct mm_struct *mm = vma->vm_mm; 1922423bad60SNick Piggin pte_t *pte, entry; 1923423bad60SNick Piggin spinlock_t *ptl; 1924423bad60SNick Piggin 1925423bad60SNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1926423bad60SNick Piggin if (!pte) 19279b5a8e00SMatthew Wilcox return VM_FAULT_OOM; 1928b2770da6SRoss Zwisler if (!pte_none(*pte)) { 1929b2770da6SRoss Zwisler if (mkwrite) { 1930b2770da6SRoss Zwisler /* 1931b2770da6SRoss Zwisler * For read faults on private mappings the PFN passed 1932b2770da6SRoss Zwisler * in may not match the PFN we have mapped if the 1933b2770da6SRoss Zwisler * mapped PFN is a writeable COW page. In the mkwrite 1934b2770da6SRoss Zwisler * case we are creating a writable PTE for a shared 1935f2c57d91SJan Kara * mapping and we expect the PFNs to match. If they 1936f2c57d91SJan Kara * don't match, we are likely racing with block 1937f2c57d91SJan Kara * allocation and mapping invalidation so just skip the 1938f2c57d91SJan Kara * update. 1939b2770da6SRoss Zwisler */ 1940f2c57d91SJan Kara if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { 1941f2c57d91SJan Kara WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); 1942423bad60SNick Piggin goto out_unlock; 1943f2c57d91SJan Kara } 1944cae85cb8SJan Kara entry = pte_mkyoung(*pte); 1945cae85cb8SJan Kara entry = maybe_mkwrite(pte_mkdirty(entry), vma); 1946cae85cb8SJan Kara if (ptep_set_access_flags(vma, addr, pte, entry, 1)) 1947cae85cb8SJan Kara update_mmu_cache(vma, addr, pte); 1948cae85cb8SJan Kara } 1949b2770da6SRoss Zwisler goto out_unlock; 1950b2770da6SRoss Zwisler } 1951423bad60SNick Piggin 1952423bad60SNick Piggin /* Ok, finally just insert the thing.. */ 195301c8f1c4SDan Williams if (pfn_t_devmap(pfn)) 195401c8f1c4SDan Williams entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); 195501c8f1c4SDan Williams else 195601c8f1c4SDan Williams entry = pte_mkspecial(pfn_t_pte(pfn, prot)); 1957b2770da6SRoss Zwisler 1958b2770da6SRoss Zwisler if (mkwrite) { 1959b2770da6SRoss Zwisler entry = pte_mkyoung(entry); 1960b2770da6SRoss Zwisler entry = maybe_mkwrite(pte_mkdirty(entry), vma); 1961b2770da6SRoss Zwisler } 1962b2770da6SRoss Zwisler 1963423bad60SNick Piggin set_pte_at(mm, addr, pte, entry); 19644b3073e1SRussell King update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ 1965423bad60SNick Piggin 1966423bad60SNick Piggin out_unlock: 1967423bad60SNick Piggin pte_unmap_unlock(pte, ptl); 19689b5a8e00SMatthew Wilcox return VM_FAULT_NOPAGE; 1969423bad60SNick Piggin } 1970423bad60SNick Piggin 1971f5e6d1d5SMatthew Wilcox /** 1972f5e6d1d5SMatthew Wilcox * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot 1973f5e6d1d5SMatthew Wilcox * @vma: user vma to map to 1974f5e6d1d5SMatthew Wilcox * @addr: target user address of this page 1975f5e6d1d5SMatthew Wilcox * @pfn: source kernel pfn 1976f5e6d1d5SMatthew Wilcox * @pgprot: pgprot flags for the inserted page 1977f5e6d1d5SMatthew Wilcox * 1978a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_pfn(), except that it allows drivers 1979f5e6d1d5SMatthew Wilcox * to override pgprot on a per-page basis. 1980f5e6d1d5SMatthew Wilcox * 1981f5e6d1d5SMatthew Wilcox * This only makes sense for IO mappings, and it makes no sense for 1982f5e6d1d5SMatthew Wilcox * COW mappings. In general, using multiple vmas is preferable; 1983ae2b01f3SMatthew Wilcox * vmf_insert_pfn_prot should only be used if using multiple VMAs is 1984f5e6d1d5SMatthew Wilcox * impractical. 1985f5e6d1d5SMatthew Wilcox * 1986574c5b3dSThomas Hellstrom * See vmf_insert_mixed_prot() for a discussion of the implication of using 1987574c5b3dSThomas Hellstrom * a value of @pgprot different from that of @vma->vm_page_prot. 1988574c5b3dSThomas Hellstrom * 1989ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 1990f5e6d1d5SMatthew Wilcox * Return: vm_fault_t value. 1991f5e6d1d5SMatthew Wilcox */ 1992f5e6d1d5SMatthew Wilcox vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, 1993f5e6d1d5SMatthew Wilcox unsigned long pfn, pgprot_t pgprot) 1994f5e6d1d5SMatthew Wilcox { 19956d958546SMatthew Wilcox /* 19966d958546SMatthew Wilcox * Technically, architectures with pte_special can avoid all these 19976d958546SMatthew Wilcox * restrictions (same for remap_pfn_range). However we would like 19986d958546SMatthew Wilcox * consistency in testing and feature parity among all, so we should 19996d958546SMatthew Wilcox * try to keep these invariants in place for everybody. 20006d958546SMatthew Wilcox */ 20016d958546SMatthew Wilcox BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); 20026d958546SMatthew Wilcox BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == 20036d958546SMatthew Wilcox (VM_PFNMAP|VM_MIXEDMAP)); 20046d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); 20056d958546SMatthew Wilcox BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); 20066d958546SMatthew Wilcox 20076d958546SMatthew Wilcox if (addr < vma->vm_start || addr >= vma->vm_end) 20086d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 20096d958546SMatthew Wilcox 20106d958546SMatthew Wilcox if (!pfn_modify_allowed(pfn, pgprot)) 20116d958546SMatthew Wilcox return VM_FAULT_SIGBUS; 20126d958546SMatthew Wilcox 20136d958546SMatthew Wilcox track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); 20146d958546SMatthew Wilcox 20159b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, 20166d958546SMatthew Wilcox false); 2017f5e6d1d5SMatthew Wilcox } 2018f5e6d1d5SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn_prot); 2019e0dc0d8fSNick Piggin 2020ae2b01f3SMatthew Wilcox /** 2021ae2b01f3SMatthew Wilcox * vmf_insert_pfn - insert single pfn into user vma 2022ae2b01f3SMatthew Wilcox * @vma: user vma to map to 2023ae2b01f3SMatthew Wilcox * @addr: target user address of this page 2024ae2b01f3SMatthew Wilcox * @pfn: source kernel pfn 2025ae2b01f3SMatthew Wilcox * 2026ae2b01f3SMatthew Wilcox * Similar to vm_insert_page, this allows drivers to insert individual pages 2027ae2b01f3SMatthew Wilcox * they've allocated into a user vma. Same comments apply. 2028ae2b01f3SMatthew Wilcox * 2029ae2b01f3SMatthew Wilcox * This function should only be called from a vm_ops->fault handler, and 2030ae2b01f3SMatthew Wilcox * in that case the handler should return the result of this function. 2031ae2b01f3SMatthew Wilcox * 2032ae2b01f3SMatthew Wilcox * vma cannot be a COW mapping. 2033ae2b01f3SMatthew Wilcox * 2034ae2b01f3SMatthew Wilcox * As this is called only for pages that do not currently exist, we 2035ae2b01f3SMatthew Wilcox * do not need to flush old virtual caches or the TLB. 2036ae2b01f3SMatthew Wilcox * 2037ae2b01f3SMatthew Wilcox * Context: Process context. May allocate using %GFP_KERNEL. 2038ae2b01f3SMatthew Wilcox * Return: vm_fault_t value. 2039ae2b01f3SMatthew Wilcox */ 2040ae2b01f3SMatthew Wilcox vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 2041ae2b01f3SMatthew Wilcox unsigned long pfn) 2042ae2b01f3SMatthew Wilcox { 2043ae2b01f3SMatthew Wilcox return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); 2044ae2b01f3SMatthew Wilcox } 2045ae2b01f3SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_pfn); 2046ae2b01f3SMatthew Wilcox 2047785a3fabSDan Williams static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) 2048785a3fabSDan Williams { 2049785a3fabSDan Williams /* these checks mirror the abort conditions in vm_normal_page */ 2050785a3fabSDan Williams if (vma->vm_flags & VM_MIXEDMAP) 2051785a3fabSDan Williams return true; 2052785a3fabSDan Williams if (pfn_t_devmap(pfn)) 2053785a3fabSDan Williams return true; 2054785a3fabSDan Williams if (pfn_t_special(pfn)) 2055785a3fabSDan Williams return true; 2056785a3fabSDan Williams if (is_zero_pfn(pfn_t_to_pfn(pfn))) 2057785a3fabSDan Williams return true; 2058785a3fabSDan Williams return false; 2059785a3fabSDan Williams } 2060785a3fabSDan Williams 206179f3aa5bSMatthew Wilcox static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, 2062574c5b3dSThomas Hellstrom unsigned long addr, pfn_t pfn, pgprot_t pgprot, 2063574c5b3dSThomas Hellstrom bool mkwrite) 2064423bad60SNick Piggin { 206579f3aa5bSMatthew Wilcox int err; 206687744ab3SDan Williams 2067785a3fabSDan Williams BUG_ON(!vm_mixed_ok(vma, pfn)); 2068423bad60SNick Piggin 2069423bad60SNick Piggin if (addr < vma->vm_start || addr >= vma->vm_end) 207079f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 2071308a047cSBorislav Petkov 2072308a047cSBorislav Petkov track_pfn_insert(vma, &pgprot, pfn); 2073423bad60SNick Piggin 207442e4089cSAndi Kleen if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) 207579f3aa5bSMatthew Wilcox return VM_FAULT_SIGBUS; 207642e4089cSAndi Kleen 2077423bad60SNick Piggin /* 2078423bad60SNick Piggin * If we don't have pte special, then we have to use the pfn_valid() 2079423bad60SNick Piggin * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* 2080423bad60SNick Piggin * refcount the page if pfn_valid is true (hence insert_page rather 208162eede62SHugh Dickins * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP 208262eede62SHugh Dickins * without pte special, it would there be refcounted as a normal page. 2083423bad60SNick Piggin */ 208400b3a331SLaurent Dufour if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && 208500b3a331SLaurent Dufour !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { 2086423bad60SNick Piggin struct page *page; 2087423bad60SNick Piggin 208803fc2da6SDan Williams /* 208903fc2da6SDan Williams * At this point we are committed to insert_page() 209003fc2da6SDan Williams * regardless of whether the caller specified flags that 209103fc2da6SDan Williams * result in pfn_t_has_page() == false. 209203fc2da6SDan Williams */ 209303fc2da6SDan Williams page = pfn_to_page(pfn_t_to_pfn(pfn)); 209479f3aa5bSMatthew Wilcox err = insert_page(vma, addr, page, pgprot); 209579f3aa5bSMatthew Wilcox } else { 20969b5a8e00SMatthew Wilcox return insert_pfn(vma, addr, pfn, pgprot, mkwrite); 2097423bad60SNick Piggin } 2098b2770da6SRoss Zwisler 20995d747637SMatthew Wilcox if (err == -ENOMEM) 21005d747637SMatthew Wilcox return VM_FAULT_OOM; 21015d747637SMatthew Wilcox if (err < 0 && err != -EBUSY) 21025d747637SMatthew Wilcox return VM_FAULT_SIGBUS; 21035d747637SMatthew Wilcox 21045d747637SMatthew Wilcox return VM_FAULT_NOPAGE; 2105423bad60SNick Piggin } 210679f3aa5bSMatthew Wilcox 2107574c5b3dSThomas Hellstrom /** 2108574c5b3dSThomas Hellstrom * vmf_insert_mixed_prot - insert single pfn into user vma with specified pgprot 2109574c5b3dSThomas Hellstrom * @vma: user vma to map to 2110574c5b3dSThomas Hellstrom * @addr: target user address of this page 2111574c5b3dSThomas Hellstrom * @pfn: source kernel pfn 2112574c5b3dSThomas Hellstrom * @pgprot: pgprot flags for the inserted page 2113574c5b3dSThomas Hellstrom * 2114a1a0aea5SRandy Dunlap * This is exactly like vmf_insert_mixed(), except that it allows drivers 2115574c5b3dSThomas Hellstrom * to override pgprot on a per-page basis. 2116574c5b3dSThomas Hellstrom * 2117574c5b3dSThomas Hellstrom * Typically this function should be used by drivers to set caching- and 2118574c5b3dSThomas Hellstrom * encryption bits different than those of @vma->vm_page_prot, because 2119574c5b3dSThomas Hellstrom * the caching- or encryption mode may not be known at mmap() time. 2120574c5b3dSThomas Hellstrom * This is ok as long as @vma->vm_page_prot is not used by the core vm 2121574c5b3dSThomas Hellstrom * to set caching and encryption bits for those vmas (except for COW pages). 2122574c5b3dSThomas Hellstrom * This is ensured by core vm only modifying these page table entries using 2123574c5b3dSThomas Hellstrom * functions that don't touch caching- or encryption bits, using pte_modify() 2124574c5b3dSThomas Hellstrom * if needed. (See for example mprotect()). 2125574c5b3dSThomas Hellstrom * Also when new page-table entries are created, this is only done using the 2126574c5b3dSThomas Hellstrom * fault() callback, and never using the value of vma->vm_page_prot, 2127574c5b3dSThomas Hellstrom * except for page-table entries that point to anonymous pages as the result 2128574c5b3dSThomas Hellstrom * of COW. 2129574c5b3dSThomas Hellstrom * 2130574c5b3dSThomas Hellstrom * Context: Process context. May allocate using %GFP_KERNEL. 2131574c5b3dSThomas Hellstrom * Return: vm_fault_t value. 2132574c5b3dSThomas Hellstrom */ 2133574c5b3dSThomas Hellstrom vm_fault_t vmf_insert_mixed_prot(struct vm_area_struct *vma, unsigned long addr, 2134574c5b3dSThomas Hellstrom pfn_t pfn, pgprot_t pgprot) 2135574c5b3dSThomas Hellstrom { 2136574c5b3dSThomas Hellstrom return __vm_insert_mixed(vma, addr, pfn, pgprot, false); 2137574c5b3dSThomas Hellstrom } 21385379e4ddSThomas Hellstrom EXPORT_SYMBOL(vmf_insert_mixed_prot); 2139574c5b3dSThomas Hellstrom 214079f3aa5bSMatthew Wilcox vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, 214179f3aa5bSMatthew Wilcox pfn_t pfn) 214279f3aa5bSMatthew Wilcox { 2143574c5b3dSThomas Hellstrom return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, false); 214479f3aa5bSMatthew Wilcox } 21455d747637SMatthew Wilcox EXPORT_SYMBOL(vmf_insert_mixed); 2146423bad60SNick Piggin 2147ab77dab4SSouptick Joarder /* 2148ab77dab4SSouptick Joarder * If the insertion of PTE failed because someone else already added a 2149ab77dab4SSouptick Joarder * different entry in the mean time, we treat that as success as we assume 2150ab77dab4SSouptick Joarder * the same entry was actually inserted. 2151ab77dab4SSouptick Joarder */ 2152ab77dab4SSouptick Joarder vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, 2153ab77dab4SSouptick Joarder unsigned long addr, pfn_t pfn) 2154b2770da6SRoss Zwisler { 2155574c5b3dSThomas Hellstrom return __vm_insert_mixed(vma, addr, pfn, vma->vm_page_prot, true); 2156b2770da6SRoss Zwisler } 2157ab77dab4SSouptick Joarder EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); 2158b2770da6SRoss Zwisler 2159a145dd41SLinus Torvalds /* 21601da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 21611da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 21621da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 21631da177e4SLinus Torvalds */ 21641da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 21651da177e4SLinus Torvalds unsigned long addr, unsigned long end, 21661da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 21671da177e4SLinus Torvalds { 21681da177e4SLinus Torvalds pte_t *pte; 2169c74df32cSHugh Dickins spinlock_t *ptl; 217042e4089cSAndi Kleen int err = 0; 21711da177e4SLinus Torvalds 2172c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 21731da177e4SLinus Torvalds if (!pte) 21741da177e4SLinus Torvalds return -ENOMEM; 21756606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 21761da177e4SLinus Torvalds do { 21771da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 217842e4089cSAndi Kleen if (!pfn_modify_allowed(pfn, prot)) { 217942e4089cSAndi Kleen err = -EACCES; 218042e4089cSAndi Kleen break; 218142e4089cSAndi Kleen } 21827e675137SNick Piggin set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); 21831da177e4SLinus Torvalds pfn++; 21841da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 21856606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 2186c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 218742e4089cSAndi Kleen return err; 21881da177e4SLinus Torvalds } 21891da177e4SLinus Torvalds 21901da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 21911da177e4SLinus Torvalds unsigned long addr, unsigned long end, 21921da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 21931da177e4SLinus Torvalds { 21941da177e4SLinus Torvalds pmd_t *pmd; 21951da177e4SLinus Torvalds unsigned long next; 219642e4089cSAndi Kleen int err; 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 21991da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 22001da177e4SLinus Torvalds if (!pmd) 22011da177e4SLinus Torvalds return -ENOMEM; 2202f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 22031da177e4SLinus Torvalds do { 22041da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 220542e4089cSAndi Kleen err = remap_pte_range(mm, pmd, addr, next, 220642e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 220742e4089cSAndi Kleen if (err) 220842e4089cSAndi Kleen return err; 22091da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 22101da177e4SLinus Torvalds return 0; 22111da177e4SLinus Torvalds } 22121da177e4SLinus Torvalds 2213c2febafcSKirill A. Shutemov static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, 22141da177e4SLinus Torvalds unsigned long addr, unsigned long end, 22151da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 22161da177e4SLinus Torvalds { 22171da177e4SLinus Torvalds pud_t *pud; 22181da177e4SLinus Torvalds unsigned long next; 221942e4089cSAndi Kleen int err; 22201da177e4SLinus Torvalds 22211da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 2222c2febafcSKirill A. Shutemov pud = pud_alloc(mm, p4d, addr); 22231da177e4SLinus Torvalds if (!pud) 22241da177e4SLinus Torvalds return -ENOMEM; 22251da177e4SLinus Torvalds do { 22261da177e4SLinus Torvalds next = pud_addr_end(addr, end); 222742e4089cSAndi Kleen err = remap_pmd_range(mm, pud, addr, next, 222842e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 222942e4089cSAndi Kleen if (err) 223042e4089cSAndi Kleen return err; 22311da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 22321da177e4SLinus Torvalds return 0; 22331da177e4SLinus Torvalds } 22341da177e4SLinus Torvalds 2235c2febafcSKirill A. Shutemov static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2236c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2237c2febafcSKirill A. Shutemov unsigned long pfn, pgprot_t prot) 2238c2febafcSKirill A. Shutemov { 2239c2febafcSKirill A. Shutemov p4d_t *p4d; 2240c2febafcSKirill A. Shutemov unsigned long next; 224142e4089cSAndi Kleen int err; 2242c2febafcSKirill A. Shutemov 2243c2febafcSKirill A. Shutemov pfn -= addr >> PAGE_SHIFT; 2244c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 2245c2febafcSKirill A. Shutemov if (!p4d) 2246c2febafcSKirill A. Shutemov return -ENOMEM; 2247c2febafcSKirill A. Shutemov do { 2248c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 224942e4089cSAndi Kleen err = remap_pud_range(mm, p4d, addr, next, 225042e4089cSAndi Kleen pfn + (addr >> PAGE_SHIFT), prot); 225142e4089cSAndi Kleen if (err) 225242e4089cSAndi Kleen return err; 2253c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2254c2febafcSKirill A. Shutemov return 0; 2255c2febafcSKirill A. Shutemov } 2256c2febafcSKirill A. Shutemov 2257bfa5bf6dSRolf Eike Beer /** 2258bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 2259bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 22600c4123e3SAlex Zhang * @addr: target page aligned user address to start at 226186a76331SWANG Wenhu * @pfn: page frame number of kernel physical memory address 2262552657b7Schenqiwu * @size: size of mapping area 2263bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 2264bfa5bf6dSRolf Eike Beer * 2265bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 2266a862f68aSMike Rapoport * 2267a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2268bfa5bf6dSRolf Eike Beer */ 22691da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 22701da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 22711da177e4SLinus Torvalds { 22721da177e4SLinus Torvalds pgd_t *pgd; 22731da177e4SLinus Torvalds unsigned long next; 22742d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 22751da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 2276d5957d2fSYongji Xie unsigned long remap_pfn = pfn; 22771da177e4SLinus Torvalds int err; 22781da177e4SLinus Torvalds 22790c4123e3SAlex Zhang if (WARN_ON_ONCE(!PAGE_ALIGNED(addr))) 22800c4123e3SAlex Zhang return -EINVAL; 22810c4123e3SAlex Zhang 22821da177e4SLinus Torvalds /* 22831da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 22841da177e4SLinus Torvalds * rest of the world about it: 22851da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 22861da177e4SLinus Torvalds * (accesses can have side effects). 22876aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 22886aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 22896aab341eSLinus Torvalds * with them. 2290314e51b9SKonstantin Khlebnikov * VM_DONTEXPAND 2291314e51b9SKonstantin Khlebnikov * Disable vma merging and expanding with mremap(). 2292314e51b9SKonstantin Khlebnikov * VM_DONTDUMP 2293314e51b9SKonstantin Khlebnikov * Omit vma from core dump, even when VM_IO turned off. 2294fb155c16SLinus Torvalds * 2295fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 2296fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 2297fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 2298b3b9c293SKonstantin Khlebnikov * See vm_normal_page() for details. 22991da177e4SLinus Torvalds */ 2300b3b9c293SKonstantin Khlebnikov if (is_cow_mapping(vma->vm_flags)) { 2301b3b9c293SKonstantin Khlebnikov if (addr != vma->vm_start || end != vma->vm_end) 2302b3b9c293SKonstantin Khlebnikov return -EINVAL; 23036aab341eSLinus Torvalds vma->vm_pgoff = pfn; 2304b3b9c293SKonstantin Khlebnikov } 2305b3b9c293SKonstantin Khlebnikov 2306d5957d2fSYongji Xie err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); 2307b3b9c293SKonstantin Khlebnikov if (err) 23083c8bb73aSvenkatesh.pallipadi@intel.com return -EINVAL; 2309fb155c16SLinus Torvalds 2310314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 23111da177e4SLinus Torvalds 23121da177e4SLinus Torvalds BUG_ON(addr >= end); 23131da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 23141da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 23151da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 23161da177e4SLinus Torvalds do { 23171da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2318c2febafcSKirill A. Shutemov err = remap_p4d_range(mm, pgd, addr, next, 23191da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 23201da177e4SLinus Torvalds if (err) 23211da177e4SLinus Torvalds break; 23221da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 23232ab64037Svenkatesh.pallipadi@intel.com 23242ab64037Svenkatesh.pallipadi@intel.com if (err) 2325d5957d2fSYongji Xie untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); 23262ab64037Svenkatesh.pallipadi@intel.com 23271da177e4SLinus Torvalds return err; 23281da177e4SLinus Torvalds } 23291da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 23301da177e4SLinus Torvalds 2331b4cbb197SLinus Torvalds /** 2332b4cbb197SLinus Torvalds * vm_iomap_memory - remap memory to userspace 2333b4cbb197SLinus Torvalds * @vma: user vma to map to 2334abd69b9eSWang Wenhu * @start: start of the physical memory to be mapped 2335b4cbb197SLinus Torvalds * @len: size of area 2336b4cbb197SLinus Torvalds * 2337b4cbb197SLinus Torvalds * This is a simplified io_remap_pfn_range() for common driver use. The 2338b4cbb197SLinus Torvalds * driver just needs to give us the physical memory range to be mapped, 2339b4cbb197SLinus Torvalds * we'll figure out the rest from the vma information. 2340b4cbb197SLinus Torvalds * 2341b4cbb197SLinus Torvalds * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get 2342b4cbb197SLinus Torvalds * whatever write-combining details or similar. 2343a862f68aSMike Rapoport * 2344a862f68aSMike Rapoport * Return: %0 on success, negative error code otherwise. 2345b4cbb197SLinus Torvalds */ 2346b4cbb197SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 2347b4cbb197SLinus Torvalds { 2348b4cbb197SLinus Torvalds unsigned long vm_len, pfn, pages; 2349b4cbb197SLinus Torvalds 2350b4cbb197SLinus Torvalds /* Check that the physical memory area passed in looks valid */ 2351b4cbb197SLinus Torvalds if (start + len < start) 2352b4cbb197SLinus Torvalds return -EINVAL; 2353b4cbb197SLinus Torvalds /* 2354b4cbb197SLinus Torvalds * You *really* shouldn't map things that aren't page-aligned, 2355b4cbb197SLinus Torvalds * but we've historically allowed it because IO memory might 2356b4cbb197SLinus Torvalds * just have smaller alignment. 2357b4cbb197SLinus Torvalds */ 2358b4cbb197SLinus Torvalds len += start & ~PAGE_MASK; 2359b4cbb197SLinus Torvalds pfn = start >> PAGE_SHIFT; 2360b4cbb197SLinus Torvalds pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; 2361b4cbb197SLinus Torvalds if (pfn + pages < pfn) 2362b4cbb197SLinus Torvalds return -EINVAL; 2363b4cbb197SLinus Torvalds 2364b4cbb197SLinus Torvalds /* We start the mapping 'vm_pgoff' pages into the area */ 2365b4cbb197SLinus Torvalds if (vma->vm_pgoff > pages) 2366b4cbb197SLinus Torvalds return -EINVAL; 2367b4cbb197SLinus Torvalds pfn += vma->vm_pgoff; 2368b4cbb197SLinus Torvalds pages -= vma->vm_pgoff; 2369b4cbb197SLinus Torvalds 2370b4cbb197SLinus Torvalds /* Can we fit all of the mapping? */ 2371b4cbb197SLinus Torvalds vm_len = vma->vm_end - vma->vm_start; 2372b4cbb197SLinus Torvalds if (vm_len >> PAGE_SHIFT > pages) 2373b4cbb197SLinus Torvalds return -EINVAL; 2374b4cbb197SLinus Torvalds 2375b4cbb197SLinus Torvalds /* Ok, let it rip */ 2376b4cbb197SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 2377b4cbb197SLinus Torvalds } 2378b4cbb197SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 2379b4cbb197SLinus Torvalds 2380aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 2381aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2382e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2383e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2384aee16b3cSJeremy Fitzhardinge { 2385aee16b3cSJeremy Fitzhardinge pte_t *pte; 2386be1db475SDaniel Axtens int err = 0; 23873f649ab7SKees Cook spinlock_t *ptl; 2388aee16b3cSJeremy Fitzhardinge 2389be1db475SDaniel Axtens if (create) { 2390aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 2391e80d3909SJoerg Roedel pte_alloc_kernel_track(pmd, addr, mask) : 2392aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 2393aee16b3cSJeremy Fitzhardinge if (!pte) 2394aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2395be1db475SDaniel Axtens } else { 2396be1db475SDaniel Axtens pte = (mm == &init_mm) ? 2397be1db475SDaniel Axtens pte_offset_kernel(pmd, addr) : 2398be1db475SDaniel Axtens pte_offset_map_lock(mm, pmd, addr, &ptl); 2399be1db475SDaniel Axtens } 2400aee16b3cSJeremy Fitzhardinge 2401aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 2402aee16b3cSJeremy Fitzhardinge 240338e0edb1SJeremy Fitzhardinge arch_enter_lazy_mmu_mode(); 240438e0edb1SJeremy Fitzhardinge 2405eeb4a05fSChristoph Hellwig if (fn) { 2406aee16b3cSJeremy Fitzhardinge do { 2407be1db475SDaniel Axtens if (create || !pte_none(*pte)) { 24088b1e0f81SAnshuman Khandual err = fn(pte++, addr, data); 2409aee16b3cSJeremy Fitzhardinge if (err) 2410aee16b3cSJeremy Fitzhardinge break; 2411be1db475SDaniel Axtens } 2412c36987e2SDaisuke Nishimura } while (addr += PAGE_SIZE, addr != end); 2413eeb4a05fSChristoph Hellwig } 2414e80d3909SJoerg Roedel *mask |= PGTBL_PTE_MODIFIED; 2415aee16b3cSJeremy Fitzhardinge 241638e0edb1SJeremy Fitzhardinge arch_leave_lazy_mmu_mode(); 241738e0edb1SJeremy Fitzhardinge 2418aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 2419aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 2420aee16b3cSJeremy Fitzhardinge return err; 2421aee16b3cSJeremy Fitzhardinge } 2422aee16b3cSJeremy Fitzhardinge 2423aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 2424aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2425e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2426e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2427aee16b3cSJeremy Fitzhardinge { 2428aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 2429aee16b3cSJeremy Fitzhardinge unsigned long next; 2430be1db475SDaniel Axtens int err = 0; 2431aee16b3cSJeremy Fitzhardinge 2432ceb86879SAndi Kleen BUG_ON(pud_huge(*pud)); 2433ceb86879SAndi Kleen 2434be1db475SDaniel Axtens if (create) { 2435e80d3909SJoerg Roedel pmd = pmd_alloc_track(mm, pud, addr, mask); 2436aee16b3cSJeremy Fitzhardinge if (!pmd) 2437aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2438be1db475SDaniel Axtens } else { 2439be1db475SDaniel Axtens pmd = pmd_offset(pud, addr); 2440be1db475SDaniel Axtens } 2441aee16b3cSJeremy Fitzhardinge do { 2442aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 2443be1db475SDaniel Axtens if (create || !pmd_none_or_clear_bad(pmd)) { 2444be1db475SDaniel Axtens err = apply_to_pte_range(mm, pmd, addr, next, fn, data, 2445e80d3909SJoerg Roedel create, mask); 2446aee16b3cSJeremy Fitzhardinge if (err) 2447aee16b3cSJeremy Fitzhardinge break; 2448be1db475SDaniel Axtens } 2449aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 2450aee16b3cSJeremy Fitzhardinge return err; 2451aee16b3cSJeremy Fitzhardinge } 2452aee16b3cSJeremy Fitzhardinge 2453c2febafcSKirill A. Shutemov static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, 2454aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 2455e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2456e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2457aee16b3cSJeremy Fitzhardinge { 2458aee16b3cSJeremy Fitzhardinge pud_t *pud; 2459aee16b3cSJeremy Fitzhardinge unsigned long next; 2460be1db475SDaniel Axtens int err = 0; 2461aee16b3cSJeremy Fitzhardinge 2462be1db475SDaniel Axtens if (create) { 2463e80d3909SJoerg Roedel pud = pud_alloc_track(mm, p4d, addr, mask); 2464aee16b3cSJeremy Fitzhardinge if (!pud) 2465aee16b3cSJeremy Fitzhardinge return -ENOMEM; 2466be1db475SDaniel Axtens } else { 2467be1db475SDaniel Axtens pud = pud_offset(p4d, addr); 2468be1db475SDaniel Axtens } 2469aee16b3cSJeremy Fitzhardinge do { 2470aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 2471be1db475SDaniel Axtens if (create || !pud_none_or_clear_bad(pud)) { 2472be1db475SDaniel Axtens err = apply_to_pmd_range(mm, pud, addr, next, fn, data, 2473e80d3909SJoerg Roedel create, mask); 2474aee16b3cSJeremy Fitzhardinge if (err) 2475aee16b3cSJeremy Fitzhardinge break; 2476be1db475SDaniel Axtens } 2477aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 2478aee16b3cSJeremy Fitzhardinge return err; 2479aee16b3cSJeremy Fitzhardinge } 2480aee16b3cSJeremy Fitzhardinge 2481c2febafcSKirill A. Shutemov static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, 2482c2febafcSKirill A. Shutemov unsigned long addr, unsigned long end, 2483e80d3909SJoerg Roedel pte_fn_t fn, void *data, bool create, 2484e80d3909SJoerg Roedel pgtbl_mod_mask *mask) 2485c2febafcSKirill A. Shutemov { 2486c2febafcSKirill A. Shutemov p4d_t *p4d; 2487c2febafcSKirill A. Shutemov unsigned long next; 2488be1db475SDaniel Axtens int err = 0; 2489c2febafcSKirill A. Shutemov 2490be1db475SDaniel Axtens if (create) { 2491e80d3909SJoerg Roedel p4d = p4d_alloc_track(mm, pgd, addr, mask); 2492c2febafcSKirill A. Shutemov if (!p4d) 2493c2febafcSKirill A. Shutemov return -ENOMEM; 2494be1db475SDaniel Axtens } else { 2495be1db475SDaniel Axtens p4d = p4d_offset(pgd, addr); 2496be1db475SDaniel Axtens } 2497c2febafcSKirill A. Shutemov do { 2498c2febafcSKirill A. Shutemov next = p4d_addr_end(addr, end); 2499be1db475SDaniel Axtens if (create || !p4d_none_or_clear_bad(p4d)) { 2500be1db475SDaniel Axtens err = apply_to_pud_range(mm, p4d, addr, next, fn, data, 2501e80d3909SJoerg Roedel create, mask); 2502c2febafcSKirill A. Shutemov if (err) 2503c2febafcSKirill A. Shutemov break; 2504be1db475SDaniel Axtens } 2505c2febafcSKirill A. Shutemov } while (p4d++, addr = next, addr != end); 2506c2febafcSKirill A. Shutemov return err; 2507c2febafcSKirill A. Shutemov } 2508c2febafcSKirill A. Shutemov 2509be1db475SDaniel Axtens static int __apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2510be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, 2511be1db475SDaniel Axtens void *data, bool create) 2512be1db475SDaniel Axtens { 2513be1db475SDaniel Axtens pgd_t *pgd; 2514e80d3909SJoerg Roedel unsigned long start = addr, next; 2515be1db475SDaniel Axtens unsigned long end = addr + size; 2516e80d3909SJoerg Roedel pgtbl_mod_mask mask = 0; 2517be1db475SDaniel Axtens int err = 0; 2518be1db475SDaniel Axtens 2519be1db475SDaniel Axtens if (WARN_ON(addr >= end)) 2520be1db475SDaniel Axtens return -EINVAL; 2521be1db475SDaniel Axtens 2522be1db475SDaniel Axtens pgd = pgd_offset(mm, addr); 2523be1db475SDaniel Axtens do { 2524be1db475SDaniel Axtens next = pgd_addr_end(addr, end); 2525be1db475SDaniel Axtens if (!create && pgd_none_or_clear_bad(pgd)) 2526be1db475SDaniel Axtens continue; 2527e80d3909SJoerg Roedel err = apply_to_p4d_range(mm, pgd, addr, next, fn, data, create, &mask); 2528be1db475SDaniel Axtens if (err) 2529be1db475SDaniel Axtens break; 2530be1db475SDaniel Axtens } while (pgd++, addr = next, addr != end); 2531be1db475SDaniel Axtens 2532e80d3909SJoerg Roedel if (mask & ARCH_PAGE_TABLE_SYNC_MASK) 2533e80d3909SJoerg Roedel arch_sync_kernel_mappings(start, start + size); 2534e80d3909SJoerg Roedel 2535be1db475SDaniel Axtens return err; 2536be1db475SDaniel Axtens } 2537be1db475SDaniel Axtens 2538aee16b3cSJeremy Fitzhardinge /* 2539aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 2540aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 2541aee16b3cSJeremy Fitzhardinge */ 2542aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 2543aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 2544aee16b3cSJeremy Fitzhardinge { 2545be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, true); 2546aee16b3cSJeremy Fitzhardinge } 2547aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 2548aee16b3cSJeremy Fitzhardinge 25491da177e4SLinus Torvalds /* 2550be1db475SDaniel Axtens * Scan a region of virtual memory, calling a provided function on 2551be1db475SDaniel Axtens * each leaf page table where it exists. 2552be1db475SDaniel Axtens * 2553be1db475SDaniel Axtens * Unlike apply_to_page_range, this does _not_ fill in page tables 2554be1db475SDaniel Axtens * where they are absent. 2555be1db475SDaniel Axtens */ 2556be1db475SDaniel Axtens int apply_to_existing_page_range(struct mm_struct *mm, unsigned long addr, 2557be1db475SDaniel Axtens unsigned long size, pte_fn_t fn, void *data) 2558be1db475SDaniel Axtens { 2559be1db475SDaniel Axtens return __apply_to_page_range(mm, addr, size, fn, data, false); 2560be1db475SDaniel Axtens } 2561be1db475SDaniel Axtens EXPORT_SYMBOL_GPL(apply_to_existing_page_range); 2562be1db475SDaniel Axtens 2563be1db475SDaniel Axtens /* 25649b4bdd2fSKirill A. Shutemov * handle_pte_fault chooses page fault handler according to an entry which was 25659b4bdd2fSKirill A. Shutemov * read non-atomically. Before making any commitment, on those architectures 25669b4bdd2fSKirill A. Shutemov * or configurations (e.g. i386 with PAE) which might give a mix of unmatched 25679b4bdd2fSKirill A. Shutemov * parts, do_swap_page must check under lock before unmapping the pte and 25689b4bdd2fSKirill A. Shutemov * proceeding (but do_wp_page is only called after already making such a check; 2569a335b2e1SRyota Ozaki * and do_anonymous_page can safely check later on). 25708f4e2101SHugh Dickins */ 25714c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 25728f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 25738f4e2101SHugh Dickins { 25748f4e2101SHugh Dickins int same = 1; 2575923717cbSThomas Gleixner #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPTION) 25768f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 25774c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 25784c21e2f2SHugh Dickins spin_lock(ptl); 25798f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 25804c21e2f2SHugh Dickins spin_unlock(ptl); 25818f4e2101SHugh Dickins } 25828f4e2101SHugh Dickins #endif 25838f4e2101SHugh Dickins pte_unmap(page_table); 25848f4e2101SHugh Dickins return same; 25858f4e2101SHugh Dickins } 25868f4e2101SHugh Dickins 258783d116c5SJia He static inline bool cow_user_page(struct page *dst, struct page *src, 258883d116c5SJia He struct vm_fault *vmf) 25896aab341eSLinus Torvalds { 259083d116c5SJia He bool ret; 259183d116c5SJia He void *kaddr; 259283d116c5SJia He void __user *uaddr; 2593c3e5ea6eSKirill A. Shutemov bool locked = false; 259483d116c5SJia He struct vm_area_struct *vma = vmf->vma; 259583d116c5SJia He struct mm_struct *mm = vma->vm_mm; 259683d116c5SJia He unsigned long addr = vmf->address; 259783d116c5SJia He 259883d116c5SJia He if (likely(src)) { 259983d116c5SJia He copy_user_highpage(dst, src, addr, vma); 260083d116c5SJia He return true; 260183d116c5SJia He } 260283d116c5SJia He 26036aab341eSLinus Torvalds /* 26046aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 26056aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 26066aab341eSLinus Torvalds * just copying from the original user address. If that 26076aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 26086aab341eSLinus Torvalds */ 260983d116c5SJia He kaddr = kmap_atomic(dst); 261083d116c5SJia He uaddr = (void __user *)(addr & PAGE_MASK); 261183d116c5SJia He 261283d116c5SJia He /* 261383d116c5SJia He * On architectures with software "accessed" bits, we would 261483d116c5SJia He * take a double page fault, so mark it accessed here. 261583d116c5SJia He */ 2616c3e5ea6eSKirill A. Shutemov if (arch_faults_on_old_pte() && !pte_young(vmf->orig_pte)) { 261783d116c5SJia He pte_t entry; 261883d116c5SJia He 261983d116c5SJia He vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2620c3e5ea6eSKirill A. Shutemov locked = true; 262183d116c5SJia He if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { 262283d116c5SJia He /* 262383d116c5SJia He * Other thread has already handled the fault 26247df67697SBibo Mao * and update local tlb only 262583d116c5SJia He */ 26267df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 262783d116c5SJia He ret = false; 262883d116c5SJia He goto pte_unlock; 262983d116c5SJia He } 263083d116c5SJia He 263183d116c5SJia He entry = pte_mkyoung(vmf->orig_pte); 263283d116c5SJia He if (ptep_set_access_flags(vma, addr, vmf->pte, entry, 0)) 263383d116c5SJia He update_mmu_cache(vma, addr, vmf->pte); 263483d116c5SJia He } 26355d2a2dbbSLinus Torvalds 26365d2a2dbbSLinus Torvalds /* 26375d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 26385d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 26395d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 26405d2a2dbbSLinus Torvalds * zeroes. 26415d2a2dbbSLinus Torvalds */ 264283d116c5SJia He if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 2643c3e5ea6eSKirill A. Shutemov if (locked) 2644c3e5ea6eSKirill A. Shutemov goto warn; 2645c3e5ea6eSKirill A. Shutemov 2646c3e5ea6eSKirill A. Shutemov /* Re-validate under PTL if the page is still mapped */ 2647c3e5ea6eSKirill A. Shutemov vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl); 2648c3e5ea6eSKirill A. Shutemov locked = true; 2649c3e5ea6eSKirill A. Shutemov if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) { 26507df67697SBibo Mao /* The PTE changed under us, update local tlb */ 26517df67697SBibo Mao update_mmu_tlb(vma, addr, vmf->pte); 2652c3e5ea6eSKirill A. Shutemov ret = false; 2653c3e5ea6eSKirill A. Shutemov goto pte_unlock; 2654c3e5ea6eSKirill A. Shutemov } 2655c3e5ea6eSKirill A. Shutemov 2656c3e5ea6eSKirill A. Shutemov /* 2657985ba004SEthon Paul * The same page can be mapped back since last copy attempt. 2658c3e5ea6eSKirill A. Shutemov * Try to copy again under PTL. 2659c3e5ea6eSKirill A. Shutemov */ 2660c3e5ea6eSKirill A. Shutemov if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) { 266183d116c5SJia He /* 266283d116c5SJia He * Give a warn in case there can be some obscure 266383d116c5SJia He * use-case 266483d116c5SJia He */ 2665c3e5ea6eSKirill A. Shutemov warn: 266683d116c5SJia He WARN_ON_ONCE(1); 26673ecb01dfSJan Beulich clear_page(kaddr); 266883d116c5SJia He } 2669c3e5ea6eSKirill A. Shutemov } 267083d116c5SJia He 267183d116c5SJia He ret = true; 267283d116c5SJia He 267383d116c5SJia He pte_unlock: 2674c3e5ea6eSKirill A. Shutemov if (locked) 267583d116c5SJia He pte_unmap_unlock(vmf->pte, vmf->ptl); 26769b04c5feSCong Wang kunmap_atomic(kaddr); 2677c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 267883d116c5SJia He 267983d116c5SJia He return ret; 26806aab341eSLinus Torvalds } 26816aab341eSLinus Torvalds 2682c20cd45eSMichal Hocko static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) 2683c20cd45eSMichal Hocko { 2684c20cd45eSMichal Hocko struct file *vm_file = vma->vm_file; 2685c20cd45eSMichal Hocko 2686c20cd45eSMichal Hocko if (vm_file) 2687c20cd45eSMichal Hocko return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; 2688c20cd45eSMichal Hocko 2689c20cd45eSMichal Hocko /* 2690c20cd45eSMichal Hocko * Special mappings (e.g. VDSO) do not have any file so fake 2691c20cd45eSMichal Hocko * a default GFP_KERNEL for them. 2692c20cd45eSMichal Hocko */ 2693c20cd45eSMichal Hocko return GFP_KERNEL; 2694c20cd45eSMichal Hocko } 2695c20cd45eSMichal Hocko 26961da177e4SLinus Torvalds /* 2697fb09a464SKirill A. Shutemov * Notify the address space that the page is about to become writable so that 2698fb09a464SKirill A. Shutemov * it can prohibit this or wait for the page to get into an appropriate state. 2699fb09a464SKirill A. Shutemov * 2700fb09a464SKirill A. Shutemov * We do this without the lock held, so that it can sleep if it needs to. 2701fb09a464SKirill A. Shutemov */ 27022b740303SSouptick Joarder static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) 2703fb09a464SKirill A. Shutemov { 27042b740303SSouptick Joarder vm_fault_t ret; 270538b8cb7fSJan Kara struct page *page = vmf->page; 270638b8cb7fSJan Kara unsigned int old_flags = vmf->flags; 2707fb09a464SKirill A. Shutemov 270838b8cb7fSJan Kara vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; 2709fb09a464SKirill A. Shutemov 2710dc617f29SDarrick J. Wong if (vmf->vma->vm_file && 2711dc617f29SDarrick J. Wong IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host)) 2712dc617f29SDarrick J. Wong return VM_FAULT_SIGBUS; 2713dc617f29SDarrick J. Wong 271411bac800SDave Jiang ret = vmf->vma->vm_ops->page_mkwrite(vmf); 271538b8cb7fSJan Kara /* Restore original flags so that caller is not surprised */ 271638b8cb7fSJan Kara vmf->flags = old_flags; 2717fb09a464SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2718fb09a464SKirill A. Shutemov return ret; 2719fb09a464SKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) { 2720fb09a464SKirill A. Shutemov lock_page(page); 2721fb09a464SKirill A. Shutemov if (!page->mapping) { 2722fb09a464SKirill A. Shutemov unlock_page(page); 2723fb09a464SKirill A. Shutemov return 0; /* retry */ 2724fb09a464SKirill A. Shutemov } 2725fb09a464SKirill A. Shutemov ret |= VM_FAULT_LOCKED; 2726fb09a464SKirill A. Shutemov } else 2727fb09a464SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 2728fb09a464SKirill A. Shutemov return ret; 2729fb09a464SKirill A. Shutemov } 2730fb09a464SKirill A. Shutemov 2731fb09a464SKirill A. Shutemov /* 273297ba0c2bSJan Kara * Handle dirtying of a page in shared file mapping on a write fault. 27334e047f89SShachar Raindel * 273497ba0c2bSJan Kara * The function expects the page to be locked and unlocks it. 27354e047f89SShachar Raindel */ 273689b15332SJohannes Weiner static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf) 27374e047f89SShachar Raindel { 273889b15332SJohannes Weiner struct vm_area_struct *vma = vmf->vma; 27394e047f89SShachar Raindel struct address_space *mapping; 274089b15332SJohannes Weiner struct page *page = vmf->page; 274197ba0c2bSJan Kara bool dirtied; 274297ba0c2bSJan Kara bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; 27434e047f89SShachar Raindel 27444e047f89SShachar Raindel dirtied = set_page_dirty(page); 27454e047f89SShachar Raindel VM_BUG_ON_PAGE(PageAnon(page), page); 274697ba0c2bSJan Kara /* 274797ba0c2bSJan Kara * Take a local copy of the address_space - page.mapping may be zeroed 274897ba0c2bSJan Kara * by truncate after unlock_page(). The address_space itself remains 274997ba0c2bSJan Kara * pinned by vma->vm_file's reference. We rely on unlock_page()'s 275097ba0c2bSJan Kara * release semantics to prevent the compiler from undoing this copying. 275197ba0c2bSJan Kara */ 275297ba0c2bSJan Kara mapping = page_rmapping(page); 27534e047f89SShachar Raindel unlock_page(page); 27544e047f89SShachar Raindel 27554e047f89SShachar Raindel if (!page_mkwrite) 27564e047f89SShachar Raindel file_update_time(vma->vm_file); 275789b15332SJohannes Weiner 275889b15332SJohannes Weiner /* 275989b15332SJohannes Weiner * Throttle page dirtying rate down to writeback speed. 276089b15332SJohannes Weiner * 276189b15332SJohannes Weiner * mapping may be NULL here because some device drivers do not 276289b15332SJohannes Weiner * set page.mapping but still dirty their pages 276389b15332SJohannes Weiner * 2764c1e8d7c6SMichel Lespinasse * Drop the mmap_lock before waiting on IO, if we can. The file 276589b15332SJohannes Weiner * is pinning the mapping, as per above. 276689b15332SJohannes Weiner */ 276789b15332SJohannes Weiner if ((dirtied || page_mkwrite) && mapping) { 276889b15332SJohannes Weiner struct file *fpin; 276989b15332SJohannes Weiner 277089b15332SJohannes Weiner fpin = maybe_unlock_mmap_for_io(vmf, NULL); 277189b15332SJohannes Weiner balance_dirty_pages_ratelimited(mapping); 277289b15332SJohannes Weiner if (fpin) { 277389b15332SJohannes Weiner fput(fpin); 277489b15332SJohannes Weiner return VM_FAULT_RETRY; 277589b15332SJohannes Weiner } 277689b15332SJohannes Weiner } 277789b15332SJohannes Weiner 277889b15332SJohannes Weiner return 0; 27794e047f89SShachar Raindel } 27804e047f89SShachar Raindel 278197ba0c2bSJan Kara /* 27824e047f89SShachar Raindel * Handle write page faults for pages that can be reused in the current vma 27834e047f89SShachar Raindel * 27844e047f89SShachar Raindel * This can happen either due to the mapping being with the VM_SHARED flag, 27854e047f89SShachar Raindel * or due to us being the last reference standing to the page. In either 27864e047f89SShachar Raindel * case, all we need to do here is to mark the page as writable and update 27874e047f89SShachar Raindel * any related book-keeping. 27884e047f89SShachar Raindel */ 2789997dd98dSJan Kara static inline void wp_page_reuse(struct vm_fault *vmf) 279082b0f8c3SJan Kara __releases(vmf->ptl) 27914e047f89SShachar Raindel { 279282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 2793a41b70d6SJan Kara struct page *page = vmf->page; 27944e047f89SShachar Raindel pte_t entry; 27954e047f89SShachar Raindel /* 27964e047f89SShachar Raindel * Clear the pages cpupid information as the existing 27974e047f89SShachar Raindel * information potentially belongs to a now completely 27984e047f89SShachar Raindel * unrelated process. 27994e047f89SShachar Raindel */ 28004e047f89SShachar Raindel if (page) 28014e047f89SShachar Raindel page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); 28024e047f89SShachar Raindel 28032994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 28042994302bSJan Kara entry = pte_mkyoung(vmf->orig_pte); 28054e047f89SShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 280682b0f8c3SJan Kara if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) 280782b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 280882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 2809798a6b87SPeter Xu count_vm_event(PGREUSE); 28104e047f89SShachar Raindel } 28114e047f89SShachar Raindel 28124e047f89SShachar Raindel /* 28132f38ab2cSShachar Raindel * Handle the case of a page which we actually need to copy to a new page. 28142f38ab2cSShachar Raindel * 2815c1e8d7c6SMichel Lespinasse * Called with mmap_lock locked and the old page referenced, but 28162f38ab2cSShachar Raindel * without the ptl held. 28172f38ab2cSShachar Raindel * 28182f38ab2cSShachar Raindel * High level logic flow: 28192f38ab2cSShachar Raindel * 28202f38ab2cSShachar Raindel * - Allocate a page, copy the content of the old page to the new one. 28212f38ab2cSShachar Raindel * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. 28222f38ab2cSShachar Raindel * - Take the PTL. If the pte changed, bail out and release the allocated page 28232f38ab2cSShachar Raindel * - If the pte is still the way we remember it, update the page table and all 28242f38ab2cSShachar Raindel * relevant references. This includes dropping the reference the page-table 28252f38ab2cSShachar Raindel * held to the old page, as well as updating the rmap. 28262f38ab2cSShachar Raindel * - In any case, unlock the PTL and drop the reference we took to the old page. 28272f38ab2cSShachar Raindel */ 28282b740303SSouptick Joarder static vm_fault_t wp_page_copy(struct vm_fault *vmf) 28292f38ab2cSShachar Raindel { 283082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 2831bae473a4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 2832a41b70d6SJan Kara struct page *old_page = vmf->page; 28332f38ab2cSShachar Raindel struct page *new_page = NULL; 28342f38ab2cSShachar Raindel pte_t entry; 28352f38ab2cSShachar Raindel int page_copied = 0; 2836ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 28372f38ab2cSShachar Raindel 28382f38ab2cSShachar Raindel if (unlikely(anon_vma_prepare(vma))) 28392f38ab2cSShachar Raindel goto oom; 28402f38ab2cSShachar Raindel 28412994302bSJan Kara if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { 284282b0f8c3SJan Kara new_page = alloc_zeroed_user_highpage_movable(vma, 284382b0f8c3SJan Kara vmf->address); 28442f38ab2cSShachar Raindel if (!new_page) 28452f38ab2cSShachar Raindel goto oom; 28462f38ab2cSShachar Raindel } else { 2847bae473a4SKirill A. Shutemov new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, 284882b0f8c3SJan Kara vmf->address); 28492f38ab2cSShachar Raindel if (!new_page) 28502f38ab2cSShachar Raindel goto oom; 285183d116c5SJia He 285283d116c5SJia He if (!cow_user_page(new_page, old_page, vmf)) { 285383d116c5SJia He /* 285483d116c5SJia He * COW failed, if the fault was solved by other, 285583d116c5SJia He * it's fine. If not, userspace would re-fault on 285683d116c5SJia He * the same address and we will handle the fault 285783d116c5SJia He * from the second attempt. 285883d116c5SJia He */ 285983d116c5SJia He put_page(new_page); 286083d116c5SJia He if (old_page) 286183d116c5SJia He put_page(old_page); 286283d116c5SJia He return 0; 286383d116c5SJia He } 28642f38ab2cSShachar Raindel } 28652f38ab2cSShachar Raindel 2866d9eb1ea2SJohannes Weiner if (mem_cgroup_charge(new_page, mm, GFP_KERNEL)) 28672f38ab2cSShachar Raindel goto oom_free_new; 28689d82c694SJohannes Weiner cgroup_throttle_swaprate(new_page, GFP_KERNEL); 28692f38ab2cSShachar Raindel 2870eb3c24f3SMel Gorman __SetPageUptodate(new_page); 2871eb3c24f3SMel Gorman 28727269f999SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, 28736f4f13e8SJérôme Glisse vmf->address & PAGE_MASK, 2874ac46d4f3SJérôme Glisse (vmf->address & PAGE_MASK) + PAGE_SIZE); 2875ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 28762f38ab2cSShachar Raindel 28772f38ab2cSShachar Raindel /* 28782f38ab2cSShachar Raindel * Re-check the pte - we dropped the lock 28792f38ab2cSShachar Raindel */ 288082b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); 28812994302bSJan Kara if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { 28822f38ab2cSShachar Raindel if (old_page) { 28832f38ab2cSShachar Raindel if (!PageAnon(old_page)) { 2884eca56ff9SJerome Marchand dec_mm_counter_fast(mm, 2885eca56ff9SJerome Marchand mm_counter_file(old_page)); 28862f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 28872f38ab2cSShachar Raindel } 28882f38ab2cSShachar Raindel } else { 28892f38ab2cSShachar Raindel inc_mm_counter_fast(mm, MM_ANONPAGES); 28902f38ab2cSShachar Raindel } 28912994302bSJan Kara flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); 28922f38ab2cSShachar Raindel entry = mk_pte(new_page, vma->vm_page_prot); 289344bf431bSBibo Mao entry = pte_sw_mkyoung(entry); 28942f38ab2cSShachar Raindel entry = maybe_mkwrite(pte_mkdirty(entry), vma); 28952f38ab2cSShachar Raindel /* 28962f38ab2cSShachar Raindel * Clear the pte entry and flush it first, before updating the 28972f38ab2cSShachar Raindel * pte with the new entry. This will avoid a race condition 28982f38ab2cSShachar Raindel * seen in the presence of one thread doing SMC and another 28992f38ab2cSShachar Raindel * thread doing COW. 29002f38ab2cSShachar Raindel */ 290182b0f8c3SJan Kara ptep_clear_flush_notify(vma, vmf->address, vmf->pte); 290282b0f8c3SJan Kara page_add_new_anon_rmap(new_page, vma, vmf->address, false); 2903b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(new_page, vma); 29042f38ab2cSShachar Raindel /* 29052f38ab2cSShachar Raindel * We call the notify macro here because, when using secondary 29062f38ab2cSShachar Raindel * mmu page tables (such as kvm shadow page tables), we want the 29072f38ab2cSShachar Raindel * new page to be mapped directly into the secondary page table. 29082f38ab2cSShachar Raindel */ 290982b0f8c3SJan Kara set_pte_at_notify(mm, vmf->address, vmf->pte, entry); 291082b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 29112f38ab2cSShachar Raindel if (old_page) { 29122f38ab2cSShachar Raindel /* 29132f38ab2cSShachar Raindel * Only after switching the pte to the new page may 29142f38ab2cSShachar Raindel * we remove the mapcount here. Otherwise another 29152f38ab2cSShachar Raindel * process may come and find the rmap count decremented 29162f38ab2cSShachar Raindel * before the pte is switched to the new page, and 29172f38ab2cSShachar Raindel * "reuse" the old page writing into it while our pte 29182f38ab2cSShachar Raindel * here still points into it and can be read by other 29192f38ab2cSShachar Raindel * threads. 29202f38ab2cSShachar Raindel * 29212f38ab2cSShachar Raindel * The critical issue is to order this 29222f38ab2cSShachar Raindel * page_remove_rmap with the ptp_clear_flush above. 29232f38ab2cSShachar Raindel * Those stores are ordered by (if nothing else,) 29242f38ab2cSShachar Raindel * the barrier present in the atomic_add_negative 29252f38ab2cSShachar Raindel * in page_remove_rmap. 29262f38ab2cSShachar Raindel * 29272f38ab2cSShachar Raindel * Then the TLB flush in ptep_clear_flush ensures that 29282f38ab2cSShachar Raindel * no process can access the old page before the 29292f38ab2cSShachar Raindel * decremented mapcount is visible. And the old page 29302f38ab2cSShachar Raindel * cannot be reused until after the decremented 29312f38ab2cSShachar Raindel * mapcount is visible. So transitively, TLBs to 29322f38ab2cSShachar Raindel * old page will be flushed before it can be reused. 29332f38ab2cSShachar Raindel */ 2934d281ee61SKirill A. Shutemov page_remove_rmap(old_page, false); 29352f38ab2cSShachar Raindel } 29362f38ab2cSShachar Raindel 29372f38ab2cSShachar Raindel /* Free the old page.. */ 29382f38ab2cSShachar Raindel new_page = old_page; 29392f38ab2cSShachar Raindel page_copied = 1; 29402f38ab2cSShachar Raindel } else { 29417df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 29422f38ab2cSShachar Raindel } 29432f38ab2cSShachar Raindel 29442f38ab2cSShachar Raindel if (new_page) 294509cbfeafSKirill A. Shutemov put_page(new_page); 29462f38ab2cSShachar Raindel 294782b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 29484645b9feSJérôme Glisse /* 29494645b9feSJérôme Glisse * No need to double call mmu_notifier->invalidate_range() callback as 29504645b9feSJérôme Glisse * the above ptep_clear_flush_notify() did already call it. 29514645b9feSJérôme Glisse */ 2952ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_only_end(&range); 29532f38ab2cSShachar Raindel if (old_page) { 29542f38ab2cSShachar Raindel /* 29552f38ab2cSShachar Raindel * Don't let another task, with possibly unlocked vma, 29562f38ab2cSShachar Raindel * keep the mlocked page. 29572f38ab2cSShachar Raindel */ 29582f38ab2cSShachar Raindel if (page_copied && (vma->vm_flags & VM_LOCKED)) { 29592f38ab2cSShachar Raindel lock_page(old_page); /* LRU manipulation */ 2960e90309c9SKirill A. Shutemov if (PageMlocked(old_page)) 29612f38ab2cSShachar Raindel munlock_vma_page(old_page); 29622f38ab2cSShachar Raindel unlock_page(old_page); 29632f38ab2cSShachar Raindel } 296409cbfeafSKirill A. Shutemov put_page(old_page); 29652f38ab2cSShachar Raindel } 29662f38ab2cSShachar Raindel return page_copied ? VM_FAULT_WRITE : 0; 29672f38ab2cSShachar Raindel oom_free_new: 296809cbfeafSKirill A. Shutemov put_page(new_page); 29692f38ab2cSShachar Raindel oom: 29702f38ab2cSShachar Raindel if (old_page) 297109cbfeafSKirill A. Shutemov put_page(old_page); 29722f38ab2cSShachar Raindel return VM_FAULT_OOM; 29732f38ab2cSShachar Raindel } 29742f38ab2cSShachar Raindel 297566a6197cSJan Kara /** 297666a6197cSJan Kara * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE 297766a6197cSJan Kara * writeable once the page is prepared 297866a6197cSJan Kara * 297966a6197cSJan Kara * @vmf: structure describing the fault 298066a6197cSJan Kara * 298166a6197cSJan Kara * This function handles all that is needed to finish a write page fault in a 298266a6197cSJan Kara * shared mapping due to PTE being read-only once the mapped page is prepared. 2983a862f68aSMike Rapoport * It handles locking of PTE and modifying it. 298466a6197cSJan Kara * 298566a6197cSJan Kara * The function expects the page to be locked or other protection against 298666a6197cSJan Kara * concurrent faults / writeback (such as DAX radix tree locks). 2987a862f68aSMike Rapoport * 2988a862f68aSMike Rapoport * Return: %VM_FAULT_WRITE on success, %0 when PTE got changed before 2989a862f68aSMike Rapoport * we acquired PTE lock. 299066a6197cSJan Kara */ 29912b740303SSouptick Joarder vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) 299266a6197cSJan Kara { 299366a6197cSJan Kara WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); 299466a6197cSJan Kara vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, 299566a6197cSJan Kara &vmf->ptl); 299666a6197cSJan Kara /* 299766a6197cSJan Kara * We might have raced with another page fault while we released the 299866a6197cSJan Kara * pte_offset_map_lock. 299966a6197cSJan Kara */ 300066a6197cSJan Kara if (!pte_same(*vmf->pte, vmf->orig_pte)) { 30017df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 300266a6197cSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3003a19e2553SJan Kara return VM_FAULT_NOPAGE; 300466a6197cSJan Kara } 300566a6197cSJan Kara wp_page_reuse(vmf); 3006a19e2553SJan Kara return 0; 300766a6197cSJan Kara } 300866a6197cSJan Kara 3009dd906184SBoaz Harrosh /* 3010dd906184SBoaz Harrosh * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED 3011dd906184SBoaz Harrosh * mapping 3012dd906184SBoaz Harrosh */ 30132b740303SSouptick Joarder static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) 3014dd906184SBoaz Harrosh { 301582b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3016bae473a4SKirill A. Shutemov 3017dd906184SBoaz Harrosh if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { 30182b740303SSouptick Joarder vm_fault_t ret; 3019dd906184SBoaz Harrosh 302082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3021fe82221fSJan Kara vmf->flags |= FAULT_FLAG_MKWRITE; 302211bac800SDave Jiang ret = vma->vm_ops->pfn_mkwrite(vmf); 30232f89dc12SJan Kara if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) 3024dd906184SBoaz Harrosh return ret; 302566a6197cSJan Kara return finish_mkwrite_fault(vmf); 3026dd906184SBoaz Harrosh } 3027997dd98dSJan Kara wp_page_reuse(vmf); 3028997dd98dSJan Kara return VM_FAULT_WRITE; 3029dd906184SBoaz Harrosh } 3030dd906184SBoaz Harrosh 30312b740303SSouptick Joarder static vm_fault_t wp_page_shared(struct vm_fault *vmf) 303282b0f8c3SJan Kara __releases(vmf->ptl) 303393e478d4SShachar Raindel { 303482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 303589b15332SJohannes Weiner vm_fault_t ret = VM_FAULT_WRITE; 303693e478d4SShachar Raindel 3037a41b70d6SJan Kara get_page(vmf->page); 303893e478d4SShachar Raindel 303993e478d4SShachar Raindel if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 30402b740303SSouptick Joarder vm_fault_t tmp; 304193e478d4SShachar Raindel 304282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 304338b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 304493e478d4SShachar Raindel if (unlikely(!tmp || (tmp & 304593e478d4SShachar Raindel (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 3046a41b70d6SJan Kara put_page(vmf->page); 304793e478d4SShachar Raindel return tmp; 304893e478d4SShachar Raindel } 304966a6197cSJan Kara tmp = finish_mkwrite_fault(vmf); 3050a19e2553SJan Kara if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { 3051a41b70d6SJan Kara unlock_page(vmf->page); 3052a41b70d6SJan Kara put_page(vmf->page); 305366a6197cSJan Kara return tmp; 305493e478d4SShachar Raindel } 305566a6197cSJan Kara } else { 3056997dd98dSJan Kara wp_page_reuse(vmf); 3057997dd98dSJan Kara lock_page(vmf->page); 305893e478d4SShachar Raindel } 305989b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 3060997dd98dSJan Kara put_page(vmf->page); 306193e478d4SShachar Raindel 306289b15332SJohannes Weiner return ret; 306393e478d4SShachar Raindel } 306493e478d4SShachar Raindel 30652f38ab2cSShachar Raindel /* 30661da177e4SLinus Torvalds * This routine handles present pages, when users try to write 30671da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 30681da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 30691da177e4SLinus Torvalds * 30701da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 30711da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 30721da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 30731da177e4SLinus Torvalds * COW. 30741da177e4SLinus Torvalds * 30751da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 30761da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 30771da177e4SLinus Torvalds * and potentially makes it more efficient. 30781da177e4SLinus Torvalds * 3079c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 30808f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 3081c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 30821da177e4SLinus Torvalds */ 30832b740303SSouptick Joarder static vm_fault_t do_wp_page(struct vm_fault *vmf) 308482b0f8c3SJan Kara __releases(vmf->ptl) 30851da177e4SLinus Torvalds { 308682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 30871da177e4SLinus Torvalds 3088292924b2SPeter Xu if (userfaultfd_pte_wp(vma, *vmf->pte)) { 3089529b930bSAndrea Arcangeli pte_unmap_unlock(vmf->pte, vmf->ptl); 3090529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 3091529b930bSAndrea Arcangeli } 3092529b930bSAndrea Arcangeli 3093a41b70d6SJan Kara vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); 3094a41b70d6SJan Kara if (!vmf->page) { 3095251b97f5SPeter Zijlstra /* 309664e45507SPeter Feiner * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a 309764e45507SPeter Feiner * VM_PFNMAP VMA. 3098251b97f5SPeter Zijlstra * 3099251b97f5SPeter Zijlstra * We should not cow pages in a shared writeable mapping. 3100dd906184SBoaz Harrosh * Just mark the pages writable and/or call ops->pfn_mkwrite. 3101251b97f5SPeter Zijlstra */ 3102251b97f5SPeter Zijlstra if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 3103251b97f5SPeter Zijlstra (VM_WRITE|VM_SHARED)) 31042994302bSJan Kara return wp_pfn_shared(vmf); 31052f38ab2cSShachar Raindel 310682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3107a41b70d6SJan Kara return wp_page_copy(vmf); 3108251b97f5SPeter Zijlstra } 31091da177e4SLinus Torvalds 3110d08b3851SPeter Zijlstra /* 3111ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 3112ee6a6457SPeter Zijlstra * not dirty accountable. 3113d08b3851SPeter Zijlstra */ 311452d1e606SKirill Tkhai if (PageAnon(vmf->page)) { 311509854ba9SLinus Torvalds struct page *page = vmf->page; 311609854ba9SLinus Torvalds 311709854ba9SLinus Torvalds /* PageKsm() doesn't necessarily raise the page refcount */ 311809854ba9SLinus Torvalds if (PageKsm(page) || page_count(page) != 1) 311952d1e606SKirill Tkhai goto copy; 312009854ba9SLinus Torvalds if (!trylock_page(page)) 312152d1e606SKirill Tkhai goto copy; 312209854ba9SLinus Torvalds if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1) { 312309854ba9SLinus Torvalds unlock_page(page); 312409854ba9SLinus Torvalds goto copy; 312552d1e606SKirill Tkhai } 3126c44b6743SRik van Riel /* 312709854ba9SLinus Torvalds * Ok, we've got the only map reference, and the only 312809854ba9SLinus Torvalds * page count reference, and the page is locked, 312909854ba9SLinus Torvalds * it's dark out, and we're wearing sunglasses. Hit it. 3130c44b6743SRik van Riel */ 313109854ba9SLinus Torvalds unlock_page(page); 3132be068f29SLinus Torvalds wp_page_reuse(vmf); 3133997dd98dSJan Kara return VM_FAULT_WRITE; 3134ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 3135d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 3136a41b70d6SJan Kara return wp_page_shared(vmf); 31371da177e4SLinus Torvalds } 313852d1e606SKirill Tkhai copy: 31391da177e4SLinus Torvalds /* 31401da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 31411da177e4SLinus Torvalds */ 3142a41b70d6SJan Kara get_page(vmf->page); 314328766805SShachar Raindel 314482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3145a41b70d6SJan Kara return wp_page_copy(vmf); 31461da177e4SLinus Torvalds } 31471da177e4SLinus Torvalds 314897a89413SPeter Zijlstra static void unmap_mapping_range_vma(struct vm_area_struct *vma, 31491da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 31501da177e4SLinus Torvalds struct zap_details *details) 31511da177e4SLinus Torvalds { 3152f5cc4eefSAl Viro zap_page_range_single(vma, start_addr, end_addr - start_addr, details); 31531da177e4SLinus Torvalds } 31541da177e4SLinus Torvalds 3155f808c13fSDavidlohr Bueso static inline void unmap_mapping_range_tree(struct rb_root_cached *root, 31561da177e4SLinus Torvalds struct zap_details *details) 31571da177e4SLinus Torvalds { 31581da177e4SLinus Torvalds struct vm_area_struct *vma; 31591da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 31601da177e4SLinus Torvalds 31616b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, 31621da177e4SLinus Torvalds details->first_index, details->last_index) { 31631da177e4SLinus Torvalds 31641da177e4SLinus Torvalds vba = vma->vm_pgoff; 3165d6e93217SLibin vea = vba + vma_pages(vma) - 1; 31661da177e4SLinus Torvalds zba = details->first_index; 31671da177e4SLinus Torvalds if (zba < vba) 31681da177e4SLinus Torvalds zba = vba; 31691da177e4SLinus Torvalds zea = details->last_index; 31701da177e4SLinus Torvalds if (zea > vea) 31711da177e4SLinus Torvalds zea = vea; 31721da177e4SLinus Torvalds 317397a89413SPeter Zijlstra unmap_mapping_range_vma(vma, 31741da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 31751da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 317697a89413SPeter Zijlstra details); 31771da177e4SLinus Torvalds } 31781da177e4SLinus Torvalds } 31791da177e4SLinus Torvalds 31801da177e4SLinus Torvalds /** 3181977fbdcdSMatthew Wilcox * unmap_mapping_pages() - Unmap pages from processes. 3182977fbdcdSMatthew Wilcox * @mapping: The address space containing pages to be unmapped. 3183977fbdcdSMatthew Wilcox * @start: Index of first page to be unmapped. 3184977fbdcdSMatthew Wilcox * @nr: Number of pages to be unmapped. 0 to unmap to end of file. 3185977fbdcdSMatthew Wilcox * @even_cows: Whether to unmap even private COWed pages. 3186977fbdcdSMatthew Wilcox * 3187977fbdcdSMatthew Wilcox * Unmap the pages in this address space from any userspace process which 3188977fbdcdSMatthew Wilcox * has them mmaped. Generally, you want to remove COWed pages as well when 3189977fbdcdSMatthew Wilcox * a file is being truncated, but not when invalidating pages from the page 3190977fbdcdSMatthew Wilcox * cache. 3191977fbdcdSMatthew Wilcox */ 3192977fbdcdSMatthew Wilcox void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, 3193977fbdcdSMatthew Wilcox pgoff_t nr, bool even_cows) 3194977fbdcdSMatthew Wilcox { 3195977fbdcdSMatthew Wilcox struct zap_details details = { }; 3196977fbdcdSMatthew Wilcox 3197977fbdcdSMatthew Wilcox details.check_mapping = even_cows ? NULL : mapping; 3198977fbdcdSMatthew Wilcox details.first_index = start; 3199977fbdcdSMatthew Wilcox details.last_index = start + nr - 1; 3200977fbdcdSMatthew Wilcox if (details.last_index < details.first_index) 3201977fbdcdSMatthew Wilcox details.last_index = ULONG_MAX; 3202977fbdcdSMatthew Wilcox 3203977fbdcdSMatthew Wilcox i_mmap_lock_write(mapping); 3204977fbdcdSMatthew Wilcox if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) 3205977fbdcdSMatthew Wilcox unmap_mapping_range_tree(&mapping->i_mmap, &details); 3206977fbdcdSMatthew Wilcox i_mmap_unlock_write(mapping); 3207977fbdcdSMatthew Wilcox } 3208977fbdcdSMatthew Wilcox 3209977fbdcdSMatthew Wilcox /** 32108a5f14a2SKirill A. Shutemov * unmap_mapping_range - unmap the portion of all mmaps in the specified 3211977fbdcdSMatthew Wilcox * address_space corresponding to the specified byte range in the underlying 32128a5f14a2SKirill A. Shutemov * file. 32138a5f14a2SKirill A. Shutemov * 32143d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 32151da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 32161da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 321725d9e2d1Snpiggin@suse.de * boundary. Note that this is different from truncate_pagecache(), which 32181da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 32191da177e4SLinus Torvalds * partial pages. 32201da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 32211da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 32221da177e4SLinus Torvalds * end of the file. 32231da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 32241da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 32251da177e4SLinus Torvalds */ 32261da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 32271da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 32281da177e4SLinus Torvalds { 32291da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 32301da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 32311da177e4SLinus Torvalds 32321da177e4SLinus Torvalds /* Check for overflow. */ 32331da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 32341da177e4SLinus Torvalds long long holeend = 32351da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 32361da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 32371da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 32381da177e4SLinus Torvalds } 32391da177e4SLinus Torvalds 3240977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, hba, hlen, even_cows); 32411da177e4SLinus Torvalds } 32421da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 32431da177e4SLinus Torvalds 32441da177e4SLinus Torvalds /* 3245c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 32468f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 32479a95f3cfSPaul Cassella * We return with pte unmapped and unlocked. 32489a95f3cfSPaul Cassella * 3249c1e8d7c6SMichel Lespinasse * We return with the mmap_lock locked or unlocked in the same cases 32509a95f3cfSPaul Cassella * as does filemap_fault(). 32511da177e4SLinus Torvalds */ 32522b740303SSouptick Joarder vm_fault_t do_swap_page(struct vm_fault *vmf) 32531da177e4SLinus Torvalds { 325482b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3255eaf649ebSMinchan Kim struct page *page = NULL, *swapcache; 325665500d23SHugh Dickins swp_entry_t entry; 32571da177e4SLinus Torvalds pte_t pte; 3258d065bd81SMichel Lespinasse int locked; 3259ad8c2ee8SRik van Riel int exclusive = 0; 32602b740303SSouptick Joarder vm_fault_t ret = 0; 3261aae466b0SJoonsoo Kim void *shadow = NULL; 32621da177e4SLinus Torvalds 3263eaf649ebSMinchan Kim if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) 32648f4e2101SHugh Dickins goto out; 326565500d23SHugh Dickins 32662994302bSJan Kara entry = pte_to_swp_entry(vmf->orig_pte); 3267d1737fdbSAndi Kleen if (unlikely(non_swap_entry(entry))) { 32680697212aSChristoph Lameter if (is_migration_entry(entry)) { 326982b0f8c3SJan Kara migration_entry_wait(vma->vm_mm, vmf->pmd, 327082b0f8c3SJan Kara vmf->address); 32715042db43SJérôme Glisse } else if (is_device_private_entry(entry)) { 3272897e6365SChristoph Hellwig vmf->page = device_private_entry_to_page(entry); 3273897e6365SChristoph Hellwig ret = vmf->page->pgmap->ops->migrate_to_ram(vmf); 3274d1737fdbSAndi Kleen } else if (is_hwpoison_entry(entry)) { 3275d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3276d1737fdbSAndi Kleen } else { 32772994302bSJan Kara print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); 3278d99be1a8SHugh Dickins ret = VM_FAULT_SIGBUS; 3279d1737fdbSAndi Kleen } 32800697212aSChristoph Lameter goto out; 32810697212aSChristoph Lameter } 32820bcac06fSMinchan Kim 32830bcac06fSMinchan Kim 32840ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 3285eaf649ebSMinchan Kim page = lookup_swap_cache(entry, vma, vmf->address); 3286f8020772SMinchan Kim swapcache = page; 3287f8020772SMinchan Kim 32881da177e4SLinus Torvalds if (!page) { 32890bcac06fSMinchan Kim struct swap_info_struct *si = swp_swap_info(entry); 32900bcac06fSMinchan Kim 3291a449bf58SQian Cai if (data_race(si->flags & SWP_SYNCHRONOUS_IO) && 3292eb085574SHuang Ying __swap_count(entry) == 1) { 32930bcac06fSMinchan Kim /* skip swapcache */ 3294e9e9b7ecSMinchan Kim page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, 3295e9e9b7ecSMinchan Kim vmf->address); 32960bcac06fSMinchan Kim if (page) { 32974c6355b2SJohannes Weiner int err; 32984c6355b2SJohannes Weiner 32990bcac06fSMinchan Kim __SetPageLocked(page); 33000bcac06fSMinchan Kim __SetPageSwapBacked(page); 33010bcac06fSMinchan Kim set_page_private(page, entry.val); 33024c6355b2SJohannes Weiner 33034c6355b2SJohannes Weiner /* Tell memcg to use swap ownership records */ 33044c6355b2SJohannes Weiner SetPageSwapCache(page); 33054c6355b2SJohannes Weiner err = mem_cgroup_charge(page, vma->vm_mm, 3306d9eb1ea2SJohannes Weiner GFP_KERNEL); 33074c6355b2SJohannes Weiner ClearPageSwapCache(page); 3308545b1b07SMichal Hocko if (err) { 3309545b1b07SMichal Hocko ret = VM_FAULT_OOM; 33104c6355b2SJohannes Weiner goto out_page; 3311545b1b07SMichal Hocko } 33124c6355b2SJohannes Weiner 3313aae466b0SJoonsoo Kim shadow = get_shadow_from_swap_cache(entry); 3314aae466b0SJoonsoo Kim if (shadow) 3315aae466b0SJoonsoo Kim workingset_refault(page, shadow); 33160076f029SJoonsoo Kim 33176058eaecSJohannes Weiner lru_cache_add(page); 33180bcac06fSMinchan Kim swap_readpage(page, true); 33190bcac06fSMinchan Kim } 3320aa8d22a1SMinchan Kim } else { 3321e9e9b7ecSMinchan Kim page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, 3322e9e9b7ecSMinchan Kim vmf); 3323aa8d22a1SMinchan Kim swapcache = page; 33240bcac06fSMinchan Kim } 33250bcac06fSMinchan Kim 3326ec560175SHuang Ying if (!page) { 33271da177e4SLinus Torvalds /* 33288f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 33298f4e2101SHugh Dickins * while we released the pte lock. 33301da177e4SLinus Torvalds */ 333182b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 333282b0f8c3SJan Kara vmf->address, &vmf->ptl); 33332994302bSJan Kara if (likely(pte_same(*vmf->pte, vmf->orig_pte))) 33341da177e4SLinus Torvalds ret = VM_FAULT_OOM; 33350ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 333665500d23SHugh Dickins goto unlock; 33371da177e4SLinus Torvalds } 33381da177e4SLinus Torvalds 33391da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 33401da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 3341f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 33422262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); 3343d1737fdbSAndi Kleen } else if (PageHWPoison(page)) { 334471f72525SWu Fengguang /* 334571f72525SWu Fengguang * hwpoisoned dirty swapcache pages are kept for killing 334671f72525SWu Fengguang * owner processes (which may be unknown at hwpoison time) 334771f72525SWu Fengguang */ 3348d1737fdbSAndi Kleen ret = VM_FAULT_HWPOISON; 3349d1737fdbSAndi Kleen delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 33504779cb31SAndi Kleen goto out_release; 33511da177e4SLinus Torvalds } 33521da177e4SLinus Torvalds 335382b0f8c3SJan Kara locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); 3354e709ffd6SRik van Riel 335520a1022dSBalbir Singh delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 3356d065bd81SMichel Lespinasse if (!locked) { 3357d065bd81SMichel Lespinasse ret |= VM_FAULT_RETRY; 3358d065bd81SMichel Lespinasse goto out_release; 3359d065bd81SMichel Lespinasse } 33601da177e4SLinus Torvalds 33614969c119SAndrea Arcangeli /* 336231c4a3d3SHugh Dickins * Make sure try_to_free_swap or reuse_swap_page or swapoff did not 336331c4a3d3SHugh Dickins * release the swapcache from under us. The page pin, and pte_same 336431c4a3d3SHugh Dickins * test below, are not enough to exclude that. Even if it is still 336531c4a3d3SHugh Dickins * swapcache, we need to check that the page's swap has not changed. 33664969c119SAndrea Arcangeli */ 33670bcac06fSMinchan Kim if (unlikely((!PageSwapCache(page) || 33680bcac06fSMinchan Kim page_private(page) != entry.val)) && swapcache) 33694969c119SAndrea Arcangeli goto out_page; 33704969c119SAndrea Arcangeli 337182b0f8c3SJan Kara page = ksm_might_need_to_copy(page, vma, vmf->address); 33724969c119SAndrea Arcangeli if (unlikely(!page)) { 33735ad64688SHugh Dickins ret = VM_FAULT_OOM; 33744969c119SAndrea Arcangeli page = swapcache; 33754969c119SAndrea Arcangeli goto out_page; 33764969c119SAndrea Arcangeli } 33775ad64688SHugh Dickins 33789d82c694SJohannes Weiner cgroup_throttle_swaprate(page, GFP_KERNEL); 3379073e587eSKAMEZAWA Hiroyuki 33801da177e4SLinus Torvalds /* 33818f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 33821da177e4SLinus Torvalds */ 338382b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 338482b0f8c3SJan Kara &vmf->ptl); 33852994302bSJan Kara if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) 3386b8107480SKirill Korotaev goto out_nomap; 3387b8107480SKirill Korotaev 3388b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 3389b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 3390b8107480SKirill Korotaev goto out_nomap; 33911da177e4SLinus Torvalds } 33921da177e4SLinus Torvalds 33938c7c6e34SKAMEZAWA Hiroyuki /* 33948c7c6e34SKAMEZAWA Hiroyuki * The page isn't present yet, go ahead with the fault. 33958c7c6e34SKAMEZAWA Hiroyuki * 33968c7c6e34SKAMEZAWA Hiroyuki * Be careful about the sequence of operations here. 33978c7c6e34SKAMEZAWA Hiroyuki * To get its accounting right, reuse_swap_page() must be called 33988c7c6e34SKAMEZAWA Hiroyuki * while the page is counted on swap but not yet in mapcount i.e. 33998c7c6e34SKAMEZAWA Hiroyuki * before page_add_anon_rmap() and swap_free(); try_to_free_swap() 34008c7c6e34SKAMEZAWA Hiroyuki * must be called after the swap_free(), or it will never succeed. 34018c7c6e34SKAMEZAWA Hiroyuki */ 34021da177e4SLinus Torvalds 3403bae473a4SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 3404bae473a4SKirill A. Shutemov dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); 34051da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 340682b0f8c3SJan Kara if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { 34071da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 340882b0f8c3SJan Kara vmf->flags &= ~FAULT_FLAG_WRITE; 34099a5b489bSAndrea Arcangeli ret |= VM_FAULT_WRITE; 3410d281ee61SKirill A. Shutemov exclusive = RMAP_EXCLUSIVE; 34111da177e4SLinus Torvalds } 34121da177e4SLinus Torvalds flush_icache_page(vma, page); 34132994302bSJan Kara if (pte_swp_soft_dirty(vmf->orig_pte)) 3414179ef71cSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 3415f45ec5ffSPeter Xu if (pte_swp_uffd_wp(vmf->orig_pte)) { 3416f45ec5ffSPeter Xu pte = pte_mkuffd_wp(pte); 3417f45ec5ffSPeter Xu pte = pte_wrprotect(pte); 3418f45ec5ffSPeter Xu } 341982b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); 3420ca827d55SKhalid Aziz arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); 34212994302bSJan Kara vmf->orig_pte = pte; 34220bcac06fSMinchan Kim 34230bcac06fSMinchan Kim /* ksm created a completely new copy */ 34240bcac06fSMinchan Kim if (unlikely(page != swapcache && swapcache)) { 342582b0f8c3SJan Kara page_add_new_anon_rmap(page, vma, vmf->address, false); 3426b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 34270bcac06fSMinchan Kim } else { 34280bcac06fSMinchan Kim do_page_add_anon_rmap(page, vma, vmf->address, exclusive); 342900501b53SJohannes Weiner } 34301da177e4SLinus Torvalds 3431c475a8abSHugh Dickins swap_free(entry); 34325ccc5abaSVladimir Davydov if (mem_cgroup_swap_full(page) || 34335ccc5abaSVladimir Davydov (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) 3434a2c43eedSHugh Dickins try_to_free_swap(page); 3435c475a8abSHugh Dickins unlock_page(page); 34360bcac06fSMinchan Kim if (page != swapcache && swapcache) { 34374969c119SAndrea Arcangeli /* 34384969c119SAndrea Arcangeli * Hold the lock to avoid the swap entry to be reused 34394969c119SAndrea Arcangeli * until we take the PT lock for the pte_same() check 34404969c119SAndrea Arcangeli * (to avoid false positives from pte_same). For 34414969c119SAndrea Arcangeli * further safety release the lock after the swap_free 34424969c119SAndrea Arcangeli * so that the swap count won't change under a 34434969c119SAndrea Arcangeli * parallel locked swapcache. 34444969c119SAndrea Arcangeli */ 34454969c119SAndrea Arcangeli unlock_page(swapcache); 344609cbfeafSKirill A. Shutemov put_page(swapcache); 34474969c119SAndrea Arcangeli } 3448c475a8abSHugh Dickins 344982b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 34502994302bSJan Kara ret |= do_wp_page(vmf); 345161469f1dSHugh Dickins if (ret & VM_FAULT_ERROR) 345261469f1dSHugh Dickins ret &= VM_FAULT_ERROR; 34531da177e4SLinus Torvalds goto out; 34541da177e4SLinus Torvalds } 34551da177e4SLinus Torvalds 34561da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 345782b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 345865500d23SHugh Dickins unlock: 345982b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 34601da177e4SLinus Torvalds out: 34611da177e4SLinus Torvalds return ret; 3462b8107480SKirill Korotaev out_nomap: 346382b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 3464bc43f75cSJohannes Weiner out_page: 3465b8107480SKirill Korotaev unlock_page(page); 34664779cb31SAndi Kleen out_release: 346709cbfeafSKirill A. Shutemov put_page(page); 34680bcac06fSMinchan Kim if (page != swapcache && swapcache) { 34694969c119SAndrea Arcangeli unlock_page(swapcache); 347009cbfeafSKirill A. Shutemov put_page(swapcache); 34714969c119SAndrea Arcangeli } 347265500d23SHugh Dickins return ret; 34731da177e4SLinus Torvalds } 34741da177e4SLinus Torvalds 34751da177e4SLinus Torvalds /* 3476c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 34778f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 3478c1e8d7c6SMichel Lespinasse * We return with mmap_lock still held, but pte unmapped and unlocked. 34791da177e4SLinus Torvalds */ 34802b740303SSouptick Joarder static vm_fault_t do_anonymous_page(struct vm_fault *vmf) 34811da177e4SLinus Torvalds { 348282b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 34838f4e2101SHugh Dickins struct page *page; 34842b740303SSouptick Joarder vm_fault_t ret = 0; 34851da177e4SLinus Torvalds pte_t entry; 34861da177e4SLinus Torvalds 34876b7339f4SKirill A. Shutemov /* File mapping without ->vm_ops ? */ 34886b7339f4SKirill A. Shutemov if (vma->vm_flags & VM_SHARED) 34896b7339f4SKirill A. Shutemov return VM_FAULT_SIGBUS; 34906b7339f4SKirill A. Shutemov 34917267ec00SKirill A. Shutemov /* 34927267ec00SKirill A. Shutemov * Use pte_alloc() instead of pte_alloc_map(). We can't run 34937267ec00SKirill A. Shutemov * pte_offset_map() on pmds where a huge pmd might be created 34947267ec00SKirill A. Shutemov * from a different thread. 34957267ec00SKirill A. Shutemov * 34963e4e28c5SMichel Lespinasse * pte_alloc_map() is safe to use under mmap_write_lock(mm) or when 34977267ec00SKirill A. Shutemov * parallel threads are excluded by other means. 34987267ec00SKirill A. Shutemov * 34993e4e28c5SMichel Lespinasse * Here we only have mmap_read_lock(mm). 35007267ec00SKirill A. Shutemov */ 35014cf58924SJoel Fernandes (Google) if (pte_alloc(vma->vm_mm, vmf->pmd)) 35027267ec00SKirill A. Shutemov return VM_FAULT_OOM; 35037267ec00SKirill A. Shutemov 35047267ec00SKirill A. Shutemov /* See the comment in pte_alloc_one_map() */ 350582b0f8c3SJan Kara if (unlikely(pmd_trans_unstable(vmf->pmd))) 35067267ec00SKirill A. Shutemov return 0; 35077267ec00SKirill A. Shutemov 350811ac5524SLinus Torvalds /* Use the zero-page for reads */ 350982b0f8c3SJan Kara if (!(vmf->flags & FAULT_FLAG_WRITE) && 3510bae473a4SKirill A. Shutemov !mm_forbids_zeropage(vma->vm_mm)) { 351182b0f8c3SJan Kara entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), 351262eede62SHugh Dickins vma->vm_page_prot)); 351382b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, 351482b0f8c3SJan Kara vmf->address, &vmf->ptl); 35157df67697SBibo Mao if (!pte_none(*vmf->pte)) { 35167df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 3517a13ea5b7SHugh Dickins goto unlock; 35187df67697SBibo Mao } 35196b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 35206b31d595SMichal Hocko if (ret) 35216b31d595SMichal Hocko goto unlock; 35226b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 35236b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 352482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 352582b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 35266b251fc9SAndrea Arcangeli } 3527a13ea5b7SHugh Dickins goto setpte; 3528a13ea5b7SHugh Dickins } 3529a13ea5b7SHugh Dickins 35301da177e4SLinus Torvalds /* Allocate our own private page. */ 35311da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 353265500d23SHugh Dickins goto oom; 353382b0f8c3SJan Kara page = alloc_zeroed_user_highpage_movable(vma, vmf->address); 35341da177e4SLinus Torvalds if (!page) 353565500d23SHugh Dickins goto oom; 3536eb3c24f3SMel Gorman 3537d9eb1ea2SJohannes Weiner if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL)) 3538eb3c24f3SMel Gorman goto oom_free_page; 35399d82c694SJohannes Weiner cgroup_throttle_swaprate(page, GFP_KERNEL); 3540eb3c24f3SMel Gorman 354152f37629SMinchan Kim /* 354252f37629SMinchan Kim * The memory barrier inside __SetPageUptodate makes sure that 3543f4f5329dSWei Yang * preceding stores to the page contents become visible before 354452f37629SMinchan Kim * the set_pte_at() write. 354552f37629SMinchan Kim */ 35460ed361deSNick Piggin __SetPageUptodate(page); 35471da177e4SLinus Torvalds 354865500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 354944bf431bSBibo Mao entry = pte_sw_mkyoung(entry); 35501ac0cb5dSHugh Dickins if (vma->vm_flags & VM_WRITE) 35511ac0cb5dSHugh Dickins entry = pte_mkwrite(pte_mkdirty(entry)); 35528f4e2101SHugh Dickins 355382b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 355482b0f8c3SJan Kara &vmf->ptl); 35557df67697SBibo Mao if (!pte_none(*vmf->pte)) { 35567df67697SBibo Mao update_mmu_cache(vma, vmf->address, vmf->pte); 35578f4e2101SHugh Dickins goto release; 35587df67697SBibo Mao } 35599ba69294SHugh Dickins 35606b31d595SMichal Hocko ret = check_stable_address_space(vma->vm_mm); 35616b31d595SMichal Hocko if (ret) 35626b31d595SMichal Hocko goto release; 35636b31d595SMichal Hocko 35646b251fc9SAndrea Arcangeli /* Deliver the page fault to userland, check inside PT lock */ 35656b251fc9SAndrea Arcangeli if (userfaultfd_missing(vma)) { 356682b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 356709cbfeafSKirill A. Shutemov put_page(page); 356882b0f8c3SJan Kara return handle_userfault(vmf, VM_UFFD_MISSING); 35696b251fc9SAndrea Arcangeli } 35706b251fc9SAndrea Arcangeli 3571bae473a4SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 357282b0f8c3SJan Kara page_add_new_anon_rmap(page, vma, vmf->address, false); 3573b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 3574a13ea5b7SHugh Dickins setpte: 357582b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); 35761da177e4SLinus Torvalds 35771da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 357882b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 357965500d23SHugh Dickins unlock: 358082b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 35816b31d595SMichal Hocko return ret; 35828f4e2101SHugh Dickins release: 358309cbfeafSKirill A. Shutemov put_page(page); 35848f4e2101SHugh Dickins goto unlock; 35858a9f3ccdSBalbir Singh oom_free_page: 358609cbfeafSKirill A. Shutemov put_page(page); 358765500d23SHugh Dickins oom: 35881da177e4SLinus Torvalds return VM_FAULT_OOM; 35891da177e4SLinus Torvalds } 35901da177e4SLinus Torvalds 35919a95f3cfSPaul Cassella /* 3592c1e8d7c6SMichel Lespinasse * The mmap_lock must have been held on entry, and may have been 35939a95f3cfSPaul Cassella * released depending on flags and vma->vm_ops->fault() return value. 35949a95f3cfSPaul Cassella * See filemap_fault() and __lock_page_retry(). 35959a95f3cfSPaul Cassella */ 35962b740303SSouptick Joarder static vm_fault_t __do_fault(struct vm_fault *vmf) 35977eae74afSKirill A. Shutemov { 359882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 35992b740303SSouptick Joarder vm_fault_t ret; 36007eae74afSKirill A. Shutemov 360163f3655fSMichal Hocko /* 360263f3655fSMichal Hocko * Preallocate pte before we take page_lock because this might lead to 360363f3655fSMichal Hocko * deadlocks for memcg reclaim which waits for pages under writeback: 360463f3655fSMichal Hocko * lock_page(A) 360563f3655fSMichal Hocko * SetPageWriteback(A) 360663f3655fSMichal Hocko * unlock_page(A) 360763f3655fSMichal Hocko * lock_page(B) 360863f3655fSMichal Hocko * lock_page(B) 3609d383807aSYanfei Xu * pte_alloc_one 361063f3655fSMichal Hocko * shrink_page_list 361163f3655fSMichal Hocko * wait_on_page_writeback(A) 361263f3655fSMichal Hocko * SetPageWriteback(B) 361363f3655fSMichal Hocko * unlock_page(B) 361463f3655fSMichal Hocko * # flush A, B to clear the writeback 361563f3655fSMichal Hocko */ 361663f3655fSMichal Hocko if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { 3617a7069ee3SYanfei Xu vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 361863f3655fSMichal Hocko if (!vmf->prealloc_pte) 361963f3655fSMichal Hocko return VM_FAULT_OOM; 362063f3655fSMichal Hocko smp_wmb(); /* See comment in __pte_alloc() */ 362163f3655fSMichal Hocko } 362263f3655fSMichal Hocko 362311bac800SDave Jiang ret = vma->vm_ops->fault(vmf); 36243917048dSJan Kara if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | 3625b1aa812bSJan Kara VM_FAULT_DONE_COW))) 36267eae74afSKirill A. Shutemov return ret; 36277eae74afSKirill A. Shutemov 3628667240e0SJan Kara if (unlikely(PageHWPoison(vmf->page))) { 36297eae74afSKirill A. Shutemov if (ret & VM_FAULT_LOCKED) 3630667240e0SJan Kara unlock_page(vmf->page); 3631667240e0SJan Kara put_page(vmf->page); 3632936ca80dSJan Kara vmf->page = NULL; 36337eae74afSKirill A. Shutemov return VM_FAULT_HWPOISON; 36347eae74afSKirill A. Shutemov } 36357eae74afSKirill A. Shutemov 36367eae74afSKirill A. Shutemov if (unlikely(!(ret & VM_FAULT_LOCKED))) 3637667240e0SJan Kara lock_page(vmf->page); 36387eae74afSKirill A. Shutemov else 3639667240e0SJan Kara VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); 36407eae74afSKirill A. Shutemov 36417eae74afSKirill A. Shutemov return ret; 36427eae74afSKirill A. Shutemov } 36437eae74afSKirill A. Shutemov 3644d0f0931dSRoss Zwisler /* 3645d0f0931dSRoss Zwisler * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. 3646d0f0931dSRoss Zwisler * If we check pmd_trans_unstable() first we will trip the bad_pmd() check 3647d0f0931dSRoss Zwisler * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly 3648d0f0931dSRoss Zwisler * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. 3649d0f0931dSRoss Zwisler */ 3650d0f0931dSRoss Zwisler static int pmd_devmap_trans_unstable(pmd_t *pmd) 3651d0f0931dSRoss Zwisler { 3652d0f0931dSRoss Zwisler return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); 3653d0f0931dSRoss Zwisler } 3654d0f0931dSRoss Zwisler 36552b740303SSouptick Joarder static vm_fault_t pte_alloc_one_map(struct vm_fault *vmf) 36567267ec00SKirill A. Shutemov { 365782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 36587267ec00SKirill A. Shutemov 365982b0f8c3SJan Kara if (!pmd_none(*vmf->pmd)) 36607267ec00SKirill A. Shutemov goto map_pte; 366182b0f8c3SJan Kara if (vmf->prealloc_pte) { 366282b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 366382b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) { 366482b0f8c3SJan Kara spin_unlock(vmf->ptl); 36657267ec00SKirill A. Shutemov goto map_pte; 36667267ec00SKirill A. Shutemov } 36677267ec00SKirill A. Shutemov 3668c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 366982b0f8c3SJan Kara pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 367082b0f8c3SJan Kara spin_unlock(vmf->ptl); 36717f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 36724cf58924SJoel Fernandes (Google) } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) { 36737267ec00SKirill A. Shutemov return VM_FAULT_OOM; 36747267ec00SKirill A. Shutemov } 36757267ec00SKirill A. Shutemov map_pte: 36767267ec00SKirill A. Shutemov /* 36777267ec00SKirill A. Shutemov * If a huge pmd materialized under us just retry later. Use 3678d0f0931dSRoss Zwisler * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of 3679d0f0931dSRoss Zwisler * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge 3680d0f0931dSRoss Zwisler * under us and then back to pmd_none, as a result of MADV_DONTNEED 3681d0f0931dSRoss Zwisler * running immediately after a huge pmd fault in a different thread of 3682d0f0931dSRoss Zwisler * this mm, in turn leading to a misleading pmd_trans_huge() retval. 3683d0f0931dSRoss Zwisler * All we have to ensure is that it is a regular pmd that we can walk 3684d0f0931dSRoss Zwisler * with pte_offset_map() and we can do that through an atomic read in 3685d0f0931dSRoss Zwisler * C, which is what pmd_trans_unstable() provides. 36867267ec00SKirill A. Shutemov */ 3687d0f0931dSRoss Zwisler if (pmd_devmap_trans_unstable(vmf->pmd)) 36887267ec00SKirill A. Shutemov return VM_FAULT_NOPAGE; 36897267ec00SKirill A. Shutemov 3690d0f0931dSRoss Zwisler /* 3691d0f0931dSRoss Zwisler * At this point we know that our vmf->pmd points to a page of ptes 3692d0f0931dSRoss Zwisler * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() 3693d0f0931dSRoss Zwisler * for the duration of the fault. If a racing MADV_DONTNEED runs and 3694d0f0931dSRoss Zwisler * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still 3695d0f0931dSRoss Zwisler * be valid and we will re-check to make sure the vmf->pte isn't 3696d0f0931dSRoss Zwisler * pte_none() under vmf->ptl protection when we return to 3697d0f0931dSRoss Zwisler * alloc_set_pte(). 3698d0f0931dSRoss Zwisler */ 369982b0f8c3SJan Kara vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, 370082b0f8c3SJan Kara &vmf->ptl); 37017267ec00SKirill A. Shutemov return 0; 37027267ec00SKirill A. Shutemov } 37037267ec00SKirill A. Shutemov 3704396bcc52SMatthew Wilcox (Oracle) #ifdef CONFIG_TRANSPARENT_HUGEPAGE 370582b0f8c3SJan Kara static void deposit_prealloc_pte(struct vm_fault *vmf) 3706953c66c2SAneesh Kumar K.V { 370782b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 3708953c66c2SAneesh Kumar K.V 370982b0f8c3SJan Kara pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); 3710953c66c2SAneesh Kumar K.V /* 3711953c66c2SAneesh Kumar K.V * We are going to consume the prealloc table, 3712953c66c2SAneesh Kumar K.V * count that as nr_ptes. 3713953c66c2SAneesh Kumar K.V */ 3714c4812909SKirill A. Shutemov mm_inc_nr_ptes(vma->vm_mm); 37157f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 3716953c66c2SAneesh Kumar K.V } 3717953c66c2SAneesh Kumar K.V 37182b740303SSouptick Joarder static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 371910102459SKirill A. Shutemov { 372082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 372182b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 372282b0f8c3SJan Kara unsigned long haddr = vmf->address & HPAGE_PMD_MASK; 372310102459SKirill A. Shutemov pmd_t entry; 37242b740303SSouptick Joarder int i; 3725d01ac3c3SMatthew Wilcox (Oracle) vm_fault_t ret = VM_FAULT_FALLBACK; 372610102459SKirill A. Shutemov 372710102459SKirill A. Shutemov if (!transhuge_vma_suitable(vma, haddr)) 3728d01ac3c3SMatthew Wilcox (Oracle) return ret; 372910102459SKirill A. Shutemov 373010102459SKirill A. Shutemov page = compound_head(page); 3731d01ac3c3SMatthew Wilcox (Oracle) if (compound_order(page) != HPAGE_PMD_ORDER) 3732d01ac3c3SMatthew Wilcox (Oracle) return ret; 373310102459SKirill A. Shutemov 3734953c66c2SAneesh Kumar K.V /* 3735953c66c2SAneesh Kumar K.V * Archs like ppc64 need additonal space to store information 3736953c66c2SAneesh Kumar K.V * related to pte entry. Use the preallocated table for that. 3737953c66c2SAneesh Kumar K.V */ 373882b0f8c3SJan Kara if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { 37394cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); 374082b0f8c3SJan Kara if (!vmf->prealloc_pte) 3741953c66c2SAneesh Kumar K.V return VM_FAULT_OOM; 3742953c66c2SAneesh Kumar K.V smp_wmb(); /* See comment in __pte_alloc() */ 3743953c66c2SAneesh Kumar K.V } 3744953c66c2SAneesh Kumar K.V 374582b0f8c3SJan Kara vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); 374682b0f8c3SJan Kara if (unlikely(!pmd_none(*vmf->pmd))) 374710102459SKirill A. Shutemov goto out; 374810102459SKirill A. Shutemov 374910102459SKirill A. Shutemov for (i = 0; i < HPAGE_PMD_NR; i++) 375010102459SKirill A. Shutemov flush_icache_page(vma, page + i); 375110102459SKirill A. Shutemov 375210102459SKirill A. Shutemov entry = mk_huge_pmd(page, vma->vm_page_prot); 375310102459SKirill A. Shutemov if (write) 3754f55e1014SLinus Torvalds entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); 375510102459SKirill A. Shutemov 3756fadae295SYang Shi add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); 375710102459SKirill A. Shutemov page_add_file_rmap(page, true); 3758953c66c2SAneesh Kumar K.V /* 3759953c66c2SAneesh Kumar K.V * deposit and withdraw with pmd lock held 3760953c66c2SAneesh Kumar K.V */ 3761953c66c2SAneesh Kumar K.V if (arch_needs_pgtable_deposit()) 376282b0f8c3SJan Kara deposit_prealloc_pte(vmf); 376310102459SKirill A. Shutemov 376482b0f8c3SJan Kara set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); 376510102459SKirill A. Shutemov 376682b0f8c3SJan Kara update_mmu_cache_pmd(vma, haddr, vmf->pmd); 376710102459SKirill A. Shutemov 376810102459SKirill A. Shutemov /* fault is handled */ 376910102459SKirill A. Shutemov ret = 0; 377095ecedcdSKirill A. Shutemov count_vm_event(THP_FILE_MAPPED); 377110102459SKirill A. Shutemov out: 377282b0f8c3SJan Kara spin_unlock(vmf->ptl); 377310102459SKirill A. Shutemov return ret; 377410102459SKirill A. Shutemov } 377510102459SKirill A. Shutemov #else 37762b740303SSouptick Joarder static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) 377710102459SKirill A. Shutemov { 377810102459SKirill A. Shutemov BUILD_BUG(); 377910102459SKirill A. Shutemov return 0; 378010102459SKirill A. Shutemov } 378110102459SKirill A. Shutemov #endif 378210102459SKirill A. Shutemov 37838c6e50b0SKirill A. Shutemov /** 37847267ec00SKirill A. Shutemov * alloc_set_pte - setup new PTE entry for given page and add reverse page 3785f1dc1685SRandy Dunlap * mapping. If needed, the function allocates page table or use pre-allocated. 37868c6e50b0SKirill A. Shutemov * 378782b0f8c3SJan Kara * @vmf: fault environment 37888c6e50b0SKirill A. Shutemov * @page: page to map 37898c6e50b0SKirill A. Shutemov * 379082b0f8c3SJan Kara * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on 379182b0f8c3SJan Kara * return. 37928c6e50b0SKirill A. Shutemov * 37938c6e50b0SKirill A. Shutemov * Target users are page handler itself and implementations of 37948c6e50b0SKirill A. Shutemov * vm_ops->map_pages. 3795a862f68aSMike Rapoport * 3796a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 37978c6e50b0SKirill A. Shutemov */ 37989d82c694SJohannes Weiner vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct page *page) 37993bb97794SKirill A. Shutemov { 380082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 380182b0f8c3SJan Kara bool write = vmf->flags & FAULT_FLAG_WRITE; 38023bb97794SKirill A. Shutemov pte_t entry; 38032b740303SSouptick Joarder vm_fault_t ret; 380410102459SKirill A. Shutemov 3805396bcc52SMatthew Wilcox (Oracle) if (pmd_none(*vmf->pmd) && PageTransCompound(page)) { 380682b0f8c3SJan Kara ret = do_set_pmd(vmf, page); 380710102459SKirill A. Shutemov if (ret != VM_FAULT_FALLBACK) 3808b0b9b3dfSHugh Dickins return ret; 380910102459SKirill A. Shutemov } 38103bb97794SKirill A. Shutemov 381182b0f8c3SJan Kara if (!vmf->pte) { 381282b0f8c3SJan Kara ret = pte_alloc_one_map(vmf); 38137267ec00SKirill A. Shutemov if (ret) 3814b0b9b3dfSHugh Dickins return ret; 38157267ec00SKirill A. Shutemov } 38167267ec00SKirill A. Shutemov 38177267ec00SKirill A. Shutemov /* Re-check under ptl */ 38187df67697SBibo Mao if (unlikely(!pte_none(*vmf->pte))) { 38197df67697SBibo Mao update_mmu_tlb(vma, vmf->address, vmf->pte); 3820b0b9b3dfSHugh Dickins return VM_FAULT_NOPAGE; 38217df67697SBibo Mao } 38227267ec00SKirill A. Shutemov 38233bb97794SKirill A. Shutemov flush_icache_page(vma, page); 38243bb97794SKirill A. Shutemov entry = mk_pte(page, vma->vm_page_prot); 382544bf431bSBibo Mao entry = pte_sw_mkyoung(entry); 38263bb97794SKirill A. Shutemov if (write) 38273bb97794SKirill A. Shutemov entry = maybe_mkwrite(pte_mkdirty(entry), vma); 3828bae473a4SKirill A. Shutemov /* copy-on-write page */ 3829bae473a4SKirill A. Shutemov if (write && !(vma->vm_flags & VM_SHARED)) { 38303bb97794SKirill A. Shutemov inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); 383182b0f8c3SJan Kara page_add_new_anon_rmap(page, vma, vmf->address, false); 3832b518154eSJoonsoo Kim lru_cache_add_inactive_or_unevictable(page, vma); 38333bb97794SKirill A. Shutemov } else { 3834eca56ff9SJerome Marchand inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); 3835dd78feddSKirill A. Shutemov page_add_file_rmap(page, false); 38363bb97794SKirill A. Shutemov } 383782b0f8c3SJan Kara set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); 38383bb97794SKirill A. Shutemov 38393bb97794SKirill A. Shutemov /* no need to invalidate: a not-present page won't be cached */ 384082b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 38417267ec00SKirill A. Shutemov 3842b0b9b3dfSHugh Dickins return 0; 38433bb97794SKirill A. Shutemov } 38443bb97794SKirill A. Shutemov 38459118c0cbSJan Kara 38469118c0cbSJan Kara /** 38479118c0cbSJan Kara * finish_fault - finish page fault once we have prepared the page to fault 38489118c0cbSJan Kara * 38499118c0cbSJan Kara * @vmf: structure describing the fault 38509118c0cbSJan Kara * 38519118c0cbSJan Kara * This function handles all that is needed to finish a page fault once the 38529118c0cbSJan Kara * page to fault in is prepared. It handles locking of PTEs, inserts PTE for 38539118c0cbSJan Kara * given page, adds reverse page mapping, handles memcg charges and LRU 3854a862f68aSMike Rapoport * addition. 38559118c0cbSJan Kara * 38569118c0cbSJan Kara * The function expects the page to be locked and on success it consumes a 38579118c0cbSJan Kara * reference of a page being mapped (for the PTE which maps it). 3858a862f68aSMike Rapoport * 3859a862f68aSMike Rapoport * Return: %0 on success, %VM_FAULT_ code in case of error. 38609118c0cbSJan Kara */ 38612b740303SSouptick Joarder vm_fault_t finish_fault(struct vm_fault *vmf) 38629118c0cbSJan Kara { 38639118c0cbSJan Kara struct page *page; 38642b740303SSouptick Joarder vm_fault_t ret = 0; 38659118c0cbSJan Kara 38669118c0cbSJan Kara /* Did we COW the page? */ 38679118c0cbSJan Kara if ((vmf->flags & FAULT_FLAG_WRITE) && 38689118c0cbSJan Kara !(vmf->vma->vm_flags & VM_SHARED)) 38699118c0cbSJan Kara page = vmf->cow_page; 38709118c0cbSJan Kara else 38719118c0cbSJan Kara page = vmf->page; 38726b31d595SMichal Hocko 38736b31d595SMichal Hocko /* 38746b31d595SMichal Hocko * check even for read faults because we might have lost our CoWed 38756b31d595SMichal Hocko * page 38766b31d595SMichal Hocko */ 38776b31d595SMichal Hocko if (!(vmf->vma->vm_flags & VM_SHARED)) 38786b31d595SMichal Hocko ret = check_stable_address_space(vmf->vma->vm_mm); 38796b31d595SMichal Hocko if (!ret) 38809d82c694SJohannes Weiner ret = alloc_set_pte(vmf, page); 38819118c0cbSJan Kara if (vmf->pte) 38829118c0cbSJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 38839118c0cbSJan Kara return ret; 38849118c0cbSJan Kara } 38859118c0cbSJan Kara 38863a91053aSKirill A. Shutemov static unsigned long fault_around_bytes __read_mostly = 38873a91053aSKirill A. Shutemov rounddown_pow_of_two(65536); 3888a9b0f861SKirill A. Shutemov 38891592eef0SKirill A. Shutemov #ifdef CONFIG_DEBUG_FS 3890a9b0f861SKirill A. Shutemov static int fault_around_bytes_get(void *data, u64 *val) 38911592eef0SKirill A. Shutemov { 3892a9b0f861SKirill A. Shutemov *val = fault_around_bytes; 38931592eef0SKirill A. Shutemov return 0; 38941592eef0SKirill A. Shutemov } 38951592eef0SKirill A. Shutemov 3896b4903d6eSAndrey Ryabinin /* 3897da391d64SWilliam Kucharski * fault_around_bytes must be rounded down to the nearest page order as it's 3898da391d64SWilliam Kucharski * what do_fault_around() expects to see. 3899b4903d6eSAndrey Ryabinin */ 3900a9b0f861SKirill A. Shutemov static int fault_around_bytes_set(void *data, u64 val) 39011592eef0SKirill A. Shutemov { 3902a9b0f861SKirill A. Shutemov if (val / PAGE_SIZE > PTRS_PER_PTE) 39031592eef0SKirill A. Shutemov return -EINVAL; 3904b4903d6eSAndrey Ryabinin if (val > PAGE_SIZE) 3905b4903d6eSAndrey Ryabinin fault_around_bytes = rounddown_pow_of_two(val); 3906b4903d6eSAndrey Ryabinin else 3907b4903d6eSAndrey Ryabinin fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ 39081592eef0SKirill A. Shutemov return 0; 39091592eef0SKirill A. Shutemov } 39100a1345f8SYevgen Pronenko DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, 3911a9b0f861SKirill A. Shutemov fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); 39121592eef0SKirill A. Shutemov 39131592eef0SKirill A. Shutemov static int __init fault_around_debugfs(void) 39141592eef0SKirill A. Shutemov { 3915d9f7979cSGreg Kroah-Hartman debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, 3916a9b0f861SKirill A. Shutemov &fault_around_bytes_fops); 39171592eef0SKirill A. Shutemov return 0; 39181592eef0SKirill A. Shutemov } 39191592eef0SKirill A. Shutemov late_initcall(fault_around_debugfs); 39201592eef0SKirill A. Shutemov #endif 39218c6e50b0SKirill A. Shutemov 39221fdb412bSKirill A. Shutemov /* 39231fdb412bSKirill A. Shutemov * do_fault_around() tries to map few pages around the fault address. The hope 39241fdb412bSKirill A. Shutemov * is that the pages will be needed soon and this will lower the number of 39251fdb412bSKirill A. Shutemov * faults to handle. 39261fdb412bSKirill A. Shutemov * 39271fdb412bSKirill A. Shutemov * It uses vm_ops->map_pages() to map the pages, which skips the page if it's 39281fdb412bSKirill A. Shutemov * not ready to be mapped: not up-to-date, locked, etc. 39291fdb412bSKirill A. Shutemov * 39301fdb412bSKirill A. Shutemov * This function is called with the page table lock taken. In the split ptlock 39311fdb412bSKirill A. Shutemov * case the page table lock only protects only those entries which belong to 39321fdb412bSKirill A. Shutemov * the page table corresponding to the fault address. 39331fdb412bSKirill A. Shutemov * 39341fdb412bSKirill A. Shutemov * This function doesn't cross the VMA boundaries, in order to call map_pages() 39351fdb412bSKirill A. Shutemov * only once. 39361fdb412bSKirill A. Shutemov * 3937da391d64SWilliam Kucharski * fault_around_bytes defines how many bytes we'll try to map. 3938da391d64SWilliam Kucharski * do_fault_around() expects it to be set to a power of two less than or equal 3939da391d64SWilliam Kucharski * to PTRS_PER_PTE. 39401fdb412bSKirill A. Shutemov * 3941da391d64SWilliam Kucharski * The virtual address of the area that we map is naturally aligned to 3942da391d64SWilliam Kucharski * fault_around_bytes rounded down to the machine page size 3943da391d64SWilliam Kucharski * (and therefore to page order). This way it's easier to guarantee 3944da391d64SWilliam Kucharski * that we don't cross page table boundaries. 39451fdb412bSKirill A. Shutemov */ 39462b740303SSouptick Joarder static vm_fault_t do_fault_around(struct vm_fault *vmf) 39478c6e50b0SKirill A. Shutemov { 394882b0f8c3SJan Kara unsigned long address = vmf->address, nr_pages, mask; 39490721ec8bSJan Kara pgoff_t start_pgoff = vmf->pgoff; 3950bae473a4SKirill A. Shutemov pgoff_t end_pgoff; 39512b740303SSouptick Joarder int off; 39522b740303SSouptick Joarder vm_fault_t ret = 0; 39538c6e50b0SKirill A. Shutemov 39544db0c3c2SJason Low nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; 3955aecd6f44SKirill A. Shutemov mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; 3956aecd6f44SKirill A. Shutemov 395782b0f8c3SJan Kara vmf->address = max(address & mask, vmf->vma->vm_start); 395882b0f8c3SJan Kara off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); 3959bae473a4SKirill A. Shutemov start_pgoff -= off; 39608c6e50b0SKirill A. Shutemov 39618c6e50b0SKirill A. Shutemov /* 3962da391d64SWilliam Kucharski * end_pgoff is either the end of the page table, the end of 3963da391d64SWilliam Kucharski * the vma or nr_pages from start_pgoff, depending what is nearest. 39648c6e50b0SKirill A. Shutemov */ 3965bae473a4SKirill A. Shutemov end_pgoff = start_pgoff - 396682b0f8c3SJan Kara ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + 39678c6e50b0SKirill A. Shutemov PTRS_PER_PTE - 1; 396882b0f8c3SJan Kara end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, 3969bae473a4SKirill A. Shutemov start_pgoff + nr_pages - 1); 39708c6e50b0SKirill A. Shutemov 397182b0f8c3SJan Kara if (pmd_none(*vmf->pmd)) { 39724cf58924SJoel Fernandes (Google) vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); 397382b0f8c3SJan Kara if (!vmf->prealloc_pte) 3974c5f88bd2SVegard Nossum goto out; 39757267ec00SKirill A. Shutemov smp_wmb(); /* See comment in __pte_alloc() */ 39768c6e50b0SKirill A. Shutemov } 39778c6e50b0SKirill A. Shutemov 397882b0f8c3SJan Kara vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); 39797267ec00SKirill A. Shutemov 39807267ec00SKirill A. Shutemov /* Huge page is mapped? Page fault is solved */ 398182b0f8c3SJan Kara if (pmd_trans_huge(*vmf->pmd)) { 39827267ec00SKirill A. Shutemov ret = VM_FAULT_NOPAGE; 39837267ec00SKirill A. Shutemov goto out; 39848c6e50b0SKirill A. Shutemov } 39858c6e50b0SKirill A. Shutemov 39867267ec00SKirill A. Shutemov /* ->map_pages() haven't done anything useful. Cold page cache? */ 398782b0f8c3SJan Kara if (!vmf->pte) 39887267ec00SKirill A. Shutemov goto out; 39897267ec00SKirill A. Shutemov 39907267ec00SKirill A. Shutemov /* check if the page fault is solved */ 399182b0f8c3SJan Kara vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); 399282b0f8c3SJan Kara if (!pte_none(*vmf->pte)) 39937267ec00SKirill A. Shutemov ret = VM_FAULT_NOPAGE; 399482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 39957267ec00SKirill A. Shutemov out: 399682b0f8c3SJan Kara vmf->address = address; 399782b0f8c3SJan Kara vmf->pte = NULL; 39987267ec00SKirill A. Shutemov return ret; 39997267ec00SKirill A. Shutemov } 40007267ec00SKirill A. Shutemov 40012b740303SSouptick Joarder static vm_fault_t do_read_fault(struct vm_fault *vmf) 4002e655fb29SKirill A. Shutemov { 400382b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40042b740303SSouptick Joarder vm_fault_t ret = 0; 40058c6e50b0SKirill A. Shutemov 40068c6e50b0SKirill A. Shutemov /* 40078c6e50b0SKirill A. Shutemov * Let's call ->map_pages() first and use ->fault() as fallback 40088c6e50b0SKirill A. Shutemov * if page by the offset is not ready to be mapped (cold cache or 40098c6e50b0SKirill A. Shutemov * something). 40108c6e50b0SKirill A. Shutemov */ 40119b4bdd2fSKirill A. Shutemov if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { 40120721ec8bSJan Kara ret = do_fault_around(vmf); 40137267ec00SKirill A. Shutemov if (ret) 40147267ec00SKirill A. Shutemov return ret; 40158c6e50b0SKirill A. Shutemov } 4016e655fb29SKirill A. Shutemov 4017936ca80dSJan Kara ret = __do_fault(vmf); 4018e655fb29SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4019e655fb29SKirill A. Shutemov return ret; 4020e655fb29SKirill A. Shutemov 40219118c0cbSJan Kara ret |= finish_fault(vmf); 4022936ca80dSJan Kara unlock_page(vmf->page); 40237267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4024936ca80dSJan Kara put_page(vmf->page); 4025e655fb29SKirill A. Shutemov return ret; 4026e655fb29SKirill A. Shutemov } 4027e655fb29SKirill A. Shutemov 40282b740303SSouptick Joarder static vm_fault_t do_cow_fault(struct vm_fault *vmf) 4029ec47c3b9SKirill A. Shutemov { 403082b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40312b740303SSouptick Joarder vm_fault_t ret; 4032ec47c3b9SKirill A. Shutemov 4033ec47c3b9SKirill A. Shutemov if (unlikely(anon_vma_prepare(vma))) 4034ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4035ec47c3b9SKirill A. Shutemov 4036936ca80dSJan Kara vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); 4037936ca80dSJan Kara if (!vmf->cow_page) 4038ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4039ec47c3b9SKirill A. Shutemov 4040d9eb1ea2SJohannes Weiner if (mem_cgroup_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL)) { 4041936ca80dSJan Kara put_page(vmf->cow_page); 4042ec47c3b9SKirill A. Shutemov return VM_FAULT_OOM; 4043ec47c3b9SKirill A. Shutemov } 40449d82c694SJohannes Weiner cgroup_throttle_swaprate(vmf->cow_page, GFP_KERNEL); 4045ec47c3b9SKirill A. Shutemov 4046936ca80dSJan Kara ret = __do_fault(vmf); 4047ec47c3b9SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4048ec47c3b9SKirill A. Shutemov goto uncharge_out; 40493917048dSJan Kara if (ret & VM_FAULT_DONE_COW) 40503917048dSJan Kara return ret; 4051ec47c3b9SKirill A. Shutemov 4052936ca80dSJan Kara copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); 4053936ca80dSJan Kara __SetPageUptodate(vmf->cow_page); 4054ec47c3b9SKirill A. Shutemov 40559118c0cbSJan Kara ret |= finish_fault(vmf); 4056936ca80dSJan Kara unlock_page(vmf->page); 4057936ca80dSJan Kara put_page(vmf->page); 40587267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 40597267ec00SKirill A. Shutemov goto uncharge_out; 4060ec47c3b9SKirill A. Shutemov return ret; 4061ec47c3b9SKirill A. Shutemov uncharge_out: 4062936ca80dSJan Kara put_page(vmf->cow_page); 4063ec47c3b9SKirill A. Shutemov return ret; 4064ec47c3b9SKirill A. Shutemov } 4065ec47c3b9SKirill A. Shutemov 40662b740303SSouptick Joarder static vm_fault_t do_shared_fault(struct vm_fault *vmf) 40671da177e4SLinus Torvalds { 406882b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 40692b740303SSouptick Joarder vm_fault_t ret, tmp; 40701d65f86dSKAMEZAWA Hiroyuki 4071936ca80dSJan Kara ret = __do_fault(vmf); 40727eae74afSKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) 4073f0c6d4d2SKirill A. Shutemov return ret; 40741da177e4SLinus Torvalds 40751da177e4SLinus Torvalds /* 4076f0c6d4d2SKirill A. Shutemov * Check if the backing address space wants to know that the page is 4077f0c6d4d2SKirill A. Shutemov * about to become writable 40781da177e4SLinus Torvalds */ 4079fb09a464SKirill A. Shutemov if (vma->vm_ops->page_mkwrite) { 4080936ca80dSJan Kara unlock_page(vmf->page); 408138b8cb7fSJan Kara tmp = do_page_mkwrite(vmf); 4082fb09a464SKirill A. Shutemov if (unlikely(!tmp || 4083fb09a464SKirill A. Shutemov (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { 4084936ca80dSJan Kara put_page(vmf->page); 4085f0c6d4d2SKirill A. Shutemov return tmp; 408669676147SMark Fasheh } 4087d0217ac0SNick Piggin } 4088fb09a464SKirill A. Shutemov 40899118c0cbSJan Kara ret |= finish_fault(vmf); 40907267ec00SKirill A. Shutemov if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | 40917267ec00SKirill A. Shutemov VM_FAULT_RETRY))) { 4092936ca80dSJan Kara unlock_page(vmf->page); 4093936ca80dSJan Kara put_page(vmf->page); 4094f0c6d4d2SKirill A. Shutemov return ret; 40959637a5efSDavid Howells } 4096d00806b1SNick Piggin 409789b15332SJohannes Weiner ret |= fault_dirty_shared_page(vmf); 4098b827e496SNick Piggin return ret; 409954cb8821SNick Piggin } 4100d00806b1SNick Piggin 41019a95f3cfSPaul Cassella /* 4102c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, 41039a95f3cfSPaul Cassella * but allow concurrent faults). 4104c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 41059a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 4106c1e8d7c6SMichel Lespinasse * If mmap_lock is released, vma may become invalid (for example 4107fc8efd2dSJan Stancek * by other thread calling munmap()). 41089a95f3cfSPaul Cassella */ 41092b740303SSouptick Joarder static vm_fault_t do_fault(struct vm_fault *vmf) 411054cb8821SNick Piggin { 411182b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 4112fc8efd2dSJan Stancek struct mm_struct *vm_mm = vma->vm_mm; 41132b740303SSouptick Joarder vm_fault_t ret; 411454cb8821SNick Piggin 4115ff09d7ecSAneesh Kumar K.V /* 4116ff09d7ecSAneesh Kumar K.V * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND 4117ff09d7ecSAneesh Kumar K.V */ 4118ff09d7ecSAneesh Kumar K.V if (!vma->vm_ops->fault) { 4119ff09d7ecSAneesh Kumar K.V /* 4120ff09d7ecSAneesh Kumar K.V * If we find a migration pmd entry or a none pmd entry, which 4121ff09d7ecSAneesh Kumar K.V * should never happen, return SIGBUS 4122ff09d7ecSAneesh Kumar K.V */ 4123ff09d7ecSAneesh Kumar K.V if (unlikely(!pmd_present(*vmf->pmd))) 4124b0b9b3dfSHugh Dickins ret = VM_FAULT_SIGBUS; 4125ff09d7ecSAneesh Kumar K.V else { 4126ff09d7ecSAneesh Kumar K.V vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, 4127ff09d7ecSAneesh Kumar K.V vmf->pmd, 4128ff09d7ecSAneesh Kumar K.V vmf->address, 4129ff09d7ecSAneesh Kumar K.V &vmf->ptl); 4130ff09d7ecSAneesh Kumar K.V /* 4131ff09d7ecSAneesh Kumar K.V * Make sure this is not a temporary clearing of pte 4132ff09d7ecSAneesh Kumar K.V * by holding ptl and checking again. A R/M/W update 4133ff09d7ecSAneesh Kumar K.V * of pte involves: take ptl, clearing the pte so that 4134ff09d7ecSAneesh Kumar K.V * we don't have concurrent modification by hardware 4135ff09d7ecSAneesh Kumar K.V * followed by an update. 4136ff09d7ecSAneesh Kumar K.V */ 4137ff09d7ecSAneesh Kumar K.V if (unlikely(pte_none(*vmf->pte))) 4138ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_SIGBUS; 4139ff09d7ecSAneesh Kumar K.V else 4140ff09d7ecSAneesh Kumar K.V ret = VM_FAULT_NOPAGE; 4141ff09d7ecSAneesh Kumar K.V 4142ff09d7ecSAneesh Kumar K.V pte_unmap_unlock(vmf->pte, vmf->ptl); 4143ff09d7ecSAneesh Kumar K.V } 4144ff09d7ecSAneesh Kumar K.V } else if (!(vmf->flags & FAULT_FLAG_WRITE)) 4145b0b9b3dfSHugh Dickins ret = do_read_fault(vmf); 4146b0b9b3dfSHugh Dickins else if (!(vma->vm_flags & VM_SHARED)) 4147b0b9b3dfSHugh Dickins ret = do_cow_fault(vmf); 4148b0b9b3dfSHugh Dickins else 4149b0b9b3dfSHugh Dickins ret = do_shared_fault(vmf); 4150b0b9b3dfSHugh Dickins 4151b0b9b3dfSHugh Dickins /* preallocated pagetable is unused: free it */ 4152b0b9b3dfSHugh Dickins if (vmf->prealloc_pte) { 4153fc8efd2dSJan Stancek pte_free(vm_mm, vmf->prealloc_pte); 41547f2b6ce8STobin C Harding vmf->prealloc_pte = NULL; 4155b0b9b3dfSHugh Dickins } 4156b0b9b3dfSHugh Dickins return ret; 415754cb8821SNick Piggin } 415854cb8821SNick Piggin 4159b19a9939SRashika Kheria static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, 416004bb2f94SRik van Riel unsigned long addr, int page_nid, 416104bb2f94SRik van Riel int *flags) 41629532fec1SMel Gorman { 41639532fec1SMel Gorman get_page(page); 41649532fec1SMel Gorman 41659532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS); 416604bb2f94SRik van Riel if (page_nid == numa_node_id()) { 41679532fec1SMel Gorman count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); 416804bb2f94SRik van Riel *flags |= TNF_FAULT_LOCAL; 416904bb2f94SRik van Riel } 41709532fec1SMel Gorman 41719532fec1SMel Gorman return mpol_misplaced(page, vma, addr); 41729532fec1SMel Gorman } 41739532fec1SMel Gorman 41742b740303SSouptick Joarder static vm_fault_t do_numa_page(struct vm_fault *vmf) 4175d10e63f2SMel Gorman { 417682b0f8c3SJan Kara struct vm_area_struct *vma = vmf->vma; 41774daae3b4SMel Gorman struct page *page = NULL; 417898fa15f3SAnshuman Khandual int page_nid = NUMA_NO_NODE; 417990572890SPeter Zijlstra int last_cpupid; 4180cbee9f88SPeter Zijlstra int target_nid; 4181b8593bfdSMel Gorman bool migrated = false; 418204a86453SAneesh Kumar K.V pte_t pte, old_pte; 4183288bc549SAneesh Kumar K.V bool was_writable = pte_savedwrite(vmf->orig_pte); 41846688cc05SPeter Zijlstra int flags = 0; 4185d10e63f2SMel Gorman 4186d10e63f2SMel Gorman /* 4187d10e63f2SMel Gorman * The "pte" at this point cannot be used safely without 4188d10e63f2SMel Gorman * validation through pte_unmap_same(). It's of NUMA type but 4189d10e63f2SMel Gorman * the pfn may be screwed if the read is non atomic. 4190d10e63f2SMel Gorman */ 419182b0f8c3SJan Kara vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); 419282b0f8c3SJan Kara spin_lock(vmf->ptl); 4193cee216a6SAneesh Kumar K.V if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { 419482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 41954daae3b4SMel Gorman goto out; 41964daae3b4SMel Gorman } 41974daae3b4SMel Gorman 4198cee216a6SAneesh Kumar K.V /* 4199cee216a6SAneesh Kumar K.V * Make it present again, Depending on how arch implementes non 4200cee216a6SAneesh Kumar K.V * accessible ptes, some can allow access by kernel mode. 4201cee216a6SAneesh Kumar K.V */ 420204a86453SAneesh Kumar K.V old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); 420304a86453SAneesh Kumar K.V pte = pte_modify(old_pte, vma->vm_page_prot); 42044d942466SMel Gorman pte = pte_mkyoung(pte); 4205b191f9b1SMel Gorman if (was_writable) 4206b191f9b1SMel Gorman pte = pte_mkwrite(pte); 420704a86453SAneesh Kumar K.V ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); 420882b0f8c3SJan Kara update_mmu_cache(vma, vmf->address, vmf->pte); 4209d10e63f2SMel Gorman 421082b0f8c3SJan Kara page = vm_normal_page(vma, vmf->address, pte); 4211d10e63f2SMel Gorman if (!page) { 421282b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4213d10e63f2SMel Gorman return 0; 4214d10e63f2SMel Gorman } 4215d10e63f2SMel Gorman 4216e81c4802SKirill A. Shutemov /* TODO: handle PTE-mapped THP */ 4217e81c4802SKirill A. Shutemov if (PageCompound(page)) { 421882b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 4219e81c4802SKirill A. Shutemov return 0; 4220e81c4802SKirill A. Shutemov } 4221e81c4802SKirill A. Shutemov 42226688cc05SPeter Zijlstra /* 4223bea66fbdSMel Gorman * Avoid grouping on RO pages in general. RO pages shouldn't hurt as 4224bea66fbdSMel Gorman * much anyway since they can be in shared cache state. This misses 4225bea66fbdSMel Gorman * the case where a mapping is writable but the process never writes 4226bea66fbdSMel Gorman * to it but pte_write gets cleared during protection updates and 4227bea66fbdSMel Gorman * pte_dirty has unpredictable behaviour between PTE scan updates, 4228bea66fbdSMel Gorman * background writeback, dirty balancing and application behaviour. 42296688cc05SPeter Zijlstra */ 4230d59dc7bcSRik van Riel if (!pte_write(pte)) 42316688cc05SPeter Zijlstra flags |= TNF_NO_GROUP; 42326688cc05SPeter Zijlstra 4233dabe1d99SRik van Riel /* 4234dabe1d99SRik van Riel * Flag if the page is shared between multiple address spaces. This 4235dabe1d99SRik van Riel * is later used when determining whether to group tasks together 4236dabe1d99SRik van Riel */ 4237dabe1d99SRik van Riel if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) 4238dabe1d99SRik van Riel flags |= TNF_SHARED; 4239dabe1d99SRik van Riel 424090572890SPeter Zijlstra last_cpupid = page_cpupid_last(page); 42418191acbdSMel Gorman page_nid = page_to_nid(page); 424282b0f8c3SJan Kara target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, 4243bae473a4SKirill A. Shutemov &flags); 424482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 424598fa15f3SAnshuman Khandual if (target_nid == NUMA_NO_NODE) { 42464daae3b4SMel Gorman put_page(page); 42474daae3b4SMel Gorman goto out; 42484daae3b4SMel Gorman } 42494daae3b4SMel Gorman 42504daae3b4SMel Gorman /* Migrate to the requested node */ 42511bc115d8SMel Gorman migrated = migrate_misplaced_page(page, vma, target_nid); 42526688cc05SPeter Zijlstra if (migrated) { 42538191acbdSMel Gorman page_nid = target_nid; 42546688cc05SPeter Zijlstra flags |= TNF_MIGRATED; 4255074c2381SMel Gorman } else 4256074c2381SMel Gorman flags |= TNF_MIGRATE_FAIL; 42574daae3b4SMel Gorman 42584daae3b4SMel Gorman out: 425998fa15f3SAnshuman Khandual if (page_nid != NUMA_NO_NODE) 42606688cc05SPeter Zijlstra task_numa_fault(last_cpupid, page_nid, 1, flags); 4261d10e63f2SMel Gorman return 0; 4262d10e63f2SMel Gorman } 4263d10e63f2SMel Gorman 42642b740303SSouptick Joarder static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) 4265b96375f7SMatthew Wilcox { 4266f4200391SDave Jiang if (vma_is_anonymous(vmf->vma)) 426782b0f8c3SJan Kara return do_huge_pmd_anonymous_page(vmf); 4268a2d58167SDave Jiang if (vmf->vma->vm_ops->huge_fault) 4269c791ace1SDave Jiang return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4270b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4271b96375f7SMatthew Wilcox } 4272b96375f7SMatthew Wilcox 4273183f24aaSGeert Uytterhoeven /* `inline' is required to avoid gcc 4.1.2 build error */ 42742b740303SSouptick Joarder static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) 4275b96375f7SMatthew Wilcox { 4276529b930bSAndrea Arcangeli if (vma_is_anonymous(vmf->vma)) { 4277292924b2SPeter Xu if (userfaultfd_huge_pmd_wp(vmf->vma, orig_pmd)) 4278529b930bSAndrea Arcangeli return handle_userfault(vmf, VM_UFFD_WP); 427982b0f8c3SJan Kara return do_huge_pmd_wp_page(vmf, orig_pmd); 4280529b930bSAndrea Arcangeli } 4281327e9fd4SThomas Hellstrom (VMware) if (vmf->vma->vm_ops->huge_fault) { 4282327e9fd4SThomas Hellstrom (VMware) vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); 4283af9e4d5fSKirill A. Shutemov 4284327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4285327e9fd4SThomas Hellstrom (VMware) return ret; 4286327e9fd4SThomas Hellstrom (VMware) } 4287327e9fd4SThomas Hellstrom (VMware) 4288327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify handled on pte level: split pmd. */ 428982b0f8c3SJan Kara __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); 4290af9e4d5fSKirill A. Shutemov 4291b96375f7SMatthew Wilcox return VM_FAULT_FALLBACK; 4292b96375f7SMatthew Wilcox } 4293b96375f7SMatthew Wilcox 42942b740303SSouptick Joarder static vm_fault_t create_huge_pud(struct vm_fault *vmf) 4295a00cc7d9SMatthew Wilcox { 4296327e9fd4SThomas Hellstrom (VMware) #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \ 4297327e9fd4SThomas Hellstrom (VMware) defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 4298a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 4299a00cc7d9SMatthew Wilcox if (vma_is_anonymous(vmf->vma)) 4300327e9fd4SThomas Hellstrom (VMware) goto split; 4301327e9fd4SThomas Hellstrom (VMware) if (vmf->vma->vm_ops->huge_fault) { 4302327e9fd4SThomas Hellstrom (VMware) vm_fault_t ret = vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 4303327e9fd4SThomas Hellstrom (VMware) 4304327e9fd4SThomas Hellstrom (VMware) if (!(ret & VM_FAULT_FALLBACK)) 4305327e9fd4SThomas Hellstrom (VMware) return ret; 4306327e9fd4SThomas Hellstrom (VMware) } 4307327e9fd4SThomas Hellstrom (VMware) split: 4308327e9fd4SThomas Hellstrom (VMware) /* COW or write-notify not handled on PUD level: split pud.*/ 4309327e9fd4SThomas Hellstrom (VMware) __split_huge_pud(vmf->vma, vmf->pud, vmf->address); 4310a00cc7d9SMatthew Wilcox #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 4311a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4312a00cc7d9SMatthew Wilcox } 4313a00cc7d9SMatthew Wilcox 43142b740303SSouptick Joarder static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) 4315a00cc7d9SMatthew Wilcox { 4316a00cc7d9SMatthew Wilcox #ifdef CONFIG_TRANSPARENT_HUGEPAGE 4317a00cc7d9SMatthew Wilcox /* No support for anonymous transparent PUD pages yet */ 4318a00cc7d9SMatthew Wilcox if (vma_is_anonymous(vmf->vma)) 4319a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4320a00cc7d9SMatthew Wilcox if (vmf->vma->vm_ops->huge_fault) 4321c791ace1SDave Jiang return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); 4322a00cc7d9SMatthew Wilcox #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 4323a00cc7d9SMatthew Wilcox return VM_FAULT_FALLBACK; 4324a00cc7d9SMatthew Wilcox } 4325a00cc7d9SMatthew Wilcox 43261da177e4SLinus Torvalds /* 43271da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 43281da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 43291da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 43301da177e4SLinus Torvalds * 43311da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 43321da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 43331da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 43341da177e4SLinus Torvalds * 4335c1e8d7c6SMichel Lespinasse * We enter with non-exclusive mmap_lock (to exclude vma changes, but allow 43367267ec00SKirill A. Shutemov * concurrent faults). 43379a95f3cfSPaul Cassella * 4338c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our return value. 43397267ec00SKirill A. Shutemov * See filemap_fault() and __lock_page_or_retry(). 43401da177e4SLinus Torvalds */ 43412b740303SSouptick Joarder static vm_fault_t handle_pte_fault(struct vm_fault *vmf) 43421da177e4SLinus Torvalds { 43431da177e4SLinus Torvalds pte_t entry; 43441da177e4SLinus Torvalds 434582b0f8c3SJan Kara if (unlikely(pmd_none(*vmf->pmd))) { 43467267ec00SKirill A. Shutemov /* 43477267ec00SKirill A. Shutemov * Leave __pte_alloc() until later: because vm_ops->fault may 43487267ec00SKirill A. Shutemov * want to allocate huge page, and if we expose page table 43497267ec00SKirill A. Shutemov * for an instant, it will be difficult to retract from 43507267ec00SKirill A. Shutemov * concurrent faults and from rmap lookups. 43517267ec00SKirill A. Shutemov */ 435282b0f8c3SJan Kara vmf->pte = NULL; 43537267ec00SKirill A. Shutemov } else { 43547267ec00SKirill A. Shutemov /* See comment in pte_alloc_one_map() */ 4355d0f0931dSRoss Zwisler if (pmd_devmap_trans_unstable(vmf->pmd)) 43567267ec00SKirill A. Shutemov return 0; 43577267ec00SKirill A. Shutemov /* 43587267ec00SKirill A. Shutemov * A regular pmd is established and it can't morph into a huge 43597267ec00SKirill A. Shutemov * pmd from under us anymore at this point because we hold the 4360c1e8d7c6SMichel Lespinasse * mmap_lock read mode and khugepaged takes it in write mode. 43617267ec00SKirill A. Shutemov * So now it's safe to run pte_offset_map(). 43627267ec00SKirill A. Shutemov */ 436382b0f8c3SJan Kara vmf->pte = pte_offset_map(vmf->pmd, vmf->address); 43642994302bSJan Kara vmf->orig_pte = *vmf->pte; 43657267ec00SKirill A. Shutemov 4366e37c6982SChristian Borntraeger /* 4367e37c6982SChristian Borntraeger * some architectures can have larger ptes than wordsize, 43687267ec00SKirill A. Shutemov * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and 4369b03a0fe0SPaul E. McKenney * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic 4370b03a0fe0SPaul E. McKenney * accesses. The code below just needs a consistent view 4371b03a0fe0SPaul E. McKenney * for the ifs and we later double check anyway with the 43727267ec00SKirill A. Shutemov * ptl lock held. So here a barrier will do. 4373e37c6982SChristian Borntraeger */ 4374e37c6982SChristian Borntraeger barrier(); 43752994302bSJan Kara if (pte_none(vmf->orig_pte)) { 437682b0f8c3SJan Kara pte_unmap(vmf->pte); 437782b0f8c3SJan Kara vmf->pte = NULL; 43787267ec00SKirill A. Shutemov } 43797267ec00SKirill A. Shutemov } 43807267ec00SKirill A. Shutemov 438182b0f8c3SJan Kara if (!vmf->pte) { 438282b0f8c3SJan Kara if (vma_is_anonymous(vmf->vma)) 438382b0f8c3SJan Kara return do_anonymous_page(vmf); 4384b5330628SOleg Nesterov else 438582b0f8c3SJan Kara return do_fault(vmf); 438665500d23SHugh Dickins } 43877267ec00SKirill A. Shutemov 43882994302bSJan Kara if (!pte_present(vmf->orig_pte)) 43892994302bSJan Kara return do_swap_page(vmf); 43901da177e4SLinus Torvalds 43912994302bSJan Kara if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) 43922994302bSJan Kara return do_numa_page(vmf); 4393d10e63f2SMel Gorman 439482b0f8c3SJan Kara vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); 439582b0f8c3SJan Kara spin_lock(vmf->ptl); 43962994302bSJan Kara entry = vmf->orig_pte; 43977df67697SBibo Mao if (unlikely(!pte_same(*vmf->pte, entry))) { 43987df67697SBibo Mao update_mmu_tlb(vmf->vma, vmf->address, vmf->pte); 43998f4e2101SHugh Dickins goto unlock; 44007df67697SBibo Mao } 440182b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) { 4402f6f37321SLinus Torvalds if (!pte_write(entry)) 44032994302bSJan Kara return do_wp_page(vmf); 44041da177e4SLinus Torvalds entry = pte_mkdirty(entry); 44051da177e4SLinus Torvalds } 44061da177e4SLinus Torvalds entry = pte_mkyoung(entry); 440782b0f8c3SJan Kara if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, 440882b0f8c3SJan Kara vmf->flags & FAULT_FLAG_WRITE)) { 440982b0f8c3SJan Kara update_mmu_cache(vmf->vma, vmf->address, vmf->pte); 44101a44e149SAndrea Arcangeli } else { 4411b7333b58SYang Shi /* Skip spurious TLB flush for retried page fault */ 4412b7333b58SYang Shi if (vmf->flags & FAULT_FLAG_TRIED) 4413b7333b58SYang Shi goto unlock; 44141a44e149SAndrea Arcangeli /* 44151a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 44161a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 44171a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 44181a44e149SAndrea Arcangeli * with threads. 44191a44e149SAndrea Arcangeli */ 442082b0f8c3SJan Kara if (vmf->flags & FAULT_FLAG_WRITE) 442182b0f8c3SJan Kara flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); 44221a44e149SAndrea Arcangeli } 44238f4e2101SHugh Dickins unlock: 442482b0f8c3SJan Kara pte_unmap_unlock(vmf->pte, vmf->ptl); 442583c54070SNick Piggin return 0; 44261da177e4SLinus Torvalds } 44271da177e4SLinus Torvalds 44281da177e4SLinus Torvalds /* 44291da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 44309a95f3cfSPaul Cassella * 4431c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 44329a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 44331da177e4SLinus Torvalds */ 44342b740303SSouptick Joarder static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, 44352b740303SSouptick Joarder unsigned long address, unsigned int flags) 44361da177e4SLinus Torvalds { 443782b0f8c3SJan Kara struct vm_fault vmf = { 4438bae473a4SKirill A. Shutemov .vma = vma, 44391a29d85eSJan Kara .address = address & PAGE_MASK, 4440bae473a4SKirill A. Shutemov .flags = flags, 44410721ec8bSJan Kara .pgoff = linear_page_index(vma, address), 4442667240e0SJan Kara .gfp_mask = __get_fault_gfp_mask(vma), 4443bae473a4SKirill A. Shutemov }; 4444fde26bedSAnshuman Khandual unsigned int dirty = flags & FAULT_FLAG_WRITE; 4445dcddffd4SKirill A. Shutemov struct mm_struct *mm = vma->vm_mm; 44461da177e4SLinus Torvalds pgd_t *pgd; 4447c2febafcSKirill A. Shutemov p4d_t *p4d; 44482b740303SSouptick Joarder vm_fault_t ret; 44491da177e4SLinus Torvalds 44501da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 4451c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, address); 4452c2febafcSKirill A. Shutemov if (!p4d) 4453c2febafcSKirill A. Shutemov return VM_FAULT_OOM; 4454a00cc7d9SMatthew Wilcox 4455c2febafcSKirill A. Shutemov vmf.pud = pud_alloc(mm, p4d, address); 4456a00cc7d9SMatthew Wilcox if (!vmf.pud) 4457c74df32cSHugh Dickins return VM_FAULT_OOM; 4458625110b5SThomas Hellstrom retry_pud: 44597635d9cbSMichal Hocko if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) { 4460a00cc7d9SMatthew Wilcox ret = create_huge_pud(&vmf); 4461a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 4462a00cc7d9SMatthew Wilcox return ret; 4463a00cc7d9SMatthew Wilcox } else { 4464a00cc7d9SMatthew Wilcox pud_t orig_pud = *vmf.pud; 4465a00cc7d9SMatthew Wilcox 4466a00cc7d9SMatthew Wilcox barrier(); 4467a00cc7d9SMatthew Wilcox if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { 4468a00cc7d9SMatthew Wilcox 4469a00cc7d9SMatthew Wilcox /* NUMA case for anonymous PUDs would go here */ 4470a00cc7d9SMatthew Wilcox 4471f6f37321SLinus Torvalds if (dirty && !pud_write(orig_pud)) { 4472a00cc7d9SMatthew Wilcox ret = wp_huge_pud(&vmf, orig_pud); 4473a00cc7d9SMatthew Wilcox if (!(ret & VM_FAULT_FALLBACK)) 4474a00cc7d9SMatthew Wilcox return ret; 4475a00cc7d9SMatthew Wilcox } else { 4476a00cc7d9SMatthew Wilcox huge_pud_set_accessed(&vmf, orig_pud); 4477a00cc7d9SMatthew Wilcox return 0; 4478a00cc7d9SMatthew Wilcox } 4479a00cc7d9SMatthew Wilcox } 4480a00cc7d9SMatthew Wilcox } 4481a00cc7d9SMatthew Wilcox 4482a00cc7d9SMatthew Wilcox vmf.pmd = pmd_alloc(mm, vmf.pud, address); 448382b0f8c3SJan Kara if (!vmf.pmd) 4484c74df32cSHugh Dickins return VM_FAULT_OOM; 4485625110b5SThomas Hellstrom 4486625110b5SThomas Hellstrom /* Huge pud page fault raced with pmd_alloc? */ 4487625110b5SThomas Hellstrom if (pud_trans_unstable(vmf.pud)) 4488625110b5SThomas Hellstrom goto retry_pud; 4489625110b5SThomas Hellstrom 44907635d9cbSMichal Hocko if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) { 4491a2d58167SDave Jiang ret = create_huge_pmd(&vmf); 4492c0292554SKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 4493c0292554SKirill A. Shutemov return ret; 449471e3aac0SAndrea Arcangeli } else { 449582b0f8c3SJan Kara pmd_t orig_pmd = *vmf.pmd; 44961f1d06c3SDavid Rientjes 449771e3aac0SAndrea Arcangeli barrier(); 449884c3fc4eSZi Yan if (unlikely(is_swap_pmd(orig_pmd))) { 449984c3fc4eSZi Yan VM_BUG_ON(thp_migration_supported() && 450084c3fc4eSZi Yan !is_pmd_migration_entry(orig_pmd)); 450184c3fc4eSZi Yan if (is_pmd_migration_entry(orig_pmd)) 450284c3fc4eSZi Yan pmd_migration_entry_wait(mm, vmf.pmd); 450384c3fc4eSZi Yan return 0; 450484c3fc4eSZi Yan } 45055c7fb56eSDan Williams if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { 450638e08854SLorenzo Stoakes if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) 450782b0f8c3SJan Kara return do_huge_pmd_numa_page(&vmf, orig_pmd); 4508d10e63f2SMel Gorman 4509f6f37321SLinus Torvalds if (dirty && !pmd_write(orig_pmd)) { 451082b0f8c3SJan Kara ret = wp_huge_pmd(&vmf, orig_pmd); 45119845cbbdSKirill A. Shutemov if (!(ret & VM_FAULT_FALLBACK)) 45121f1d06c3SDavid Rientjes return ret; 4513a1dd450bSWill Deacon } else { 451482b0f8c3SJan Kara huge_pmd_set_accessed(&vmf, orig_pmd); 451571e3aac0SAndrea Arcangeli return 0; 451671e3aac0SAndrea Arcangeli } 451771e3aac0SAndrea Arcangeli } 45189845cbbdSKirill A. Shutemov } 451971e3aac0SAndrea Arcangeli 452082b0f8c3SJan Kara return handle_pte_fault(&vmf); 45211da177e4SLinus Torvalds } 45221da177e4SLinus Torvalds 4523bce617edSPeter Xu /** 4524bce617edSPeter Xu * mm_account_fault - Do page fault accountings 4525bce617edSPeter Xu * 4526bce617edSPeter Xu * @regs: the pt_regs struct pointer. When set to NULL, will skip accounting 4527bce617edSPeter Xu * of perf event counters, but we'll still do the per-task accounting to 4528bce617edSPeter Xu * the task who triggered this page fault. 4529bce617edSPeter Xu * @address: the faulted address. 4530bce617edSPeter Xu * @flags: the fault flags. 4531bce617edSPeter Xu * @ret: the fault retcode. 4532bce617edSPeter Xu * 4533bce617edSPeter Xu * This will take care of most of the page fault accountings. Meanwhile, it 4534bce617edSPeter Xu * will also include the PERF_COUNT_SW_PAGE_FAULTS_[MAJ|MIN] perf counter 4535bce617edSPeter Xu * updates. However note that the handling of PERF_COUNT_SW_PAGE_FAULTS should 4536bce617edSPeter Xu * still be in per-arch page fault handlers at the entry of page fault. 4537bce617edSPeter Xu */ 4538bce617edSPeter Xu static inline void mm_account_fault(struct pt_regs *regs, 4539bce617edSPeter Xu unsigned long address, unsigned int flags, 4540bce617edSPeter Xu vm_fault_t ret) 4541bce617edSPeter Xu { 4542bce617edSPeter Xu bool major; 4543bce617edSPeter Xu 4544bce617edSPeter Xu /* 4545bce617edSPeter Xu * We don't do accounting for some specific faults: 4546bce617edSPeter Xu * 4547bce617edSPeter Xu * - Unsuccessful faults (e.g. when the address wasn't valid). That 4548bce617edSPeter Xu * includes arch_vma_access_permitted() failing before reaching here. 4549bce617edSPeter Xu * So this is not a "this many hardware page faults" counter. We 4550bce617edSPeter Xu * should use the hw profiling for that. 4551bce617edSPeter Xu * 4552bce617edSPeter Xu * - Incomplete faults (VM_FAULT_RETRY). They will only be counted 4553bce617edSPeter Xu * once they're completed. 4554bce617edSPeter Xu */ 4555bce617edSPeter Xu if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY)) 4556bce617edSPeter Xu return; 4557bce617edSPeter Xu 4558bce617edSPeter Xu /* 4559bce617edSPeter Xu * We define the fault as a major fault when the final successful fault 4560bce617edSPeter Xu * is VM_FAULT_MAJOR, or if it retried (which implies that we couldn't 4561bce617edSPeter Xu * handle it immediately previously). 4562bce617edSPeter Xu */ 4563bce617edSPeter Xu major = (ret & VM_FAULT_MAJOR) || (flags & FAULT_FLAG_TRIED); 4564bce617edSPeter Xu 4565a2beb5f1SPeter Xu if (major) 4566a2beb5f1SPeter Xu current->maj_flt++; 4567a2beb5f1SPeter Xu else 4568a2beb5f1SPeter Xu current->min_flt++; 4569a2beb5f1SPeter Xu 4570bce617edSPeter Xu /* 4571a2beb5f1SPeter Xu * If the fault is done for GUP, regs will be NULL. We only do the 4572a2beb5f1SPeter Xu * accounting for the per thread fault counters who triggered the 4573a2beb5f1SPeter Xu * fault, and we skip the perf event updates. 4574bce617edSPeter Xu */ 4575bce617edSPeter Xu if (!regs) 4576bce617edSPeter Xu return; 4577bce617edSPeter Xu 4578a2beb5f1SPeter Xu if (major) 4579bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 4580a2beb5f1SPeter Xu else 4581bce617edSPeter Xu perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 4582bce617edSPeter Xu } 4583bce617edSPeter Xu 45849a95f3cfSPaul Cassella /* 45859a95f3cfSPaul Cassella * By the time we get here, we already hold the mm semaphore 45869a95f3cfSPaul Cassella * 4587c1e8d7c6SMichel Lespinasse * The mmap_lock may have been released depending on flags and our 45889a95f3cfSPaul Cassella * return value. See filemap_fault() and __lock_page_or_retry(). 45899a95f3cfSPaul Cassella */ 45902b740303SSouptick Joarder vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, 4591bce617edSPeter Xu unsigned int flags, struct pt_regs *regs) 4592519e5247SJohannes Weiner { 45932b740303SSouptick Joarder vm_fault_t ret; 4594519e5247SJohannes Weiner 4595519e5247SJohannes Weiner __set_current_state(TASK_RUNNING); 4596519e5247SJohannes Weiner 4597519e5247SJohannes Weiner count_vm_event(PGFAULT); 45982262185cSRoman Gushchin count_memcg_event_mm(vma->vm_mm, PGFAULT); 4599519e5247SJohannes Weiner 4600519e5247SJohannes Weiner /* do counter updates before entering really critical section. */ 4601519e5247SJohannes Weiner check_sync_rss_stat(current); 4602519e5247SJohannes Weiner 4603de0c799bSLaurent Dufour if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, 4604de0c799bSLaurent Dufour flags & FAULT_FLAG_INSTRUCTION, 4605de0c799bSLaurent Dufour flags & FAULT_FLAG_REMOTE)) 4606de0c799bSLaurent Dufour return VM_FAULT_SIGSEGV; 4607de0c799bSLaurent Dufour 4608519e5247SJohannes Weiner /* 4609519e5247SJohannes Weiner * Enable the memcg OOM handling for faults triggered in user 4610519e5247SJohannes Weiner * space. Kernel faults are handled more gracefully. 4611519e5247SJohannes Weiner */ 4612519e5247SJohannes Weiner if (flags & FAULT_FLAG_USER) 461329ef680aSMichal Hocko mem_cgroup_enter_user_fault(); 4614519e5247SJohannes Weiner 4615bae473a4SKirill A. Shutemov if (unlikely(is_vm_hugetlb_page(vma))) 4616bae473a4SKirill A. Shutemov ret = hugetlb_fault(vma->vm_mm, vma, address, flags); 4617bae473a4SKirill A. Shutemov else 4618dcddffd4SKirill A. Shutemov ret = __handle_mm_fault(vma, address, flags); 4619519e5247SJohannes Weiner 462049426420SJohannes Weiner if (flags & FAULT_FLAG_USER) { 462129ef680aSMichal Hocko mem_cgroup_exit_user_fault(); 462249426420SJohannes Weiner /* 462349426420SJohannes Weiner * The task may have entered a memcg OOM situation but 462449426420SJohannes Weiner * if the allocation error was handled gracefully (no 462549426420SJohannes Weiner * VM_FAULT_OOM), there is no need to kill anything. 462649426420SJohannes Weiner * Just clean up the OOM state peacefully. 462749426420SJohannes Weiner */ 462849426420SJohannes Weiner if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) 462949426420SJohannes Weiner mem_cgroup_oom_synchronize(false); 463049426420SJohannes Weiner } 46313812c8c8SJohannes Weiner 4632bce617edSPeter Xu mm_account_fault(regs, address, flags, ret); 4633bce617edSPeter Xu 4634519e5247SJohannes Weiner return ret; 4635519e5247SJohannes Weiner } 4636e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(handle_mm_fault); 4637519e5247SJohannes Weiner 463890eceff1SKirill A. Shutemov #ifndef __PAGETABLE_P4D_FOLDED 463990eceff1SKirill A. Shutemov /* 464090eceff1SKirill A. Shutemov * Allocate p4d page table. 464190eceff1SKirill A. Shutemov * We've already handled the fast-path in-line. 464290eceff1SKirill A. Shutemov */ 464390eceff1SKirill A. Shutemov int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 464490eceff1SKirill A. Shutemov { 464590eceff1SKirill A. Shutemov p4d_t *new = p4d_alloc_one(mm, address); 464690eceff1SKirill A. Shutemov if (!new) 464790eceff1SKirill A. Shutemov return -ENOMEM; 464890eceff1SKirill A. Shutemov 464990eceff1SKirill A. Shutemov smp_wmb(); /* See comment in __pte_alloc */ 465090eceff1SKirill A. Shutemov 465190eceff1SKirill A. Shutemov spin_lock(&mm->page_table_lock); 465290eceff1SKirill A. Shutemov if (pgd_present(*pgd)) /* Another has populated it */ 465390eceff1SKirill A. Shutemov p4d_free(mm, new); 465490eceff1SKirill A. Shutemov else 465590eceff1SKirill A. Shutemov pgd_populate(mm, pgd, new); 465690eceff1SKirill A. Shutemov spin_unlock(&mm->page_table_lock); 465790eceff1SKirill A. Shutemov return 0; 465890eceff1SKirill A. Shutemov } 465990eceff1SKirill A. Shutemov #endif /* __PAGETABLE_P4D_FOLDED */ 466090eceff1SKirill A. Shutemov 46611da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 46621da177e4SLinus Torvalds /* 46631da177e4SLinus Torvalds * Allocate page upper directory. 4664872fec16SHugh Dickins * We've already handled the fast-path in-line. 46651da177e4SLinus Torvalds */ 4666c2febafcSKirill A. Shutemov int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) 46671da177e4SLinus Torvalds { 4668c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 4669c74df32cSHugh Dickins if (!new) 46701bb3630eSHugh Dickins return -ENOMEM; 46711da177e4SLinus Torvalds 4672362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 4673362a61adSNick Piggin 4674872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 4675b4e98d9aSKirill A. Shutemov if (!p4d_present(*p4d)) { 4676b4e98d9aSKirill A. Shutemov mm_inc_nr_puds(mm); 4677c2febafcSKirill A. Shutemov p4d_populate(mm, p4d, new); 4678b4e98d9aSKirill A. Shutemov } else /* Another has populated it */ 4679c2febafcSKirill A. Shutemov pud_free(mm, new); 4680872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 46811bb3630eSHugh Dickins return 0; 46821da177e4SLinus Torvalds } 46831da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 46841da177e4SLinus Torvalds 46851da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 46861da177e4SLinus Torvalds /* 46871da177e4SLinus Torvalds * Allocate page middle directory. 4688872fec16SHugh Dickins * We've already handled the fast-path in-line. 46891da177e4SLinus Torvalds */ 46901bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 46911da177e4SLinus Torvalds { 4692a00cc7d9SMatthew Wilcox spinlock_t *ptl; 4693c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 4694c74df32cSHugh Dickins if (!new) 46951bb3630eSHugh Dickins return -ENOMEM; 46961da177e4SLinus Torvalds 4697362a61adSNick Piggin smp_wmb(); /* See comment in __pte_alloc */ 4698362a61adSNick Piggin 4699a00cc7d9SMatthew Wilcox ptl = pud_lock(mm, pud); 4700dc6c9a35SKirill A. Shutemov if (!pud_present(*pud)) { 4701dc6c9a35SKirill A. Shutemov mm_inc_nr_pmds(mm); 47021da177e4SLinus Torvalds pud_populate(mm, pud, new); 4703dc6c9a35SKirill A. Shutemov } else /* Another has populated it */ 47045e541973SBenjamin Herrenschmidt pmd_free(mm, new); 4705a00cc7d9SMatthew Wilcox spin_unlock(ptl); 47061bb3630eSHugh Dickins return 0; 47071da177e4SLinus Torvalds } 47081da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 47091da177e4SLinus Torvalds 471009796395SRoss Zwisler static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, 4711ac46d4f3SJérôme Glisse struct mmu_notifier_range *range, 471209796395SRoss Zwisler pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) 4713f8ad0f49SJohannes Weiner { 4714f8ad0f49SJohannes Weiner pgd_t *pgd; 4715c2febafcSKirill A. Shutemov p4d_t *p4d; 4716f8ad0f49SJohannes Weiner pud_t *pud; 4717f8ad0f49SJohannes Weiner pmd_t *pmd; 4718f8ad0f49SJohannes Weiner pte_t *ptep; 4719f8ad0f49SJohannes Weiner 4720f8ad0f49SJohannes Weiner pgd = pgd_offset(mm, address); 4721f8ad0f49SJohannes Weiner if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 4722f8ad0f49SJohannes Weiner goto out; 4723f8ad0f49SJohannes Weiner 4724c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, address); 4725c2febafcSKirill A. Shutemov if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) 4726c2febafcSKirill A. Shutemov goto out; 4727c2febafcSKirill A. Shutemov 4728c2febafcSKirill A. Shutemov pud = pud_offset(p4d, address); 4729f8ad0f49SJohannes Weiner if (pud_none(*pud) || unlikely(pud_bad(*pud))) 4730f8ad0f49SJohannes Weiner goto out; 4731f8ad0f49SJohannes Weiner 4732f8ad0f49SJohannes Weiner pmd = pmd_offset(pud, address); 4733f66055abSAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 473409796395SRoss Zwisler 473509796395SRoss Zwisler if (pmd_huge(*pmd)) { 473609796395SRoss Zwisler if (!pmdpp) 4737f8ad0f49SJohannes Weiner goto out; 4738f8ad0f49SJohannes Weiner 4739ac46d4f3SJérôme Glisse if (range) { 47407269f999SJérôme Glisse mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, 47416f4f13e8SJérôme Glisse NULL, mm, address & PMD_MASK, 4742ac46d4f3SJérôme Glisse (address & PMD_MASK) + PMD_SIZE); 4743ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(range); 4744a4d1a885SJérôme Glisse } 474509796395SRoss Zwisler *ptlp = pmd_lock(mm, pmd); 474609796395SRoss Zwisler if (pmd_huge(*pmd)) { 474709796395SRoss Zwisler *pmdpp = pmd; 474809796395SRoss Zwisler return 0; 474909796395SRoss Zwisler } 475009796395SRoss Zwisler spin_unlock(*ptlp); 4751ac46d4f3SJérôme Glisse if (range) 4752ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(range); 475309796395SRoss Zwisler } 475409796395SRoss Zwisler 475509796395SRoss Zwisler if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 4756f8ad0f49SJohannes Weiner goto out; 4757f8ad0f49SJohannes Weiner 4758ac46d4f3SJérôme Glisse if (range) { 47597269f999SJérôme Glisse mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, NULL, mm, 47606f4f13e8SJérôme Glisse address & PAGE_MASK, 47611ed7293aSMatthew Wilcox (address & PAGE_MASK) + PAGE_SIZE); 4762ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(range); 4763a4d1a885SJérôme Glisse } 4764f8ad0f49SJohannes Weiner ptep = pte_offset_map_lock(mm, pmd, address, ptlp); 4765f8ad0f49SJohannes Weiner if (!pte_present(*ptep)) 4766f8ad0f49SJohannes Weiner goto unlock; 4767f8ad0f49SJohannes Weiner *ptepp = ptep; 4768f8ad0f49SJohannes Weiner return 0; 4769f8ad0f49SJohannes Weiner unlock: 4770f8ad0f49SJohannes Weiner pte_unmap_unlock(ptep, *ptlp); 4771ac46d4f3SJérôme Glisse if (range) 4772ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(range); 4773f8ad0f49SJohannes Weiner out: 4774f8ad0f49SJohannes Weiner return -EINVAL; 4775f8ad0f49SJohannes Weiner } 4776f8ad0f49SJohannes Weiner 4777f729c8c9SRoss Zwisler static inline int follow_pte(struct mm_struct *mm, unsigned long address, 4778f729c8c9SRoss Zwisler pte_t **ptepp, spinlock_t **ptlp) 47791b36ba81SNamhyung Kim { 47801b36ba81SNamhyung Kim int res; 47811b36ba81SNamhyung Kim 47821b36ba81SNamhyung Kim /* (void) is needed to make gcc happy */ 47831b36ba81SNamhyung Kim (void) __cond_lock(*ptlp, 4784ac46d4f3SJérôme Glisse !(res = __follow_pte_pmd(mm, address, NULL, 4785a4d1a885SJérôme Glisse ptepp, NULL, ptlp))); 47861b36ba81SNamhyung Kim return res; 47871b36ba81SNamhyung Kim } 47881b36ba81SNamhyung Kim 478909796395SRoss Zwisler int follow_pte_pmd(struct mm_struct *mm, unsigned long address, 4790ac46d4f3SJérôme Glisse struct mmu_notifier_range *range, 479109796395SRoss Zwisler pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) 479209796395SRoss Zwisler { 479309796395SRoss Zwisler int res; 479409796395SRoss Zwisler 479509796395SRoss Zwisler /* (void) is needed to make gcc happy */ 479609796395SRoss Zwisler (void) __cond_lock(*ptlp, 4797ac46d4f3SJérôme Glisse !(res = __follow_pte_pmd(mm, address, range, 4798a4d1a885SJérôme Glisse ptepp, pmdpp, ptlp))); 479909796395SRoss Zwisler return res; 480009796395SRoss Zwisler } 480109796395SRoss Zwisler EXPORT_SYMBOL(follow_pte_pmd); 480209796395SRoss Zwisler 48033b6748e2SJohannes Weiner /** 48043b6748e2SJohannes Weiner * follow_pfn - look up PFN at a user virtual address 48053b6748e2SJohannes Weiner * @vma: memory mapping 48063b6748e2SJohannes Weiner * @address: user virtual address 48073b6748e2SJohannes Weiner * @pfn: location to store found PFN 48083b6748e2SJohannes Weiner * 48093b6748e2SJohannes Weiner * Only IO mappings and raw PFN mappings are allowed. 48103b6748e2SJohannes Weiner * 4811a862f68aSMike Rapoport * Return: zero and the pfn at @pfn on success, -ve otherwise. 48123b6748e2SJohannes Weiner */ 48133b6748e2SJohannes Weiner int follow_pfn(struct vm_area_struct *vma, unsigned long address, 48143b6748e2SJohannes Weiner unsigned long *pfn) 48153b6748e2SJohannes Weiner { 48163b6748e2SJohannes Weiner int ret = -EINVAL; 48173b6748e2SJohannes Weiner spinlock_t *ptl; 48183b6748e2SJohannes Weiner pte_t *ptep; 48193b6748e2SJohannes Weiner 48203b6748e2SJohannes Weiner if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 48213b6748e2SJohannes Weiner return ret; 48223b6748e2SJohannes Weiner 48233b6748e2SJohannes Weiner ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); 48243b6748e2SJohannes Weiner if (ret) 48253b6748e2SJohannes Weiner return ret; 48263b6748e2SJohannes Weiner *pfn = pte_pfn(*ptep); 48273b6748e2SJohannes Weiner pte_unmap_unlock(ptep, ptl); 48283b6748e2SJohannes Weiner return 0; 48293b6748e2SJohannes Weiner } 48303b6748e2SJohannes Weiner EXPORT_SYMBOL(follow_pfn); 48313b6748e2SJohannes Weiner 483228b2ee20SRik van Riel #ifdef CONFIG_HAVE_IOREMAP_PROT 4833d87fe660Svenkatesh.pallipadi@intel.com int follow_phys(struct vm_area_struct *vma, 483428b2ee20SRik van Riel unsigned long address, unsigned int flags, 4835d87fe660Svenkatesh.pallipadi@intel.com unsigned long *prot, resource_size_t *phys) 483628b2ee20SRik van Riel { 483703668a4dSJohannes Weiner int ret = -EINVAL; 483828b2ee20SRik van Riel pte_t *ptep, pte; 483928b2ee20SRik van Riel spinlock_t *ptl; 484028b2ee20SRik van Riel 4841d87fe660Svenkatesh.pallipadi@intel.com if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 4842d87fe660Svenkatesh.pallipadi@intel.com goto out; 484328b2ee20SRik van Riel 484403668a4dSJohannes Weiner if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) 4845d87fe660Svenkatesh.pallipadi@intel.com goto out; 484628b2ee20SRik van Riel pte = *ptep; 484703668a4dSJohannes Weiner 4848f6f37321SLinus Torvalds if ((flags & FOLL_WRITE) && !pte_write(pte)) 484928b2ee20SRik van Riel goto unlock; 485028b2ee20SRik van Riel 485128b2ee20SRik van Riel *prot = pgprot_val(pte_pgprot(pte)); 485203668a4dSJohannes Weiner *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; 485328b2ee20SRik van Riel 485403668a4dSJohannes Weiner ret = 0; 485528b2ee20SRik van Riel unlock: 485628b2ee20SRik van Riel pte_unmap_unlock(ptep, ptl); 485728b2ee20SRik van Riel out: 4858d87fe660Svenkatesh.pallipadi@intel.com return ret; 485928b2ee20SRik van Riel } 486028b2ee20SRik van Riel 486128b2ee20SRik van Riel int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, 486228b2ee20SRik van Riel void *buf, int len, int write) 486328b2ee20SRik van Riel { 486428b2ee20SRik van Riel resource_size_t phys_addr; 486528b2ee20SRik van Riel unsigned long prot = 0; 48662bc7273bSKOSAKI Motohiro void __iomem *maddr; 486728b2ee20SRik van Riel int offset = addr & (PAGE_SIZE-1); 486828b2ee20SRik van Riel 4869d87fe660Svenkatesh.pallipadi@intel.com if (follow_phys(vma, addr, write, &prot, &phys_addr)) 487028b2ee20SRik van Riel return -EINVAL; 487128b2ee20SRik van Riel 48729cb12d7bSGrazvydas Ignotas maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); 487324eee1e4Sjie@chenjie6@huwei.com if (!maddr) 487424eee1e4Sjie@chenjie6@huwei.com return -ENOMEM; 487524eee1e4Sjie@chenjie6@huwei.com 487628b2ee20SRik van Riel if (write) 487728b2ee20SRik van Riel memcpy_toio(maddr + offset, buf, len); 487828b2ee20SRik van Riel else 487928b2ee20SRik van Riel memcpy_fromio(buf, maddr + offset, len); 488028b2ee20SRik van Riel iounmap(maddr); 488128b2ee20SRik van Riel 488228b2ee20SRik van Riel return len; 488328b2ee20SRik van Riel } 48845a73633eSUwe Kleine-König EXPORT_SYMBOL_GPL(generic_access_phys); 488528b2ee20SRik van Riel #endif 488628b2ee20SRik van Riel 48870ec76a11SDavid Howells /* 4888206cb636SStephen Wilson * Access another process' address space as given in mm. If non-NULL, use the 4889206cb636SStephen Wilson * given task for page fault accounting. 48900ec76a11SDavid Howells */ 489184d77d3fSEric W. Biederman int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 4892442486ecSLorenzo Stoakes unsigned long addr, void *buf, int len, unsigned int gup_flags) 48930ec76a11SDavid Howells { 48940ec76a11SDavid Howells struct vm_area_struct *vma; 48950ec76a11SDavid Howells void *old_buf = buf; 4896442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 48970ec76a11SDavid Howells 4898d8ed45c5SMichel Lespinasse if (mmap_read_lock_killable(mm)) 48991e426fe2SKonstantin Khlebnikov return 0; 49001e426fe2SKonstantin Khlebnikov 4901183ff22bSSimon Arlott /* ignore errors, just check how much was successfully transferred */ 49020ec76a11SDavid Howells while (len) { 49030ec76a11SDavid Howells int bytes, ret, offset; 49040ec76a11SDavid Howells void *maddr; 490528b2ee20SRik van Riel struct page *page = NULL; 49060ec76a11SDavid Howells 490764019a2eSPeter Xu ret = get_user_pages_remote(mm, addr, 1, 49085b56d49fSLorenzo Stoakes gup_flags, &page, &vma, NULL); 490928b2ee20SRik van Riel if (ret <= 0) { 4910dbffcd03SRik van Riel #ifndef CONFIG_HAVE_IOREMAP_PROT 4911dbffcd03SRik van Riel break; 4912dbffcd03SRik van Riel #else 491328b2ee20SRik van Riel /* 491428b2ee20SRik van Riel * Check if this is a VM_IO | VM_PFNMAP VMA, which 491528b2ee20SRik van Riel * we can access using slightly different code. 491628b2ee20SRik van Riel */ 491728b2ee20SRik van Riel vma = find_vma(mm, addr); 4918fe936dfcSMichael Ellerman if (!vma || vma->vm_start > addr) 49190ec76a11SDavid Howells break; 492028b2ee20SRik van Riel if (vma->vm_ops && vma->vm_ops->access) 492128b2ee20SRik van Riel ret = vma->vm_ops->access(vma, addr, buf, 492228b2ee20SRik van Riel len, write); 492328b2ee20SRik van Riel if (ret <= 0) 492428b2ee20SRik van Riel break; 492528b2ee20SRik van Riel bytes = ret; 4926dbffcd03SRik van Riel #endif 492728b2ee20SRik van Riel } else { 49280ec76a11SDavid Howells bytes = len; 49290ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 49300ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 49310ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 49320ec76a11SDavid Howells 49330ec76a11SDavid Howells maddr = kmap(page); 49340ec76a11SDavid Howells if (write) { 49350ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 49360ec76a11SDavid Howells maddr + offset, buf, bytes); 49370ec76a11SDavid Howells set_page_dirty_lock(page); 49380ec76a11SDavid Howells } else { 49390ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 49400ec76a11SDavid Howells buf, maddr + offset, bytes); 49410ec76a11SDavid Howells } 49420ec76a11SDavid Howells kunmap(page); 494309cbfeafSKirill A. Shutemov put_page(page); 494428b2ee20SRik van Riel } 49450ec76a11SDavid Howells len -= bytes; 49460ec76a11SDavid Howells buf += bytes; 49470ec76a11SDavid Howells addr += bytes; 49480ec76a11SDavid Howells } 4949d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 49500ec76a11SDavid Howells 49510ec76a11SDavid Howells return buf - old_buf; 49520ec76a11SDavid Howells } 495303252919SAndi Kleen 49545ddd36b9SStephen Wilson /** 4955ae91dbfcSRandy Dunlap * access_remote_vm - access another process' address space 49565ddd36b9SStephen Wilson * @mm: the mm_struct of the target address space 49575ddd36b9SStephen Wilson * @addr: start address to access 49585ddd36b9SStephen Wilson * @buf: source or destination buffer 49595ddd36b9SStephen Wilson * @len: number of bytes to transfer 49606347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 49615ddd36b9SStephen Wilson * 49625ddd36b9SStephen Wilson * The caller must hold a reference on @mm. 4963a862f68aSMike Rapoport * 4964a862f68aSMike Rapoport * Return: number of bytes copied from source to destination. 49655ddd36b9SStephen Wilson */ 49665ddd36b9SStephen Wilson int access_remote_vm(struct mm_struct *mm, unsigned long addr, 49676347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 49685ddd36b9SStephen Wilson { 49696347e8d5SLorenzo Stoakes return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); 49705ddd36b9SStephen Wilson } 49715ddd36b9SStephen Wilson 497203252919SAndi Kleen /* 4973206cb636SStephen Wilson * Access another process' address space. 4974206cb636SStephen Wilson * Source/target buffer must be kernel space, 4975206cb636SStephen Wilson * Do not walk the page table directly, use get_user_pages 4976206cb636SStephen Wilson */ 4977206cb636SStephen Wilson int access_process_vm(struct task_struct *tsk, unsigned long addr, 4978f307ab6dSLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 4979206cb636SStephen Wilson { 4980206cb636SStephen Wilson struct mm_struct *mm; 4981206cb636SStephen Wilson int ret; 4982206cb636SStephen Wilson 4983206cb636SStephen Wilson mm = get_task_mm(tsk); 4984206cb636SStephen Wilson if (!mm) 4985206cb636SStephen Wilson return 0; 4986206cb636SStephen Wilson 4987f307ab6dSLorenzo Stoakes ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); 4988442486ecSLorenzo Stoakes 4989206cb636SStephen Wilson mmput(mm); 4990206cb636SStephen Wilson 4991206cb636SStephen Wilson return ret; 4992206cb636SStephen Wilson } 4993fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 4994206cb636SStephen Wilson 499503252919SAndi Kleen /* 499603252919SAndi Kleen * Print the name of a VMA. 499703252919SAndi Kleen */ 499803252919SAndi Kleen void print_vma_addr(char *prefix, unsigned long ip) 499903252919SAndi Kleen { 500003252919SAndi Kleen struct mm_struct *mm = current->mm; 500103252919SAndi Kleen struct vm_area_struct *vma; 500203252919SAndi Kleen 5003e8bff74aSIngo Molnar /* 50040a7f682dSMichal Hocko * we might be running from an atomic context so we cannot sleep 5005e8bff74aSIngo Molnar */ 5006d8ed45c5SMichel Lespinasse if (!mmap_read_trylock(mm)) 5007e8bff74aSIngo Molnar return; 5008e8bff74aSIngo Molnar 500903252919SAndi Kleen vma = find_vma(mm, ip); 501003252919SAndi Kleen if (vma && vma->vm_file) { 501103252919SAndi Kleen struct file *f = vma->vm_file; 50120a7f682dSMichal Hocko char *buf = (char *)__get_free_page(GFP_NOWAIT); 501303252919SAndi Kleen if (buf) { 50142fbc57c5SAndy Shevchenko char *p; 501503252919SAndi Kleen 50169bf39ab2SMiklos Szeredi p = file_path(f, buf, PAGE_SIZE); 501703252919SAndi Kleen if (IS_ERR(p)) 501803252919SAndi Kleen p = "?"; 50192fbc57c5SAndy Shevchenko printk("%s%s[%lx+%lx]", prefix, kbasename(p), 502003252919SAndi Kleen vma->vm_start, 502103252919SAndi Kleen vma->vm_end - vma->vm_start); 502203252919SAndi Kleen free_page((unsigned long)buf); 502303252919SAndi Kleen } 502403252919SAndi Kleen } 5025d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 502603252919SAndi Kleen } 50273ee1afa3SNick Piggin 5028662bbcb2SMichael S. Tsirkin #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) 50299ec23531SDavid Hildenbrand void __might_fault(const char *file, int line) 50303ee1afa3SNick Piggin { 503195156f00SPeter Zijlstra /* 503295156f00SPeter Zijlstra * Some code (nfs/sunrpc) uses socket ops on kernel memory while 5033c1e8d7c6SMichel Lespinasse * holding the mmap_lock, this is safe because kernel memory doesn't 503495156f00SPeter Zijlstra * get paged out, therefore we'll never actually fault, and the 503595156f00SPeter Zijlstra * below annotations will generate false positives. 503695156f00SPeter Zijlstra */ 5037db68ce10SAl Viro if (uaccess_kernel()) 503895156f00SPeter Zijlstra return; 50399ec23531SDavid Hildenbrand if (pagefault_disabled()) 5040662bbcb2SMichael S. Tsirkin return; 50419ec23531SDavid Hildenbrand __might_sleep(file, line, 0); 50429ec23531SDavid Hildenbrand #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) 5043662bbcb2SMichael S. Tsirkin if (current->mm) 5044da1c55f1SMichel Lespinasse might_lock_read(¤t->mm->mmap_lock); 50459ec23531SDavid Hildenbrand #endif 50463ee1afa3SNick Piggin } 50479ec23531SDavid Hildenbrand EXPORT_SYMBOL(__might_fault); 50483ee1afa3SNick Piggin #endif 504947ad8475SAndrea Arcangeli 505047ad8475SAndrea Arcangeli #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) 5051c6ddfb6cSHuang Ying /* 5052c6ddfb6cSHuang Ying * Process all subpages of the specified huge page with the specified 5053c6ddfb6cSHuang Ying * operation. The target subpage will be processed last to keep its 5054c6ddfb6cSHuang Ying * cache lines hot. 5055c6ddfb6cSHuang Ying */ 5056c6ddfb6cSHuang Ying static inline void process_huge_page( 5057c6ddfb6cSHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page, 5058c6ddfb6cSHuang Ying void (*process_subpage)(unsigned long addr, int idx, void *arg), 5059c6ddfb6cSHuang Ying void *arg) 5060c6ddfb6cSHuang Ying { 5061c6ddfb6cSHuang Ying int i, n, base, l; 5062c6ddfb6cSHuang Ying unsigned long addr = addr_hint & 5063c6ddfb6cSHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5064c6ddfb6cSHuang Ying 5065c6ddfb6cSHuang Ying /* Process target subpage last to keep its cache lines hot */ 5066c6ddfb6cSHuang Ying might_sleep(); 5067c6ddfb6cSHuang Ying n = (addr_hint - addr) / PAGE_SIZE; 5068c6ddfb6cSHuang Ying if (2 * n <= pages_per_huge_page) { 5069c6ddfb6cSHuang Ying /* If target subpage in first half of huge page */ 5070c6ddfb6cSHuang Ying base = 0; 5071c6ddfb6cSHuang Ying l = n; 5072c6ddfb6cSHuang Ying /* Process subpages at the end of huge page */ 5073c6ddfb6cSHuang Ying for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { 5074c6ddfb6cSHuang Ying cond_resched(); 5075c6ddfb6cSHuang Ying process_subpage(addr + i * PAGE_SIZE, i, arg); 5076c6ddfb6cSHuang Ying } 5077c6ddfb6cSHuang Ying } else { 5078c6ddfb6cSHuang Ying /* If target subpage in second half of huge page */ 5079c6ddfb6cSHuang Ying base = pages_per_huge_page - 2 * (pages_per_huge_page - n); 5080c6ddfb6cSHuang Ying l = pages_per_huge_page - n; 5081c6ddfb6cSHuang Ying /* Process subpages at the begin of huge page */ 5082c6ddfb6cSHuang Ying for (i = 0; i < base; i++) { 5083c6ddfb6cSHuang Ying cond_resched(); 5084c6ddfb6cSHuang Ying process_subpage(addr + i * PAGE_SIZE, i, arg); 5085c6ddfb6cSHuang Ying } 5086c6ddfb6cSHuang Ying } 5087c6ddfb6cSHuang Ying /* 5088c6ddfb6cSHuang Ying * Process remaining subpages in left-right-left-right pattern 5089c6ddfb6cSHuang Ying * towards the target subpage 5090c6ddfb6cSHuang Ying */ 5091c6ddfb6cSHuang Ying for (i = 0; i < l; i++) { 5092c6ddfb6cSHuang Ying int left_idx = base + i; 5093c6ddfb6cSHuang Ying int right_idx = base + 2 * l - 1 - i; 5094c6ddfb6cSHuang Ying 5095c6ddfb6cSHuang Ying cond_resched(); 5096c6ddfb6cSHuang Ying process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); 5097c6ddfb6cSHuang Ying cond_resched(); 5098c6ddfb6cSHuang Ying process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); 5099c6ddfb6cSHuang Ying } 5100c6ddfb6cSHuang Ying } 5101c6ddfb6cSHuang Ying 510247ad8475SAndrea Arcangeli static void clear_gigantic_page(struct page *page, 510347ad8475SAndrea Arcangeli unsigned long addr, 510447ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 510547ad8475SAndrea Arcangeli { 510647ad8475SAndrea Arcangeli int i; 510747ad8475SAndrea Arcangeli struct page *p = page; 510847ad8475SAndrea Arcangeli 510947ad8475SAndrea Arcangeli might_sleep(); 511047ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; 511147ad8475SAndrea Arcangeli i++, p = mem_map_next(p, page, i)) { 511247ad8475SAndrea Arcangeli cond_resched(); 511347ad8475SAndrea Arcangeli clear_user_highpage(p, addr + i * PAGE_SIZE); 511447ad8475SAndrea Arcangeli } 511547ad8475SAndrea Arcangeli } 5116c6ddfb6cSHuang Ying 5117c6ddfb6cSHuang Ying static void clear_subpage(unsigned long addr, int idx, void *arg) 5118c6ddfb6cSHuang Ying { 5119c6ddfb6cSHuang Ying struct page *page = arg; 5120c6ddfb6cSHuang Ying 5121c6ddfb6cSHuang Ying clear_user_highpage(page + idx, addr); 5122c6ddfb6cSHuang Ying } 5123c6ddfb6cSHuang Ying 512447ad8475SAndrea Arcangeli void clear_huge_page(struct page *page, 5125c79b57e4SHuang Ying unsigned long addr_hint, unsigned int pages_per_huge_page) 512647ad8475SAndrea Arcangeli { 5127c79b57e4SHuang Ying unsigned long addr = addr_hint & 5128c79b57e4SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 512947ad8475SAndrea Arcangeli 513047ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 513147ad8475SAndrea Arcangeli clear_gigantic_page(page, addr, pages_per_huge_page); 513247ad8475SAndrea Arcangeli return; 513347ad8475SAndrea Arcangeli } 513447ad8475SAndrea Arcangeli 5135c6ddfb6cSHuang Ying process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); 513647ad8475SAndrea Arcangeli } 513747ad8475SAndrea Arcangeli 513847ad8475SAndrea Arcangeli static void copy_user_gigantic_page(struct page *dst, struct page *src, 513947ad8475SAndrea Arcangeli unsigned long addr, 514047ad8475SAndrea Arcangeli struct vm_area_struct *vma, 514147ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 514247ad8475SAndrea Arcangeli { 514347ad8475SAndrea Arcangeli int i; 514447ad8475SAndrea Arcangeli struct page *dst_base = dst; 514547ad8475SAndrea Arcangeli struct page *src_base = src; 514647ad8475SAndrea Arcangeli 514747ad8475SAndrea Arcangeli for (i = 0; i < pages_per_huge_page; ) { 514847ad8475SAndrea Arcangeli cond_resched(); 514947ad8475SAndrea Arcangeli copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); 515047ad8475SAndrea Arcangeli 515147ad8475SAndrea Arcangeli i++; 515247ad8475SAndrea Arcangeli dst = mem_map_next(dst, dst_base, i); 515347ad8475SAndrea Arcangeli src = mem_map_next(src, src_base, i); 515447ad8475SAndrea Arcangeli } 515547ad8475SAndrea Arcangeli } 515647ad8475SAndrea Arcangeli 5157c9f4cd71SHuang Ying struct copy_subpage_arg { 5158c9f4cd71SHuang Ying struct page *dst; 5159c9f4cd71SHuang Ying struct page *src; 5160c9f4cd71SHuang Ying struct vm_area_struct *vma; 5161c9f4cd71SHuang Ying }; 5162c9f4cd71SHuang Ying 5163c9f4cd71SHuang Ying static void copy_subpage(unsigned long addr, int idx, void *arg) 5164c9f4cd71SHuang Ying { 5165c9f4cd71SHuang Ying struct copy_subpage_arg *copy_arg = arg; 5166c9f4cd71SHuang Ying 5167c9f4cd71SHuang Ying copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, 5168c9f4cd71SHuang Ying addr, copy_arg->vma); 5169c9f4cd71SHuang Ying } 5170c9f4cd71SHuang Ying 517147ad8475SAndrea Arcangeli void copy_user_huge_page(struct page *dst, struct page *src, 5172c9f4cd71SHuang Ying unsigned long addr_hint, struct vm_area_struct *vma, 517347ad8475SAndrea Arcangeli unsigned int pages_per_huge_page) 517447ad8475SAndrea Arcangeli { 5175c9f4cd71SHuang Ying unsigned long addr = addr_hint & 5176c9f4cd71SHuang Ying ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); 5177c9f4cd71SHuang Ying struct copy_subpage_arg arg = { 5178c9f4cd71SHuang Ying .dst = dst, 5179c9f4cd71SHuang Ying .src = src, 5180c9f4cd71SHuang Ying .vma = vma, 5181c9f4cd71SHuang Ying }; 518247ad8475SAndrea Arcangeli 518347ad8475SAndrea Arcangeli if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { 518447ad8475SAndrea Arcangeli copy_user_gigantic_page(dst, src, addr, vma, 518547ad8475SAndrea Arcangeli pages_per_huge_page); 518647ad8475SAndrea Arcangeli return; 518747ad8475SAndrea Arcangeli } 518847ad8475SAndrea Arcangeli 5189c9f4cd71SHuang Ying process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); 519047ad8475SAndrea Arcangeli } 5191fa4d75c1SMike Kravetz 5192fa4d75c1SMike Kravetz long copy_huge_page_from_user(struct page *dst_page, 5193fa4d75c1SMike Kravetz const void __user *usr_src, 5194810a56b9SMike Kravetz unsigned int pages_per_huge_page, 5195810a56b9SMike Kravetz bool allow_pagefault) 5196fa4d75c1SMike Kravetz { 5197fa4d75c1SMike Kravetz void *src = (void *)usr_src; 5198fa4d75c1SMike Kravetz void *page_kaddr; 5199fa4d75c1SMike Kravetz unsigned long i, rc = 0; 5200fa4d75c1SMike Kravetz unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; 5201fa4d75c1SMike Kravetz 5202fa4d75c1SMike Kravetz for (i = 0; i < pages_per_huge_page; i++) { 5203810a56b9SMike Kravetz if (allow_pagefault) 5204810a56b9SMike Kravetz page_kaddr = kmap(dst_page + i); 5205810a56b9SMike Kravetz else 5206fa4d75c1SMike Kravetz page_kaddr = kmap_atomic(dst_page + i); 5207fa4d75c1SMike Kravetz rc = copy_from_user(page_kaddr, 5208fa4d75c1SMike Kravetz (const void __user *)(src + i * PAGE_SIZE), 5209fa4d75c1SMike Kravetz PAGE_SIZE); 5210810a56b9SMike Kravetz if (allow_pagefault) 5211810a56b9SMike Kravetz kunmap(dst_page + i); 5212810a56b9SMike Kravetz else 5213fa4d75c1SMike Kravetz kunmap_atomic(page_kaddr); 5214fa4d75c1SMike Kravetz 5215fa4d75c1SMike Kravetz ret_val -= (PAGE_SIZE - rc); 5216fa4d75c1SMike Kravetz if (rc) 5217fa4d75c1SMike Kravetz break; 5218fa4d75c1SMike Kravetz 5219fa4d75c1SMike Kravetz cond_resched(); 5220fa4d75c1SMike Kravetz } 5221fa4d75c1SMike Kravetz return ret_val; 5222fa4d75c1SMike Kravetz } 522347ad8475SAndrea Arcangeli #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ 522449076ec2SKirill A. Shutemov 522540b64acdSOlof Johansson #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS 5226b35f1819SKirill A. Shutemov 5227b35f1819SKirill A. Shutemov static struct kmem_cache *page_ptl_cachep; 5228b35f1819SKirill A. Shutemov 5229b35f1819SKirill A. Shutemov void __init ptlock_cache_init(void) 5230b35f1819SKirill A. Shutemov { 5231b35f1819SKirill A. Shutemov page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, 5232b35f1819SKirill A. Shutemov SLAB_PANIC, NULL); 5233b35f1819SKirill A. Shutemov } 5234b35f1819SKirill A. Shutemov 5235539edb58SPeter Zijlstra bool ptlock_alloc(struct page *page) 523649076ec2SKirill A. Shutemov { 523749076ec2SKirill A. Shutemov spinlock_t *ptl; 523849076ec2SKirill A. Shutemov 5239b35f1819SKirill A. Shutemov ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); 524049076ec2SKirill A. Shutemov if (!ptl) 524149076ec2SKirill A. Shutemov return false; 5242539edb58SPeter Zijlstra page->ptl = ptl; 524349076ec2SKirill A. Shutemov return true; 524449076ec2SKirill A. Shutemov } 524549076ec2SKirill A. Shutemov 5246539edb58SPeter Zijlstra void ptlock_free(struct page *page) 524749076ec2SKirill A. Shutemov { 5248b35f1819SKirill A. Shutemov kmem_cache_free(page_ptl_cachep, page->ptl); 524949076ec2SKirill A. Shutemov } 525049076ec2SKirill A. Shutemov #endif 5251