Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/memory.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* |
| 8 | * demand-loading started 01.12.91 - seems it is high on the list of |
| 9 | * things wanted, and it should be easy to implement. - Linus |
| 10 | */ |
| 11 | |
| 12 | /* |
| 13 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared |
| 14 | * pages started 02.12.91, seems to work. - Linus. |
| 15 | * |
| 16 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it |
| 17 | * would have taken more than the 6M I have free, but it worked well as |
| 18 | * far as I could see. |
| 19 | * |
| 20 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. |
| 21 | */ |
| 22 | |
| 23 | /* |
| 24 | * Real VM (paging to/from disk) started 18.12.91. Much more work and |
| 25 | * thought has to go into this. Oh, well.. |
| 26 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. |
| 27 | * Found it. Everything seems to work now. |
| 28 | * 20.12.91 - Ok, making the swap-device changeable like the root. |
| 29 | */ |
| 30 | |
| 31 | /* |
| 32 | * 05.04.94 - Multi-page memory management added for v1.1. |
| 33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
| 34 | * |
| 35 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG |
| 36 | * (Gerhard.Wichert@pdb.siemens.de) |
| 37 | * |
| 38 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) |
| 39 | */ |
| 40 | |
| 41 | #include <linux/kernel_stat.h> |
| 42 | #include <linux/mm.h> |
| 43 | #include <linux/hugetlb.h> |
| 44 | #include <linux/mman.h> |
| 45 | #include <linux/swap.h> |
| 46 | #include <linux/highmem.h> |
| 47 | #include <linux/pagemap.h> |
Hugh Dickins | 9a84089 | 2009-09-21 17:02:01 -0700 | [diff] [blame] | 48 | #include <linux/ksm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | #include <linux/rmap.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 50 | #include <linux/export.h> |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 51 | #include <linux/delayacct.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | #include <linux/init.h> |
Dan Williams | 01c8f1c | 2016-01-15 16:56:40 -0800 | [diff] [blame] | 53 | #include <linux/pfn_t.h> |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 54 | #include <linux/writeback.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 55 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 56 | #include <linux/mmu_notifier.h> |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 57 | #include <linux/kallsyms.h> |
| 58 | #include <linux/swapops.h> |
| 59 | #include <linux/elf.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 60 | #include <linux/gfp.h> |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 61 | #include <linux/migrate.h> |
Andy Shevchenko | 2fbc57c | 2012-12-17 16:01:23 -0800 | [diff] [blame] | 62 | #include <linux/string.h> |
Dan Williams | 0abdd7a | 2014-01-21 15:48:12 -0800 | [diff] [blame] | 63 | #include <linux/dma-debug.h> |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 64 | #include <linux/debugfs.h> |
Andrea Arcangeli | 6b251fc | 2015-09-04 15:46:20 -0700 | [diff] [blame] | 65 | #include <linux/userfaultfd_k.h> |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 66 | #include <linux/dax.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | |
Alexey Dobriyan | 6952b61 | 2009-09-18 23:55:55 +0400 | [diff] [blame] | 68 | #include <asm/io.h> |
Dave Hansen | 33a709b | 2016-02-12 13:02:19 -0800 | [diff] [blame] | 69 | #include <asm/mmu_context.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | #include <asm/pgalloc.h> |
| 71 | #include <asm/uaccess.h> |
| 72 | #include <asm/tlb.h> |
| 73 | #include <asm/tlbflush.h> |
| 74 | #include <asm/pgtable.h> |
| 75 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 76 | #include "internal.h" |
| 77 | |
Peter Zijlstra | 9057289 | 2013-10-07 11:29:20 +0100 | [diff] [blame] | 78 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
| 79 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
Peter Zijlstra | 75980e9 | 2013-02-22 16:34:32 -0800 | [diff] [blame] | 80 | #endif |
| 81 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 82 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
| 84 | unsigned long max_mapnr; |
| 85 | struct page *mem_map; |
| 86 | |
| 87 | EXPORT_SYMBOL(max_mapnr); |
| 88 | EXPORT_SYMBOL(mem_map); |
| 89 | #endif |
| 90 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | /* |
| 92 | * A number of key systems in x86 including ioremap() rely on the assumption |
| 93 | * that high_memory defines the upper bound on direct map memory, then end |
| 94 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and |
| 95 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL |
| 96 | * and ZONE_HIGHMEM. |
| 97 | */ |
| 98 | void * high_memory; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | EXPORT_SYMBOL(high_memory); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | |
Ingo Molnar | 32a9323 | 2008-02-06 22:39:44 +0100 | [diff] [blame] | 102 | /* |
| 103 | * Randomize the address space (stacks, mmaps, brk, etc.). |
| 104 | * |
| 105 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, |
| 106 | * as ancient (libc5 based) binaries can segfault. ) |
| 107 | */ |
| 108 | int randomize_va_space __read_mostly = |
| 109 | #ifdef CONFIG_COMPAT_BRK |
| 110 | 1; |
| 111 | #else |
| 112 | 2; |
| 113 | #endif |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 114 | |
| 115 | static int __init disable_randmaps(char *s) |
| 116 | { |
| 117 | randomize_va_space = 0; |
OGAWA Hirofumi | 9b41046 | 2006-03-31 02:30:33 -0800 | [diff] [blame] | 118 | return 1; |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 119 | } |
| 120 | __setup("norandmaps", disable_randmaps); |
| 121 | |
Hugh Dickins | 62eede6 | 2009-09-21 17:03:34 -0700 | [diff] [blame] | 122 | unsigned long zero_pfn __read_mostly; |
Hugh Dickins | 03f6462 | 2009-09-21 17:03:35 -0700 | [diff] [blame] | 123 | unsigned long highest_memmap_pfn __read_mostly; |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 124 | |
Ard Biesheuvel | 0b70068 | 2014-09-12 22:17:23 +0200 | [diff] [blame] | 125 | EXPORT_SYMBOL(zero_pfn); |
| 126 | |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 127 | /* |
| 128 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() |
| 129 | */ |
| 130 | static int __init init_zero_pfn(void) |
| 131 | { |
| 132 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); |
| 133 | return 0; |
| 134 | } |
| 135 | core_initcall(init_zero_pfn); |
Andi Kleen | a62eaf1 | 2006-02-16 23:41:58 +0100 | [diff] [blame] | 136 | |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 137 | |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 138 | #if defined(SPLIT_RSS_COUNTING) |
| 139 | |
David Rientjes | ea48cf7 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 140 | void sync_mm_rss(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 141 | { |
| 142 | int i; |
| 143 | |
| 144 | for (i = 0; i < NR_MM_COUNTERS; i++) { |
David Rientjes | 05af2e1 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 145 | if (current->rss_stat.count[i]) { |
| 146 | add_mm_counter(mm, i, current->rss_stat.count[i]); |
| 147 | current->rss_stat.count[i] = 0; |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 148 | } |
| 149 | } |
David Rientjes | 05af2e1 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 150 | current->rss_stat.events = 0; |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 151 | } |
| 152 | |
| 153 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) |
| 154 | { |
| 155 | struct task_struct *task = current; |
| 156 | |
| 157 | if (likely(task->mm == mm)) |
| 158 | task->rss_stat.count[member] += val; |
| 159 | else |
| 160 | add_mm_counter(mm, member, val); |
| 161 | } |
| 162 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) |
| 163 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) |
| 164 | |
| 165 | /* sync counter once per 64 page faults */ |
| 166 | #define TASK_RSS_EVENTS_THRESH (64) |
| 167 | static void check_sync_rss_stat(struct task_struct *task) |
| 168 | { |
| 169 | if (unlikely(task != current)) |
| 170 | return; |
| 171 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) |
David Rientjes | ea48cf7 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 172 | sync_mm_rss(task->mm); |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 173 | } |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 174 | #else /* SPLIT_RSS_COUNTING */ |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 175 | |
| 176 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) |
| 177 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) |
| 178 | |
| 179 | static void check_sync_rss_stat(struct task_struct *task) |
| 180 | { |
| 181 | } |
| 182 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 183 | #endif /* SPLIT_RSS_COUNTING */ |
| 184 | |
| 185 | #ifdef HAVE_GENERIC_MMU_GATHER |
| 186 | |
Nicholas Krause | ca1d6c7 | 2015-09-04 15:48:22 -0700 | [diff] [blame] | 187 | static bool tlb_next_batch(struct mmu_gather *tlb) |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 188 | { |
| 189 | struct mmu_gather_batch *batch; |
| 190 | |
| 191 | batch = tlb->active; |
| 192 | if (batch->next) { |
| 193 | tlb->active = batch->next; |
Nicholas Krause | ca1d6c7 | 2015-09-04 15:48:22 -0700 | [diff] [blame] | 194 | return true; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 195 | } |
| 196 | |
Michal Hocko | 53a59fc | 2013-01-04 15:35:12 -0800 | [diff] [blame] | 197 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
Nicholas Krause | ca1d6c7 | 2015-09-04 15:48:22 -0700 | [diff] [blame] | 198 | return false; |
Michal Hocko | 53a59fc | 2013-01-04 15:35:12 -0800 | [diff] [blame] | 199 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 200 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
| 201 | if (!batch) |
Nicholas Krause | ca1d6c7 | 2015-09-04 15:48:22 -0700 | [diff] [blame] | 202 | return false; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 203 | |
Michal Hocko | 53a59fc | 2013-01-04 15:35:12 -0800 | [diff] [blame] | 204 | tlb->batch_count++; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 205 | batch->next = NULL; |
| 206 | batch->nr = 0; |
| 207 | batch->max = MAX_GATHER_BATCH; |
| 208 | |
| 209 | tlb->active->next = batch; |
| 210 | tlb->active = batch; |
| 211 | |
Nicholas Krause | ca1d6c7 | 2015-09-04 15:48:22 -0700 | [diff] [blame] | 212 | return true; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 213 | } |
| 214 | |
| 215 | /* tlb_gather_mmu |
| 216 | * Called to initialize an (on-stack) mmu_gather structure for page-table |
| 217 | * tear-down from @mm. The @fullmm argument is used when @mm is without |
| 218 | * users and we're going to destroy the full address space (exit/execve). |
| 219 | */ |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 220 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 221 | { |
| 222 | tlb->mm = mm; |
| 223 | |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 224 | /* Is it from 0 to ~0? */ |
| 225 | tlb->fullmm = !(start | (end+1)); |
Dave Hansen | 1de14c3 | 2013-04-12 16:23:54 -0700 | [diff] [blame] | 226 | tlb->need_flush_all = 0; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 227 | tlb->local.next = NULL; |
| 228 | tlb->local.nr = 0; |
| 229 | tlb->local.max = ARRAY_SIZE(tlb->__pages); |
| 230 | tlb->active = &tlb->local; |
Michal Hocko | 53a59fc | 2013-01-04 15:35:12 -0800 | [diff] [blame] | 231 | tlb->batch_count = 0; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 232 | |
| 233 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
| 234 | tlb->batch = NULL; |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 235 | #endif |
Aneesh Kumar K.V | e77b085 | 2016-07-26 15:24:12 -0700 | [diff] [blame] | 236 | tlb->page_size = 0; |
Will Deacon | fb7332a | 2014-10-29 10:03:09 +0000 | [diff] [blame] | 237 | |
| 238 | __tlb_reset_range(tlb); |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 239 | } |
| 240 | |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 241 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 242 | { |
Will Deacon | 721c21c | 2015-01-12 19:10:55 +0000 | [diff] [blame] | 243 | if (!tlb->end) |
| 244 | return; |
| 245 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 246 | tlb_flush(tlb); |
Joerg Roedel | 34ee645 | 2014-11-13 13:46:09 +1100 | [diff] [blame] | 247 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 248 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
| 249 | tlb_table_flush(tlb); |
| 250 | #endif |
Will Deacon | fb7332a | 2014-10-29 10:03:09 +0000 | [diff] [blame] | 251 | __tlb_reset_range(tlb); |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 252 | } |
| 253 | |
| 254 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) |
| 255 | { |
| 256 | struct mmu_gather_batch *batch; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 257 | |
Will Deacon | 721c21c | 2015-01-12 19:10:55 +0000 | [diff] [blame] | 258 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 259 | free_pages_and_swap_cache(batch->pages, batch->nr); |
| 260 | batch->nr = 0; |
| 261 | } |
| 262 | tlb->active = &tlb->local; |
| 263 | } |
| 264 | |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 265 | void tlb_flush_mmu(struct mmu_gather *tlb) |
| 266 | { |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 267 | tlb_flush_mmu_tlbonly(tlb); |
| 268 | tlb_flush_mmu_free(tlb); |
| 269 | } |
| 270 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 271 | /* tlb_finish_mmu |
| 272 | * Called at the end of the shootdown operation to free up any resources |
| 273 | * that were required. |
| 274 | */ |
| 275 | void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) |
| 276 | { |
| 277 | struct mmu_gather_batch *batch, *next; |
| 278 | |
| 279 | tlb_flush_mmu(tlb); |
| 280 | |
| 281 | /* keep the page table cache within bounds */ |
| 282 | check_pgt_cache(); |
| 283 | |
| 284 | for (batch = tlb->local.next; batch; batch = next) { |
| 285 | next = batch->next; |
| 286 | free_pages((unsigned long)batch, 0); |
| 287 | } |
| 288 | tlb->local.next = NULL; |
| 289 | } |
| 290 | |
| 291 | /* __tlb_remove_page |
| 292 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while |
| 293 | * handling the additional races in SMP caused by other CPUs caching valid |
| 294 | * mappings in their TLBs. Returns the number of free page slots left. |
| 295 | * When out of page slots we must call tlb_flush_mmu(). |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 296 | *returns true if the caller should flush. |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 297 | */ |
Aneesh Kumar K.V | e77b085 | 2016-07-26 15:24:12 -0700 | [diff] [blame] | 298 | bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size) |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 299 | { |
| 300 | struct mmu_gather_batch *batch; |
| 301 | |
Will Deacon | fb7332a | 2014-10-29 10:03:09 +0000 | [diff] [blame] | 302 | VM_BUG_ON(!tlb->end); |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 303 | |
Aneesh Kumar K.V | e77b085 | 2016-07-26 15:24:12 -0700 | [diff] [blame] | 304 | if (!tlb->page_size) |
| 305 | tlb->page_size = page_size; |
| 306 | else { |
| 307 | if (page_size != tlb->page_size) |
| 308 | return true; |
| 309 | } |
| 310 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 311 | batch = tlb->active; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 312 | if (batch->nr == batch->max) { |
| 313 | if (!tlb_next_batch(tlb)) |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 314 | return true; |
Shaohua Li | 0b43c3a | 2011-07-08 15:39:41 -0700 | [diff] [blame] | 315 | batch = tlb->active; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 316 | } |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 317 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 318 | |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 319 | batch->pages[batch->nr++] = page; |
| 320 | return false; |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 321 | } |
| 322 | |
| 323 | #endif /* HAVE_GENERIC_MMU_GATHER */ |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 324 | |
Peter Zijlstra | 2672391 | 2011-05-24 17:12:00 -0700 | [diff] [blame] | 325 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
| 326 | |
| 327 | /* |
| 328 | * See the comment near struct mmu_table_batch. |
| 329 | */ |
| 330 | |
| 331 | static void tlb_remove_table_smp_sync(void *arg) |
| 332 | { |
| 333 | /* Simply deliver the interrupt */ |
| 334 | } |
| 335 | |
| 336 | static void tlb_remove_table_one(void *table) |
| 337 | { |
| 338 | /* |
| 339 | * This isn't an RCU grace period and hence the page-tables cannot be |
| 340 | * assumed to be actually RCU-freed. |
| 341 | * |
| 342 | * It is however sufficient for software page-table walkers that rely on |
| 343 | * IRQ disabling. See the comment near struct mmu_table_batch. |
| 344 | */ |
| 345 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); |
| 346 | __tlb_remove_table(table); |
| 347 | } |
| 348 | |
| 349 | static void tlb_remove_table_rcu(struct rcu_head *head) |
| 350 | { |
| 351 | struct mmu_table_batch *batch; |
| 352 | int i; |
| 353 | |
| 354 | batch = container_of(head, struct mmu_table_batch, rcu); |
| 355 | |
| 356 | for (i = 0; i < batch->nr; i++) |
| 357 | __tlb_remove_table(batch->tables[i]); |
| 358 | |
| 359 | free_page((unsigned long)batch); |
| 360 | } |
| 361 | |
| 362 | void tlb_table_flush(struct mmu_gather *tlb) |
| 363 | { |
| 364 | struct mmu_table_batch **batch = &tlb->batch; |
| 365 | |
| 366 | if (*batch) { |
| 367 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); |
| 368 | *batch = NULL; |
| 369 | } |
| 370 | } |
| 371 | |
| 372 | void tlb_remove_table(struct mmu_gather *tlb, void *table) |
| 373 | { |
| 374 | struct mmu_table_batch **batch = &tlb->batch; |
| 375 | |
Peter Zijlstra | 2672391 | 2011-05-24 17:12:00 -0700 | [diff] [blame] | 376 | /* |
| 377 | * When there's less then two users of this mm there cannot be a |
| 378 | * concurrent page-table walk. |
| 379 | */ |
| 380 | if (atomic_read(&tlb->mm->mm_users) < 2) { |
| 381 | __tlb_remove_table(table); |
| 382 | return; |
| 383 | } |
| 384 | |
| 385 | if (*batch == NULL) { |
| 386 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); |
| 387 | if (*batch == NULL) { |
| 388 | tlb_remove_table_one(table); |
| 389 | return; |
| 390 | } |
| 391 | (*batch)->nr = 0; |
| 392 | } |
| 393 | (*batch)->tables[(*batch)->nr++] = table; |
| 394 | if ((*batch)->nr == MAX_TABLE_BATCH) |
| 395 | tlb_table_flush(tlb); |
| 396 | } |
| 397 | |
Peter Zijlstra | 9547d01 | 2011-05-24 17:12:14 -0700 | [diff] [blame] | 398 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
Peter Zijlstra | 2672391 | 2011-05-24 17:12:00 -0700 | [diff] [blame] | 399 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | * Note: this doesn't free the actual pages themselves. That |
| 402 | * has been handled earlier when unmapping all the memory regions. |
| 403 | */ |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 404 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
| 405 | unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | { |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 407 | pgtable_t token = pmd_pgtable(*pmd); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 408 | pmd_clear(pmd); |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 409 | pte_free_tlb(tlb, token, addr); |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 410 | atomic_long_dec(&tlb->mm->nr_ptes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | } |
| 412 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 413 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
| 414 | unsigned long addr, unsigned long end, |
| 415 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | { |
| 417 | pmd_t *pmd; |
| 418 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 419 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 421 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | pmd = pmd_offset(pud, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | do { |
| 424 | next = pmd_addr_end(addr, end); |
| 425 | if (pmd_none_or_clear_bad(pmd)) |
| 426 | continue; |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 427 | free_pte_range(tlb, pmd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | } while (pmd++, addr = next, addr != end); |
| 429 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 430 | start &= PUD_MASK; |
| 431 | if (start < floor) |
| 432 | return; |
| 433 | if (ceiling) { |
| 434 | ceiling &= PUD_MASK; |
| 435 | if (!ceiling) |
| 436 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 438 | if (end - 1 > ceiling - 1) |
| 439 | return; |
| 440 | |
| 441 | pmd = pmd_offset(pud, start); |
| 442 | pud_clear(pud); |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 443 | pmd_free_tlb(tlb, pmd, start); |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 444 | mm_dec_nr_pmds(tlb->mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | } |
| 446 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 447 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
| 448 | unsigned long addr, unsigned long end, |
| 449 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | { |
| 451 | pud_t *pud; |
| 452 | unsigned long next; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 453 | unsigned long start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 455 | start = addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | pud = pud_offset(pgd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | do { |
| 458 | next = pud_addr_end(addr, end); |
| 459 | if (pud_none_or_clear_bad(pud)) |
| 460 | continue; |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 461 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | } while (pud++, addr = next, addr != end); |
| 463 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 464 | start &= PGDIR_MASK; |
| 465 | if (start < floor) |
| 466 | return; |
| 467 | if (ceiling) { |
| 468 | ceiling &= PGDIR_MASK; |
| 469 | if (!ceiling) |
| 470 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 472 | if (end - 1 > ceiling - 1) |
| 473 | return; |
| 474 | |
| 475 | pud = pud_offset(pgd, start); |
| 476 | pgd_clear(pgd); |
Benjamin Herrenschmidt | 9e1b32c | 2009-07-22 15:44:28 +1000 | [diff] [blame] | 477 | pud_free_tlb(tlb, pud, start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | } |
| 479 | |
| 480 | /* |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 481 | * This function frees user-level page tables of a process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | */ |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 483 | void free_pgd_range(struct mmu_gather *tlb, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 484 | unsigned long addr, unsigned long end, |
| 485 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | { |
| 487 | pgd_t *pgd; |
| 488 | unsigned long next; |
| 489 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 490 | /* |
| 491 | * The next few lines have given us lots of grief... |
| 492 | * |
| 493 | * Why are we testing PMD* at this top level? Because often |
| 494 | * there will be no work to do at all, and we'd prefer not to |
| 495 | * go all the way down to the bottom just to discover that. |
| 496 | * |
| 497 | * Why all these "- 1"s? Because 0 represents both the bottom |
| 498 | * of the address space and the top of it (using -1 for the |
| 499 | * top wouldn't help much: the masks would do the wrong thing). |
| 500 | * The rule is that addr 0 and floor 0 refer to the bottom of |
| 501 | * the address space, but end 0 and ceiling 0 refer to the top |
| 502 | * Comparisons need to use "end - 1" and "ceiling - 1" (though |
| 503 | * that end 0 case should be mythical). |
| 504 | * |
| 505 | * Wherever addr is brought up or ceiling brought down, we must |
| 506 | * be careful to reject "the opposite 0" before it confuses the |
| 507 | * subsequent tests. But what about where end is brought down |
| 508 | * by PMD_SIZE below? no, end can't go down to 0 there. |
| 509 | * |
| 510 | * Whereas we round start (addr) and ceiling down, by different |
| 511 | * masks at different levels, in order to test whether a table |
| 512 | * now has no other vmas using it, so can be freed, we don't |
| 513 | * bother to round floor or end up - the tests don't need that. |
| 514 | */ |
| 515 | |
| 516 | addr &= PMD_MASK; |
| 517 | if (addr < floor) { |
| 518 | addr += PMD_SIZE; |
| 519 | if (!addr) |
| 520 | return; |
| 521 | } |
| 522 | if (ceiling) { |
| 523 | ceiling &= PMD_MASK; |
| 524 | if (!ceiling) |
| 525 | return; |
| 526 | } |
| 527 | if (end - 1 > ceiling - 1) |
| 528 | end -= PMD_SIZE; |
| 529 | if (addr > end - 1) |
| 530 | return; |
| 531 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 532 | pgd = pgd_offset(tlb->mm, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | do { |
| 534 | next = pgd_addr_end(addr, end); |
| 535 | if (pgd_none_or_clear_bad(pgd)) |
| 536 | continue; |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 537 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | } while (pgd++, addr = next, addr != end); |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 539 | } |
| 540 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 541 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 542 | unsigned long floor, unsigned long ceiling) |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 543 | { |
| 544 | while (vma) { |
| 545 | struct vm_area_struct *next = vma->vm_next; |
| 546 | unsigned long addr = vma->vm_start; |
| 547 | |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 548 | /* |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 549 | * Hide vma from rmap and truncate_pagecache before freeing |
| 550 | * pgtables |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 551 | */ |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 552 | unlink_anon_vmas(vma); |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 553 | unlink_file_vma(vma); |
| 554 | |
David Gibson | 9da61ae | 2006-03-22 00:08:57 -0800 | [diff] [blame] | 555 | if (is_vm_hugetlb_page(vma)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 556 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 557 | floor, next? next->vm_start: ceiling); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 558 | } else { |
| 559 | /* |
| 560 | * Optimization: gather nearby vmas into one call down |
| 561 | */ |
| 562 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE |
David Gibson | 4866920 | 2006-03-22 00:08:58 -0800 | [diff] [blame] | 563 | && !is_vm_hugetlb_page(next)) { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 564 | vma = next; |
| 565 | next = vma->vm_next; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 566 | unlink_anon_vmas(vma); |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 567 | unlink_file_vma(vma); |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 568 | } |
| 569 | free_pgd_range(tlb, addr, vma->vm_end, |
| 570 | floor, next? next->vm_start: ceiling); |
| 571 | } |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 572 | vma = next; |
| 573 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | } |
| 575 | |
Kirill A. Shutemov | 3ed3a4f | 2016-03-17 14:19:11 -0700 | [diff] [blame] | 576 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | { |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 578 | spinlock_t *ptl; |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 579 | pgtable_t new = pte_alloc_one(mm, address); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 580 | if (!new) |
| 581 | return -ENOMEM; |
| 582 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 583 | /* |
| 584 | * Ensure all pte setup (eg. pte page lock and page clearing) are |
| 585 | * visible before the pte is made visible to other CPUs by being |
| 586 | * put into page tables. |
| 587 | * |
| 588 | * The other side of the story is the pointer chasing in the page |
| 589 | * table walking code (when walking the page table without locking; |
| 590 | * ie. most of the time). Fortunately, these data accesses consist |
| 591 | * of a chain of data-dependent loads, meaning most CPUs (alpha |
| 592 | * being the notable exception) will already guarantee loads are |
| 593 | * seen in-order. See the alpha page table accessors for the |
| 594 | * smp_read_barrier_depends() barriers in page table walking code. |
| 595 | */ |
| 596 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ |
| 597 | |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 598 | ptl = pmd_lock(mm, pmd); |
Andrea Arcangeli | 8ac1f83 | 2011-01-13 15:46:43 -0800 | [diff] [blame] | 599 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 600 | atomic_long_inc(&mm->nr_ptes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 601 | pmd_populate(mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 602 | new = NULL; |
Kirill A. Shutemov | 4b471e8 | 2016-01-15 16:53:39 -0800 | [diff] [blame] | 603 | } |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 604 | spin_unlock(ptl); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 605 | if (new) |
| 606 | pte_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 607 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | } |
| 609 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 610 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | { |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 612 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
| 613 | if (!new) |
| 614 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 616 | smp_wmb(); /* See comment in __pte_alloc */ |
| 617 | |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 618 | spin_lock(&init_mm.page_table_lock); |
Andrea Arcangeli | 8ac1f83 | 2011-01-13 15:46:43 -0800 | [diff] [blame] | 619 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 620 | pmd_populate_kernel(&init_mm, pmd, new); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 621 | new = NULL; |
Kirill A. Shutemov | 4b471e8 | 2016-01-15 16:53:39 -0800 | [diff] [blame] | 622 | } |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 623 | spin_unlock(&init_mm.page_table_lock); |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 624 | if (new) |
| 625 | pte_free_kernel(&init_mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 626 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | } |
| 628 | |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 629 | static inline void init_rss_vec(int *rss) |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 630 | { |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 631 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); |
| 632 | } |
| 633 | |
| 634 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) |
| 635 | { |
| 636 | int i; |
| 637 | |
KAMEZAWA Hiroyuki | 34e5523 | 2010-03-05 13:41:40 -0800 | [diff] [blame] | 638 | if (current->mm == mm) |
David Rientjes | 05af2e1 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 639 | sync_mm_rss(mm); |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 640 | for (i = 0; i < NR_MM_COUNTERS; i++) |
| 641 | if (rss[i]) |
| 642 | add_mm_counter(mm, i, rss[i]); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 643 | } |
| 644 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | /* |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 646 | * This function is called to print an error when a bad pte |
| 647 | * is found. For example, we might have a PFN-mapped pte in |
| 648 | * a region that doesn't allow it. |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 649 | * |
| 650 | * The calling function must still handle the error. |
| 651 | */ |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 652 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
| 653 | pte_t pte, struct page *page) |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 654 | { |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 655 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
| 656 | pud_t *pud = pud_offset(pgd, addr); |
| 657 | pmd_t *pmd = pmd_offset(pud, addr); |
| 658 | struct address_space *mapping; |
| 659 | pgoff_t index; |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 660 | static unsigned long resume; |
| 661 | static unsigned long nr_shown; |
| 662 | static unsigned long nr_unshown; |
| 663 | |
| 664 | /* |
| 665 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 666 | * or allow a steady drip of one report per second. |
| 667 | */ |
| 668 | if (nr_shown == 60) { |
| 669 | if (time_before(jiffies, resume)) { |
| 670 | nr_unshown++; |
| 671 | return; |
| 672 | } |
| 673 | if (nr_unshown) { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 674 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
| 675 | nr_unshown); |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 676 | nr_unshown = 0; |
| 677 | } |
| 678 | nr_shown = 0; |
| 679 | } |
| 680 | if (nr_shown++ == 0) |
| 681 | resume = jiffies + 60 * HZ; |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 682 | |
| 683 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; |
| 684 | index = linear_page_index(vma, addr); |
| 685 | |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 686 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
| 687 | current->comm, |
| 688 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 689 | if (page) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 690 | dump_page(page, "bad pte"); |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 691 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
| 692 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 693 | /* |
| 694 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y |
| 695 | */ |
Konstantin Khlebnikov | 2682582 | 2015-04-15 16:15:08 -0700 | [diff] [blame] | 696 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
| 697 | vma->vm_file, |
| 698 | vma->vm_ops ? vma->vm_ops->fault : NULL, |
| 699 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, |
| 700 | mapping ? mapping->a_ops->readpage : NULL); |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 701 | dump_stack(); |
Rusty Russell | 373d4d0 | 2013-01-21 17:17:39 +1030 | [diff] [blame] | 702 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 703 | } |
| 704 | |
| 705 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 706 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 707 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 708 | * "Special" mappings do not wish to be associated with a "struct page" (either |
| 709 | * it doesn't exist, or it exists but they don't want to touch it). In this |
| 710 | * case, NULL is returned here. "Normal" mappings do have a struct page. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 711 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 712 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
| 713 | * pte bit, in which case this function is trivial. Secondly, an architecture |
| 714 | * may not have a spare pte bit, which requires a more complicated scheme, |
| 715 | * described below. |
| 716 | * |
| 717 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a |
| 718 | * special mapping (even if there are underlying and valid "struct pages"). |
| 719 | * COWed pages of a VM_PFNMAP are always normal. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 720 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 721 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
| 722 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 723 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
| 724 | * mapping will always honor the rule |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 725 | * |
| 726 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) |
| 727 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 728 | * And for normal mappings this is false. |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 729 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 730 | * This restricts such mappings to be a linear translation from virtual address |
| 731 | * to pfn. To get around this restriction, we allow arbitrary mappings so long |
| 732 | * as the vma is not a COW mapping; in that case, we know that all ptes are |
| 733 | * special (because none can have been COWed). |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 734 | * |
| 735 | * |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 736 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
| 737 | * |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 738 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct |
| 739 | * page" backing, however the difference is that _all_ pages with a struct |
| 740 | * page (that is, those where pfn_valid is true) are refcounted and considered |
| 741 | * normal pages by the VM. The disadvantage is that pages are refcounted |
| 742 | * (which can be slower and simply not an option for some PFNMAP users). The |
| 743 | * advantage is that we don't have to follow the strict linearity rule of |
| 744 | * PFNMAP mappings in order to support COWable mappings. |
| 745 | * |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 746 | */ |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 747 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
| 748 | # define HAVE_PTE_SPECIAL 1 |
| 749 | #else |
| 750 | # define HAVE_PTE_SPECIAL 0 |
| 751 | #endif |
| 752 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
| 753 | pte_t pte) |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 754 | { |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 755 | unsigned long pfn = pte_pfn(pte); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 756 | |
| 757 | if (HAVE_PTE_SPECIAL) { |
Hugh Dickins | b38af47 | 2014-08-29 15:18:44 -0700 | [diff] [blame] | 758 | if (likely(!pte_special(pte))) |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 759 | goto check_pfn; |
David Vrabel | 667a0a0 | 2014-12-18 14:48:15 +0000 | [diff] [blame] | 760 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
| 761 | return vma->vm_ops->find_special_page(vma, addr); |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 762 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
| 763 | return NULL; |
Hugh Dickins | 62eede6 | 2009-09-21 17:03:34 -0700 | [diff] [blame] | 764 | if (!is_zero_pfn(pfn)) |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 765 | print_bad_pte(vma, addr, pte, NULL); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 766 | return NULL; |
| 767 | } |
| 768 | |
| 769 | /* !HAVE_PTE_SPECIAL case follows: */ |
| 770 | |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 771 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
| 772 | if (vma->vm_flags & VM_MIXEDMAP) { |
| 773 | if (!pfn_valid(pfn)) |
| 774 | return NULL; |
| 775 | goto out; |
| 776 | } else { |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 777 | unsigned long off; |
| 778 | off = (addr - vma->vm_start) >> PAGE_SHIFT; |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 779 | if (pfn == vma->vm_pgoff + off) |
| 780 | return NULL; |
| 781 | if (!is_cow_mapping(vma->vm_flags)) |
| 782 | return NULL; |
| 783 | } |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 784 | } |
| 785 | |
Hugh Dickins | b38af47 | 2014-08-29 15:18:44 -0700 | [diff] [blame] | 786 | if (is_zero_pfn(pfn)) |
| 787 | return NULL; |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 788 | check_pfn: |
| 789 | if (unlikely(pfn > highest_memmap_pfn)) { |
| 790 | print_bad_pte(vma, addr, pte, NULL); |
| 791 | return NULL; |
| 792 | } |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 793 | |
| 794 | /* |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 795 | * NOTE! We still have PageReserved() pages in the page tables. |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 796 | * eg. VDSO mappings can cause them to exist. |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 797 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 798 | out: |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 799 | return pfn_to_page(pfn); |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 800 | } |
| 801 | |
Gerald Schaefer | 28093f9 | 2016-04-28 16:18:35 -0700 | [diff] [blame] | 802 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 803 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, |
| 804 | pmd_t pmd) |
| 805 | { |
| 806 | unsigned long pfn = pmd_pfn(pmd); |
| 807 | |
| 808 | /* |
| 809 | * There is no pmd_special() but there may be special pmds, e.g. |
| 810 | * in a direct-access (dax) mapping, so let's just replicate the |
| 811 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. |
| 812 | */ |
| 813 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
| 814 | if (vma->vm_flags & VM_MIXEDMAP) { |
| 815 | if (!pfn_valid(pfn)) |
| 816 | return NULL; |
| 817 | goto out; |
| 818 | } else { |
| 819 | unsigned long off; |
| 820 | off = (addr - vma->vm_start) >> PAGE_SHIFT; |
| 821 | if (pfn == vma->vm_pgoff + off) |
| 822 | return NULL; |
| 823 | if (!is_cow_mapping(vma->vm_flags)) |
| 824 | return NULL; |
| 825 | } |
| 826 | } |
| 827 | |
| 828 | if (is_zero_pfn(pfn)) |
| 829 | return NULL; |
| 830 | if (unlikely(pfn > highest_memmap_pfn)) |
| 831 | return NULL; |
| 832 | |
| 833 | /* |
| 834 | * NOTE! We still have PageReserved() pages in the page tables. |
| 835 | * eg. VDSO mappings can cause them to exist. |
| 836 | */ |
| 837 | out: |
| 838 | return pfn_to_page(pfn); |
| 839 | } |
| 840 | #endif |
| 841 | |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 842 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | * copy one vm_area from one task to the other. Assumes the page tables |
| 844 | * already present in the new task to be cleared in the whole range |
| 845 | * covered by this vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | */ |
| 847 | |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 848 | static inline unsigned long |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 850 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
Hugh Dickins | 8c10376 | 2005-10-29 18:16:13 -0700 | [diff] [blame] | 851 | unsigned long addr, int *rss) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 853 | unsigned long vm_flags = vma->vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | pte_t pte = *src_pte; |
| 855 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | |
| 857 | /* pte contains position in swap or file, so copy. */ |
| 858 | if (unlikely(!pte_present(pte))) { |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 859 | swp_entry_t entry = pte_to_swp_entry(pte); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 860 | |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 861 | if (likely(!non_swap_entry(entry))) { |
| 862 | if (swap_duplicate(entry) < 0) |
| 863 | return entry.val; |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 864 | |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 865 | /* make sure dst_mm is on swapoff's mmlist. */ |
| 866 | if (unlikely(list_empty(&dst_mm->mmlist))) { |
| 867 | spin_lock(&mmlist_lock); |
| 868 | if (list_empty(&dst_mm->mmlist)) |
| 869 | list_add(&dst_mm->mmlist, |
| 870 | &src_mm->mmlist); |
| 871 | spin_unlock(&mmlist_lock); |
| 872 | } |
| 873 | rss[MM_SWAPENTS]++; |
| 874 | } else if (is_migration_entry(entry)) { |
| 875 | page = migration_entry_to_page(entry); |
Konstantin Khlebnikov | 9f9f1ac | 2012-01-20 14:34:24 -0800 | [diff] [blame] | 876 | |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 877 | rss[mm_counter(page)]++; |
Konstantin Khlebnikov | 9f9f1ac | 2012-01-20 14:34:24 -0800 | [diff] [blame] | 878 | |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 879 | if (is_write_migration_entry(entry) && |
| 880 | is_cow_mapping(vm_flags)) { |
| 881 | /* |
| 882 | * COW mappings require pages in both |
| 883 | * parent and child to be set to read. |
| 884 | */ |
| 885 | make_migration_entry_read(&entry); |
| 886 | pte = swp_entry_to_pte(entry); |
| 887 | if (pte_swp_soft_dirty(*src_pte)) |
| 888 | pte = pte_swp_mksoft_dirty(pte); |
| 889 | set_pte_at(src_mm, addr, src_pte, pte); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 890 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 892 | goto out_set_pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | } |
| 894 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | /* |
| 896 | * If it's a COW mapping, write protect it both |
| 897 | * in the parent and the child |
| 898 | */ |
Linus Torvalds | 6712117 | 2005-12-11 20:38:17 -0800 | [diff] [blame] | 899 | if (is_cow_mapping(vm_flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | ptep_set_wrprotect(src_mm, addr, src_pte); |
Zachary Amsden | 3dc9079 | 2006-09-30 23:29:30 -0700 | [diff] [blame] | 901 | pte = pte_wrprotect(pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | } |
| 903 | |
| 904 | /* |
| 905 | * If it's a shared mapping, mark it clean in |
| 906 | * the child |
| 907 | */ |
| 908 | if (vm_flags & VM_SHARED) |
| 909 | pte = pte_mkclean(pte); |
| 910 | pte = pte_mkold(pte); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 911 | |
| 912 | page = vm_normal_page(vma, addr, pte); |
| 913 | if (page) { |
| 914 | get_page(page); |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 915 | page_dup_rmap(page, false); |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 916 | rss[mm_counter(page)]++; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 917 | } |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 918 | |
| 919 | out_set_pte: |
| 920 | set_pte_at(dst_mm, addr, dst_pte, pte); |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 921 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | } |
| 923 | |
Jerome Marchand | 21bda26 | 2014-08-06 16:06:56 -0700 | [diff] [blame] | 924 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 925 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
| 926 | unsigned long addr, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | { |
Daisuke Nishimura | c36987e | 2009-10-26 16:50:23 -0700 | [diff] [blame] | 928 | pte_t *orig_src_pte, *orig_dst_pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | pte_t *src_pte, *dst_pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 930 | spinlock_t *src_ptl, *dst_ptl; |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 931 | int progress = 0; |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 932 | int rss[NR_MM_COUNTERS]; |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 933 | swp_entry_t entry = (swp_entry_t){0}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | |
| 935 | again: |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 936 | init_rss_vec(rss); |
| 937 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 938 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | if (!dst_pte) |
| 940 | return -ENOMEM; |
Peter Zijlstra | ece0e2b | 2010-10-26 14:21:52 -0700 | [diff] [blame] | 941 | src_pte = pte_offset_map(src_pmd, addr); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 942 | src_ptl = pte_lockptr(src_mm, src_pmd); |
Ingo Molnar | f20dc5f | 2006-07-03 00:25:08 -0700 | [diff] [blame] | 943 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
Daisuke Nishimura | c36987e | 2009-10-26 16:50:23 -0700 | [diff] [blame] | 944 | orig_src_pte = src_pte; |
| 945 | orig_dst_pte = dst_pte; |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 946 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 947 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | do { |
| 949 | /* |
| 950 | * We are holding two locks at this point - either of them |
| 951 | * could generate latencies in another task on another CPU. |
| 952 | */ |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 953 | if (progress >= 32) { |
| 954 | progress = 0; |
| 955 | if (need_resched() || |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 956 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
Hugh Dickins | e040f21 | 2005-10-29 18:15:53 -0700 | [diff] [blame] | 957 | break; |
| 958 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | if (pte_none(*src_pte)) { |
| 960 | progress++; |
| 961 | continue; |
| 962 | } |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 963 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
| 964 | vma, addr, rss); |
| 965 | if (entry.val) |
| 966 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | progress += 8; |
| 968 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 970 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 971 | spin_unlock(src_ptl); |
Peter Zijlstra | ece0e2b | 2010-10-26 14:21:52 -0700 | [diff] [blame] | 972 | pte_unmap(orig_src_pte); |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 973 | add_mm_rss_vec(dst_mm, rss); |
Daisuke Nishimura | c36987e | 2009-10-26 16:50:23 -0700 | [diff] [blame] | 974 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 975 | cond_resched(); |
Hugh Dickins | 570a335b | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 976 | |
| 977 | if (entry.val) { |
| 978 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) |
| 979 | return -ENOMEM; |
| 980 | progress = 0; |
| 981 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | if (addr != end) |
| 983 | goto again; |
| 984 | return 0; |
| 985 | } |
| 986 | |
| 987 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 988 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, |
| 989 | unsigned long addr, unsigned long end) |
| 990 | { |
| 991 | pmd_t *src_pmd, *dst_pmd; |
| 992 | unsigned long next; |
| 993 | |
| 994 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); |
| 995 | if (!dst_pmd) |
| 996 | return -ENOMEM; |
| 997 | src_pmd = pmd_offset(src_pud, addr); |
| 998 | do { |
| 999 | next = pmd_addr_end(addr, end); |
Dan Williams | 5c7fb56 | 2016-01-15 16:56:52 -0800 | [diff] [blame] | 1000 | if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) { |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 1001 | int err; |
Andrea Arcangeli | 14d1a55 | 2011-01-13 15:47:15 -0800 | [diff] [blame] | 1002 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 1003 | err = copy_huge_pmd(dst_mm, src_mm, |
| 1004 | dst_pmd, src_pmd, addr, vma); |
| 1005 | if (err == -ENOMEM) |
| 1006 | return -ENOMEM; |
| 1007 | if (!err) |
| 1008 | continue; |
| 1009 | /* fall through */ |
| 1010 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | if (pmd_none_or_clear_bad(src_pmd)) |
| 1012 | continue; |
| 1013 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, |
| 1014 | vma, addr, next)) |
| 1015 | return -ENOMEM; |
| 1016 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); |
| 1017 | return 0; |
| 1018 | } |
| 1019 | |
| 1020 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 1021 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, |
| 1022 | unsigned long addr, unsigned long end) |
| 1023 | { |
| 1024 | pud_t *src_pud, *dst_pud; |
| 1025 | unsigned long next; |
| 1026 | |
| 1027 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); |
| 1028 | if (!dst_pud) |
| 1029 | return -ENOMEM; |
| 1030 | src_pud = pud_offset(src_pgd, addr); |
| 1031 | do { |
| 1032 | next = pud_addr_end(addr, end); |
| 1033 | if (pud_none_or_clear_bad(src_pud)) |
| 1034 | continue; |
| 1035 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, |
| 1036 | vma, addr, next)) |
| 1037 | return -ENOMEM; |
| 1038 | } while (dst_pud++, src_pud++, addr = next, addr != end); |
| 1039 | return 0; |
| 1040 | } |
| 1041 | |
| 1042 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
| 1043 | struct vm_area_struct *vma) |
| 1044 | { |
| 1045 | pgd_t *src_pgd, *dst_pgd; |
| 1046 | unsigned long next; |
| 1047 | unsigned long addr = vma->vm_start; |
| 1048 | unsigned long end = vma->vm_end; |
Sagi Grimberg | 2ec74c3 | 2012-10-08 16:33:33 -0700 | [diff] [blame] | 1049 | unsigned long mmun_start; /* For mmu_notifiers */ |
| 1050 | unsigned long mmun_end; /* For mmu_notifiers */ |
| 1051 | bool is_cow; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1052 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 1054 | /* |
| 1055 | * Don't copy ptes where a page fault will fill them correctly. |
| 1056 | * Fork becomes much lighter when there are big shared or private |
| 1057 | * readonly mappings. The tradeoff is that copy_page_range is more |
| 1058 | * efficient than faulting. |
| 1059 | */ |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 1060 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
| 1061 | !vma->anon_vma) |
| 1062 | return 0; |
Nick Piggin | d992895 | 2005-08-28 16:49:11 +1000 | [diff] [blame] | 1063 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | if (is_vm_hugetlb_page(vma)) |
| 1065 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); |
| 1066 | |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1067 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1068 | /* |
| 1069 | * We do not free on error cases below as remove_vma |
| 1070 | * gets called on error from higher level routine |
| 1071 | */ |
Suresh Siddha | 5180da4 | 2012-10-08 16:28:29 -0700 | [diff] [blame] | 1072 | ret = track_pfn_copy(vma); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1073 | if (ret) |
| 1074 | return ret; |
| 1075 | } |
| 1076 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1077 | /* |
| 1078 | * We need to invalidate the secondary MMU mappings only when |
| 1079 | * there could be a permission downgrade on the ptes of the |
| 1080 | * parent mm. And a permission downgrade will only happen if |
| 1081 | * is_cow_mapping() returns true. |
| 1082 | */ |
Sagi Grimberg | 2ec74c3 | 2012-10-08 16:33:33 -0700 | [diff] [blame] | 1083 | is_cow = is_cow_mapping(vma->vm_flags); |
| 1084 | mmun_start = addr; |
| 1085 | mmun_end = end; |
| 1086 | if (is_cow) |
| 1087 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, |
| 1088 | mmun_end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1089 | |
| 1090 | ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1091 | dst_pgd = pgd_offset(dst_mm, addr); |
| 1092 | src_pgd = pgd_offset(src_mm, addr); |
| 1093 | do { |
| 1094 | next = pgd_addr_end(addr, end); |
| 1095 | if (pgd_none_or_clear_bad(src_pgd)) |
| 1096 | continue; |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1097 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
| 1098 | vma, addr, next))) { |
| 1099 | ret = -ENOMEM; |
| 1100 | break; |
| 1101 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1102 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1103 | |
Sagi Grimberg | 2ec74c3 | 2012-10-08 16:33:33 -0700 | [diff] [blame] | 1104 | if (is_cow) |
| 1105 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1106 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | } |
| 1108 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1109 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1110 | struct vm_area_struct *vma, pmd_t *pmd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | unsigned long addr, unsigned long end, |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1112 | struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1114 | struct mm_struct *mm = tlb->mm; |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 1115 | int force_flush = 0; |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1116 | int rss[NR_MM_COUNTERS]; |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1117 | spinlock_t *ptl; |
Steven Rostedt | 5f1a190 | 2011-06-15 15:08:23 -0700 | [diff] [blame] | 1118 | pte_t *start_pte; |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1119 | pte_t *pte; |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1120 | swp_entry_t entry; |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 1121 | struct page *pending_page = NULL; |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1122 | |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 1123 | again: |
Peter Zijlstra | e303297 | 2011-05-24 17:12:01 -0700 | [diff] [blame] | 1124 | init_rss_vec(rss); |
Steven Rostedt | 5f1a190 | 2011-06-15 15:08:23 -0700 | [diff] [blame] | 1125 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
| 1126 | pte = start_pte; |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1127 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | do { |
| 1129 | pte_t ptent = *pte; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1130 | if (pte_none(ptent)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | continue; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1132 | } |
Hugh Dickins | 6f5e6b9 | 2006-03-16 23:04:09 -0800 | [diff] [blame] | 1133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | if (pte_present(ptent)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 1135 | struct page *page; |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1136 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1137 | page = vm_normal_page(vma, addr, ptent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | if (unlikely(details) && page) { |
| 1139 | /* |
| 1140 | * unmap_shared_mapping_pages() wants to |
| 1141 | * invalidate cache without truncating: |
| 1142 | * unmap shared but keep private pages. |
| 1143 | */ |
| 1144 | if (details->check_mapping && |
Kirill A. Shutemov | 800d8c6 | 2016-07-26 15:26:18 -0700 | [diff] [blame] | 1145 | details->check_mapping != page_rmapping(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1147 | } |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1148 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
Zachary Amsden | a600388 | 2005-09-03 15:55:04 -0700 | [diff] [blame] | 1149 | tlb->fullmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | tlb_remove_tlb_entry(tlb, pte, addr); |
| 1151 | if (unlikely(!page)) |
| 1152 | continue; |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 1153 | |
| 1154 | if (!PageAnon(page)) { |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1155 | if (pte_dirty(ptent)) { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 1156 | /* |
| 1157 | * oom_reaper cannot tear down dirty |
| 1158 | * pages |
| 1159 | */ |
| 1160 | if (unlikely(details && details->ignore_dirty)) |
| 1161 | continue; |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1162 | force_flush = 1; |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 1163 | set_page_dirty(page); |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1164 | } |
Johannes Weiner | 4917e5d | 2009-01-06 14:39:17 -0800 | [diff] [blame] | 1165 | if (pte_young(ptent) && |
Joe Perches | 64363aa | 2013-07-08 16:00:18 -0700 | [diff] [blame] | 1166 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
Nick Piggin | bf3f3bc | 2009-01-06 14:38:55 -0800 | [diff] [blame] | 1167 | mark_page_accessed(page); |
Hugh Dickins | 6237bcd | 2005-10-29 18:15:54 -0700 | [diff] [blame] | 1168 | } |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 1169 | rss[mm_counter(page)]--; |
Kirill A. Shutemov | d281ee6 | 2016-01-15 16:52:16 -0800 | [diff] [blame] | 1170 | page_remove_rmap(page, false); |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 1171 | if (unlikely(page_mapcount(page) < 0)) |
| 1172 | print_bad_pte(vma, addr, ptent, page); |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 1173 | if (unlikely(__tlb_remove_page(tlb, page))) { |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1174 | force_flush = 1; |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 1175 | pending_page = page; |
Will Deacon | ce9ec37 | 2014-10-28 13:16:28 -0700 | [diff] [blame] | 1176 | addr += PAGE_SIZE; |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 1177 | break; |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1178 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | continue; |
| 1180 | } |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 1181 | /* only check swap_entries if explicitly asked for in details */ |
| 1182 | if (unlikely(details && !details->check_swap_entries)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1183 | continue; |
KAMEZAWA Hiroyuki | b084d43 | 2010-03-05 13:41:42 -0800 | [diff] [blame] | 1184 | |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1185 | entry = pte_to_swp_entry(ptent); |
| 1186 | if (!non_swap_entry(entry)) |
| 1187 | rss[MM_SWAPENTS]--; |
| 1188 | else if (is_migration_entry(entry)) { |
| 1189 | struct page *page; |
Konstantin Khlebnikov | 9f9f1ac | 2012-01-20 14:34:24 -0800 | [diff] [blame] | 1190 | |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1191 | page = migration_entry_to_page(entry); |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 1192 | rss[mm_counter(page)]--; |
KAMEZAWA Hiroyuki | b084d43 | 2010-03-05 13:41:42 -0800 | [diff] [blame] | 1193 | } |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1194 | if (unlikely(!free_swap_and_cache(entry))) |
| 1195 | print_bad_pte(vma, addr, ptent, NULL); |
Zachary Amsden | 9888a1c | 2006-09-30 23:29:31 -0700 | [diff] [blame] | 1196 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1197 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Hugh Dickins | ae85976 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 1198 | |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1199 | add_mm_rss_vec(mm, rss); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1200 | arch_leave_lazy_mmu_mode(); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1201 | |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1202 | /* Do the actual TLB flush before dropping ptl */ |
Will Deacon | fb7332a | 2014-10-29 10:03:09 +0000 | [diff] [blame] | 1203 | if (force_flush) |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1204 | tlb_flush_mmu_tlbonly(tlb); |
Linus Torvalds | 1cf35d4 | 2014-04-25 16:05:40 -0700 | [diff] [blame] | 1205 | pte_unmap_unlock(start_pte, ptl); |
| 1206 | |
| 1207 | /* |
| 1208 | * If we forced a TLB flush (either due to running out of |
| 1209 | * batch buffers or because we needed to flush dirty TLB |
| 1210 | * entries before releasing the ptl), free the batched |
| 1211 | * memory too. Restart if we didn't do everything. |
| 1212 | */ |
| 1213 | if (force_flush) { |
| 1214 | force_flush = 0; |
| 1215 | tlb_flush_mmu_free(tlb); |
Aneesh Kumar K.V | e9d55e1 | 2016-07-26 15:24:09 -0700 | [diff] [blame] | 1216 | if (pending_page) { |
| 1217 | /* remove the page with new size */ |
| 1218 | __tlb_remove_pte_page(tlb, pending_page); |
| 1219 | pending_page = NULL; |
| 1220 | } |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 1221 | if (addr != end) |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 1222 | goto again; |
| 1223 | } |
| 1224 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1225 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | } |
| 1227 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1228 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1229 | struct vm_area_struct *vma, pud_t *pud, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | unsigned long addr, unsigned long end, |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1231 | struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | { |
| 1233 | pmd_t *pmd; |
| 1234 | unsigned long next; |
| 1235 | |
| 1236 | pmd = pmd_offset(pud, addr); |
| 1237 | do { |
| 1238 | next = pmd_addr_end(addr, end); |
Dan Williams | 5c7fb56 | 2016-01-15 16:56:52 -0800 | [diff] [blame] | 1239 | if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 1240 | if (next - addr != HPAGE_PMD_SIZE) { |
Hugh Dickins | 6842839 | 2016-05-05 16:22:09 -0700 | [diff] [blame] | 1241 | VM_BUG_ON_VMA(vma_is_anonymous(vma) && |
| 1242 | !rwsem_is_locked(&tlb->mm->mmap_sem), vma); |
Kirill A. Shutemov | 78ddc53 | 2016-01-15 16:52:42 -0800 | [diff] [blame] | 1243 | split_huge_pmd(vma, pmd, addr); |
Shaohua Li | f21760b | 2012-01-12 17:19:16 -0800 | [diff] [blame] | 1244 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 1245 | goto next; |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 1246 | /* fall through */ |
| 1247 | } |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 1248 | /* |
| 1249 | * Here there can be other concurrent MADV_DONTNEED or |
| 1250 | * trans huge page faults running, and if the pmd is |
| 1251 | * none or trans huge it can change under us. This is |
| 1252 | * because MADV_DONTNEED holds the mmap_sem in read |
| 1253 | * mode. |
| 1254 | */ |
| 1255 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
| 1256 | goto next; |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1257 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 1258 | next: |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1259 | cond_resched(); |
| 1260 | } while (pmd++, addr = next, addr != end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1261 | |
| 1262 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1263 | } |
| 1264 | |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1265 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 1266 | struct vm_area_struct *vma, pgd_t *pgd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | unsigned long addr, unsigned long end, |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1268 | struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1269 | { |
| 1270 | pud_t *pud; |
| 1271 | unsigned long next; |
| 1272 | |
| 1273 | pud = pud_offset(pgd, addr); |
| 1274 | do { |
| 1275 | next = pud_addr_end(addr, end); |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1276 | if (pud_none_or_clear_bad(pud)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | continue; |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1278 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
| 1279 | } while (pud++, addr = next, addr != end); |
Robin Holt | 51c6f66 | 2005-11-13 16:06:42 -0800 | [diff] [blame] | 1280 | |
| 1281 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | } |
| 1283 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 1284 | void unmap_page_range(struct mmu_gather *tlb, |
Al Viro | 038c7aa | 2012-03-05 13:25:09 -0500 | [diff] [blame] | 1285 | struct vm_area_struct *vma, |
| 1286 | unsigned long addr, unsigned long end, |
| 1287 | struct zap_details *details) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | { |
| 1289 | pgd_t *pgd; |
| 1290 | unsigned long next; |
| 1291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | BUG_ON(addr >= end); |
| 1293 | tlb_start_vma(tlb, vma); |
| 1294 | pgd = pgd_offset(vma->vm_mm, addr); |
| 1295 | do { |
| 1296 | next = pgd_addr_end(addr, end); |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1297 | if (pgd_none_or_clear_bad(pgd)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1298 | continue; |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 1299 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
| 1300 | } while (pgd++, addr = next, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1301 | tlb_end_vma(tlb, vma); |
| 1302 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1304 | |
| 1305 | static void unmap_single_vma(struct mmu_gather *tlb, |
| 1306 | struct vm_area_struct *vma, unsigned long start_addr, |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 1307 | unsigned long end_addr, |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1308 | struct zap_details *details) |
| 1309 | { |
| 1310 | unsigned long start = max(vma->vm_start, start_addr); |
| 1311 | unsigned long end; |
| 1312 | |
| 1313 | if (start >= vma->vm_end) |
| 1314 | return; |
| 1315 | end = min(vma->vm_end, end_addr); |
| 1316 | if (end <= vma->vm_start) |
| 1317 | return; |
| 1318 | |
Srikar Dronamraju | cbc91f7 | 2012-04-11 16:05:27 +0530 | [diff] [blame] | 1319 | if (vma->vm_file) |
| 1320 | uprobe_munmap(vma, start, end); |
| 1321 | |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1322 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
Suresh Siddha | 5180da4 | 2012-10-08 16:28:29 -0700 | [diff] [blame] | 1323 | untrack_pfn(vma, 0, 0); |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1324 | |
| 1325 | if (start != end) { |
| 1326 | if (unlikely(is_vm_hugetlb_page(vma))) { |
| 1327 | /* |
| 1328 | * It is undesirable to test vma->vm_file as it |
| 1329 | * should be non-null for valid hugetlb area. |
| 1330 | * However, vm_file will be NULL in the error |
Davidlohr Bueso | 7aa6b4a | 2014-04-07 15:37:01 -0700 | [diff] [blame] | 1331 | * cleanup path of mmap_region. When |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1332 | * hugetlbfs ->mmap method fails, |
Davidlohr Bueso | 7aa6b4a | 2014-04-07 15:37:01 -0700 | [diff] [blame] | 1333 | * mmap_region() nullifies vma->vm_file |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1334 | * before calling this function to clean up. |
| 1335 | * Since no pte has actually been setup, it is |
| 1336 | * safe to do nothing in this case. |
| 1337 | */ |
Aneesh Kumar K.V | 24669e5 | 2012-07-31 16:42:03 -0700 | [diff] [blame] | 1338 | if (vma->vm_file) { |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 1339 | i_mmap_lock_write(vma->vm_file->f_mapping); |
Mel Gorman | d833352 | 2012-07-31 16:46:20 -0700 | [diff] [blame] | 1340 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 1341 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
Aneesh Kumar K.V | 24669e5 | 2012-07-31 16:42:03 -0700 | [diff] [blame] | 1342 | } |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1343 | } else |
| 1344 | unmap_page_range(tlb, vma, start, end, details); |
| 1345 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | } |
| 1347 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | /** |
| 1349 | * unmap_vmas - unmap a range of memory covered by a list of vma's |
Randy Dunlap | 0164f69 | 2011-06-15 15:08:09 -0700 | [diff] [blame] | 1350 | * @tlb: address of the caller's struct mmu_gather |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | * @vma: the starting vma |
| 1352 | * @start_addr: virtual address at which to start unmapping |
| 1353 | * @end_addr: virtual address at which to end unmapping |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | * |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 1355 | * Unmap all pages in the vma list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | * Only addresses between `start' and `end' will be unmapped. |
| 1358 | * |
| 1359 | * The VMA list must be sorted in ascending virtual address order. |
| 1360 | * |
| 1361 | * unmap_vmas() assumes that the caller will flush the whole unmapped address |
| 1362 | * range after unmap_vmas() returns. So the only responsibility here is to |
| 1363 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() |
| 1364 | * drops the lock and schedules. |
| 1365 | */ |
Al Viro | 6e8bb01 | 2012-03-05 13:41:15 -0500 | [diff] [blame] | 1366 | void unmap_vmas(struct mmu_gather *tlb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1367 | struct vm_area_struct *vma, unsigned long start_addr, |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 1368 | unsigned long end_addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | { |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1370 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1371 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1372 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1373 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 1374 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1375 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | } |
| 1377 | |
| 1378 | /** |
| 1379 | * zap_page_range - remove user pages in a given range |
| 1380 | * @vma: vm_area_struct holding the applicable pages |
Randy Dunlap | eb4546b | 2012-06-20 12:53:02 -0700 | [diff] [blame] | 1381 | * @start: starting address of pages to zap |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | * @size: number of bytes to zap |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1383 | * @details: details of shared cache invalidation |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1384 | * |
| 1385 | * Caller must protect the VMA list |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1386 | */ |
Linus Torvalds | 7e027b1 | 2012-05-06 13:43:15 -0700 | [diff] [blame] | 1387 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | unsigned long size, struct zap_details *details) |
| 1389 | { |
| 1390 | struct mm_struct *mm = vma->vm_mm; |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 1391 | struct mmu_gather tlb; |
Linus Torvalds | 7e027b1 | 2012-05-06 13:43:15 -0700 | [diff] [blame] | 1392 | unsigned long end = start + size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | lru_add_drain(); |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 1395 | tlb_gather_mmu(&tlb, mm, start, end); |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 1396 | update_hiwater_rss(mm); |
Linus Torvalds | 7e027b1 | 2012-05-06 13:43:15 -0700 | [diff] [blame] | 1397 | mmu_notifier_invalidate_range_start(mm, start, end); |
| 1398 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 1399 | unmap_single_vma(&tlb, vma, start, end, details); |
Linus Torvalds | 7e027b1 | 2012-05-06 13:43:15 -0700 | [diff] [blame] | 1400 | mmu_notifier_invalidate_range_end(mm, start, end); |
| 1401 | tlb_finish_mmu(&tlb, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1402 | } |
| 1403 | |
Jack Steiner | c627f9c | 2008-07-29 22:33:53 -0700 | [diff] [blame] | 1404 | /** |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1405 | * zap_page_range_single - remove user pages in a given range |
| 1406 | * @vma: vm_area_struct holding the applicable pages |
| 1407 | * @address: starting address of pages to zap |
| 1408 | * @size: number of bytes to zap |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 1409 | * @details: details of shared cache invalidation |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1410 | * |
| 1411 | * The range must fit into one VMA. |
| 1412 | */ |
| 1413 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
| 1414 | unsigned long size, struct zap_details *details) |
| 1415 | { |
| 1416 | struct mm_struct *mm = vma->vm_mm; |
| 1417 | struct mmu_gather tlb; |
| 1418 | unsigned long end = address + size; |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1419 | |
| 1420 | lru_add_drain(); |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 1421 | tlb_gather_mmu(&tlb, mm, address, end); |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1422 | update_hiwater_rss(mm); |
| 1423 | mmu_notifier_invalidate_range_start(mm, address, end); |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 1424 | unmap_single_vma(&tlb, vma, address, end, details); |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1425 | mmu_notifier_invalidate_range_end(mm, address, end); |
| 1426 | tlb_finish_mmu(&tlb, address, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1427 | } |
| 1428 | |
Jack Steiner | c627f9c | 2008-07-29 22:33:53 -0700 | [diff] [blame] | 1429 | /** |
| 1430 | * zap_vma_ptes - remove ptes mapping the vma |
| 1431 | * @vma: vm_area_struct holding ptes to be zapped |
| 1432 | * @address: starting address of pages to zap |
| 1433 | * @size: number of bytes to zap |
| 1434 | * |
| 1435 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. |
| 1436 | * |
| 1437 | * The entire address range must be fully contained within the vma. |
| 1438 | * |
| 1439 | * Returns 0 if successful. |
| 1440 | */ |
| 1441 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
| 1442 | unsigned long size) |
| 1443 | { |
| 1444 | if (address < vma->vm_start || address + size > vma->vm_end || |
| 1445 | !(vma->vm_flags & VM_PFNMAP)) |
| 1446 | return -1; |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 1447 | zap_page_range_single(vma, address, size, NULL); |
Jack Steiner | c627f9c | 2008-07-29 22:33:53 -0700 | [diff] [blame] | 1448 | return 0; |
| 1449 | } |
| 1450 | EXPORT_SYMBOL_GPL(zap_vma_ptes); |
| 1451 | |
Namhyung Kim | 25ca1d6 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 1452 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 1453 | spinlock_t **ptl) |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1454 | { |
| 1455 | pgd_t * pgd = pgd_offset(mm, addr); |
| 1456 | pud_t * pud = pud_alloc(mm, pgd, addr); |
| 1457 | if (pud) { |
Trond Myklebust | 49c91fb | 2005-11-29 19:27:22 -0500 | [diff] [blame] | 1458 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
Andrea Arcangeli | f66055ab | 2011-01-13 15:46:54 -0800 | [diff] [blame] | 1459 | if (pmd) { |
| 1460 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1461 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
Andrea Arcangeli | f66055ab | 2011-01-13 15:46:54 -0800 | [diff] [blame] | 1462 | } |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1463 | } |
| 1464 | return NULL; |
| 1465 | } |
| 1466 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | /* |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1468 | * This is the old fallback for page remapping. |
| 1469 | * |
| 1470 | * For historical reasons, it only allows reserved pages. Only |
| 1471 | * old drivers should use this, and they needed to mark their |
| 1472 | * pages reserved for the old functions anyway. |
| 1473 | */ |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1474 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1475 | struct page *page, pgprot_t prot) |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1476 | { |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1477 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1478 | int retval; |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1479 | pte_t *pte; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1480 | spinlock_t *ptl; |
| 1481 | |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1482 | retval = -EINVAL; |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1483 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 1484 | goto out; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1485 | retval = -ENOMEM; |
| 1486 | flush_dcache_page(page); |
Linus Torvalds | c9cfcdd | 2005-11-29 14:03:14 -0800 | [diff] [blame] | 1487 | pte = get_locked_pte(mm, addr, &ptl); |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1488 | if (!pte) |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 1489 | goto out; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1490 | retval = -EBUSY; |
| 1491 | if (!pte_none(*pte)) |
| 1492 | goto out_unlock; |
| 1493 | |
| 1494 | /* Ok, finally just insert the thing.. */ |
| 1495 | get_page(page); |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 1496 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
Kirill A. Shutemov | dd78fed | 2016-07-26 15:25:26 -0700 | [diff] [blame] | 1497 | page_add_file_rmap(page, false); |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1498 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
| 1499 | |
| 1500 | retval = 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1501 | pte_unmap_unlock(pte, ptl); |
| 1502 | return retval; |
Linus Torvalds | 238f58d | 2005-11-29 13:01:56 -0800 | [diff] [blame] | 1503 | out_unlock: |
| 1504 | pte_unmap_unlock(pte, ptl); |
| 1505 | out: |
| 1506 | return retval; |
| 1507 | } |
| 1508 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1509 | /** |
| 1510 | * vm_insert_page - insert single page into user vma |
| 1511 | * @vma: user vma to map to |
| 1512 | * @addr: target user address of this page |
| 1513 | * @page: source kernel page |
| 1514 | * |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1515 | * This allows drivers to insert individual pages they've allocated |
| 1516 | * into a user vma. |
| 1517 | * |
| 1518 | * The page has to be a nice clean _individual_ kernel allocation. |
| 1519 | * If you allocate a compound page, you need to have marked it as |
| 1520 | * such (__GFP_COMP), or manually just split the page up yourself |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1521 | * (see split_page()). |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1522 | * |
| 1523 | * NOTE! Traditionally this was done with "remap_pfn_range()" which |
| 1524 | * took an arbitrary page protection parameter. This doesn't allow |
| 1525 | * that. Your vma protection will have to be set up correctly, which |
| 1526 | * means that if you want a shared writable mapping, you'd better |
| 1527 | * ask for a shared writable mapping! |
| 1528 | * |
| 1529 | * The page does not need to be reserved. |
Konstantin Khlebnikov | 4b6e1e3 | 2012-10-08 16:28:40 -0700 | [diff] [blame] | 1530 | * |
| 1531 | * Usually this function is called from f_op->mmap() handler |
| 1532 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. |
| 1533 | * Caller must set VM_MIXEDMAP on vma if it wants to call this |
| 1534 | * function from other places, for example from page-fault handler. |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1535 | */ |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1536 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 1537 | struct page *page) |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1538 | { |
| 1539 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1540 | return -EFAULT; |
| 1541 | if (!page_count(page)) |
| 1542 | return -EINVAL; |
Konstantin Khlebnikov | 4b6e1e3 | 2012-10-08 16:28:40 -0700 | [diff] [blame] | 1543 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
| 1544 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); |
| 1545 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
| 1546 | vma->vm_flags |= VM_MIXEDMAP; |
| 1547 | } |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1548 | return insert_page(vma, addr, page, vma->vm_page_prot); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1549 | } |
Linus Torvalds | e3c3374 | 2005-12-03 20:48:11 -0800 | [diff] [blame] | 1550 | EXPORT_SYMBOL(vm_insert_page); |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1551 | |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1552 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
Dan Williams | 01c8f1c | 2016-01-15 16:56:40 -0800 | [diff] [blame] | 1553 | pfn_t pfn, pgprot_t prot) |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1554 | { |
| 1555 | struct mm_struct *mm = vma->vm_mm; |
| 1556 | int retval; |
| 1557 | pte_t *pte, entry; |
| 1558 | spinlock_t *ptl; |
| 1559 | |
| 1560 | retval = -ENOMEM; |
| 1561 | pte = get_locked_pte(mm, addr, &ptl); |
| 1562 | if (!pte) |
| 1563 | goto out; |
| 1564 | retval = -EBUSY; |
| 1565 | if (!pte_none(*pte)) |
| 1566 | goto out_unlock; |
| 1567 | |
| 1568 | /* Ok, finally just insert the thing.. */ |
Dan Williams | 01c8f1c | 2016-01-15 16:56:40 -0800 | [diff] [blame] | 1569 | if (pfn_t_devmap(pfn)) |
| 1570 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); |
| 1571 | else |
| 1572 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1573 | set_pte_at(mm, addr, pte, entry); |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 1574 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1575 | |
| 1576 | retval = 0; |
| 1577 | out_unlock: |
| 1578 | pte_unmap_unlock(pte, ptl); |
| 1579 | out: |
| 1580 | return retval; |
| 1581 | } |
| 1582 | |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1583 | /** |
| 1584 | * vm_insert_pfn - insert single pfn into user vma |
| 1585 | * @vma: user vma to map to |
| 1586 | * @addr: target user address of this page |
| 1587 | * @pfn: source kernel pfn |
| 1588 | * |
Robert P. J. Day | c462f17 | 2012-10-08 16:33:43 -0700 | [diff] [blame] | 1589 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1590 | * they've allocated into a user vma. Same comments apply. |
| 1591 | * |
| 1592 | * This function should only be called from a vm_ops->fault handler, and |
| 1593 | * in that case the handler should return NULL. |
Nick Piggin | 0d71d10a | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 1594 | * |
| 1595 | * vma cannot be a COW mapping. |
| 1596 | * |
| 1597 | * As this is called only for pages that do not currently exist, we |
| 1598 | * do not need to flush old virtual caches or the TLB. |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1599 | */ |
| 1600 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1601 | unsigned long pfn) |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1602 | { |
Andy Lutomirski | 1745cbc | 2015-12-29 20:12:20 -0800 | [diff] [blame] | 1603 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); |
| 1604 | } |
| 1605 | EXPORT_SYMBOL(vm_insert_pfn); |
| 1606 | |
| 1607 | /** |
| 1608 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot |
| 1609 | * @vma: user vma to map to |
| 1610 | * @addr: target user address of this page |
| 1611 | * @pfn: source kernel pfn |
| 1612 | * @pgprot: pgprot flags for the inserted page |
| 1613 | * |
| 1614 | * This is exactly like vm_insert_pfn, except that it allows drivers to |
| 1615 | * to override pgprot on a per-page basis. |
| 1616 | * |
| 1617 | * This only makes sense for IO mappings, and it makes no sense for |
| 1618 | * cow mappings. In general, using multiple vmas is preferable; |
| 1619 | * vm_insert_pfn_prot should only be used if using multiple VMAs is |
| 1620 | * impractical. |
| 1621 | */ |
| 1622 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, |
| 1623 | unsigned long pfn, pgprot_t pgprot) |
| 1624 | { |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1625 | int ret; |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1626 | /* |
| 1627 | * Technically, architectures with pte_special can avoid all these |
| 1628 | * restrictions (same for remap_pfn_range). However we would like |
| 1629 | * consistency in testing and feature parity among all, so we should |
| 1630 | * try to keep these invariants in place for everybody. |
| 1631 | */ |
Jared Hulbert | b379d79 | 2008-04-28 02:12:58 -0700 | [diff] [blame] | 1632 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
| 1633 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == |
| 1634 | (VM_PFNMAP|VM_MIXEDMAP)); |
| 1635 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); |
| 1636 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1637 | |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1638 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1639 | return -EFAULT; |
Dan Williams | f25748e3 | 2016-01-15 16:56:43 -0800 | [diff] [blame] | 1640 | if (track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV))) |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1641 | return -EINVAL; |
| 1642 | |
Dan Williams | 01c8f1c | 2016-01-15 16:56:40 -0800 | [diff] [blame] | 1643 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1644 | |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1645 | return ret; |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1646 | } |
Andy Lutomirski | 1745cbc | 2015-12-29 20:12:20 -0800 | [diff] [blame] | 1647 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
Nick Piggin | e0dc0d8 | 2007-02-12 00:51:36 -0800 | [diff] [blame] | 1648 | |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1649 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
Dan Williams | 01c8f1c | 2016-01-15 16:56:40 -0800 | [diff] [blame] | 1650 | pfn_t pfn) |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1651 | { |
Dan Williams | 87744ab | 2016-10-07 17:00:18 -0700 | [diff] [blame] | 1652 | pgprot_t pgprot = vma->vm_page_prot; |
| 1653 | |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1654 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
| 1655 | |
| 1656 | if (addr < vma->vm_start || addr >= vma->vm_end) |
| 1657 | return -EFAULT; |
Dan Williams | 87744ab | 2016-10-07 17:00:18 -0700 | [diff] [blame] | 1658 | if (track_pfn_insert(vma, &pgprot, pfn)) |
| 1659 | return -EINVAL; |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1660 | |
| 1661 | /* |
| 1662 | * If we don't have pte special, then we have to use the pfn_valid() |
| 1663 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* |
| 1664 | * refcount the page if pfn_valid is true (hence insert_page rather |
Hugh Dickins | 62eede6 | 2009-09-21 17:03:34 -0700 | [diff] [blame] | 1665 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
| 1666 | * without pte special, it would there be refcounted as a normal page. |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1667 | */ |
Dan Williams | 03fc2da | 2016-01-26 09:48:05 -0800 | [diff] [blame] | 1668 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1669 | struct page *page; |
| 1670 | |
Dan Williams | 03fc2da | 2016-01-26 09:48:05 -0800 | [diff] [blame] | 1671 | /* |
| 1672 | * At this point we are committed to insert_page() |
| 1673 | * regardless of whether the caller specified flags that |
| 1674 | * result in pfn_t_has_page() == false. |
| 1675 | */ |
| 1676 | page = pfn_to_page(pfn_t_to_pfn(pfn)); |
Dan Williams | 87744ab | 2016-10-07 17:00:18 -0700 | [diff] [blame] | 1677 | return insert_page(vma, addr, page, pgprot); |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1678 | } |
Dan Williams | 87744ab | 2016-10-07 17:00:18 -0700 | [diff] [blame] | 1679 | return insert_pfn(vma, addr, pfn, pgprot); |
Nick Piggin | 423bad60 | 2008-04-28 02:13:01 -0700 | [diff] [blame] | 1680 | } |
| 1681 | EXPORT_SYMBOL(vm_insert_mixed); |
| 1682 | |
Linus Torvalds | a145dd4 | 2005-11-30 09:35:19 -0800 | [diff] [blame] | 1683 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | * maps a range of physical memory into the requested pages. the old |
| 1685 | * mappings are removed. any references to nonexistent pages results |
| 1686 | * in null mappings (currently treated as "copy-on-access") |
| 1687 | */ |
| 1688 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1689 | unsigned long addr, unsigned long end, |
| 1690 | unsigned long pfn, pgprot_t prot) |
| 1691 | { |
| 1692 | pte_t *pte; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1693 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1695 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | if (!pte) |
| 1697 | return -ENOMEM; |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1698 | arch_enter_lazy_mmu_mode(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | do { |
| 1700 | BUG_ON(!pte_none(*pte)); |
Nick Piggin | 7e67513 | 2008-04-28 02:13:00 -0700 | [diff] [blame] | 1701 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1702 | pfn++; |
| 1703 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Zachary Amsden | 6606c3e | 2006-09-30 23:29:33 -0700 | [diff] [blame] | 1704 | arch_leave_lazy_mmu_mode(); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 1705 | pte_unmap_unlock(pte - 1, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1706 | return 0; |
| 1707 | } |
| 1708 | |
| 1709 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1710 | unsigned long addr, unsigned long end, |
| 1711 | unsigned long pfn, pgprot_t prot) |
| 1712 | { |
| 1713 | pmd_t *pmd; |
| 1714 | unsigned long next; |
| 1715 | |
| 1716 | pfn -= addr >> PAGE_SHIFT; |
| 1717 | pmd = pmd_alloc(mm, pud, addr); |
| 1718 | if (!pmd) |
| 1719 | return -ENOMEM; |
Andrea Arcangeli | f66055ab | 2011-01-13 15:46:54 -0800 | [diff] [blame] | 1720 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1721 | do { |
| 1722 | next = pmd_addr_end(addr, end); |
| 1723 | if (remap_pte_range(mm, pmd, addr, next, |
| 1724 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1725 | return -ENOMEM; |
| 1726 | } while (pmd++, addr = next, addr != end); |
| 1727 | return 0; |
| 1728 | } |
| 1729 | |
| 1730 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1731 | unsigned long addr, unsigned long end, |
| 1732 | unsigned long pfn, pgprot_t prot) |
| 1733 | { |
| 1734 | pud_t *pud; |
| 1735 | unsigned long next; |
| 1736 | |
| 1737 | pfn -= addr >> PAGE_SHIFT; |
| 1738 | pud = pud_alloc(mm, pgd, addr); |
| 1739 | if (!pud) |
| 1740 | return -ENOMEM; |
| 1741 | do { |
| 1742 | next = pud_addr_end(addr, end); |
| 1743 | if (remap_pmd_range(mm, pud, addr, next, |
| 1744 | pfn + (addr >> PAGE_SHIFT), prot)) |
| 1745 | return -ENOMEM; |
| 1746 | } while (pud++, addr = next, addr != end); |
| 1747 | return 0; |
| 1748 | } |
| 1749 | |
Rolf Eike Beer | bfa5bf6 | 2006-09-25 23:31:22 -0700 | [diff] [blame] | 1750 | /** |
| 1751 | * remap_pfn_range - remap kernel memory to userspace |
| 1752 | * @vma: user vma to map to |
| 1753 | * @addr: target user address to start at |
| 1754 | * @pfn: physical address of kernel memory |
| 1755 | * @size: size of map area |
| 1756 | * @prot: page protection flags for this mapping |
| 1757 | * |
| 1758 | * Note: this is only safe if the mm semaphore is held when called. |
| 1759 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
| 1761 | unsigned long pfn, unsigned long size, pgprot_t prot) |
| 1762 | { |
| 1763 | pgd_t *pgd; |
| 1764 | unsigned long next; |
Hugh Dickins | 2d15cab | 2005-06-25 14:54:33 -0700 | [diff] [blame] | 1765 | unsigned long end = addr + PAGE_ALIGN(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1766 | struct mm_struct *mm = vma->vm_mm; |
Yongji Xie | d5957d2 | 2016-05-20 16:57:41 -0700 | [diff] [blame] | 1767 | unsigned long remap_pfn = pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1768 | int err; |
| 1769 | |
| 1770 | /* |
| 1771 | * Physically remapped pages are special. Tell the |
| 1772 | * rest of the world about it: |
| 1773 | * VM_IO tells people not to look at these pages |
| 1774 | * (accesses can have side effects). |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1775 | * VM_PFNMAP tells the core MM that the base pages are just |
| 1776 | * raw PFN mappings, and do not have a "struct page" associated |
| 1777 | * with them. |
Konstantin Khlebnikov | 314e51b | 2012-10-08 16:29:02 -0700 | [diff] [blame] | 1778 | * VM_DONTEXPAND |
| 1779 | * Disable vma merging and expanding with mremap(). |
| 1780 | * VM_DONTDUMP |
| 1781 | * Omit vma from core dump, even when VM_IO turned off. |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1782 | * |
| 1783 | * There's a horrible special case to handle copy-on-write |
| 1784 | * behaviour that some programs depend on. We mark the "original" |
| 1785 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1786 | * See vm_normal_page() for details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | */ |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1788 | if (is_cow_mapping(vma->vm_flags)) { |
| 1789 | if (addr != vma->vm_start || end != vma->vm_end) |
| 1790 | return -EINVAL; |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1791 | vma->vm_pgoff = pfn; |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1792 | } |
| 1793 | |
Yongji Xie | d5957d2 | 2016-05-20 16:57:41 -0700 | [diff] [blame] | 1794 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
Konstantin Khlebnikov | b3b9c29 | 2012-10-08 16:28:34 -0700 | [diff] [blame] | 1795 | if (err) |
venkatesh.pallipadi@intel.com | 3c8bb73 | 2008-12-18 11:41:27 -0800 | [diff] [blame] | 1796 | return -EINVAL; |
Linus Torvalds | fb155c1 | 2005-12-11 19:46:02 -0800 | [diff] [blame] | 1797 | |
Konstantin Khlebnikov | 314e51b | 2012-10-08 16:29:02 -0700 | [diff] [blame] | 1798 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | |
| 1800 | BUG_ON(addr >= end); |
| 1801 | pfn -= addr >> PAGE_SHIFT; |
| 1802 | pgd = pgd_offset(mm, addr); |
| 1803 | flush_cache_range(vma, addr, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1804 | do { |
| 1805 | next = pgd_addr_end(addr, end); |
| 1806 | err = remap_pud_range(mm, pgd, addr, next, |
| 1807 | pfn + (addr >> PAGE_SHIFT), prot); |
| 1808 | if (err) |
| 1809 | break; |
| 1810 | } while (pgd++, addr = next, addr != end); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1811 | |
| 1812 | if (err) |
Yongji Xie | d5957d2 | 2016-05-20 16:57:41 -0700 | [diff] [blame] | 1813 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
venkatesh.pallipadi@intel.com | 2ab6403 | 2008-12-18 11:41:29 -0800 | [diff] [blame] | 1814 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1815 | return err; |
| 1816 | } |
| 1817 | EXPORT_SYMBOL(remap_pfn_range); |
| 1818 | |
Linus Torvalds | b4cbb19 | 2013-04-16 13:45:37 -0700 | [diff] [blame] | 1819 | /** |
| 1820 | * vm_iomap_memory - remap memory to userspace |
| 1821 | * @vma: user vma to map to |
| 1822 | * @start: start of area |
| 1823 | * @len: size of area |
| 1824 | * |
| 1825 | * This is a simplified io_remap_pfn_range() for common driver use. The |
| 1826 | * driver just needs to give us the physical memory range to be mapped, |
| 1827 | * we'll figure out the rest from the vma information. |
| 1828 | * |
| 1829 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get |
| 1830 | * whatever write-combining details or similar. |
| 1831 | */ |
| 1832 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
| 1833 | { |
| 1834 | unsigned long vm_len, pfn, pages; |
| 1835 | |
| 1836 | /* Check that the physical memory area passed in looks valid */ |
| 1837 | if (start + len < start) |
| 1838 | return -EINVAL; |
| 1839 | /* |
| 1840 | * You *really* shouldn't map things that aren't page-aligned, |
| 1841 | * but we've historically allowed it because IO memory might |
| 1842 | * just have smaller alignment. |
| 1843 | */ |
| 1844 | len += start & ~PAGE_MASK; |
| 1845 | pfn = start >> PAGE_SHIFT; |
| 1846 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; |
| 1847 | if (pfn + pages < pfn) |
| 1848 | return -EINVAL; |
| 1849 | |
| 1850 | /* We start the mapping 'vm_pgoff' pages into the area */ |
| 1851 | if (vma->vm_pgoff > pages) |
| 1852 | return -EINVAL; |
| 1853 | pfn += vma->vm_pgoff; |
| 1854 | pages -= vma->vm_pgoff; |
| 1855 | |
| 1856 | /* Can we fit all of the mapping? */ |
| 1857 | vm_len = vma->vm_end - vma->vm_start; |
| 1858 | if (vm_len >> PAGE_SHIFT > pages) |
| 1859 | return -EINVAL; |
| 1860 | |
| 1861 | /* Ok, let it rip */ |
| 1862 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); |
| 1863 | } |
| 1864 | EXPORT_SYMBOL(vm_iomap_memory); |
| 1865 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1866 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
| 1867 | unsigned long addr, unsigned long end, |
| 1868 | pte_fn_t fn, void *data) |
| 1869 | { |
| 1870 | pte_t *pte; |
| 1871 | int err; |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1872 | pgtable_t token; |
Borislav Petkov | 9490991 | 2007-05-06 14:49:17 -0700 | [diff] [blame] | 1873 | spinlock_t *uninitialized_var(ptl); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1874 | |
| 1875 | pte = (mm == &init_mm) ? |
| 1876 | pte_alloc_kernel(pmd, addr) : |
| 1877 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
| 1878 | if (!pte) |
| 1879 | return -ENOMEM; |
| 1880 | |
| 1881 | BUG_ON(pmd_huge(*pmd)); |
| 1882 | |
Jeremy Fitzhardinge | 38e0edb | 2009-01-06 14:39:21 -0800 | [diff] [blame] | 1883 | arch_enter_lazy_mmu_mode(); |
| 1884 | |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 1885 | token = pmd_pgtable(*pmd); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1886 | |
| 1887 | do { |
Daisuke Nishimura | c36987e | 2009-10-26 16:50:23 -0700 | [diff] [blame] | 1888 | err = fn(pte++, token, addr, data); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1889 | if (err) |
| 1890 | break; |
Daisuke Nishimura | c36987e | 2009-10-26 16:50:23 -0700 | [diff] [blame] | 1891 | } while (addr += PAGE_SIZE, addr != end); |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1892 | |
Jeremy Fitzhardinge | 38e0edb | 2009-01-06 14:39:21 -0800 | [diff] [blame] | 1893 | arch_leave_lazy_mmu_mode(); |
| 1894 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1895 | if (mm != &init_mm) |
| 1896 | pte_unmap_unlock(pte-1, ptl); |
| 1897 | return err; |
| 1898 | } |
| 1899 | |
| 1900 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, |
| 1901 | unsigned long addr, unsigned long end, |
| 1902 | pte_fn_t fn, void *data) |
| 1903 | { |
| 1904 | pmd_t *pmd; |
| 1905 | unsigned long next; |
| 1906 | int err; |
| 1907 | |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 1908 | BUG_ON(pud_huge(*pud)); |
| 1909 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1910 | pmd = pmd_alloc(mm, pud, addr); |
| 1911 | if (!pmd) |
| 1912 | return -ENOMEM; |
| 1913 | do { |
| 1914 | next = pmd_addr_end(addr, end); |
| 1915 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); |
| 1916 | if (err) |
| 1917 | break; |
| 1918 | } while (pmd++, addr = next, addr != end); |
| 1919 | return err; |
| 1920 | } |
| 1921 | |
| 1922 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, |
| 1923 | unsigned long addr, unsigned long end, |
| 1924 | pte_fn_t fn, void *data) |
| 1925 | { |
| 1926 | pud_t *pud; |
| 1927 | unsigned long next; |
| 1928 | int err; |
| 1929 | |
| 1930 | pud = pud_alloc(mm, pgd, addr); |
| 1931 | if (!pud) |
| 1932 | return -ENOMEM; |
| 1933 | do { |
| 1934 | next = pud_addr_end(addr, end); |
| 1935 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); |
| 1936 | if (err) |
| 1937 | break; |
| 1938 | } while (pud++, addr = next, addr != end); |
| 1939 | return err; |
| 1940 | } |
| 1941 | |
| 1942 | /* |
| 1943 | * Scan a region of virtual memory, filling in page tables as necessary |
| 1944 | * and calling a provided function on each leaf page table. |
| 1945 | */ |
| 1946 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
| 1947 | unsigned long size, pte_fn_t fn, void *data) |
| 1948 | { |
| 1949 | pgd_t *pgd; |
| 1950 | unsigned long next; |
Jeremy Fitzhardinge | 57250a5 | 2010-08-09 17:19:52 -0700 | [diff] [blame] | 1951 | unsigned long end = addr + size; |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1952 | int err; |
| 1953 | |
Mika Penttilä | 9cb65bc | 2016-03-15 14:56:45 -0700 | [diff] [blame] | 1954 | if (WARN_ON(addr >= end)) |
| 1955 | return -EINVAL; |
| 1956 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1957 | pgd = pgd_offset(mm, addr); |
| 1958 | do { |
| 1959 | next = pgd_addr_end(addr, end); |
| 1960 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); |
| 1961 | if (err) |
| 1962 | break; |
| 1963 | } while (pgd++, addr = next, addr != end); |
Jeremy Fitzhardinge | 57250a5 | 2010-08-09 17:19:52 -0700 | [diff] [blame] | 1964 | |
Jeremy Fitzhardinge | aee16b3 | 2007-05-06 14:48:54 -0700 | [diff] [blame] | 1965 | return err; |
| 1966 | } |
| 1967 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
| 1968 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | /* |
Kirill A. Shutemov | 9b4bdd2 | 2015-02-10 14:09:51 -0800 | [diff] [blame] | 1970 | * handle_pte_fault chooses page fault handler according to an entry which was |
| 1971 | * read non-atomically. Before making any commitment, on those architectures |
| 1972 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched |
| 1973 | * parts, do_swap_page must check under lock before unmapping the pte and |
| 1974 | * proceeding (but do_wp_page is only called after already making such a check; |
Ryota Ozaki | a335b2e | 2011-02-10 13:56:28 +0900 | [diff] [blame] | 1975 | * and do_anonymous_page can safely check later on). |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1976 | */ |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1977 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1978 | pte_t *page_table, pte_t orig_pte) |
| 1979 | { |
| 1980 | int same = 1; |
| 1981 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) |
| 1982 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1983 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
| 1984 | spin_lock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1985 | same = pte_same(*page_table, orig_pte); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1986 | spin_unlock(ptl); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 1987 | } |
| 1988 | #endif |
| 1989 | pte_unmap(page_table); |
| 1990 | return same; |
| 1991 | } |
| 1992 | |
Atsushi Nemoto | 9de455b | 2006-12-12 17:14:55 +0000 | [diff] [blame] | 1993 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1994 | { |
Dan Williams | 0abdd7a | 2014-01-21 15:48:12 -0800 | [diff] [blame] | 1995 | debug_dma_assert_idle(src); |
| 1996 | |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1997 | /* |
| 1998 | * If the source page was a PFN mapping, we don't have |
| 1999 | * a "struct page" for it. We do a best-effort copy by |
| 2000 | * just copying from the original user address. If that |
| 2001 | * fails, we just zero-fill it. Live with it. |
| 2002 | */ |
| 2003 | if (unlikely(!src)) { |
Cong Wang | 9b04c5f | 2011-11-25 23:14:39 +0800 | [diff] [blame] | 2004 | void *kaddr = kmap_atomic(dst); |
Linus Torvalds | 5d2a2dbbc | 2005-11-29 14:07:55 -0800 | [diff] [blame] | 2005 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
| 2006 | |
| 2007 | /* |
| 2008 | * This really shouldn't fail, because the page is there |
| 2009 | * in the page tables. But it might just be unreadable, |
| 2010 | * in which case we just give up and fill the result with |
| 2011 | * zeroes. |
| 2012 | */ |
| 2013 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) |
Jan Beulich | 3ecb01d | 2010-10-26 14:22:27 -0700 | [diff] [blame] | 2014 | clear_page(kaddr); |
Cong Wang | 9b04c5f | 2011-11-25 23:14:39 +0800 | [diff] [blame] | 2015 | kunmap_atomic(kaddr); |
Dmitriy Monakhov | c4ec7b0 | 2006-10-19 23:29:08 -0700 | [diff] [blame] | 2016 | flush_dcache_page(dst); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2017 | } else |
| 2018 | copy_user_highpage(dst, src, va, vma); |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 2019 | } |
| 2020 | |
Michal Hocko | c20cd45 | 2016-01-14 15:20:12 -0800 | [diff] [blame] | 2021 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
| 2022 | { |
| 2023 | struct file *vm_file = vma->vm_file; |
| 2024 | |
| 2025 | if (vm_file) |
| 2026 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; |
| 2027 | |
| 2028 | /* |
| 2029 | * Special mappings (e.g. VDSO) do not have any file so fake |
| 2030 | * a default GFP_KERNEL for them. |
| 2031 | */ |
| 2032 | return GFP_KERNEL; |
| 2033 | } |
| 2034 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2035 | /* |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 2036 | * Notify the address space that the page is about to become writable so that |
| 2037 | * it can prohibit this or wait for the page to get into an appropriate state. |
| 2038 | * |
| 2039 | * We do this without the lock held, so that it can sleep if it needs to. |
| 2040 | */ |
| 2041 | static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, |
| 2042 | unsigned long address) |
| 2043 | { |
| 2044 | struct vm_fault vmf; |
| 2045 | int ret; |
| 2046 | |
| 2047 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
| 2048 | vmf.pgoff = page->index; |
| 2049 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
Michal Hocko | c20cd45 | 2016-01-14 15:20:12 -0800 | [diff] [blame] | 2050 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 2051 | vmf.page = page; |
Matthew Wilcox | 2e4cdab | 2015-02-16 15:58:50 -0800 | [diff] [blame] | 2052 | vmf.cow_page = NULL; |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 2053 | |
| 2054 | ret = vma->vm_ops->page_mkwrite(vma, &vmf); |
| 2055 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
| 2056 | return ret; |
| 2057 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { |
| 2058 | lock_page(page); |
| 2059 | if (!page->mapping) { |
| 2060 | unlock_page(page); |
| 2061 | return 0; /* retry */ |
| 2062 | } |
| 2063 | ret |= VM_FAULT_LOCKED; |
| 2064 | } else |
| 2065 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 2066 | return ret; |
| 2067 | } |
| 2068 | |
| 2069 | /* |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2070 | * Handle write page faults for pages that can be reused in the current vma |
| 2071 | * |
| 2072 | * This can happen either due to the mapping being with the VM_SHARED flag, |
| 2073 | * or due to us being the last reference standing to the page. In either |
| 2074 | * case, all we need to do here is to mark the page as writable and update |
| 2075 | * any related book-keeping. |
| 2076 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2077 | static inline int wp_page_reuse(struct fault_env *fe, pte_t orig_pte, |
| 2078 | struct page *page, int page_mkwrite, int dirty_shared) |
| 2079 | __releases(fe->ptl) |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2080 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2081 | struct vm_area_struct *vma = fe->vma; |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2082 | pte_t entry; |
| 2083 | /* |
| 2084 | * Clear the pages cpupid information as the existing |
| 2085 | * information potentially belongs to a now completely |
| 2086 | * unrelated process. |
| 2087 | */ |
| 2088 | if (page) |
| 2089 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); |
| 2090 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2091 | flush_cache_page(vma, fe->address, pte_pfn(orig_pte)); |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2092 | entry = pte_mkyoung(orig_pte); |
| 2093 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2094 | if (ptep_set_access_flags(vma, fe->address, fe->pte, entry, 1)) |
| 2095 | update_mmu_cache(vma, fe->address, fe->pte); |
| 2096 | pte_unmap_unlock(fe->pte, fe->ptl); |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2097 | |
| 2098 | if (dirty_shared) { |
| 2099 | struct address_space *mapping; |
| 2100 | int dirtied; |
| 2101 | |
| 2102 | if (!page_mkwrite) |
| 2103 | lock_page(page); |
| 2104 | |
| 2105 | dirtied = set_page_dirty(page); |
| 2106 | VM_BUG_ON_PAGE(PageAnon(page), page); |
| 2107 | mapping = page->mapping; |
| 2108 | unlock_page(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2109 | put_page(page); |
Shachar Raindel | 4e047f8 | 2015-04-14 15:46:25 -0700 | [diff] [blame] | 2110 | |
| 2111 | if ((dirtied || page_mkwrite) && mapping) { |
| 2112 | /* |
| 2113 | * Some device drivers do not set page.mapping |
| 2114 | * but still dirty their pages |
| 2115 | */ |
| 2116 | balance_dirty_pages_ratelimited(mapping); |
| 2117 | } |
| 2118 | |
| 2119 | if (!page_mkwrite) |
| 2120 | file_update_time(vma->vm_file); |
| 2121 | } |
| 2122 | |
| 2123 | return VM_FAULT_WRITE; |
| 2124 | } |
| 2125 | |
| 2126 | /* |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2127 | * Handle the case of a page which we actually need to copy to a new page. |
| 2128 | * |
| 2129 | * Called with mmap_sem locked and the old page referenced, but |
| 2130 | * without the ptl held. |
| 2131 | * |
| 2132 | * High level logic flow: |
| 2133 | * |
| 2134 | * - Allocate a page, copy the content of the old page to the new one. |
| 2135 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. |
| 2136 | * - Take the PTL. If the pte changed, bail out and release the allocated page |
| 2137 | * - If the pte is still the way we remember it, update the page table and all |
| 2138 | * relevant references. This includes dropping the reference the page-table |
| 2139 | * held to the old page, as well as updating the rmap. |
| 2140 | * - In any case, unlock the PTL and drop the reference we took to the old page. |
| 2141 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2142 | static int wp_page_copy(struct fault_env *fe, pte_t orig_pte, |
| 2143 | struct page *old_page) |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2144 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2145 | struct vm_area_struct *vma = fe->vma; |
| 2146 | struct mm_struct *mm = vma->vm_mm; |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2147 | struct page *new_page = NULL; |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2148 | pte_t entry; |
| 2149 | int page_copied = 0; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2150 | const unsigned long mmun_start = fe->address & PAGE_MASK; |
| 2151 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2152 | struct mem_cgroup *memcg; |
| 2153 | |
| 2154 | if (unlikely(anon_vma_prepare(vma))) |
| 2155 | goto oom; |
| 2156 | |
| 2157 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2158 | new_page = alloc_zeroed_user_highpage_movable(vma, fe->address); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2159 | if (!new_page) |
| 2160 | goto oom; |
| 2161 | } else { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2162 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
| 2163 | fe->address); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2164 | if (!new_page) |
| 2165 | goto oom; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2166 | cow_user_page(new_page, old_page, fe->address, vma); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2167 | } |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2168 | |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2169 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2170 | goto oom_free_new; |
| 2171 | |
Mel Gorman | eb3c24f | 2015-06-24 16:57:27 -0700 | [diff] [blame] | 2172 | __SetPageUptodate(new_page); |
| 2173 | |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2174 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
| 2175 | |
| 2176 | /* |
| 2177 | * Re-check the pte - we dropped the lock |
| 2178 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2179 | fe->pte = pte_offset_map_lock(mm, fe->pmd, fe->address, &fe->ptl); |
| 2180 | if (likely(pte_same(*fe->pte, orig_pte))) { |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2181 | if (old_page) { |
| 2182 | if (!PageAnon(old_page)) { |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 2183 | dec_mm_counter_fast(mm, |
| 2184 | mm_counter_file(old_page)); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2185 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
| 2186 | } |
| 2187 | } else { |
| 2188 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
| 2189 | } |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2190 | flush_cache_page(vma, fe->address, pte_pfn(orig_pte)); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2191 | entry = mk_pte(new_page, vma->vm_page_prot); |
| 2192 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
| 2193 | /* |
| 2194 | * Clear the pte entry and flush it first, before updating the |
| 2195 | * pte with the new entry. This will avoid a race condition |
| 2196 | * seen in the presence of one thread doing SMC and another |
| 2197 | * thread doing COW. |
| 2198 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2199 | ptep_clear_flush_notify(vma, fe->address, fe->pte); |
| 2200 | page_add_new_anon_rmap(new_page, vma, fe->address, false); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2201 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2202 | lru_cache_add_active_or_unevictable(new_page, vma); |
| 2203 | /* |
| 2204 | * We call the notify macro here because, when using secondary |
| 2205 | * mmu page tables (such as kvm shadow page tables), we want the |
| 2206 | * new page to be mapped directly into the secondary page table. |
| 2207 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2208 | set_pte_at_notify(mm, fe->address, fe->pte, entry); |
| 2209 | update_mmu_cache(vma, fe->address, fe->pte); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2210 | if (old_page) { |
| 2211 | /* |
| 2212 | * Only after switching the pte to the new page may |
| 2213 | * we remove the mapcount here. Otherwise another |
| 2214 | * process may come and find the rmap count decremented |
| 2215 | * before the pte is switched to the new page, and |
| 2216 | * "reuse" the old page writing into it while our pte |
| 2217 | * here still points into it and can be read by other |
| 2218 | * threads. |
| 2219 | * |
| 2220 | * The critical issue is to order this |
| 2221 | * page_remove_rmap with the ptp_clear_flush above. |
| 2222 | * Those stores are ordered by (if nothing else,) |
| 2223 | * the barrier present in the atomic_add_negative |
| 2224 | * in page_remove_rmap. |
| 2225 | * |
| 2226 | * Then the TLB flush in ptep_clear_flush ensures that |
| 2227 | * no process can access the old page before the |
| 2228 | * decremented mapcount is visible. And the old page |
| 2229 | * cannot be reused until after the decremented |
| 2230 | * mapcount is visible. So transitively, TLBs to |
| 2231 | * old page will be flushed before it can be reused. |
| 2232 | */ |
Kirill A. Shutemov | d281ee6 | 2016-01-15 16:52:16 -0800 | [diff] [blame] | 2233 | page_remove_rmap(old_page, false); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2234 | } |
| 2235 | |
| 2236 | /* Free the old page.. */ |
| 2237 | new_page = old_page; |
| 2238 | page_copied = 1; |
| 2239 | } else { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2240 | mem_cgroup_cancel_charge(new_page, memcg, false); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2241 | } |
| 2242 | |
| 2243 | if (new_page) |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2244 | put_page(new_page); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2245 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2246 | pte_unmap_unlock(fe->pte, fe->ptl); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2247 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
| 2248 | if (old_page) { |
| 2249 | /* |
| 2250 | * Don't let another task, with possibly unlocked vma, |
| 2251 | * keep the mlocked page. |
| 2252 | */ |
| 2253 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { |
| 2254 | lock_page(old_page); /* LRU manipulation */ |
Kirill A. Shutemov | e90309c | 2016-01-15 16:54:33 -0800 | [diff] [blame] | 2255 | if (PageMlocked(old_page)) |
| 2256 | munlock_vma_page(old_page); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2257 | unlock_page(old_page); |
| 2258 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2259 | put_page(old_page); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2260 | } |
| 2261 | return page_copied ? VM_FAULT_WRITE : 0; |
| 2262 | oom_free_new: |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2263 | put_page(new_page); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2264 | oom: |
| 2265 | if (old_page) |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2266 | put_page(old_page); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2267 | return VM_FAULT_OOM; |
| 2268 | } |
| 2269 | |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2270 | /* |
| 2271 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED |
| 2272 | * mapping |
| 2273 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2274 | static int wp_pfn_shared(struct fault_env *fe, pte_t orig_pte) |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2275 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2276 | struct vm_area_struct *vma = fe->vma; |
| 2277 | |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2278 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
| 2279 | struct vm_fault vmf = { |
| 2280 | .page = NULL, |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2281 | .pgoff = linear_page_index(vma, fe->address), |
| 2282 | .virtual_address = |
| 2283 | (void __user *)(fe->address & PAGE_MASK), |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2284 | .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE, |
| 2285 | }; |
| 2286 | int ret; |
| 2287 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2288 | pte_unmap_unlock(fe->pte, fe->ptl); |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2289 | ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); |
| 2290 | if (ret & VM_FAULT_ERROR) |
| 2291 | return ret; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2292 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2293 | &fe->ptl); |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2294 | /* |
| 2295 | * We might have raced with another page fault while we |
| 2296 | * released the pte_offset_map_lock. |
| 2297 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2298 | if (!pte_same(*fe->pte, orig_pte)) { |
| 2299 | pte_unmap_unlock(fe->pte, fe->ptl); |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2300 | return 0; |
| 2301 | } |
| 2302 | } |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2303 | return wp_page_reuse(fe, orig_pte, NULL, 0, 0); |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2304 | } |
| 2305 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2306 | static int wp_page_shared(struct fault_env *fe, pte_t orig_pte, |
| 2307 | struct page *old_page) |
| 2308 | __releases(fe->ptl) |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2309 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2310 | struct vm_area_struct *vma = fe->vma; |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2311 | int page_mkwrite = 0; |
| 2312 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2313 | get_page(old_page); |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2314 | |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2315 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
| 2316 | int tmp; |
| 2317 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2318 | pte_unmap_unlock(fe->pte, fe->ptl); |
| 2319 | tmp = do_page_mkwrite(vma, old_page, fe->address); |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2320 | if (unlikely(!tmp || (tmp & |
| 2321 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2322 | put_page(old_page); |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2323 | return tmp; |
| 2324 | } |
| 2325 | /* |
| 2326 | * Since we dropped the lock we need to revalidate |
| 2327 | * the PTE as someone else may have changed it. If |
| 2328 | * they did, we just return, as we can count on the |
| 2329 | * MMU to tell us if they didn't also make it writable. |
| 2330 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2331 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2332 | &fe->ptl); |
| 2333 | if (!pte_same(*fe->pte, orig_pte)) { |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2334 | unlock_page(old_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2335 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2336 | put_page(old_page); |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2337 | return 0; |
| 2338 | } |
| 2339 | page_mkwrite = 1; |
| 2340 | } |
| 2341 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2342 | return wp_page_reuse(fe, orig_pte, old_page, page_mkwrite, 1); |
Shachar Raindel | 93e478d | 2015-04-14 15:46:35 -0700 | [diff] [blame] | 2343 | } |
| 2344 | |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2345 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | * This routine handles present pages, when users try to write |
| 2347 | * to a shared page. It is done by copying the page to a new address |
| 2348 | * and decrementing the shared-page counter for the old page. |
| 2349 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | * Note that this routine assumes that the protection checks have been |
| 2351 | * done by the caller (the low-level page fault routine in most cases). |
| 2352 | * Thus we can safely just mark it writable once we've done any necessary |
| 2353 | * COW. |
| 2354 | * |
| 2355 | * We also mark the page dirty at this point even though the page will |
| 2356 | * change only once the write actually happens. This avoids a few races, |
| 2357 | * and potentially makes it more efficient. |
| 2358 | * |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2359 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2360 | * but allow concurrent faults), with pte both mapped and locked. |
| 2361 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2362 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2363 | static int do_wp_page(struct fault_env *fe, pte_t orig_pte) |
| 2364 | __releases(fe->ptl) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2366 | struct vm_area_struct *vma = fe->vma; |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2367 | struct page *old_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2368 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2369 | old_page = vm_normal_page(vma, fe->address, orig_pte); |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 2370 | if (!old_page) { |
| 2371 | /* |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 2372 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
| 2373 | * VM_PFNMAP VMA. |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 2374 | * |
| 2375 | * We should not cow pages in a shared writeable mapping. |
Boaz Harrosh | dd90618 | 2015-04-15 16:15:11 -0700 | [diff] [blame] | 2376 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 2377 | */ |
| 2378 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
| 2379 | (VM_WRITE|VM_SHARED)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2380 | return wp_pfn_shared(fe, orig_pte); |
Shachar Raindel | 2f38ab2 | 2015-04-14 15:46:32 -0700 | [diff] [blame] | 2381 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2382 | pte_unmap_unlock(fe->pte, fe->ptl); |
| 2383 | return wp_page_copy(fe, orig_pte, old_page); |
Peter Zijlstra | 251b97f | 2008-07-04 09:59:24 -0700 | [diff] [blame] | 2384 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2385 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2386 | /* |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 2387 | * Take out anonymous pages first, anonymous shared vmas are |
| 2388 | * not dirty accountable. |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2389 | */ |
Hugh Dickins | 9a84089 | 2009-09-21 17:02:01 -0700 | [diff] [blame] | 2390 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
Andrea Arcangeli | 6d0a07e | 2016-05-12 15:42:25 -0700 | [diff] [blame] | 2391 | int total_mapcount; |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 2392 | if (!trylock_page(old_page)) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2393 | get_page(old_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2394 | pte_unmap_unlock(fe->pte, fe->ptl); |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 2395 | lock_page(old_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2396 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, |
| 2397 | fe->address, &fe->ptl); |
| 2398 | if (!pte_same(*fe->pte, orig_pte)) { |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 2399 | unlock_page(old_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2400 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2401 | put_page(old_page); |
Shachar Raindel | 2876680 | 2015-04-14 15:46:29 -0700 | [diff] [blame] | 2402 | return 0; |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 2403 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2404 | put_page(old_page); |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 2405 | } |
Andrea Arcangeli | 6d0a07e | 2016-05-12 15:42:25 -0700 | [diff] [blame] | 2406 | if (reuse_swap_page(old_page, &total_mapcount)) { |
| 2407 | if (total_mapcount == 1) { |
| 2408 | /* |
| 2409 | * The page is all ours. Move it to |
| 2410 | * our anon_vma so the rmap code will |
| 2411 | * not search our parent or siblings. |
| 2412 | * Protected against the rmap code by |
| 2413 | * the page lock. |
| 2414 | */ |
Hugh Dickins | 5a49973 | 2016-07-14 12:07:38 -0700 | [diff] [blame] | 2415 | page_move_anon_rmap(old_page, vma); |
Andrea Arcangeli | 6d0a07e | 2016-05-12 15:42:25 -0700 | [diff] [blame] | 2416 | } |
Michel Lespinasse | b009c02 | 2011-01-13 15:46:07 -0800 | [diff] [blame] | 2417 | unlock_page(old_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2418 | return wp_page_reuse(fe, orig_pte, old_page, 0, 0); |
Michel Lespinasse | b009c02 | 2011-01-13 15:46:07 -0800 | [diff] [blame] | 2419 | } |
Hugh Dickins | ab967d8 | 2009-01-06 14:39:33 -0800 | [diff] [blame] | 2420 | unlock_page(old_page); |
Peter Zijlstra | ee6a645 | 2006-09-25 23:31:00 -0700 | [diff] [blame] | 2421 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 2422 | (VM_WRITE|VM_SHARED))) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2423 | return wp_page_shared(fe, orig_pte, old_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2424 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2425 | |
| 2426 | /* |
| 2427 | * Ok, we need to copy. Oh, well.. |
| 2428 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2429 | get_page(old_page); |
Shachar Raindel | 2876680 | 2015-04-14 15:46:29 -0700 | [diff] [blame] | 2430 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2431 | pte_unmap_unlock(fe->pte, fe->ptl); |
| 2432 | return wp_page_copy(fe, orig_pte, old_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2433 | } |
| 2434 | |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 2435 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2436 | unsigned long start_addr, unsigned long end_addr, |
| 2437 | struct zap_details *details) |
| 2438 | { |
Al Viro | f5cc4ee | 2012-03-05 14:14:20 -0500 | [diff] [blame] | 2439 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | } |
| 2441 | |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 2442 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2443 | struct zap_details *details) |
| 2444 | { |
| 2445 | struct vm_area_struct *vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2446 | pgoff_t vba, vea, zba, zea; |
| 2447 | |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 2448 | vma_interval_tree_foreach(vma, root, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2449 | details->first_index, details->last_index) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | |
| 2451 | vba = vma->vm_pgoff; |
Libin | d6e9321 | 2013-07-03 15:01:26 -0700 | [diff] [blame] | 2452 | vea = vba + vma_pages(vma) - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2453 | zba = details->first_index; |
| 2454 | if (zba < vba) |
| 2455 | zba = vba; |
| 2456 | zea = details->last_index; |
| 2457 | if (zea > vea) |
| 2458 | zea = vea; |
| 2459 | |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 2460 | unmap_mapping_range_vma(vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2461 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
| 2462 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, |
Peter Zijlstra | 97a8941 | 2011-05-24 17:12:04 -0700 | [diff] [blame] | 2463 | details); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2464 | } |
| 2465 | } |
| 2466 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | /** |
Kirill A. Shutemov | 8a5f14a | 2015-02-10 14:09:49 -0800 | [diff] [blame] | 2468 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
| 2469 | * address_space corresponding to the specified page range in the underlying |
| 2470 | * file. |
| 2471 | * |
Martin Waitz | 3d41088 | 2005-06-23 22:05:21 -0700 | [diff] [blame] | 2472 | * @mapping: the address space containing mmaps to be unmapped. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2473 | * @holebegin: byte in first page to unmap, relative to the start of |
| 2474 | * the underlying file. This will be rounded down to a PAGE_SIZE |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 2475 | * boundary. Note that this is different from truncate_pagecache(), which |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2476 | * must keep the partial page. In contrast, we must get rid of |
| 2477 | * partial pages. |
| 2478 | * @holelen: size of prospective hole in bytes. This will be rounded |
| 2479 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the |
| 2480 | * end of the file. |
| 2481 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; |
| 2482 | * but 0 when invalidating pagecache, don't throw away private data. |
| 2483 | */ |
| 2484 | void unmap_mapping_range(struct address_space *mapping, |
| 2485 | loff_t const holebegin, loff_t const holelen, int even_cows) |
| 2486 | { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 2487 | struct zap_details details = { }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2488 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
| 2489 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2490 | |
| 2491 | /* Check for overflow. */ |
| 2492 | if (sizeof(holelen) > sizeof(hlen)) { |
| 2493 | long long holeend = |
| 2494 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2495 | if (holeend & ~(long long)ULONG_MAX) |
| 2496 | hlen = ULONG_MAX - hba + 1; |
| 2497 | } |
| 2498 | |
| 2499 | details.check_mapping = even_cows? NULL: mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2500 | details.first_index = hba; |
| 2501 | details.last_index = hba + hlen - 1; |
| 2502 | if (details.last_index < details.first_index) |
| 2503 | details.last_index = ULONG_MAX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2504 | |
Kirill A. Shutemov | 46c043e | 2015-09-08 14:59:42 -0700 | [diff] [blame] | 2505 | i_mmap_lock_write(mapping); |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 2506 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2507 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
Kirill A. Shutemov | 46c043e | 2015-09-08 14:59:42 -0700 | [diff] [blame] | 2508 | i_mmap_unlock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2509 | } |
| 2510 | EXPORT_SYMBOL(unmap_mapping_range); |
| 2511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2512 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2513 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2514 | * but allow concurrent faults), and pte mapped but not yet locked. |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 2515 | * We return with pte unmapped and unlocked. |
| 2516 | * |
| 2517 | * We return with the mmap_sem locked or unlocked in the same cases |
| 2518 | * as does filemap_fault(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2519 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2520 | int do_swap_page(struct fault_env *fe, pte_t orig_pte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2521 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2522 | struct vm_area_struct *vma = fe->vma; |
Hugh Dickins | 56f3180 | 2013-02-22 16:36:10 -0800 | [diff] [blame] | 2523 | struct page *page, *swapcache; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2524 | struct mem_cgroup *memcg; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2525 | swp_entry_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2526 | pte_t pte; |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 2527 | int locked; |
Rik van Riel | ad8c2ee | 2010-08-09 17:19:48 -0700 | [diff] [blame] | 2528 | int exclusive = 0; |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2529 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2530 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2531 | if (!pte_unmap_same(vma->vm_mm, fe->pmd, fe->pte, orig_pte)) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2532 | goto out; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2533 | |
| 2534 | entry = pte_to_swp_entry(orig_pte); |
Andi Kleen | d1737fd | 2009-09-16 11:50:06 +0200 | [diff] [blame] | 2535 | if (unlikely(non_swap_entry(entry))) { |
| 2536 | if (is_migration_entry(entry)) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2537 | migration_entry_wait(vma->vm_mm, fe->pmd, fe->address); |
Andi Kleen | d1737fd | 2009-09-16 11:50:06 +0200 | [diff] [blame] | 2538 | } else if (is_hwpoison_entry(entry)) { |
| 2539 | ret = VM_FAULT_HWPOISON; |
| 2540 | } else { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2541 | print_bad_pte(vma, fe->address, orig_pte, NULL); |
Hugh Dickins | d99be1a | 2009-12-14 17:59:04 -0800 | [diff] [blame] | 2542 | ret = VM_FAULT_SIGBUS; |
Andi Kleen | d1737fd | 2009-09-16 11:50:06 +0200 | [diff] [blame] | 2543 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 2544 | goto out; |
| 2545 | } |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2546 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2547 | page = lookup_swap_cache(entry); |
| 2548 | if (!page) { |
Hugh Dickins | 02098fe | 2008-02-04 22:28:42 -0800 | [diff] [blame] | 2549 | page = swapin_readahead(entry, |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2550 | GFP_HIGHUSER_MOVABLE, vma, fe->address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2551 | if (!page) { |
| 2552 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2553 | * Back out if somebody else faulted in this pte |
| 2554 | * while we released the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2555 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2556 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, |
| 2557 | fe->address, &fe->ptl); |
| 2558 | if (likely(pte_same(*fe->pte, orig_pte))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | ret = VM_FAULT_OOM; |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 2560 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2561 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2562 | } |
| 2563 | |
| 2564 | /* Had to read the page from swap area: Major fault */ |
| 2565 | ret = VM_FAULT_MAJOR; |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2566 | count_vm_event(PGMAJFAULT); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2567 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
Andi Kleen | d1737fd | 2009-09-16 11:50:06 +0200 | [diff] [blame] | 2568 | } else if (PageHWPoison(page)) { |
Wu Fengguang | 71f7252 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 2569 | /* |
| 2570 | * hwpoisoned dirty swapcache pages are kept for killing |
| 2571 | * owner processes (which may be unknown at hwpoison time) |
| 2572 | */ |
Andi Kleen | d1737fd | 2009-09-16 11:50:06 +0200 | [diff] [blame] | 2573 | ret = VM_FAULT_HWPOISON; |
| 2574 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Hugh Dickins | 56f3180 | 2013-02-22 16:36:10 -0800 | [diff] [blame] | 2575 | swapcache = page; |
Andi Kleen | 4779cb3 | 2009-10-14 01:51:41 +0200 | [diff] [blame] | 2576 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2577 | } |
| 2578 | |
Hugh Dickins | 56f3180 | 2013-02-22 16:36:10 -0800 | [diff] [blame] | 2579 | swapcache = page; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2580 | locked = lock_page_or_retry(page, vma->vm_mm, fe->flags); |
Rik van Riel | e709ffd | 2012-05-29 15:06:18 -0700 | [diff] [blame] | 2581 | |
Balbir Singh | 20a1022 | 2007-11-14 17:00:33 -0800 | [diff] [blame] | 2582 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 2583 | if (!locked) { |
| 2584 | ret |= VM_FAULT_RETRY; |
| 2585 | goto out_release; |
| 2586 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2587 | |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2588 | /* |
Hugh Dickins | 31c4a3d | 2010-09-19 19:40:22 -0700 | [diff] [blame] | 2589 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
| 2590 | * release the swapcache from under us. The page pin, and pte_same |
| 2591 | * test below, are not enough to exclude that. Even if it is still |
| 2592 | * swapcache, we need to check that the page's swap has not changed. |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2593 | */ |
Hugh Dickins | 31c4a3d | 2010-09-19 19:40:22 -0700 | [diff] [blame] | 2594 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2595 | goto out_page; |
| 2596 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2597 | page = ksm_might_need_to_copy(page, vma, fe->address); |
Hugh Dickins | cbf86cf | 2013-02-22 16:35:08 -0800 | [diff] [blame] | 2598 | if (unlikely(!page)) { |
| 2599 | ret = VM_FAULT_OOM; |
| 2600 | page = swapcache; |
Hugh Dickins | cbf86cf | 2013-02-22 16:35:08 -0800 | [diff] [blame] | 2601 | goto out_page; |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 2602 | } |
| 2603 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2604 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
| 2605 | &memcg, false)) { |
KAMEZAWA Hiroyuki | 073e587 | 2008-10-18 20:28:08 -0700 | [diff] [blame] | 2606 | ret = VM_FAULT_OOM; |
Johannes Weiner | bc43f75 | 2009-04-30 15:08:08 -0700 | [diff] [blame] | 2607 | goto out_page; |
KAMEZAWA Hiroyuki | 073e587 | 2008-10-18 20:28:08 -0700 | [diff] [blame] | 2608 | } |
| 2609 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2610 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2611 | * Back out if somebody else already faulted in this pte. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2612 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2613 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2614 | &fe->ptl); |
| 2615 | if (unlikely(!pte_same(*fe->pte, orig_pte))) |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2616 | goto out_nomap; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2617 | |
| 2618 | if (unlikely(!PageUptodate(page))) { |
| 2619 | ret = VM_FAULT_SIGBUS; |
| 2620 | goto out_nomap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2621 | } |
| 2622 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2623 | /* |
| 2624 | * The page isn't present yet, go ahead with the fault. |
| 2625 | * |
| 2626 | * Be careful about the sequence of operations here. |
| 2627 | * To get its accounting right, reuse_swap_page() must be called |
| 2628 | * while the page is counted on swap but not yet in mapcount i.e. |
| 2629 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() |
| 2630 | * must be called after the swap_free(), or it will never succeed. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2631 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2632 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2633 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
| 2634 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2635 | pte = mk_pte(page, vma->vm_page_prot); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2636 | if ((fe->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2637 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2638 | fe->flags &= ~FAULT_FLAG_WRITE; |
Andrea Arcangeli | 9a5b489 | 2010-08-09 17:19:49 -0700 | [diff] [blame] | 2639 | ret |= VM_FAULT_WRITE; |
Kirill A. Shutemov | d281ee6 | 2016-01-15 16:52:16 -0800 | [diff] [blame] | 2640 | exclusive = RMAP_EXCLUSIVE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2641 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2642 | flush_icache_page(vma, page); |
Cyrill Gorcunov | 179ef71 | 2013-08-13 16:00:49 -0700 | [diff] [blame] | 2643 | if (pte_swp_soft_dirty(orig_pte)) |
| 2644 | pte = pte_mksoft_dirty(pte); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2645 | set_pte_at(vma->vm_mm, fe->address, fe->pte, pte); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2646 | if (page == swapcache) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2647 | do_page_add_anon_rmap(page, vma, fe->address, exclusive); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2648 | mem_cgroup_commit_charge(page, memcg, true, false); |
Minchan Kim | 1a8018f | 2016-08-02 14:02:25 -0700 | [diff] [blame] | 2649 | activate_page(page); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2650 | } else { /* ksm created a completely new copy */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2651 | page_add_new_anon_rmap(page, vma, fe->address, false); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2652 | mem_cgroup_commit_charge(page, memcg, false, false); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2653 | lru_cache_add_active_or_unevictable(page, vma); |
| 2654 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2655 | |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2656 | swap_free(entry); |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 2657 | if (mem_cgroup_swap_full(page) || |
| 2658 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 2659 | try_to_free_swap(page); |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2660 | unlock_page(page); |
Hugh Dickins | 56f3180 | 2013-02-22 16:36:10 -0800 | [diff] [blame] | 2661 | if (page != swapcache) { |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2662 | /* |
| 2663 | * Hold the lock to avoid the swap entry to be reused |
| 2664 | * until we take the PT lock for the pte_same() check |
| 2665 | * (to avoid false positives from pte_same). For |
| 2666 | * further safety release the lock after the swap_free |
| 2667 | * so that the swap count won't change under a |
| 2668 | * parallel locked swapcache. |
| 2669 | */ |
| 2670 | unlock_page(swapcache); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2671 | put_page(swapcache); |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2672 | } |
Hugh Dickins | c475a8a | 2005-06-21 17:15:12 -0700 | [diff] [blame] | 2673 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2674 | if (fe->flags & FAULT_FLAG_WRITE) { |
| 2675 | ret |= do_wp_page(fe, pte); |
Hugh Dickins | 61469f1 | 2008-03-04 14:29:04 -0800 | [diff] [blame] | 2676 | if (ret & VM_FAULT_ERROR) |
| 2677 | ret &= VM_FAULT_ERROR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2678 | goto out; |
| 2679 | } |
| 2680 | |
| 2681 | /* No need to invalidate - it was non-present before */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2682 | update_mmu_cache(vma, fe->address, fe->pte); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2683 | unlock: |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2684 | pte_unmap_unlock(fe->pte, fe->ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2685 | out: |
| 2686 | return ret; |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2687 | out_nomap: |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2688 | mem_cgroup_cancel_charge(page, memcg, false); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2689 | pte_unmap_unlock(fe->pte, fe->ptl); |
Johannes Weiner | bc43f75 | 2009-04-30 15:08:08 -0700 | [diff] [blame] | 2690 | out_page: |
Kirill Korotaev | b810748 | 2005-05-16 21:53:50 -0700 | [diff] [blame] | 2691 | unlock_page(page); |
Andi Kleen | 4779cb3 | 2009-10-14 01:51:41 +0200 | [diff] [blame] | 2692 | out_release: |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2693 | put_page(page); |
Hugh Dickins | 56f3180 | 2013-02-22 16:36:10 -0800 | [diff] [blame] | 2694 | if (page != swapcache) { |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2695 | unlock_page(swapcache); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2696 | put_page(swapcache); |
Andrea Arcangeli | 4969c11 | 2010-09-09 16:37:52 -0700 | [diff] [blame] | 2697 | } |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2698 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2699 | } |
| 2700 | |
| 2701 | /* |
Luck, Tony | 8ca3eb0 | 2010-08-24 11:44:18 -0700 | [diff] [blame] | 2702 | * This is like a special single-page "expand_{down|up}wards()", |
| 2703 | * except we must first make sure that 'address{-|+}PAGE_SIZE' |
Linus Torvalds | 320b2b8 | 2010-08-12 17:54:33 -0700 | [diff] [blame] | 2704 | * doesn't hit another vma. |
Linus Torvalds | 320b2b8 | 2010-08-12 17:54:33 -0700 | [diff] [blame] | 2705 | */ |
| 2706 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) |
| 2707 | { |
| 2708 | address &= PAGE_MASK; |
| 2709 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { |
Linus Torvalds | 0e8e50e | 2010-08-20 16:49:40 -0700 | [diff] [blame] | 2710 | struct vm_area_struct *prev = vma->vm_prev; |
Linus Torvalds | 320b2b8 | 2010-08-12 17:54:33 -0700 | [diff] [blame] | 2711 | |
Linus Torvalds | 0e8e50e | 2010-08-20 16:49:40 -0700 | [diff] [blame] | 2712 | /* |
| 2713 | * Is there a mapping abutting this one below? |
| 2714 | * |
| 2715 | * That's only ok if it's the same stack mapping |
| 2716 | * that has gotten split.. |
| 2717 | */ |
| 2718 | if (prev && prev->vm_end == address) |
| 2719 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; |
| 2720 | |
Linus Torvalds | fee7e49 | 2015-01-06 13:00:05 -0800 | [diff] [blame] | 2721 | return expand_downwards(vma, address - PAGE_SIZE); |
Linus Torvalds | 320b2b8 | 2010-08-12 17:54:33 -0700 | [diff] [blame] | 2722 | } |
Luck, Tony | 8ca3eb0 | 2010-08-24 11:44:18 -0700 | [diff] [blame] | 2723 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
| 2724 | struct vm_area_struct *next = vma->vm_next; |
| 2725 | |
| 2726 | /* As VM_GROWSDOWN but s/below/above/ */ |
| 2727 | if (next && next->vm_start == address + PAGE_SIZE) |
| 2728 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; |
| 2729 | |
Linus Torvalds | fee7e49 | 2015-01-06 13:00:05 -0800 | [diff] [blame] | 2730 | return expand_upwards(vma, address + PAGE_SIZE); |
Luck, Tony | 8ca3eb0 | 2010-08-24 11:44:18 -0700 | [diff] [blame] | 2731 | } |
Linus Torvalds | 320b2b8 | 2010-08-12 17:54:33 -0700 | [diff] [blame] | 2732 | return 0; |
| 2733 | } |
| 2734 | |
| 2735 | /* |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2736 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 2737 | * but allow concurrent faults), and pte mapped but not yet locked. |
| 2738 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2739 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2740 | static int do_anonymous_page(struct fault_env *fe) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2741 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2742 | struct vm_area_struct *vma = fe->vma; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2743 | struct mem_cgroup *memcg; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2744 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2745 | pte_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2746 | |
Kirill A. Shutemov | 6b7339f | 2015-07-06 23:18:37 +0300 | [diff] [blame] | 2747 | /* File mapping without ->vm_ops ? */ |
| 2748 | if (vma->vm_flags & VM_SHARED) |
| 2749 | return VM_FAULT_SIGBUS; |
| 2750 | |
Linus Torvalds | 11ac552 | 2010-08-14 11:44:56 -0700 | [diff] [blame] | 2751 | /* Check if we need to add a guard page to the stack */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2752 | if (check_stack_guard_page(vma, fe->address) < 0) |
Linus Torvalds | 9c145c56 | 2015-01-29 11:15:17 -0800 | [diff] [blame] | 2753 | return VM_FAULT_SIGSEGV; |
Linus Torvalds | 11ac552 | 2010-08-14 11:44:56 -0700 | [diff] [blame] | 2754 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 2755 | /* |
| 2756 | * Use pte_alloc() instead of pte_alloc_map(). We can't run |
| 2757 | * pte_offset_map() on pmds where a huge pmd might be created |
| 2758 | * from a different thread. |
| 2759 | * |
| 2760 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when |
| 2761 | * parallel threads are excluded by other means. |
| 2762 | * |
| 2763 | * Here we only have down_read(mmap_sem). |
| 2764 | */ |
| 2765 | if (pte_alloc(vma->vm_mm, fe->pmd, fe->address)) |
| 2766 | return VM_FAULT_OOM; |
| 2767 | |
| 2768 | /* See the comment in pte_alloc_one_map() */ |
| 2769 | if (unlikely(pmd_trans_unstable(fe->pmd))) |
| 2770 | return 0; |
| 2771 | |
Linus Torvalds | 11ac552 | 2010-08-14 11:44:56 -0700 | [diff] [blame] | 2772 | /* Use the zero-page for reads */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2773 | if (!(fe->flags & FAULT_FLAG_WRITE) && |
| 2774 | !mm_forbids_zeropage(vma->vm_mm)) { |
| 2775 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(fe->address), |
Hugh Dickins | 62eede6 | 2009-09-21 17:03:34 -0700 | [diff] [blame] | 2776 | vma->vm_page_prot)); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2777 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2778 | &fe->ptl); |
| 2779 | if (!pte_none(*fe->pte)) |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 2780 | goto unlock; |
Andrea Arcangeli | 6b251fc | 2015-09-04 15:46:20 -0700 | [diff] [blame] | 2781 | /* Deliver the page fault to userland, check inside PT lock */ |
| 2782 | if (userfaultfd_missing(vma)) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2783 | pte_unmap_unlock(fe->pte, fe->ptl); |
| 2784 | return handle_userfault(fe, VM_UFFD_MISSING); |
Andrea Arcangeli | 6b251fc | 2015-09-04 15:46:20 -0700 | [diff] [blame] | 2785 | } |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 2786 | goto setpte; |
| 2787 | } |
| 2788 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2789 | /* Allocate our own private page. */ |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2790 | if (unlikely(anon_vma_prepare(vma))) |
| 2791 | goto oom; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2792 | page = alloc_zeroed_user_highpage_movable(vma, fe->address); |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2793 | if (!page) |
| 2794 | goto oom; |
Mel Gorman | eb3c24f | 2015-06-24 16:57:27 -0700 | [diff] [blame] | 2795 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2796 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
Mel Gorman | eb3c24f | 2015-06-24 16:57:27 -0700 | [diff] [blame] | 2797 | goto oom_free_page; |
| 2798 | |
Minchan Kim | 52f3762 | 2013-04-29 15:08:15 -0700 | [diff] [blame] | 2799 | /* |
| 2800 | * The memory barrier inside __SetPageUptodate makes sure that |
| 2801 | * preceeding stores to the page contents become visible before |
| 2802 | * the set_pte_at() write. |
| 2803 | */ |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2804 | __SetPageUptodate(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2805 | |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2806 | entry = mk_pte(page, vma->vm_page_prot); |
Hugh Dickins | 1ac0cb5 | 2009-09-21 17:03:29 -0700 | [diff] [blame] | 2807 | if (vma->vm_flags & VM_WRITE) |
| 2808 | entry = pte_mkwrite(pte_mkdirty(entry)); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2809 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2810 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2811 | &fe->ptl); |
| 2812 | if (!pte_none(*fe->pte)) |
Nick Piggin | 557ed1f | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 2813 | goto release; |
Hugh Dickins | 9ba6929 | 2009-09-21 17:02:20 -0700 | [diff] [blame] | 2814 | |
Andrea Arcangeli | 6b251fc | 2015-09-04 15:46:20 -0700 | [diff] [blame] | 2815 | /* Deliver the page fault to userland, check inside PT lock */ |
| 2816 | if (userfaultfd_missing(vma)) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2817 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2818 | mem_cgroup_cancel_charge(page, memcg, false); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2819 | put_page(page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2820 | return handle_userfault(fe, VM_UFFD_MISSING); |
Andrea Arcangeli | 6b251fc | 2015-09-04 15:46:20 -0700 | [diff] [blame] | 2821 | } |
| 2822 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2823 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
| 2824 | page_add_new_anon_rmap(page, vma, fe->address, false); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2825 | mem_cgroup_commit_charge(page, memcg, false, false); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2826 | lru_cache_add_active_or_unevictable(page, vma); |
Hugh Dickins | a13ea5b | 2009-09-21 17:03:30 -0700 | [diff] [blame] | 2827 | setpte: |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2828 | set_pte_at(vma->vm_mm, fe->address, fe->pte, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2829 | |
| 2830 | /* No need to invalidate - it was non-present before */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2831 | update_mmu_cache(vma, fe->address, fe->pte); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2832 | unlock: |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2833 | pte_unmap_unlock(fe->pte, fe->ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2834 | return 0; |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2835 | release: |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 2836 | mem_cgroup_cancel_charge(page, memcg, false); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2837 | put_page(page); |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 2838 | goto unlock; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2839 | oom_free_page: |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2840 | put_page(page); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 2841 | oom: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2842 | return VM_FAULT_OOM; |
| 2843 | } |
| 2844 | |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 2845 | /* |
| 2846 | * The mmap_sem must have been held on entry, and may have been |
| 2847 | * released depending on flags and vma->vm_ops->fault() return value. |
| 2848 | * See filemap_fault() and __lock_page_retry(). |
| 2849 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2850 | static int __do_fault(struct fault_env *fe, pgoff_t pgoff, |
| 2851 | struct page *cow_page, struct page **page, void **entry) |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2852 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2853 | struct vm_area_struct *vma = fe->vma; |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2854 | struct vm_fault vmf; |
| 2855 | int ret; |
| 2856 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2857 | vmf.virtual_address = (void __user *)(fe->address & PAGE_MASK); |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2858 | vmf.pgoff = pgoff; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2859 | vmf.flags = fe->flags; |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2860 | vmf.page = NULL; |
Michal Hocko | c20cd45 | 2016-01-14 15:20:12 -0800 | [diff] [blame] | 2861 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
Matthew Wilcox | 2e4cdab | 2015-02-16 15:58:50 -0800 | [diff] [blame] | 2862 | vmf.cow_page = cow_page; |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2863 | |
| 2864 | ret = vma->vm_ops->fault(vma, &vmf); |
| 2865 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
| 2866 | return ret; |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 2867 | if (ret & VM_FAULT_DAX_LOCKED) { |
| 2868 | *entry = vmf.entry; |
| 2869 | return ret; |
| 2870 | } |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2871 | |
| 2872 | if (unlikely(PageHWPoison(vmf.page))) { |
| 2873 | if (ret & VM_FAULT_LOCKED) |
| 2874 | unlock_page(vmf.page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2875 | put_page(vmf.page); |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 2876 | return VM_FAULT_HWPOISON; |
| 2877 | } |
| 2878 | |
| 2879 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
| 2880 | lock_page(vmf.page); |
| 2881 | else |
| 2882 | VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); |
| 2883 | |
| 2884 | *page = vmf.page; |
| 2885 | return ret; |
| 2886 | } |
| 2887 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 2888 | static int pte_alloc_one_map(struct fault_env *fe) |
| 2889 | { |
| 2890 | struct vm_area_struct *vma = fe->vma; |
| 2891 | |
| 2892 | if (!pmd_none(*fe->pmd)) |
| 2893 | goto map_pte; |
| 2894 | if (fe->prealloc_pte) { |
| 2895 | fe->ptl = pmd_lock(vma->vm_mm, fe->pmd); |
| 2896 | if (unlikely(!pmd_none(*fe->pmd))) { |
| 2897 | spin_unlock(fe->ptl); |
| 2898 | goto map_pte; |
| 2899 | } |
| 2900 | |
| 2901 | atomic_long_inc(&vma->vm_mm->nr_ptes); |
| 2902 | pmd_populate(vma->vm_mm, fe->pmd, fe->prealloc_pte); |
| 2903 | spin_unlock(fe->ptl); |
| 2904 | fe->prealloc_pte = 0; |
| 2905 | } else if (unlikely(pte_alloc(vma->vm_mm, fe->pmd, fe->address))) { |
| 2906 | return VM_FAULT_OOM; |
| 2907 | } |
| 2908 | map_pte: |
| 2909 | /* |
| 2910 | * If a huge pmd materialized under us just retry later. Use |
| 2911 | * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd |
| 2912 | * didn't become pmd_trans_huge under us and then back to pmd_none, as |
| 2913 | * a result of MADV_DONTNEED running immediately after a huge pmd fault |
| 2914 | * in a different thread of this mm, in turn leading to a misleading |
| 2915 | * pmd_trans_huge() retval. All we have to ensure is that it is a |
| 2916 | * regular pmd that we can walk with pte_offset_map() and we can do that |
| 2917 | * through an atomic read in C, which is what pmd_trans_unstable() |
| 2918 | * provides. |
| 2919 | */ |
| 2920 | if (pmd_trans_unstable(fe->pmd) || pmd_devmap(*fe->pmd)) |
| 2921 | return VM_FAULT_NOPAGE; |
| 2922 | |
| 2923 | fe->pte = pte_offset_map_lock(vma->vm_mm, fe->pmd, fe->address, |
| 2924 | &fe->ptl); |
| 2925 | return 0; |
| 2926 | } |
| 2927 | |
Kirill A. Shutemov | e496cf3 | 2016-07-26 15:26:35 -0700 | [diff] [blame] | 2928 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
Kirill A. Shutemov | 1010245 | 2016-07-26 15:25:29 -0700 | [diff] [blame] | 2929 | |
| 2930 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) |
| 2931 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, |
| 2932 | unsigned long haddr) |
| 2933 | { |
| 2934 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != |
| 2935 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) |
| 2936 | return false; |
| 2937 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) |
| 2938 | return false; |
| 2939 | return true; |
| 2940 | } |
| 2941 | |
| 2942 | static int do_set_pmd(struct fault_env *fe, struct page *page) |
| 2943 | { |
| 2944 | struct vm_area_struct *vma = fe->vma; |
| 2945 | bool write = fe->flags & FAULT_FLAG_WRITE; |
| 2946 | unsigned long haddr = fe->address & HPAGE_PMD_MASK; |
| 2947 | pmd_t entry; |
| 2948 | int i, ret; |
| 2949 | |
| 2950 | if (!transhuge_vma_suitable(vma, haddr)) |
| 2951 | return VM_FAULT_FALLBACK; |
| 2952 | |
| 2953 | ret = VM_FAULT_FALLBACK; |
| 2954 | page = compound_head(page); |
| 2955 | |
| 2956 | fe->ptl = pmd_lock(vma->vm_mm, fe->pmd); |
| 2957 | if (unlikely(!pmd_none(*fe->pmd))) |
| 2958 | goto out; |
| 2959 | |
| 2960 | for (i = 0; i < HPAGE_PMD_NR; i++) |
| 2961 | flush_icache_page(vma, page + i); |
| 2962 | |
| 2963 | entry = mk_huge_pmd(page, vma->vm_page_prot); |
| 2964 | if (write) |
| 2965 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
| 2966 | |
| 2967 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); |
| 2968 | page_add_file_rmap(page, true); |
| 2969 | |
| 2970 | set_pmd_at(vma->vm_mm, haddr, fe->pmd, entry); |
| 2971 | |
| 2972 | update_mmu_cache_pmd(vma, haddr, fe->pmd); |
| 2973 | |
| 2974 | /* fault is handled */ |
| 2975 | ret = 0; |
Kirill A. Shutemov | 95ecedc | 2016-07-26 15:25:31 -0700 | [diff] [blame] | 2976 | count_vm_event(THP_FILE_MAPPED); |
Kirill A. Shutemov | 1010245 | 2016-07-26 15:25:29 -0700 | [diff] [blame] | 2977 | out: |
| 2978 | spin_unlock(fe->ptl); |
| 2979 | return ret; |
| 2980 | } |
| 2981 | #else |
| 2982 | static int do_set_pmd(struct fault_env *fe, struct page *page) |
| 2983 | { |
| 2984 | BUILD_BUG(); |
| 2985 | return 0; |
| 2986 | } |
| 2987 | #endif |
| 2988 | |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 2989 | /** |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 2990 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
| 2991 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 2992 | * |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 2993 | * @fe: fault environment |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 2994 | * @memcg: memcg to charge page (only for private mappings) |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 2995 | * @page: page to map |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 2996 | * |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 2997 | * Caller must take care of unlocking fe->ptl, if fe->pte is non-NULL on return. |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 2998 | * |
| 2999 | * Target users are page handler itself and implementations of |
| 3000 | * vm_ops->map_pages. |
| 3001 | */ |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3002 | int alloc_set_pte(struct fault_env *fe, struct mem_cgroup *memcg, |
| 3003 | struct page *page) |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3004 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3005 | struct vm_area_struct *vma = fe->vma; |
| 3006 | bool write = fe->flags & FAULT_FLAG_WRITE; |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3007 | pte_t entry; |
Kirill A. Shutemov | 1010245 | 2016-07-26 15:25:29 -0700 | [diff] [blame] | 3008 | int ret; |
| 3009 | |
Kirill A. Shutemov | e496cf3 | 2016-07-26 15:26:35 -0700 | [diff] [blame] | 3010 | if (pmd_none(*fe->pmd) && PageTransCompound(page) && |
| 3011 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
Kirill A. Shutemov | 1010245 | 2016-07-26 15:25:29 -0700 | [diff] [blame] | 3012 | /* THP on COW? */ |
| 3013 | VM_BUG_ON_PAGE(memcg, page); |
| 3014 | |
| 3015 | ret = do_set_pmd(fe, page); |
| 3016 | if (ret != VM_FAULT_FALLBACK) |
| 3017 | return ret; |
| 3018 | } |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3019 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3020 | if (!fe->pte) { |
Kirill A. Shutemov | 1010245 | 2016-07-26 15:25:29 -0700 | [diff] [blame] | 3021 | ret = pte_alloc_one_map(fe); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3022 | if (ret) |
| 3023 | return ret; |
| 3024 | } |
| 3025 | |
| 3026 | /* Re-check under ptl */ |
| 3027 | if (unlikely(!pte_none(*fe->pte))) |
| 3028 | return VM_FAULT_NOPAGE; |
| 3029 | |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3030 | flush_icache_page(vma, page); |
| 3031 | entry = mk_pte(page, vma->vm_page_prot); |
| 3032 | if (write) |
| 3033 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3034 | /* copy-on-write page */ |
| 3035 | if (write && !(vma->vm_flags & VM_SHARED)) { |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3036 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3037 | page_add_new_anon_rmap(page, vma, fe->address, false); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3038 | mem_cgroup_commit_charge(page, memcg, false, false); |
| 3039 | lru_cache_add_active_or_unevictable(page, vma); |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3040 | } else { |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 3041 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
Kirill A. Shutemov | dd78fed | 2016-07-26 15:25:26 -0700 | [diff] [blame] | 3042 | page_add_file_rmap(page, false); |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3043 | } |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3044 | set_pte_at(vma->vm_mm, fe->address, fe->pte, entry); |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3045 | |
| 3046 | /* no need to invalidate: a not-present page won't be cached */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3047 | update_mmu_cache(vma, fe->address, fe->pte); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3048 | |
| 3049 | return 0; |
Kirill A. Shutemov | 3bb9779 | 2014-04-03 14:48:16 -0700 | [diff] [blame] | 3050 | } |
| 3051 | |
Kirill A. Shutemov | 3a91053 | 2014-08-06 16:08:07 -0700 | [diff] [blame] | 3052 | static unsigned long fault_around_bytes __read_mostly = |
| 3053 | rounddown_pow_of_two(65536); |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3054 | |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3055 | #ifdef CONFIG_DEBUG_FS |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3056 | static int fault_around_bytes_get(void *data, u64 *val) |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3057 | { |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3058 | *val = fault_around_bytes; |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3059 | return 0; |
| 3060 | } |
| 3061 | |
Andrey Ryabinin | b4903d6 | 2014-07-30 16:08:35 -0700 | [diff] [blame] | 3062 | /* |
| 3063 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes |
| 3064 | * rounded down to nearest page order. It's what do_fault_around() expects to |
| 3065 | * see. |
| 3066 | */ |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3067 | static int fault_around_bytes_set(void *data, u64 val) |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3068 | { |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3069 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3070 | return -EINVAL; |
Andrey Ryabinin | b4903d6 | 2014-07-30 16:08:35 -0700 | [diff] [blame] | 3071 | if (val > PAGE_SIZE) |
| 3072 | fault_around_bytes = rounddown_pow_of_two(val); |
| 3073 | else |
| 3074 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3075 | return 0; |
| 3076 | } |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3077 | DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, |
| 3078 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3079 | |
| 3080 | static int __init fault_around_debugfs(void) |
| 3081 | { |
| 3082 | void *ret; |
| 3083 | |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3084 | ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, |
| 3085 | &fault_around_bytes_fops); |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3086 | if (!ret) |
Kirill A. Shutemov | a9b0f86 | 2014-06-04 16:10:54 -0700 | [diff] [blame] | 3087 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3088 | return 0; |
| 3089 | } |
| 3090 | late_initcall(fault_around_debugfs); |
Kirill A. Shutemov | 1592eef | 2014-04-07 15:37:22 -0700 | [diff] [blame] | 3091 | #endif |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3092 | |
Kirill A. Shutemov | 1fdb412 | 2014-06-04 16:10:55 -0700 | [diff] [blame] | 3093 | /* |
| 3094 | * do_fault_around() tries to map few pages around the fault address. The hope |
| 3095 | * is that the pages will be needed soon and this will lower the number of |
| 3096 | * faults to handle. |
| 3097 | * |
| 3098 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's |
| 3099 | * not ready to be mapped: not up-to-date, locked, etc. |
| 3100 | * |
| 3101 | * This function is called with the page table lock taken. In the split ptlock |
| 3102 | * case the page table lock only protects only those entries which belong to |
| 3103 | * the page table corresponding to the fault address. |
| 3104 | * |
| 3105 | * This function doesn't cross the VMA boundaries, in order to call map_pages() |
| 3106 | * only once. |
| 3107 | * |
| 3108 | * fault_around_pages() defines how many pages we'll try to map. |
| 3109 | * do_fault_around() expects it to return a power of two less than or equal to |
| 3110 | * PTRS_PER_PTE. |
| 3111 | * |
| 3112 | * The virtual address of the area that we map is naturally aligned to the |
| 3113 | * fault_around_pages() value (and therefore to page order). This way it's |
| 3114 | * easier to guarantee that we don't cross page table boundaries. |
| 3115 | */ |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3116 | static int do_fault_around(struct fault_env *fe, pgoff_t start_pgoff) |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3117 | { |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3118 | unsigned long address = fe->address, nr_pages, mask; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3119 | pgoff_t end_pgoff; |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3120 | int off, ret = 0; |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3121 | |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 3122 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
Kirill A. Shutemov | aecd6f4 | 2014-08-06 16:08:05 -0700 | [diff] [blame] | 3123 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
| 3124 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3125 | fe->address = max(address & mask, fe->vma->vm_start); |
| 3126 | off = ((address - fe->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3127 | start_pgoff -= off; |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3128 | |
| 3129 | /* |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3130 | * end_pgoff is either end of page table or end of vma |
| 3131 | * or fault_around_pages() from start_pgoff, depending what is nearest. |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3132 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3133 | end_pgoff = start_pgoff - |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3134 | ((fe->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3135 | PTRS_PER_PTE - 1; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3136 | end_pgoff = min3(end_pgoff, vma_pages(fe->vma) + fe->vma->vm_pgoff - 1, |
| 3137 | start_pgoff + nr_pages - 1); |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3138 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3139 | if (pmd_none(*fe->pmd)) { |
| 3140 | fe->prealloc_pte = pte_alloc_one(fe->vma->vm_mm, fe->address); |
Vegard Nossum | c5f88bd | 2016-08-02 14:02:22 -0700 | [diff] [blame] | 3141 | if (!fe->prealloc_pte) |
| 3142 | goto out; |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3143 | smp_wmb(); /* See comment in __pte_alloc() */ |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3144 | } |
| 3145 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3146 | fe->vma->vm_ops->map_pages(fe, start_pgoff, end_pgoff); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3147 | |
| 3148 | /* preallocated pagetable is unused: free it */ |
| 3149 | if (fe->prealloc_pte) { |
| 3150 | pte_free(fe->vma->vm_mm, fe->prealloc_pte); |
| 3151 | fe->prealloc_pte = 0; |
| 3152 | } |
| 3153 | /* Huge page is mapped? Page fault is solved */ |
| 3154 | if (pmd_trans_huge(*fe->pmd)) { |
| 3155 | ret = VM_FAULT_NOPAGE; |
| 3156 | goto out; |
| 3157 | } |
| 3158 | |
| 3159 | /* ->map_pages() haven't done anything useful. Cold page cache? */ |
| 3160 | if (!fe->pte) |
| 3161 | goto out; |
| 3162 | |
| 3163 | /* check if the page fault is solved */ |
| 3164 | fe->pte -= (fe->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
| 3165 | if (!pte_none(*fe->pte)) |
| 3166 | ret = VM_FAULT_NOPAGE; |
| 3167 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3168 | out: |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3169 | fe->address = address; |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3170 | fe->pte = NULL; |
| 3171 | return ret; |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3172 | } |
| 3173 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3174 | static int do_read_fault(struct fault_env *fe, pgoff_t pgoff) |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3175 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3176 | struct vm_area_struct *vma = fe->vma; |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3177 | struct page *fault_page; |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3178 | int ret = 0; |
| 3179 | |
| 3180 | /* |
| 3181 | * Let's call ->map_pages() first and use ->fault() as fallback |
| 3182 | * if page by the offset is not ready to be mapped (cold cache or |
| 3183 | * something). |
| 3184 | */ |
Kirill A. Shutemov | 9b4bdd2 | 2015-02-10 14:09:51 -0800 | [diff] [blame] | 3185 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3186 | ret = do_fault_around(fe, pgoff); |
| 3187 | if (ret) |
| 3188 | return ret; |
Kirill A. Shutemov | 8c6e50b | 2014-04-07 15:37:18 -0700 | [diff] [blame] | 3189 | } |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3190 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3191 | ret = __do_fault(fe, pgoff, NULL, &fault_page, NULL); |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3192 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
| 3193 | return ret; |
| 3194 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3195 | ret |= alloc_set_pte(fe, NULL, fault_page); |
| 3196 | if (fe->pte) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3197 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3198 | unlock_page(fault_page); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3199 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
| 3200 | put_page(fault_page); |
Kirill A. Shutemov | e655fb2 | 2014-04-03 14:48:11 -0700 | [diff] [blame] | 3201 | return ret; |
| 3202 | } |
| 3203 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3204 | static int do_cow_fault(struct fault_env *fe, pgoff_t pgoff) |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3205 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3206 | struct vm_area_struct *vma = fe->vma; |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3207 | struct page *fault_page, *new_page; |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 3208 | void *fault_entry; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 3209 | struct mem_cgroup *memcg; |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3210 | int ret; |
| 3211 | |
| 3212 | if (unlikely(anon_vma_prepare(vma))) |
| 3213 | return VM_FAULT_OOM; |
| 3214 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3215 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, fe->address); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3216 | if (!new_page) |
| 3217 | return VM_FAULT_OOM; |
| 3218 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3219 | if (mem_cgroup_try_charge(new_page, vma->vm_mm, GFP_KERNEL, |
| 3220 | &memcg, false)) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3221 | put_page(new_page); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3222 | return VM_FAULT_OOM; |
| 3223 | } |
| 3224 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3225 | ret = __do_fault(fe, pgoff, new_page, &fault_page, &fault_entry); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3226 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
| 3227 | goto uncharge_out; |
| 3228 | |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 3229 | if (!(ret & VM_FAULT_DAX_LOCKED)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3230 | copy_user_highpage(new_page, fault_page, fe->address, vma); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3231 | __SetPageUptodate(new_page); |
| 3232 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3233 | ret |= alloc_set_pte(fe, memcg, new_page); |
| 3234 | if (fe->pte) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3235 | pte_unmap_unlock(fe->pte, fe->ptl); |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 3236 | if (!(ret & VM_FAULT_DAX_LOCKED)) { |
Matthew Wilcox | 2e4cdab | 2015-02-16 15:58:50 -0800 | [diff] [blame] | 3237 | unlock_page(fault_page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3238 | put_page(fault_page); |
Matthew Wilcox | 2e4cdab | 2015-02-16 15:58:50 -0800 | [diff] [blame] | 3239 | } else { |
Jan Kara | bc2466e | 2016-05-12 18:29:19 +0200 | [diff] [blame] | 3240 | dax_unlock_mapping_entry(vma->vm_file->f_mapping, pgoff); |
Matthew Wilcox | 2e4cdab | 2015-02-16 15:58:50 -0800 | [diff] [blame] | 3241 | } |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3242 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
| 3243 | goto uncharge_out; |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3244 | return ret; |
| 3245 | uncharge_out: |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 3246 | mem_cgroup_cancel_charge(new_page, memcg, false); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3247 | put_page(new_page); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3248 | return ret; |
| 3249 | } |
| 3250 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3251 | static int do_shared_fault(struct fault_env *fe, pgoff_t pgoff) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3252 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3253 | struct vm_area_struct *vma = fe->vma; |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3254 | struct page *fault_page; |
| 3255 | struct address_space *mapping; |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3256 | int dirtied = 0; |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3257 | int ret, tmp; |
KAMEZAWA Hiroyuki | 1d65f86 | 2011-07-25 17:12:27 -0700 | [diff] [blame] | 3258 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3259 | ret = __do_fault(fe, pgoff, NULL, &fault_page, NULL); |
Kirill A. Shutemov | 7eae74a | 2014-04-03 14:48:10 -0700 | [diff] [blame] | 3260 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3261 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | |
| 3263 | /* |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3264 | * Check if the backing address space wants to know that the page is |
| 3265 | * about to become writable |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3266 | */ |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 3267 | if (vma->vm_ops->page_mkwrite) { |
| 3268 | unlock_page(fault_page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3269 | tmp = do_page_mkwrite(vma, fault_page, fe->address); |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 3270 | if (unlikely(!tmp || |
| 3271 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3272 | put_page(fault_page); |
Kirill A. Shutemov | fb09a46 | 2014-04-03 14:48:15 -0700 | [diff] [blame] | 3273 | return tmp; |
| 3274 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3275 | } |
| 3276 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3277 | ret |= alloc_set_pte(fe, NULL, fault_page); |
| 3278 | if (fe->pte) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3279 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3280 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
| 3281 | VM_FAULT_RETRY))) { |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3282 | unlock_page(fault_page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3283 | put_page(fault_page); |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3284 | return ret; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 3285 | } |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3286 | |
| 3287 | if (set_page_dirty(fault_page)) |
| 3288 | dirtied = 1; |
Andrew Morton | d82fa87 | 2014-12-18 16:17:29 -0800 | [diff] [blame] | 3289 | /* |
| 3290 | * Take a local copy of the address_space - page.mapping may be zeroed |
| 3291 | * by truncate after unlock_page(). The address_space itself remains |
| 3292 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s |
| 3293 | * release semantics to prevent the compiler from undoing this copying. |
| 3294 | */ |
Kirill A. Shutemov | 1c290f6 | 2016-01-15 16:52:07 -0800 | [diff] [blame] | 3295 | mapping = page_rmapping(fault_page); |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3296 | unlock_page(fault_page); |
| 3297 | if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { |
| 3298 | /* |
| 3299 | * Some device drivers do not set page.mapping but still |
| 3300 | * dirty their pages |
| 3301 | */ |
| 3302 | balance_dirty_pages_ratelimited(mapping); |
KAMEZAWA Hiroyuki | 1d65f86 | 2011-07-25 17:12:27 -0700 | [diff] [blame] | 3303 | } |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3304 | |
Johannes Weiner | 74ec675 | 2015-02-10 14:11:28 -0800 | [diff] [blame] | 3305 | if (!vma->vm_ops->page_mkwrite) |
Kirill A. Shutemov | f0c6d4d | 2014-04-03 14:48:13 -0700 | [diff] [blame] | 3306 | file_update_time(vma->vm_file); |
| 3307 | |
KAMEZAWA Hiroyuki | 1d65f86 | 2011-07-25 17:12:27 -0700 | [diff] [blame] | 3308 | return ret; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3309 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 3310 | |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3311 | /* |
| 3312 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
| 3313 | * but allow concurrent faults). |
| 3314 | * The mmap_sem may have been released depending on flags and our |
| 3315 | * return value. See filemap_fault() and __lock_page_or_retry(). |
| 3316 | */ |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3317 | static int do_fault(struct fault_env *fe) |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3318 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3319 | struct vm_area_struct *vma = fe->vma; |
| 3320 | pgoff_t pgoff = linear_page_index(vma, fe->address); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3321 | |
Kirill A. Shutemov | 6b7339f | 2015-07-06 23:18:37 +0300 | [diff] [blame] | 3322 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
| 3323 | if (!vma->vm_ops->fault) |
| 3324 | return VM_FAULT_SIGBUS; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3325 | if (!(fe->flags & FAULT_FLAG_WRITE)) |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3326 | return do_read_fault(fe, pgoff); |
Kirill A. Shutemov | ec47c3b | 2014-04-03 14:48:12 -0700 | [diff] [blame] | 3327 | if (!(vma->vm_flags & VM_SHARED)) |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3328 | return do_cow_fault(fe, pgoff); |
| 3329 | return do_shared_fault(fe, pgoff); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3330 | } |
| 3331 | |
Rashika Kheria | b19a993 | 2014-04-03 14:48:02 -0700 | [diff] [blame] | 3332 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
Rik van Riel | 04bb2f9 | 2013-10-07 11:29:36 +0100 | [diff] [blame] | 3333 | unsigned long addr, int page_nid, |
| 3334 | int *flags) |
Mel Gorman | 9532fec | 2012-11-15 01:24:32 +0000 | [diff] [blame] | 3335 | { |
| 3336 | get_page(page); |
| 3337 | |
| 3338 | count_vm_numa_event(NUMA_HINT_FAULTS); |
Rik van Riel | 04bb2f9 | 2013-10-07 11:29:36 +0100 | [diff] [blame] | 3339 | if (page_nid == numa_node_id()) { |
Mel Gorman | 9532fec | 2012-11-15 01:24:32 +0000 | [diff] [blame] | 3340 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
Rik van Riel | 04bb2f9 | 2013-10-07 11:29:36 +0100 | [diff] [blame] | 3341 | *flags |= TNF_FAULT_LOCAL; |
| 3342 | } |
Mel Gorman | 9532fec | 2012-11-15 01:24:32 +0000 | [diff] [blame] | 3343 | |
| 3344 | return mpol_misplaced(page, vma, addr); |
| 3345 | } |
| 3346 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3347 | static int do_numa_page(struct fault_env *fe, pte_t pte) |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3348 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3349 | struct vm_area_struct *vma = fe->vma; |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 3350 | struct page *page = NULL; |
Mel Gorman | 8191acb | 2013-10-07 11:28:45 +0100 | [diff] [blame] | 3351 | int page_nid = -1; |
Peter Zijlstra | 9057289 | 2013-10-07 11:29:20 +0100 | [diff] [blame] | 3352 | int last_cpupid; |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 3353 | int target_nid; |
Mel Gorman | b8593bf | 2012-11-21 01:18:23 +0000 | [diff] [blame] | 3354 | bool migrated = false; |
Mel Gorman | b191f9b | 2015-03-25 15:55:40 -0700 | [diff] [blame] | 3355 | bool was_writable = pte_write(pte); |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3356 | int flags = 0; |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3357 | |
| 3358 | /* |
| 3359 | * The "pte" at this point cannot be used safely without |
| 3360 | * validation through pte_unmap_same(). It's of NUMA type but |
| 3361 | * the pfn may be screwed if the read is non atomic. |
| 3362 | * |
Mel Gorman | 4d94246 | 2015-02-12 14:58:28 -0800 | [diff] [blame] | 3363 | * We can safely just do a "set_pte_at()", because the old |
| 3364 | * page table entry is not accessible, so there would be no |
| 3365 | * concurrent hardware modifications to the PTE. |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3366 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3367 | fe->ptl = pte_lockptr(vma->vm_mm, fe->pmd); |
| 3368 | spin_lock(fe->ptl); |
| 3369 | if (unlikely(!pte_same(*fe->pte, pte))) { |
| 3370 | pte_unmap_unlock(fe->pte, fe->ptl); |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 3371 | goto out; |
| 3372 | } |
| 3373 | |
Mel Gorman | 4d94246 | 2015-02-12 14:58:28 -0800 | [diff] [blame] | 3374 | /* Make it present again */ |
| 3375 | pte = pte_modify(pte, vma->vm_page_prot); |
| 3376 | pte = pte_mkyoung(pte); |
Mel Gorman | b191f9b | 2015-03-25 15:55:40 -0700 | [diff] [blame] | 3377 | if (was_writable) |
| 3378 | pte = pte_mkwrite(pte); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3379 | set_pte_at(vma->vm_mm, fe->address, fe->pte, pte); |
| 3380 | update_mmu_cache(vma, fe->address, fe->pte); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3381 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3382 | page = vm_normal_page(vma, fe->address, pte); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3383 | if (!page) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3384 | pte_unmap_unlock(fe->pte, fe->ptl); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3385 | return 0; |
| 3386 | } |
| 3387 | |
Kirill A. Shutemov | e81c480 | 2016-01-15 16:53:49 -0800 | [diff] [blame] | 3388 | /* TODO: handle PTE-mapped THP */ |
| 3389 | if (PageCompound(page)) { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3390 | pte_unmap_unlock(fe->pte, fe->ptl); |
Kirill A. Shutemov | e81c480 | 2016-01-15 16:53:49 -0800 | [diff] [blame] | 3391 | return 0; |
| 3392 | } |
| 3393 | |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3394 | /* |
Mel Gorman | bea66fb | 2015-03-25 15:55:37 -0700 | [diff] [blame] | 3395 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
| 3396 | * much anyway since they can be in shared cache state. This misses |
| 3397 | * the case where a mapping is writable but the process never writes |
| 3398 | * to it but pte_write gets cleared during protection updates and |
| 3399 | * pte_dirty has unpredictable behaviour between PTE scan updates, |
| 3400 | * background writeback, dirty balancing and application behaviour. |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3401 | */ |
Rik van Riel | d59dc7b | 2016-09-08 21:30:53 -0400 | [diff] [blame] | 3402 | if (!pte_write(pte)) |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3403 | flags |= TNF_NO_GROUP; |
| 3404 | |
Rik van Riel | dabe1d9 | 2013-10-07 11:29:34 +0100 | [diff] [blame] | 3405 | /* |
| 3406 | * Flag if the page is shared between multiple address spaces. This |
| 3407 | * is later used when determining whether to group tasks together |
| 3408 | */ |
| 3409 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) |
| 3410 | flags |= TNF_SHARED; |
| 3411 | |
Peter Zijlstra | 9057289 | 2013-10-07 11:29:20 +0100 | [diff] [blame] | 3412 | last_cpupid = page_cpupid_last(page); |
Mel Gorman | 8191acb | 2013-10-07 11:28:45 +0100 | [diff] [blame] | 3413 | page_nid = page_to_nid(page); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3414 | target_nid = numa_migrate_prep(page, vma, fe->address, page_nid, |
| 3415 | &flags); |
| 3416 | pte_unmap_unlock(fe->pte, fe->ptl); |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 3417 | if (target_nid == -1) { |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 3418 | put_page(page); |
| 3419 | goto out; |
| 3420 | } |
| 3421 | |
| 3422 | /* Migrate to the requested node */ |
Mel Gorman | 1bc115d | 2013-10-07 11:29:05 +0100 | [diff] [blame] | 3423 | migrated = migrate_misplaced_page(page, vma, target_nid); |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3424 | if (migrated) { |
Mel Gorman | 8191acb | 2013-10-07 11:28:45 +0100 | [diff] [blame] | 3425 | page_nid = target_nid; |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3426 | flags |= TNF_MIGRATED; |
Mel Gorman | 074c238 | 2015-03-25 15:55:42 -0700 | [diff] [blame] | 3427 | } else |
| 3428 | flags |= TNF_MIGRATE_FAIL; |
Mel Gorman | 4daae3b | 2012-11-02 11:33:45 +0000 | [diff] [blame] | 3429 | |
| 3430 | out: |
Mel Gorman | 8191acb | 2013-10-07 11:28:45 +0100 | [diff] [blame] | 3431 | if (page_nid != -1) |
Peter Zijlstra | 6688cc0 | 2013-10-07 11:29:24 +0100 | [diff] [blame] | 3432 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3433 | return 0; |
| 3434 | } |
| 3435 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3436 | static int create_huge_pmd(struct fault_env *fe) |
Matthew Wilcox | b96375f | 2015-09-08 14:58:48 -0700 | [diff] [blame] | 3437 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3438 | struct vm_area_struct *vma = fe->vma; |
Kirill A. Shutemov | fb6dd5f | 2015-09-09 15:39:35 -0700 | [diff] [blame] | 3439 | if (vma_is_anonymous(vma)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3440 | return do_huge_pmd_anonymous_page(fe); |
Matthew Wilcox | b96375f | 2015-09-08 14:58:48 -0700 | [diff] [blame] | 3441 | if (vma->vm_ops->pmd_fault) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3442 | return vma->vm_ops->pmd_fault(vma, fe->address, fe->pmd, |
| 3443 | fe->flags); |
Matthew Wilcox | b96375f | 2015-09-08 14:58:48 -0700 | [diff] [blame] | 3444 | return VM_FAULT_FALLBACK; |
| 3445 | } |
| 3446 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3447 | static int wp_huge_pmd(struct fault_env *fe, pmd_t orig_pmd) |
Matthew Wilcox | b96375f | 2015-09-08 14:58:48 -0700 | [diff] [blame] | 3448 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3449 | if (vma_is_anonymous(fe->vma)) |
| 3450 | return do_huge_pmd_wp_page(fe, orig_pmd); |
| 3451 | if (fe->vma->vm_ops->pmd_fault) |
| 3452 | return fe->vma->vm_ops->pmd_fault(fe->vma, fe->address, fe->pmd, |
| 3453 | fe->flags); |
Kirill A. Shutemov | af9e4d5 | 2016-07-26 15:25:40 -0700 | [diff] [blame] | 3454 | |
| 3455 | /* COW handled on pte level: split pmd */ |
| 3456 | VM_BUG_ON_VMA(fe->vma->vm_flags & VM_SHARED, fe->vma); |
| 3457 | split_huge_pmd(fe->vma, fe->pmd, fe->address); |
| 3458 | |
Matthew Wilcox | b96375f | 2015-09-08 14:58:48 -0700 | [diff] [blame] | 3459 | return VM_FAULT_FALLBACK; |
| 3460 | } |
| 3461 | |
Lorenzo Stoakes | 38e0885 | 2016-09-11 23:54:25 +0100 | [diff] [blame] | 3462 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
| 3463 | { |
| 3464 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); |
| 3465 | } |
| 3466 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3467 | /* |
| 3468 | * These routines also need to handle stuff like marking pages dirty |
| 3469 | * and/or accessed for architectures that don't do it in hardware (most |
| 3470 | * RISC architectures). The early dirtying is also good on the i386. |
| 3471 | * |
| 3472 | * There is also a hook called "update_mmu_cache()" that architectures |
| 3473 | * with external mmu caches can use to update those (ie the Sparc or |
| 3474 | * PowerPC hashed page tables that act as extended TLBs). |
| 3475 | * |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3476 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
| 3477 | * concurrent faults). |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3478 | * |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3479 | * The mmap_sem may have been released depending on flags and our return value. |
| 3480 | * See filemap_fault() and __lock_page_or_retry(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3481 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3482 | static int handle_pte_fault(struct fault_env *fe) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3483 | { |
| 3484 | pte_t entry; |
| 3485 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3486 | if (unlikely(pmd_none(*fe->pmd))) { |
| 3487 | /* |
| 3488 | * Leave __pte_alloc() until later: because vm_ops->fault may |
| 3489 | * want to allocate huge page, and if we expose page table |
| 3490 | * for an instant, it will be difficult to retract from |
| 3491 | * concurrent faults and from rmap lookups. |
| 3492 | */ |
| 3493 | fe->pte = NULL; |
| 3494 | } else { |
| 3495 | /* See comment in pte_alloc_one_map() */ |
| 3496 | if (pmd_trans_unstable(fe->pmd) || pmd_devmap(*fe->pmd)) |
| 3497 | return 0; |
| 3498 | /* |
| 3499 | * A regular pmd is established and it can't morph into a huge |
| 3500 | * pmd from under us anymore at this point because we hold the |
| 3501 | * mmap_sem read mode and khugepaged takes it in write mode. |
| 3502 | * So now it's safe to run pte_offset_map(). |
| 3503 | */ |
| 3504 | fe->pte = pte_offset_map(fe->pmd, fe->address); |
| 3505 | |
| 3506 | entry = *fe->pte; |
| 3507 | |
| 3508 | /* |
| 3509 | * some architectures can have larger ptes than wordsize, |
| 3510 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and |
| 3511 | * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee |
| 3512 | * atomic accesses. The code below just needs a consistent |
| 3513 | * view for the ifs and we later double check anyway with the |
| 3514 | * ptl lock held. So here a barrier will do. |
| 3515 | */ |
| 3516 | barrier(); |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 3517 | if (pte_none(entry)) { |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3518 | pte_unmap(fe->pte); |
| 3519 | fe->pte = NULL; |
Hugh Dickins | 65500d2 | 2005-10-29 18:15:59 -0700 | [diff] [blame] | 3520 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3521 | } |
| 3522 | |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3523 | if (!fe->pte) { |
| 3524 | if (vma_is_anonymous(fe->vma)) |
| 3525 | return do_anonymous_page(fe); |
| 3526 | else |
| 3527 | return do_fault(fe); |
| 3528 | } |
| 3529 | |
| 3530 | if (!pte_present(entry)) |
| 3531 | return do_swap_page(fe, entry); |
| 3532 | |
Lorenzo Stoakes | 38e0885 | 2016-09-11 23:54:25 +0100 | [diff] [blame] | 3533 | if (pte_protnone(entry) && vma_is_accessible(fe->vma)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3534 | return do_numa_page(fe, entry); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3535 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3536 | fe->ptl = pte_lockptr(fe->vma->vm_mm, fe->pmd); |
| 3537 | spin_lock(fe->ptl); |
| 3538 | if (unlikely(!pte_same(*fe->pte, entry))) |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 3539 | goto unlock; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3540 | if (fe->flags & FAULT_FLAG_WRITE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3541 | if (!pte_write(entry)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3542 | return do_wp_page(fe, entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3543 | entry = pte_mkdirty(entry); |
| 3544 | } |
| 3545 | entry = pte_mkyoung(entry); |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3546 | if (ptep_set_access_flags(fe->vma, fe->address, fe->pte, entry, |
| 3547 | fe->flags & FAULT_FLAG_WRITE)) { |
| 3548 | update_mmu_cache(fe->vma, fe->address, fe->pte); |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 3549 | } else { |
| 3550 | /* |
| 3551 | * This is needed only for protection faults but the arch code |
| 3552 | * is not yet telling us if this is a protection fault or not. |
| 3553 | * This still avoids useless tlb flushes for .text page faults |
| 3554 | * with threads. |
| 3555 | */ |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3556 | if (fe->flags & FAULT_FLAG_WRITE) |
| 3557 | flush_tlb_fix_spurious_fault(fe->vma, fe->address); |
Andrea Arcangeli | 1a44e14 | 2005-10-29 18:16:48 -0700 | [diff] [blame] | 3558 | } |
Hugh Dickins | 8f4e210 | 2005-10-29 18:16:26 -0700 | [diff] [blame] | 3559 | unlock: |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3560 | pte_unmap_unlock(fe->pte, fe->ptl); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 3561 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3562 | } |
| 3563 | |
| 3564 | /* |
| 3565 | * By the time we get here, we already hold the mm semaphore |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3566 | * |
| 3567 | * The mmap_sem may have been released depending on flags and our |
| 3568 | * return value. See filemap_fault() and __lock_page_or_retry(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3569 | */ |
Kirill A. Shutemov | dcddffd | 2016-07-26 15:25:18 -0700 | [diff] [blame] | 3570 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
| 3571 | unsigned int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3572 | { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3573 | struct fault_env fe = { |
| 3574 | .vma = vma, |
| 3575 | .address = address, |
| 3576 | .flags = flags, |
| 3577 | }; |
Kirill A. Shutemov | dcddffd | 2016-07-26 15:25:18 -0700 | [diff] [blame] | 3578 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3579 | pgd_t *pgd; |
| 3580 | pud_t *pud; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3581 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3582 | pgd = pgd_offset(mm, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3583 | pud = pud_alloc(mm, pgd, address); |
| 3584 | if (!pud) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3585 | return VM_FAULT_OOM; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3586 | fe.pmd = pmd_alloc(mm, pud, address); |
| 3587 | if (!fe.pmd) |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3588 | return VM_FAULT_OOM; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3589 | if (pmd_none(*fe.pmd) && transparent_hugepage_enabled(vma)) { |
| 3590 | int ret = create_huge_pmd(&fe); |
Kirill A. Shutemov | c029255 | 2013-09-12 15:14:05 -0700 | [diff] [blame] | 3591 | if (!(ret & VM_FAULT_FALLBACK)) |
| 3592 | return ret; |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 3593 | } else { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3594 | pmd_t orig_pmd = *fe.pmd; |
David Rientjes | 1f1d06c | 2012-05-29 15:06:23 -0700 | [diff] [blame] | 3595 | int ret; |
| 3596 | |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 3597 | barrier(); |
Dan Williams | 5c7fb56 | 2016-01-15 16:56:52 -0800 | [diff] [blame] | 3598 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
Lorenzo Stoakes | 38e0885 | 2016-09-11 23:54:25 +0100 | [diff] [blame] | 3599 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3600 | return do_huge_pmd_numa_page(&fe, orig_pmd); |
Mel Gorman | d10e63f | 2012-10-25 14:16:31 +0200 | [diff] [blame] | 3601 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3602 | if ((fe.flags & FAULT_FLAG_WRITE) && |
| 3603 | !pmd_write(orig_pmd)) { |
| 3604 | ret = wp_huge_pmd(&fe, orig_pmd); |
Kirill A. Shutemov | 9845cbb | 2014-02-25 15:01:42 -0800 | [diff] [blame] | 3605 | if (!(ret & VM_FAULT_FALLBACK)) |
| 3606 | return ret; |
Will Deacon | a1dd450 | 2012-12-11 16:01:27 -0800 | [diff] [blame] | 3607 | } else { |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3608 | huge_pmd_set_accessed(&fe, orig_pmd); |
Kirill A. Shutemov | 9845cbb | 2014-02-25 15:01:42 -0800 | [diff] [blame] | 3609 | return 0; |
David Rientjes | 1f1d06c | 2012-05-29 15:06:23 -0700 | [diff] [blame] | 3610 | } |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 3611 | } |
| 3612 | } |
| 3613 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3614 | return handle_pte_fault(&fe); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | } |
| 3616 | |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3617 | /* |
| 3618 | * By the time we get here, we already hold the mm semaphore |
| 3619 | * |
| 3620 | * The mmap_sem may have been released depending on flags and our |
| 3621 | * return value. See filemap_fault() and __lock_page_or_retry(). |
| 3622 | */ |
Kirill A. Shutemov | dcddffd | 2016-07-26 15:25:18 -0700 | [diff] [blame] | 3623 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
| 3624 | unsigned int flags) |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3625 | { |
| 3626 | int ret; |
| 3627 | |
| 3628 | __set_current_state(TASK_RUNNING); |
| 3629 | |
| 3630 | count_vm_event(PGFAULT); |
Kirill A. Shutemov | dcddffd | 2016-07-26 15:25:18 -0700 | [diff] [blame] | 3631 | mem_cgroup_count_vm_event(vma->vm_mm, PGFAULT); |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3632 | |
| 3633 | /* do counter updates before entering really critical section. */ |
| 3634 | check_sync_rss_stat(current); |
| 3635 | |
| 3636 | /* |
| 3637 | * Enable the memcg OOM handling for faults triggered in user |
| 3638 | * space. Kernel faults are handled more gracefully. |
| 3639 | */ |
| 3640 | if (flags & FAULT_FLAG_USER) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 3641 | mem_cgroup_oom_enable(); |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3642 | |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3643 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
| 3644 | flags & FAULT_FLAG_INSTRUCTION, |
| 3645 | flags & FAULT_FLAG_REMOTE)) |
| 3646 | return VM_FAULT_SIGSEGV; |
| 3647 | |
| 3648 | if (unlikely(is_vm_hugetlb_page(vma))) |
| 3649 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); |
| 3650 | else |
| 3651 | ret = __handle_mm_fault(vma, address, flags); |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3652 | |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 3653 | if (flags & FAULT_FLAG_USER) { |
| 3654 | mem_cgroup_oom_disable(); |
| 3655 | /* |
| 3656 | * The task may have entered a memcg OOM situation but |
| 3657 | * if the allocation error was handled gracefully (no |
| 3658 | * VM_FAULT_OOM), there is no need to kill anything. |
| 3659 | * Just clean up the OOM state peacefully. |
| 3660 | */ |
| 3661 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) |
| 3662 | mem_cgroup_oom_synchronize(false); |
| 3663 | } |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 3664 | |
Michal Hocko | 3f70dc3 | 2016-10-07 16:59:06 -0700 | [diff] [blame] | 3665 | /* |
| 3666 | * This mm has been already reaped by the oom reaper and so the |
| 3667 | * refault cannot be trusted in general. Anonymous refaults would |
| 3668 | * lose data and give a zero page instead e.g. This is especially |
| 3669 | * problem for use_mm() because regular tasks will just die and |
| 3670 | * the corrupted data will not be visible anywhere while kthread |
| 3671 | * will outlive the oom victim and potentially propagate the date |
| 3672 | * further. |
| 3673 | */ |
| 3674 | if (unlikely((current->flags & PF_KTHREAD) && !(ret & VM_FAULT_ERROR) |
| 3675 | && test_bit(MMF_UNSTABLE, &vma->vm_mm->flags))) |
| 3676 | ret = VM_FAULT_SIGBUS; |
| 3677 | |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3678 | return ret; |
| 3679 | } |
Jesse Barnes | e1d6d01 | 2014-12-12 16:55:27 -0800 | [diff] [blame] | 3680 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
Johannes Weiner | 519e524 | 2013-09-12 15:13:42 -0700 | [diff] [blame] | 3681 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3682 | #ifndef __PAGETABLE_PUD_FOLDED |
| 3683 | /* |
| 3684 | * Allocate page upper directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 3685 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3686 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3687 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3688 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3689 | pud_t *new = pud_alloc_one(mm, address); |
| 3690 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3691 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3692 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 3693 | smp_wmb(); /* See comment in __pte_alloc */ |
| 3694 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 3695 | spin_lock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3696 | if (pgd_present(*pgd)) /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 3697 | pud_free(mm, new); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3698 | else |
| 3699 | pgd_populate(mm, pgd, new); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3700 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3701 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3702 | } |
| 3703 | #endif /* __PAGETABLE_PUD_FOLDED */ |
| 3704 | |
| 3705 | #ifndef __PAGETABLE_PMD_FOLDED |
| 3706 | /* |
| 3707 | * Allocate page middle directory. |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 3708 | * We've already handled the fast-path in-line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3709 | */ |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3710 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3711 | { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3712 | pmd_t *new = pmd_alloc_one(mm, address); |
| 3713 | if (!new) |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3714 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3715 | |
Nick Piggin | 362a61a | 2008-05-14 06:37:36 +0200 | [diff] [blame] | 3716 | smp_wmb(); /* See comment in __pte_alloc */ |
| 3717 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 3718 | spin_lock(&mm->page_table_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3719 | #ifndef __ARCH_HAS_4LEVEL_HACK |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 3720 | if (!pud_present(*pud)) { |
| 3721 | mm_inc_nr_pmds(mm); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3722 | pud_populate(mm, pud, new); |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 3723 | } else /* Another has populated it */ |
Benjamin Herrenschmidt | 5e54197 | 2008-02-04 22:29:14 -0800 | [diff] [blame] | 3724 | pmd_free(mm, new); |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 3725 | #else |
| 3726 | if (!pgd_present(*pud)) { |
| 3727 | mm_inc_nr_pmds(mm); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3728 | pgd_populate(mm, pud, new); |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 3729 | } else /* Another has populated it */ |
| 3730 | pmd_free(mm, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3731 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 3732 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 1bb3630 | 2005-10-29 18:16:22 -0700 | [diff] [blame] | 3733 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3734 | } |
| 3735 | #endif /* __PAGETABLE_PMD_FOLDED */ |
| 3736 | |
Namhyung Kim | 1b36ba8 | 2010-10-26 14:22:00 -0700 | [diff] [blame] | 3737 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
Johannes Weiner | f8ad0f49 | 2009-06-16 15:32:33 -0700 | [diff] [blame] | 3738 | pte_t **ptepp, spinlock_t **ptlp) |
| 3739 | { |
| 3740 | pgd_t *pgd; |
| 3741 | pud_t *pud; |
| 3742 | pmd_t *pmd; |
| 3743 | pte_t *ptep; |
| 3744 | |
| 3745 | pgd = pgd_offset(mm, address); |
| 3746 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
| 3747 | goto out; |
| 3748 | |
| 3749 | pud = pud_offset(pgd, address); |
| 3750 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
| 3751 | goto out; |
| 3752 | |
| 3753 | pmd = pmd_offset(pud, address); |
Andrea Arcangeli | f66055ab | 2011-01-13 15:46:54 -0800 | [diff] [blame] | 3754 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
Johannes Weiner | f8ad0f49 | 2009-06-16 15:32:33 -0700 | [diff] [blame] | 3755 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
| 3756 | goto out; |
| 3757 | |
| 3758 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ |
| 3759 | if (pmd_huge(*pmd)) |
| 3760 | goto out; |
| 3761 | |
| 3762 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
| 3763 | if (!ptep) |
| 3764 | goto out; |
| 3765 | if (!pte_present(*ptep)) |
| 3766 | goto unlock; |
| 3767 | *ptepp = ptep; |
| 3768 | return 0; |
| 3769 | unlock: |
| 3770 | pte_unmap_unlock(ptep, *ptlp); |
| 3771 | out: |
| 3772 | return -EINVAL; |
| 3773 | } |
| 3774 | |
Namhyung Kim | 1b36ba8 | 2010-10-26 14:22:00 -0700 | [diff] [blame] | 3775 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
| 3776 | pte_t **ptepp, spinlock_t **ptlp) |
| 3777 | { |
| 3778 | int res; |
| 3779 | |
| 3780 | /* (void) is needed to make gcc happy */ |
| 3781 | (void) __cond_lock(*ptlp, |
| 3782 | !(res = __follow_pte(mm, address, ptepp, ptlp))); |
| 3783 | return res; |
| 3784 | } |
| 3785 | |
Johannes Weiner | 3b6748e | 2009-06-16 15:32:35 -0700 | [diff] [blame] | 3786 | /** |
| 3787 | * follow_pfn - look up PFN at a user virtual address |
| 3788 | * @vma: memory mapping |
| 3789 | * @address: user virtual address |
| 3790 | * @pfn: location to store found PFN |
| 3791 | * |
| 3792 | * Only IO mappings and raw PFN mappings are allowed. |
| 3793 | * |
| 3794 | * Returns zero and the pfn at @pfn on success, -ve otherwise. |
| 3795 | */ |
| 3796 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, |
| 3797 | unsigned long *pfn) |
| 3798 | { |
| 3799 | int ret = -EINVAL; |
| 3800 | spinlock_t *ptl; |
| 3801 | pte_t *ptep; |
| 3802 | |
| 3803 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
| 3804 | return ret; |
| 3805 | |
| 3806 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); |
| 3807 | if (ret) |
| 3808 | return ret; |
| 3809 | *pfn = pte_pfn(*ptep); |
| 3810 | pte_unmap_unlock(ptep, ptl); |
| 3811 | return 0; |
| 3812 | } |
| 3813 | EXPORT_SYMBOL(follow_pfn); |
| 3814 | |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3815 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3816 | int follow_phys(struct vm_area_struct *vma, |
| 3817 | unsigned long address, unsigned int flags, |
| 3818 | unsigned long *prot, resource_size_t *phys) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3819 | { |
Johannes Weiner | 03668a4 | 2009-06-16 15:32:34 -0700 | [diff] [blame] | 3820 | int ret = -EINVAL; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3821 | pte_t *ptep, pte; |
| 3822 | spinlock_t *ptl; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3823 | |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3824 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
| 3825 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3826 | |
Johannes Weiner | 03668a4 | 2009-06-16 15:32:34 -0700 | [diff] [blame] | 3827 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3828 | goto out; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3829 | pte = *ptep; |
Johannes Weiner | 03668a4 | 2009-06-16 15:32:34 -0700 | [diff] [blame] | 3830 | |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3831 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
| 3832 | goto unlock; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3833 | |
| 3834 | *prot = pgprot_val(pte_pgprot(pte)); |
Johannes Weiner | 03668a4 | 2009-06-16 15:32:34 -0700 | [diff] [blame] | 3835 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3836 | |
Johannes Weiner | 03668a4 | 2009-06-16 15:32:34 -0700 | [diff] [blame] | 3837 | ret = 0; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3838 | unlock: |
| 3839 | pte_unmap_unlock(ptep, ptl); |
| 3840 | out: |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3841 | return ret; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3842 | } |
| 3843 | |
| 3844 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
| 3845 | void *buf, int len, int write) |
| 3846 | { |
| 3847 | resource_size_t phys_addr; |
| 3848 | unsigned long prot = 0; |
KOSAKI Motohiro | 2bc7273 | 2009-01-06 14:39:43 -0800 | [diff] [blame] | 3849 | void __iomem *maddr; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3850 | int offset = addr & (PAGE_SIZE-1); |
| 3851 | |
venkatesh.pallipadi@intel.com | d87fe66 | 2008-12-19 13:47:27 -0800 | [diff] [blame] | 3852 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3853 | return -EINVAL; |
| 3854 | |
Grazvydas Ignotas | 9cb12d7 | 2015-02-12 15:00:19 -0800 | [diff] [blame] | 3855 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3856 | if (write) |
| 3857 | memcpy_toio(maddr + offset, buf, len); |
| 3858 | else |
| 3859 | memcpy_fromio(buf, maddr + offset, len); |
| 3860 | iounmap(maddr); |
| 3861 | |
| 3862 | return len; |
| 3863 | } |
Uwe Kleine-König | 5a73633 | 2013-08-07 13:02:52 +0200 | [diff] [blame] | 3864 | EXPORT_SYMBOL_GPL(generic_access_phys); |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3865 | #endif |
| 3866 | |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3867 | /* |
Stephen Wilson | 206cb63 | 2011-03-13 15:49:19 -0400 | [diff] [blame] | 3868 | * Access another process' address space as given in mm. If non-NULL, use the |
| 3869 | * given task for page fault accounting. |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3870 | */ |
Stephen Wilson | 206cb63 | 2011-03-13 15:49:19 -0400 | [diff] [blame] | 3871 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
Lorenzo Stoakes | 442486e | 2016-10-13 01:20:18 +0100 | [diff] [blame] | 3872 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3873 | { |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3874 | struct vm_area_struct *vma; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3875 | void *old_buf = buf; |
Lorenzo Stoakes | 442486e | 2016-10-13 01:20:18 +0100 | [diff] [blame] | 3876 | int write = gup_flags & FOLL_WRITE; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3877 | |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3878 | down_read(&mm->mmap_sem); |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 3879 | /* ignore errors, just check how much was successfully transferred */ |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3880 | while (len) { |
| 3881 | int bytes, ret, offset; |
| 3882 | void *maddr; |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3883 | struct page *page = NULL; |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3884 | |
Dave Hansen | 1e98779 | 2016-02-12 13:01:54 -0800 | [diff] [blame] | 3885 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
Lorenzo Stoakes | 442486e | 2016-10-13 01:20:18 +0100 | [diff] [blame] | 3886 | gup_flags, &page, &vma); |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3887 | if (ret <= 0) { |
Rik van Riel | dbffcd0 | 2014-08-06 16:08:12 -0700 | [diff] [blame] | 3888 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
| 3889 | break; |
| 3890 | #else |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3891 | /* |
| 3892 | * Check if this is a VM_IO | VM_PFNMAP VMA, which |
| 3893 | * we can access using slightly different code. |
| 3894 | */ |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3895 | vma = find_vma(mm, addr); |
Michael Ellerman | fe936df | 2011-04-14 15:22:10 -0700 | [diff] [blame] | 3896 | if (!vma || vma->vm_start > addr) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3897 | break; |
| 3898 | if (vma->vm_ops && vma->vm_ops->access) |
| 3899 | ret = vma->vm_ops->access(vma, addr, buf, |
| 3900 | len, write); |
| 3901 | if (ret <= 0) |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3902 | break; |
| 3903 | bytes = ret; |
Rik van Riel | dbffcd0 | 2014-08-06 16:08:12 -0700 | [diff] [blame] | 3904 | #endif |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3905 | } else { |
Rik van Riel | 28b2ee2 | 2008-07-23 21:27:05 -0700 | [diff] [blame] | 3906 | bytes = len; |
| 3907 | offset = addr & (PAGE_SIZE-1); |
| 3908 | if (bytes > PAGE_SIZE-offset) |
| 3909 | bytes = PAGE_SIZE-offset; |
| 3910 | |
| 3911 | maddr = kmap(page); |
| 3912 | if (write) { |
| 3913 | copy_to_user_page(vma, page, addr, |
| 3914 | maddr + offset, buf, bytes); |
| 3915 | set_page_dirty_lock(page); |
| 3916 | } else { |
| 3917 | copy_from_user_page(vma, page, addr, |
| 3918 | buf, maddr + offset, bytes); |
| 3919 | } |
| 3920 | kunmap(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3921 | put_page(page); |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3922 | } |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3923 | len -= bytes; |
| 3924 | buf += bytes; |
| 3925 | addr += bytes; |
| 3926 | } |
| 3927 | up_read(&mm->mmap_sem); |
David Howells | 0ec76a1 | 2006-09-27 01:50:15 -0700 | [diff] [blame] | 3928 | |
| 3929 | return buf - old_buf; |
| 3930 | } |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3931 | |
Stephen Wilson | 5ddd36b | 2011-03-13 15:49:20 -0400 | [diff] [blame] | 3932 | /** |
Randy Dunlap | ae91dbf | 2011-03-26 13:27:01 -0700 | [diff] [blame] | 3933 | * access_remote_vm - access another process' address space |
Stephen Wilson | 5ddd36b | 2011-03-13 15:49:20 -0400 | [diff] [blame] | 3934 | * @mm: the mm_struct of the target address space |
| 3935 | * @addr: start address to access |
| 3936 | * @buf: source or destination buffer |
| 3937 | * @len: number of bytes to transfer |
Lorenzo Stoakes | 6347e8d | 2016-10-13 01:20:19 +0100 | [diff] [blame] | 3938 | * @gup_flags: flags modifying lookup behaviour |
Stephen Wilson | 5ddd36b | 2011-03-13 15:49:20 -0400 | [diff] [blame] | 3939 | * |
| 3940 | * The caller must hold a reference on @mm. |
| 3941 | */ |
| 3942 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, |
Lorenzo Stoakes | 6347e8d | 2016-10-13 01:20:19 +0100 | [diff] [blame] | 3943 | void *buf, int len, unsigned int gup_flags) |
Stephen Wilson | 5ddd36b | 2011-03-13 15:49:20 -0400 | [diff] [blame] | 3944 | { |
Lorenzo Stoakes | 6347e8d | 2016-10-13 01:20:19 +0100 | [diff] [blame] | 3945 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
Stephen Wilson | 5ddd36b | 2011-03-13 15:49:20 -0400 | [diff] [blame] | 3946 | } |
| 3947 | |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3948 | /* |
Stephen Wilson | 206cb63 | 2011-03-13 15:49:19 -0400 | [diff] [blame] | 3949 | * Access another process' address space. |
| 3950 | * Source/target buffer must be kernel space, |
| 3951 | * Do not walk the page table directly, use get_user_pages |
| 3952 | */ |
| 3953 | int access_process_vm(struct task_struct *tsk, unsigned long addr, |
Lorenzo Stoakes | f307ab6 | 2016-10-13 01:20:20 +0100 | [diff] [blame] | 3954 | void *buf, int len, unsigned int gup_flags) |
Stephen Wilson | 206cb63 | 2011-03-13 15:49:19 -0400 | [diff] [blame] | 3955 | { |
| 3956 | struct mm_struct *mm; |
| 3957 | int ret; |
| 3958 | |
| 3959 | mm = get_task_mm(tsk); |
| 3960 | if (!mm) |
| 3961 | return 0; |
| 3962 | |
Lorenzo Stoakes | f307ab6 | 2016-10-13 01:20:20 +0100 | [diff] [blame] | 3963 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
Lorenzo Stoakes | 442486e | 2016-10-13 01:20:18 +0100 | [diff] [blame] | 3964 | |
Stephen Wilson | 206cb63 | 2011-03-13 15:49:19 -0400 | [diff] [blame] | 3965 | mmput(mm); |
| 3966 | |
| 3967 | return ret; |
| 3968 | } |
| 3969 | |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3970 | /* |
| 3971 | * Print the name of a VMA. |
| 3972 | */ |
| 3973 | void print_vma_addr(char *prefix, unsigned long ip) |
| 3974 | { |
| 3975 | struct mm_struct *mm = current->mm; |
| 3976 | struct vm_area_struct *vma; |
| 3977 | |
Ingo Molnar | e8bff74 | 2008-02-13 20:21:06 +0100 | [diff] [blame] | 3978 | /* |
| 3979 | * Do not print if we are in atomic |
| 3980 | * contexts (in exception stacks, etc.): |
| 3981 | */ |
| 3982 | if (preempt_count()) |
| 3983 | return; |
| 3984 | |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3985 | down_read(&mm->mmap_sem); |
| 3986 | vma = find_vma(mm, ip); |
| 3987 | if (vma && vma->vm_file) { |
| 3988 | struct file *f = vma->vm_file; |
| 3989 | char *buf = (char *)__get_free_page(GFP_KERNEL); |
| 3990 | if (buf) { |
Andy Shevchenko | 2fbc57c | 2012-12-17 16:01:23 -0800 | [diff] [blame] | 3991 | char *p; |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3992 | |
Miklos Szeredi | 9bf39ab | 2015-06-19 10:29:13 +0200 | [diff] [blame] | 3993 | p = file_path(f, buf, PAGE_SIZE); |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3994 | if (IS_ERR(p)) |
| 3995 | p = "?"; |
Andy Shevchenko | 2fbc57c | 2012-12-17 16:01:23 -0800 | [diff] [blame] | 3996 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 3997 | vma->vm_start, |
| 3998 | vma->vm_end - vma->vm_start); |
| 3999 | free_page((unsigned long)buf); |
| 4000 | } |
| 4001 | } |
Jeff Liu | 51a07e5 | 2012-07-31 16:43:18 -0700 | [diff] [blame] | 4002 | up_read(&mm->mmap_sem); |
Andi Kleen | 0325291 | 2008-01-30 13:33:18 +0100 | [diff] [blame] | 4003 | } |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 4004 | |
Michael S. Tsirkin | 662bbcb | 2013-05-26 17:32:23 +0300 | [diff] [blame] | 4005 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
David Hildenbrand | 9ec2353 | 2015-05-11 17:52:07 +0200 | [diff] [blame] | 4006 | void __might_fault(const char *file, int line) |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 4007 | { |
Peter Zijlstra | 95156f0 | 2009-01-12 13:02:11 +0100 | [diff] [blame] | 4008 | /* |
| 4009 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while |
| 4010 | * holding the mmap_sem, this is safe because kernel memory doesn't |
| 4011 | * get paged out, therefore we'll never actually fault, and the |
| 4012 | * below annotations will generate false positives. |
| 4013 | */ |
| 4014 | if (segment_eq(get_fs(), KERNEL_DS)) |
| 4015 | return; |
David Hildenbrand | 9ec2353 | 2015-05-11 17:52:07 +0200 | [diff] [blame] | 4016 | if (pagefault_disabled()) |
Michael S. Tsirkin | 662bbcb | 2013-05-26 17:32:23 +0300 | [diff] [blame] | 4017 | return; |
David Hildenbrand | 9ec2353 | 2015-05-11 17:52:07 +0200 | [diff] [blame] | 4018 | __might_sleep(file, line, 0); |
| 4019 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
Michael S. Tsirkin | 662bbcb | 2013-05-26 17:32:23 +0300 | [diff] [blame] | 4020 | if (current->mm) |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 4021 | might_lock_read(¤t->mm->mmap_sem); |
David Hildenbrand | 9ec2353 | 2015-05-11 17:52:07 +0200 | [diff] [blame] | 4022 | #endif |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 4023 | } |
David Hildenbrand | 9ec2353 | 2015-05-11 17:52:07 +0200 | [diff] [blame] | 4024 | EXPORT_SYMBOL(__might_fault); |
Nick Piggin | 3ee1afa | 2008-09-10 13:37:17 +0200 | [diff] [blame] | 4025 | #endif |
Andrea Arcangeli | 47ad847 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 4026 | |
| 4027 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) |
| 4028 | static void clear_gigantic_page(struct page *page, |
| 4029 | unsigned long addr, |
| 4030 | unsigned int pages_per_huge_page) |
| 4031 | { |
| 4032 | int i; |
| 4033 | struct page *p = page; |
| 4034 | |
| 4035 | might_sleep(); |
| 4036 | for (i = 0; i < pages_per_huge_page; |
| 4037 | i++, p = mem_map_next(p, page, i)) { |
| 4038 | cond_resched(); |
| 4039 | clear_user_highpage(p, addr + i * PAGE_SIZE); |
| 4040 | } |
| 4041 | } |
| 4042 | void clear_huge_page(struct page *page, |
| 4043 | unsigned long addr, unsigned int pages_per_huge_page) |
| 4044 | { |
| 4045 | int i; |
| 4046 | |
| 4047 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { |
| 4048 | clear_gigantic_page(page, addr, pages_per_huge_page); |
| 4049 | return; |
| 4050 | } |
| 4051 | |
| 4052 | might_sleep(); |
| 4053 | for (i = 0; i < pages_per_huge_page; i++) { |
| 4054 | cond_resched(); |
| 4055 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); |
| 4056 | } |
| 4057 | } |
| 4058 | |
| 4059 | static void copy_user_gigantic_page(struct page *dst, struct page *src, |
| 4060 | unsigned long addr, |
| 4061 | struct vm_area_struct *vma, |
| 4062 | unsigned int pages_per_huge_page) |
| 4063 | { |
| 4064 | int i; |
| 4065 | struct page *dst_base = dst; |
| 4066 | struct page *src_base = src; |
| 4067 | |
| 4068 | for (i = 0; i < pages_per_huge_page; ) { |
| 4069 | cond_resched(); |
| 4070 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); |
| 4071 | |
| 4072 | i++; |
| 4073 | dst = mem_map_next(dst, dst_base, i); |
| 4074 | src = mem_map_next(src, src_base, i); |
| 4075 | } |
| 4076 | } |
| 4077 | |
| 4078 | void copy_user_huge_page(struct page *dst, struct page *src, |
| 4079 | unsigned long addr, struct vm_area_struct *vma, |
| 4080 | unsigned int pages_per_huge_page) |
| 4081 | { |
| 4082 | int i; |
| 4083 | |
| 4084 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { |
| 4085 | copy_user_gigantic_page(dst, src, addr, vma, |
| 4086 | pages_per_huge_page); |
| 4087 | return; |
| 4088 | } |
| 4089 | |
| 4090 | might_sleep(); |
| 4091 | for (i = 0; i < pages_per_huge_page; i++) { |
| 4092 | cond_resched(); |
| 4093 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); |
| 4094 | } |
| 4095 | } |
| 4096 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4097 | |
Olof Johansson | 40b64ac | 2013-12-20 14:28:05 -0800 | [diff] [blame] | 4098 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
Kirill A. Shutemov | b35f181 | 2014-01-21 15:49:07 -0800 | [diff] [blame] | 4099 | |
| 4100 | static struct kmem_cache *page_ptl_cachep; |
| 4101 | |
| 4102 | void __init ptlock_cache_init(void) |
| 4103 | { |
| 4104 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, |
| 4105 | SLAB_PANIC, NULL); |
| 4106 | } |
| 4107 | |
Peter Zijlstra | 539edb5 | 2013-11-14 14:31:52 -0800 | [diff] [blame] | 4108 | bool ptlock_alloc(struct page *page) |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4109 | { |
| 4110 | spinlock_t *ptl; |
| 4111 | |
Kirill A. Shutemov | b35f181 | 2014-01-21 15:49:07 -0800 | [diff] [blame] | 4112 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4113 | if (!ptl) |
| 4114 | return false; |
Peter Zijlstra | 539edb5 | 2013-11-14 14:31:52 -0800 | [diff] [blame] | 4115 | page->ptl = ptl; |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4116 | return true; |
| 4117 | } |
| 4118 | |
Peter Zijlstra | 539edb5 | 2013-11-14 14:31:52 -0800 | [diff] [blame] | 4119 | void ptlock_free(struct page *page) |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4120 | { |
Kirill A. Shutemov | b35f181 | 2014-01-21 15:49:07 -0800 | [diff] [blame] | 4121 | kmem_cache_free(page_ptl_cachep, page->ptl); |
Kirill A. Shutemov | 49076ec | 2013-11-14 14:31:51 -0800 | [diff] [blame] | 4122 | } |
| 4123 | #endif |