fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/hugetlb.h>
   3 #include <linux/huge_mm.h>
   4 #include <linux/mount.h>
   5 #include <linux/seq_file.h>
   6 #include <linux/highmem.h>
   7 #include <linux/ptrace.h>
   8 #include <linux/slab.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/mempolicy.h>
  11 #include <linux/rmap.h>
  12 #include <linux/swap.h>
  13 #include <linux/swapops.h>
  14
  15 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
  16 #include <linux/mm_inline.h>
  17 #endif
  18
  19 #include <asm/elf.h>
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22 #include "internal.h"
  23
  24 void task_mem(struct seq_file *m, struct mm_struct *mm)
  25 {
  26         unsigned long data, text, lib, swap;
  27         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  28
  29         /*
  30          * Note: to minimize their overhead, mm maintains hiwater_vm and
  31          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  32          * collector of these hiwater stats must therefore get total_vm
  33          * and rss too, which will usually be the higher.  Barriers? not
  34          * worth the effort, such snapshots can always be inconsistent.
  35          */
  36         hiwater_vm = total_vm = mm->total_vm;
  37         if (hiwater_vm < mm->hiwater_vm)
  38                 hiwater_vm = mm->hiwater_vm;
  39         hiwater_rss = total_rss = get_mm_rss(mm);
  40         if (hiwater_rss < mm->hiwater_rss)
  41                 hiwater_rss = mm->hiwater_rss;
  42
  43         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  44         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  45         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  46         swap = get_mm_counter(mm, MM_SWAPENTS);
  47         seq_printf(m,
  48                 "VmPeak:\t%8lu kB\n"
  49                 "VmSize:\t%8lu kB\n"
  50                 "VmLck:\t%8lu kB\n"
  51                 "VmPin:\t%8lu kB\n"
  52                 "VmHWM:\t%8lu kB\n"
  53                 "VmRSS:\t%8lu kB\n"
  54                 "VmData:\t%8lu kB\n"
  55                 "VmStk:\t%8lu kB\n"
  56                 "VmExe:\t%8lu kB\n"
  57                 "VmLib:\t%8lu kB\n"
  58                 "VmPTE:\t%8lu kB\n"
  59                 "VmSwap:\t%8lu kB\n",
  60                 hiwater_vm << (PAGE_SHIFT-10),
  61                 total_vm << (PAGE_SHIFT-10),
  62                 mm->locked_vm << (PAGE_SHIFT-10),
  63                 mm->pinned_vm << (PAGE_SHIFT-10),
  64                 hiwater_rss << (PAGE_SHIFT-10),
  65                 total_rss << (PAGE_SHIFT-10),
  66                 data << (PAGE_SHIFT-10),
  67                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  68                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
  69                 swap << (PAGE_SHIFT-10));
  70 }
  71
  72 unsigned long task_vsize(struct mm_struct *mm)
  73 {
  74         return PAGE_SIZE * mm->total_vm;
  75 }
  76
  77 unsigned long task_statm(struct mm_struct *mm,
  78                          unsigned long *shared, unsigned long *text,
  79                          unsigned long *data, unsigned long *resident)
  80 {
  81         *shared = get_mm_counter(mm, MM_FILEPAGES);
  82         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  83                                                                 >> PAGE_SHIFT;
  84         *data = mm->total_vm - mm->shared_vm;
  85         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  86         return mm->total_vm;
  87 }
  88
  89 static void pad_len_spaces(struct seq_file *m, int len)
  90 {
  91         len = 25 + sizeof(void*) * 6 - len;
  92         if (len < 1)
  93                 len = 1;
  94         seq_printf(m, "%*c", len, ' ');
  95 }
  96
  97 #ifdef CONFIG_NUMA
  98 /*
  99  * These functions are for numa_maps but called in generic **maps seq_file
 100  * ->start(), ->stop() ops.
 101  *
 102  * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
 103  * Each mempolicy object is controlled by reference counting. The problem here
 104  * is how to avoid accessing dead mempolicy object.
 105  *
 106  * Because we're holding mmap_sem while reading seq_file, it's safe to access
 107  * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
 108  *
 109  * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
 110  * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
 111  * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
 112  * gurantee the task never exits under us. But taking task_lock() around
 113  * get_vma_plicy() causes lock order problem.
 114  *
 115  * To access task->mempolicy without lock, we hold a reference count of an
 116  * object pointed by task->mempolicy and remember it. This will guarantee
 117  * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
 118  */
 119 static void hold_task_mempolicy(struct proc_maps_private *priv)
 120 {
 121         struct task_struct *task = priv->task;
 122
 123         task_lock(task);
 124         priv->task_mempolicy = task->mempolicy;
 125         mpol_get(priv->task_mempolicy);
 126         task_unlock(task);
 127 }
 128 static void release_task_mempolicy(struct proc_maps_private *priv)
 129 {
 130         mpol_put(priv->task_mempolicy);
 131 }
 132 #else
 133 static void hold_task_mempolicy(struct proc_maps_private *priv)
 134 {
 135 }
 136 static void release_task_mempolicy(struct proc_maps_private *priv)
 137 {
 138 }
 139 #endif
 140
 141 static void seq_print_vma_name(struct seq_file *m, struct vm_area_struct *vma)
 142 {
 143         const char __user *name = vma_get_anon_name(vma);
 144         struct mm_struct *mm = vma->vm_mm;
 145
 146         unsigned long page_start_vaddr;
 147         unsigned long page_offset;
 148         unsigned long num_pages;
 149         unsigned long max_len = NAME_MAX;
 150         int i;
 151
 152         page_start_vaddr = (unsigned long)name & PAGE_MASK;
 153         page_offset = (unsigned long)name - page_start_vaddr;
 154         num_pages = DIV_ROUND_UP(page_offset + max_len, PAGE_SIZE);
 155
 156         seq_puts(m, "[anon:");
 157
 158         for (i = 0; i < num_pages; i++) {
 159                 int len;
 160                 int write_len;
 161                 const char *kaddr;
 162                 long pages_pinned;
 163                 struct page *page;
 164
 165                 pages_pinned = get_user_pages(current, mm, page_start_vaddr,
 166                                 1, 0, 0, &page, NULL);
 167                 if (pages_pinned < 1) {
 168                         seq_puts(m, "<fault>]");
 169                         return;
 170                 }
 171
 172                 kaddr = (const char *)kmap(page);
 173                 len = min(max_len, PAGE_SIZE - page_offset);
 174                 write_len = strnlen(kaddr + page_offset, len);
 175                 seq_write(m, kaddr + page_offset, write_len);
 176                 kunmap(page);
 177                 put_page(page);
 178
 179                 /* if strnlen hit a null terminator then we're done */
 180                 if (write_len != len)
 181                         break;
 182
 183                 max_len -= len;
 184                 page_offset = 0;
 185                 page_start_vaddr += PAGE_SIZE;
 186         }
 187
 188         seq_putc(m, ']');
 189 }
 190
 191 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
 192 {
 193         if (vma && vma != priv->tail_vma) {
 194                 struct mm_struct *mm = vma->vm_mm;
 195                 release_task_mempolicy(priv);
 196                 up_read(&mm->mmap_sem);
 197                 mmput(mm);
 198         }
 199 }
 200
 201 static void *m_start(struct seq_file *m, loff_t *pos)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204         unsigned long last_addr = m->version;
 205         struct mm_struct *mm;
 206         struct vm_area_struct *vma, *tail_vma = NULL;
 207         loff_t l = *pos;
 208
 209         /* Clear the per syscall fields in priv */
 210         priv->task = NULL;
 211         priv->tail_vma = NULL;
 212
 213         /*
 214          * We remember last_addr rather than next_addr to hit with
 215          * mmap_cache most of the time. We have zero last_addr at
 216          * the beginning and also after lseek. We will have -1 last_addr
 217          * after the end of the vmas.
 218          */
 219
 220         if (last_addr == -1UL)
 221                 return NULL;
 222
 223         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
 224         if (!priv->task)
 225                 return ERR_PTR(-ESRCH);
 226
 227         mm = mm_access(priv->task, PTRACE_MODE_READ);
 228         if (!mm || IS_ERR(mm))
 229                 return mm;
 230         down_read(&mm->mmap_sem);
 231
 232         tail_vma = get_gate_vma(priv->task->mm);
 233         priv->tail_vma = tail_vma;
 234         hold_task_mempolicy(priv);
 235         /* Start with last addr hint */
 236         vma = find_vma(mm, last_addr);
 237         if (last_addr && vma) {
 238                 vma = vma->vm_next;
 239                 goto out;
 240         }
 241
 242         /*
 243          * Check the vma index is within the range and do
 244          * sequential scan until m_index.
 245          */
 246         vma = NULL;
 247         if ((unsigned long)l < mm->map_count) {
 248                 vma = mm->mmap;
 249                 while (l-- && vma)
 250                         vma = vma->vm_next;
 251                 goto out;
 252         }
 253
 254         if (l != mm->map_count)
 255                 tail_vma = NULL; /* After gate vma */
 256
 257 out:
 258         if (vma)
 259                 return vma;
 260
 261         release_task_mempolicy(priv);
 262         /* End of vmas has been reached */
 263         m->version = (tail_vma != NULL)? 0: -1UL;
 264         up_read(&mm->mmap_sem);
 265         mmput(mm);
 266         return tail_vma;
 267 }
 268
 269 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 270 {
 271         struct proc_maps_private *priv = m->private;
 272         struct vm_area_struct *vma = v;
 273         struct vm_area_struct *tail_vma = priv->tail_vma;
 274
 275         (*pos)++;
 276         if (vma && (vma != tail_vma) && vma->vm_next)
 277                 return vma->vm_next;
 278         vma_stop(priv, vma);
 279         return (vma != tail_vma)? tail_vma: NULL;
 280 }
 281
 282 static void m_stop(struct seq_file *m, void *v)
 283 {
 284         struct proc_maps_private *priv = m->private;
 285         struct vm_area_struct *vma = v;
 286
 287         if (!IS_ERR(vma))
 288                 vma_stop(priv, vma);
 289         if (priv->task)
 290                 put_task_struct(priv->task);
 291 }
 292
 293 static int do_maps_open(struct inode *inode, struct file *file,
 294                         const struct seq_operations *ops)
 295 {
 296         struct proc_maps_private *priv;
 297         int ret = -ENOMEM;
 298         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 299         if (priv) {
 300                 priv->pid = proc_pid(inode);
 301                 ret = seq_open(file, ops);
 302                 if (!ret) {
 303                         struct seq_file *m = file->private_data;
 304                         m->private = priv;
 305 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 306                         priv->rss = 0;
 307                         priv->pss = 0;
 308                         priv->uss = 0;
 309                         priv->filecache_pss = 0;
 310                         priv->anonymous_pss = 0;
 311                         priv->filecache_uss = 0;
 312                         priv->anonymous_uss = 0;
 313                         priv->swap = 0;
 314                         priv->pswap = 0;
 315                         priv->anonymous_uswap = 0;
 316                         priv->is_last = false;
 317 #endif
 318                 } else {
 319                         kfree(priv);
 320                 }
 321         }
 322         return ret;
 323 }
 324
 325 static void
 326 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 327 {
 328         struct mm_struct *mm = vma->vm_mm;
 329         struct file *file = vma->vm_file;
 330         struct proc_maps_private *priv = m->private;
 331         struct task_struct *task = priv->task;
 332         vm_flags_t flags = vma->vm_flags;
 333         unsigned long ino = 0;
 334         unsigned long long pgoff = 0;
 335         unsigned long start, end;
 336         dev_t dev = 0;
 337         int len;
 338         const char *name = NULL;
 339
 340         if (file) {
 341                 struct inode *inode = file_inode(vma->vm_file);
 342                 dev = inode->i_sb->s_dev;
 343                 ino = inode->i_ino;
 344                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 345         }
 346
 347         /* We don't show the stack guard page in /proc/maps */
 348         start = vma->vm_start;
 349         if (stack_guard_page_start(vma, start))
 350                 start += PAGE_SIZE;
 351         end = vma->vm_end;
 352         if (stack_guard_page_end(vma, end))
 353                 end -= PAGE_SIZE;
 354
 355         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
 356                         start,
 357                         end,
 358                         flags & VM_READ ? 'r' : '-',
 359                         flags & VM_WRITE ? 'w' : '-',
 360                         flags & VM_EXEC ? 'x' : '-',
 361                         flags & VM_MAYSHARE ? 's' : 'p',
 362                         pgoff,
 363                         MAJOR(dev), MINOR(dev), ino, &len);
 364
 365         /*
 366          * Print the dentry name for named mappings, and a
 367          * special [heap] marker for the heap:
 368          */
 369         if (file) {
 370                 pad_len_spaces(m, len);
 371                 seq_path(m, &file->f_path, "\n");
 372                 goto done;
 373         }
 374
 375         name = arch_vma_name(vma);
 376         if (!name) {
 377                 pid_t tid;
 378
 379                 if (!mm) {
 380                         name = "[vdso]";
 381                         goto done;
 382                 }
 383
 384                 if (vma->vm_start <= mm->brk &&
 385                     vma->vm_end >= mm->start_brk) {
 386                         name = "[heap]";
 387                         goto done;
 388                 }
 389
 390                 tid = vm_is_stack(task, vma, is_pid);
 391
 392                 if (tid != 0) {
 393                         /*
 394                          * Thread stack in /proc/PID/task/TID/maps or
 395                          * the main process stack.
 396                          */
 397                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
 398                             vma->vm_end >= mm->start_stack)) {
 399                                 name = "[stack]";
 400                         } else {
 401                                 /* Thread stack in /proc/PID/maps */
 402                                 pad_len_spaces(m, len);
 403                                 seq_printf(m, "[stack:%d]", tid);
 404                         }
 405                         goto done;
 406                 }
 407
 408                 if (vma_get_anon_name(vma)) {
 409                         pad_len_spaces(m, len);
 410                         seq_print_vma_name(m, vma);
 411                 }
 412         }
 413
 414 done:
 415         if (name) {
 416                 pad_len_spaces(m, len);
 417                 seq_puts(m, name);
 418         }
 419         seq_putc(m, '\n');
 420 }
 421
 422 static int show_map(struct seq_file *m, void *v, int is_pid)
 423 {
 424         struct vm_area_struct *vma = v;
 425         struct proc_maps_private *priv = m->private;
 426         struct task_struct *task = priv->task;
 427
 428         show_map_vma(m, vma, is_pid);
 429
 430         if (m->count < m->size)  /* vma is copied successfully */
 431                 m->version = (vma != get_gate_vma(task->mm))
 432                         ? vma->vm_start : 0;
 433         return 0;
 434 }
 435
 436 static int show_pid_map(struct seq_file *m, void *v)
 437 {
 438         return show_map(m, v, 1);
 439 }
 440
 441 static int show_tid_map(struct seq_file *m, void *v)
 442 {
 443         return show_map(m, v, 0);
 444 }
 445
 446 static const struct seq_operations proc_pid_maps_op = {
 447         .start  = m_start,
 448         .next   = m_next,
 449         .stop   = m_stop,
 450         .show   = show_pid_map
 451 };
 452
 453 static const struct seq_operations proc_tid_maps_op = {
 454         .start  = m_start,
 455         .next   = m_next,
 456         .stop   = m_stop,
 457         .show   = show_tid_map
 458 };
 459
 460 static int pid_maps_open(struct inode *inode, struct file *file)
 461 {
 462         return do_maps_open(inode, file, &proc_pid_maps_op);
 463 }
 464
 465 static int tid_maps_open(struct inode *inode, struct file *file)
 466 {
 467         return do_maps_open(inode, file, &proc_tid_maps_op);
 468 }
 469
 470 const struct file_operations proc_pid_maps_operations = {
 471         .open           = pid_maps_open,
 472         .read           = seq_read,
 473         .llseek         = seq_lseek,
 474         .release        = seq_release_private,
 475 };
 476
 477 const struct file_operations proc_tid_maps_operations = {
 478         .open           = tid_maps_open,
 479         .read           = seq_read,
 480         .llseek         = seq_lseek,
 481         .release        = seq_release_private,
 482 };
 483
 484 /*
 485  * Proportional Set Size(PSS): my share of RSS.
 486  *
 487  * PSS of a process is the count of pages it has in memory, where each
 488  * page is divided by the number of processes sharing it.  So if a
 489  * process has 1000 pages all to itself, and 1000 shared with one other
 490  * process, its PSS will be 1500.
 491  *
 492  * To keep (accumulated) division errors low, we adopt a 64bit
 493  * fixed-point pss counter to minimize division errors. So (pss >>
 494  * PSS_SHIFT) would be the real byte count.
 495  *
 496  * A shift of 12 before division means (assuming 4K page size):
 497  *      - 1M 3-user-pages add up to 8KB errors;
 498  *      - supports mapcount up to 2^24, or 16M;
 499  *      - supports PSS up to 2^52 bytes, or 4PB.
 500  */
 501 #define PSS_SHIFT 12
 502
 503 #ifdef CONFIG_PROC_PAGE_MONITOR
 504 struct mem_size_stats {
 505         struct vm_area_struct *vma;
 506         unsigned long resident;
 507         unsigned long shared_clean;
 508         unsigned long shared_dirty;
 509         unsigned long private_clean;
 510         unsigned long private_dirty;
 511         unsigned long referenced;
 512         unsigned long anonymous;
 513         unsigned long anonymous_thp;
 514         unsigned long swap;
 515         unsigned long nonlinear;
 516 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 517         unsigned long filecache;
 518         unsigned long shared_filecache;
 519         unsigned long private_filecache;
 520         unsigned long shared_anonymous;
 521         unsigned long private_anonymous;
 522         u64 pss_filecache;
 523         u64 pss_anonymous;
 524         u64 uswap;
 525 #endif
 526         u64 pss;
 527         u64 pswap;
 528
 529 };
 530
 531 #ifdef CONFIG_SWAP
 532 extern struct swap_info_struct *swap_info_get(swp_entry_t entry);
 533
 534 static inline unsigned char swap_count(unsigned char ent)
 535 {
 536         return ent & ~SWAP_HAS_CACHE;   /* may include SWAP_HAS_CONT flag */
 537 }
 538 #endif
 539
 540 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
 541                 unsigned long ptent_size, struct mm_walk *walk)
 542 {
 543         struct mem_size_stats *mss = walk->private;
 544         struct vm_area_struct *vma = mss->vma;
 545         pgoff_t pgoff = linear_page_index(vma, addr);
 546         struct page *page = NULL;
 547         int mapcount;
 548
 549         if (pte_present(ptent)) {
 550                 page = vm_normal_page(vma, addr, ptent);
 551         } else if (is_swap_pte(ptent)) {
 552                 swp_entry_t swpent = pte_to_swp_entry(ptent);
 553                 struct swap_info_struct *p;
 554
 555                 if (!non_swap_entry(swpent)) {
 556                         mss->swap += ptent_size;
 557 #ifdef CONFIG_SWAP
 558                         p = swap_info_get(swpent);
 559                         if (p) {
 560                                 int swapcount = swap_count(
 561                                         p->swap_map[swp_offset(swpent)]);
 562                                 if (!swapcount)
 563                                         swapcount = 1;
 564                                 mss->pswap +=
 565                                         (ptent_size << PSS_SHIFT) / swapcount;
 566 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 567                                 if (1 == swapcount)
 568                                         mss->uswap += ptent_size << PSS_SHIFT;
 569 #endif
 570                                 spin_unlock(&p->lock);
 571                         }
 572 #endif
 573                 }
 574                 else if (is_migration_entry(swpent))
 575                         page = migration_entry_to_page(swpent);
 576         } else if (pte_file(ptent)) {
 577                 if (pte_to_pgoff(ptent) != pgoff)
 578                         mss->nonlinear += ptent_size;
 579         }
 580
 581         if (!page)
 582                 return;
 583
 584         if (PageAnon(page))
 585                 mss->anonymous += ptent_size;
 586 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 587         else if(page_is_file_cache(page))
 588                 mss->filecache += ptent_size;
 589 #endif
 590
 591         if (page->index != pgoff)
 592                 mss->nonlinear += ptent_size;
 593
 594         mss->resident += ptent_size;
 595         /* Accumulate the size in pages that have been accessed. */
 596         if (pte_young(ptent) || PageReferenced(page))
 597                 mss->referenced += ptent_size;
 598         mapcount = page_mapcount(page);
 599         if (mapcount >= 2) {
 600                 if (pte_dirty(ptent) || PageDirty(page))
 601                         mss->shared_dirty += ptent_size;
 602                 else
 603                         mss->shared_clean += ptent_size;
 604                 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
 605 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 606                 if (PageAnon(page)) {
 607                         mss->shared_anonymous += ptent_size;
 608                         mss->pss_anonymous += (ptent_size << PSS_SHIFT) / mapcount;
 609                 }
 610                 else if(page_is_file_cache(page)) {
 611                         mss->shared_filecache += ptent_size;
 612                         mss->pss_filecache += (ptent_size << PSS_SHIFT) / mapcount;
 613                 }
 614 #endif
 615         } else {
 616                 if (pte_dirty(ptent) || PageDirty(page))
 617                         mss->private_dirty += ptent_size;
 618                 else
 619                         mss->private_clean += ptent_size;
 620                 mss->pss += (ptent_size << PSS_SHIFT);
 621 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 622                 if (PageAnon(page)) {
 623                         mss->private_anonymous += ptent_size;
 624                         mss->pss_anonymous += (ptent_size << PSS_SHIFT);
 625                 }
 626                 else if(page_is_file_cache(page)) {
 627                         mss->private_filecache += ptent_size;
 628                         mss->pss_filecache += (ptent_size << PSS_SHIFT);
 629                 }
 630 #endif
 631         }
 632 }
 633
 634 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 635                            struct mm_walk *walk)
 636 {
 637         struct mem_size_stats *mss = walk->private;
 638         struct vm_area_struct *vma = mss->vma;
 639         pte_t *pte;
 640         spinlock_t *ptl;
 641
 642         if (pmd_trans_huge_lock(pmd, vma) == 1) {
 643                 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
 644                 spin_unlock(&walk->mm->page_table_lock);
 645                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 646                 return 0;
 647         }
 648
 649         if (pmd_trans_unstable(pmd))
 650                 return 0;
 651         /*
 652          * The mmap_sem held all the way back in m_start() is what
 653          * keeps khugepaged out of here and from collapsing things
 654          * in here.
 655          */
 656         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 657         for (; addr != end; pte++, addr += PAGE_SIZE)
 658                 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
 659         pte_unmap_unlock(pte - 1, ptl);
 660         cond_resched();
 661         return 0;
 662 }
 663
 664 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 665 {
 666         /*
 667          * Don't forget to update Documentation/ on changes.
 668          */
 669         static const char mnemonics[BITS_PER_LONG][2] = {
 670                 /*
 671                  * In case if we meet a flag we don't know about.
 672                  */
 673                 [0 ... (BITS_PER_LONG-1)] = "??",
 674
 675                 [ilog2(VM_READ)]        = "rd",
 676                 [ilog2(VM_WRITE)]       = "wr",
 677                 [ilog2(VM_EXEC)]        = "ex",
 678                 [ilog2(VM_SHARED)]      = "sh",
 679                 [ilog2(VM_MAYREAD)]     = "mr",
 680                 [ilog2(VM_MAYWRITE)]    = "mw",
 681                 [ilog2(VM_MAYEXEC)]     = "me",
 682                 [ilog2(VM_MAYSHARE)]    = "ms",
 683                 [ilog2(VM_GROWSDOWN)]   = "gd",
 684                 [ilog2(VM_PFNMAP)]      = "pf",
 685                 [ilog2(VM_DENYWRITE)]   = "dw",
 686                 [ilog2(VM_LOCKED)]      = "lo",
 687                 [ilog2(VM_IO)]          = "io",
 688                 [ilog2(VM_SEQ_READ)]    = "sr",
 689                 [ilog2(VM_RAND_READ)]   = "rr",
 690                 [ilog2(VM_DONTCOPY)]    = "dc",
 691                 [ilog2(VM_DONTEXPAND)]  = "de",
 692                 [ilog2(VM_ACCOUNT)]     = "ac",
 693                 [ilog2(VM_NORESERVE)]   = "nr",
 694                 [ilog2(VM_HUGETLB)]     = "ht",
 695                 [ilog2(VM_NONLINEAR)]   = "nl",
 696                 [ilog2(VM_ARCH_1)]      = "ar",
 697                 [ilog2(VM_DONTDUMP)]    = "dd",
 698                 [ilog2(VM_MIXEDMAP)]    = "mm",
 699                 [ilog2(VM_HUGEPAGE)]    = "hg",
 700                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 701                 [ilog2(VM_MERGEABLE)]   = "mg",
 702         };
 703         size_t i;
 704
 705         seq_puts(m, "VmFlags: ");
 706         for (i = 0; i < BITS_PER_LONG; i++) {
 707                 if (vma->vm_flags & (1UL << i)) {
 708                         seq_printf(m, "%c%c ",
 709                                    mnemonics[i][0], mnemonics[i][1]);
 710                 }
 711         }
 712         seq_putc(m, '\n');
 713 }
 714
 715 static int show_smap(struct seq_file *m, void *v, int is_pid)
 716 {
 717         struct proc_maps_private *priv = m->private;
 718         struct task_struct *task = priv->task;
 719         struct vm_area_struct *vma = v;
 720         struct mem_size_stats mss;
 721         struct mm_walk smaps_walk = {
 722                 .pmd_entry = smaps_pte_range,
 723                 .mm = vma->vm_mm,
 724                 .private = &mss,
 725         };
 726
 727         memset(&mss, 0, sizeof mss);
 728         mss.vma = vma;
 729         /* mmap_sem is held in m_start */
 730         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 731                 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
 732
 733         show_map_vma(m, vma, is_pid);
 734 #ifdef CONFIG_SPRD_DEBUG_MORE_SMAPS_INFO
 735         seq_printf(m,
 736                    "Size:             %8lu kB\n"
 737                    "Rss:              %8lu kB\n"
 738                    "Pss:              %8lu kB\n"
 739                    "Uss:              %8lu kB\n"
 740                    "Shared_Clean:     %8lu kB\n"
 741                    "Shared_Dirty:     %8lu kB\n"
 742                    "Private_Clean:    %8lu kB\n"
 743                    "Private_Dirty:    %8lu kB\n"
 744                    "Referenced:       %8lu kB\n"
 745                    "Filecache:        %8lu kB\n"
 746                    "Pss_Filecache:    %8lu kB\n"
 747                    "Shared_Filecache: %8lu kB\n"
 748                    "Private_Filecache:%8lu kB\n"
 749                    "Anonymous:        %8lu kB\n"
 750                    "Pss_Anonymous:    %8lu kB\n"
 751                    "Shared_Anonymous: %8lu kB\n"
 752                    "Private_Anonymous:%8lu kB\n"
 753                    "AnonHugePages:    %8lu kB\n"
 754                    "Swap:             %8lu kB\n"
 755                    "PSwap:            %8lu kB\n"
 756                    "USwap:            %8lu kB\n"
 757                    "KernelPageSize:   %8lu kB\n"
 758                    "MMUPageSize:      %8lu kB\n"
 759                    "Locked:           %8lu kB\n",
 760                    (vma->vm_end - vma->vm_start) >> 10,
 761                    mss.resident >> 10,
 762                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 763                    (mss.private_clean + mss.private_dirty) >> 10,
 764                    mss.shared_clean  >> 10,
 765                    mss.shared_dirty  >> 10,
 766                    mss.private_clean >> 10,
 767                    mss.private_dirty >> 10,
 768                    mss.referenced >> 10,
 769                    mss.filecache >> 10,
 770                    (unsigned long)(mss.pss_filecache >> (10 + PSS_SHIFT)),
 771                    mss.shared_filecache >> 10,
 772                    mss.private_filecache >> 10,
 773                    mss.anonymous >> 10,
 774                    (unsigned long)(mss.pss_anonymous >> (10 + PSS_SHIFT)),
 775                    mss.shared_anonymous >> 10,
 776                    mss.private_anonymous >> 10,
 777                    mss.anonymous_thp >> 10,
 778                    mss.swap >> 10,
 779                    (unsigned long)(mss.pswap >> (10 + PSS_SHIFT)),
 780                    (unsigned long)(mss.uswap >> (10 + PSS_SHIFT)),
 781                    vma_kernel_pagesize(vma) >> 10,
 782                    vma_mmu_pagesize(vma) >> 10,
 783                    (vma->vm_flags & VM_LOCKED) ?
 784                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 785
 786         if (vma->vm_flags & VM_NONLINEAR)
 787                 seq_printf(m, "Nonlinear:      %8lu kB\n",
 788                                 mss.nonlinear >> 10);
 789
 790         show_smap_vma_flags(m, vma);
 791
 792         if (vma_get_anon_name(vma)) {
 793                 seq_puts(m, "Name:           ");
 794                 seq_print_vma_name(m, vma);
 795                 seq_putc(m, '\n');
 796         }
 797
 798         if (m->count < m->size) {  /* vma is copied successfully */
 799                 priv->rss += mss.resident;
 800                 priv->pss += mss.pss;
 801                 priv->uss += mss.private_clean + mss.private_dirty;
 802                 priv->filecache_rss += mss.filecache;
 803                 priv->anonymous_rss += mss.anonymous;
 804                 priv->filecache_pss += mss.pss_filecache;
 805                 priv->anonymous_pss += mss.pss_anonymous;
 806                 priv->filecache_uss += mss.private_filecache;
 807                 priv->anonymous_uss += mss.private_anonymous;
 808                 priv->swap += mss.swap;
 809                 priv->pswap += mss.pswap;
 810                 priv->anonymous_uswap += mss.uswap;
 811
 812                 if ((vma == priv->tail_vma) && (priv->is_last == false)) {
 813                         /* This seq_printf will increases m->count*/
 814                         seq_printf(m,
 815                                 "Rss_All:                %8lu kB\n"
 816                                 "Pss_All:                %8lu kB\n"
 817                                 "Uss_All:                %8lu kB\n"
 818                                 "Rss_Filecache_All:      %8lu kB\n"
 819                                 "Rss_Anonymous_All:      %8lu kB\n"
 820                                 "Pss_Filecache_All:      %8lu kB\n"
 821                                 "Pss_Anonymous_All:      %8lu kB\n"
 822                                 "Uss_Filecache_All:      %8lu kB\n"
 823                                 "Uss_Anonymous_All:      %8lu kB\n"
 824                                 "Swap_All:               %8lu kB\n"
 825                                 "Pswap_All:              %8lu kB\n"
 826                                 "USwap_All:              %8lu kB\n",
 827                                 (unsigned long)(priv->rss >> 10),
 828                                 (unsigned long)(priv->pss >> (10 + PSS_SHIFT)),
 829                                 (unsigned long)(priv->uss >> 10),
 830                                 (unsigned long)(priv->filecache_rss >> 10),
 831                                 (unsigned long)(priv->anonymous_rss >> 10),
 832                                 (unsigned long)(priv->filecache_pss
 833                                         >> (10 + PSS_SHIFT)),
 834                                 (unsigned long)(priv->anonymous_pss
 835                                         >> (10 + PSS_SHIFT)),
 836                                 (unsigned long)(priv->filecache_uss >> 10),
 837                                 (unsigned long)(priv->anonymous_uss >> 10),
 838                                 (unsigned long)(priv->swap >> 10),
 839                                 (unsigned long)(priv->pswap
 840                                         >> (10 + PSS_SHIFT)),
 841                                 (unsigned long)(priv->anonymous_uswap
 842                                         >> (10 + PSS_SHIFT)));
 843
 844                         /* vma is still copied successfully */
 845                         if (m->count < m->size) {
 846                                 priv->is_last = true;
 847                         } else {
 848                                 priv->rss -= mss.resident;
 849                                 priv->pss -= mss.pss;
 850                                 priv->uss -= mss.private_clean
 851                                         + mss.private_dirty;
 852                                 priv->filecache_rss -= mss.filecache;
 853                                 priv->anonymous_rss -= mss.anonymous;
 854                                 priv->filecache_pss -= mss.pss_filecache;
 855                                 priv->anonymous_pss -= mss.pss_anonymous;
 856                                 priv->filecache_uss -= mss.private_filecache;
 857                                 priv->anonymous_uss -= mss.private_anonymous;
 858                                 priv->swap -= mss.swap;
 859                                 priv->pswap -= mss.pswap;
 860                                 priv->anonymous_uswap -= mss.uswap;
 861                         }
 862                 }
 863         }
 864
 865         if (m->count < m->size) {  /* vma is copied successfully */
 866                 m->version = (vma != get_gate_vma(task->mm))
 867                         ? vma->vm_start : 0;
 868         }
 869 #else
 870         seq_printf(m,
 871                    "Size:           %8lu kB\n"
 872                    "Rss:            %8lu kB\n"
 873                    "Pss:            %8lu kB\n"
 874                    "Shared_Clean:   %8lu kB\n"
 875                    "Shared_Dirty:   %8lu kB\n"
 876                    "Private_Clean:  %8lu kB\n"
 877                    "Private_Dirty:  %8lu kB\n"
 878                    "Referenced:     %8lu kB\n"
 879                    "Anonymous:      %8lu kB\n"
 880                    "AnonHugePages:  %8lu kB\n"
 881                    "Swap:           %8lu kB\n"
 882                    "PSwap:          %8lu kB\n"
 883                    "KernelPageSize: %8lu kB\n"
 884                    "MMUPageSize:    %8lu kB\n"
 885                    "Locked:         %8lu kB\n",
 886                    (vma->vm_end - vma->vm_start) >> 10,
 887                    mss.resident >> 10,
 888                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 889                    mss.shared_clean  >> 10,
 890                    mss.shared_dirty  >> 10,
 891                    mss.private_clean >> 10,
 892                    mss.private_dirty >> 10,
 893                    mss.referenced >> 10,
 894                    mss.anonymous >> 10,
 895                    mss.anonymous_thp >> 10,
 896                    mss.swap >> 10,
 897                    (unsigned long)(mss.pswap >> (10 + PSS_SHIFT)),
 898                    vma_kernel_pagesize(vma) >> 10,
 899                    vma_mmu_pagesize(vma) >> 10,
 900                    (vma->vm_flags & VM_LOCKED) ?
 901                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 902
 903         if (vma->vm_flags & VM_NONLINEAR)
 904                 seq_printf(m, "Nonlinear:      %8lu kB\n",
 905                                 mss.nonlinear >> 10);
 906
 907         show_smap_vma_flags(m, vma);
 908
 909         if (vma_get_anon_name(vma)) {
 910                 seq_puts(m, "Name:           ");
 911                 seq_print_vma_name(m, vma);
 912                 seq_putc(m, '\n');
 913         }
 914
 915         if (m->count < m->size)  /* vma is copied successfully */
 916                 m->version = (vma != get_gate_vma(task->mm))
 917                         ? vma->vm_start : 0;
 918 #endif
 919         return 0;
 920 }
 921
 922 static int show_pid_smap(struct seq_file *m, void *v)
 923 {
 924         return show_smap(m, v, 1);
 925 }
 926
 927 static int show_tid_smap(struct seq_file *m, void *v)
 928 {
 929         return show_smap(m, v, 0);
 930 }
 931
 932 static const struct seq_operations proc_pid_smaps_op = {
 933         .start  = m_start,
 934         .next   = m_next,
 935         .stop   = m_stop,
 936         .show   = show_pid_smap
 937 };
 938
 939 static const struct seq_operations proc_tid_smaps_op = {
 940         .start  = m_start,
 941         .next   = m_next,
 942         .stop   = m_stop,
 943         .show   = show_tid_smap
 944 };
 945
 946 static int pid_smaps_open(struct inode *inode, struct file *file)
 947 {
 948         return do_maps_open(inode, file, &proc_pid_smaps_op);
 949 }
 950
 951 static int tid_smaps_open(struct inode *inode, struct file *file)
 952 {
 953         return do_maps_open(inode, file, &proc_tid_smaps_op);
 954 }
 955
 956 const struct file_operations proc_pid_smaps_operations = {
 957         .open           = pid_smaps_open,
 958         .read           = seq_read,
 959         .llseek         = seq_lseek,
 960         .release        = seq_release_private,
 961 };
 962
 963 const struct file_operations proc_tid_smaps_operations = {
 964         .open           = tid_smaps_open,
 965         .read           = seq_read,
 966         .llseek         = seq_lseek,
 967         .release        = seq_release_private,
 968 };
 969
 970 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 971                                 unsigned long end, struct mm_walk *walk)
 972 {
 973         struct vm_area_struct *vma = walk->private;
 974         pte_t *pte, ptent;
 975         spinlock_t *ptl;
 976         struct page *page;
 977
 978         split_huge_page_pmd(vma, addr, pmd);
 979         if (pmd_trans_unstable(pmd))
 980                 return 0;
 981
 982         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 983         for (; addr != end; pte++, addr += PAGE_SIZE) {
 984                 ptent = *pte;
 985                 if (!pte_present(ptent))
 986                         continue;
 987
 988                 page = vm_normal_page(vma, addr, ptent);
 989                 if (!page)
 990                         continue;
 991
 992                 /* Clear accessed and referenced bits. */
 993                 ptep_test_and_clear_young(vma, addr, pte);
 994                 ClearPageReferenced(page);
 995         }
 996         pte_unmap_unlock(pte - 1, ptl);
 997         cond_resched();
 998         return 0;
 999 }
1000
1001 #define CLEAR_REFS_ALL 1
1002 #define CLEAR_REFS_ANON 2
1003 #define CLEAR_REFS_MAPPED 3
1004
1005 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1006                                 size_t count, loff_t *ppos)
1007 {
1008         struct task_struct *task;
1009         char buffer[PROC_NUMBUF];
1010         struct mm_struct *mm;
1011         struct vm_area_struct *vma;
1012         int type;
1013         int rv;
1014
1015         memset(buffer, 0, sizeof(buffer));
1016         if (count > sizeof(buffer) - 1)
1017                 count = sizeof(buffer) - 1;
1018         if (copy_from_user(buffer, buf, count))
1019                 return -EFAULT;
1020         rv = kstrtoint(strstrip(buffer), 10, &type);
1021         if (rv < 0)
1022                 return rv;
1023         if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
1024                 return -EINVAL;
1025         task = get_proc_task(file_inode(file));
1026         if (!task)
1027                 return -ESRCH;
1028         mm = get_task_mm(task);
1029         if (mm) {
1030                 struct mm_walk clear_refs_walk = {
1031                         .pmd_entry = clear_refs_pte_range,
1032                         .mm = mm,
1033                 };
1034                 down_read(&mm->mmap_sem);
1035                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1036                         clear_refs_walk.private = vma;
1037                         if (is_vm_hugetlb_page(vma))
1038                                 continue;
1039                         /*
1040                          * Writing 1 to /proc/pid/clear_refs affects all pages.
1041                          *
1042                          * Writing 2 to /proc/pid/clear_refs only affects
1043                          * Anonymous pages.
1044                          *
1045                          * Writing 3 to /proc/pid/clear_refs only affects file
1046                          * mapped pages.
1047                          */
1048                         if (type == CLEAR_REFS_ANON && vma->vm_file)
1049                                 continue;
1050                         if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
1051                                 continue;
1052                         walk_page_range(vma->vm_start, vma->vm_end,
1053                                         &clear_refs_walk);
1054                 }
1055                 flush_tlb_mm(mm);
1056                 up_read(&mm->mmap_sem);
1057                 mmput(mm);
1058         }
1059         put_task_struct(task);
1060
1061         return count;
1062 }
1063
1064 const struct file_operations proc_clear_refs_operations = {
1065         .write          = clear_refs_write,
1066         .llseek         = noop_llseek,
1067 };
1068
1069 typedef struct {
1070         u64 pme;
1071 } pagemap_entry_t;
1072
1073 struct pagemapread {
1074         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1075         pagemap_entry_t *buffer;
1076 };
1077
1078 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1079 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1080
1081 #define PM_ENTRY_BYTES      sizeof(pagemap_entry_t)
1082 #define PM_STATUS_BITS      3
1083 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
1084 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
1085 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
1086 #define PM_PSHIFT_BITS      6
1087 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
1088 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
1089 #define PM_PSHIFT(x)        (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
1090 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
1091 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
1092
1093 #define PM_PRESENT          PM_STATUS(4LL)
1094 #define PM_SWAP             PM_STATUS(2LL)
1095 #define PM_FILE             PM_STATUS(1LL)
1096 #define PM_NOT_PRESENT      PM_PSHIFT(PAGE_SHIFT)
1097 #define PM_END_OF_BUFFER    1
1098
1099 static inline pagemap_entry_t make_pme(u64 val)
1100 {
1101         return (pagemap_entry_t) { .pme = val };
1102 }
1103
1104 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1105                           struct pagemapread *pm)
1106 {
1107         pm->buffer[pm->pos++] = *pme;
1108         if (pm->pos >= pm->len)
1109                 return PM_END_OF_BUFFER;
1110         return 0;
1111 }
1112
1113 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1114                                 struct mm_walk *walk)
1115 {
1116         struct pagemapread *pm = walk->private;
1117         unsigned long addr;
1118         int err = 0;
1119         pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
1120
1121         for (addr = start; addr < end; addr += PAGE_SIZE) {
1122                 err = add_to_pagemap(addr, &pme, pm);
1123                 if (err)
1124                         break;
1125         }
1126         return err;
1127 }
1128
1129 static void pte_to_pagemap_entry(pagemap_entry_t *pme,
1130                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1131 {
1132         u64 frame, flags;
1133         struct page *page = NULL;
1134
1135         if (pte_present(pte)) {
1136                 frame = pte_pfn(pte);
1137                 flags = PM_PRESENT;
1138                 page = vm_normal_page(vma, addr, pte);
1139         } else if (is_swap_pte(pte)) {
1140                 swp_entry_t entry = pte_to_swp_entry(pte);
1141
1142                 frame = swp_type(entry) |
1143                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1144                 flags = PM_SWAP;
1145                 if (is_migration_entry(entry))
1146                         page = migration_entry_to_page(entry);
1147         } else {
1148                 *pme = make_pme(PM_NOT_PRESENT);
1149                 return;
1150         }
1151
1152         if (page && !PageAnon(page))
1153                 flags |= PM_FILE;
1154
1155         *pme = make_pme(PM_PFRAME(frame) | PM_PSHIFT(PAGE_SHIFT) | flags);
1156 }
1157
1158 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1159 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
1160                                         pmd_t pmd, int offset)
1161 {
1162         /*
1163          * Currently pmd for thp is always present because thp can not be
1164          * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1165          * This if-check is just to prepare for future implementation.
1166          */
1167         if (pmd_present(pmd))
1168                 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1169                                 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
1170         else
1171                 *pme = make_pme(PM_NOT_PRESENT);
1172 }
1173 #else
1174 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
1175                                                 pmd_t pmd, int offset)
1176 {
1177 }
1178 #endif
1179
1180 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1181                              struct mm_walk *walk)
1182 {
1183         struct vm_area_struct *vma;
1184         struct pagemapread *pm = walk->private;
1185         pte_t *pte;
1186         int err = 0;
1187         pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
1188
1189         /* find the first VMA at or above 'addr' */
1190         vma = find_vma(walk->mm, addr);
1191         if (vma && pmd_trans_huge_lock(pmd, vma) == 1) {
1192                 for (; addr != end; addr += PAGE_SIZE) {
1193                         unsigned long offset;
1194
1195                         offset = (addr & ~PAGEMAP_WALK_MASK) >>
1196                                         PAGE_SHIFT;
1197                         thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
1198                         err = add_to_pagemap(addr, &pme, pm);
1199                         if (err)
1200                                 break;
1201                 }
1202                 spin_unlock(&walk->mm->page_table_lock);
1203                 return err;
1204         }
1205
1206         if (pmd_trans_unstable(pmd))
1207                 return 0;
1208         for (; addr != end; addr += PAGE_SIZE) {
1209
1210                 /* check to see if we've left 'vma' behind
1211                  * and need a new, higher one */
1212                 if (vma && (addr >= vma->vm_end)) {
1213                         vma = find_vma(walk->mm, addr);
1214                         pme = make_pme(PM_NOT_PRESENT);
1215                 }
1216
1217                 /* check that 'vma' actually covers this address,
1218                  * and that it isn't a huge page vma */
1219                 if (vma && (vma->vm_start <= addr) &&
1220                     !is_vm_hugetlb_page(vma)) {
1221                         pte = pte_offset_map(pmd, addr);
1222                         pte_to_pagemap_entry(&pme, vma, addr, *pte);
1223                         /* unmap before userspace copy */
1224                         pte_unmap(pte);
1225                 }
1226                 err = add_to_pagemap(addr, &pme, pm);
1227                 if (err)
1228                         return err;
1229         }
1230
1231         cond_resched();
1232
1233         return err;
1234 }
1235
1236 #ifdef CONFIG_HUGETLB_PAGE
1237 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
1238                                         pte_t pte, int offset)
1239 {
1240         if (pte_present(pte))
1241                 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
1242                                 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
1243         else
1244                 *pme = make_pme(PM_NOT_PRESENT);
1245 }
1246
1247 /* This function walks within one hugetlb entry in the single call */
1248 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1249                                  unsigned long addr, unsigned long end,
1250                                  struct mm_walk *walk)
1251 {
1252         struct pagemapread *pm = walk->private;
1253         int err = 0;
1254         pagemap_entry_t pme;
1255
1256         for (; addr != end; addr += PAGE_SIZE) {
1257                 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1258                 huge_pte_to_pagemap_entry(&pme, *pte, offset);
1259                 err = add_to_pagemap(addr, &pme, pm);
1260                 if (err)
1261                         return err;
1262         }
1263
1264         cond_resched();
1265
1266         return err;
1267 }
1268 #endif /* HUGETLB_PAGE */
1269
1270 /*
1271  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1272  *
1273  * For each page in the address space, this file contains one 64-bit entry
1274  * consisting of the following:
1275  *
1276  * Bits 0-54  page frame number (PFN) if present
1277  * Bits 0-4   swap type if swapped
1278  * Bits 5-54  swap offset if swapped
1279  * Bits 55-60 page shift (page size = 1<<page shift)
1280  * Bit  61    page is file-page or shared-anon
1281  * Bit  62    page swapped
1282  * Bit  63    page present
1283  *
1284  * If the page is not present but in swap, then the PFN contains an
1285  * encoding of the swap file number and the page's offset into the
1286  * swap. Unmapped pages return a null PFN. This allows determining
1287  * precisely which pages are mapped (or in swap) and comparing mapped
1288  * pages between processes.
1289  *
1290  * Efficient users of this interface will use /proc/pid/maps to
1291  * determine which areas of memory are actually mapped and llseek to
1292  * skip over unmapped regions.
1293  */
1294 static ssize_t pagemap_read(struct file *file, char __user *buf,
1295                             size_t count, loff_t *ppos)
1296 {
1297         struct task_struct *task = get_proc_task(file_inode(file));
1298         struct mm_struct *mm;
1299         struct pagemapread pm;
1300         int ret = -ESRCH;
1301         struct mm_walk pagemap_walk = {};
1302         unsigned long src;
1303         unsigned long svpfn;
1304         unsigned long start_vaddr;
1305         unsigned long end_vaddr;
1306         int copied = 0;
1307
1308         if (!task)
1309                 goto out;
1310
1311         ret = -EINVAL;
1312         /* file position must be aligned */
1313         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1314                 goto out_task;
1315
1316         ret = 0;
1317         if (!count)
1318                 goto out_task;
1319
1320         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1321         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1322         ret = -ENOMEM;
1323         if (!pm.buffer)
1324                 goto out_task;
1325
1326         mm = mm_access(task, PTRACE_MODE_READ);
1327         ret = PTR_ERR(mm);
1328         if (!mm || IS_ERR(mm))
1329                 goto out_free;
1330
1331         pagemap_walk.pmd_entry = pagemap_pte_range;
1332         pagemap_walk.pte_hole = pagemap_pte_hole;
1333 #ifdef CONFIG_HUGETLB_PAGE
1334         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1335 #endif
1336         pagemap_walk.mm = mm;
1337         pagemap_walk.private = &pm;
1338
1339         src = *ppos;
1340         svpfn = src / PM_ENTRY_BYTES;
1341         start_vaddr = svpfn << PAGE_SHIFT;
1342         end_vaddr = TASK_SIZE_OF(task);
1343
1344         /* watch out for wraparound */
1345         if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1346                 start_vaddr = end_vaddr;
1347
1348         /*
1349          * The odds are that this will stop walking way
1350          * before end_vaddr, because the length of the
1351          * user buffer is tracked in "pm", and the walk
1352          * will stop when we hit the end of the buffer.
1353          */
1354         ret = 0;
1355         while (count && (start_vaddr < end_vaddr)) {
1356                 int len;
1357                 unsigned long end;
1358
1359                 pm.pos = 0;
1360                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1361                 /* overflow ? */
1362                 if (end < start_vaddr || end > end_vaddr)
1363                         end = end_vaddr;
1364                 down_read(&mm->mmap_sem);
1365                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1366                 up_read(&mm->mmap_sem);
1367                 start_vaddr = end;
1368
1369                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1370                 if (copy_to_user(buf, pm.buffer, len)) {
1371                         ret = -EFAULT;
1372                         goto out_mm;
1373                 }
1374                 copied += len;
1375                 buf += len;
1376                 count -= len;
1377         }
1378         *ppos += copied;
1379         if (!ret || ret == PM_END_OF_BUFFER)
1380                 ret = copied;
1381
1382 out_mm:
1383         mmput(mm);
1384 out_free:
1385         kfree(pm.buffer);
1386 out_task:
1387         put_task_struct(task);
1388 out:
1389         return ret;
1390 }
1391
1392 const struct file_operations proc_pagemap_operations = {
1393         .llseek         = mem_lseek, /* borrow this */
1394         .read           = pagemap_read,
1395 };
1396 #endif /* CONFIG_PROC_PAGE_MONITOR */
1397
1398 #ifdef CONFIG_NUMA
1399
1400 struct numa_maps {
1401         struct vm_area_struct *vma;
1402         unsigned long pages;
1403         unsigned long anon;
1404         unsigned long active;
1405         unsigned long writeback;
1406         unsigned long mapcount_max;
1407         unsigned long dirty;
1408         unsigned long swapcache;
1409         unsigned long node[MAX_NUMNODES];
1410 };
1411
1412 struct numa_maps_private {
1413         struct proc_maps_private proc_maps;
1414         struct numa_maps md;
1415 };
1416
1417 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1418                         unsigned long nr_pages)
1419 {
1420         int count = page_mapcount(page);
1421
1422         md->pages += nr_pages;
1423         if (pte_dirty || PageDirty(page))
1424                 md->dirty += nr_pages;
1425
1426         if (PageSwapCache(page))
1427                 md->swapcache += nr_pages;
1428
1429         if (PageActive(page) || PageUnevictable(page))
1430                 md->active += nr_pages;
1431
1432         if (PageWriteback(page))
1433                 md->writeback += nr_pages;
1434
1435         if (PageAnon(page))
1436                 md->anon += nr_pages;
1437
1438         if (count > md->mapcount_max)
1439                 md->mapcount_max = count;
1440
1441         md->node[page_to_nid(page)] += nr_pages;
1442 }
1443
1444 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1445                 unsigned long addr)
1446 {
1447         struct page *page;
1448         int nid;
1449
1450         if (!pte_present(pte))
1451                 return NULL;
1452
1453         page = vm_normal_page(vma, addr, pte);
1454         if (!page)
1455                 return NULL;
1456
1457         if (PageReserved(page))
1458                 return NULL;
1459
1460         nid = page_to_nid(page);
1461         if (!node_isset(nid, node_states[N_MEMORY]))
1462                 return NULL;
1463
1464         return page;
1465 }
1466
1467 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1468                 unsigned long end, struct mm_walk *walk)
1469 {
1470         struct numa_maps *md;
1471         spinlock_t *ptl;
1472         pte_t *orig_pte;
1473         pte_t *pte;
1474
1475         md = walk->private;
1476
1477         if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1478                 pte_t huge_pte = *(pte_t *)pmd;
1479                 struct page *page;
1480
1481                 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1482                 if (page)
1483                         gather_stats(page, md, pte_dirty(huge_pte),
1484                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1485                 spin_unlock(&walk->mm->page_table_lock);
1486                 return 0;
1487         }
1488
1489         if (pmd_trans_unstable(pmd))
1490                 return 0;
1491         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1492         do {
1493                 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1494                 if (!page)
1495                         continue;
1496                 gather_stats(page, md, pte_dirty(*pte), 1);
1497
1498         } while (pte++, addr += PAGE_SIZE, addr != end);
1499         pte_unmap_unlock(orig_pte, ptl);
1500         return 0;
1501 }
1502 #ifdef CONFIG_HUGETLB_PAGE
1503 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1504                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1505 {
1506         struct numa_maps *md;
1507         struct page *page;
1508
1509         if (pte_none(*pte))
1510                 return 0;
1511
1512         page = pte_page(*pte);
1513         if (!page)
1514                 return 0;
1515
1516         md = walk->private;
1517         gather_stats(page, md, pte_dirty(*pte), 1);
1518         return 0;
1519 }
1520
1521 #else
1522 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1523                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1524 {
1525         return 0;
1526 }
1527 #endif
1528
1529 /*
1530  * Display pages allocated per node and memory policy via /proc.
1531  */
1532 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1533 {
1534         struct numa_maps_private *numa_priv = m->private;
1535         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1536         struct vm_area_struct *vma = v;
1537         struct numa_maps *md = &numa_priv->md;
1538         struct file *file = vma->vm_file;
1539         struct task_struct *task = proc_priv->task;
1540         struct mm_struct *mm = vma->vm_mm;
1541         struct mm_walk walk = {};
1542         struct mempolicy *pol;
1543         int n;
1544         char buffer[50];
1545
1546         if (!mm)
1547                 return 0;
1548
1549         /* Ensure we start with an empty set of numa_maps statistics. */
1550         memset(md, 0, sizeof(*md));
1551
1552         md->vma = vma;
1553
1554         walk.hugetlb_entry = gather_hugetbl_stats;
1555         walk.pmd_entry = gather_pte_stats;
1556         walk.private = md;
1557         walk.mm = mm;
1558
1559         pol = get_vma_policy(task, vma, vma->vm_start);
1560         mpol_to_str(buffer, sizeof(buffer), pol);
1561         mpol_cond_put(pol);
1562
1563         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1564
1565         if (file) {
1566                 seq_printf(m, " file=");
1567                 seq_path(m, &file->f_path, "\n\t= ");
1568         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1569                 seq_printf(m, " heap");
1570         } else {
1571                 pid_t tid = vm_is_stack(task, vma, is_pid);
1572                 if (tid != 0) {
1573                         /*
1574                          * Thread stack in /proc/PID/task/TID/maps or
1575                          * the main process stack.
1576                          */
1577                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
1578                             vma->vm_end >= mm->start_stack))
1579                                 seq_printf(m, " stack");
1580                         else
1581                                 seq_printf(m, " stack:%d", tid);
1582                 }
1583         }
1584
1585         if (is_vm_hugetlb_page(vma))
1586                 seq_printf(m, " huge");
1587
1588         walk_page_range(vma->vm_start, vma->vm_end, &walk);
1589
1590         if (!md->pages)
1591                 goto out;
1592
1593         if (md->anon)
1594                 seq_printf(m, " anon=%lu", md->anon);
1595
1596         if (md->dirty)
1597                 seq_printf(m, " dirty=%lu", md->dirty);
1598
1599         if (md->pages != md->anon && md->pages != md->dirty)
1600                 seq_printf(m, " mapped=%lu", md->pages);
1601
1602         if (md->mapcount_max > 1)
1603                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1604
1605         if (md->swapcache)
1606                 seq_printf(m, " swapcache=%lu", md->swapcache);
1607
1608         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1609                 seq_printf(m, " active=%lu", md->active);
1610
1611         if (md->writeback)
1612                 seq_printf(m, " writeback=%lu", md->writeback);
1613
1614         for_each_node_state(n, N_MEMORY)
1615                 if (md->node[n])
1616                         seq_printf(m, " N%d=%lu", n, md->node[n]);
1617 out:
1618         seq_putc(m, '\n');
1619
1620         if (m->count < m->size)
1621                 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1622         return 0;
1623 }
1624
1625 static int show_pid_numa_map(struct seq_file *m, void *v)
1626 {
1627         return show_numa_map(m, v, 1);
1628 }
1629
1630 static int show_tid_numa_map(struct seq_file *m, void *v)
1631 {
1632         return show_numa_map(m, v, 0);
1633 }
1634
1635 static const struct seq_operations proc_pid_numa_maps_op = {
1636         .start  = m_start,
1637         .next   = m_next,
1638         .stop   = m_stop,
1639         .show   = show_pid_numa_map,
1640 };
1641
1642 static const struct seq_operations proc_tid_numa_maps_op = {
1643         .start  = m_start,
1644         .next   = m_next,
1645         .stop   = m_stop,
1646         .show   = show_tid_numa_map,
1647 };
1648
1649 static int numa_maps_open(struct inode *inode, struct file *file,
1650                           const struct seq_operations *ops)
1651 {
1652         struct numa_maps_private *priv;
1653         int ret = -ENOMEM;
1654         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1655         if (priv) {
1656                 priv->proc_maps.pid = proc_pid(inode);
1657                 ret = seq_open(file, ops);
1658                 if (!ret) {
1659                         struct seq_file *m = file->private_data;
1660                         m->private = priv;
1661                 } else {
1662                         kfree(priv);
1663                 }
1664         }
1665         return ret;
1666 }
1667
1668 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1669 {
1670         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1671 }
1672
1673 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1674 {
1675         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1676 }
1677
1678 const struct file_operations proc_pid_numa_maps_operations = {
1679         .open           = pid_numa_maps_open,
1680         .read           = seq_read,
1681         .llseek         = seq_lseek,
1682         .release        = seq_release_private,
1683 };
1684
1685 const struct file_operations proc_tid_numa_maps_operations = {
1686         .open           = tid_numa_maps_open,
1687         .read           = seq_read,
1688         .llseek         = seq_lseek,
1689         .release        = seq_release_private,
1690 };
1691 #endif /* CONFIG_NUMA */