1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/pagewalk.h>
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 #include <linux/pkeys.h>
25 #include <asm/tlbflush.h>
28 #define SEQ_PUT_DEC(str, val) \
29 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30 void task_mem(struct seq_file *m, struct mm_struct *mm)
32 unsigned long text, lib, swap, anon, file, shmem;
33 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
35 anon = get_mm_counter(mm, MM_ANONPAGES);
36 file = get_mm_counter(mm, MM_FILEPAGES);
37 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
40 * Note: to minimize their overhead, mm maintains hiwater_vm and
41 * hiwater_rss only when about to *lower* total_vm or rss. Any
42 * collector of these hiwater stats must therefore get total_vm
43 * and rss too, which will usually be the higher. Barriers? not
44 * worth the effort, such snapshots can always be inconsistent.
46 hiwater_vm = total_vm = mm->total_vm;
47 if (hiwater_vm < mm->hiwater_vm)
48 hiwater_vm = mm->hiwater_vm;
49 hiwater_rss = total_rss = anon + file + shmem;
50 if (hiwater_rss < mm->hiwater_rss)
51 hiwater_rss = mm->hiwater_rss;
53 /* split executable areas between text and lib */
54 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
55 text = min(text, mm->exec_vm << PAGE_SHIFT);
56 lib = (mm->exec_vm << PAGE_SHIFT) - text;
58 swap = get_mm_counter(mm, MM_SWAPENTS);
59 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
60 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
61 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
62 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
63 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
64 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
65 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
66 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
67 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
68 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
69 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
70 seq_put_decimal_ull_width(m,
71 " kB\nVmExe:\t", text >> 10, 8);
72 seq_put_decimal_ull_width(m,
73 " kB\nVmLib:\t", lib >> 10, 8);
74 seq_put_decimal_ull_width(m,
75 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
76 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
78 hugetlb_report_usage(m, mm);
82 unsigned long task_vsize(struct mm_struct *mm)
84 return PAGE_SIZE * mm->total_vm;
87 unsigned long task_statm(struct mm_struct *mm,
88 unsigned long *shared, unsigned long *text,
89 unsigned long *data, unsigned long *resident)
91 *shared = get_mm_counter(mm, MM_FILEPAGES) +
92 get_mm_counter(mm, MM_SHMEMPAGES);
93 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
95 *data = mm->data_vm + mm->stack_vm;
96 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
102 * Save get_task_policy() for show_numa_map().
104 static void hold_task_mempolicy(struct proc_maps_private *priv)
106 struct task_struct *task = priv->task;
109 priv->task_mempolicy = get_task_policy(task);
110 mpol_get(priv->task_mempolicy);
113 static void release_task_mempolicy(struct proc_maps_private *priv)
115 mpol_put(priv->task_mempolicy);
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
121 static void release_task_mempolicy(struct proc_maps_private *priv)
126 static void *m_start(struct seq_file *m, loff_t *ppos)
128 struct proc_maps_private *priv = m->private;
129 unsigned long last_addr = *ppos;
130 struct mm_struct *mm;
131 struct vm_area_struct *vma;
133 /* See m_next(). Zero at the start or after lseek. */
134 if (last_addr == -1UL)
137 priv->task = get_proc_task(priv->inode);
139 return ERR_PTR(-ESRCH);
142 if (!mm || !mmget_not_zero(mm)) {
143 put_task_struct(priv->task);
148 if (mmap_read_lock_killable(mm)) {
150 put_task_struct(priv->task);
152 return ERR_PTR(-EINTR);
155 hold_task_mempolicy(priv);
156 priv->tail_vma = get_gate_vma(mm);
158 vma = find_vma(mm, last_addr);
162 return priv->tail_vma;
165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
167 struct proc_maps_private *priv = m->private;
168 struct vm_area_struct *next, *vma = v;
170 if (vma == priv->tail_vma)
172 else if (vma->vm_next)
175 next = priv->tail_vma;
177 *ppos = next ? next->vm_start : -1UL;
182 static void m_stop(struct seq_file *m, void *v)
184 struct proc_maps_private *priv = m->private;
185 struct mm_struct *mm = priv->mm;
190 release_task_mempolicy(priv);
191 mmap_read_unlock(mm);
193 put_task_struct(priv->task);
197 static int proc_maps_open(struct inode *inode, struct file *file,
198 const struct seq_operations *ops, int psize)
200 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
206 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
207 if (IS_ERR(priv->mm)) {
208 int err = PTR_ERR(priv->mm);
210 seq_release_private(inode, file);
217 static int proc_map_release(struct inode *inode, struct file *file)
219 struct seq_file *seq = file->private_data;
220 struct proc_maps_private *priv = seq->private;
225 return seq_release_private(inode, file);
228 static int do_maps_open(struct inode *inode, struct file *file,
229 const struct seq_operations *ops)
231 return proc_maps_open(inode, file, ops,
232 sizeof(struct proc_maps_private));
236 * Indicate if the VMA is a stack for the given task; for
237 * /proc/PID/maps that is the stack of the main task.
239 static int is_stack(struct vm_area_struct *vma)
242 * We make no effort to guess what a given thread considers to be
243 * its "stack". It's not even well-defined for programs written
246 return vma->vm_start <= vma->vm_mm->start_stack &&
247 vma->vm_end >= vma->vm_mm->start_stack;
250 static void show_vma_header_prefix(struct seq_file *m,
251 unsigned long start, unsigned long end,
252 vm_flags_t flags, unsigned long long pgoff,
253 dev_t dev, unsigned long ino)
255 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
256 seq_put_hex_ll(m, NULL, start, 8);
257 seq_put_hex_ll(m, "-", end, 8);
259 seq_putc(m, flags & VM_READ ? 'r' : '-');
260 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
261 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
262 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
263 seq_put_hex_ll(m, " ", pgoff, 8);
264 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
265 seq_put_hex_ll(m, ":", MINOR(dev), 2);
266 seq_put_decimal_ull(m, " ", ino);
271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
273 struct mm_struct *mm = vma->vm_mm;
274 struct file *file = vma->vm_file;
275 vm_flags_t flags = vma->vm_flags;
276 unsigned long ino = 0;
277 unsigned long long pgoff = 0;
278 unsigned long start, end;
280 const char *name = NULL;
283 struct inode *inode = file_inode(vma->vm_file);
284 dev = inode->i_sb->s_dev;
286 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
289 start = vma->vm_start;
291 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
294 * Print the dentry name for named mappings, and a
295 * special [heap] marker for the heap:
299 seq_file_path(m, file, "\n");
303 if (vma->vm_ops && vma->vm_ops->name) {
304 name = vma->vm_ops->name(vma);
309 name = arch_vma_name(vma);
316 if (vma->vm_start <= mm->brk &&
317 vma->vm_end >= mm->start_brk) {
334 static int show_map(struct seq_file *m, void *v)
340 static const struct seq_operations proc_pid_maps_op = {
347 static int pid_maps_open(struct inode *inode, struct file *file)
349 return do_maps_open(inode, file, &proc_pid_maps_op);
352 const struct file_operations proc_pid_maps_operations = {
353 .open = pid_maps_open,
356 .release = proc_map_release,
360 * Proportional Set Size(PSS): my share of RSS.
362 * PSS of a process is the count of pages it has in memory, where each
363 * page is divided by the number of processes sharing it. So if a
364 * process has 1000 pages all to itself, and 1000 shared with one other
365 * process, its PSS will be 1500.
367 * To keep (accumulated) division errors low, we adopt a 64bit
368 * fixed-point pss counter to minimize division errors. So (pss >>
369 * PSS_SHIFT) would be the real byte count.
371 * A shift of 12 before division means (assuming 4K page size):
372 * - 1M 3-user-pages add up to 8KB errors;
373 * - supports mapcount up to 2^24, or 16M;
374 * - supports PSS up to 2^52 bytes, or 4PB.
378 #ifdef CONFIG_PROC_PAGE_MONITOR
379 struct mem_size_stats {
380 unsigned long resident;
381 unsigned long shared_clean;
382 unsigned long shared_dirty;
383 unsigned long private_clean;
384 unsigned long private_dirty;
385 unsigned long referenced;
386 unsigned long anonymous;
387 unsigned long lazyfree;
388 unsigned long anonymous_thp;
389 unsigned long shmem_thp;
390 unsigned long file_thp;
392 unsigned long shared_hugetlb;
393 unsigned long private_hugetlb;
400 bool check_shmem_swap;
403 static void smaps_page_accumulate(struct mem_size_stats *mss,
404 struct page *page, unsigned long size, unsigned long pss,
405 bool dirty, bool locked, bool private)
410 mss->pss_anon += pss;
411 else if (PageSwapBacked(page))
412 mss->pss_shmem += pss;
414 mss->pss_file += pss;
417 mss->pss_locked += pss;
419 if (dirty || PageDirty(page)) {
421 mss->private_dirty += size;
423 mss->shared_dirty += size;
426 mss->private_clean += size;
428 mss->shared_clean += size;
432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
433 bool compound, bool young, bool dirty, bool locked,
436 int i, nr = compound ? compound_nr(page) : 1;
437 unsigned long size = nr * PAGE_SIZE;
440 * First accumulate quantities that depend only on |size| and the type
441 * of the compound page.
443 if (PageAnon(page)) {
444 mss->anonymous += size;
445 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
446 mss->lazyfree += size;
449 mss->resident += size;
450 /* Accumulate the size in pages that have been accessed. */
451 if (young || page_is_young(page) || PageReferenced(page))
452 mss->referenced += size;
455 * Then accumulate quantities that may depend on sharing, or that may
456 * differ page-by-page.
458 * page_count(page) == 1 guarantees the page is mapped exactly once.
459 * If any subpage of the compound page mapped with PTE it would elevate
462 * The page_mapcount() is called to get a snapshot of the mapcount.
463 * Without holding the page lock this snapshot can be slightly wrong as
464 * we cannot always read the mapcount atomically. It is not safe to
465 * call page_mapcount() even with PTL held if the page is not mapped,
466 * especially for migration entries. Treat regular migration entries
469 if ((page_count(page) == 1) || migration) {
470 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
474 for (i = 0; i < nr; i++, page++) {
475 int mapcount = page_mapcount(page);
476 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
479 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
485 static int smaps_pte_hole(unsigned long addr, unsigned long end,
486 __always_unused int depth, struct mm_walk *walk)
488 struct mem_size_stats *mss = walk->private;
490 mss->swap += shmem_partial_swap_usage(
491 walk->vma->vm_file->f_mapping, addr, end);
496 #define smaps_pte_hole NULL
497 #endif /* CONFIG_SHMEM */
499 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
500 struct mm_walk *walk)
502 struct mem_size_stats *mss = walk->private;
503 struct vm_area_struct *vma = walk->vma;
504 bool locked = !!(vma->vm_flags & VM_LOCKED);
505 struct page *page = NULL;
506 bool migration = false, young = false, dirty = false;
508 if (pte_present(*pte)) {
509 page = vm_normal_page(vma, addr, *pte);
510 young = pte_young(*pte);
511 dirty = pte_dirty(*pte);
512 } else if (is_swap_pte(*pte)) {
513 swp_entry_t swpent = pte_to_swp_entry(*pte);
515 if (!non_swap_entry(swpent)) {
518 mss->swap += PAGE_SIZE;
519 mapcount = swp_swapcount(swpent);
521 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
523 do_div(pss_delta, mapcount);
524 mss->swap_pss += pss_delta;
526 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
528 } else if (is_pfn_swap_entry(swpent)) {
529 if (is_migration_entry(swpent))
531 page = pfn_swap_entry_to_page(swpent);
533 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
534 && pte_none(*pte))) {
535 page = xa_load(&vma->vm_file->f_mapping->i_pages,
536 linear_page_index(vma, addr));
537 if (xa_is_value(page))
538 mss->swap += PAGE_SIZE;
545 smaps_account(mss, page, false, young, dirty, locked, migration);
548 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
549 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
550 struct mm_walk *walk)
552 struct mem_size_stats *mss = walk->private;
553 struct vm_area_struct *vma = walk->vma;
554 bool locked = !!(vma->vm_flags & VM_LOCKED);
555 struct page *page = NULL;
556 bool migration = false;
558 if (pmd_present(*pmd)) {
559 /* FOLL_DUMP will return -EFAULT on huge zero page */
560 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
561 } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
562 swp_entry_t entry = pmd_to_swp_entry(*pmd);
564 if (is_migration_entry(entry)) {
566 page = pfn_swap_entry_to_page(entry);
569 if (IS_ERR_OR_NULL(page))
572 mss->anonymous_thp += HPAGE_PMD_SIZE;
573 else if (PageSwapBacked(page))
574 mss->shmem_thp += HPAGE_PMD_SIZE;
575 else if (is_zone_device_page(page))
578 mss->file_thp += HPAGE_PMD_SIZE;
580 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
584 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
585 struct mm_walk *walk)
590 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
591 struct mm_walk *walk)
593 struct vm_area_struct *vma = walk->vma;
597 ptl = pmd_trans_huge_lock(pmd, vma);
599 smaps_pmd_entry(pmd, addr, walk);
604 if (pmd_trans_unstable(pmd))
607 * The mmap_lock held all the way back in m_start() is what
608 * keeps khugepaged out of here and from collapsing things
611 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
612 for (; addr != end; pte++, addr += PAGE_SIZE)
613 smaps_pte_entry(pte, addr, walk);
614 pte_unmap_unlock(pte - 1, ptl);
620 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
623 * Don't forget to update Documentation/ on changes.
625 static const char mnemonics[BITS_PER_LONG][2] = {
627 * In case if we meet a flag we don't know about.
629 [0 ... (BITS_PER_LONG-1)] = "??",
631 [ilog2(VM_READ)] = "rd",
632 [ilog2(VM_WRITE)] = "wr",
633 [ilog2(VM_EXEC)] = "ex",
634 [ilog2(VM_SHARED)] = "sh",
635 [ilog2(VM_MAYREAD)] = "mr",
636 [ilog2(VM_MAYWRITE)] = "mw",
637 [ilog2(VM_MAYEXEC)] = "me",
638 [ilog2(VM_MAYSHARE)] = "ms",
639 [ilog2(VM_GROWSDOWN)] = "gd",
640 [ilog2(VM_PFNMAP)] = "pf",
641 [ilog2(VM_LOCKED)] = "lo",
642 [ilog2(VM_IO)] = "io",
643 [ilog2(VM_SEQ_READ)] = "sr",
644 [ilog2(VM_RAND_READ)] = "rr",
645 [ilog2(VM_DONTCOPY)] = "dc",
646 [ilog2(VM_DONTEXPAND)] = "de",
647 [ilog2(VM_ACCOUNT)] = "ac",
648 [ilog2(VM_NORESERVE)] = "nr",
649 [ilog2(VM_HUGETLB)] = "ht",
650 [ilog2(VM_SYNC)] = "sf",
651 [ilog2(VM_ARCH_1)] = "ar",
652 [ilog2(VM_WIPEONFORK)] = "wf",
653 [ilog2(VM_DONTDUMP)] = "dd",
654 #ifdef CONFIG_ARM64_BTI
655 [ilog2(VM_ARM64_BTI)] = "bt",
657 #ifdef CONFIG_MEM_SOFT_DIRTY
658 [ilog2(VM_SOFTDIRTY)] = "sd",
660 [ilog2(VM_MIXEDMAP)] = "mm",
661 [ilog2(VM_HUGEPAGE)] = "hg",
662 [ilog2(VM_NOHUGEPAGE)] = "nh",
663 [ilog2(VM_MERGEABLE)] = "mg",
664 [ilog2(VM_UFFD_MISSING)]= "um",
665 [ilog2(VM_UFFD_WP)] = "uw",
666 #ifdef CONFIG_ARM64_MTE
667 [ilog2(VM_MTE)] = "mt",
668 [ilog2(VM_MTE_ALLOWED)] = "",
670 #ifdef CONFIG_ARCH_HAS_PKEYS
671 /* These come out via ProtectionKey: */
672 [ilog2(VM_PKEY_BIT0)] = "",
673 [ilog2(VM_PKEY_BIT1)] = "",
674 [ilog2(VM_PKEY_BIT2)] = "",
675 [ilog2(VM_PKEY_BIT3)] = "",
677 [ilog2(VM_PKEY_BIT4)] = "",
679 #endif /* CONFIG_ARCH_HAS_PKEYS */
680 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
681 [ilog2(VM_UFFD_MINOR)] = "ui",
682 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
686 seq_puts(m, "VmFlags: ");
687 for (i = 0; i < BITS_PER_LONG; i++) {
688 if (!mnemonics[i][0])
690 if (vma->vm_flags & (1UL << i)) {
691 seq_putc(m, mnemonics[i][0]);
692 seq_putc(m, mnemonics[i][1]);
699 #ifdef CONFIG_HUGETLB_PAGE
700 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
701 unsigned long addr, unsigned long end,
702 struct mm_walk *walk)
704 struct mem_size_stats *mss = walk->private;
705 struct vm_area_struct *vma = walk->vma;
706 struct page *page = NULL;
708 if (pte_present(*pte)) {
709 page = vm_normal_page(vma, addr, *pte);
710 } else if (is_swap_pte(*pte)) {
711 swp_entry_t swpent = pte_to_swp_entry(*pte);
713 if (is_pfn_swap_entry(swpent))
714 page = pfn_swap_entry_to_page(swpent);
717 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
718 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
720 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
725 #define smaps_hugetlb_range NULL
726 #endif /* HUGETLB_PAGE */
728 static const struct mm_walk_ops smaps_walk_ops = {
729 .pmd_entry = smaps_pte_range,
730 .hugetlb_entry = smaps_hugetlb_range,
733 static const struct mm_walk_ops smaps_shmem_walk_ops = {
734 .pmd_entry = smaps_pte_range,
735 .hugetlb_entry = smaps_hugetlb_range,
736 .pte_hole = smaps_pte_hole,
740 * Gather mem stats from @vma with the indicated beginning
741 * address @start, and keep them in @mss.
743 * Use vm_start of @vma as the beginning address if @start is 0.
745 static void smap_gather_stats(struct vm_area_struct *vma,
746 struct mem_size_stats *mss, unsigned long start)
748 const struct mm_walk_ops *ops = &smaps_walk_ops;
751 if (start >= vma->vm_end)
755 /* In case of smaps_rollup, reset the value from previous vma */
756 mss->check_shmem_swap = false;
757 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
759 * For shared or readonly shmem mappings we know that all
760 * swapped out pages belong to the shmem object, and we can
761 * obtain the swap value much more efficiently. For private
762 * writable mappings, we might have COW pages that are
763 * not affected by the parent swapped out pages of the shmem
764 * object, so we have to distinguish them during the page walk.
765 * Unless we know that the shmem object (or the part mapped by
766 * our VMA) has no swapped out pages at all.
768 unsigned long shmem_swapped = shmem_swap_usage(vma);
770 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
771 !(vma->vm_flags & VM_WRITE))) {
772 mss->swap += shmem_swapped;
774 mss->check_shmem_swap = true;
775 ops = &smaps_shmem_walk_ops;
779 /* mmap_lock is held in m_start */
781 walk_page_vma(vma, ops, mss);
783 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
786 #define SEQ_PUT_DEC(str, val) \
787 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
789 /* Show the contents common for smaps and smaps_rollup */
790 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
793 SEQ_PUT_DEC("Rss: ", mss->resident);
794 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
797 * These are meaningful only for smaps_rollup, otherwise two of
798 * them are zero, and the other one is the same as Pss.
800 SEQ_PUT_DEC(" kB\nPss_Anon: ",
801 mss->pss_anon >> PSS_SHIFT);
802 SEQ_PUT_DEC(" kB\nPss_File: ",
803 mss->pss_file >> PSS_SHIFT);
804 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
805 mss->pss_shmem >> PSS_SHIFT);
807 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
808 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
809 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
810 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
811 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
812 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
813 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
814 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
815 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
816 SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
817 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
818 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
819 mss->private_hugetlb >> 10, 7);
820 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
821 SEQ_PUT_DEC(" kB\nSwapPss: ",
822 mss->swap_pss >> PSS_SHIFT);
823 SEQ_PUT_DEC(" kB\nLocked: ",
824 mss->pss_locked >> PSS_SHIFT);
825 seq_puts(m, " kB\n");
828 static int show_smap(struct seq_file *m, void *v)
830 struct vm_area_struct *vma = v;
831 struct mem_size_stats mss;
833 memset(&mss, 0, sizeof(mss));
835 smap_gather_stats(vma, &mss, 0);
837 show_map_vma(m, vma);
839 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
840 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
841 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
842 seq_puts(m, " kB\n");
844 __show_smap(m, &mss, false);
846 seq_printf(m, "THPeligible: %d\n",
847 transparent_hugepage_active(vma));
849 if (arch_pkeys_enabled())
850 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
851 show_smap_vma_flags(m, vma);
856 static int show_smaps_rollup(struct seq_file *m, void *v)
858 struct proc_maps_private *priv = m->private;
859 struct mem_size_stats mss;
860 struct mm_struct *mm;
861 struct vm_area_struct *vma;
862 unsigned long last_vma_end = 0;
865 priv->task = get_proc_task(priv->inode);
870 if (!mm || !mmget_not_zero(mm)) {
875 memset(&mss, 0, sizeof(mss));
877 ret = mmap_read_lock_killable(mm);
881 hold_task_mempolicy(priv);
883 for (vma = priv->mm->mmap; vma;) {
884 smap_gather_stats(vma, &mss, 0);
885 last_vma_end = vma->vm_end;
888 * Release mmap_lock temporarily if someone wants to
889 * access it for write request.
891 if (mmap_lock_is_contended(mm)) {
892 mmap_read_unlock(mm);
893 ret = mmap_read_lock_killable(mm);
895 release_task_mempolicy(priv);
900 * After dropping the lock, there are four cases to
901 * consider. See the following example for explanation.
903 * +------+------+-----------+
904 * | VMA1 | VMA2 | VMA3 |
905 * +------+------+-----------+
909 * Suppose we drop the lock after reading VMA2 due to
910 * contention, then we get:
914 * 1) VMA2 is freed, but VMA3 exists:
916 * find_vma(mm, 16k - 1) will return VMA3.
917 * In this case, just continue from VMA3.
919 * 2) VMA2 still exists:
921 * find_vma(mm, 16k - 1) will return VMA2.
922 * Iterate the loop like the original one.
924 * 3) No more VMAs can be found:
926 * find_vma(mm, 16k - 1) will return NULL.
927 * No more things to do, just break.
929 * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
931 * find_vma(mm, 16k - 1) will return VMA' whose range
932 * contains last_vma_end.
933 * Iterate VMA' from last_vma_end.
935 vma = find_vma(mm, last_vma_end - 1);
941 if (vma->vm_start >= last_vma_end)
945 if (vma->vm_end > last_vma_end)
946 smap_gather_stats(vma, &mss, last_vma_end);
952 show_vma_header_prefix(m, priv->mm->mmap ? priv->mm->mmap->vm_start : 0,
953 last_vma_end, 0, 0, 0, 0);
955 seq_puts(m, "[rollup]\n");
957 __show_smap(m, &mss, true);
959 release_task_mempolicy(priv);
960 mmap_read_unlock(mm);
965 put_task_struct(priv->task);
972 static const struct seq_operations proc_pid_smaps_op = {
979 static int pid_smaps_open(struct inode *inode, struct file *file)
981 return do_maps_open(inode, file, &proc_pid_smaps_op);
984 static int smaps_rollup_open(struct inode *inode, struct file *file)
987 struct proc_maps_private *priv;
989 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
993 ret = single_open(file, show_smaps_rollup, priv);
998 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
999 if (IS_ERR(priv->mm)) {
1000 ret = PTR_ERR(priv->mm);
1002 single_release(inode, file);
1013 static int smaps_rollup_release(struct inode *inode, struct file *file)
1015 struct seq_file *seq = file->private_data;
1016 struct proc_maps_private *priv = seq->private;
1022 return single_release(inode, file);
1025 const struct file_operations proc_pid_smaps_operations = {
1026 .open = pid_smaps_open,
1028 .llseek = seq_lseek,
1029 .release = proc_map_release,
1032 const struct file_operations proc_pid_smaps_rollup_operations = {
1033 .open = smaps_rollup_open,
1035 .llseek = seq_lseek,
1036 .release = smaps_rollup_release,
1039 enum clear_refs_types {
1043 CLEAR_REFS_SOFT_DIRTY,
1044 CLEAR_REFS_MM_HIWATER_RSS,
1048 struct clear_refs_private {
1049 enum clear_refs_types type;
1052 #ifdef CONFIG_MEM_SOFT_DIRTY
1054 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1058 if (!pte_write(pte))
1060 if (!is_cow_mapping(vma->vm_flags))
1062 if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1064 page = vm_normal_page(vma, addr, pte);
1067 return page_maybe_dma_pinned(page);
1070 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1071 unsigned long addr, pte_t *pte)
1074 * The soft-dirty tracker uses #PF-s to catch writes
1075 * to pages, so write-protect the pte as well. See the
1076 * Documentation/admin-guide/mm/soft-dirty.rst for full description
1077 * of how soft-dirty works.
1081 if (pte_present(ptent)) {
1084 if (pte_is_pinned(vma, addr, ptent))
1086 old_pte = ptep_modify_prot_start(vma, addr, pte);
1087 ptent = pte_wrprotect(old_pte);
1088 ptent = pte_clear_soft_dirty(ptent);
1089 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1090 } else if (is_swap_pte(ptent)) {
1091 ptent = pte_swp_clear_soft_dirty(ptent);
1092 set_pte_at(vma->vm_mm, addr, pte, ptent);
1096 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1097 unsigned long addr, pte_t *pte)
1102 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1103 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1104 unsigned long addr, pmd_t *pmdp)
1106 pmd_t old, pmd = *pmdp;
1108 if (pmd_present(pmd)) {
1109 /* See comment in change_huge_pmd() */
1110 old = pmdp_invalidate(vma, addr, pmdp);
1112 pmd = pmd_mkdirty(pmd);
1114 pmd = pmd_mkyoung(pmd);
1116 pmd = pmd_wrprotect(pmd);
1117 pmd = pmd_clear_soft_dirty(pmd);
1119 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1120 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1121 pmd = pmd_swp_clear_soft_dirty(pmd);
1122 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1126 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1127 unsigned long addr, pmd_t *pmdp)
1132 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1133 unsigned long end, struct mm_walk *walk)
1135 struct clear_refs_private *cp = walk->private;
1136 struct vm_area_struct *vma = walk->vma;
1141 ptl = pmd_trans_huge_lock(pmd, vma);
1143 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1144 clear_soft_dirty_pmd(vma, addr, pmd);
1148 if (!pmd_present(*pmd))
1151 page = pmd_page(*pmd);
1153 /* Clear accessed and referenced bits. */
1154 pmdp_test_and_clear_young(vma, addr, pmd);
1155 test_and_clear_page_young(page);
1156 ClearPageReferenced(page);
1162 if (pmd_trans_unstable(pmd))
1165 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1166 for (; addr != end; pte++, addr += PAGE_SIZE) {
1169 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1170 clear_soft_dirty(vma, addr, pte);
1174 if (!pte_present(ptent))
1177 page = vm_normal_page(vma, addr, ptent);
1181 /* Clear accessed and referenced bits. */
1182 ptep_test_and_clear_young(vma, addr, pte);
1183 test_and_clear_page_young(page);
1184 ClearPageReferenced(page);
1186 pte_unmap_unlock(pte - 1, ptl);
1191 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1192 struct mm_walk *walk)
1194 struct clear_refs_private *cp = walk->private;
1195 struct vm_area_struct *vma = walk->vma;
1197 if (vma->vm_flags & VM_PFNMAP)
1201 * Writing 1 to /proc/pid/clear_refs affects all pages.
1202 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1203 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1204 * Writing 4 to /proc/pid/clear_refs affects all pages.
1206 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1208 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1213 static const struct mm_walk_ops clear_refs_walk_ops = {
1214 .pmd_entry = clear_refs_pte_range,
1215 .test_walk = clear_refs_test_walk,
1218 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1219 size_t count, loff_t *ppos)
1221 struct task_struct *task;
1222 char buffer[PROC_NUMBUF];
1223 struct mm_struct *mm;
1224 struct vm_area_struct *vma;
1225 enum clear_refs_types type;
1229 memset(buffer, 0, sizeof(buffer));
1230 if (count > sizeof(buffer) - 1)
1231 count = sizeof(buffer) - 1;
1232 if (copy_from_user(buffer, buf, count))
1234 rv = kstrtoint(strstrip(buffer), 10, &itype);
1237 type = (enum clear_refs_types)itype;
1238 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1241 task = get_proc_task(file_inode(file));
1244 mm = get_task_mm(task);
1246 struct mmu_notifier_range range;
1247 struct clear_refs_private cp = {
1251 if (mmap_write_lock_killable(mm)) {
1255 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1257 * Writing 5 to /proc/pid/clear_refs resets the peak
1258 * resident set size to this mm's current rss value.
1260 reset_mm_hiwater_rss(mm);
1264 if (type == CLEAR_REFS_SOFT_DIRTY) {
1265 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1266 if (!(vma->vm_flags & VM_SOFTDIRTY))
1268 vma->vm_flags &= ~VM_SOFTDIRTY;
1269 vma_set_page_prot(vma);
1272 inc_tlb_flush_pending(mm);
1273 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1274 0, NULL, mm, 0, -1UL);
1275 mmu_notifier_invalidate_range_start(&range);
1277 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1279 if (type == CLEAR_REFS_SOFT_DIRTY) {
1280 mmu_notifier_invalidate_range_end(&range);
1282 dec_tlb_flush_pending(mm);
1285 mmap_write_unlock(mm);
1289 put_task_struct(task);
1294 const struct file_operations proc_clear_refs_operations = {
1295 .write = clear_refs_write,
1296 .llseek = noop_llseek,
1303 struct pagemapread {
1304 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1305 pagemap_entry_t *buffer;
1309 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1310 #define PAGEMAP_WALK_MASK (PMD_MASK)
1312 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1313 #define PM_PFRAME_BITS 55
1314 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1315 #define PM_SOFT_DIRTY BIT_ULL(55)
1316 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1317 #define PM_UFFD_WP BIT_ULL(57)
1318 #define PM_FILE BIT_ULL(61)
1319 #define PM_SWAP BIT_ULL(62)
1320 #define PM_PRESENT BIT_ULL(63)
1322 #define PM_END_OF_BUFFER 1
1324 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1326 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1329 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1330 struct pagemapread *pm)
1332 pm->buffer[pm->pos++] = *pme;
1333 if (pm->pos >= pm->len)
1334 return PM_END_OF_BUFFER;
1338 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1339 __always_unused int depth, struct mm_walk *walk)
1341 struct pagemapread *pm = walk->private;
1342 unsigned long addr = start;
1345 while (addr < end) {
1346 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1347 pagemap_entry_t pme = make_pme(0, 0);
1348 /* End of address space hole, which we mark as non-present. */
1349 unsigned long hole_end;
1352 hole_end = min(end, vma->vm_start);
1356 for (; addr < hole_end; addr += PAGE_SIZE) {
1357 err = add_to_pagemap(addr, &pme, pm);
1365 /* Addresses in the VMA. */
1366 if (vma->vm_flags & VM_SOFTDIRTY)
1367 pme = make_pme(0, PM_SOFT_DIRTY);
1368 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1369 err = add_to_pagemap(addr, &pme, pm);
1378 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1379 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1381 u64 frame = 0, flags = 0;
1382 struct page *page = NULL;
1383 bool migration = false;
1385 if (pte_present(pte)) {
1387 frame = pte_pfn(pte);
1388 flags |= PM_PRESENT;
1389 page = vm_normal_page(vma, addr, pte);
1390 if (pte_soft_dirty(pte))
1391 flags |= PM_SOFT_DIRTY;
1392 if (pte_uffd_wp(pte))
1393 flags |= PM_UFFD_WP;
1394 } else if (is_swap_pte(pte)) {
1396 if (pte_swp_soft_dirty(pte))
1397 flags |= PM_SOFT_DIRTY;
1398 if (pte_swp_uffd_wp(pte))
1399 flags |= PM_UFFD_WP;
1400 entry = pte_to_swp_entry(pte);
1402 frame = swp_type(entry) |
1403 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1405 migration = is_migration_entry(entry);
1406 if (is_pfn_swap_entry(entry))
1407 page = pfn_swap_entry_to_page(entry);
1410 if (page && !PageAnon(page))
1412 if (page && !migration && page_mapcount(page) == 1)
1413 flags |= PM_MMAP_EXCLUSIVE;
1414 if (vma->vm_flags & VM_SOFTDIRTY)
1415 flags |= PM_SOFT_DIRTY;
1417 return make_pme(frame, flags);
1420 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1421 struct mm_walk *walk)
1423 struct vm_area_struct *vma = walk->vma;
1424 struct pagemapread *pm = walk->private;
1426 pte_t *pte, *orig_pte;
1428 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1429 bool migration = false;
1431 ptl = pmd_trans_huge_lock(pmdp, vma);
1433 u64 flags = 0, frame = 0;
1435 struct page *page = NULL;
1437 if (vma->vm_flags & VM_SOFTDIRTY)
1438 flags |= PM_SOFT_DIRTY;
1440 if (pmd_present(pmd)) {
1441 page = pmd_page(pmd);
1443 flags |= PM_PRESENT;
1444 if (pmd_soft_dirty(pmd))
1445 flags |= PM_SOFT_DIRTY;
1446 if (pmd_uffd_wp(pmd))
1447 flags |= PM_UFFD_WP;
1449 frame = pmd_pfn(pmd) +
1450 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1452 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1453 else if (is_swap_pmd(pmd)) {
1454 swp_entry_t entry = pmd_to_swp_entry(pmd);
1455 unsigned long offset;
1458 offset = swp_offset(entry) +
1459 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1460 frame = swp_type(entry) |
1461 (offset << MAX_SWAPFILES_SHIFT);
1464 if (pmd_swp_soft_dirty(pmd))
1465 flags |= PM_SOFT_DIRTY;
1466 if (pmd_swp_uffd_wp(pmd))
1467 flags |= PM_UFFD_WP;
1468 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1469 migration = is_migration_entry(entry);
1470 page = pfn_swap_entry_to_page(entry);
1474 if (page && !migration && page_mapcount(page) == 1)
1475 flags |= PM_MMAP_EXCLUSIVE;
1477 for (; addr != end; addr += PAGE_SIZE) {
1478 pagemap_entry_t pme = make_pme(frame, flags);
1480 err = add_to_pagemap(addr, &pme, pm);
1484 if (flags & PM_PRESENT)
1486 else if (flags & PM_SWAP)
1487 frame += (1 << MAX_SWAPFILES_SHIFT);
1494 if (pmd_trans_unstable(pmdp))
1496 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1499 * We can assume that @vma always points to a valid one and @end never
1500 * goes beyond vma->vm_end.
1502 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1503 for (; addr < end; pte++, addr += PAGE_SIZE) {
1504 pagemap_entry_t pme;
1506 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1507 err = add_to_pagemap(addr, &pme, pm);
1511 pte_unmap_unlock(orig_pte, ptl);
1518 #ifdef CONFIG_HUGETLB_PAGE
1519 /* This function walks within one hugetlb entry in the single call */
1520 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1521 unsigned long addr, unsigned long end,
1522 struct mm_walk *walk)
1524 struct pagemapread *pm = walk->private;
1525 struct vm_area_struct *vma = walk->vma;
1526 u64 flags = 0, frame = 0;
1530 if (vma->vm_flags & VM_SOFTDIRTY)
1531 flags |= PM_SOFT_DIRTY;
1533 pte = huge_ptep_get(ptep);
1534 if (pte_present(pte)) {
1535 struct page *page = pte_page(pte);
1537 if (!PageAnon(page))
1540 if (page_mapcount(page) == 1)
1541 flags |= PM_MMAP_EXCLUSIVE;
1543 flags |= PM_PRESENT;
1545 frame = pte_pfn(pte) +
1546 ((addr & ~hmask) >> PAGE_SHIFT);
1549 for (; addr != end; addr += PAGE_SIZE) {
1550 pagemap_entry_t pme = make_pme(frame, flags);
1552 err = add_to_pagemap(addr, &pme, pm);
1555 if (pm->show_pfn && (flags & PM_PRESENT))
1564 #define pagemap_hugetlb_range NULL
1565 #endif /* HUGETLB_PAGE */
1567 static const struct mm_walk_ops pagemap_ops = {
1568 .pmd_entry = pagemap_pmd_range,
1569 .pte_hole = pagemap_pte_hole,
1570 .hugetlb_entry = pagemap_hugetlb_range,
1574 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1576 * For each page in the address space, this file contains one 64-bit entry
1577 * consisting of the following:
1579 * Bits 0-54 page frame number (PFN) if present
1580 * Bits 0-4 swap type if swapped
1581 * Bits 5-54 swap offset if swapped
1582 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1583 * Bit 56 page exclusively mapped
1584 * Bit 57 pte is uffd-wp write-protected
1586 * Bit 61 page is file-page or shared-anon
1587 * Bit 62 page swapped
1588 * Bit 63 page present
1590 * If the page is not present but in swap, then the PFN contains an
1591 * encoding of the swap file number and the page's offset into the
1592 * swap. Unmapped pages return a null PFN. This allows determining
1593 * precisely which pages are mapped (or in swap) and comparing mapped
1594 * pages between processes.
1596 * Efficient users of this interface will use /proc/pid/maps to
1597 * determine which areas of memory are actually mapped and llseek to
1598 * skip over unmapped regions.
1600 static ssize_t pagemap_read(struct file *file, char __user *buf,
1601 size_t count, loff_t *ppos)
1603 struct mm_struct *mm = file->private_data;
1604 struct pagemapread pm;
1606 unsigned long svpfn;
1607 unsigned long start_vaddr;
1608 unsigned long end_vaddr;
1609 int ret = 0, copied = 0;
1611 if (!mm || !mmget_not_zero(mm))
1615 /* file position must be aligned */
1616 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1623 /* do not disclose physical addresses: attack vector */
1624 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1626 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1627 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1633 svpfn = src / PM_ENTRY_BYTES;
1634 end_vaddr = mm->task_size;
1636 /* watch out for wraparound */
1637 start_vaddr = end_vaddr;
1638 if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1639 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1641 /* Ensure the address is inside the task */
1642 if (start_vaddr > mm->task_size)
1643 start_vaddr = end_vaddr;
1646 * The odds are that this will stop walking way
1647 * before end_vaddr, because the length of the
1648 * user buffer is tracked in "pm", and the walk
1649 * will stop when we hit the end of the buffer.
1652 while (count && (start_vaddr < end_vaddr)) {
1657 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1659 if (end < start_vaddr || end > end_vaddr)
1661 ret = mmap_read_lock_killable(mm);
1664 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1665 mmap_read_unlock(mm);
1668 len = min(count, PM_ENTRY_BYTES * pm.pos);
1669 if (copy_to_user(buf, pm.buffer, len)) {
1678 if (!ret || ret == PM_END_OF_BUFFER)
1689 static int pagemap_open(struct inode *inode, struct file *file)
1691 struct mm_struct *mm;
1693 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1696 file->private_data = mm;
1700 static int pagemap_release(struct inode *inode, struct file *file)
1702 struct mm_struct *mm = file->private_data;
1709 const struct file_operations proc_pagemap_operations = {
1710 .llseek = mem_lseek, /* borrow this */
1711 .read = pagemap_read,
1712 .open = pagemap_open,
1713 .release = pagemap_release,
1715 #endif /* CONFIG_PROC_PAGE_MONITOR */
1720 unsigned long pages;
1722 unsigned long active;
1723 unsigned long writeback;
1724 unsigned long mapcount_max;
1725 unsigned long dirty;
1726 unsigned long swapcache;
1727 unsigned long node[MAX_NUMNODES];
1730 struct numa_maps_private {
1731 struct proc_maps_private proc_maps;
1732 struct numa_maps md;
1735 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1736 unsigned long nr_pages)
1738 int count = page_mapcount(page);
1740 md->pages += nr_pages;
1741 if (pte_dirty || PageDirty(page))
1742 md->dirty += nr_pages;
1744 if (PageSwapCache(page))
1745 md->swapcache += nr_pages;
1747 if (PageActive(page) || PageUnevictable(page))
1748 md->active += nr_pages;
1750 if (PageWriteback(page))
1751 md->writeback += nr_pages;
1754 md->anon += nr_pages;
1756 if (count > md->mapcount_max)
1757 md->mapcount_max = count;
1759 md->node[page_to_nid(page)] += nr_pages;
1762 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1768 if (!pte_present(pte))
1771 page = vm_normal_page(vma, addr, pte);
1775 if (PageReserved(page))
1778 nid = page_to_nid(page);
1779 if (!node_isset(nid, node_states[N_MEMORY]))
1785 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1786 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1787 struct vm_area_struct *vma,
1793 if (!pmd_present(pmd))
1796 page = vm_normal_page_pmd(vma, addr, pmd);
1800 if (PageReserved(page))
1803 nid = page_to_nid(page);
1804 if (!node_isset(nid, node_states[N_MEMORY]))
1811 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1812 unsigned long end, struct mm_walk *walk)
1814 struct numa_maps *md = walk->private;
1815 struct vm_area_struct *vma = walk->vma;
1820 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1821 ptl = pmd_trans_huge_lock(pmd, vma);
1825 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1827 gather_stats(page, md, pmd_dirty(*pmd),
1828 HPAGE_PMD_SIZE/PAGE_SIZE);
1833 if (pmd_trans_unstable(pmd))
1836 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1838 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1841 gather_stats(page, md, pte_dirty(*pte), 1);
1843 } while (pte++, addr += PAGE_SIZE, addr != end);
1844 pte_unmap_unlock(orig_pte, ptl);
1848 #ifdef CONFIG_HUGETLB_PAGE
1849 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1850 unsigned long addr, unsigned long end, struct mm_walk *walk)
1852 pte_t huge_pte = huge_ptep_get(pte);
1853 struct numa_maps *md;
1856 if (!pte_present(huge_pte))
1859 page = pte_page(huge_pte);
1864 gather_stats(page, md, pte_dirty(huge_pte), 1);
1869 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1870 unsigned long addr, unsigned long end, struct mm_walk *walk)
1876 static const struct mm_walk_ops show_numa_ops = {
1877 .hugetlb_entry = gather_hugetlb_stats,
1878 .pmd_entry = gather_pte_stats,
1882 * Display pages allocated per node and memory policy via /proc.
1884 static int show_numa_map(struct seq_file *m, void *v)
1886 struct numa_maps_private *numa_priv = m->private;
1887 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1888 struct vm_area_struct *vma = v;
1889 struct numa_maps *md = &numa_priv->md;
1890 struct file *file = vma->vm_file;
1891 struct mm_struct *mm = vma->vm_mm;
1892 struct mempolicy *pol;
1899 /* Ensure we start with an empty set of numa_maps statistics. */
1900 memset(md, 0, sizeof(*md));
1902 pol = __get_vma_policy(vma, vma->vm_start);
1904 mpol_to_str(buffer, sizeof(buffer), pol);
1907 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1910 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1913 seq_puts(m, " file=");
1914 seq_file_path(m, file, "\n\t= ");
1915 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1916 seq_puts(m, " heap");
1917 } else if (is_stack(vma)) {
1918 seq_puts(m, " stack");
1921 if (is_vm_hugetlb_page(vma))
1922 seq_puts(m, " huge");
1924 /* mmap_lock is held by m_start */
1925 walk_page_vma(vma, &show_numa_ops, md);
1931 seq_printf(m, " anon=%lu", md->anon);
1934 seq_printf(m, " dirty=%lu", md->dirty);
1936 if (md->pages != md->anon && md->pages != md->dirty)
1937 seq_printf(m, " mapped=%lu", md->pages);
1939 if (md->mapcount_max > 1)
1940 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1943 seq_printf(m, " swapcache=%lu", md->swapcache);
1945 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1946 seq_printf(m, " active=%lu", md->active);
1949 seq_printf(m, " writeback=%lu", md->writeback);
1951 for_each_node_state(nid, N_MEMORY)
1953 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1955 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1961 static const struct seq_operations proc_pid_numa_maps_op = {
1965 .show = show_numa_map,
1968 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1970 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1971 sizeof(struct numa_maps_private));
1974 const struct file_operations proc_pid_numa_maps_operations = {
1975 .open = pid_numa_maps_open,
1977 .llseek = seq_lseek,
1978 .release = proc_map_release,
1981 #endif /* CONFIG_NUMA */