1 // SPDX-License-Identifier: GPL-2.0
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>
24 #include <asm/tlbflush.h>
27 void task_mem(struct seq_file *m, struct mm_struct *mm)
29 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
30 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
32 anon = get_mm_counter(mm, MM_ANONPAGES);
33 file = get_mm_counter(mm, MM_FILEPAGES);
34 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
37 * Note: to minimize their overhead, mm maintains hiwater_vm and
38 * hiwater_rss only when about to *lower* total_vm or rss. Any
39 * collector of these hiwater stats must therefore get total_vm
40 * and rss too, which will usually be the higher. Barriers? not
41 * worth the effort, such snapshots can always be inconsistent.
43 hiwater_vm = total_vm = mm->total_vm;
44 if (hiwater_vm < mm->hiwater_vm)
45 hiwater_vm = mm->hiwater_vm;
46 hiwater_rss = total_rss = anon + file + shmem;
47 if (hiwater_rss < mm->hiwater_rss)
48 hiwater_rss = mm->hiwater_rss;
50 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
51 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
52 swap = get_mm_counter(mm, MM_SWAPENTS);
53 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
54 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
64 "RssShmem:\t%8lu kB\n"
72 hiwater_vm << (PAGE_SHIFT-10),
73 total_vm << (PAGE_SHIFT-10),
74 mm->locked_vm << (PAGE_SHIFT-10),
75 mm->pinned_vm << (PAGE_SHIFT-10),
76 hiwater_rss << (PAGE_SHIFT-10),
77 total_rss << (PAGE_SHIFT-10),
78 anon << (PAGE_SHIFT-10),
79 file << (PAGE_SHIFT-10),
80 shmem << (PAGE_SHIFT-10),
81 mm->data_vm << (PAGE_SHIFT-10),
82 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
85 swap << (PAGE_SHIFT-10));
86 hugetlb_report_usage(m, mm);
89 unsigned long task_vsize(struct mm_struct *mm)
91 return PAGE_SIZE * mm->total_vm;
94 unsigned long task_statm(struct mm_struct *mm,
95 unsigned long *shared, unsigned long *text,
96 unsigned long *data, unsigned long *resident)
98 *shared = get_mm_counter(mm, MM_FILEPAGES) +
99 get_mm_counter(mm, MM_SHMEMPAGES);
100 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
102 *data = mm->data_vm + mm->stack_vm;
103 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
109 * Save get_task_policy() for show_numa_map().
111 static void hold_task_mempolicy(struct proc_maps_private *priv)
113 struct task_struct *task = priv->task;
116 priv->task_mempolicy = get_task_policy(task);
117 mpol_get(priv->task_mempolicy);
120 static void release_task_mempolicy(struct proc_maps_private *priv)
122 mpol_put(priv->task_mempolicy);
125 static void hold_task_mempolicy(struct proc_maps_private *priv)
128 static void release_task_mempolicy(struct proc_maps_private *priv)
133 static void vma_stop(struct proc_maps_private *priv)
135 struct mm_struct *mm = priv->mm;
137 release_task_mempolicy(priv);
138 up_read(&mm->mmap_sem);
142 static struct vm_area_struct *
143 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
145 if (vma == priv->tail_vma)
147 return vma->vm_next ?: priv->tail_vma;
150 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
152 if (m->count < m->size) /* vma is copied successfully */
153 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
156 static void *m_start(struct seq_file *m, loff_t *ppos)
158 struct proc_maps_private *priv = m->private;
159 unsigned long last_addr = m->version;
160 struct mm_struct *mm;
161 struct vm_area_struct *vma;
162 unsigned int pos = *ppos;
164 /* See m_cache_vma(). Zero at the start or after lseek. */
165 if (last_addr == -1UL)
168 priv->task = get_proc_task(priv->inode);
170 return ERR_PTR(-ESRCH);
173 if (!mm || !mmget_not_zero(mm))
176 down_read(&mm->mmap_sem);
177 hold_task_mempolicy(priv);
178 priv->tail_vma = get_gate_vma(mm);
181 vma = find_vma(mm, last_addr - 1);
182 if (vma && vma->vm_start <= last_addr)
183 vma = m_next_vma(priv, vma);
189 if (pos < mm->map_count) {
190 for (vma = mm->mmap; pos; pos--) {
191 m->version = vma->vm_start;
197 /* we do not bother to update m->version in this case */
198 if (pos == mm->map_count && priv->tail_vma)
199 return priv->tail_vma;
205 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
207 struct proc_maps_private *priv = m->private;
208 struct vm_area_struct *next;
211 next = m_next_vma(priv, v);
217 static void m_stop(struct seq_file *m, void *v)
219 struct proc_maps_private *priv = m->private;
221 if (!IS_ERR_OR_NULL(v))
224 put_task_struct(priv->task);
229 static int proc_maps_open(struct inode *inode, struct file *file,
230 const struct seq_operations *ops, int psize)
232 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
238 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
239 if (IS_ERR(priv->mm)) {
240 int err = PTR_ERR(priv->mm);
242 seq_release_private(inode, file);
249 static int proc_map_release(struct inode *inode, struct file *file)
251 struct seq_file *seq = file->private_data;
252 struct proc_maps_private *priv = seq->private;
258 return seq_release_private(inode, file);
261 static int do_maps_open(struct inode *inode, struct file *file,
262 const struct seq_operations *ops)
264 return proc_maps_open(inode, file, ops,
265 sizeof(struct proc_maps_private));
269 * Indicate if the VMA is a stack for the given task; for
270 * /proc/PID/maps that is the stack of the main task.
272 static int is_stack(struct vm_area_struct *vma)
275 * We make no effort to guess what a given thread considers to be
276 * its "stack". It's not even well-defined for programs written
279 return vma->vm_start <= vma->vm_mm->start_stack &&
280 vma->vm_end >= vma->vm_mm->start_stack;
283 static void show_vma_header_prefix(struct seq_file *m,
284 unsigned long start, unsigned long end,
285 vm_flags_t flags, unsigned long long pgoff,
286 dev_t dev, unsigned long ino)
288 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
289 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
292 flags & VM_READ ? 'r' : '-',
293 flags & VM_WRITE ? 'w' : '-',
294 flags & VM_EXEC ? 'x' : '-',
295 flags & VM_MAYSHARE ? 's' : 'p',
297 MAJOR(dev), MINOR(dev), ino);
301 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
303 struct mm_struct *mm = vma->vm_mm;
304 struct file *file = vma->vm_file;
305 vm_flags_t flags = vma->vm_flags;
306 unsigned long ino = 0;
307 unsigned long long pgoff = 0;
308 unsigned long start, end;
310 const char *name = NULL;
313 struct inode *inode = file_inode(vma->vm_file);
314 dev = inode->i_sb->s_dev;
316 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
319 start = vma->vm_start;
321 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
324 * Print the dentry name for named mappings, and a
325 * special [heap] marker for the heap:
329 seq_file_path(m, file, "\n");
333 if (vma->vm_ops && vma->vm_ops->name) {
334 name = vma->vm_ops->name(vma);
339 name = arch_vma_name(vma);
346 if (vma->vm_start <= mm->brk &&
347 vma->vm_end >= mm->start_brk) {
364 static int show_map(struct seq_file *m, void *v, int is_pid)
366 show_map_vma(m, v, is_pid);
371 static int show_pid_map(struct seq_file *m, void *v)
373 return show_map(m, v, 1);
376 static int show_tid_map(struct seq_file *m, void *v)
378 return show_map(m, v, 0);
381 static const struct seq_operations proc_pid_maps_op = {
388 static const struct seq_operations proc_tid_maps_op = {
395 static int pid_maps_open(struct inode *inode, struct file *file)
397 return do_maps_open(inode, file, &proc_pid_maps_op);
400 static int tid_maps_open(struct inode *inode, struct file *file)
402 return do_maps_open(inode, file, &proc_tid_maps_op);
405 const struct file_operations proc_pid_maps_operations = {
406 .open = pid_maps_open,
409 .release = proc_map_release,
412 const struct file_operations proc_tid_maps_operations = {
413 .open = tid_maps_open,
416 .release = proc_map_release,
420 * Proportional Set Size(PSS): my share of RSS.
422 * PSS of a process is the count of pages it has in memory, where each
423 * page is divided by the number of processes sharing it. So if a
424 * process has 1000 pages all to itself, and 1000 shared with one other
425 * process, its PSS will be 1500.
427 * To keep (accumulated) division errors low, we adopt a 64bit
428 * fixed-point pss counter to minimize division errors. So (pss >>
429 * PSS_SHIFT) would be the real byte count.
431 * A shift of 12 before division means (assuming 4K page size):
432 * - 1M 3-user-pages add up to 8KB errors;
433 * - supports mapcount up to 2^24, or 16M;
434 * - supports PSS up to 2^52 bytes, or 4PB.
438 #ifdef CONFIG_PROC_PAGE_MONITOR
439 struct mem_size_stats {
441 unsigned long resident;
442 unsigned long shared_clean;
443 unsigned long shared_dirty;
444 unsigned long private_clean;
445 unsigned long private_dirty;
446 unsigned long referenced;
447 unsigned long anonymous;
448 unsigned long lazyfree;
449 unsigned long anonymous_thp;
450 unsigned long shmem_thp;
452 unsigned long shared_hugetlb;
453 unsigned long private_hugetlb;
454 unsigned long first_vma_start;
458 bool check_shmem_swap;
461 static void smaps_account(struct mem_size_stats *mss, struct page *page,
462 bool compound, bool young, bool dirty, bool locked)
464 int i, nr = compound ? 1 << compound_order(page) : 1;
465 unsigned long size = nr * PAGE_SIZE;
467 if (PageAnon(page)) {
468 mss->anonymous += size;
469 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
470 mss->lazyfree += size;
473 mss->resident += size;
474 /* Accumulate the size in pages that have been accessed. */
475 if (young || page_is_young(page) || PageReferenced(page))
476 mss->referenced += size;
479 * page_count(page) == 1 guarantees the page is mapped exactly once.
480 * If any subpage of the compound page mapped with PTE it would elevate
483 if (page_count(page) == 1) {
484 if (dirty || PageDirty(page))
485 mss->private_dirty += size;
487 mss->private_clean += size;
488 mss->pss += (u64)size << PSS_SHIFT;
490 mss->pss_locked += (u64)size << PSS_SHIFT;
494 for (i = 0; i < nr; i++, page++) {
495 int mapcount = page_mapcount(page);
496 unsigned long pss = (PAGE_SIZE << PSS_SHIFT);
499 if (dirty || PageDirty(page))
500 mss->shared_dirty += PAGE_SIZE;
502 mss->shared_clean += PAGE_SIZE;
503 mss->pss += pss / mapcount;
505 mss->pss_locked += pss / mapcount;
507 if (dirty || PageDirty(page))
508 mss->private_dirty += PAGE_SIZE;
510 mss->private_clean += PAGE_SIZE;
513 mss->pss_locked += pss;
519 static int smaps_pte_hole(unsigned long addr, unsigned long end,
520 struct mm_walk *walk)
522 struct mem_size_stats *mss = walk->private;
524 mss->swap += shmem_partial_swap_usage(
525 walk->vma->vm_file->f_mapping, addr, end);
531 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
532 struct mm_walk *walk)
534 struct mem_size_stats *mss = walk->private;
535 struct vm_area_struct *vma = walk->vma;
536 bool locked = !!(vma->vm_flags & VM_LOCKED);
537 struct page *page = NULL;
539 if (pte_present(*pte)) {
540 page = vm_normal_page(vma, addr, *pte);
541 } else if (is_swap_pte(*pte)) {
542 swp_entry_t swpent = pte_to_swp_entry(*pte);
544 if (!non_swap_entry(swpent)) {
547 mss->swap += PAGE_SIZE;
548 mapcount = swp_swapcount(swpent);
550 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
552 do_div(pss_delta, mapcount);
553 mss->swap_pss += pss_delta;
555 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
557 } else if (is_migration_entry(swpent))
558 page = migration_entry_to_page(swpent);
559 else if (is_device_private_entry(swpent))
560 page = device_private_entry_to_page(swpent);
561 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
562 && pte_none(*pte))) {
563 page = find_get_entry(vma->vm_file->f_mapping,
564 linear_page_index(vma, addr));
568 if (radix_tree_exceptional_entry(page))
569 mss->swap += PAGE_SIZE;
579 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
582 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
583 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
584 struct mm_walk *walk)
586 struct mem_size_stats *mss = walk->private;
587 struct vm_area_struct *vma = walk->vma;
588 bool locked = !!(vma->vm_flags & VM_LOCKED);
591 /* FOLL_DUMP will return -EFAULT on huge zero page */
592 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
593 if (IS_ERR_OR_NULL(page))
596 mss->anonymous_thp += HPAGE_PMD_SIZE;
597 else if (PageSwapBacked(page))
598 mss->shmem_thp += HPAGE_PMD_SIZE;
599 else if (is_zone_device_page(page))
602 VM_BUG_ON_PAGE(1, page);
603 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
606 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
607 struct mm_walk *walk)
612 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
613 struct mm_walk *walk)
615 struct vm_area_struct *vma = walk->vma;
619 ptl = pmd_trans_huge_lock(pmd, vma);
621 if (pmd_present(*pmd))
622 smaps_pmd_entry(pmd, addr, walk);
627 if (pmd_trans_unstable(pmd))
630 * The mmap_sem held all the way back in m_start() is what
631 * keeps khugepaged out of here and from collapsing things
634 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
635 for (; addr != end; pte++, addr += PAGE_SIZE)
636 smaps_pte_entry(pte, addr, walk);
637 pte_unmap_unlock(pte - 1, ptl);
643 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
646 * Don't forget to update Documentation/ on changes.
648 static const char mnemonics[BITS_PER_LONG][2] = {
650 * In case if we meet a flag we don't know about.
652 [0 ... (BITS_PER_LONG-1)] = "??",
654 [ilog2(VM_READ)] = "rd",
655 [ilog2(VM_WRITE)] = "wr",
656 [ilog2(VM_EXEC)] = "ex",
657 [ilog2(VM_SHARED)] = "sh",
658 [ilog2(VM_MAYREAD)] = "mr",
659 [ilog2(VM_MAYWRITE)] = "mw",
660 [ilog2(VM_MAYEXEC)] = "me",
661 [ilog2(VM_MAYSHARE)] = "ms",
662 [ilog2(VM_GROWSDOWN)] = "gd",
663 [ilog2(VM_PFNMAP)] = "pf",
664 [ilog2(VM_DENYWRITE)] = "dw",
665 #ifdef CONFIG_X86_INTEL_MPX
666 [ilog2(VM_MPX)] = "mp",
668 [ilog2(VM_LOCKED)] = "lo",
669 [ilog2(VM_IO)] = "io",
670 [ilog2(VM_SEQ_READ)] = "sr",
671 [ilog2(VM_RAND_READ)] = "rr",
672 [ilog2(VM_DONTCOPY)] = "dc",
673 [ilog2(VM_DONTEXPAND)] = "de",
674 [ilog2(VM_ACCOUNT)] = "ac",
675 [ilog2(VM_NORESERVE)] = "nr",
676 [ilog2(VM_HUGETLB)] = "ht",
677 [ilog2(VM_ARCH_1)] = "ar",
678 [ilog2(VM_WIPEONFORK)] = "wf",
679 [ilog2(VM_DONTDUMP)] = "dd",
680 #ifdef CONFIG_MEM_SOFT_DIRTY
681 [ilog2(VM_SOFTDIRTY)] = "sd",
683 [ilog2(VM_MIXEDMAP)] = "mm",
684 [ilog2(VM_HUGEPAGE)] = "hg",
685 [ilog2(VM_NOHUGEPAGE)] = "nh",
686 [ilog2(VM_MERGEABLE)] = "mg",
687 [ilog2(VM_UFFD_MISSING)]= "um",
688 [ilog2(VM_UFFD_WP)] = "uw",
689 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
690 /* These come out via ProtectionKey: */
691 [ilog2(VM_PKEY_BIT0)] = "",
692 [ilog2(VM_PKEY_BIT1)] = "",
693 [ilog2(VM_PKEY_BIT2)] = "",
694 [ilog2(VM_PKEY_BIT3)] = "",
699 seq_puts(m, "VmFlags: ");
700 for (i = 0; i < BITS_PER_LONG; i++) {
701 if (!mnemonics[i][0])
703 if (vma->vm_flags & (1UL << i)) {
704 seq_printf(m, "%c%c ",
705 mnemonics[i][0], mnemonics[i][1]);
711 #ifdef CONFIG_HUGETLB_PAGE
712 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
713 unsigned long addr, unsigned long end,
714 struct mm_walk *walk)
716 struct mem_size_stats *mss = walk->private;
717 struct vm_area_struct *vma = walk->vma;
718 struct page *page = NULL;
720 if (pte_present(*pte)) {
721 page = vm_normal_page(vma, addr, *pte);
722 } else if (is_swap_pte(*pte)) {
723 swp_entry_t swpent = pte_to_swp_entry(*pte);
725 if (is_migration_entry(swpent))
726 page = migration_entry_to_page(swpent);
727 else if (is_device_private_entry(swpent))
728 page = device_private_entry_to_page(swpent);
731 if (page_mapcount(page) >= 2 || hugetlb_pmd_shared(pte))
732 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
734 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
738 #endif /* HUGETLB_PAGE */
740 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
744 static int show_smap(struct seq_file *m, void *v, int is_pid)
746 struct proc_maps_private *priv = m->private;
747 struct vm_area_struct *vma = v;
748 struct mem_size_stats mss_stack;
749 struct mem_size_stats *mss;
750 struct mm_walk smaps_walk = {
751 .pmd_entry = smaps_pte_range,
752 #ifdef CONFIG_HUGETLB_PAGE
753 .hugetlb_entry = smaps_hugetlb_range,
765 mss->first_vma_start = vma->vm_start;
768 last_vma = !m_next_vma(priv, vma);
771 memset(&mss_stack, 0, sizeof(mss_stack));
775 smaps_walk.private = mss;
778 /* In case of smaps_rollup, reset the value from previous vma */
779 mss->check_shmem_swap = false;
780 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
782 * For shared or readonly shmem mappings we know that all
783 * swapped out pages belong to the shmem object, and we can
784 * obtain the swap value much more efficiently. For private
785 * writable mappings, we might have COW pages that are
786 * not affected by the parent swapped out pages of the shmem
787 * object, so we have to distinguish them during the page walk.
788 * Unless we know that the shmem object (or the part mapped by
789 * our VMA) has no swapped out pages at all.
791 unsigned long shmem_swapped = shmem_swap_usage(vma);
793 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
794 !(vma->vm_flags & VM_WRITE)) {
795 mss->swap += shmem_swapped;
797 mss->check_shmem_swap = true;
798 smaps_walk.pte_hole = smaps_pte_hole;
802 /* mmap_sem is held in m_start */
803 walk_page_vma(vma, &smaps_walk);
806 show_map_vma(m, vma, is_pid);
807 } else if (last_vma) {
808 show_vma_header_prefix(
809 m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
811 seq_puts(m, "[rollup]\n");
819 "KernelPageSize: %8lu kB\n"
820 "MMUPageSize: %8lu kB\n",
821 (vma->vm_end - vma->vm_start) >> 10,
822 vma_kernel_pagesize(vma) >> 10,
823 vma_mmu_pagesize(vma) >> 10);
826 if (!rollup_mode || last_vma)
830 "Shared_Clean: %8lu kB\n"
831 "Shared_Dirty: %8lu kB\n"
832 "Private_Clean: %8lu kB\n"
833 "Private_Dirty: %8lu kB\n"
834 "Referenced: %8lu kB\n"
835 "Anonymous: %8lu kB\n"
836 "LazyFree: %8lu kB\n"
837 "AnonHugePages: %8lu kB\n"
838 "ShmemPmdMapped: %8lu kB\n"
839 "Shared_Hugetlb: %8lu kB\n"
840 "Private_Hugetlb: %7lu kB\n"
845 (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
846 mss->shared_clean >> 10,
847 mss->shared_dirty >> 10,
848 mss->private_clean >> 10,
849 mss->private_dirty >> 10,
850 mss->referenced >> 10,
851 mss->anonymous >> 10,
853 mss->anonymous_thp >> 10,
854 mss->shmem_thp >> 10,
855 mss->shared_hugetlb >> 10,
856 mss->private_hugetlb >> 10,
858 (unsigned long)(mss->swap_pss >> (10 + PSS_SHIFT)),
859 (unsigned long)(mss->pss_locked >> (10 + PSS_SHIFT)));
862 arch_show_smap(m, vma);
863 show_smap_vma_flags(m, vma);
869 static int show_pid_smap(struct seq_file *m, void *v)
871 return show_smap(m, v, 1);
874 static int show_tid_smap(struct seq_file *m, void *v)
876 return show_smap(m, v, 0);
879 static const struct seq_operations proc_pid_smaps_op = {
883 .show = show_pid_smap
886 static const struct seq_operations proc_tid_smaps_op = {
890 .show = show_tid_smap
893 static int pid_smaps_open(struct inode *inode, struct file *file)
895 return do_maps_open(inode, file, &proc_pid_smaps_op);
898 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
900 struct seq_file *seq;
901 struct proc_maps_private *priv;
902 int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
906 seq = file->private_data;
908 priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
910 proc_map_release(inode, file);
913 priv->rollup->first = true;
917 static int tid_smaps_open(struct inode *inode, struct file *file)
919 return do_maps_open(inode, file, &proc_tid_smaps_op);
922 const struct file_operations proc_pid_smaps_operations = {
923 .open = pid_smaps_open,
926 .release = proc_map_release,
929 const struct file_operations proc_pid_smaps_rollup_operations = {
930 .open = pid_smaps_rollup_open,
933 .release = proc_map_release,
936 const struct file_operations proc_tid_smaps_operations = {
937 .open = tid_smaps_open,
940 .release = proc_map_release,
943 enum clear_refs_types {
947 CLEAR_REFS_SOFT_DIRTY,
948 CLEAR_REFS_MM_HIWATER_RSS,
952 struct clear_refs_private {
953 enum clear_refs_types type;
956 #ifdef CONFIG_MEM_SOFT_DIRTY
957 static inline void clear_soft_dirty(struct vm_area_struct *vma,
958 unsigned long addr, pte_t *pte)
961 * The soft-dirty tracker uses #PF-s to catch writes
962 * to pages, so write-protect the pte as well. See the
963 * Documentation/vm/soft-dirty.txt for full description
964 * of how soft-dirty works.
968 if (pte_present(ptent)) {
969 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
970 ptent = pte_wrprotect(ptent);
971 ptent = pte_clear_soft_dirty(ptent);
972 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
973 } else if (is_swap_pte(ptent)) {
974 ptent = pte_swp_clear_soft_dirty(ptent);
975 set_pte_at(vma->vm_mm, addr, pte, ptent);
979 static inline void clear_soft_dirty(struct vm_area_struct *vma,
980 unsigned long addr, pte_t *pte)
985 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
986 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
987 unsigned long addr, pmd_t *pmdp)
991 if (pmd_present(pmd)) {
992 /* See comment in change_huge_pmd() */
993 pmdp_invalidate(vma, addr, pmdp);
994 if (pmd_dirty(*pmdp))
995 pmd = pmd_mkdirty(pmd);
996 if (pmd_young(*pmdp))
997 pmd = pmd_mkyoung(pmd);
999 pmd = pmd_wrprotect(pmd);
1000 pmd = pmd_clear_soft_dirty(pmd);
1002 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1003 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1004 pmd = pmd_swp_clear_soft_dirty(pmd);
1005 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1009 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1010 unsigned long addr, pmd_t *pmdp)
1015 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1016 unsigned long end, struct mm_walk *walk)
1018 struct clear_refs_private *cp = walk->private;
1019 struct vm_area_struct *vma = walk->vma;
1024 ptl = pmd_trans_huge_lock(pmd, vma);
1026 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1027 clear_soft_dirty_pmd(vma, addr, pmd);
1031 if (!pmd_present(*pmd))
1034 page = pmd_page(*pmd);
1036 /* Clear accessed and referenced bits. */
1037 pmdp_test_and_clear_young(vma, addr, pmd);
1038 test_and_clear_page_young(page);
1039 ClearPageReferenced(page);
1045 if (pmd_trans_unstable(pmd))
1048 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1049 for (; addr != end; pte++, addr += PAGE_SIZE) {
1052 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1053 clear_soft_dirty(vma, addr, pte);
1057 if (!pte_present(ptent))
1060 page = vm_normal_page(vma, addr, ptent);
1064 /* Clear accessed and referenced bits. */
1065 ptep_test_and_clear_young(vma, addr, pte);
1066 test_and_clear_page_young(page);
1067 ClearPageReferenced(page);
1069 pte_unmap_unlock(pte - 1, ptl);
1074 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1075 struct mm_walk *walk)
1077 struct clear_refs_private *cp = walk->private;
1078 struct vm_area_struct *vma = walk->vma;
1080 if (vma->vm_flags & VM_PFNMAP)
1084 * Writing 1 to /proc/pid/clear_refs affects all pages.
1085 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1086 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1087 * Writing 4 to /proc/pid/clear_refs affects all pages.
1089 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1091 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1096 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1097 size_t count, loff_t *ppos)
1099 struct task_struct *task;
1100 char buffer[PROC_NUMBUF];
1101 struct mm_struct *mm;
1102 struct vm_area_struct *vma;
1103 enum clear_refs_types type;
1104 struct mmu_gather tlb;
1108 memset(buffer, 0, sizeof(buffer));
1109 if (count > sizeof(buffer) - 1)
1110 count = sizeof(buffer) - 1;
1111 if (copy_from_user(buffer, buf, count))
1113 rv = kstrtoint(strstrip(buffer), 10, &itype);
1116 type = (enum clear_refs_types)itype;
1117 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1120 task = get_proc_task(file_inode(file));
1123 mm = get_task_mm(task);
1125 struct clear_refs_private cp = {
1128 struct mm_walk clear_refs_walk = {
1129 .pmd_entry = clear_refs_pte_range,
1130 .test_walk = clear_refs_test_walk,
1135 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1136 if (down_write_killable(&mm->mmap_sem)) {
1142 * Writing 5 to /proc/pid/clear_refs resets the peak
1143 * resident set size to this mm's current rss value.
1145 reset_mm_hiwater_rss(mm);
1146 up_write(&mm->mmap_sem);
1150 down_read(&mm->mmap_sem);
1151 tlb_gather_mmu(&tlb, mm, 0, -1);
1152 if (type == CLEAR_REFS_SOFT_DIRTY) {
1153 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1154 if (!(vma->vm_flags & VM_SOFTDIRTY))
1156 up_read(&mm->mmap_sem);
1157 if (down_write_killable(&mm->mmap_sem)) {
1162 * Avoid to modify vma->vm_flags
1163 * without locked ops while the
1164 * coredump reads the vm_flags.
1166 if (!mmget_still_valid(mm)) {
1168 * Silently return "count"
1169 * like if get_task_mm()
1170 * failed. FIXME: should this
1171 * function have returned
1172 * -ESRCH if get_task_mm()
1174 * get_proc_task() fails?
1176 up_write(&mm->mmap_sem);
1179 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1180 vma->vm_flags &= ~VM_SOFTDIRTY;
1181 vma_set_page_prot(vma);
1183 downgrade_write(&mm->mmap_sem);
1186 mmu_notifier_invalidate_range_start(mm, 0, -1);
1188 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1189 if (type == CLEAR_REFS_SOFT_DIRTY)
1190 mmu_notifier_invalidate_range_end(mm, 0, -1);
1191 tlb_finish_mmu(&tlb, 0, -1);
1192 up_read(&mm->mmap_sem);
1196 put_task_struct(task);
1201 const struct file_operations proc_clear_refs_operations = {
1202 .write = clear_refs_write,
1203 .llseek = noop_llseek,
1210 struct pagemapread {
1211 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1212 pagemap_entry_t *buffer;
1216 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1217 #define PAGEMAP_WALK_MASK (PMD_MASK)
1219 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1220 #define PM_PFRAME_BITS 55
1221 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1222 #define PM_SOFT_DIRTY BIT_ULL(55)
1223 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1224 #define PM_FILE BIT_ULL(61)
1225 #define PM_SWAP BIT_ULL(62)
1226 #define PM_PRESENT BIT_ULL(63)
1228 #define PM_END_OF_BUFFER 1
1230 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1232 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1235 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1236 struct pagemapread *pm)
1238 pm->buffer[pm->pos++] = *pme;
1239 if (pm->pos >= pm->len)
1240 return PM_END_OF_BUFFER;
1244 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1245 struct mm_walk *walk)
1247 struct pagemapread *pm = walk->private;
1248 unsigned long addr = start;
1251 while (addr < end) {
1252 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1253 pagemap_entry_t pme = make_pme(0, 0);
1254 /* End of address space hole, which we mark as non-present. */
1255 unsigned long hole_end;
1258 hole_end = min(end, vma->vm_start);
1262 for (; addr < hole_end; addr += PAGE_SIZE) {
1263 err = add_to_pagemap(addr, &pme, pm);
1271 /* Addresses in the VMA. */
1272 if (vma->vm_flags & VM_SOFTDIRTY)
1273 pme = make_pme(0, PM_SOFT_DIRTY);
1274 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1275 err = add_to_pagemap(addr, &pme, pm);
1284 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1285 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1287 u64 frame = 0, flags = 0;
1288 struct page *page = NULL;
1290 if (pte_present(pte)) {
1292 frame = pte_pfn(pte);
1293 flags |= PM_PRESENT;
1294 page = _vm_normal_page(vma, addr, pte, true);
1295 if (pte_soft_dirty(pte))
1296 flags |= PM_SOFT_DIRTY;
1297 } else if (is_swap_pte(pte)) {
1299 if (pte_swp_soft_dirty(pte))
1300 flags |= PM_SOFT_DIRTY;
1301 entry = pte_to_swp_entry(pte);
1303 frame = swp_type(entry) |
1304 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1306 if (is_migration_entry(entry))
1307 page = migration_entry_to_page(entry);
1309 if (is_device_private_entry(entry))
1310 page = device_private_entry_to_page(entry);
1313 if (page && !PageAnon(page))
1315 if (page && page_mapcount(page) == 1)
1316 flags |= PM_MMAP_EXCLUSIVE;
1317 if (vma->vm_flags & VM_SOFTDIRTY)
1318 flags |= PM_SOFT_DIRTY;
1320 return make_pme(frame, flags);
1323 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1324 struct mm_walk *walk)
1326 struct vm_area_struct *vma = walk->vma;
1327 struct pagemapread *pm = walk->private;
1329 pte_t *pte, *orig_pte;
1332 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1333 ptl = pmd_trans_huge_lock(pmdp, vma);
1335 u64 flags = 0, frame = 0;
1337 struct page *page = NULL;
1339 if (vma->vm_flags & VM_SOFTDIRTY)
1340 flags |= PM_SOFT_DIRTY;
1342 if (pmd_present(pmd)) {
1343 page = pmd_page(pmd);
1345 flags |= PM_PRESENT;
1346 if (pmd_soft_dirty(pmd))
1347 flags |= PM_SOFT_DIRTY;
1349 frame = pmd_pfn(pmd) +
1350 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1352 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1353 else if (is_swap_pmd(pmd)) {
1354 swp_entry_t entry = pmd_to_swp_entry(pmd);
1355 unsigned long offset;
1358 offset = swp_offset(entry) +
1359 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1360 frame = swp_type(entry) |
1361 (offset << MAX_SWAPFILES_SHIFT);
1364 if (pmd_swp_soft_dirty(pmd))
1365 flags |= PM_SOFT_DIRTY;
1366 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1367 page = migration_entry_to_page(entry);
1371 if (page && page_mapcount(page) == 1)
1372 flags |= PM_MMAP_EXCLUSIVE;
1374 for (; addr != end; addr += PAGE_SIZE) {
1375 pagemap_entry_t pme = make_pme(frame, flags);
1377 err = add_to_pagemap(addr, &pme, pm);
1381 if (flags & PM_PRESENT)
1383 else if (flags & PM_SWAP)
1384 frame += (1 << MAX_SWAPFILES_SHIFT);
1391 if (pmd_trans_unstable(pmdp))
1393 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1396 * We can assume that @vma always points to a valid one and @end never
1397 * goes beyond vma->vm_end.
1399 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1400 for (; addr < end; pte++, addr += PAGE_SIZE) {
1401 pagemap_entry_t pme;
1403 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1404 err = add_to_pagemap(addr, &pme, pm);
1408 pte_unmap_unlock(orig_pte, ptl);
1415 #ifdef CONFIG_HUGETLB_PAGE
1416 /* This function walks within one hugetlb entry in the single call */
1417 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1418 unsigned long addr, unsigned long end,
1419 struct mm_walk *walk)
1421 struct pagemapread *pm = walk->private;
1422 struct vm_area_struct *vma = walk->vma;
1423 u64 flags = 0, frame = 0;
1427 if (vma->vm_flags & VM_SOFTDIRTY)
1428 flags |= PM_SOFT_DIRTY;
1430 pte = huge_ptep_get(ptep);
1431 if (pte_present(pte)) {
1432 struct page *page = pte_page(pte);
1434 if (!PageAnon(page))
1437 if (page_mapcount(page) == 1)
1438 flags |= PM_MMAP_EXCLUSIVE;
1440 flags |= PM_PRESENT;
1442 frame = pte_pfn(pte) +
1443 ((addr & ~hmask) >> PAGE_SHIFT);
1446 for (; addr != end; addr += PAGE_SIZE) {
1447 pagemap_entry_t pme = make_pme(frame, flags);
1449 err = add_to_pagemap(addr, &pme, pm);
1452 if (pm->show_pfn && (flags & PM_PRESENT))
1460 #endif /* HUGETLB_PAGE */
1463 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1465 * For each page in the address space, this file contains one 64-bit entry
1466 * consisting of the following:
1468 * Bits 0-54 page frame number (PFN) if present
1469 * Bits 0-4 swap type if swapped
1470 * Bits 5-54 swap offset if swapped
1471 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1472 * Bit 56 page exclusively mapped
1474 * Bit 61 page is file-page or shared-anon
1475 * Bit 62 page swapped
1476 * Bit 63 page present
1478 * If the page is not present but in swap, then the PFN contains an
1479 * encoding of the swap file number and the page's offset into the
1480 * swap. Unmapped pages return a null PFN. This allows determining
1481 * precisely which pages are mapped (or in swap) and comparing mapped
1482 * pages between processes.
1484 * Efficient users of this interface will use /proc/pid/maps to
1485 * determine which areas of memory are actually mapped and llseek to
1486 * skip over unmapped regions.
1488 static ssize_t pagemap_read(struct file *file, char __user *buf,
1489 size_t count, loff_t *ppos)
1491 struct mm_struct *mm = file->private_data;
1492 struct pagemapread pm;
1493 struct mm_walk pagemap_walk = {};
1495 unsigned long svpfn;
1496 unsigned long start_vaddr;
1497 unsigned long end_vaddr;
1498 int ret = 0, copied = 0;
1500 if (!mm || !mmget_not_zero(mm))
1504 /* file position must be aligned */
1505 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1512 /* do not disclose physical addresses: attack vector */
1513 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1515 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1516 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1521 pagemap_walk.pmd_entry = pagemap_pmd_range;
1522 pagemap_walk.pte_hole = pagemap_pte_hole;
1523 #ifdef CONFIG_HUGETLB_PAGE
1524 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1526 pagemap_walk.mm = mm;
1527 pagemap_walk.private = ±
1530 svpfn = src / PM_ENTRY_BYTES;
1531 start_vaddr = svpfn << PAGE_SHIFT;
1532 end_vaddr = mm->task_size;
1534 /* watch out for wraparound */
1535 if (svpfn > mm->task_size >> PAGE_SHIFT)
1536 start_vaddr = end_vaddr;
1539 * The odds are that this will stop walking way
1540 * before end_vaddr, because the length of the
1541 * user buffer is tracked in "pm", and the walk
1542 * will stop when we hit the end of the buffer.
1545 while (count && (start_vaddr < end_vaddr)) {
1550 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1552 if (end < start_vaddr || end > end_vaddr)
1554 down_read(&mm->mmap_sem);
1555 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1556 up_read(&mm->mmap_sem);
1559 len = min(count, PM_ENTRY_BYTES * pm.pos);
1560 if (copy_to_user(buf, pm.buffer, len)) {
1569 if (!ret || ret == PM_END_OF_BUFFER)
1580 static int pagemap_open(struct inode *inode, struct file *file)
1582 struct mm_struct *mm;
1584 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1587 file->private_data = mm;
1591 static int pagemap_release(struct inode *inode, struct file *file)
1593 struct mm_struct *mm = file->private_data;
1600 const struct file_operations proc_pagemap_operations = {
1601 .llseek = mem_lseek, /* borrow this */
1602 .read = pagemap_read,
1603 .open = pagemap_open,
1604 .release = pagemap_release,
1606 #endif /* CONFIG_PROC_PAGE_MONITOR */
1611 unsigned long pages;
1613 unsigned long active;
1614 unsigned long writeback;
1615 unsigned long mapcount_max;
1616 unsigned long dirty;
1617 unsigned long swapcache;
1618 unsigned long node[MAX_NUMNODES];
1621 struct numa_maps_private {
1622 struct proc_maps_private proc_maps;
1623 struct numa_maps md;
1626 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1627 unsigned long nr_pages)
1629 int count = page_mapcount(page);
1631 md->pages += nr_pages;
1632 if (pte_dirty || PageDirty(page))
1633 md->dirty += nr_pages;
1635 if (PageSwapCache(page))
1636 md->swapcache += nr_pages;
1638 if (PageActive(page) || PageUnevictable(page))
1639 md->active += nr_pages;
1641 if (PageWriteback(page))
1642 md->writeback += nr_pages;
1645 md->anon += nr_pages;
1647 if (count > md->mapcount_max)
1648 md->mapcount_max = count;
1650 md->node[page_to_nid(page)] += nr_pages;
1653 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1659 if (!pte_present(pte))
1662 page = vm_normal_page(vma, addr, pte);
1666 if (PageReserved(page))
1669 nid = page_to_nid(page);
1670 if (!node_isset(nid, node_states[N_MEMORY]))
1676 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1677 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1678 struct vm_area_struct *vma,
1684 if (!pmd_present(pmd))
1687 page = vm_normal_page_pmd(vma, addr, pmd);
1691 if (PageReserved(page))
1694 nid = page_to_nid(page);
1695 if (!node_isset(nid, node_states[N_MEMORY]))
1702 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1703 unsigned long end, struct mm_walk *walk)
1705 struct numa_maps *md = walk->private;
1706 struct vm_area_struct *vma = walk->vma;
1711 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1712 ptl = pmd_trans_huge_lock(pmd, vma);
1716 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1718 gather_stats(page, md, pmd_dirty(*pmd),
1719 HPAGE_PMD_SIZE/PAGE_SIZE);
1724 if (pmd_trans_unstable(pmd))
1727 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1729 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1732 gather_stats(page, md, pte_dirty(*pte), 1);
1734 } while (pte++, addr += PAGE_SIZE, addr != end);
1735 pte_unmap_unlock(orig_pte, ptl);
1739 #ifdef CONFIG_HUGETLB_PAGE
1740 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1741 unsigned long addr, unsigned long end, struct mm_walk *walk)
1743 pte_t huge_pte = huge_ptep_get(pte);
1744 struct numa_maps *md;
1747 if (!pte_present(huge_pte))
1750 page = pte_page(huge_pte);
1755 gather_stats(page, md, pte_dirty(huge_pte), 1);
1760 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1761 unsigned long addr, unsigned long end, struct mm_walk *walk)
1768 * Display pages allocated per node and memory policy via /proc.
1770 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1772 struct numa_maps_private *numa_priv = m->private;
1773 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1774 struct vm_area_struct *vma = v;
1775 struct numa_maps *md = &numa_priv->md;
1776 struct file *file = vma->vm_file;
1777 struct mm_struct *mm = vma->vm_mm;
1778 struct mm_walk walk = {
1779 .hugetlb_entry = gather_hugetlb_stats,
1780 .pmd_entry = gather_pte_stats,
1784 struct mempolicy *pol;
1791 /* Ensure we start with an empty set of numa_maps statistics. */
1792 memset(md, 0, sizeof(*md));
1794 pol = __get_vma_policy(vma, vma->vm_start);
1796 mpol_to_str(buffer, sizeof(buffer), pol);
1799 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1802 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1805 seq_puts(m, " file=");
1806 seq_file_path(m, file, "\n\t= ");
1807 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1808 seq_puts(m, " heap");
1809 } else if (is_stack(vma)) {
1810 seq_puts(m, " stack");
1813 if (is_vm_hugetlb_page(vma))
1814 seq_puts(m, " huge");
1816 /* mmap_sem is held by m_start */
1817 walk_page_vma(vma, &walk);
1823 seq_printf(m, " anon=%lu", md->anon);
1826 seq_printf(m, " dirty=%lu", md->dirty);
1828 if (md->pages != md->anon && md->pages != md->dirty)
1829 seq_printf(m, " mapped=%lu", md->pages);
1831 if (md->mapcount_max > 1)
1832 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1835 seq_printf(m, " swapcache=%lu", md->swapcache);
1837 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1838 seq_printf(m, " active=%lu", md->active);
1841 seq_printf(m, " writeback=%lu", md->writeback);
1843 for_each_node_state(nid, N_MEMORY)
1845 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1847 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1850 m_cache_vma(m, vma);
1854 static int show_pid_numa_map(struct seq_file *m, void *v)
1856 return show_numa_map(m, v, 1);
1859 static int show_tid_numa_map(struct seq_file *m, void *v)
1861 return show_numa_map(m, v, 0);
1864 static const struct seq_operations proc_pid_numa_maps_op = {
1868 .show = show_pid_numa_map,
1871 static const struct seq_operations proc_tid_numa_maps_op = {
1875 .show = show_tid_numa_map,
1878 static int numa_maps_open(struct inode *inode, struct file *file,
1879 const struct seq_operations *ops)
1881 return proc_maps_open(inode, file, ops,
1882 sizeof(struct numa_maps_private));
1885 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1887 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1890 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1892 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1895 const struct file_operations proc_pid_numa_maps_operations = {
1896 .open = pid_numa_maps_open,
1898 .llseek = seq_lseek,
1899 .release = proc_map_release,
1902 const struct file_operations proc_tid_numa_maps_operations = {
1903 .open = tid_numa_maps_open,
1905 .llseek = seq_lseek,
1906 .release = proc_map_release,
1908 #endif /* CONFIG_NUMA */